Search is not available for this dataset
article
stringlengths 4.36k
149k
| summary
stringlengths 32
3.35k
| section_headings
listlengths 1
91
| keywords
listlengths 0
141
| year
stringclasses 13
values | title
stringlengths 20
281
|
|---|---|---|---|---|---|
The role of the pore-forming Staphylococcus aureus toxin Panton-Valentine leukocidin ( PVL ) in severe necrotizing diseases is debated due to conflicting data from epidemiological studies of community-associated methicillin-resistant S . aureus ( CA-MRSA ) infections and various murine disease-models . In this study , we used neutrophils isolated from different species to evaluate the cytotoxic effect of PVL in comparison to other staphylococcal cytolytic components . Furthermore , to study the impact of PVL we expressed it heterologously in a non-virulent staphylococcal species and examined pvl-positive and pvl-negative clinical isolates as well as the strain USA300 and its pvl-negative mutant . We demonstrate that PVL induces rapid activation and cell death in human and rabbit neutrophils , but not in murine or simian cells . By contrast , the phenol-soluble modulins ( PSMs ) , a newly identified group of cytolytic staphylococcal components , lack species-specificity . In general , after phagocytosis of bacteria different pvl-positive and pvl-negative staphylococcal strains , expressing a variety of other virulence factors ( such as surface proteins ) , induced cell death in neutrophils , which is most likely associated with the physiological clearing function of these cells . However , the release of PVL by staphylococcal strains caused rapid and premature cell death , which is different from the physiological ( and programmed ) cell death of neutrophils following phagocytosis and degradation of virulent bacteria . Taken together , our results question the value of infection-models in mice and non-human primates to elucidate the impact of PVL . Our data clearly demonstrate that PVL acts differentially on neutrophils of various species and suggests that PVL has an important cytotoxic role in human neutrophils , which has major implications for the pathogenesis of CA-MRSA infections .
Staphylococcus aureus is an important human pathogen that can cause serious diseases [1] . In the last few years , there was a dramatic increase in the incidence of community-associated methicillin-resistant S . aureus ( CA-MRSA ) infections in otherwise healthy individuals and resistance to multiple antibiotic classes largely limits therapeutic options . Especially the MRSA strain USA300 has widely spread within the United States and has become the cause of more unusually severe diseases , including necrotizing pneumonia , skin infections , osteomyelitis and necrotizing fasciitis [2] , [3] . Necrotizing pneumonia seems to be a specific disease entity and often follows infection with influenza virus [4] , [5] . To combat these life-threatening infections , there is a need to better understand the bacteria-host interaction and virulence factors involved . Clinical studies propose the exotoxin Panton-Valentine leukocidin ( PVL ) as a crucial virulence factor in necrotizing diseases [4] , [6] . PVL is a two component pore-forming toxin , which mainly acts on neutrophils [7] . It is expressed by only a small percentage of S . aureus wild-type isolates ( 2–3% ) [8] , but it is highly prevalent in S . aureus strains isolated from necrotizing infections [4] , [6] . However , several studies that used a diversity of animal models have created conflicting results concerning the role of PVL . One study , applying a mouse acute pneumonia model , suggests PVL as major virulence factor [9] . By contrast , other groups fail to detect a pathogenic function of PVL in murine lung and skin infections and in cell culture experiments , but demonstrate a predominant role of α-hemolysin ( α-toxin ) and a possible relevance of the bacterial surface protein A ( Spa ) [10]–[12] . Both factors are expressed at high prevalence among clinical isolates and are considered to contribute to various disease entities [1] , [13] , [14] . Yet , when a rabbit bacteremia model was used , a transient effect of PVL in the acute phase of infection could be demonstrated [15] . Furthermore , a recent study identified a group of S . aureus peptides , the phenol-soluble modulins ( PSMs ) , with strong cytolytic activity on human neutrophils . As PSMs are released at high concentrations by CA-MRSA strains and contribute to disease development in murine models , the authors propose that PSMs account for the enhanced virulence of CA-MRSA [16] . However , there is some evidence that the actions of S . aureus toxins can be strongly dependent on the animal species used , which should be analysed in detail to better interpret disease-models . In particular , the host cell response to PVL may be species-specific [17] , whereas the effects of other staphylococcal factors , such as PSMs , might be species-independent . In this study , we used polymorphonuclear cells ( neutrophils ) from different species including humans , mice , rabbits and monkeys to test the effect of several virulence factors . As neutrophils are the major defending cells against bacterial invasion , their excessive cell death most likely largely promotes disease development .
First , we challenged human neutrophils with purified S . aureus components , including PVL , α-toxin , protein A and PSMs . For PVL , doses≥40 ng/ml ( 0 . 04 µg/ml ) were sufficient to induce cell damage ( Figure 1A ) . Cell death occurred rapidly , within 1 h , and most likely due to necrosis , as we could not detect characteristic apoptotic features ( Figure S2 ) [18] . In contrast to PVL , α-toxin or protein A did not cause cell death , even when applied at high concentrations , which have pro-inflammatory or cytotoxic effects in other cell types [19] , [20] . As recently published [16] , three different forms of S . aureus PSMs ( PSMα1 , PSMα2 , PSMα3 ) were able to provoke cell-lysis . However , cell death induction required relatively high doses of PSMs ( ≥40 µg/ml ) in comparison to PVL ( ≥40 ng/ml ) ( Figure 1A ) . In previous studies , the impact of PVL was mainly tested on human or rabbit neutrophils , as cells from both species were reported to be susceptible to PVL [17] . In line with published data , we found similar responses of human and rabbit neutrophils to low doses of PVL ( Figure 1B ) . The action of PVL appears to be tightly restricted to these species , as neutrophils isolated from Java monkeys ( Macaca fascicularis , cynomolgus ) , the most commonly used non-human primate in biomedical research , were not killed in response to PVL ( Figure 1C ) . In recent reports , models of severe staphylococcal infections were mainly performed in the murine strains BALB/c or C57/BL6 [9] , [10] . However , murine neutrophils from both strains were largely resistant to PVL ( Figure 1D , E ) , irrespective of their maturation and inflammatory state ( Figure S3 ) . In contrast to PVL , all PSM-types tested ( PSMα1–3 ) lysed neutrophils from different species at concentrations≥40 µg/ml , indicating that the actions of PSMs apparently lack species-specificity ( Figure 1A–E ) . Further on , we detected additional differences between PVL- and PSM-induced cell death . Incubation with PVL caused changes in cell morphology , including rounding and swelling of cells and nuclei ( Figure 2A , 2B ) , which persisted for several hours ( data not shown ) . By contrast , PSM-stimulated cells were rapidly destroyed without characteristic changes in morphology ( Figure 2A ) . In PVL-treated neutrophils , an oxidative burst reaction ( Figure 2C ) and pro-inflammatory activation ( Figure S4 ) accompanied cell death induction , whereas incubation with PSMs did not cause an oxidative burst ( Figure 2C ) . These results point to completely different mechanisms of action provoked by the S . aureus cytotoxic components PVL and PSMs . To investigate the impact of defined virulence factor expression we transformed S . carnosus TM300 with a plasmid encoding the genes for PVL , α-toxin , protein A ( Spa ) or PSMs , respectively ( Table 1 ) . Using live bacteria with these constructs revealed that the expression of PVL most efficiently induced neutrophils cell death ( Figure 3A ) . The effect of TM300+PVL was comparable to the cytotoxic potential of clonally independent MRSA ( ST239 ) and MSSA ( 6850 ) strains ( Figure 3B ) and of pvl-positive clinical isolates , which were recovered from severe invasive ( including necrotizing pneumonia ) diseases ( Figure 3C ) . However , cytotoxicity was not restricted to PVL-expressing strains , as live bacteria of some pvl-negative isolates compromised cell viability to a similar extent ( Figure 3D ) . Moreover , we could not detect differences between strain USA300 and the corresponding mutant USA300ΔPVL ( Figure 3B ) , indicating that the presence of the pvl-gene does not necessarily contribute to neutrophils cell death following phagocytosis of bacteria . We also failed to block the cytotoxic effect of USA300 by the use of antibodies against PVL ( Figure S5 ) . These findings indicate that other staphylococcal factors can also induce cell death , which might mask the cytotoxic function of PVL . However , the expression of α-toxin and PSMs in TM300 had no effect on neutrophils . Apparently , PSMs need to accumulate to lyse neutrophils , as the corresponding bacterial supernatants , which contained PSMs , were cytolytic ( Figure S6 ) [16] . Besides PVL , the expression of protein A moderately decreased the number of intact cells ( Figure 3A ) . This is further demonstrated by using strain Cowan I , which is a high producer of protein A , whereas two isogenic mutants ( Δspa ) were much less cytotoxic ( Figure 3E ) . Although protein A is known to be a cell wall-anchored protein with an anti-phagocytic effect [14] , we observed an increased rate of cell death . In our experiments , the action of protein A was dependent on the expression by bacteria , which exhibit protein A on the bacterial surface . This phenomenon was not specific for protein A , as the expression of another wall-associated protein , namely fibronectin-binding protein A ( FnBPA ) , also reduced the number of intact neutrophils ( Figure 3F ) . In general , phagocytosis of pathogens triggers mechanisms to kill ingested bacteria . Further on , it has been shown that phagocytosis significantly accelerates neutrophils apoptosis , which appears to contribute to the resolution of the inflammatory response [21] , [22] . These processes promote healthy resolution and could be an explanation for the enhanced rate of cell death caused by bacteria holding virulent surface proteins . This assumption is further confirmed by apoptotic features detected in neutrophils ( Figure S7 , annexin V-positive cells ) . However , bacterial toxins , such as PVL and PSMs , could interfere with the physiological functions of neutrophils , by rapidly and prematurely killing cells . To investigate this possibility we analysed neutrophils cell death in a time-dependent manner . Challenge with PVL ( ≥40 ng/ml ) induced cell death within the first 20 min ( Figure 4A ) , whereas incubation of neutrophils with live bacteria resulted in a much slower rate of death induction ( within 2–3 h ) , which is most likely associated with the neutrophils physiological function [21] . Using PVL-expressing ( USA300 ) or non PVL-expressing ( ST239 , 6850 ) strains did not reveal any differences ( Figure 4B ) . PVL is a bacterial exotoxin , which is rapidly released and could act on cells at the infection sites . To mimic this situation , we stimulated neutrophils with sterile-filtered bacterial supernatants from overnight cultures . Culture media from strain TM300+PVL induced rapid cell death within 20 min , whereas supernatants from the control strain TM300 did not affect cell integrity . Further on , comparing supernatants from the wild-type strain USA300 with supernatants from the corresponding knock-out mutant USA300ΔPVL revealed that culture media from the PVL-deletion strain had a much reduced ability to induce cell death , as the majority of cells remained intact ( Figure 5A ) . The impact of PVL release was further strengthened by testing clinical isolates . Supernatants from four out of six pvl-positive strains , recovered from severe ( including necrotizing ) diseases , had a much higher cytotoxic activity than supernatants from pvl-negative strains ( also recovered from severe invasive diseases ) . It is of particular importance , that PVL secretion of the strains ( measured by Western blot in the bacterial supernatants; Figure S1C ) clearly corresponded to the cytotoxic activity in all cases ( Figure 5B ) .
The role of PVL in severe CA-MRSA infections is debated due to conflicting data from epidemiological studies , in vitro cell culture experiments , and different animal disease models [9] , [10] , [12] , [23] . As PVL was found in almost all MRSA strains that cause CA-MRSA infections , such as necrotizing pneumonia , skin- and soft tissue infections , it was assumed to be a crucial virulence factor [4] , [6] . These disease entities are characterized by massive tissue necrosis and leukopenia , which has been linked to the ability of PVL to kill neutrophils , the primary defending cells against invading bacteria . However , different disease models in mice , in which USA300 and the corresponding knock-out mutant were used , failed to detect a pathogenic function for PVL [10] , [12] . In line with these data , we found that murine neutrophils , isolated from different commonly used mice strains , were quite insensitive to PVL . Neutrophils from Java monkeys , a species much more closely related to humans , were not affected by PVL . The reason for the differential sensitivity of cells from various species is completely unknown , but receptors/signal transduction pathways , which are confined to certain species , might be involved . By contrast , all PSM-types tested lysed neutrophils from different species equally efficient and induced membrane damaging effects [16] . Since murine and simian cells are largely resistant to PVL , PSMs might play a more dominant role in S . aureus infections in mice or non-human primates than in humans , especially when high S . aureus inocula are needed to cause diseases . Our data strongly suggest that animal models using mice or non-human primates do not correctly replicate S . aureus diseases in humans , at least if the role of PVL is elucidated . As neutrophils from rabbits are much more susceptible to PVL , this species is most likely more appropriate to study the function of PVL in necrotizing diseases . Very recently , a rabbit bacteremia model has been published , which describes a modest and transient effect of PVL in the acute phase of infection [15] . However , this type of infection might not show the full pathogenicity of PVL expression , as during bacteremia staphylococci are directly exposed to cells of the immune system . In our experiments , we could not detect differences in virulence between PVL-expressing and pvl-negative ( knock-out mutants or wild-type isolates ) strains , when live bacteria were directly phagocytized by neutrophils . This is in line with other published data , demonstrating that disruption or absence of the pvl-gene in S . aureus wild-type isolates ( including USA300 ) did not alter their capacity to induce neutrophils cell death [12] , [24] . Nevertheless , as cell death of neutrophils is part of the physiological immune response following phagocytosis of bacteria [21] , [22] and as S . aureus wild-type isolates express a wide variety of factors promoting this process ( e . g . diverse surface proteins ) [1] , the effect of secreted PVL on human neutrophils might be masked in this model . Furthermore , it is reasonable to suspect that PVL is not ( highly ) expressed , when staphylococci are instantly phagocytized by neutrophils , as toxic virulence factors were found to be down-regulated after internalization of bacteria [25] . Like other toxins , PVL is mainly expressed in the post-exponential bacterial growth phase [26] , which is most likely reached in encapsulated infection foci , e . g . folliculitis , abscesses , tissue necrosis . Only recently , high expression of PVL was found directly in clinical samples from cutaneous abscesses of invasive CA-MRSA infections [27] . Here , PVL most likely accumulates and can also exert systemic pathogenic actions upon entering the bloodstream . In human neutrophils , low doses of PVL were sufficient to cause cell death , which correspond to amounts produced by clinical CA-MRSA strains [28] . Granted that the action of PVL involves yet unknown host receptors/signal transduction pathways , PVL might interfere with various functions of susceptible cells . Furthermore , it is reasonable to speculate that host organisms can become even more vulnerable against PVL , e . g . following an infection with influenza virus . Additional studies on human cells and in susceptible animal models ( rabbits ) will be necessary to clarify these possibilities and to better define the functions of PVL in staphylococcal infections . Taken together , our results clearly demonstrate that PVL is a strong cytotoxic factor for human neutrophils , which can play an important role in CA-MRSA infections . Our results do not contradict previously published work , as we could not find an effect of PVL on murine neutrophils or when bacteria were directly phagocytised by neutrophils . However , under certain pathogenic conditions , such as necrosis and abscesses , which are characteristic for severe invasive S . aureus diseases , PVL could exert its function as a cytotoxic exotoxin in susceptible organisms . The premature cell death of neutrophils may be extremely relevant in the virulence of CA-MRSA . As neutrophils are the major defense against invading bacteria , their excessive cell death most likely largely compromises the host's immune system . Furthermore , uncontrolled neutrophils cell damage discharges many pro-inflammatory components within the host tissue , which could also essentially promote disease development . These results are important for ongoing efforts to find therapeutics against S . aureus infections . Due to the rapid spread of CA-MRSA strains and situations , which favour S . aureus infections at a large scale , e . g . epidemic of influenza , there is an urgent need for efficient preventive and therapeutic strategies .
Taking of blood samples from humans and animals and cell isolation were conducted with approval of the local ethics committee ( Ethik-Komission der Ärztekammer Westfalen-Lippe und der Medizinischen Fakultät der Westfälischen Wilhelms-Universität Münster ) . Human blood samples were taken from healthy blood donors , who provided written informed consent for the collection of samples and subsequent neutrophil isolation and analysis . All animals were handled in strict accordance with good animal practice and animal keeping and taking of blood samples were supervised by the veterinary office of Münster ( Veterinäramt der Stadt Münster ) . Bacterial strains used in this study are listed in table 1 . They were all characterized for presence of genes encoding PVL and α-toxin by PCR . Gene expression was investigated by Western blots for PVL ( Figure S1B+C ) or by hemolysis on sheep blood agar plates ( sign for α-toxin production ) . For cell culture and animal experiments with live staphylococci , bacteria were grown overnight at 37°C in Müller-Hinton medium ( MH , containing antibiotics/xylose , if mutants are used ) without shaking . Bacteria were washed in PBS and resuspended in PBS with 1% HSA . Neutrophils were incubated with bacterial suspensions , resulting in a multiplicity of infection ( MOI ) as indicated . Bacterial supernatants were prepared by growing bacteria in 5 ml of brain-heart infusion ( BHI ) broth ( Merck ) in a rotary shaker ( 160 rpm ) at 37°C for 12–14 h and pelleted for 10 min at 3350 g . Supernatants were sterile-filtered through a Millex-GP filter unit ( 0 . 22 µm; Millipore ) and used for the experiments . For PVL isolation , E . coli TG1 strains containing expression vectors for lukF-PV and lukS-PV were grown in Luria Bertani ( LB ) -media with IPTG ( 1 mM ) and ampicillin ( 100 mg/ml ) and cell lysates were used to purify PVL ( Figure S1A ) . Different genes were amplified by PCR using chromosomal DNA from different strains ( Table S1 ) as template . To create S . carnosus strains , which express virulence factors of S . aureus , we used two basic vectors , the xylose inducible pXR100 and the pNXR100 , which is a non-inducible derivate of the pXR100 . For the expression of lukF-PV and lukS-PV in E . coli TG1 the commercial IPTG inducible pQE30UA was used . For creation of the expression vectors the respective genes were amplified by PCR , purified and digested . The basic vectors were also digested corresponding to the genes . After ligation S . carnosus TM300 and E . coli were transformed by protoplast transformation or CaCl–method . The Six-histagged lukF-PV and lukS-PV proteins from E . coli were purified by nickel-nitrilotriacetic acid affinity resin ( Qiagen , Germany ) . α-toxin and Protein A ( P3838 ) were obtained from Sigma-Aldrich Chemie GmbH ( Germany ) . PSMα1 – PSMα3 were synthesized by Genosphere Biotechnology ( France ) . Polyclonal antibodies against lukF-PV and lukS-PV were raised separately and together in rabbits by standard procedures and this was performed by Genosphere-Biotechnology ( France ) . Human , rabbit and Java monkey polymorphonuclear cells ( neutrophils ) were freshly isolated from Na citrate-treated blood of healthy donors . Neutrophils from BALB/c and C57/BL6 mice were prepared from bone marrow . For neutrophil-isolation , dextran-sedimentation and density gradient centrifugation using Ficoll-Paque Plus ( Amersham Bioscience ) was used according to the manufacturer's instruction . Cell purity was determined by Giemsa staining and was always above 99% . For murine cells , sedimented cells were used as neutrophils and , in addition , were further deprived of CD3+ ( T cells ) , CD19+ ( B cells ) , and CD11c+ ( dendritic cells ) cells using MACS technology ( Miltenyi Biotech , Bergisch- Gladbach ) according to the manufacturer’s instruction . Resulting cells were <0 . 1% CD3+ , CD19+ , or CD11c+ and <95% CD11b+ and Gr1+ . Neutrophils were resuspended at a final density of 1×106 cells/0 . 5 ml in RPMI 1640 culture medium ( PAA Laboratories GmbH ) supplemented with 10% heat-inactivated FCS ( PAA Laboratories GmbH ) and immediately used for the experiments . All incubations were performed at 37°C in humidified air with 5% CO2 . All experiments were performed in 24-well plates and neutrophils were incubated with PVL , α-toxin , PSMs , live bacteria or bacterial supernatants at the indicated concentrations . Oxidative burst activity was determined after 10 min of incubation using a phagoburst test ( Orpegen Pharma ) according to the manufacturer's instruction . Measurement of cell death was performed after 1 h of incubation followed by washing and double staining of cells with annexin V-FITC and propidium iodide ( PI ) ( taking 1 hour ) and then cells were analyzed in a FACScalibur flow cytometer using an annexin V-FITC apoptosis detection kit ( Becton Dickinson ) . For analysis of time-dependent cell death inductions , cells were incubated for the indicated time periods , followed by washing and single staining with PI ( taking 10 min ) and then cells were immediately analysed by flow cytometry . A live cell imaging system ( Zeiss ) was used to obtain light micrographs . For transmission electron microscopy , 5×106/2 . 5 ml neutrophils were incubated with PVL at the indicated concentrations for 1 h . Then the cells were washed three times with PBS , fixed in 3% glutaraldehyde , stained in 1% osmium tetroxide and embedded in epoxy resin in the culture dish in situ . Electron micrographs were obtained using imaging plate technology . Unpaired Student’s t-test was performed to compare cell survival . A value of P≤0 . 05 was considered significant in all cases .
|
Staphylococcus aureus can cause serious diseases , including necrotizing pneumonia , which often affects young immunocompetent patients and has a high lethality rate . Several clinical studies demonstrated a clear association between this form of pneumonia and S . aureus strains carrying the gene for the pore-forming toxin Panton-Valentine leukocidin ( PVL ) . However , laboratory work , which mainly used murine disease models , has created very contrasting results and often fails to show a pathogenic role for PVL . In this study , we demonstrate that the expression of PVL by staphylococcal strains confers strong and rapid cytotoxic activity against neutrophils . However , this action was basically restricted to human cells and could not be reproduced in murine or Java monkeys’ cells . These results indicate that infection-models in mice and in non-human primates fail to replicate the pathogenic activity of PVL seen in human cells . Our data with human neutrophils clearly show that PVL has a major cytotoxic effect , as the release of PVL by staphylococcal strains caused rapid and premature cell death , which is different from the physiological ( and programmed ) cell death of neutrophils following phagocytosis and degradation of virulent bacteria . These results have important implications especially for infections with CA-MRSA strains , which often carry the gene for PVL and have spread widely in the community .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"microbiology/cellular",
"microbiology",
"and",
"pathogenesis",
"microbiology/innate",
"immunity"
] |
2010
|
Staphylococcus aureus Panton-Valentine Leukocidin Is a Very Potent Cytotoxic Factor for Human Neutrophils
|
The Gram-negative soil dwelling bacterium Burkholderia pseudomallei is the etiological agent of melioidosis . The disease is endemic in most parts of Southeast Asia and northern Australia . Over last few years , there has been an increase in number of melioidosis cases from India; however the disease epidemiology is less clearly understood . Multi-locus sequence typing ( MLST ) is a powerful genotypic method used to characterize the genetic diversity of B . Pseudomallei both within and across the geographic regions . In this study , MLST analysis was performed on 64 B . pseudomallei clinical isolates . These isolates were obtained between 2008–2014 from southwestern coastal region of India . Broad population patterns of Indian B . pseudomallei isolates in context with isolates of Southeast Asia or global collection was determined using in silico phylogenetic tools . A total of 32 Sequence types ( STs ) were reported among these isolates of which 17 STs ( 53% ) were found to be novel . ST1368 was found as group founder and the most predominant genotype ( n = 11 , 17% ) . Most of the B . pseudomallei isolates reported in this study ( or other Indian isolates available in MLST database ) clustered in one major group suggesting clonality in Indian isolates; however , there were a few outliers . When analyzed by measure of genetic differentiation ( FST ) and other phylogenetic tools ( e . g . PHYLOViZ ) , Indian STs were found closer to Southeast Asian isolates than Australian isolates . The phylogenetic analysis further revealed that within Asian clade , Indian isolates grouped more closely with isolates from Sri Lanka , Vietnam , Bangladesh and Thailand . Overall , the results of this study suggest that the Indian B . pseudomallei isolates are closely related with lesser heterogeneity among them and cluster in one major group suggesting clonality of the isolates . However , it appears that there are a few outliers which are distantly related to the majority of Indian STs . Phylogenetic analysis suggest that Indian isolates are closely related to isolates from Southeast Asia , particularly from South Asia .
Melioidosis , caused by soil saprophyte Gram-negative bacterium Burkholderia pseudomallei , is classically characterized by pneumonia , septicemia and multiple abscesses . There are various predisposing factors for melioidosis with diabetes mellitus as one of the most important factors . Until recently , the disease was considered endemic only to Southeast Asia and northern Australia , however now it has been reported from tropical , subtropical and temperate regions [1 , 2] . Reports of melioidosis from India had been few and sporadic [3 , 4] , however , over the past few years , there has been an increase in number of melioidosis cases . It has been reported from various states of India including Karnataka , Kerala , Maharashtra , Tamil Nadu and Puducherry [5] . The disease is quite prevalent in southwestern costal part of the country and is strongly associated with rainfall and diabetes mellitus [5 , 6] . A mathematical modeling study of 2015 has predicted global annual burden of melioidosis to be 165 , 000 cases with 89 , 000 deaths with Indian subcontinent to have highest burden of the disease [7] . Regional variations in melioidosis signs and symptoms have been reported . It has been seen that while prostatic abscess and encephalomyelitis are common in Australians whereas parotid abscess and hepatosplenic suppuration are most frequently seen in patients from Thailand [8] . The reasons for this diversity remain unclear but it could be due to host , bacterial and environmental factors [9] . Study of epidemiology by molecular methods provides insight about bacterial diversity and distribution . Various molecular methods e . g . pulse field electrophoresis , ribotyping have been used for phylogenetic reconstruction with different levels of success , however , multi-locus sequence typing ( MLST ) is a proven tool for the molecular typing of B . pseudomallei . MLST not only helps to explore the sequence types ( STs ) in a population or helps tracing an outbreak [10] but also helps to understand the microbial evolution . There is large MLST database for B . pseudomallei ( http://pubmlst . org/bpseudomallei/ ) . In India , melioidosis is an emerging endemic infection and potentially fatal as early diagnosis is missed due to its varied manifestations such as localized or disseminated infection . Here , it has largely been reported from the coastal regions and it is generally believed that the disease is underreported or misdiagnosed . Using MLST , we had previously reported in a pilot study that Indian isolates were genetically diverse from the Australian or Southeast Asian isolates [11] . In this work , we aimed to study the genetic diversity among larger number of B . pseudomallei isolates using MLST . Further , using sequence data we attempted to find broad population patterns of Indian isolates with global collection of 5541 isolates and construct phylogeny of B . pseudomallei Indian isolates and their close relatives from Southeast Asia .
This study was approved by Institutional Ethical Committee of Kasturba Hospital , Manipal under protocol number IEC 141/2011 . Sixty-four ( n = 64 ) clinical isolates of B . pseudomallei isolated from Karnataka and adjacent states ( Kerala , Goa and Puducherry ) of southern India were included in this study . These isolates were collected from melioidosis patients at Kasturba Medical College , Manipal , Karnataka during 2008–2014 . All the isolates were identified using API20NE and later confirmed using TTS1-PCR assay [12] and latex agglutination test from the colonies . The isolates were also characterized for genetic markers linked to geographic origin Yersinia-like fimbriae ( YLF ) and B . thailendensis-like flagellum and chemotaxis ( BTFC ) gene [13] . For preparation of DNA , bacteria were grown overnight in Luria- Bertani ( LB ) broth at 37°C in high containment facility , a biosafety level 3 facility at Defence Research & Development Establishment ( DRDE ) , Gwalior . Genomic DNA was isolated from culture using DNeasy blood and tissue genomic DNA kit ( Qiagen Gmbh , Hilden ) , according to the manufacturer’s instructions . The genomic DNAs were stored at -20°C till further used . MLST was carried out according to the method of Godoy et al [14] . Primer sequences targeting the conserved regions of seven housekeeping genes ( ace- gltB- gmhD- lepA- lipA- narK- ndh ) of B . pseudomallei were used as shown in MLST site ( http://pubmlst . org/bpseudomallei/ ) . PCR amplification , sequence analysis and determination of ST for each isolate were carried out by our earlier reported procedure [11] . The purified PCR amplicons were double pass sequenced using commercial SANGER sequencing services ( M/s Genotypic Technology Pvt . Ltd . , Bengaluru ) . The relatedness of MLST profiles of isolates of this study or Indian isolates in B . pseudomallei MLST database was performed using eBURST software [15] ( launched at http://eburst . mlst . net/ ) with single locus variant ( SLV ) selected . eBURST or global optimal based upon related sequence type ( goeBURST ) allows for an unrooted tree-based representation of the relationship of the analyzed isolates . The diversification of the "founding" genotype is reflected in the appearance of STs differing only in one housekeeping gene sequence from the founder genotype–single locus variants ( SLVs ) . Further diversification of those SLVs results in the appearance of variations of the original genotype with more than one difference in the allelic profile: double locus variants ( DLVs ) , triple locus variants ( TLVs ) and so on . The final eBURST tree provides a hypothetical pattern of descent for the strains analyzed . Basic quantities such as number of alleles , number of variable sites per allele , number and frequency of single nucleotide polymorphism ( SNPs ) in each locus were determined . Nucleotide sequence diversity ( π ) was calculated using DNAsp V5 . 1 software [16] . The association of individual STs with the type of infection among the study population was carried out using Fisher exact test ( Graph Pad Prism software , La Jolla , USA ) . Measures of genetic differentiation ( FST ) between the concatenated sequences of Indian origin and from Asia or Australia was estimated using DNAsp V5 . 1 [16] . Relationship of Indian STs with global collection of STs was studied using goeBURST [17] implemented in PHYLOViZ programme [18] available at MLST site ( http://pubmlst . org/bpseudomallei ) . PHYLOViZ is a flexible and expandable plugin based tool that is able to handle large datasets and builds on goeBURST implementation . In order to further determine the relationship of Indian STs with STs from Asia and Australia , topology and grouping of all Indian STs were displayed on constructed boot strapped phylogenetic tree using Unweighted Pair Group Method with Arithmetic average ( UPGMA ) method in molecular evolutionary genetic analysis version -6 ( MEGA 6 ) software [19] . Indian STs including STs of this study were analyzed with selected 57 STs from other Asian countries e . g . Sri Lanka , Vietnam , Bangladesh , Cambodia , Malaysia , Thailand , China and Laos and , 6 most predominant STs from Australia . The 57 STs were either SLV or DLV of the Indian STs . Concatenated sequences of all STs used for this study were downloaded from MLST site ( http://pubmlst . org/bpseudomallei/ ) .
Among 64 isolates of this study , a total of 32 STs were identified . The frequencies of STs among isolates ranged from 1–11 with ST1368 ( n = 11 ) , ST42 , ST1373 , ST1478 ( n = 4 ) , ST124 , ST1512 , ST1507 , ST1375 ( n = 3 ) being the most predominant STs . Seventeen STs identified in this study ( ST1373 , ST1374 , ST1506 , ST1507 , ST1508 , ST1509 , ST1510 , ST1511 , ST1512 , ST1513 , ST1514 , ST1515 , ST1516 , ST1517 , ST1518 , ST1519 , ST1520 ) were found to be novel that were not reported previously and 15 STs were previously documented , five of which were reported as novel in our previous study [11] . When analyzed by eBURST , almost all STs of this study formed in one large group with ST1368 as predicted founder; only ST1141 , ST1374 , ST1506 and ST859 were found to be singleton . ST1368 had 7 SLV , 4 DLV , 7 TLV and 9 satellites STs . ST550 , ST124 , ST42 , ST1511 , ST1512 , and ST1518 were found to be subgroup founders . Interestingly , all novel 17 STs of this study except ST1506 and ST1374 also clustered in the same group ( Fig 1 , S2 Table ) . A total of 130 isolates are now available ( as on 16th October , 2018 ) in B . pseudomallei MLST database from India which include data from 64 isolates of this study . All isolates from India except one , which was isolated from soil , were clinical isolates . Of the 65 STs , 47 STs ( 109 isolates ) cluster in one group with the remaining 18 STs ( 21 isolates ) being singleton . ST1368 is dispersed all over the four Indian states of Karnataka , Kerala , Goa and Puducherry ( S1 Table ) , [11] . It had a frequency of 26 with 7 SLV , 10 DLV , 11 TLV and 18 satellites STs . The present study has expanded the clonal cluster of Indian isolates by adding more branching STs . eBURST analysis of total Indian isolates revealed ST1513 and ST1372 to be as additional sub-group founders ( Fig 2 ) . Measure of genetic differentiation ( FST ) between Indian and Australian or Asian isolates was determined and was found to be 0 . 1561 and 0 . 09082 respectively , suggesting that Indian isolates are closer to Asian clade rather than Australian clade . When Indian STs were analyzed by PHYLOViZ with the global collection of 5541 isolates in the B . pseudomallei MLST isolates database ( as on 19th May , 2018 ) , majority of Indian isolates grouped in four groups ( Fig 3 ) . Group A ( ST42 , ST550 , ST1051 , ST1370 , ST1555 ) and B ( ST405 , ST856 , ST1637 ) clustered with Asian clade and accounted for about 43% ( 28/65 ) of the known STs . Indian isolates appeared to be different from the main Thailand cluster and grouped closure to isolates from South Asia e . g . Sri Lanka , Bangladesh . There was overlap of Indian STs with STs of other Asian countries e . g . Bangladesh ( ST42 , ST43 , ST300 ) ; Vietnam ( ST550 , ST858 , ST1051 ) ; Sri Lanka ( ST293 , ST1141 ) ; Thailand ( ST300 , ST405 ) and China ( ST405 ) . About 57% of Indian STs grouped with Australian STs in two groups ( Group C–ST124 , ST1375 , ST1507 , ST1512 , ST1552 , ST1554 and Group D—ST1368 , ST1372 , ST1373 ) . ST468 and ST1051 were the only Indian STs that overlapped with Australia . However , all isolates of this study were YLF positive , a gene cluster found predominantly among isolates of Southeast Asian origin . Some of the Indian STs also had overlapping STs with USA; these included ST1427 , ST550 and ST960 . B . pseudomallei has never been found in the US environment , thus all of these USA patient melioidosis infections would have been acquired overseas . We also wanted to find out whether predominant Indian STs had SLVs with STs of other countries . SLV of ST1368 , ST1373 ( ST293 ) , ST1478 ( ST1147 , ST1137 ) , ST1507 ( ST1138 , ST1152 ) and ST1512 ( ST1152 ) were reported from Sri Lanka . A few Indian STs had SLVs with STs from Thailand ( ST42-ST405 , ST501 , ST371 ) , Bangladesh ( ST42-ST43 ) and Vietnam ( ST1373- ST1568 ) . Only a few predominant STs of other countries had SLVs with Indian STs which included STs from Sri Lanka ( ST1132-ST1514 , ST1517 , ST1630 ) , Thailand ( ST371-ST1514 ) and Vietnam ( ST41- ST1051 ) . The topology and grouping of all STs from India was displayed with selected STs from Asia and Australia and results are shown in Fig 4 . This type of cluster analysis presents several advantages , such as the ease of interpretation and the creation of an hierarchical grouping of the isolates that can provide a global overview of the relatedness of the isolates under study and how the defined clusters are connected to each other . It was found that majority of the Indian isolates grouped in ‘Group 1’along with STs from Sri Lanka , Vietnam , Bangladesh , Cambodia , Malaysia , Thailand , China and Laos . ‘Group 4’ was another group in which STs from India and other Asian countries clustered together . Both of these groups had one ST each from Australia . ‘Groups 2 , 6 & 7’ which were relatively smaller groups , had STs only from India and Australia . ‘Groups 8 , 9 & 10’ had STs only from India , which probably represent some distantly related STs from main Indian cluster . Attempts were also made to find association between the STs and the clinical conditions . In this study , ST1368 was found to be predominantly associated with localized infections ( 8 of 27 ) , however this association was not statistically significant ( P = 0 . 913 ) . We also did not observe any association of STs with geographical location and year of isolation .
Overall , results of this study suggest that Indian B . pseudomallei isolates are closely related with lesser heterogeneity among them and belong to single group . However , it appears that there are a few outliers which are distantly related to the majority of Indian STs . The results further suggest that Indian isolates are closely related to B . pseudomallei isolates from Southeast Asia , particularly from South Asia , and there seems to be migration of bacteria between India and other Southeast Asian countries . However , due to low resolution of MLST , future studies should focus on WGS based SNP typing to find out ( or rule out ) the clustering of Indian B . pseudomallei with Asian or Australian isolates .
|
Burkholderia pseudomallei , a gram negative bacterium , is the causative agent of melioidosis . B . pseudomallei is a soil saprophyte and causes infections in humans by percutaneous inoculation , inhalation or ingestion . Melioidosis is a life threatening disease , which requires prolonged antibiotic treatment and is classically characterized by pneumonia , septicemia and multiple abscesses . Melioidosis is widely prevalent in Southeast Asia and northern Australia . Of late it has been reported from tropical , subtropical and temperate regions . The predicted annual global burden of melioidosis is 165 , 000 cases . B . pseudomallei has been classified as a Category B threat agent by US Center for Disease Control . Melioidosis is an emerging disease in India that affects many regions . Over the past few years , there has been an increase in number of melioidosis cases , mainly from southwestern costal part of India . This study provides new insights into molecular epidemiology of melioidosis in India . By use of multi locus sequence typing ( MLST ) , we show that Indian isolates are closely related and cluster in one major group suggesting clonality of the isolates . We further show that Indian isolates are more closely related to isolates from Asian countries particularly from South Asia .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"biogeography",
"taxonomy",
"medicine",
"and",
"health",
"sciences",
"ecology",
"and",
"environmental",
"sciences",
"melioidosis",
"population",
"genetics",
"geographical",
"locations",
"india",
"australia",
"bacterial",
"diseases",
"phylogenetics",
"data",
"management",
"population",
"biology",
"research",
"and",
"analysis",
"methods",
"sequence",
"analysis",
"infectious",
"diseases",
"geography",
"computer",
"and",
"information",
"sciences",
"bioinformatics",
"biological",
"databases",
"phylogeography",
"evolutionary",
"systematics",
"people",
"and",
"places",
"sequence",
"databases",
"asia",
"oceania",
"database",
"and",
"informatics",
"methods",
"earth",
"sciences",
"genetics",
"biology",
"and",
"life",
"sciences",
"evolutionary",
"biology"
] |
2018
|
Molecular analysis of clinical Burkholderia pseudomallei isolates from southwestern coastal region of India, using multi-locus sequence typing
|
Since the two eutherian sex chromosomes diverged from an ancestral autosomal pair , the X has remained relatively gene-rich , while the Y has lost most of its genes through the accumulation of deleterious mutations in nonrecombining regions . Presently , it is unclear what is distinctive about genes that remain on the Y chromosome , when the sex chromosomes acquired their unique evolutionary rates , and whether X-Y gene divergence paralleled that of paralogs located on autosomes . To tackle these questions , here we juxtaposed the evolution of X and Y homologous genes ( gametologs ) in eutherian mammals with their autosomal orthologs in marsupial and monotreme mammals . We discovered that genes on the X and Y acquired distinct evolutionary rates immediately following the suppression of recombination between the two sex chromosomes . The Y-linked genes evolved at higher rates , while the X-linked genes maintained the lower evolutionary rates of the ancestral autosomal genes . These distinct rates have been maintained throughout the evolution of X and Y . Specifically , in humans , most X gametologs and , curiously , also most Y gametologs evolved under stronger purifying selection than similarly aged autosomal paralogs . Finally , after evaluating the current experimental data from the literature , we concluded that unique mRNA/protein expression patterns and functions acquired by Y ( versus X ) gametologs likely contributed to their retention . Our results also suggest that either the boundary between sex chromosome strata 3 and 4 should be shifted or that stratum 3 should be divided into two strata .
Therian sex chromosomes , X and Y , evolved from a pair of homologous autosomes and thus originally harbored an identical set of genes [1]–[3] . Driven by a male-determining locus ( SRY ) , the stepwise suppression of recombination between the Y and the X led to evolutionary strata corresponding to individual suppression events [1] . Suppression of recombination between the Y and the X also resulted in their current dramatically different gene numbers [2] , ∼1 , 100 and <200 genes on the human X and Y , respectively [4] , [5] . While many X-linked genes have been preserved , the majority of Y-linked genes have been pseudogenized or deleted . Purifying selection is predicted to be inefficient in nonrecombining regions of the Y , causing an accumulation of deleterious mutations; eventually , genes are expected to be lost by means of Muller's ratchet , background selection , the Hill-Robertson effect , and/or genetic hitchhiking of beneficial mutations [6] , [7] . The already gene-poor mammalian Y continues to deteriorate [8] , and it has been proposed that within a few million years the human Y will lose all of its genes , with major consequences for mankind [2] , [9] . The human Y has retained a meager 16 functional single-copy protein-coding genes described as X-degenerate [10] , i . e . possessing divergent X chromosome gametologs ( gametologs are X-Y homologs [11] ) . Therefore , these genes represent relics of ancient autosomal genes ( the remaining functional Y-linked genes are classified as pseudoautosomal , ampliconic , and recently X-transposed [5] ) . What evolutionary forces have been maintaining these X-degenerate genes on the Y ? The first possibility is that the surviving genes might carry out essential functions where purifying selection maintains the amino acid sequence of the encoded protein leading to a low rate ratio of nonsynonymous to synonymous substitutions ( KA/KS ) . However , decreased efficacy of such selection on the Y would elevate KA/KS for Y vs . X gametologs [8] . The second possibility is that recombination suppression between the X and the Y can be viewed , effectively , as a duplication event . There are several proposed scenarios for how paralogs diverge from one another , including asymmetric evolution , where one copy is presumed to maintain the ancestral function , and thus experiences stronger purifying selection , while the other copy can undergo neofunctionalization or pseudogenization [12] and thus might experience positive selection or evolve neutrally . If this scenario holds true with respect to X and Y divergence , we expect that X gametologs will maintain the ancestral somatic functions necessary to both males and females ( because the X is present in both sexes ) , and will evolve under purifying selection . Purifying selection might be strong on the X because it is hemizygous in males and thus recessive alleles are readily available for such selection to operate there . Y-linked genes , present only in males may undergo neofunctionalization , or , as has often been observed , may undergo pseudogenization [4] , [5] , [10] . Purifying selection is expected to be weak for genes on the Y because of the lack of recombination there ( see above ) . Thus , similar to paralogs , divergence in function and expression between Y- and X-gametologs might actually contribute to the survival , in addition to the accelerated evolution [13] , of Y chromosome genes . Previous studies have observed elevated evolutionary rates for Y- versus X-linked genes . For instance , evolutionary rates were found to be higher for human and mouse Y chromosome genes compared with their gametologs on the X [13] . However , without available outgroup sequences , the incipient stages of X- and Y-linked gene evolution remained ambiguous , i . e . , the ancestral sex chromosome branch could not be broken into X- and Y-specific segments . In a different study , not only was purifying selection shown to be less potent in exons of three primate Y than X chromosome genes , but positive selection was also evident at several sites of Y chromosome exons [8] . Nevertheless , as both sex chromosomes carry genes with a nonrandom assortment of functions ( e . g . , genes involved in spermatogenesis are enriched on the Y [14] , whereas genes important for reproduction and brain function are overrepresented on the X [2] ) , contrasting only the X- and Y-linked genes might not represent an ideal way to study the evolution of either gene group . When feasible , a direct comparison of sex chromosome genes with homologous autosomal genes is therefore warranted . Tied to the understanding of sex chromosome evolution are hypotheses of how X and Y diverged from each other forming different evolutionary strata . Each stratum corresponds to a distinct recombination suppression event , thus , gametologs belonging to the same stratum have similar divergence [1] . In eutherian mammals , five strata of increasing age are observed linearly along the X chromosome , with the youngest near its proximal end and the oldest near its distal end , suggesting that suppression of recombination occurred in a stepwise manner between X and Y [1] , [4] . The arrangement of homologous sequences on the Y chromosome has been scrambled , supporting the hypothesis about the role of inversions in Y chromosome evolution [1] , [4] . While some X-degenerate Y chromosome genes were retained from the original autosomal pair , others were added later . After eutherian-marsupial divergence ( ∼166 MYA [15] ) , the eutherian sex chromosomes acquired the X-/Y-added region ( XAR/YAR ) , through a translocation from an autosome [16] . This segment remains autosomal in marsupials and monotremes [16] , [17] and provides a direct comparison of homologous genes between autosomes and sex chromosomes . Such a comparison allows us to infer the eutherian proto-sex chromosome branch and separate the ancestral sex chromosome branch into X- and Y-specific portions , i . e . to investigate emergent eutherian sex chromosome evolution . In eutherian mammals , the XAR/YAR continued to recombine between X and Y until the formation of strata 3 and 4 , app roximately 80–130 MYA and 30–50 MYA , respectively [1] . Primates and rodents diverged ∼85–90 MYA [18] , and thus genes belonging to stratum 3 putatively began evolving as X- and Y-specific in the ancestor of eutherian mammals . It is expected that stratum 4 genes only evolved as X- and Y-specific along the primate lineage . Only 12 human gametologous pairs with functional Y homologs are left in the human XAR/YAR [1] , [4]: TMSB4X/Y , CX/YORF15A , CX/YORF15B , EIF1AX/Y , ZFX/Y , USP9X/Y , DDX3X/Y , and UTX/Y are classified in stratum 3 [1] , [4]; but there has been some debate whether stratum 4 contains PRKX/Y , NLGN4X/Y , TBL1X/Y , and AMELX/Y ( classified based on sequence divergence [1] ) or whether TBL1X/Y and AMELXY/Y belong , instead , to stratum 3 ( based on analysis of parsimonious inversions [4] ) . Here , in our attempt to analyze the early stages of sex chromosome evolution , as well as to address what evolutionary forces lead to preservation of functional Y chromosomal gametologs , we analyzed 12 XAR/YAR gametologous pairs in eutherians along with their autosomal orthologs in opossum and platypus . A direct comparison of homologs decreased biases due to sequence composition , gene size , and ancestral functional constraints possible in studies juxtaposing Y- and X-linked genes against nonhomologous autosomal genes ( e . g . , [19] ) . Specifically , we tested the following hypotheses: 1 ) whether X and Y evolved unique evolutionary rates immediately after the suppression of recombination between them; 2 ) whether the evolutionary rates along both the X and Y branches have been constant throughout their evolutionary histories , and , 3 ) whether gametolog evolution parallels paralog evolution in terms of rates and functional constraints . Additionally , by utilizing whole-genome transcriptome and other published experimental data , we examined whether the expression and functional divergence of Y from X gametologs correlated with their evolution and potentially contributed to their survival on the sex chromosomes . Because of the use of opossum and platypus sequences , for the first time we are able to get a glimpse of how the ancestral eutherian sex chromosomes evolved .
To test the hypotheses stated above , we studied the evolution of all 12 available XAR/YAR human functional gametologs [4]: PRKX/Y , NLGN4X/Y , TBL1X/Y , AMELX/Y , TMSB4X/Y , CX/YORF15A , CX/YORF15B , EIF1AX/Y , ZFX/Y , USP9X/Y , DDX3X/Y , and UTX/Y , here listed starting from the Xpter ( Figure 1; the Y-linked gametolog of CXorf15 in human and chimpanzee has been split into two genes , CYorf15A and CYorf15B [10] , which we investigate separately ) . We included sequences from eight eutherian mammals ( human , chimpanzee , rhesus , horse , cow , dog , mouse and rat ) that had sufficient sequence coverage for robust analysis of all of the genes in the XAR ( Figure 2 , Figure 3 , and Materials and Methods ) as well as human , chimpanzee and ( when available ) mouse YAR gene sequences . To isolate chromosome-specific effects and to delineate the ancestral and proto-sex chromosomes branches , we included the orthologous autosomal gene sequences from opossum and platypus . In opossum , the order of genes found in the XAR/YAR is the same as in eutherians , but the sequences are split between chromosomes 4 and 7 [20] . The platypus genome is not yet assembled , however , the presence of the orthologous genes on a single chicken chromosome ( chromosome 1 ) [4] , in the same order , suggests that the original translocation likely occurred in one event . The phylogenetic analysis of the coding region within each homologous XAR/YAR gene group usually resulted in one of two separate tree topologies . For DDX3X/Y , USP9X/Y , and UTX/Y , we observed the pre-radiation tree topology ( Figure 1 , Figure 2 , Figure S1 ) , in which X- and Y-linked genes formed two distinct clades , and thus these gametologs diverged from one another in the common ancestor of boreoeutherian mammals [21] , forming stratum 3 , believed to be shared among all eutherian mammals [1] . For PRKX/Y , NLGN4X/Y , TBL1X/Y , AMELX/Y , and TMSB4X/Y , we observed the post-radiation tree topology ( Figure 1 , Figure 3 , Figure S1 ) , in which primate gametologs clustered together , and therefore recombination suppression between them followed the boreoeutherian radiation and presumably occurred along the primate lineage , forming stratum 4 . For genes with the post-radiation topology , consistent with previous experimental assays [22]–[24] , we did not identify the homologous mouse Y genes , suggesting that they have been deleted , pseudogenized beyond the recognition of the alignment algorithms utilized , or are yet unsequenced ( Materials and Methods ) . For each gene with either the pre- or post-radiation topology , the observed topology was significantly different from the alternative topology ( Table S1 ) . Genes for which the topology could not be confidently determined , CX/Yorf15A , CX/Yorf15B , EIF1AX/Y and ZFX/Y ( Figure S1 ) , were excluded from the concatenated analysis ( Table S1 ) , along with NLGN4X/Y ( Figure S1 ) , because its murid X orthologs could not be identified reliably [25] . To test for gene conversion , we conducted a phylogenetic analysis of each exon individually . Exons where the X and Y sequence from the same species formed a unique clade have putatively undergone gene conversion and were excluded from further analysis ( Table S2 ) . In most cases though , the phylogenetic trees produced for each exon were identical to the topology of the parent gene . When exons following the post- and pre-radiation topology were mapped to the X chromosome , they grouped closest and furthest from the Xpter , respectively ( Figure 1 ) in a significantly non-random distribution ( P<2 . 2×10−16; Wilcoxon rank-sum test ) . Although gene conversion was detected for isolated exons ( Table S2 ) , the observed distribution is more parsimoniously explained by two distinct evolutionary strata . Thus , either the boundary separating strata 3 and 4 , is closer to the position suggested in [1] , i . e . between TMSB4X and AMELX , or it is located between TBL1X and NLGN4X , as proposed in [4] , but stratum 3 should be split into two sub-strata with a second boundary somewhere between USP9X and TMSB4X ( Figure 4 ) . Homologous marsupial and monotreme sequences have allowed us to expand upon previous efforts investigating sex chromosome evolution [13] . In particular , for the pre-radiation topology , we were able to separate the ancestral sex chromosome branch ( preceding the boreoeutherian divergence ) into X- and Y-specific portions ( labeled Ancestral X and Ancestral Y , respectively , Figure 2A ) and to delineate the eutherian proto-sex chromosome branch ( labeled Proto-Sex , Figure 2A ) , preceding the Y chromosome inversion that led to formation of stratum 3 . Similarly , for primates in the post-radiation topology , we were able to investigate the evolution of X- and Y-linked sequences before ( identified by the Proto-SexPrimate branch ) and after the recombination suppression event that led to the formation of stratum 4 ( indicated on the AncestralPrimateX and AncestralPrimateY branches ) . To study differences in evolutionary rates of X , Y , and autosomal genes , we concatenated the coding regions of genes following the pre-radiation ( PRKX/Y , TBL1X/Y , AMELX/Y and TMSB4X/Y; a total of 2700 bp ) and post-radiation ( USP9X/Y , DDX3X/Y and UTX/Y; a total of 6108 bp ) topology separately ( Materials and Methods , Table S1; bootstrap values shown in Figure S2 ) , to reduce the confounding influences of comparing genes from potentially different strata . Further , we masked out exons from the exon-by-exon analysis described above that ( 1 ) did not conform to the topology characteristic for the majority of the exons of the gene ( these are likely gene conversion events ) , ( 2 ) produced an ambiguous tree topology , or ( 3 ) lacked sufficient data ( see Materials and Methods ) . First , we investigated how synonymous rates differ among the two sex chromosomes and the homologous autosomal sequence . Synonymous rates for genes with the pre-radiation topology ( Figure 2 ) were significantly higher for Y than X gametologs ( between the sum of branches to the common ancestor between human X and Y , P = 1 . 01×10−3; chimpanzee X and Y , P = 1 . 31×10−3; and mouse X and Y , P = 4 . 40×10−6 ) , reflecting male mutation bias [26] . Genes with this topology had significantly higher synonymous rates for mouse than human ( compared between the sum of branches to the common ancestor , P = 2 . 43×10−10 for mouse X - human X , P = 2 . 54×10−10 for mouse Y - human Y ) , in agreement with previous studies ( e . g . , [27] ) . Synonymous rates for genes with the post-radiation topology ( Figure 2B ) were ( not significantly ) higher between mouse X vs . human X , and similar between human Y and X sums of branches ( data not shown ) . Synonymous rates were lower in the opossum lineage ( 0 . 282 and 0 . 530 for the pre- and post-radiation topology , respectively ) than in even the shortest eutherian lineages ( 0 . 469 and 1 . 227; calculated as the sum of eutherian-specific branches leading to Human X for the pre-radiation topology and Horse X for the post-radiation topology , respectively ) . This can be explained by the lower GC content and reduced recombination rates of opossum vs . eutherian chromosomes [20] , [28] . The differences in opossum rates between the pre- and post-radiation topologies might either result from interchromosomal rate variation [29] , since most of the genes following the former and latter topologies are found on opossum chromosomes 4 and 7 , respectively , or be driven by local genomic factors [30] . Second , we studied variation in the KA/KS ratio among branches . For every comparison in both topologies , the KA/KS ratio was significantly higher for the Y than the X branch ( Figure 2B , Figure 3B ) . Our data set allowed us to investigate when these differences between X and Y chromosome evolution emerged , i . e . whether the elevated evolutionary rates observed on the Y versus the X occurred immediately after recombination suppression or just recently , after million years of suppressed recombination . For both topologies , immediately after recombination suppression , the Y chromosome ( Ancestral Y and Ancestralprimate Y branches for pre- and post-radiation , respectively ) acquired significantly higher KA/KS ratios as compared with the Proto-Sex branch ( Figure 2B , Figure 3B ) . This increase could be due to relaxed purifying selection on the Y in the absence of recombination and/or due to positive selection of Y-linked genes that acquired new functions [8] . Positive selection was not detected on any branches or sites in these seven genes ( see Materials and Methods ) and , consequently , KA/KS ratios were interpreted as varying degrees of purifying selection , reflecting the level of functional constraints . Thus , purifying selection was weaker on the Ancestral Y branch than on the Proto-Sex branch ( or the Ancestral X branch ) for trees with both topologies ( Figure 2B , Figure 3B ) . In contrast , the intensity of purifying selection did not differ significantly between the Proto-Sex and Ancestral X branches for gametologs following the pre-radiation topology , implying that these X-linked genes have retained the level of functional constraints of their autosomal ancestors ( Figure 2B ) . Interestingly , X and Y lineages of the pre-radiation topology maintained relatively constant KA/KS ratios since the suppression of recombination between them ( Figure 2B; recent gametolog separation in the post-radiation topology prevented us from conducting a similar analysis there ) . Indeed , the KA/KS ratio was not significantly different between the Ancestral X branch and either the ape or the mouse X branches , again suggesting preservation of functional constraints of X gametologs . Similarly , the KA/KS ratio did not differ significantly between the Ancestral Y branch and either the ape or the mouse Y branches , indicating that Y rapidly settled on its own equilibrium evolutionary rate [13] . We next asked whether divergence between gametologs mimicked the divergence between paralogs . To answer this question , we compared the evolution of human gametologs ( here all 12 gametologous pairs were considered ) against pairs of similarly aged human autosomal paralogs . Using the synonymous rate ( KS ) as an estimate of evolutionary age , for each gametolog , we compiled a set of similarly aged autosomal trios composed of a pair of human paralogs , duplicated after human-opossum divergence , aligned with the orthologous autosomal sequence in opossum ( a total of 470 trios; Materials and Methods ) . The distribution of pairwise KA/KS ratios was significantly different between gametologs and similarly aged autosomal paralogs ( P = 0 . 0001 , Wilcoxon test ) . The impact of positive selection was minor ( only 13 sites of CYorf15B and 5 sites of ZFY exhibited signatures of positive selection; Materials and Methods ) , and thus we again interpreted the KA/KS ratio as the strength of purifying selection . Pairwise KA/KS ratios were lower for nine out of 12 gametologs than for autosomal paralogs ( Table 1 ) , suggesting stronger purifying selection acting on gametologs . The higher pairwise KA/KS ratios observed for AMELX/Y , CX/Yorf15A and CX/Yorf15B might reflect the initial stages of Y gametolog pseudogenization [10] , [31] or positive selection acting on some CYorf15B sites . Stronger purifying selection between most gametologs than paralogs contradicts the hypothesis of sexual selection driving more rapid divergence between gametologs than autosomal paralogs [32] . Using opossum sequence to polarize substitutions , we determined that most gametologs displayed asymmetric functional constraints , meaning that the KA/KS ratios differed between the two gametologs , often by an order of magnitude , although not always significantly so , and all gametologs had a lower KA/KS ratio for the X than Y copy ( Table 1 ) . Thus , gametologs likely followed an evolutionary scenario proposed for paralogs , in which purifying selection was stronger for one than the other paralogous copy [12] . And , consistent with our expectation ( see Introduction ) , purifying selection was always stronger for the X than the Y copy . We next asked whether X and Y gametologs evolved at rates similar to these for slowly and quickly evolving paralogous copies , respectively ( slowly and quickly evolving paralogous copies were determined using opossum as an outgroup ) . In contrast to expectations of inefficient purifying selection on the Y [6] , all but three Y gametologs had lower KA/KS ratios and thus may have evolved under stronger purifying selection than the quickly evolving copies of paralogs ( Table 1 ) . This might represent a mechanism of Y gametolog preservation; either a gene must be maintained through purifying selection , or , as evident again for AMELY , CYorf15A , and CYorf15B , that had higher KA/KS ratios than the similarly aged quickly evolving paralogs , they may become prey to pseudogenization . Relatively strong purifying selection observed for Y gametologs might also , in part , be explained by genetic hitchhiking due to selection acting on other Y chromosome genes ( e . g . , ampliconic genes ) ; genetic hitchhiking is expected to be particularly potent on the Y because it does not undergo recombination outside of the pseudoautosomal regions . Similar to Y gametologs , all but two X gametologs had lower KA/KS ratios than the slowly evolving paralogous copies ( Table 1 ) . Intense purifying selection acting on X gametologs can be explained by the fact that X is hemizygous in males ( thus recessive alleles are instantly open to selection ) and by the preservation of somatic functions important for both sexes . To test a hypothesis that the expression and functional divergence of Y gametologs from their X counterparts potentially contributed to the survival of the former on the sex chromosomes , we compiled and analyzed whole-genome transcriptome and other published experimental data . Expression divergence between X and Y gametologs was inferred from human and mouse transcriptome microarray data produced by Su and colleagues [33] . In humans we studied 11 tissue samples collected from males in that study . In over three quarters of gametolog-tissue combinations , either the X and Y gametologs in a pair were expressed at unequal levels ( at least 25% different ) or one copy was completely turned off ( Figure 5 ) . Thus , gametologs acquired expression patterns distinct from one another . We found no significant difference in the expression divergence between human gametologous pairs and similarly aged human autosomal paralogs ( Table S3 ) , implying that the expression patterns of gametologous pairs diverge from one another at a similar rate as those of paralogous pairs . Next , using the proportion of tissues in which both the X and Y gametolog are similarly expressed ( white boxes with a number in Figure 5 ) among all tissues with detected expression as a measure of gametolog expression similarity , we determined that there is no significant difference in expression patterns between gametologs following the pre- versus post-radiation topologies ( Wilcoxon rank sum test , P = 0 . 3018 ) , and there is no significant correlation ( P = 0 . 622 ) between gametolog expression similarity and the distance from the Xpter . The non-significance may be due to both the limited number of genes , as well as the limited number of tissues available for the analysis . However , given that expression patterns diverge very rapidly , frequently outpacing sequence divergence [34] , [35] , the genes considered here may already have diverged past any threshold of observing certain correlations . Mouse samples used in the study of Su and colleagues [33] , were all pooled from tissues collected from both males and females , thus it was impossible to distinguish levels of X and Y expression unambiguously . Still , two mouse Y-linked genes included in microarrays analyzed by Su and colleagues [33] - Ddx3y and Usp9y - had undetectable expression across all 61 tissues analyzed , while their gametologs , Ddx3x and Usp9x were expressed in all and one of the tissues examined , respectively ( the other gametologs present on the array studied , Utx/y and Zfx/y , were not expressed [33] ) . Therefore , we do observe unique expression patterns between at least some mouse and most human X and Y gametologs . These differences in expression might have contributed to the retention of Y gametologs . Additionally , mining and compiling nearly 15 years of experimental data gathered from the literature allowed us to conclude that the majority of human X and Y gametologs acquired unique protein expression patterns and/or functions ( Table S4 ) , sometimes not detectable from studies of gene expression alone . For instance , in the case of human DDX3X/Y , although both gametologs are widely transcribed , only the X-linked copy , DDX3X , is also widely translated , while DDX3Y is translated exclusively in the male germ line [36] . This is accompanied by distinct temporal protein expression patterns , at least in spermatogenesis , where the two protein products are present at different stages [36] . In another example , the TBL1X/Y gametologs differ in both mRNA expression and protein function . TBL1X mRNA is ubiquitously expressed [37] , while TBL1Y mRNA expression is limited to only a few tissues [38] . The dissimilarity is also evident in function as the TBL1X protein represses transcription [39] , while the TBL1Y protein has no such activity [38] . As a final example , AMELY deletions cause no detectable phenotypic changes [40] , but deletion of AMELX causes amelogenesis imperfecta [31] , [41] . Such differences in protein expression and function between gametologs might have also contributed to retention of X degenerate Y chromosome genes . To the best of our knowledge , we present the first analysis of the ancestral proto-sex evolutionary rates in eutherian mammals . We observed that immediately following the suppression of recombination between X and Y , likely due to their importance in both sexes , X gametologs largely maintained the ancestral autosomal sequence and functional constraints . In contrast , Y gametologs , as predicted due to absence of recombination [6] , evolved under weaker purifying selection than X gametologs . Further , these different rates have been roughly maintained through evolutionary time by each of the sex chromosomes . Both X and Y gametologs , on average , acquired functional constraints stronger than quickly and slowly evolving copies of autosomal paralogs , respectively . This might have contributed to the survival of these gametologs . We also observe that the divergence between of X and Y gametolog sequences after recombination suppression , in some ways mimics that of paralogous genes , were one copy maintains a lower , more conservative , rate of evolution while the other is allowed a higher substitution rate , and may eventually evolve a new function or become prey to pseudogenization . Our analysis of the sequence evolution combined with experimental observations suggests that to withstand the evolutionary vulnerability on the Y chromosome , most Y-linked genes diverged in expression and function from their X gametologs to become separately valuable . Although Y chromosome sequencing and assembly is an undeniably challenging endeavor [5] , [10] , [42] , it provides invaluable and otherwise impossible insights into mammalian evolution . Further studies investigating gametologs will critically depend on the availability of Y chromosome sequences for several mammals , in addition to human [5] and chimpanzee [42] .
Eutherian X-linked and corresponding autosomal nucleotide sequences for opossum and platypus were extracted from the 28-way vertebrate alignments [43] available through Galaxy [44] , using the human X homolog as a reference . We initially considered X-linked sequences from all 18 eutherian species included in the 28-way genomic alignments [43] , but retained only eight due to limited coverage in the other species ( Figure 2 and Figure 3 ) . Only complete human and chimpanzee Y [5] , [10] , and partial mouse Y chromosome sequences are available . Human , chimpanzee and mouse Y-linked sequences were downloaded from Genbank ( see Table S5 ) . Of the 12 gametologs , we identified only four ( Zfy , Usp9y , Ddx3y , and Uty ) annotated on the mouse Y chromosome in Genbank . Since the mouse Y chromosome has yet to be completely sequenced and assembled , we searched the available 533 mouse Y BACs ( a total of ∼90 Mb ) for the seven missing genes . Using BlastZ [45] , we identified the four previously annotated genes ( see above ) , but were unable to locate the unannotated genes . The coding nucleotide sequences for each homologous gene group ( sex-linked gametologs and autosomal homologs ) were aligned using ClustalW [46] . The phylogenetic trees were built according to the Neighbor-Joining method [47] as implemented in PHYLIP [48] using X-linked sequences from human , chimpanzee , rhesus , mouse , rat , cow , dog , horse , Y-linked sequences from human , chimpanzee , and mouse , when available , and autosomal sequences from opossum and platypus . These species were chosen among the 18 mammals represented in [43] because for each of them at least nine of the 12 genes had greater than 50% sequence coverage . 1000 bootstrap replicates were generated first for each gene and then for each coding exon . Exons with less than 50% bootstrap support for clades with either the pre- or post-radiation topology , fewer than 24 nucleotides aligned across all species , or inconsistent with the topology of the whole gene ( a total of 92 exons ) were excluded from this portion of the analysis . In addition to Neighbor-Joining analysis , we used Maximum Likelihood and Maximum Parsimony tree building methods [48] . The three approaches led to similar results ( data not shown ) . Our results represent gene trees , not necessarily species trees ( see discussion of primate , rodent , and carnivore groupings in [49] ) , and so we advise against using these groupings to support arguments for or against contentious species groupings . The exon by exon analysis described above led us to identify known cases of gene conversion ( e . g . in ZFX/Y [50] ) . To further test for gene conversion , we aligned human X with human Y , chimp X with chimp Y and mouse X with mouse Y sequences using PipMaker [51] , a software that utilizes a local alignment algorithm to output regions of similar sequence identity . Higher identity of a particular stretch of an alignment in relation to the entire alignment can be suggestive of gene conversion [52] . New instances of gene conversion were not detected either with this method nor with GENECONV [53] . HyPhy was used to estimate the branch-specific KS and KA under the GY94_3×4 model and to test for statistical significance of differences in the synonymous rates among branches using a Likelihood Ratio Test ( LRT ) , testing the likelihood that two branches had the same vs . different KS values [54] . Tests conducted with the MG94_3×4 and MG94xHKY_3×4 models yielded similar statistically significant results . To compute the probability that the KA/KS ratio was significantly different between two branches , we used the GAbranch analysis [55] in the online version of HyPhy ( www . datamonkey . org ) , which computes the model-averaged probability that two branches have the same KA/KS ratio [56] . To determine the significance of the difference between sums of branches , we re-ran our analyses excluding the species that broke the branches we intended to compare ( e . g . , in the pre-radiation topology , we excluded rat X to be able to compare mouse X and Y branches ) . To examine a possibility of positive selection , we first used the GAbranch analysis [55] , [56] to compute the model-averaged probability that KA was significantly greater than KS along each branch . Second , we tested for significant differences between site-specific models M1 ( neutral ) and M2 ( selection ) , and between M7 ( beta ) and M8 ( beta and omega >1 ) in the codeml module of PAML [57] . Selection was not detected by these two methods . In a third test for positive selection , using the random effects likelihood ( REL ) approach [56] , [58] to identify specific sites that might have been acted on by positive selection , there was evidence for positive selection at 13 sites of CYorf15B and at 5 sites of ZFY . Using the FASTA method [59] , 6 , 536 autosomal paralogous pairs were identified among 48 , 218 protein sequences of consensus CDS , known , and novel genes in Ensembl ( release 38 of NCBI build 36 ) . Each human protein in a paralogous pair was used as a blastp query against all known opossum proteins [45] . An opossum homolog was identified if it was the highest scoring hit to both human paralogs with an e-value <1×10−10 . A pair of human paralogs together with the opossum homolog formed a trio that was retained if , after computing branch-specific KA and KS in the codeml module of PAML [57] , KS was <1 along the sum of the two human branches , to ensure that the human paralogs were duplicated after human-opossum divergence [20] . Finally , gene trios were excluded if any of the three genes were sex-linked in their respective species , or if the absolute value of the difference between the KA/KS ratios of human paralogs , Δ ( KA/KS ) , was greater than 10 . As a result , a total of 470 trios were retained . Pairwise KA and KS were estimated for each gametologous pair ( without masking any exons ) and for each paralogous pair , using the codeml module of PAML [57] . Using the opossum homolog as an outgroup to polarize the changes , we then identified the slowly and quickly evolving copies for each gametologous or paralogous gene pair as the gene having a lower or higher KA/KS ratio relative to each other , respectively . The KA/KS ratio for each X-linked gametolog was compared against the distribution of these ratios calculated for the slowly evolving paralogous gene copies , and the KA/KS ratio for each Y gametolog was compared against the distribution of these ratios calculated for the fast evolving paralogous gene copies . We computed the probability that the observed pairwise or branch-specific KA/KS ratio for each gametolog was significantly lower than these values calculated for paralogs by calculating a left-tailed empirical P value , equal to the number of paralogs having a lower ratio than a gametologous pair under consideration , divided by the total number of paralogs . Empirical distributions for the autosomal paralogs , determined individually for each gametolog , were composed of all autosomal paralogs with a KS value within ±0 . 1 of the pairwise or branch-specific KS of the gametolog ( s ) . The significance of the results did not change if we used a range of ±0 . 05 , and only changed for one pair if we used a range of ±0 . 5 . Final P values were corrected for multiple comparisons according to the Bonferroni method . The probability that the X- and Y-specific branches for each gametologous pair had significantly different KA/KS ratios was estimated using the GAbranch analysis [55] implemented in the online version of HyPhy [56] . To analyze human and mouse gametologous gene expression , we used the data from [33] . Probe sets were mapped to genes and screened for potential cross-hybridization to both gametologs in each pair following the methods described in [60] . Reliable probe sets were identified for all human and mouse gametologous pairs ( Table S6 ) . For humans , all but 13 of the 79 tissues analyzed in [33] were either female-specific or pooled between females and males . Of the remaining 13 , we used only 11 that were non-redundant tissues [33] . For a gene to be considered expressed in a particular tissue , we required the average difference ( AD ) to be greater than 200 in that tissue , following a method described by Su and colleagues [33] . If both genes in a pair were expressed , we calculated the fold change , Fk , computed as the log of the ratio of X and Y expression , log2 ( X/Y ) . If the Y-linked gene is more highly expressed than its X gametolog , Fk will be negative , whereas if the X gametolog is more highly expressed , Fk will be positive . For −0 . 25<Fk<0 . 25 , we considered X and Y to be similarly expressed . The results did not change qualitatively if we used a larger range of −0 . 5<Fk<0 . 5 . Distributions of autosomal paralogs were taken from the pairwise analysis , described above ( so that we compare the expression divergence of each gametologous pair with similarly aged autosomal paralogs , as measured by KS ) . Reliable probe sets and expression values were identified following the methods described above . Empirical P values were computed as explained for paralogs . Gametolog functional and protein expression data were retrieved from the iHOP ( Information Hyperlinked Over Proteins ) database ( http://www . ihop-net . org/UniPub/iHOP/ ) , the OMIM ( Online Mendelian Inheritance of Man ) database ( http://www . ncbi . nlm . nih . gov/omim/ ) , and PubMed ( http://www . ncbi . nlm . nih . gov/PubMed/ ) .
|
Using recently available marsupial and monotreme genomes , we investigated nascent sex chromosome evolution in mammals . We show that , in eutherian mammals , X and Y genes acquired distinct evolutionary rates and functional constraints immediately after recombination suppression; X-linked genes maintained lower , ancestral ( autosomal ) , rates , whereas the evolutionary rates of Y-linked genes increased . Most X and , unexpectedly , Y genes evolved under stronger purifying selection than similarly aged autosomal paralogs . However , we also observed that the divergence of gametologs and paralogs shared similar features . In addition , many Y-linked copies evolved unique functions and expression patterns compared to their counterparts on the X chromosome . Therefore , our results suggest that to be retained on the Y chromosome , genes need to acquire separately valuable expression and/or functions to be safeguarded by purifying selection .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Materials",
"and",
"Methods"
] |
[
"evolutionary",
"biology/bioinformatics",
"genetics",
"and",
"genomics/bioinformatics",
"computational",
"biology/genomics"
] |
2009
|
Evolution and Survival on Eutherian Sex Chromosomes
|
Although many human diseases have a genetic component involving many loci , the majority of studies are statistically underpowered to isolate the many contributing variants , raising the question of the existence of alternate processes to identify disease mutations . To address this question , we collect ancestral transcription factor binding sites disrupted by an individual’s variants and then look for their most significant congregation next to a group of functionally related genes . Strikingly , when the method is applied to five different full human genomes , the top enriched function for each is invariably reflective of their very different medical histories . For example , our method implicates “abnormal cardiac output” for a patient with a longstanding family history of heart disease , “decreased circulating sodium level” for an individual with hypertension , and other biologically appealing links for medical histories spanning narcolepsy to axonal neuropathy . Our results suggest that erosion of gene regulation by mutation load significantly contributes to observed heritable phenotypes that manifest in the medical history . The test we developed exposes a hitherto hidden layer of personal variants that promise to shed new light on human disease penetrance , expressivity and the sensitivity with which we can detect them .
The advent of high-throughput genotyping spurred the rise of genome-wide association studies ( GWAS ) aimed at identifying the basis of genetic diseases . GWAS variants , over 90% of which have been found to localize outside of protein-coding sequences [1] , and the growing body of non-coding genome annotations have helped improve our understanding of the genetic basis of diseases by shifting the focus from protein coding and copy number variations [2–4] , to the non-coding genome . Though GWAS have been instrumental in suggesting a gene regulatory component to human disease susceptibility [5 , 6] , they have been plagued by the “missing heritability problem” , which observes that loci detected by GWAS in general only explain a small fraction of the genetic variance responsible for phenotype [3 , 7] . Suggested models of genetic variance responsible for the “missing heritability problem” include “the infinitesimal model”–a large number of small effect common variants and “the rare allele model”–a large number of large-effect rare variants [7] . In the case of the infinitesimal model , the missing heritability can be explained due to additive or epistatic interactions between variants rather than independent polymorphisms [8] . But , selecting and evaluating all sets of variants results in a combinatorial explosion of sets that we are currently statistically underpowered to evaluate . In this work , we will show how to not only successfully avoid the combinatorial explosion , but also simultaneously address the crucial role of additive and epistatic noncoding variation in human disease . Specifically , we develop a novel statistical framework to identify putatively deleterious noncoding variation in personal genomes that en masse , confers disease risk by dysregulating key genes involved in a common biological process . A central role of the non-coding genome lies in cis-regulation of gene expression . GREAT ( Genomic Regions Enrichment of Annotations Tool ) is a tool commonly used to address the functional enrichment of a set of cis-regulatory genomic regions [9] . GREAT tests whether an arbitrary set of genomic regions , most of which are thought to regulate the expression of nearby genes , congregate next to genes of particular functions or pathways . GREAT assigns different genes variable length gene regulatory domains , accounts for distal regulatory elements and rewards observing multiple elements next to the same gene–reflecting observed properties of vertebrate gene regulation . GREAT has been shown to be superior to gene based tests ( following the one probe—one gene paradigm of transcript analysis ) in analyzing different types of ChIP-seq and related data [9] . Since we are interested in identifying disease-associated noncoding variation to interpret personal genomes , we asked whether disease-associated noncoding mutations would be functionally enriched for key biological pathways using GREAT . First , we subjected noncoding GWAS significant SNPs associated with several phenotypes ranging from Crohn’s disease to fasting glucose traits to GREAT ( S1 Table ) . Not all GWAS tag-SNPs are themselves causal , but because they lie in proximity to the causal mutation , we can assume that GREAT will in most cases associate the tag-SNP with the same affected gene/s it would associate the underlying causal mutation . For example , if we subject the 40 non-exonic non-linked GWAS SNPs associated with cholesterol levels to GREAT analysis , the topmost enriched term ( P = 3 x 10−5 ) in the entire GO ontology is genes involved in “abnormal circulating cholesterol level” . We show similar results across several different GWAS sets in S1 Table . In each case , we see that the non-linked tag-SNPs are most enriched next to genes of functional categories strikingly relevant to the assayed phenotype , providing in-silico assurances for the quality of the study and the validity of a GREAT analysis , but also suggesting that multiple of these mutations may accumulate in afflicted individuals . As such , we hypothesized that much more signal may hide in cis-regulatory variants beyond what GWAS may reveal . The coherence of target gene enrichment for GWAS variants suggests additive and/or epistatic effects of variations to confer phenotype . Modeling such interactions is generally limited to heuristic search of pairs due to the high computational requirement and lack of statistical power [10] . The statistical power for identifying causal variants is further weakened in non-coding regions due to most variants resulting from neutral evolution of the genome [11] . Thus , to obtain a high quality set of variants on which GREAT can be applied we require a method of obtaining a set of functionally relevant noncoding variants without enumerating all possible sets . To obtain a set of functionally relevant , putatively deleterious noncoding variants , we make use of transcription factor ( TF ) binding site prediction . Novel high throughput technologies , such as HT-SELEX and Protein Binding Microarrays , are revealing the precise DNA binding preferences of the majority of human transcription factors [12 , 13] . Using these preferences to predict TF binding in a single genome is notoriously hard . However , if one is willing to predict only a subset of binding sites , namely those conserved through evolution , one can then predict the existence of a binding site only if one sees the site in orthologous locations in a number of different mammals [14] . Such a scheme will naturally miss many evolutionarily newer binding sites , but , as we and others have shown , those conserved binding sites that we do predict are predicted with great precision and are useful for downstream analysis such as functional enrichment and protein complex prediction [14–16] . As shown in the GREAT paper , while a ChIP-seq experiment reveals that a TF binds non-specifically to many genomic locations , the strongest GREAT gene enrichment reflects the process or function the TF is regulating , highlighting the subset of binding sites involved in the regulatory process [9] . Previously , in our binding site prediction ( PRISM ) paper , we predicted the conserved subset of binding sites of a given TF motif and subjected this set , in place of ChIP-seq peaks , to GREAT analysis . In many cases , such as for transcription factors REST , GABPA , SRF , and STAT3 , such analysis revealed multiple functional contexts in which the TF was involved without requiring a cell-type matched TF ChIP-seq experiment [14] . Additionally , in our previous binding site prediction work , we intersected our conserved binding site predictions with GWAS tag SNPs . To maximize the chance a GWAS tag SNP was indeed the functional , causal mutation , we set out to search for the following: A GWAS tag SNP overlapped by a conserved binding site prediction , such that: 1 ) the two observed alleles significantly differed in the predicted TF’s ability to bind to the motif , and 2 ) the TF we predict to bind has been previously implicated in the GWAS phenotype . In our paper , we highlighted only five such predictions ( Table 1 in [14] ) . One striking example , in the context of prostate cancer , is our prediction that a GWAS risk allele at 6q22 modifies the conserved binding site of HOX13 , thus modifying the expression of the downstream RFX6 gene . Our prediction was later beautifully experimentally validated by Taipale and colleagues , illustrating the utility of PRISM predictions in assessing the impact of noncoding variation on disease [17] . As exemplified above , the confluence of binding site prediction with PRISM and functional assessment of cis-regulatory regions with GREAT suggests a potent combination to understand the role of noncoding variation in disease . Accordingly , in this study we looked at personal genomes , tallied all the locations where the individual carries a SNP that disrupts an evolutionarily conserved binding site , and asked ( using GREAT ) which biological function or process these mutations aggregate next to most . Guided by our hypothesis that a pathway with the most unexpected mutational load may contribute to a person’s medical history , we then assessed our pathway predictions for relevance to the person’s health record .
Using a large library of unique high quality binding motifs for 657 different transcription factors , covering all major human DNA binding domain families and a multiple alignment of 33 primates and mammals , we first predict cross-species conserved binding sites present in the reference human genome ( see Materials and Methods ) . We then examine the genetic variants of a human individual against the reference genome . We focus on the subset of variants ( heterozygous or homozygous ) that overlap conserved binding site predictions . From these , we pick only variants where the human reference base is identical to its chimpanzee orthologous base ( and thus most likely ancestral ) , and the individual variant base differs from both . Finally , of these we keep only the binding sites where the individual ( derived ) variant is predicted to significantly decrease binding affinity compared to the ancestral base–we call these conserved binding site eroding loci , or CoBELs ( see Fig 1 and Materials and Methods ) . We downloaded from UCSC whole genome variant files for all four individuals for whom public medical history summaries are also available: Stephen Quake [18] , and three individuals from the personal genome project ( PGP10 ) [19] . An additional file was obtained for James Lupski [20] . We then compared each separately to the reference genome to obtain 6 , 321 CoBELs for Stephen Quake , 5 , 291 for George Church , 5 , 775 for Misha Angrist , 5 , 861 for Rosalynn Gill , and 6 , 447 for James Lupski ( S4–S8 Tables ) . Because CoBELs weaken conserved ancestral binding sites , we asked whether an individual’s set is found preferentially next to genes encoding any particular function , and if so , whether this function relates to the individual’s medical history ( Fig 1C ) . GREAT , as described earlier , is an approach devised specifically to assess enriched functions within a set of genomic regions thought to regulate the adjacent genes [9] by associating with each gene in the genome a variable length regulatory domain , bracketed by its two neighboring genes . GREAT also holds a large body of knowledge about gene functions and phenotypes–here we use over 1 . 1 million such gene annotations ( see Materials and Methods ) . For a given set of CoBELs , GREAT iterates over 16 , 000 different biological functions and phenotypes , asking whether CoBELs are particularly enriched in the regulatory domains of genes of any particular function . For example , 33 genes in the human genome are annotated for “abnormal cardiac output” . Their GREAT assigned regulatory domains cover 0 . 45% of the genome . Of the 6 , 321 Quake CoBELs , 28 ( 0 . 45% ) are expected in the regulatory domains of these 33 genes by chance , but 57 CoBELs , over twice as many , are in fact observed . To determine statistical significance , GREAT computes two statistics for this enrichment , and corrects them for multiple hypothesis testing ( see Materials and Methods ) . Prominent in Stephen Quake’s medical records is a family history of arrhythmogenic right ventricular dysplasia/cardiomyopathy , including a possible case of sudden cardiac death [18] . Strikingly , when Quake’s set of CoBELs is analyzed using GREAT , the top phenotype enrichment ( using default parameter settings , optimized for inference power in the original GREAT paper [9] ) is “abnormal cardiac output” ( 57 CoBELs , false discovery rate Q = 1 . 69 x 10−4 ) . This enrichment is suggestive of susceptibility to heart diseases responsible for reduced cardiac output [21] . Meaningful associations between CoBELs and personal medical records are in fact observed for all five genomes ( Table 1 and S9–S13 Tables ) . The top enrichment for George Church , who suffers from narcolepsy , is “preganglionic parasympathetic nervous system development” ( 23 CoBELs , Q = 1 . 18 x 10−4 ) . The autonomic nervous system is strongly suspected to be involved in narcolepsy [22] . Misha Angrist , whose personal reporting indicates possible keratosis pilaris , a follicular condition manifested by the appearance of rough , slightly red , bumps on the skin , has “epithelial cell morphogenesis” as his top biological process enrichment [23] ( 60 CoBELs , Q = 1 . 38 x 10−5 ) . For Rosalynn Gill , who suffers from hypertension , the top enriched phenotype is “decreased circulating sodium level” ( 32 CoBELs , Q = 4 . 94 x 10−6 ) . Sodium intake is strongly associated with hypertension [24] . Intriguingly , the top biological process enrichment we obtain for James Lupski , whose family has a history of axonal neuropathies in the peripheral nervous system ( PNS ) [20] , is “regulation of oligodendrocyte differentiation” ( 59 CoBELs , Q = 2 . 93 x 10−5 ) . Oligodendrocytes are the neuroglia that create the myelin sheath around axons in the central nervous system ( CNS ) and maintain long-term axonal integrity [25 , 26] . While a statistically significant functional enrichment from GREAT rejects the null hypothesis of uniformly random distribution of the CoBELs in the regulatory domains of the function-associated genes , it does not check whether there is an inherit bias in the distribution of conserved binding sites ( eroded or not ) in the regulatory domains of genes involved in the enriched functions . Thus to further assess the significance of our results we replaced every CoBEL with a random binding site prediction for the same transcription factor of same affinity and similar cross-species conservation . Using 10 , 000 random control sets , the likelihood of obtaining the functions reported in Table 1 as top prediction due to bias in the distribution of binding sites in the genome is low ( Quake P = 3 x 10−4 , Church P = 5 . 7 x 10−3 , Angrist P = 4 . 8 x 10−3 , Gill P = 1 x 10−4 , Lupski P = 1 . 9 x 10−3 , and combined P = 1 . 6 x 10−15 ) . Significance remains high when we relax the requirement to recover each exact same term with matching any one of a broader group of 12–60 related functions as a top prediction ( Quake P = 1 . 1 x 10−3 , Church P = 1 . 3 x 10−2 , Angrist P = 7 . 7 x 10−3 , Gill P = 7 . 4 x 10−3 , Lupski P = 6 . 5 x 10−3 , and combined P = 5 . 2 x 10−12; see Materials and Methods ) . While phenotypic data is not available for the 1 , 000 genomes project subjects [27] , the availability of whole genome sequences for 1 , 094 individuals allows us to ask how unique are our top predictions for the five phenotyped individuals against a large background of controls . We asked whether the phenotype predictions were unique to a given personal genome by testing whether they rarely appeared in control individuals from the 1 , 000 genomes project , thereby testing the specificity of our screen . This control analysis was performed due to the inclusion of both common and rare variants in our analysis . We wanted to verify that enrichments observed in our five genomes were not dominated by common CoBELs shared with many other individuals . Thus , we computed the frequency of the observed enrichments in all control , un-phenotyped 1 , 094 genomes sequenced by the 1 , 000 genomes project [27] . We verified the CoBEL set size of the 1 , 094 genomes were comparable to those of the five analyzed genomes ( min 6 , 121; European median 6 , 385 ) , submitted the CoBELs to GREAT and noted top enrichments . Each one of our observed top enrichments for the five individuals had an occurrence rate less than 0 . 05 ( S2A Table ) and the enrichment’s p-value and fold statistics placed them as significantly removed from the 1 , 000 genomes cohort ( Fig 2 ) . Next , we performed PCA to verify that the five genomes analyzed in our study are both predominantly of the expected ( European ) ancestry , and not an outlier compared to the 1 , 000 genomes project data ( Fig 3A ) . We then recomputed the occurrence rate for the enrichments using only the 381 European genomes and only the 181 admixed genomes to correct for any population specific enrichments . Again , all the enriched terms had an occurrence rate less than 0 . 05 ( S2A Table ) . Since ontology terms in GREAT are related in a directed acyclic graph ( DAG ) structure , terms such as “abnormal cardiac output” ( the Quake genome prediction ) share similar gene sets to their umbrella term “abnormal cardiovascular output” , which a control patient from the 1 , 000 Genomes project may exhibit . To account for the case when two such related terms are predicted , we calculated the false discovery rate for a term by counting its broader group of related functions as well . Still , the occurrence rate for the findings remained less than 0 . 05 ( S2B Table ) when we repeated both the full 1 , 094 genomes , 381 European genomes and 181 admixed genomes calculations for the broader group of related functions , except for slightly higher p-values ( up to 0 . 088 ) for the more common heart and hypertension disorders . Indeed , 8% of the un-phenotyped 1 , 000 genome subjects ( who may themselves suffer or be predisposed to various complex diseases , especially the more common ones ) had a top enrichment in the broader set of terms associated with hypertension , and 5% were similarly most enriched for a heart term . Finally , we assessed the specificity of associating the CoBEL enrichments of five individuals with their medical histories ( Fig 1C and S14 Table ) . This test was performed to verify that the predicted top enrichments were not so broad that they would match different medical histories and likewise that the individuals selected did not have such a broad range of disease phenotypes as to match different possible top enrichments . We defined an association matrix linking enrichment and medical history , with the phenotypes observed in the five individuals as rows , and top enriched terms in all as columns . A cell in the matrix would be marked “true” only where the enriched term ( of any individual ) is thought to be related to the etiology of the phenotype ( of any individual; see Materials and Methods ) . One instance of this matrix was filled by a medical doctor based on their medical knowledge and training ( S15 Table ) and another instance was independently filled using a literature survey ( S16 Table ) . The objective was to compute the chance of associating a set of five individuals with random medical histories with the observed enrichments using one of the two association matrices as the “gold” association . We generated 1 , 000 sets of five individuals with random medical histories composed of similar disease profiles and assessed the likelihood of being able to associate them with enrichments ( see Materials and Methods ) . Successfully linking five random individuals with enrichments was highly significant using the association matrix generated by the medical doctor ( P = 3 . 0 x 10−3 ) and by the matrix generated by literature survey ( P = 3 . 0 x 10−2 ) suggesting our links between enrichment and medical histories are not just a function of the listed histories . The literature survey derived association matrix potentially offers a stricter null model since it includes associations that are currently research topics hinting at associations that may or may not become clinically relevant in the future . Our CoBEL predictions are distinct from known GWAS associations . The 238 variant alleles that underlie all Table 1 predictions overlap a single , phenotype irrelevant , GWAS SNP , suggesting our method as a complementary method to discovering disease loci . While GWAS aims to find loci most likely to individually distinguish disease cohorts from matched controls , our method tries to identify the sum of both common and rare loci that can contribute to disease . GWAS is underpowered to find such combinatorial interactions . Similarly , none of the CoBELs responsible for the 238 variants intersect with a HGMD [28] disease variant ( a large set of very rare , highly penetrant variants thought to individually trigger the underlying disease ) . When the overlap analysis is extend to include GWAS SNPs in possible linkage disequilibrium ( LD ) , only two possible phenotype matches arise: “cardiac hypertrophy” associated [29] SNP rs3729931 for Quake , and “multiple sclerosis” ( another demyelination disease [26] ) associated [30] SNP rs882300 for Lupski . Indeed , nearly half the total number of CoBEL variant alleles we predict ( 7 , 115 , 49% ) are unique to only one of our five individuals . Similarly , for each of the five top function predictions in Table 1 , of sixteen possible subsets ( CoBELs shared or not with each of the other four individuals ) , the biggest contribution ( 17–34% ) always comes from private sites ( S1 Fig ) . When the CoBEL frequencies are examined at the population level , Quake and Gill’s enriched CoBELs show higher population frequencies ( Fig 3B and 3E ) for their presumably more common enriched phenotypes of heart disease and hypertension . Conversely , Church , Lupski and even Angrist to a lesser extent , show more enriched CoBEL with low population frequencies ( Fig 3C , 3D and 3F ) . To examine the population frequency dependence of the CoBEL analysis , we restricted ourselves to rare CoBELs , defined as those with frequency less than or equal to 0 . 01 in the 1 , 000 genomes . None of our functional enrichments are significant for the rare CoBELs . Even when we increase the 1 , 000 genome frequency 10-fold to 0 . 1 , only Angrist’s “epithelial cell morphogenesis” enrichment is rescued , albeit with diminished enrichment statistics ( 16 CoBELs , Q = 1 . 85 x 10−2 ) compared to the full set ( 60 CoBELs , Q = 1 . 38 x 10−5 ) . This further corroborates that our enrichments are a combination of both common and rare variants .
The screen we perform is underpowered: we do not have the binding affinities of all human transcription factors or all functional ( ancestral or not ) binding sites; variant mapping may miss more complex gene regulatory mutations; and in particular our knowledge of phenotype to gene associations is far from complete . Additionally , we focus only on the top enrichment obtained rather than all enrichments to maintain the ability to test for statistical rigor of the associations . All these limitations , however , only reduce our power to detect true associations , but do not elevate the likelihood of false predictions . In contrast , by focusing on deeply conserved binding sites , we greatly increase the likelihood that their disruption carries a fitness cost . Indeed , considering that GREAT tests over 16 , 000 different biological processes or phenotypes ( from “abdominal aorta aneurysm” to “zymogen granule exocytosis” ) , the links we obtain between genomic prediction and medical phenotype seem highly significant . Our CoBEL predictions compliment known disease alleles . For example , a particular human leukocyte antigen ( HLA ) allele is found in a vast majority of narcolepsy patients who suffer from cataplexy , and is also common in narcolepsy patients who do not [31] . The affected Church genome is homozygous for a different HLA allele ( see Supplementary Methods ) . Four GWAS SNPs , all with modest effect size ( OR = 1 . 29–1 . 79 ) are currently associated with narcolepsy . Church carries two of these , but the other four unaffected genomes we analyze each carry 2–3 narcolepsy risk alleles as well , due to their common prevalence ( see S3 Table ) . The Quake genome was previously analyzed for coding and GWAS variants [18] . While no single strong mutation emerged , the sum of collected mutations was enough to assess heart disease as a relatively large risk . The evaluation process of the many personal variants however was biased towards genic variants and previously determined risk loci with a focus on explaining the family history of heart disease . The enrichment we obtain for cardiac output not only comes from novel , non-genic loci , it is also obtained in a completely agnostic fashion . Our analysis is complementary to state of the art analyses that focus on searching for the primary disease causing variant by intersecting with known ( predominantly coding ) variant databases , exploring rare or novel coding or splicing variants in known disease associated genes and prioritizing coding candidate variants using computational tools such as SIFT [32] , PolyPhen2 [33] and VAAST [34] . Few such tools exist for the non-coding genome , none of which to the best of our knowledge focuses explicitly on binding site disruption . Methods such as CADD [35] score the pathogenicity of non-coding variants , but train their model on positive sets only weakly enriched for deleterious non-coding mutations . Because the non-coding portion of the genome is so large ( 97% ) , and because most such tools do not aggregate mutations on functional or any other categories , most usage is restricted to splice variants or non-coding RNA . This is exemplified by the genome analyses performed by Lupski et al . [20] and Ashley et al . [18] , for Lupski and Quake , respectively . Both works focused primarily on coding variants in known disease associated genes . They identified non-synonymous SNPs and searched for matches in known pathogenic variant databases such as HGMD [28] and OMIM [36] . When known disease variants were not identified , the search was expanded to include rare and novel variants in genes relevant to their patient ( neuropathies in the case of Lupski et al . [20] and cardiovascular disease in the case of Ashley et al . [18] ) . Neither study pursued any potential gene regulatory mutations . In addition to the enrichment obtained by our analysis , the accumulation of binding sites in our top enrichments is also revealing: First , each target gene in Table 1 is affected , on average , by more than three CoBELs , chipping away at the gene’s presumed regulatory robustness [37] . Second , Table 1 also shows that in all five cases , CoBELs affect a majority ( 58–89% ) of all human genes annotated for said function/phenotype . Together , our observations suggest the gradual erosion of gene regulation over both ( human generation ) time and ( gene regulation ) space , ultimately manifesting as medical history . These observations corroborate a long held notion that lineage accumulation of small deleterious mutations , even when combined with different lifestyles and environments , ultimately increase the likelihood of familial disease phenotypes [38] . Depending on the selection coefficient of these deleterious mutations and their genetic background , these mutations may eventually be swept out of the population , but are currently visible due to nonrandom human mating patterns and the relatively short timescales since erosion . Our screen provides an exciting glimpse of the latent genetic load of human gene regulation contribution to personal medical histories . As our ability to characterize individual genetic load improves , so will our understanding of genome–environment interactions , and the thresholds that are crossed to trigger onset of human disease .
Our transcription factor binding motif library , represented as a position weight matrix ( PWM ) , contains 917 unique high quality monomer and dimer motifs for 657 transcription factors from the UniPROBE [39] , JASPAR [40] , and TransFac [41] databases , secondary UniPROBE motifs , motifs from published ChIP-seq datasets and from other primary literature [16] . We included both monomeric and dimeric ( where the TF complexes either with itself or with another TF ) motifs to improve our sensitivity since previous work has found that complexes tend to have modified binding affinities [16] . We downloaded variant calls mapped to the human reference assembly hg19 ( GRCh37 ) from the UCSC genome browser [42] . The tables were pgQuake for Stephen Quake , pgChurch for George Church , pgAngrist for Misha Angrist and pgGill for Rosalynn Gill . The variants for James Lupski were downloaded from dbSNP [43] and processed to remove non-single nucleotide polymorphism and those that had ambiguous mapping to the reference genome . The medical history summaries for Stephen Quake and James Lupski were obtained from Ashley et al . [18] and Lupski et al . [20] , respectively . Medical history summaries for the remaining individuals were obtained from their public profiles on the Personal Genome Project [19] website . We identified conserved binding sites using the UCSC human reference assembly hg19 ( GRCh37 ) based multiple alignment of 33 primates and mammals [42] . Binding site prediction was done by identifying conserved binding site matches using PRISM [14] . We chose the default PRISM thresholds of a minimum of five species preserving each site prediction , with the total phylogenetic ( neutral ) branch length [44] of the preserving species amounting to two substitutions per site or more . Additionally , we kept only the top 0–5 , 000 binding site predictions that had a conservation p-value less than or equal to 0 . 05 . The conservation p-value was computed by comparing the binding conservation for ( CpG preserving ) shuffled versions of the motif in similarly conserved regions of the genome . All parameter settings we used have been previously optimized in the PRISM paper for predictive power [14] , including against multiple ENCODE [6] datasets . Next , we identified all the heterozygous or homozygous variants in an individual genome where the human reference ( hg19 ) base is identical to the orthologous chimp ( panTro2 ) base , and thus most likely human ancestral . We then found all human reference genome conserved binding sites affected by our individual specific variants . Of these we kept only sites where replacing the reference human ( ancestral ) base ( s ) with the individual derived variant ( s ) lowers binding affinity by five per cent or more . Binding affinity was computed using the MATCH scoring scheme [45] . Overlapping binding sites were combined to obtain our final set of conserved binding site eroding loci ( CoBELs ) . Define set of human conserved binding sites , TFBS ← PRISM ( motif library ) For each individual with genomic variants Vi: Intersect TFBS with Vi ( using overlapSelect from UCSC genome browser ) For each TFBS tfbs in intersection: Compute tfbs MATCH score difference between ancestral and variant D If D decreases the MATCH score more than 5% add binding site to set of CoBELs Run GREAT ( set of CoBELs ) -> Output Top Enrichment The majority of the motifs used in our screen are public , and can be obtained , along with their predictions , from the PRISM website at PRISM . stanford . edu . A small fraction of motifs comes from the proprietary Transfac database . A list of these will be provided upon request . We also include all five CoBEL sets in S4–S8 Tables , which can be processed using GREAT at GREAT . stanford . edu to reproduce the results of Table 1 and S9–S13 Tables . Each set of CoBELs was submitted to GREAT ( for Genomic Regions Enrichment of Annotations Tool ) v2 . 0 . 2 [9] . As explained in the main text , GREAT searches for statistically significant genomic regions ( in this case CoBELs ) accumulation in the regulatory domains of genes that share the same annotation . For this study , we used GREAT’s default regulatory domain definition: a constitutive 5 , 000 bases upstream and 1 , 000 bases downstream of a gene’s canonical transcription start site ( TSS ) , extended up to the constitutive regulatory domain of the adjacent genes on either side , or up to one million bases . Significance was also defined using the default GREAT thresholds: 0 . 05 FDR threshold for both binomial and hypergeometric test and binomial fold greater than 2 . These parameter settings have all been optimized for inference power in the original GREAT paper [9] . We queried the GO Biological Processes [46] and MGI Phenotype [47] ontologies allowing GREAT to test for possible enrichment of any of 16 , 054 different functions , using 1 , 140 , 682 gene to function mappings . The CoBEL methodology was applied to each of the 1 , 094 genomes and the top enrichment satisfying the default GREAT filters in the GO Biological Processes and MGI Phenotype ontologies was tracked . For each of the enrichments highlighted for the five genomes analyzed in this report , the frequency of the enrichment in the full 1 , 094 genomes was computed . Additionally , the frequency of the enrichments in the 381 European ( EUR ) subset and 181 admixed ( AMR ) subset was measured since principal component analysis revealed that the five genomes analyzed in this report are closest to these two population subgroups ( Fig 3A ) . All SNPs from the NHGRI GWAS catalog [48] were downloaded from a build containing 8 , 967 records in hg19 ( GRCh37 ) co-ordinates , and intersected with the set of enriched CoBEL variant alleles from Table 1 . Quake , Angrist , Gill and Lupski had no overlaps . Church had a single , phenotype irrelevant , overlap with rs10808265 which is GWAS associated with pulmonary function decline [49] . To assess linkage disequilibrium ( LD ) between the enriched CoBEL variants and GWAS SNPs we used HapMap [50] rel27 LD data for the CEU ( Utah residents with Northern and Western European ancestry ) population . CoBEL variant alleles from Table 1 were mapped to HapMap by taking the HapMap provided hg18 ( NCBI Build 36 . 1 ) coordinates , lifting them to hg19 using the UCSC browser liftOver utility [45] and intersecting with the CoBEL variants . Nearly half ( 49% , 112/227 ) the enriched variants sites could be mapped to HapMap probes . NHGRI GWAS SNPs were mapped to HapMap SNPs using rsIDs . A GWAS SNP and a CoBEL variant were called in LD , using a maximalist approach , if either D’ > 0 . 99 or r2 ≥ 0 . 8 or LOD ( log odds ) ≥ 2 between their matching HapMap probes . All enriched CoBELs from the five individuals were overlapped with the HGMD PRO 2015 . 2 set containing 130 , 218 disease mutations using overlapSelect from UCSC genome browser . Over 90% of narcolepsy patients with cataplexy , and around 40% of narcolepsy patients without cataplexy carry human leukocyte antigen ( HLA ) type DQB1*06:02 [31] . The crystal structure of HLA-DQB1*06:02 ( PDB ID: 1UVQ ) [51] identified the representative amino acid haplotype of DQB1*0602 as F9G13L26Y30Y37A38D57 ( subscript represents amino acid number in exon 2 of HLA-DQB1 ) . Based on the variant call file , the haplotype present is George Church is different: Y9G13L26H30Y37A38D57 . When we used BLAST to search the Church version of exon 2 against the IMGT/HLA Database [52] , the allele closest to the observed haplotype was DQB1*06:03 , not found associated with narcolepsy patients [53] .
|
A central goal of personal genomics is to interpret an individual’s genome to identify variants that confer disease risk , an aim that has far-reaching implications for personalized , precision medicine . Here , we leverage next generation sequencing , health records , and functional genome annotations to develop statistical methods that predict disease risk from a single genome . Motivated by the fact that about 90% of genome-wide association study disease-associated variants lie in the non-coding genome , we identify personal variants that mutate conserved transcription factor binding sites . To identify if such non-coding personal variants collectively dysregulate a key biological process , we employ the enrichment analysis tool GREAT to identify if a person’s noncoding mutations are over-represented in the regulatory domains of genes involved in a common biological pathway . Notably , in five unrelated genomes we analyzed , the most statistically significant , seemingly dysregulated pathway is indicative of that person’s medical history , ranging from neuropathy to heart disease . Statistical analysis confirms that associations from our predicted pathway to an individual’s medical record are rigorous and significant in the context of the un-phenotyped , race-matched 1 , 000 Genomes cohort . As such , we present a novel method that leverages the contribution of multifactorial non-coding variation to predict disease risk in individual genomes .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2016
|
Erosion of Conserved Binding Sites in Personal Genomes Points to Medical Histories
|
This study describes two longitudinal serological surveys of European Bat Lyssavirus type 1 ( EBLV-1 ) antibodies in serotine bat ( Eptesicus serotinus ) maternity colonies located in the North-East of France . This species is currently considered as the main EBLV-1 reservoir . Multievent capture-recapture models were used to determine the factors influencing bat rabies transmission as this method accounts for imperfect detection and uncertainty in disease states . Considering the period of study , analyses revealed that survival and recapture probabilities were not affected by the serological status of individuals , confirming the capacity of bats to be exposed to lyssaviruses without dying . Five bats have been found with EBLV-1 RNA in the saliva at the start of the study , suggesting they were caught during virus excretion period . Among these bats , one was interestingly recaptured one year later and harbored a seropositive status . Along the survey , some others bats have been observed to both seroconvert ( i . e . move from a negative to a positive serological status ) and serorevert ( i . e . move from a positive to a negative serological status ) . Peak of seroprevalence reached 34% and 70% in site A and B respectively . On one of the 2 sites , global decrease of seroprevalence was observed all along the study period nuanced by oscillation intervals of approximately 2–3 years supporting the oscillation infection dynamics hypothesized during a previous EBLV-1 study in a Myotis myotis colony . Seroprevalence were affected by significantly higher seroprevalence in summer than in spring . The maximum time observed between successive positive serological statuses of a bat demonstrated the potential persistence of neutralizing antibodies for at least 4 years . At last , EBLV-1 serological status transitions have been shown driven by age category with higher seroreversion frequencies in adults than in juvenile . Juveniles and female adults seemed indeed acting as distinct drivers of the rabies virus dynamics , hypothesis have been addressed but their exact role in the EBLV-1 transmission still need to be specified .
Chiroptera is the second largest order of mammals after Rodentia . They have a worldwide geographical range , with the exception of Antarctica and are represented by more than 1 100 different species [1] , of which 36 are found in Europe . Of these 36 , 34 are reported in France , and all of them are strictly protected by national [2 , 3] and international [4] legislation as they are sensitive to the destruction of their habitat . With their long lifespan regardless of their size , their unique flying ability as a mammal , and their overactive immune system , they are considered exceptional mammals and such fundamental innate abilities have recently attracted the interest of the scientific community [5–8] . More than 200 viruses have been associated with bats [9] . They were recently discovered to be potentially at the origin of the Zaire Ebola virus [10 , 11] , and they have also been linked to other illnesses related to coronaviruses ( Severe Acute Respiratory Syndrome , Middle Eastern Respiratory Syndrome ) , filoviruses ( Ebola and Marburg ) , henipaviruses ( Hendra and Nipah ) , and Lyssaviruses [12–14] . Rabies is a severe and lethal disease transmitted by the saliva of an infected animal through bite , dogs being the main source of human infection . The current lyssavirus taxonomy includes 14 lyssavirus species of the Rhabdoviridae family , order Mononegavirales [15] , of which 12 species have been isolated in bats ( reservoirs of the Mokola virus ( MOKV ) and Ikoma lyssavirus ( IKOV ) still remain to be identified [16 , 17] ) . Phylogenetic analyses suggest that all these lyssaviruses have bat origins [18–20] . The number of recorded species will certainly increase in the future as suggested by the latest isolations of Gannoruwa Bat Lyssavirus in a fruit bat ( Pteropus medius ) in Sri Lanka , a new candidate in the formal classification of lyssaviruses [21] . In Europe , bat lyssavirus was documented for the first time in 1954 in Hamburg , Germany [22] . From 1977 to 2016 , 1 , 175 bat lyssavirus cases were recorded from the North to the South of the continent [23] . To date , 4 different lyssavirus species have been isolated in European bats . Initially , European bat lyssaviruses were genetically described into 2 different groups named European bat lyssavirus type 1 ( EBLV-1 ) and European bat lyssavirus type 2 ( EBLV-2 ) [24] . Recently , 2 new lyssavirus species represented by the Bokeloh Bat Lyssavirus ( BBLV ) located in Germany and in France [25 , 26] and the West Caucasian Bat Virus ( WCBV ) located in southern Russia [27] have been identified . A putative Lleida bat virus was detected in Spain in Miniopterus schreibersii but does not yet have a taxonomic status [28] . Most European bat cases have been recorded as belonging to EBLV-1 ( >95% ) , which is associated with the serotine bat , Eptesicus serotinus [29] , and with E . isabellinus in Spain , a sibling species of E . serotinus [30] . EBLV-1 molecular characterization has separated this species into 2 sublineages , EBLV-1a and EBLV-1b [31] . Lineage 1a shows a western-eastern European distribution from Russia to central France , while variant 1b exhibits a southern-northern European distribution from Spain to Denmark [32] . Except for 5 EBLV-2 cases in Pond bats ( Myotis dasycneme ) in the Netherlands [33] , all other EBLV-2 cases were isolated from Daubenton’s bats ( Myotis daubentonii ) within a distribution area including the Netherlands , United Kingdom , Switzerland , Germany and Finland [34–36] . Among this viruses , only EBLV-1 and EBLV-2 have been associated with human cases with two identified case per virus species [37] . In France , bat lyssavirus was identified for the first time in 1989 in the Lorraine region ( North-East France ) ( Briey and Bainville ) and a bat rabies surveillance program was consequently initiated [38] . Epidemiosurveillance and research programs to estimate the public health risks associated with the infection of native bats by Lyssavirus were then strengthened following the report of the French Ministry of Agriculture [39] , leading to the consolidation of the network involving both local veterinary services and the French National Bat Conservation Network ( SFEPM ) . From 1989 to present , 78 bat lyssavirus cases—75 EBLV-1 cases in common serotine bats , 1 EBLV-1 case in common pipistrelle ( Pipistrellus pipistrellus ) and 2 cases of BBLV in Natterer's bats ( Myotis nattereri ) —have been diagnosed in France ( E . Picard-Meyer , under revision ) and the issue of seasonality in the probability of detecting cases has been raised recently [40] . To gain a better understanding of virus transmission , active surveillance programs during population monitoring were set up in addition to the passive surveillance program . As shown in the synthesis made by Picard-Meyer for the 2004–2009 period , the sampling of such programs involved blood and saliva samples from more than 300 bats on 18 sites [41] . In such cross-sectional surveys , the bats were sampled on various sites throughout France and no data from marked individuals were available to allow a longitudinal study . This study proposes the first longitudinal survey of EBLV-1 in mono-specific serotine colonies , the main bat species found infected by this lyssavirus . Multi-state models , a category of capture-recapture analysis , have been used to attempt explaining EBLV-1 virus exposure . Originally developed for estimating the abundance of animal populations , capture-recapture methods have recently attracted attention in the field of veterinary epidemiology [42 , 43] . When appropriate data are available , capture-recapture models can also be used directly to estimate disease-associated mortality and epidemiological parameters , such as infection and recovery rates [44] . In this study , as the serological status of individuals could change , data were analyzed using a multistate capture-recapture approach [45] . Multistate models indeed allow individuals in a population to be distributed across multiple sites or among different disease states [46 , 47] . More precisely , we used multievent capture-recapture models [48] , an extension of multi-state models , to determine the factors influencing bat rabies transmission while accounting for imperfect detection and uncertainty in disease states . Survival , capture , transition and judgment probabilities were assessed by hypothesizing , based on lyssaviruses literature , that EBLV-1 exposure in bat maternity colony was driven differently according to the age of individuals and the period of time . To our knowledge , this is the first attempt of describing EBLV-1 circulation in its reservoir , over time , and by such approach and methodology .
Two maternity roost sites of serotine bat colonies located in the East of France ( Fig 1 ) were monitored . The sites were located in Universal Transverse Mercator ( UTM ) 32U zone in villages bordered by Moselle River and surrounded by hardwood forest , cropland and grassland . The climate is semi-continental and the landscape relatively flat with altitude ranging from 167 to 374 meters . Site A was the roof of a house in Ancy-sur-Moselle in the Moselle department , while site B—8 . 5 km from site A—was the garden shed of a house in Pagny-sur-Moselle in the Meurthe-et-Moselle department . Both sites were chosen following the detection of bat cadavers in 2009 and 2012 for site A and 2011 for site B ( E . Picard-Meyer , under revision ) . On site A , 6 dead animals in 2009 and one dead animal in 2012 tested positive for lyssavirus with reference techniques [49–51] . On site B , 2 dead individuals tested positive in 2011 . The infection was shown to be caused by the EBLV-1b variant , which is endemic in the region . They were the first detection of EBLV-1 infections in these municipalities . Capture-recapture sessions were completed in summer ( July and August ) and spring ( May ) between 2009 and 2015 ( during 7 years ) for site A , and between 2011 and 2015 ( during 5 years ) for site B . Capture sessions were organized by the French Agency for Food , Environmental and Occupational Health & Safety ( ANSES ) and the Commission for the Protection of Water , Heritage , Environment , Subsoil and Chiroptera ( CPEPESC ) of Lorraine Region , the naturalist association in charge of the study and protection of bats in the region . Residents provided full informed consent to have their residences used in the study . The trapping session dates were set up to avoid disturbing the bats during the parturition period . Captures were held at nightfall , when serotine bats are known to leave their roost to forage . Harp traps were used because they are the most suitable device when a large number of animals can be expected [52] . Moreover , these traps are considered the most effective for capturing bats without harming them [53] . Traps were placed at the exits of the roof from where the bats usually emerged . To avoid injury , they were handled carefully and firmly by trained people wearing gloves and adequately vaccinated against rabies . As soon as the bats were removed from the traps , they were placed temporarily in cotton bags and then held in the palm of the hand with fingers curled around the body [54] . The sex and age class were recorded for each animal and biological samples collected . Dry synthetic fiber swabs ( classiqSwabs , COPAN , France ) were soaked with saliva to assess EBLV-1 virus excretion as well as viral RNA detection . Blood was collected from the antebrachial vein along the propatagium , and more recently on the uropatagium , a method found more effective , to evaluate the serological status of each individual with respect to EBLV-1 exposure . Blood was collected using filter paper as described by Wasniewski et al . [55] and subsequently stored at -16°C till the analysis . Swabs were stored in 0 . 3 mL of DMEM culture medium ( Dulbecco’s minimum essential medium , Invitrogen , France ) at -80°C for further testing in the laboratory . Lipped bat bands ( split metal bat rings , PORZANAZTD , East Sussex , United Kingdom ) positioned on the forearm were used to mark the animals [56] . Each bat captured was assigned a single record number , allowing for follow-up over the successive capture sessions . After sampling , all the bats were immediately released at the site of night-time capture . None of the bats appeared sick or were euthanized during the study period . All the animals were handled in strict accordance with good animal practices and according to the EUROBAT guideline [57] . Field work and animal sampling were performed in accordance with French legislation . Because bats are protected species in France , prior formal authorization by the French Ministry of the Environment was granted for their trapping , handling , and sampling [58] and colony monitoring was undertaken following local authorization by the Prefect of the Lorraine Region [59] . In France and within the European Union , the legal framework for using under experimentation purposes is governed by Regulation 2010/63/EU of the European parliament and of the council of 22 September 2010 ( applicable and translated in French in 2013 ) and handling of wildlife animal in the field does not require any prior specific ethical approval . During each capture session , the captured animals were recorded and classified into different serological states concerning EBLV-1 neutralizing antibodies , “S” ( “NEG”/ “POS”/ “INC” ) . The “NEG” state included EBLV-1 seronegative individuals , i . e . bats that had never been in contact with the virus , and consequently susceptible to future infection or previously exposed but with a non-detectable level of EBLV-1 antibodies . The “POS” state included EBLV-1 seropositive animals . Seropositive animals were defined as animals that had been in contact with the virus and seroconverted . This state included both bats that were potentially protected against infection by antibodies , and sick bats . Because the serological test was occasionally inconclusive ( analysis not feasible ) , or no blood was sampled , an “INC” state was included . To address this particular issue and to allow the use of such a dataset , an extension of the multistate capture-recapture framework was used . This extension is known as the multievent model [47 , 67] . When an individual is observed in the field , its status can still remain unknown or uncertain , e . g . sex status [68] , reproductive status [69] but also epidemiological status [42 , 47] and such a model accounts for uncertainties in the assessment of a state . The different probabilities assessed during the study were as follows: The models were fitted using the E-SURGE program [70] . Both sites were maternity colonies mainly composed of females with roost site fidelity and juvenile males leaving the colony at the end of their first summer [71] , males were consequently discarded from the dataset to avoid bias in the survival analysis . Multiple capture sessions were conducted occasionally within the same season ( between 2 and 5 ) , and the detection/non-detection data were merged into a single capture session per year and season . Survival , recapture , transition and judgment probabilities were all computed by considering the serological status“S” ( POS/NEG/INC ) as explanatory variable . Age class “a” ( juvenile/adult ) was also considered as a potential explanatory variable for survival probabilities as juveniles could harbor higher mortality rate as demonstrated in serotine bat biology study [72] . Regarding serological transition probabilities , a previous EBLV-1 study suggested seasonal fluctuation in Myotis myotis colonies [73] , we consequently hypothesized that serotine colony could by driveen by a comparable dynamic and included the season “s” ( spring/summer ) as explanatory variable . This study being the only known EBVL-1 longitudinal studies on serotine monospecific colonies , we also assumed based on classical bat rabies virus ( RABV ) studies that transmission rate could vary according the age [74] and included age class “a” ( juvenile/adult ) in candidates models . The year “y” and/or season “s” ( spring/summer ) effects and their interaction were considered with regard to recapture probabilities as weather variations are suspected to impact trapping efficiency . Possible interactions with the serological status were also assessed to determine whether there were any specific infection patterns . All model combinations to estimate survival , transition , capture and judgment probabilities fit accordingly . Akaike's Information Criterion with a correction for small sample sizes ( AICc ) was used to assess the relative model fit . The model with the lowest AICc was selected as the model that fitted the data best [75] . When the ΔAICc was lower than 2 ( Δi = difference between AICc and the lowest AICc value ) , the most parsimonious model was selected ( i . e . the one with the fewest variables ) . To compute antibody prevalence and its standard error , we used the traditional abundance estimate and corrected the number of animals that tested positive or negative in each session by the corresponding recapture probability [46] . To account for “INC” observations , bats were assigned a “POS” or “NEG” status using the Viterbi algorithm [76] . For each site , a logistic regression was used to assess the effect of season and year on the estimated prevalence . The number of positive and negatives cases was used as the response variable , and the AICc was used to compare models either incorporating or excluding time variables .
On site A , 15 capture sessions were undertaken between 2009 and 2015 , corresponding to a total of 320 bat captures ( including single captures and recaptures ) . The distribution of the number of captures and recaptures per year and season is presented in Table 1 . Among the 214 marked animals , 81 individuals ( 38% ) were recaptured once , 19 individuals ( 9% ) were recaptured twice , 5 individuals ( 2% ) were recaptured 3 times and 1 individual was captured 5 times within the study period . Within the studied 201 individuals were females ( 94% ) and 13 were males ( 6% ) . All males but 2 were identified as juveniles . Both adults had a single capture history . By comparison , on site B , where 12 capture sessions were undertaken between 2009 and 2015 , there was a total of 473 bat captures , single captures and recaptures combined . The distribution of the number of captures and recaptures per year and season is also presented in Tables 1 and 2 . Among the 221 marked animals , 125 individuals ( 57% ) were recaptured once , 60 individuals ( 27% ) were recaptured twice , 36 individuals ( 16% ) were recaptured 3 times , 21 individuals ( 10% ) were recaptured 4 times , 7 individuals ( 3% ) were recaptured 5 times and 1 individual were captured 7 times , 8 times and 9 times within the study period . 156 captured bats were females ( 71% ) while 65 were males ( 30% ) . It should be noted that no other bat species have been identified within the study period , excepted one time where a Miniopterus schreibersii individual was trapped . Many different serological status histories were observed during the study ( S1 Table ) . Thus , some animals evolved from a negative to positive status , some did the reverse , and occasionally some changed several times ( See Supplementary material describing the frequencies of the different capture histories , inconclusive results were ignored for a better clarity ) . This process revealed that it was more frequent for a seropositive status to become seronegative than the opposite ( Table 3 ) . Indeed , 5 and 9 animals seroconverted ( from NEG to POS ) while 10 and 21 animals seroreverted ( from POS to NEG ) on sites A and B respectively . When analyzing oral swabs , all the tested animals were found negative for RNA detection in the saliva , apart from 5 individuals all captured in July 2009 on site A . Viral RNA was detected during a first capture for 4 animals and during a second capture for one animal . Among the animals captured for the first time , 3 animals ( 2 females—one adult , and one juvenile—and one juvenile male ) were only captured once . Among the two recaptured bats , one adult female was sampled again during the next capture session but was surprisingly found to be seronegative , with no RNA detection . The history of the second recaptured bat , found RNA-positive in the second capture of the same year , was notable . The adult female bat was indeed negative for both RNA and serology in July 2009 then , 3 days later , positive for RNA but inconclusive as to its serological status ( the blood sample could not be assessed ) and negative for virus excretion yet seropositive for EBLV-1 antibody detection in August 2010 , meaning one year later saliva has been detected RNA-positive . All the virus isolation tests failed to detect a live virus except for one seropositive juvenile female captured only once in July 2009 on site A , just after the positive testing of dead bats . All the samples from site B were negative for the presence of an infectious virus . Individual movements between the 2 colonies , 8 kilometers away , were possible but difficult to quantify as only one female marked on site B in July 2012 was found on site A in August 2013 . The best models for both sites A and B included effects for the year and season on recapture probabilities ( Tables 4 and 5 ) . No specific pattern was detected for survival probability on either site , neither age category nor serological status affecting survival . We found an effect of the interaction of serological status and age on the transition probability for site B , while the judgment probability depended on the serological status for site A only . On site A , the best-ranked model indicated that the survival probability of female bats was 0 . 86 [0 . 76–0 . 93] . The transition probability from seropositive to seronegative was 0 . 99 [0 . 02–0 . 99] ( a boundary estimate that was difficult to interpret ) and 0 . 21 [0 . 09–0 . 40] from seronegative to seropositive . The probability of judging a positive result as positive was 1 while the probability of judging a negative result as negative was 0 . 81 [0 . 75–0 . 86] . On site B , the best-ranked model indicated that the survival probability of female bats was 0 . 78 [0 . 73–0 . 83] . The transition probability from seropositive to seronegative was 0 . 89 [0 . 53–0 . 98] and 0 . 25 [0 . 06–0 . 61] from seronegative to seropositive for adult female bats and 0 . 15 [0 . 07–0 . 27] and 0 . 05 [0 . 02–0 . 12] for juveniles female bats respectively . The evolution of corrected bat EBLV-1 seroprevalence on both sites A and B are presented in Fig 2 . On site A , seroprevalence varied from 34% in summer 2010 [28 . 0–41 . 0] and summer 2012 [27 . 5–41 . 9] to 0% in spring 2013 [0–1 . 5] and spring 2015 [0–6 . 1] . The trend appeared to indicate a constant decrease in seroprevalence over time during the study , with an approximate 2–3 year oscillation interval ( 2011–2013 ) . The logistic regression detected a significantly lower frequency of seropositive cases from 2011 to 2013 than in 2010 ( OR2011 = 0 . 26 [0 . 19–0 . 37]; OR2012 = 0 . 65 [0 . 47–0 . 90]; OR2013 = 0 . 40 [0 . 29–0 . 55] ) and a lower frequency of seropositive cases in spring than in summer ( ORspring = 0 . 46 [0 . 35–0 . 60] ) . On site B , seroprevalence peaked at 70% [59 . 9–78 . 9] in spring 2013 then progressively decreased to reach 21 . 6% [11 . 5–34 . 9] in May 2015 . The logistic regression did not detect any significant differences between seasons , but the frequency of seropositive cases in 2013 was higher than in 2012 ( OR2013 = 2 . 40 [1 . 61–3 . 60] ) .
The analysis of the two roost site colonies using multievent models within the study period did not evidence any impact of the serological status on individuals’ survival or recapture probabilities . This supports previous observations that bats could harbor exposure events without any impact on their mortality rate . These results are indeed comparable to the survival analysis computed for big brown bats ( Eptesicus fuscus ) affected by RABV in the United States [77] and for a Myotis myotis colony affected by EBLV-1 in Spain [73] . The detection rates of bats were also demonstrated to be uncorrelated to the serological status , indicating that seropositivity does not induce a potential behavioral change in bats that could impact the recapture probability . This finding supports the hypothesis that , in our study , observed seroprevalence of a capture session can be regarded as an unbiased estimation of the percentage of animals wo have been exposed to EBLV-1 in the colony . Recapture probabilities on both sites were shown to be affected by seasonal and annual variations . This temporal dependency could be due to changes in climate and weather conditions over seasons and years , known to affect the emergence of bats and consequently the effectiveness of captures using a harp trap [78] . EBLV-1 virus-neutralizing antibodies have been found in various bat field studies , principally through single captures [41 , 79 , 80] or in successive captures of mono-specific colonies like Myotis myotis [73] , Eptesicus isabellinus [81] and also in longitudinal studies of multi-species colonies [82–84] . In these previous studies , seroprevalence varied greatly according to the site location , species and time ( month and year ) . This study , to our knowledge , is the first extensive longitudinal analysis of 2 mono-species serotine colonies , a species currently considered as the main EBLV-1 reservoir . Our study demonstrated that individual serological transition scenarios are highly variable . We found seroconversions ( from seronegative to seropositive ) , seroreversions ( from seropositive to seronegative ) in addition to occasional multiple seroreversions and seroconversions in succession ( about 10/393 individuals ) . It should be noted that such multiple reversions could be questionable as they may also reflect limitations of the serological tests [85] performed furthermore on small amounts of blood . Globally , on both sites , seroreversions were more frequent than seroconversions , suggesting that the 2 studies could have occurred at the end of the rabies epizootic wave . This hypothesis could be supported by the fact that prior to the first established EBLV-1 case in July 2009 , approximately 30 to 40 individuals were found dead by the house owner , but the animals were unfortunately not collected and analyzed . Previous longitudinal studies in Spain have shown the seropositive status of Myotis Myotis over 3 years [84] . The maximum length of time observed between positive serological statuses in our own study was 4 years ( 2 individuals on site A ) , suggesting the possible persistence of seropositivity over several years . In this study , cut-off level used to discriminate positive from negative animals was determined to minimize the risk of false positive results . This caution was taken to provide reliable identification of positive individuals and to avoid false conclusion in the statistical analysis , but could have resulted in an underestimation of seroprevalence and in low statistical power . Lyssaviruses are excreted only at certain periods , and the chance of finding the virus or RNA in bat saliva during active surveillance field studies is relatively poor [41 , 79] . Interestingly , and for the first time in EBLV-1 longitudinal study , 5 individuals from site A were found with viral RNA in saliva in July 2009 ( four were sampled in the same capture session , in the beginning of July and 1 was sampled three days later ) , during the same period in which bat mortality was observed . Of the 5 positive samples , one was demonstrated as effectively infectious , showing that RNA in the mouth cavity can be concomitant to virus excretion . One initially seronegative animal was captured several times in succession . Saliva was found RNA positive 3 days later and , when captured again 1 year later , the animal was again found seropositive , supporting the hypothesis that seropositivity persists for a long time after infection or that alternate subclinical infection occurred . The risk of cross-contamination regarding the four RNA positive samples collected during the same capture session can’t be completely ruled out with certainty although usual precautions to avoid false-positive PCR results were strictly followed during bat handling and in the laboratory . However , the raw-data of this capture shown that the RT-PCR positive swabs were collected and analyzed intercalated with RT-PCR negative swabs , suggesting that laboratory cross-contamination is unlikely . The first attempt to define the temporal dynamic infection of serotine bat colonies was undertaken through a one-year study [86] . In this latter study , seroprevalence declined from 74% to below 10% within a few months ( from spring to fall ) . In contrast , Sera-Cobo at al . ( 2013 ) found a significantly higher antibody prevalence in summer when maternity colonies are present in most localities . This pattern was magnified by the presence of multi-species colonies compared to mono-specific colonies , with social contacts between bats . Colony formation , conferring thermodynamic and social advantages to reproductive females during pregnancy and lactation , could indeed increase the rate of rabies exposure due to hypothetically higher probabilities of inter-individual and inter-species interactions . In our study , site A data revealed higher prevalence in summer than in spring , supporting the conclusion that numerous inter-individual interactions of the colony during the post-partum season ( care for the juveniles ) could increase the probability of exposure . On site A , corrected seroprevalence decreased over time with significantly higher seropositive frequencies in 2010 ( 34% of seropositive bats ) than in 2011–2013 , while on site B , a peak of infection was observed in 2013 ( 70% of seropositive bats ) , midway through the 5-year-study . Our data on serotine colonies thus appear to confirm the cyclic temporal hypothesis of bat infections already proposed for Myotis Myotis , with an estimated 2–3 year cycle for site A at the time of the study [73] . The model suggests that after the initial introduction of EBLV-1 into the susceptible bat colony , the seroprevalence of the colony increases then , depending on the period , tends to oscillate , its amplitude decreasing year after year . Ecological studies performed in the bordering of Luxembourg and Germany , a hundred kilometers far from the study area , have shown that females were forming maternity colonies at the middle of April and had a philopatric behavior , meaning that each year the breeding colony invests the same maternity site [72]: This eight years study also shown that median period of birth was happening in the middle of June . The young bats usually make their first flights at around three weeks old , and at six weeks they can forage for themselves . Breeding colonies usually disperse by early September , although a few bats may use the colony site as a roost until early October [72] . Reproduction seems to take place in the autumn , but very little is known about the mating behavior . Hibernation of serotine bats occurs between October and end of March . However , very few information is indeed known about this period . Based on RABV rabies model transmission in the United States , the potential impact of hibernation on the virus’s capacity to remain in animals populations has been raised [74] . The hypothesis is that hibernation could allow infected individuals and their pathogens to survive , infected virus particles being potentially hosted and preserved in brown fat [87] . The relationship between the incubation period , hibernation season and annual birth pulse could indeed generate complex dynamics that should attract more attention in bat rabies studies . With a long incubation period , infected bats could survive long enough to enter hibernation and be responsible for infectious contacts in the main transmission season that follows , maintaining a reservoir until the birth pulse provides a new supply of immunologically naïve bats . Further model predictions fitted this assumption and showed that adult female bats were infectious earlier in the year , whereas infectious juveniles appeared later in the summer [74] . In a previous study , female serotine bats were shown to be more exposed to EBLV-1 than males , probably due to their gregarious social behavior , males being more solitary [88] . Similar findings were also highlighted in the framework of RABV transmission in Brazilian free-tailed bats [89] big brown bats [77] and in vampire bats [90] . In our study , occurring in breeding colonies , only females were assessed and we have shown evidence from site B that seroreversions were significantly more frequent than seroconversions . The seroreversion frequencies of adult females were higher than those of other transition states in juvenile females . Hence , adult female serotine bats appear to be a good indicator of EBLV-1 epizootic dynamics . This raises the question of whether adult female bats are more exposed to the virus due to the mating period in September and whether they could play a major role in virus maintenance , acting as a potential source of virus transmission . Because the reproductive status of adult females could potentially drive inter-individual exposure and transmission differently , it would be valuable to consider the reproductive status as ‘pregnant’ , ‘lactating’ and ‘non reproductive’ for further studies . However , such age-related rabies dynamics we detected could also reflect the greater difficulties in characterizing EBLV-1 dynamics in juveniles due to their lower occurrence than adults in the population . The possibility of maternal antibodies transfer in juveniles via the placenta or during lactation also raises questions . Indeed , although this phenomenon has been known and measured for in experimental animals or domestic animals [91–93] , the situation regarding bats and EBLV-1 is unknown . Its impact on the antibody level in juveniles , and therefore , on the evaluation of exposure , would need to be clarified . All the EBLV-1 cases detected on sites A and B were detected from end of June to start of August and all determined in dead bats identified morphologically as juveniles ( E . Picard-Meyer , under revision ) , again raising the question of the key role of age in the virus’s transmission . The influx of susceptible young in summer could act as a crucial driver of EBLV-1 dynamics . The role of susceptible young in transmission dynamics has indeed already been raised in previous discussion on zoonotic diseases [94 , 95] . Such biological enigmas need to be clarified and additional studies are still needed , especially in the framework of age-related bat EBLV-1characterization . The means and rate of bat-to-bat transmission in serotine populations still need to be clarified . This is a difficult question to solve in field studies and the legal status of bats in Europe due to the decline of bat populations ( all the 36 species are strongly protected by European regulation ( Council Directive 92/43/EEC 1992 ) [96] ) has made experimental studies difficult to implement . Only one experimental study of EBLV-1 infection in caged serotine bats was carried out in 2009 through different means of inoculation [97] . It appeared that the environmental contamination of bats is unlikely as none of the intranasally-inoculated bats seroconverted . Infection through bites was indicated as having the greatest potential for inter-bat transmission , as the subcutaneous route of inoculation was found to be relatively efficient [97] . Despite living in close proximity to humans , human contacts with serotine bats are rarely reported . It should be noted that during this study , despite the discovery of two infected bat colonies , no sanitary incidence has been reported , nor in human neither in domestic animals . To date , only 2 EBLV-1 induced human deaths have been reported in Voroshilovgrad , Ukraine ( 1977 ) and in Belgorod , Russia ( 1985 ) [98] . Experimental infections have also shown evidence of a very limited risk of an EBLV spillover from bat to fox [99] . A few natural EBLV-1 spillover cases have been so far reported in a sheep , a stone marten , 2 cats and a fruit bat [29 , 100] . The risk of transmission to other species thus appears very low . However , to avoid any risk of contamination , protective measures such as personal protective equipment , post-exposure rabies prophylaxis or a booster dose in the event of exposure have been established for bat biologists in Europe [96] and France [101 , 102] . EBLV-1 antibody carriage in serotine bats was not correlated with mortality probability . In both site A and B , peak-seroprevalences ( 34 and 70% respectively ) were detected one or two years after the first detection of EBLV-1 positive carcasses . While we detected oscillation seroprevalences in time , at annual level , seroprevalences were found higher in summer compare to spring , suggesting that rearing period could increase virus circulation . We pointed out differences of serological statuses between adult female and juveniles and the need for further assessment . A better understanding of this mechanism , whether of ecological , biological and/or immunological origin , is indeed a real challenge and of great interest as elucidating zoonotic virus persistence in bats concomitant to unaffected survival could help to solve human health challenges .
|
A multi-annual survey of two serotine bat ( Eptesicus serotinus ) maternity colonies previously found exposed to European Bat Lyssavirus type 1 ( EBLV-1 ) was assessed using capture-recapture methodology . The two roosting site were located in the North-East of France . Animals were trapped , banded , and blood samples were collected to study their status regarding EBLV-1 exposure . Using capture-recapture models , the authors found that seropositive status of bats did not affect the survival abilities of individuals . Seroprevalence of EBLV-1 antibodies within the study showed an oscillation interval of approximately 2–3 years and a higher evidence of contact with the virus in summer than in spring . The maximum duration observed between successive positive serological statuses in the bat population also demonstrated a survival for at least 4 years after the exposition . This study confirms the ability of bats to survive despite circulation of lyssaviruses within the colony . Bats could indeed provide a valuable key to improving human health , currently facing numerous zoonotic epidemic issues .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] |
[
"medicine",
"and",
"health",
"sciences",
"body",
"fluids",
"immune",
"physiology",
"pathology",
"and",
"laboratory",
"medicine",
"european",
"union",
"pathogens",
"immunology",
"geographical",
"locations",
"microbiology",
"vertebrates",
"saliva",
"animals",
"mammals",
"viruses",
"animal",
"behavior",
"rna",
"viruses",
"antibodies",
"zoology",
"animal",
"sociality",
"fruit",
"bats",
"immune",
"system",
"proteins",
"serology",
"proteins",
"behavior",
"medical",
"microbiology",
"microbial",
"pathogens",
"people",
"and",
"places",
"biochemistry",
"france",
"lyssavirus",
"eukaryota",
"anatomy",
"viral",
"pathogens",
"bats",
"physiology",
"biology",
"and",
"life",
"sciences",
"europe",
"amniotes",
"organisms"
] |
2017
|
Longitudinal survey of two serotine bat (Eptesicus serotinus) maternity colonies exposed to EBLV-1 (European Bat Lyssavirus type 1): Assessment of survival and serological status variations using capture-recapture models
|
Clustered , Regularly Interspaced Short Palindromic Repeats ( CRISPR ) abound in the genomes of almost all archaebacteria and nearly half the eubacteria sequenced . Through a genetic interference mechanism , bacteria with CRISPR regions carrying copies of the DNA of previously encountered phage and plasmids abort the replication of phage and plasmids with these sequences . Thus it would seem that protection against infecting phage and plasmids is the selection pressure responsible for establishing and maintaining CRISPR in bacterial populations . But is it ? To address this question and provide a framework and hypotheses for the experimental study of the ecology and evolution of CRISPR , I use mathematical models of the population dynamics of CRISPR-encoding bacteria with lytic phage and conjugative plasmids . The results of the numerical ( computer simulation ) analysis of the properties of these models with parameters in the ranges estimated for Escherichia coli and its phage and conjugative plasmids indicate: ( 1 ) In the presence of lytic phage there are broad conditions where bacteria with CRISPR-mediated immunity will have an advantage in competition with non-CRISPR bacteria with otherwise higher Malthusian fitness . ( 2 ) These conditions for the existence of CRISPR are narrower when there is envelope resistance to the phage . ( 3 ) While there are situations where CRISPR-mediated immunity can provide bacteria an advantage in competition with higher Malthusian fitness bacteria bearing deleterious conjugative plasmids , the conditions for this to obtain are relatively narrow and the intensity of selection favoring CRISPR weak . The parameters of these models can be independently estimated , the assumption behind their construction validated , and the hypotheses generated from the analysis of their properties tested in experimental populations of bacteria with lytic phage and conjugative plasmids . I suggest protocols for estimating these parameters and outline the design of experiments to evaluate the validity of these models and test these hypotheses .
For many species of bacteria , adaptive evolution is through the expression of chromosomal and extrachromosomal ( plasmid- and prophage - borne ) genes or clusters of genes ( pathogenicity and nicer islands ) acquired by horizontal gene transfer ( HGT ) from the same or even quite distant species [1] , [2] . Thus , on first consideration it may seem that bacteria and their accessory genetic elements would have mechanism to promote the acquisition , incorporation and expression of genes from without . And , indeed there are mechanisms like integrons [3]–[7] that appear to have that function . On the other side , DNA acquired from external sources may be deleterious . This is certainly the case when that DNA is borne on lytic bacteriophage , but also for plasmids that engender fitness costs [8] , [9] or chromosomal DNA from the wrong source [10] , [11] . To deal with these contingencies , it would seem that bacteria would have mechanisms to protect themselves against infection by deleterious foreign DNA [12] . And indeed there are systems like restriction-modification ( restriction endonucleases ) which appear to have that role [13] , [14] . The most recently discovered mechanism postulated to provide bacteria immunity to infectious genetic elements are Clustered Regularly Interspaced Short Palindromic Repeats ( CRISPR ) . For recent reviews see [15] , [16] . CRISPR is particularly intriguing because of its ubiquity , appearing in ∼90% and ∼40% of archaeal and eubacterial sequenced genomes , respectively , and because of the adaptive mechanism by which it provides immunity to infections by a virtually indefinite diversity of bacteriophage and plasmids . DNA from infecting phage and plasmids is incorporated into the CRISPR array . Through a yet to be fully elucidated mechanism , bacteria abort the replication of infecting phage [17] or the establishment of conjugative plasmids [18] bearing copies of the DNA incorporated into their CRISPR arrays , also see [19] . Further support for CRISPR being an adaptive immune system that is maintained because it protects bacteria from infection with phage comes from studies of the community ecology of bacteria and phage; DNA in the CRISPR regions of the bacteria from those communities corresponds to that in the co-existing phage [20]–[23] . For an intriguing perspective on CRISPR as a witness to the coevolutionary history of bacteria and phage , see [24] . CRISPR-mediated immunity has been likened to a Lamarckian mechanism [25] , because the selection pressure , the infecting phage and plasmids , determine the genotype . This analogy however does not account for the evolution and maintenance of the machinery responsible for taking up the infecting phage and plasmid DNA and the mechanism employed to prevent the replication or establishment of infecting genetic elements with those sequences . Under what conditions will adaptive immunity to phage and plasmid infection be the selection pressure responsible for establishing and maintaining CRISPR-mediated immunity in populations of archeae and bacteria ? What about other mechanisms of resistance , like structural modification blocking phage adsorption ( envelope resistance ) and restriction-modification ? How do these mechanisms interact with CRISPR – acquired immunity and contribute to its establishment and maintenance ? To address these questions and provide a framework and hypotheses for their study experimentally , I use mathematical models of the population dynamics of bacteria , phage and plasmids to explore the conditions under which a CRISPR–like adaptive immune mechanism will provide bacteria a selective advantage in competition with bacteria without this immune system . The results of the numerical analysis of the properties of these models suggest that with bacterial replication and phage infection parameters in realistic ranges , there are broad but not universal conditions where a CRISPR–like adaptive immune system can be favored and will be maintained in populations of bacteria confronted with lytic phage . While this model predicts conditions where CRISPR-mediated immunity will be favored when bacteria compete with populations bearing conjugative plasmids , these conditions are relatively restrictive . The parameters of these models can be independently estimated , the validity of the assumptions behind their construction and the hypotheses generated from the analysis of the properties can be tested in experimental populations of bacteria with lytic phage and conjugative plasmids . Procedures for doing these experiments are outlined and their potential outcomes described and/or speculated upon . Also discussed are the broader implications of CRISR-mediated adaptive immunity to the population and evolutionary biology and ecology of bacteria and phage .
Both the lytic phage and conjugative plasmid models used here assume a chemostat-like habitat . The bacteria grow at a rate that is a monotonically increasing function of the concentration of a limiting resource , R µg/ml [26] . where Vi hr−1 is the maximum growth rate of the ith strain of bacteria and k the concentration of the resource when the growth rate is half its maximum value ( the “Monod constant” ) . The populations are maintained in a vessel of unit volume , ( 1ml ) into which medium containing the limiting resource from a reservoir where it is maintained at a concentration A µg/ml flows in at a rate w per hour . Excess resource and wastes are removed from the vessel at the same rate . As in [27] , the rate of uptake of the resource by the bacteria is proportional to the density , the resource concentration-dependent growth rates of the different populations of bacteria and a conversion efficiency parameter , e µg/per cell . The model developed here is an extension of that in [28] . There are four populations of bacteria . Two are sensitive to the phage , N , non–CRISPR and C , CRISPR and two that are either fully resistant ( envelope resistance ) , or immune because of CRISPR , NR and CR , respectively . The variables N , C , NR and CR are the both the densities ( bacteria per ml ) of these populations and used as their designations . There is one population of phage , with density and designation , P particles per ml . The phage adsorb to the N and C and CR bacteria with rate constants , δN and δC ( ml per phage per cell per hour ) respectively . Phage do not adsorb to bacteria with envelope resistant , i . e . the NR cells . To account for a possible multiplicity of infection ( MOI ) effect on survival of phage-infected CR , the effective killing rate constant for phage adsorption to CRISPR can be an increasing function of the ratio of free phage and CR cells , M = P/CR . ( 1 ) where δMIN and δMAX are the minimum and maximum adsorption rates . The parameter x is a coefficient ( 0≤x≤1 ) that specifies the magnitude of the MOI effect , q is the MOI where the adsorption rate is half its maximum value and n is an exponent which contributes to the shape of the distribution . At low multiplicities , δCR ( M ) the CRISPR cells would be effectively immune ( resistant ) ( Figure 1 ) . At high multiplicities , however , immune CRISPR cells can be overburdened by phage , their immunity would be overridden , and the phage would replicate , killing the cells . On the other side , we assume that the phage are removed from the population by adsorption to immune CRISPR cells at the maximum adsorption rate , δMAX . For convenience I neglect the latent periods of the phage infection but assume that the phage have potentially different burst sizes , βN , βC , and βCR particles per cell , for N , C and CR cells , respectively . Phage-immune CRISPR cells , CR are produced from C at a rate proportional to the rate at which the phage adsorb to them and a constant m ( 0≤m≤1 ) which is the probability that a phage infection will be aborted and a CRISPR strain will be produced . At a rate v per cell per hour , CRISPR lose their immunity , CR→C . For the N and C populations the loss of the adsorbed phage is subsumed in the value of the burst size ( which is one less than the number of phage produced ) . For the CR population , the loss of the phage due to adsorption is specifically considered because only a small fraction of the adsorbed phage replicate when the MOI is low . In Table 1 , I separately define these parameters and in Figure 2 , illustrate the interactions between the different populations of bacteria and the phage . The equations for this model follow . The model developed here is an extension of that in [29] . There are five bacterial populations . Two populations do not code for CRISPR , N and NP , and three populations code for CRISPR , C and CP and CX . The NP and CP populations bear the conjugative plasmid and CX , carries CRISPR and plasmid sequences that make it completely immune to the receipt of these plasmids . Plasmids are transferred by conjugation at rates proportional to the product of the densities of the plasmid-bearing and plasmid-free populations and rate constants , γNN , γNC , γCN and γCC ( ml per cell per hour ) respectively for the transfer of the plasmid from NP to N , NP to C , CP to N and CP to C . , respectively . Plasmids are lost by vegetative segregation at rates τN and τC per cell per hour , with NP→N and Cp→C . C are converted to CX at a rate proportional to the rate at which C acquires the plasmid and a probability m ( 0≤m≤1 ) . Cx lose the CRISPR plasmid immunity region and become C at rate ν per cell per hour . Each of the cell lines , have a maximum growth rate , VN , VNP , VC , and VCP , and VX per hour . In Figure 3 , I illustrate the interactions between the different cell lines in this model , and , in Table 2 , I separately define the parameters and variables . The equations for this model are: For the numerical solutions to these equations ( computer simulations ) I use a differential equation-solving software package , Berkeley Madonna . For the phage simulations there is a refuge density , below which the phage are unable to adsorb to the bacteria . The purpose of this is to control the system from oscillating without limits , see [30] . In these simulations , if the phage density falls below 10−1 particles per ml , the phage are considered to be lost . Copies of these simulations are available online , www . eclf . net/programs .
The bacterial growth , resource-uptake , phage adsorption parameters and burst sizes used in these simulations ( Table 1 ) are in a range similar to that which we observed for E . coli and the phages T2 and T7 [28] , [31] . In accord with [34] , conjugative plasmids will be maintained as long as the rate of infectious transfer exceeds the rates of loss of the plasmid due to selection against the cells carrying it , vegetative segregation , and the rate of flow through the chemostat . In terms of the above parameters , the plasmid will be maintained in an N-NP population as long as ( 2 ) where N* is the density of plasmid-free cells at the chemostat equilibrium . For example , if VN = 1 . 0 , VNP = 0 . 95 , w = 0 . 2 , τN = 10−3 , the plasmid will be maintained in a population of density N* = 108 as long as γNN>1 . 1×10−10 . If the plasmid augments the growth rate ( which in this model is the sole parameter of cell fitness ) of the bacteria that carry it , VNP>VN , as we would anticipate for antibiotic resistance encoding plasmids in the presence of the selecting antibiotic , bacteria bearing the plasmid will be able to invade even without transfer , as long as the segregation rate , τN , is sufficiently small .
The models developed in this report incorporate what has been learned about CRISPR-mediated adaptive immunity to phage and conjugative plasmids , primarily from the studies of Barrangou and colleagues [17] and Marraffini and Sontheimer [18] , into models of the population dynamics of lytic phage [28] and conjugative plasmids [29] . Although they may appear complex , at best they are simplistic caricatures of interactions between these infectious genetic elements and bacteria with CRISPR-mediated adaptive immunity . These models are not intended or anticipated to be numerically precise analogs of these processes and dynamics . The role of these mathematical models is similar to that of the diagrammatic models ( cartoons ) used to illustrate the molecular basis and mode of action of CRISPR , i . e . , to provide a framework for understanding these processes , designing experiments , and interpreting their results . In this case , these experiments are on population and evolutionary dynamics of bacteria with CRISPR-mediated immunity confronted with lytic phage and competing bacteria bearing conjugative plasmids . The purpose of these models for this experimental enterprise is: ( i ) to identify and , in a quantitative way , evaluate the role of the different factors ( parameters ) contributing to these dynamics and the conditions for the establishment and maintenance of CRISPR in bacterial populations , and ( ii ) to generate hypotheses about these dynamics and existence conditions that can be tested ( and rejected ) in experimental populations . The results of the analysis of the properties of the phage - CRISPR model are consistent with the proposition that in the presence of lytic bateriophage there are broad conditions under which a CRISPR–like adaptive immune system can become established and will be maintained in bacterial populations . With population densities , growth rates , and phage infection parameters in realistic ranges , these models predict that despite a growth rate disadvantage , bacteria with CRISPR–like acquired immunity to infecting phage will increase in frequency when initially rare and will be maintained . The necessary condition for this is that the phage population continues to persist at a sufficiently high density for CRISPR-mediated adaptive immunity to overcome an intrinsic disadvantage associated with the costs of carrying and expressing these genes . When will the phage maintain their populations at sufficient levels for this outcome ? With the parameters used to address this question , the phage will be maintained under broad conditions , but may eventually be lost if a population with envelope or other resistance ascends to dominance . I emphasized the word may for two reasons . The first is theoretical , if the relative growth rate of the resistant population is adequately low , the phage and thereby CRISPR will be maintained . The second is empirical , even when resistant bacteria dominate experimental populations of bacteria and phage , in general the phage continue to be maintained [30] , [31] , [36] . The CRISPR plasmid model predicts that because of the immunity to infection with conjugative plasmids , a lower growth rate ( Malthusian fitness ) CRISPR population can become established and will be maintained when competing with bacteria with a greater Malthusian fitness but bearing deleterious ( fitness-reducing ) conjugative plasmids . Although these conditions are met with plasmid fitness costs in the range estimated for “laboratory” plasmids [9] , [37] , it is not clear that naturally occurring plasmids would be as burdensome as those maintained in the Lab . The greater the Malthusian fitness burden attributed to the plasmid , the greater the advantage of CRISPR-mediated immunity . The rate constants of plasmid transfer used in these simulations are those for plasmids with permanently derepressed conjugative pili synthesis . Wild type conjugative plasmids are more likely to be repressed for the production of these transfer organelles and would have substantially lower rates of transmission than plasmids that are permanently derepressed for plasmid transfer [38] , [39] . Indeed , it is not clear whether in natural populations conjugative plasmids that engender fitness cost can be maintained by transfer alone . Their persistence may require periodic episodes where bacteria carrying them have an advantage [34] , [40] , but also see [41] . If the rate of infectious transfer is not sufficient to maintain deleterious plasmid in a population and they persist by continually or periodically enhancing the cells Malthusian fitness , immunity to these plasmids would not be sufficient to maintain CRISPR-encoding cells that have an intrinsic fitness disadvantage . It would be nearly impossible to determine whether the quantitative conditions predicted by these models for the establishment and maintenance of CRISPR-mediated immunity are met in natural populations . On the other hand , the values of the parameters of these models can be estimated and the validity of the assumptions behind their construction and hypotheses generated from the analysis of their properties can be tested in laboratory culture using CRISPR–positive and CRISPR–negative bacterial constructs , phage and plasmids of the types used respectively by Barrangou and colleagues [17] and Marraffini and Sontheimer , [18] in chemostat culture . In this report , I elected to restrict the model and its analysis to the simplest cases with lowest realistic number of states of bacteria , phage and plasmids . I have done so because at this time these minimum number of states models and the predictions generated from their analysis are more amenable to evaluating and testing experimentally than models with more states of bacteria , phage and plasmids . Moreover , these tests , and particularly the population dynamic experiments , should indicate the importance of the generation of additional population states by mutation , like host range phage and host range plasmids , are to these dynamics . Be that as it may , I also realize that this minimum number of states model will not account for what may turn out to be the most important contributions of CRISPR-mediated immunity to the ecology as well as the population and evolutionary biology of bacteria and phage .
|
CRISPR is the acronym for the adaptive immune system that has been found in almost all archaebacteria and nearly half the eubacteria examined . Unlike the other defenses bacteria have for protection from phage and other deleterious DNAs , CRISPR has the virtues of specificity , memory , and the capacity to abort infections with a virtually indefinite diversity of deleterious DNAs . In this report , mathematical models of the population dynamics of bacteria , phage , and plasmids are used to determine the conditions under which CRISPR can become established and will be maintained in bacterial populations and the contribution of this adaptive immune system to the ecology and ( co ) evolution of bacteria and bacteriophage . The models predict realistic and broad conditions under which bacteria bearing CRISPR regions can invade and be maintained in populations of higher fitness bacteria confronted with bacteriophage and narrower conditions when the confrontation is with competitors carrying conjugative plasmids . The models predict that CRISPR can facilitate long-term co-evolutionary arms races between phage and bacteria and between phage- rather than resource-limited bacterial communities . The parameters of these models can be independently estimated , the assumptions behind their construction validated , and the hypotheses generated from the analysis of their properties tested with experimental populations of bacteria .
|
[
"Abstract",
"Introduction",
"Model",
"Results",
"Discussion"
] |
[
"evolutionary",
"biology/microbial",
"evolution",
"and",
"genomics",
"molecular",
"biology/molecular",
"evolution",
"microbiology/microbial",
"evolution",
"and",
"genomics",
"ecology/population",
"ecology",
"genetics",
"and",
"genomics/population",
"genetics"
] |
2010
|
Nasty Viruses, Costly Plasmids, Population Dynamics, and the Conditions for Establishing and Maintaining CRISPR-Mediated Adaptive Immunity in Bacteria
|
Polar auxin transport lies at the core of many self-organizing phenomena sustaining continuous plant organogenesis . In angiosperms , the shoot apical meristem is a potentially unique system in which the two main modes of auxin-driven patterning—convergence and canalization—co-occur in a coordinated manner and in a fully three-dimensional geometry . In the epidermal layer , convergence points form , from which auxin is canalized towards inner tissue . Each of these two patterning processes has been extensively investigated separately , but the integration of both in the shoot apical meristem remains poorly understood . We present here a first attempt of a three-dimensional model of auxin-driven patterning during phyllotaxis . We base our simulations on a biochemically plausible mechanism of auxin transport proposed by Cieslak et al . ( 2015 ) which generates both convergence and canalization patterns . We are able to reproduce most of the dynamics of PIN1 polarization in the meristem , and we explore how the epidermal and inner cell layers act in concert during phyllotaxis . In addition , we discuss the mechanism by which initiating veins connect to the already existing vascular system .
In plants , most developmental processes are driven by the spatiotemporal distribution of the growth regulator auxin . The versatility of the morphogenetic role played by auxin relies on its self-regulated polar transport , in which auxin transport feedbacks on auxin efflux carriers , the PIN proteins [1] . This process can lead to various distribution patterns , depending on the specific geometry of the organ considered . Although auxin transport and patterning occur fundamentally in three-dimensional tissues , they are usually explored in one- or two-dimensional models . These restrictions are commonly justified by geometrical considerations . For instance , models of phyllotactic patterning in the shoot apical meristem ( SAM ) of Arabidopsis assume that the formation of convergence points of PIN1 polarization at primordia takes place in the single epidermal cell layer L1 [2–5] , based on observations by Reinhardt et al . [6] . However , it is known that inner tissues are essential for positioning primordia [7–9] . Bayer et al . [10] proposed an auxin transport model integrating both the formation of convergence points in the L1 and the patterning of vascular strands in the subepidermal layers . Their model is implemented on a 2D cellular template representing a longitudinal section through the meristem . Thus , primordia positioning and midvein development are simulated in , respectively , one and two dimensions . These restrictions in dimensionality limit the range of potential behaviors displayed by the model , and therefore its capacity to integrate and assess current knowledge on phyllotaxis . Although fully three-dimensional models would be a significant step forward , they still present technical challenges [11] . In a notable effort in this direction , Gruel et al . [12] modeled gene expression dynamics and cell-to-cell diffusion of signals on a 3D cellular template to explore positioning and maintenance of the stem cell niche in the meristem . Modeling auxin polar transport is arguably more challenging , since it requires in addition a description of PIN exo- and endocytosis between cytoplasm and membranes , and , ideally , auxin diffusion in the apoplast . Several technical difficulties pertaining to the modeling of 3D plant tissues could be alleviated by using the topological notion of cell complexes . It provides a framework for locally describing how the components of a plant tissue ( cell interiors , membrane elements , cell walls ) are connected to each other . Based on such a description , efficient algorithms can be developed for simulating changes in the structure of the tissue or flows within the tissue [13] . An implementation of this paradigm has been utilized by Yoshida et al . [14] to investigate the control of division orientation in early Arabidopsis embryogenesis . Cell complexes are likely to turn out even more useful when fluxes between tissue components are involved . Another defying peculiarity of the SAM is the co-occurrence of convergence and canalization at midvein initiation . In the L1 , PIN1 polarize towards convergence points , where auxin accumulates . From these points , strands of cells with high PIN1 expression extend into the subepidermal layers , with PINs displaying a canalization pattern . This co-occurrence suggests that these two modes of auxin transport regulation do not rely on completely different mechanisms , but instead relate somehow to each other . Convergence and canalization are commonly conceptualized using two distinct polarization models , respectively referred as “up-the-gradient” and “with-the-flux” . There has been interest to go beyond this dichotomy . Some models attempted to explain phyllotaxis with either purely up-the-gradient [15] or with-the-flux [5] polarization , but they lack biological plausibility or contradict experimental data [16] . Following another approach , Bayer et al . [10] put forward the concept of “dual polarization” , whereby both polarization mechanisms operate concurrently , with a continuous transition from up-the-gradient to with-the-flux polarization depending on local auxin concentration . Although the dual-polarization model displays good agreement with observations , it remains difficult to explain how two qualitatively different mechanisms of PIN1 localization can coexist within the same cells . Shifting attention from Arabidopsis to Brachypodium , O’Connor et al . [17] took advantage of the existence of the PIN1 duplicate sister-of-PIN1 ( SoPIN1 ) proteins in grasses by attributing different roles to the two proteins . However , no such duplication exists in Arabidopsis , where both convergence and canalization are apparently performed by PIN1 protein alone . Moreover , there are fundamental problems with most polarization models [16] . In with-the-flux polarization , it seems unrealistic that cells directly sense net fluxes of auxin through their membranes . Regarding up-the-gradient polarization , it is not clear how a cell could react to the auxin concentration in their neighbors . Several alternative mechanisms have been proposed . Wabnik et al . [18] assumed that auxin gradients in the apoplast are informative enough to drive polarization . Although their model is capable of transitioning between up-the-gradient and with-the-flux regimes , and might thus also account for phyllotaxis , it is yet to be determined whether significant auxin gradients can form in the very narrow spaces between meristematic cells . Abley et al . [19] proposed a model for tissue polarity based on cell-cell coupling through a diffusive mediating molecule . They used it to reproduce several polarization behaviors [20] , but phyllotactic patterns and midveins initiation seem out of reach . The central role of auxin transport as a polarity patterning factor has been questioned in the light of several findings ( see a review in [21] ) . For instance , it has been shown that PIN1 polarities can be oriented by the activity of a transcription factor [9] , or that PIN1 polarities are not disrupted when polar auxin transport is blocked [22 , 23] . As an alternative mechanism , it has been proposed that PIN polarities are influenced by mechanical perturbations [24 , 22 , 25] . A model based on this approach can generate a whorled pattern of auxin maxima [22] , but it is not yet known whether it could reproduce the range of polarizations observed in phyllotaxis . In the present work , we adopt the point of view that the auxin/PIN1 system is the central regulator , with feedbacks from downstream effectors ( transcription factors , cytokinins , mechanics ) . In all up-the-gradient models , PIN polarization is assumed to be fast compared to the production and turnover of PINs , as well as to changes in cellular auxin concentration , so that PIN concentrations at membranes are set to their steady-state values [11] . Since the cycling rates of PINs are completely unknown [16] , this assumption may not be valid . If there is indeed significant latency between changes in auxin concentration and PIN polarization , up-the-gradient models could fail to capture some aspects of convergence point formation . In an attempt to explain how cells can measure the direction and magnitude of auxin fluxes , Coen et al . [26] hypothesized the existence of “tally molecules” produced or consumed at the membrane when auxin enters/exits the cell . The concentration of these molecules at a membrane would act as a proxy for the magnitude of auxin influx and efflux through the membrane . Cieslak et al . [27] presented several biochemically plausible implementations of this concept , assuming that tally molecules modulate PIN allocation to membrane , thus giving rise to a feedback of auxin fluxes on PIN localization . When a local increase in auxin influx decreases the abundance of PIN proteins in the corresponding part of the membrane , canalization patterns emerge . Conversely , when auxin influx locally increases PIN allocation , convergence points form . The transition between these two regimes is controlled by variations in intracellular auxin concentration: An excess of auxin , relative to a threshold concentration , causes a sharp switch of the rate of a reaction occurring inside the membrane . This , in turn , provokes the transition from up-the-gradient to with-the-flux polarization . In addition , thanks to its detailed description of the underlying biochemical reactions , the model does not set PIN concentrations at membranes to their steady-state values , and can thus capture potential transient states during fast polarization events . However , it has been tested only on two-dimensional square grids . In order to explore three-dimensional aspects of auxin patterning during phyllotaxis , we adapt Cieslak’s model to arbitrary irregular 3D tissue geometries , and implement it using the cell complex paradigm . We begin by running the model on a single layer of cells to reproduce convergence point formation . We then investigate how this process is affected by inner cell layers . Finally , we determine under which conditions midveins initiated at convergent points in the epidermal layer can progress in the inner tissue and eventually connect to the already existing vascular system . We show that Cieslak’s mechanism produces dual polarization in three dimensions , and discuss the conditions under which it can happen . Furthermore , in line with previous studies , we highlight the necessity of an additional mechanism for guiding developing veins towards preexisting vasculature .
We adapted to 3D tissues a network of reactions and transport initially formulated by Cieslak and colleagues for two-dimensional grids of square cells ( see figure 9 in [27] ) . The resulting network is represented as a Petri net in Fig 1A ( the system of differential equations derived from this net is detailed in S1 Appendix ) . It is based on the following hypotheses: Since the reaction network used in the model is putative , no experimental values are available for most of the parameters . PIN exo- and endocytosis are established processes , but the associated rates are not known . For parameters already present in Cieslak’s model , we took the same values , except for auxin production ( σa ) , auxin turnover ( μa ) , and the threshold auxin concentration ( ath ) for the shift in the value of νapin . In our 3D model , we had to decrease auxin production , increase auxin turnover , and raise the threshold auxin concentration . Because simulations on 3D templates take very long time , it was not possible to run thousands of simulations to evaluate parameter uncertainties and make a sensitivity analysis . However , we did such an analysis on a 2D square grid to know whether there were correlations between parameters before moving to 3D templates . It turned out that strong correlations were numerous . For instance , the constitutive rate of PIN endocytosis ( μp ) anticorrelates with the rate of APIN spontaneous dissociation ( Tout2 ) , while the latter correlates with the rate of PIN exocytosis by APIN ( σaaux ) . Or the auxin threshold ( ath ) for the shift in the value of νapin anticorrelates with auxin decay ( see S1 Appendix for a full description of correlations ) . However , as explained above , a few parameters had different values in the 3D simulations . Thus , 2D analysis does not fully replace 3D analysis . To complement the 2D analysis , we assessed the sensitivity of the 3D model to two critical parameters: ath and the transition steepness ( see S1 Appendix ) . We constructed a 3D model of meristematic tissue . Each tissue template was built in two steps . First , we tessellated a given volume with truncated octohedra . The truncated octohedron has the advantage of filling space , while having a more complex shape than other convex space-filling polyhedra such as cube and prisms , which makes it more realistic as a meristematic cell . In a second step , the vertexes of our mesh were moved by random amounts to introduce irregularities in the template . Each face of the polyhedral cells represents a discrete element of cell membrane . When two cells share a face , this face defines two discrete elements of membrane ( one for each cell ) and a discrete element of apoplast ( shared by both cells ) . The model was implemented in C++ using the VVe modeling environment , an extension of the VV system [33] . The tissue is represented as a 3D cell complex . The representation and associated topological operations have been adapted from the work of Brisson [34 , 14] . The simulations are visualized using the following graphical convention ( see Fig 1B ) :
We first run our model on a tissue template consisting of 261 cells arranged in a single planar layer with a roughly circular shape . Initial auxin concentration was set to zero in all cells . There was initially no PIN in the cells and very little PIN on membranes . Note that the resulting patterns were independent from these initial conditions , as long as auxin initial concentration was below the auxin threshold . In the first part of the simulation , auxin concentration increased steadily until instabilities occurred and PINs started to polarize . Then , convergence points formed rapidly ( Fig 2A and S1 Movie ) . A visible unrealistic feature in the PIN convergence pattern obtained from this simulation was that some membrane elements with a small area were favored over elements with a larger area in terms of PIN allocation . PINs tended to accumulate on them . This especially happened when two cells with very different auxin concentrations shared only a very small element of cell wall . Then , an auxin flux tended to establish , which promoted PIN allocation to the corresponding membrane element ( s ) . Since the membrane elements of the two adjacent cells were very small , the flux kept a high magnitude , so equilibrium was reached very slowly . In the meantime , membrane-bound PIN concentration reached unrealistically high values . When the membrane elements involved were extremely small , this could even lead to numerical instabilities . That aberration has various causes . First , discretizing cell membranes and walls , as it is commonly done in tissue modeling , introduces some distortions in the way fluxes are modeled . This does not represent such a problem in regular ( square or hexagonal ) cell grids , in which all discrete elements have the same size . Three-dimensionality makes the problem more acute because it spontaneously generates a broader range of areas . Second , it is unrealistic not to set an upper limit to the concentration of membrane-bound PINs since PIN proteins occupy some space on a membrane and thus can not reach too high densities due to steric limitations . A solution could have been to assume that PIN allocation follows some limiting kinetics , for instance Michaelis-Menten . But as this sounded a bit ad hoc , we chose instead to assume that membrane-bound PINs diffuse between adjacent membrane elements of the cell ( Fig 1C; see mathematical details in S1 Appendix ) . Indeed , although some membrane proteins are tethered to the cell wall , it has been shown that PINs are actually diffusing in the membrane [35] . Through diffusion , PINs would move away from higher-density membrane elements to neighboring lower-density elements , and thus could not over-accumulate on a single element . We ran another simulation on the same tissue template , but this time assuming significant lateral diffusion of membrane-bound PIN proteins . We obtained a similar pattern of convergence points , except for the tendency of PINs to accumulate on small membrane elements , which was no longer observed , as expected ( Fig 2B and S2 Movie ) . The PIN lateral diffusion hypothesis did not only eliminate over-accumulation of PINs on small parts of a membrane ( since , as expected , PINs diffused from regions with higher concentrations to adjacent regions with lower concentrations ) , but it also mitigated the distortions caused by artificial membrane discretization , since the separation between membrane elements was made less strict . Furthermore , it made the model of PIN polarization more dynamic: membrane-bound PINs could then also be reallocated through lateral diffusion , not only through endo- and exocytosis . We considered a tissue template with four layers of cells to address the question of how auxin is kept confined in the L1 until venation begins . We ran several simulations on this template , setting a high auxin biosynthesis rate in L1 cells ( except from cells at the border and center , as previously ) and a low auxin biosynthesis rate in inner cells ( L2 , L3 , and L4 ) . We first assumed that only epidermal cells can synthesize PINs , while inner cells are devoid of PINs . Our simulations could not reproduce convergence points in the epidermal layer due to significant leaks of auxin into inner layers . This prevented auxin concentration in the L1 to reach the critical value at which convergence points form . To investigate whether PIN proteins in inner layers could influence auxin patterning in the epidermal layer , we amended our assumption by assigning auxin-dependent PIN biosynthesis to all cells . We could then recover the epidermal patterning ( Fig 3 and S3 Movie ) . This was only possible if the rate of auxin-dependent PIN biosynthesis was high enough , so that PINs were quickly produced in the L2 as soon as auxin was entering L2 cells from the L1 . The newly produced PINs polarized towards the L1 , obeying up-the-gradient polarization , and successfully countered inward auxin flow until convergence points formed . These results point to an active contribution of L2 PINs to auxin patterning in the epidermal layer . This is in line with observations and simulations by Bayer et al , [10] , who also reported upward PIN polarization in the L2 . It should be noted , though , that the convergence pattern obtained with four cell layers is a bit blurred compared with the pattern obtained with a single layer . Using the same template and parameter values as in the previous simulation , we investigated vein initiation . We found that veins developed from convergence points ( Fig 4 and S4 and S5 Movies ) . Very early in the formation of a convergence point , most PIN proteins were allocated to the bottom membrane of the cells in which auxin converged . Auxin flow was thus locally directed towards the L2 and a midvein was initiated . It turned out that the initiation of veins was possible only if auxin importer concentration in inner cell layers was at least as high as in the L1 . This is in contrast to experimental results by Kierzkowski et al . [8] and Bainbridge et al . [36] , who did not observe auxin influx carrier AUX1 nor LAX1 outside the L1 . This suggests the existence of other auxin importers than AUX1 and LAX1 in the inner tissue . In order to investigate how veins are progressing in the inner tissue and potentially connect to the pre-existing vasculature , we modified the previous template by assigning three evenly-spaced cells in the bottom layer ( L4 ) as sinks . Sink cells had a high auxin turnover rate , which was intended to mimic the effect of a functional vein draining auxin . Thus , each sink cell induced a local gradient of auxin concentration centered around itself . In our simulations , when a cell was exporting auxin towards a neighboring cell in the inner tissue , the auxin concentration in the latter cell was increasing until it started to polarize in turn towards a third cell , and so on . Veins developed this way , from convergence points towards regions with lower auxin concentration . Therefore , they progressed globally downward , away from L1 auxin-producing cells ( Fig 5 and S6 , S7 and S8 Movies ) . Often , a vein did not develop as a single sequence of cells , but instead as a bundle of such parallel sequences . Here , lateral diffusion of PIN proteins makes a clear difference . Without lateral diffusion of PINs , each vein formed as a single cell sequence , and this sequence followed preferentially paths connecting cells through very small apoplast element . This was the same effect we had seen in convergence point formation in the epidermal layer , where small membranes were favored over larger ones . Again , this did not fit observations , so lateral diffusion of PINs appears as a necessary feature for modeling venation in a 3D irregular tissue . Since developing veins globally progressed towards lower auxin concentrations , and since sink cells laid at the center of low auxin concentration regions , it was expected that all veins were going to eventually reach a sink . However , local auxin gradients surrounding sinks were very shallow and only had a short-range attraction power on developing veins . Veins which did not pass close to a sink missed their target and got lost in the tissue ( Fig 5C and S8 Movie ) . Therefore , although Cieslak’s mechanism does reproduce several features of vein formation in a 3D tissue , it does not provide an efficient way to find sinks . The poor capacity of developing veins to find sinks in our simulation is suggestive of an additional mechanism for guiding veins towards the already existing vasculature . Bayer et al . [10] reached a similar conclusion . This mechanism could take the form of facilitated diffusion of auxin through plasmodesmata , which has been proven by Smith and Bayer [37] to reliably and robustly connect sources to sinks . Both facilitated diffusion and polar transport could operate in parallel , with facilitated diffusion first defining a preferred route to the closest sink , and polar auxin transport then establishing polarization along this route . But there is little experimental support for facilitated diffusion of auxin in plant tissue . Therefore , following Bayer et al . [10] , we resorted to an alternative model which posits a vein attraction factor ( VAF ) emitted by existing vasculature . In the original model by Bayer et al . [10] , the vein attraction factor diffuses from cell to cell . Every cell can measure VAF concentrations in its neighbors and biases PIN allocation towards the membrane elements facing the cells with highest VAF concentrations . In Cieslak’s model , however , there is no such remote sensing between cells . Its biochemical plausibility rests on the fact that all reactions occur locally , between a cytoplasm and a membrane , or between a membrane and neighboring apoplast . We propose an alternative VAF behavior , which is compatible with the locality of Cieslak’s model . In our scheme , sink cells release VAF in their neighboring apoplast at a fixed rate . The VAF diffuses within the apoplast continuum , never reentering cells . However , it can bind to membranes and unbind from them , at some fixed rates . The concentration of bound VAF on a membrane element favors PIN exocytosis toward this membrane element . We implemented this mechanism ( see details in S1 Appendix ) and found in our simulations that VAF gradients could establish and set a local polarity around every sink cell . The spatial range of VAF-induced polarity depends on VAF production rate and diffusion coefficient . If these values were adequately set , every initiating vein eventually headed and connected to the nearest sink cell ( Fig 6 and S9 , S10 and S11 Movies ) . The VAF hypothesis introduced an additional feature to vein development . Although the cells surrounding a sink had , at the beginning , low auxin and PIN contents , they rapidly polarized towards their neighbor sink . As time went by and VAF diffused farther away , second-order neighbor cells started to polarize towards the sink . When an initiating vein reached a cell already polarized towards a sink , this polarization got reinforced thanks to the high influx of auxin from the vein and the PIN production induced . PIN polarization was then firmly established between the source and the sink . To sum up , the complete vein was the result of the encounter of two opposed movements: the progression of an initiating vein towards a sink , and the expansion of a polarized region centered around this sink . In the initiating vein , cells had high auxin concentrations since each cell had to reach a threshold concentration before it could polarize towards another cell and made the vein progress . But , when the initiating vein reached the polarized region near the sink , auxin could flow rapidly to the sink , without accumulating . Thus , at the moment the polarized path was established , cells in the upper part of the vein had high auxin concentrations while cells closer to the sinks had low concentrations . The former cells kept their high concentration whereas the latter cells only very slowly increased their concentration . To gain further support for the model , we tried to reproduce defective or artificially disturbed phenotypes . As a first example , the pin1 phenotype is characterized by the absence of primordia [38] , In the model , we mimicked this mutation by a two-fold reduction in PIN production ( both constitutive and auxin-dependent ) . Although this alteration seemed moderate , it turned out to be sufficient to inhibit primordium formation ( S12 Movie ) . Most residual PINs in the L1 were oriented toward the L2 , but without formation of distinct veins . As a second example , we tried to recreate the effect of the laser ablation of an incipient primordium . In such experiments , a new primordium forms in the vicinity of the ablated site [39] . In the model , we removed L1 cells of a newly formed primordium and we could observe the rapid emergence of a new primordium next to the ablated one ( Fig 7 and S13 Movie ) . However , unlike what has been reported in ablation experiments [22 , 24] , we did not observe a strong PIN polarization away from the wound . Such a wounding-induced PIN repolarization pattern is believed to be due to altered stress distribution and mediated through microtubules [22 , 24] .
The shoot apical meristem is a challenging system in our understanding of auxin patterning due to the overlap in time and space of different auxin transport regimes in an irreducible three-dimensional geometry . It poses both a conceptual and a technical problem . From a conceptual point of view , the co-localization of up-the-gradient and with-the-flux polarizations in the same cells , with the same efflux carriers , strongly suggests the existence of a common mechanism , while most models so far have addressed each polarization regime separately with incompatible hypotheses . From a technical point of view , building a 3D computational model of plant tissue including both symplast and apoplast , plus polar auxin transport , is a challenge which had never been taken up . As a significant first step towards that objective , we built a model of simplified meristematic tissue on which a biochemically plausible mechanism of auxin transport was implemented . According to this mechanism , the switch between the two polarization regimes depends on the auxin concentration . This approach reproduced key features of 3D phyllotactic patterning and offered new insights in the dynamics of PIN1 polarization in the shoot apical meristems . We emphasize the importance of membrane-bound PIN diffusion in auxin patterning , for both convergence point and canalization . This has been overlooked in 2D models , in which the sizes of the discrete membrane elements are approximately of the same order of magnitude , and usually quite large . But in 3D geometry , the higher complexity of the topology results in a very wide range of areas for discrete membrane elements , with occasionally extremely small values . This compartmentalization becomes then overly artificial and ignores the intrinsic continuity of the lipid cell membrane in which bound PIN proteins can move laterally . It leads to aberrant accumulation of PINs on a few very tiny membrane elements . Assuming lateral diffusion of PIN proteins partially overcomes this issue by smoothing the barriers between neighbor discretized elements of the same cell membrane . It should be noted , however , that membrane-bound PINs display a more complex behavior , due to mechanisms limiting their lateral movements [40] . They tend to form clusters in which lateral mobility is strongly reduced . Only a minor fraction of membrane-bound PINs are unclustered and laterally mobile [41] . Future models could investigate in more details the significance of PIN clustering for cell polarization at tissue scale . We demonstrated that inner cell layers , especially the L2 , significantly contribute to auxin patterning in the L1 . The higher auxin biosynthesis rate in the epidermal layer is not sufficient by itself to reach the critical concentration required for the formation of convergence points . In the early stages of phyllotactic patterning , auxin has to be kept confined in the L1 . We showed that this confinement can be performed by up-the-gradient polarization in the L2 , where PIN proteins polarize upward , towards auxin-rich L1 cells . A necessary condition is that auxin-dependent PIN production is fast enough to react quickly to the first auxin leaks into the L2 . Bayer et al . [10] assumed reduced symplasmic communication between L1 and inner tissues , based on experimental evidence [42] . We did not make that assumption and found that PIN polarization alone can efficiently confine auxin in the L1 until midvein initiation . We also found that mutual influences between cell layers have a slight but non-negligible impact on the final pattern . This suggests that precise and realistic descriptions of surface patterning should take into account effects from the underlying tissue . It turned out that midveins cannot develop if the concentration of auxin influx carriers is lower in the inner tissue than in the L1 . This seemingly contradicts experimental results [8 , 36] showing that auxin influx carriers AUX1 and LAX1 are almost exclusively present in the epidermal layer . But at the same time , Bainbridge et al . [36] reported the expression of another auxin importer , LAX2 , in the initiating vein . LAX2 has also been shown to regulate vascular patterning in cotyledons [43] . Our model does not exclude the possibility that auxin influx carriers present in the inner tissue could be restricted to the cells forming the initiating veins . However , in this case , they have to be expressed from the very beginning of vein initiation to make it possible . Since the mechanism inducing LAX2 expression is not known , amending the model in this direction seems both speculative and premature . Our investigation also pointed out that auxin gradients induced by the existing vascular system are too shallow to efficiently attract initiating veins . This fact had been already noted in 2D simulations performed by Bayer et al . [10] , and is linked to the more general problem of establishing polarity within a tissue . Various models of tissue polarity have been developed ( see a comparative study in [20] ) . In the shoot apical meristem of Arabidopsis , the region in which a polarity has to be established to guide an initiating vein is relatively small . Therefore , a simple mechanism based on a putative diffusive molecule is sufficient , as already outlined by Bayer et al . [10] . We proposed a detailed process for the action of such a Vein Attraction Factor . We assumed that the VAF is transported apoplastically , which seemed to us more realistic than the symplastic transport first proposed by Bayer et al . [10] . This vein attraction mechanism raises new questions about source-sink connections . Classically , an initiating vein is viewed as progressing in a tissue from a source until it reaches a sink . In our simulations , each sink builds up an expanding polarity field centered around itself . When the progressing vein meets the expanding polarity field , the connection is almost fully established . Then , the vein progresses very quickly and directly to the sink , locally increasing polarity in its wake . The relative contribution of each process—free progression of the initiating vein and expansion of the polarity field—depends on the biosynthesis rate and diffusion coefficient of the VAF , but also on the precise timing of vein initiation and VAF release . Further experimental studies are needed to improve our understanding of how primordia connect to the vascular system . Modeling efforts are hindered by the unknown nature of the putative VAF . Another obvious limitation to our model is that it does not reproduce phyllotactic patterns . This is essentially because it does not include tissue growth , which plays a crucial role in the dynamics of phyllotactic patterning . However , simulating meristem growth with mechanical processes would add another layer of technical complexity and requires much more development and computational resources . Making the size and shape of the template closer to an actual meristem is also a necessary improvement for future models to gain more realism . It can also be argued that the PIN polarization mechanism used in our model is questionable since there is no evidence for auxin complexes . Yet , the core of the mechanism is more generic than it looks at first sight . Cieslak et al . [27] proposed various other implementations of it , reflecting different biochemical assumptions . Since the precise reactions underlying the feedback of auxin transport on PIN localization are unknown , it is difficult to discriminate between these implementations . Other ones can be designed based on new experimental findings . For instance , the fact that convergence point formation and vein initiation seem to be associated with two distinct groups of auxin influx carriers could be exploited to amend the current reaction network . Other ingredients could be introduced , such as the inhibiting effects of auxin on PIN endocytosis [44] and PIN vacuolar degradation [45 , 46] . Alternatives will probably emerge in the near future from experimental studies , and 3D models like ours will be valuable tools to thoroughly assess their explanatory powers in terms of auxin pattern formation .
|
The regularity of leaf arrangement around stems has long puzzled scientists . The key role played by the plant hormone auxin is now well established . On the surface of the tissue responsible for leaf formation , auxin accumulates at several points , from which new leaves eventually emerge . Auxin also guides the progression of new veins from the nascent leaves to the vascular system of the plant . Models of auxin transport have been developed to explain either auxin accumulation or auxin-driven venation . We propose the first three-dimensional model embracing both phenomena using a unifying mechanism of auxin transport . This integrative approach allows an assessment of our present knowledge on how auxin contributes to the early development of leaves . Our model reproduces many observations of auxin dynamics . It highlights how the inner and epidermal tissues act together to position new leaves . We also show that an additional , yet unknown , mechanism is required to attract new developing veins towards the main vasculature of the plant .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] |
[
"plant",
"anatomy",
"apoplastic",
"space",
"cell",
"processes",
"hormones",
"plant",
"science",
"plant",
"hormones",
"cellular",
"structures",
"and",
"organelles",
"leaf",
"veins",
"leaves",
"biophysics",
"exocytosis",
"cell",
"membranes",
"physics",
"biochemistry",
"plant",
"biochemistry",
"biochemical",
"simulations",
"cell",
"biology",
"secretory",
"pathway",
"biology",
"and",
"life",
"sciences",
"auxins",
"biosynthesis",
"computational",
"biology",
"physical",
"sciences",
"biophysical",
"simulations"
] |
2019
|
Toward a 3D model of phyllotaxis based on a biochemically plausible auxin-transport mechanism
|
Tropical infectious diseases like dengue , scrub typhus , murine typhus , leptospirosis , and enteric fever continue to contribute substantially to the febrile disease burden throughout Southeast Asia while malaria is declining . Recently , there has been increasing focus on biomarkers ( i . e . C-reactive protein ( CRP ) and procalcitonin ) in delineating bacterial from viral infections . A prospective observational study was performed to investigate the causes of acute undifferentiated fever ( AUF ) in adults admitted to Chiangrai Prachanukroh hospital , northern Thailand , which included an evaluation of CRP and procalcitonin as diagnostic tools . In total , 200 patients with AUF were recruited . Scrub typhus was the leading bacterial cause of AUF ( 45/200 , 22 . 5% ) followed by leptospirosis ( 15/200 , 7 . 5% ) and murine typhus ( 7/200 , 3 . 5% ) , while dengue was the leading viral cause ( 23/200 , 11 . 5% ) . Bloodstream infections contributed to 7/200 ( 3 . 5% ) of the study cohort . There were 9 deaths during this study ( 4 . 5% ) : 3 cases of scrub typhus , 2 with septicaemia ( Talaromyces marneffei and Haemophilus influenzae ) , and 4 of unknown aetiologies . Rickettsioses , leptospirosis and culture-attributed bacterial infections , received a combination of 3rd generation cephalosporin plus a rickettsia-active drug in 53% , 73% and 67% of cases , respectively . Low CRP and white blood count were significant predictors of a viral infection ( mainly dengue ) while the presence of an eschar and elevated aspartate aminotransferase and alkaline phosphatase were important predictors of scrub typhus . Scrub typhus and dengue are the leading causes of AUF in Chiangrai , Thailand . Eschar , white blood count and CRP were beneficial in differentiating between bacterial and viral infections in this study . CRP outperformed procalcitonin although cut-offs for positivity require further assessment . The study provides evidence that accurate , pathogen-specific rapid diagnostic tests coupled with biomarker point-of-care tests such as CRP can inform the correct use of antibiotics and improve antimicrobial stewardship in this setting .
Acute undifferentiated fever ( AUF ) remains the leading cause of hospitalisation among adults and children in urban and rural regions of Southeast Asia . The causes include common diseases such as dengue , scrub typhus , murine typhus , leptospirosis , and enteric fever , which continue to contribute significantly to the febrile disease burden [1–4] . Although malaria may present similarly , its overall incidence and impact on health in this region is declining [5] . In Laos , a prospective multicentre study investigating the causes of non-malarial fever revealed dengue , scrub typhus , Japanese encephalitis and leptospirosis as the major aetiologies in hospitalised adults and children once influenza was excluded [6] . In rural Thailand , dengue , scrub typhus , leptospirosis , murine typhus , and influenza have been identified as the most common causes of AUF among adults and children [4 , 7] . Scrub typhus , enteric fever , flavivirus infection , leptospirosis and malaria were the main causes of fever in adults and children in the 1970s in rural Malaysia [8] . In febrile pregnant women on the Thai-Burmese border and in Laos , malaria , rickettsial infections , dengue , leptospirosis , typhoid and pyelonephritis predominate [9 , 10] . Adverse neonatal and maternal outcomes were high in this group , particularly in those diagnosed with rickettsial infections [10 , 11] . In Cambodian children , dengue , scrub typhus , bacteraemia ( Salmonella enterica serovar Typhi was the commonest pathogen ) and Japanese encephalitis were the major diagnoses [3] . These “causes-of-fever” studies that address a wide range of infectious diseases in diverse geographies are useful in informing clinicians and epidemiologists alike [12] . However , the majority of currently available fever studies suffer from selection bias , often rely on suboptimal diagnostic tools and non-uniform positivity criteria—limiting estimates of disease incidence and burden [13] . Performing these prospective studies correctly is costly , difficult and challenging–especially if representative geographical coverage is desired [12] . The current literature highlights a panel of AUF that represents the leading causes of fever in Asia and similarities in their clinical presentation and poor access to high-quality , affordable diagnostic tools frequently result in sub-optimal management [14] . Although progress in developing accurate , validated and cost-effective diagnostic tools for non-malarial pathogens , such as disease-specific rapid diagnostic tests ( RDTs ) , have been made ( e . g . combining NS1-antigen with IgM detection in dengue ) , recent modelling approaches suggest that testing for viral infections is unlikely to be cost-effective when considering direct health benefits , whereas RDTs for the detection of prevalent bacterial pathogens could be [15 , 16] . Biomarkers such as C-reactive protein ( CRP ) and procalcitonin have some utility in delineating between bacterial and viral infections and guiding healthcare workers on the appropriate use of antibiotics in patients with respiratory tract infections in high income settings [17] . A retrospective study based on well-characterised samples of adults and children with febrile illnesses from Cambodia , Laos and Thailand demonstrated CRP was highly sensitive and moderately specific for discriminating between bacterial and viral infections [18] . Recently , CRP testing has been shown to reduce antibiotic prescription for acute respiratory illnesses in adults and children in primary healthcare settings in Vietnam [19] . In resource-constrained tropical settings , common treatable infections are being missed and inappropriate use of antibiotics is widespread . This highlights the potential impact of CRP RDTs on the precision of antibiotic use and contribution to the global strategy to combat antimicrobial resistance . In this prospective study , we investigated the causes of AUF in adults admitted to the provincial hospital in Chiangrai , northern Thailand , and evaluated the use of CRP and procalcitonin tests in guiding appropriate antibiotic use .
The Chiangrai Hospital Ethical Committee , Thai Ministry of Public Health and the Faculty of Tropical Medicine Ethics Committee , Mahidol University , Bangkok , granted ethical approval for this study ( MUTM 2006–035 ) . All patients provided written informed consent prior to sample collection , and parents or guardians provided informed consent on behalf of all child participants . Chiangrai Prachanukroh hospital is located in Chiangrai province , the northernmost province in Thailand , and near “the Golden Triangle” where Thailand , Laos and Myanmar converge . The province population of 1 . 2 million consists mainly of ethnic Thais with 12 . 5% belonging to hill tribes and other minority ethnic groups . Between August 2006 and October 2008 , we prospectively recruited a total of n = 231 patients age ≥15 years old at Chiangrai Prachanukroh hospital with a fever >37 . 5°C or a history of fever within the past 21 days , no evidence of a primary focus of infection ( e . g . consolidation on chest X-ray , symptoms and signs of a urinary tract infection , cellulitis ) and negative for malaria on blood film . Demographic , clinical and laboratory data related to the admission were collected individually on study case-record forms ( CRFs ) from patient notes and hospital records . Demographic data included age , sex , and occupation . A rural/agricultural occupation was defined as those working as farmers , gardeners , agricultural/plantation workers , or fish and animal farm workers . Clinical data included symptoms , examination findings and vital signs on admission along with details of the current illness , prior antibiotic use , antibiotic treatment during admission , and illness outcome ( e . g . fever days , death ) . Laboratory data included haematology ( complete blood count ) and biochemistry ( renal and liver blood tests ) results from admission samples . Chest x-ray findings were also recorded if performed . An acute study blood sample was collected by study staff on enrolment in addition to the routine tests requested by the treating physician ( 10ml EDTA whole blood and 10ml clotted blood for serum ) . Blood and other routine cultures were performed if requested by the local clinician and processed using conventional techniques at the hospital microbiology laboratory . HIV testing was performed as part of routine hospital work using RDTs at the discretion of the treating physician . Follow-up was carried out by study staff 7–14 days after enrolment and involved a clinical review and collection of a convalescent blood sample ( 10ml clotted blood for serum ) . There were 19 patients with incomplete CRFs/datasets and 12 patients with incomplete sample collections . These 31 patients were excluded resulting in a total of 200 study eligible patients . Of these , 171/200 ( 86% ) provided paired samples obtained on admission and follow-up between days 7–14 , and 29/200 patients had a confirmatory diagnosis made from admission samples alone . Both admission and follow-up samples were used for the diagnostic assays outlined below . Inflammatory biomarkers were tested on acute samples only . The clotted blood samples were processed for serum , aliquoted , stored locally at -30°C , and batch transported on dry-ice for storage ( -80°C ) and subsequent analysis at the central laboratory of Mahidol-Oxford Tropical Medicine Research Unit ( MORU ) in Bangkok . EDTA whole blood samples were transported at ambient temperature on the day of collection to Bangkok for further analysis . Some whole blood samples were processed immediately for culture for leptospirosis and scrub typhus ( see below ) with the remainder stored as aliquots of whole blood , plasma and buffy coat at -80°C . In addition , meteorological data comprising average monthly temperatures and total monthly rainfalls were retrospectively collected for the study period from the local Thai Meteorological Department office of Mueang district in Chiangrai province . The data was collected from the district’s weather station near the airport . Chiangrai Prachanukroh Hospital is located within this central district , which is its main catchment area , but the hospital also admits severely ill patients from surrounding districts . The diagnostic panel included diagnosis of dengue , scrub typhus , murine typhus , leptospirosis and Japanese encephalitis . Dengue diagnosis was performed in paired sera using the following ELISA tests: PanBio Dengue Early NS1 ( Alere ) , PanBio Dengue IgM capture ( Alere ) , PanBio Dengue IgG capture ( Alere ) , and PanBio Japanese Encephalitis/Dengue IgM combo ( Alere ) . An admission titer ≥10 U of NS1 PanBio units and/or ≥4-fold increase of IgM antibodies in the convalescent sample was considered diagnostic of acute primary dengue virus infection . Patients with anti-JEV IgM levels of >40 U were classified as having acute JEV infections only if anti-dengue IgM levels were <40 U using the combination ELISA test . Leptospirosis culture was performed at MORU within 24–48 hours by injecting 100μL of whole blood and 200μL of plasma sediment ( the bottom fraction obtained from centrifuging 500μL of heparinized plasma ) into 3 mL of Ellinghausen , McCullough , Johnson , and Harris ( EMJH ) medium , supplemented with 3% rabbit serum and 0 . 1% agarose . Both culture tubes were incubated aerobically at 25°C–30°C and examined every week for 3 months for evidence of growth . The leptospirosis SD Bioline RDTs were used for detecting anti-leptospira IgM and IgG . Scrub typhus and murine typhus were diagnosed using the indirect immunofluorescence assay ( IFA ) to detect IgM antibody titers in paired sera ( or in admission samples only if convalescent samples unavailable ) against Orientia tsutsugamushi antigens ( Karp , Kato and Gilliam strains for scrub typhus ) and Rickettsia typhi antigens ( Wilmington strain for murine typhus ) , respectively . The new diagnostic IFA cut-off titer of ≥1:3 , 200 in an admission sample or ≥4-fold rise to ≥1:3 , 200 in a convalescent-phase sample was used [20] . For scrub typhus , culture and polymerase chain reaction ( PCR ) assays were also performed as previously described [21] . Briefly , the PCR assays included conventional PCR assay to detect the 56kDa gene and real-time PCR assays to detect the 47kDa htra and groEL genes . To fulfil the PCR criteria for diagnosis , a consensus of two out of three PCR assays was required . The inflammatory biomarker procalcitonin was measured by the ELISA-based VIDAS PCT kit with a detection range of 0 . 05-195ng/ml ( BioMérieux , France ) , and CRP serum levels were measured with the NycoCard Reader II ( Axis Shield , Norway ) , with a detection range of 5-150mg/L in serum [22 , 23] . Testing was performed on admission samples and two independent operators , blinded to the microbiological diagnoses , performed the procalcitonin and CRP assays in duplicate . Control reagents were provided with each test kit and calibration performed as per manufacturers’ instructions . The following thresholds were evaluated for their usefulness in predicting bacterial causes of fever; for procalcitonin 0 . 25ng/mL and 0 . 5ng/mL , and for CRP 20mg/L and 40mg/L plasma levels upon admission , respectively [24–26] . The diagnostic results were considered in relation to each other , and a final diagnosis was attributed to each case by the strength of evidence supporting each diagnosis , as previously described; ( I ) PCR/antigen/culture positivity > ( II ) dynamic serology ( 4-fold rise ) > ( III ) single titer and/or unjustified serological cut-off titer [27] . Blood , urine , sputum and stool culture results from admission were collected from the hospital reporting system if performed . A final conservative diagnosis of culture-attributed infection ( CAI ) was made on the balance of clinical information , haematological and biochemical results , and results of our diagnostic panel . Proportions , percentages and averages ( median and interquartile range [IQR] or mean and standard deviation [SD] ) were calculated controlling for any missing data . Seasonality was assessed by calculating proportions of patients ( and 95% confidence intervals ) admitted during discrete time-periods and assessing for overlap as well as performing two-sample tests of proportions . Univariate and multivariate logistic regression analysis were performed to determine predictor variables independently associated with the outcomes ( e . g . viral/bacterial/unknown aetiologies or specific diagnoses such as scrub typhus or dengue ) . Categorical data were analysed using Pearson’s Chi-squared test or Fisher’s exact test as appropriate where specified . Comparisons of receiver operating characteristic ( ROC ) curves evaluated the sensitivity , specificity and likelihood ratios for procalcitonin and CRP in differentiating between bacterial and viral aetiologies . Classification and regression trees were generated for scrub typhus and dengue using Salford Predictive Modeler Software Suite v8 . 2 ( Salford Systems , San Diego , CA , USA ) . Other analyses were performed using STATA 14 software ( College Station , Texas , USA ) .
Our study cohort of 200 adult patients with AUF admitted to Chiangrai Prachanukroh hospital between August 2006 and October 2008 was predominantly male ( 114/194 , 58 . 8% ) , had a median age of 41 ( IQR 29–52 ) , and most had a rural/agricultural occupation ( 64/136 , 47 . 1% ) . 34/200 patients ( 17% ) received antibiotic therapy prior to admission to the provincial hospital and the median number of days from onset of fever to admission was 4 ( IQR 3–7 ) . 77/200 patients ( 38 . 5% ) had a bacterial aetiology for their fever , 24/200 ( 12% ) a viral aetiology , and 97/200 ( 48 . 5% ) had an unknown aetiology ( the 2 remaining patients were diagnosed with invasive fungal infection , details below ) . Scrub typhus was the leading bacterial cause of AUF with 45/200 ( 22 . 5% ) , followed by leptospirosis with 15/200 ( 7 . 5% ) and murine typhus 7/200 ( 3 . 5% ) , while dengue was the leading viral cause with 23/200 ( 11 . 5% ) and there was a solitary JEV patient ( 0 . 5% ) . A total of 12/200 ( 6% ) cases had multiple positive tests ( 11 dual , 1 triple ) that required scrutinizing with the criteria described above . Anti-JEV IgM positive cases were superseded by scrub typhus PCR positivity +/- dynamic serology in three cases and dengue NS1 antigen +/- dynamic serology in four cases . One case had weakly positive scrub typhus PCR for a single target ( 2 out of 3 targets required to fulfil the diagnostic criteria ) with negative serology and was superseded by dengue NS1 antigen and IgM positivity . Two leptospirosis RDT positive cases were overruled by scrub typhus PCR-positivity in one case and dynamic murine typhus serology in the other . One case with dynamic rise in anti-dengue IgM but negative NS1 antigen was assigned a diagnosis of scrub typhus on the basis of PCR-positivity and dynamic serology . Finally , one case with positive leptospirosis RDT and anti-dengue IgM dynamic serology with negative NS1 antigen was diagnosed with scrub typhus on the basis of positive PCR assays . 142/200 ( 71% ) patients had blood cultures performed of which , 126 were reported as no growth , 9 had microbiologically non-significant growth ( mainly Gram positive organisms e . g . coagulase-negative staphylococci , aerobic spore bearers ) , and 7 had microbiologically significant growth ( 3 . 5% ) . Blood culture findings included 2 Talaromyces marneffei , 1 Haemophilus influenzae , 1 Staphylococcus aureus , 1 Burkholderia pseudomallei , 1 Escherischia coli , and 1 Enterococcus faecium . The patients with talaromycosis and Haemophilus influenzae bacteraemia tested positive for HIV antibodies using in-house RDTs . In addition , there were 2 significant urine cultures ( heavy growth of E . coli ) , 2 significant sputum cultures ( Klebsiella pneumoniae in patients with severe respiratory syndromes ) , and 1 significant stool culture ( Salmonella spp . ) . In summary , there were 12 additional diagnoses in the culture-attributed infections group ( CAI ) , 10 due to bacteria and 2 due to fungi . Table 1 summarises the characteristics of patients in the viral , bacterial and unknown aetiology groups . Patients who were younger ( OR 0 . 966 , 95%CI 0 . 937–0 . 996 , p = 0 . 026 ) , had lower CRP ( OR 0 . 967 , 95% CI 0 . 953–0 . 981 , p = 0 . 000 ) , lower white blood count ( OR 0 . 713 , 95%CI 0 . 615–0 . 828 , p = 0 . 000 ) , lower neutrophil count ( OR 0 . 694 , 95%CI 0 . 586–0 . 822 , p = 0 . 000 ) or higher haemoglobin ( OR 1 . 259 , 95%CI 1 . 023–1 . 549 , p = 0 . 029 ) were significantly more likely to be diagnosed with a viral aetiology on univariate logistic regression analyses . Only low CRP ( aOR 0 . 972 , 95%CI 0 . 957–0 . 987 , p = 0 . 000 ) and low white blood count ( aOR 0 . 573 , 95%CI 0 . 331–0 . 992 , p = 0 . 047 ) remained as significant predictors for viral infection on multivariate logistic regression analysis . Significant predictor variables for bacterial infection on univariate analyses included the presence of an eschar ( OR 11 . 74 . , 95%CI 3 . 849–35 . 807 , p = 0 . 000 ) and a higher lymphocyte count ( OR 1 . 366 , 95%CI 1 . 027–1 . 816 , p = 0 . 032 ) but only the eschar remained a significant predictor on multivariate analysis ( aOR 11 . 590 , 95%CI 3 . 754–35 . 784 , p = 0 . 000 ) . The finding of an eschar within the bacterial aetiology group was almost exclusively seen in patients diagnosed with scrub typhus ( 21/22 , 95 . 5% ) , the exception being one patient with Staphylococcus aureus bacteraemia ( 1/22 , 4 . 5% ) . Significant predictor variables for the unknown aetiology group are shown in Table 1 but are clinically less useful . Details of univariate and multivariate analyses of the predictor variables in Table 1 can be found in S1 Table . When comparing the viral and bacterial aetiology groups directly ( excluding unknown group ) , eschar , CRP , Hb , WBC , neutrophil count and lymphocyte count were significant variables on univariate analyses . A lower CRP ( aOR0 . 969 95%CI 0 . 951–0 . 987 , p = 0 . 001 ) was an important predictor for viral infection while presence of an eschar ( completely absent in the viral group ) and a higher CRP ( aOR1 . 032 95%CI 1 . 014–1 . 052 , p = 0 . 001 ) remained as significant predictor variables for bacterial infection on multivariate analysis . For a breakdown of demographics , symptoms and signs , chest x-ray findings and detailed laboratory results for patients diagnosed with scrub typhus , dengue , leptospirosis and murine typhus , please refer to S2 Table . Significant predictors for scrub typhus on multivariate logistic regression analysis included the presence of an eschar ( aOR 42 . 408 , 95%CI 4 . 956–362 . 905 , p = 0 . 001 ) , a higher lymphocyte count ( aOR 2 . 063 , 95%CI 1 . 146–3 . 713 , p = 0 . 016 ) , and elevated aspartate aminotransferase ( AST , aOR 1 . 014 , 95%CI 1 . 004–1 . 023 , p = 0 . 004 ) and alkaline phosphatase ( ALP , aOR 1 . 004 , 1 . 000–1 . 008 , p = 0 . 036 ) . For dengue , a lower CRP ( aOR 0 . 956 , 95%CI 0 . 927–0 . 986 , p = 0 . 005 ) was the only consistently significant predictor variable on multivariate analysis . Elevated creatinine was significantly associated with leptospirosis on univariate analysis ( OR 1 . 132 , 95%CI 1 . 001–1 . 279 , p = 0 . 048 ) but was not significant in multivariate analysis . Details of the analysis can be found in S3 Table . In addition , classification and regression trees ( CART ) were generated for scrub typhus ( S1 Fig , panel A ) and dengue ( S1 Fig , panel B ) which revealed a similar set of significant variables when compared with the multivariate logistic regression analyses above . The presence of an eschar , ALP>289IU/L and AST>88IU/L were used as decision nodes for scrub typhus while CRP≤37mg/L and WBC≤7 . 9x103/mm3 were used for dengue virus . The majority of cases occurred during the months of June to November , coinciding with the rainy and early winter seasons . Proportions of patients ( 95% confidence intervals ) admitted from June to November and from December to January were calculated for the study cohort , scrub typhus , dengue , leptospirosis , murine typhus , CAI and unknown groups: total 0 . 82 ( 0 . 77–0 . 88 ) :0 . 18 ( 0 . 12–0 . 23 ) p<0 . 001 , scrub typhus 0 . 91 ( 0 . 83–0 . 99 ) :0 . 09 ( 0 . 01–0 . 17 ) p<0 . 001 , dengue 0 . 96 ( 0 . 87–1 . 00 ) :0 . 04 ( 0 . 00–0 . 13 ) p<0 . 001 , leptospirosis 0 . 80 ( 0 . 60–1 . 00 ) :0 . 20 ( 0 . 00–0 . 40 ) p = 0 . 001 , murine typhus 0 . 57 ( 0 . 20–0 . 94 ) :0 . 43 ( 0 . 06–0 . 80 ) p = 0 . 595 , CAI 0 . 83 ( 0 . 62–1 . 00 ) :0 . 17 ( 0 . 00–0 . 38 ) p = 0 . 04 and unknown 0 . 77 ( 0 . 68–0 . 86 ) :0 . 23 ( 0 . 14–0 . 32 ) p<0 . 001 . Apart from murine typhus , there were no overlaps of 95% confidence intervals . To illustrate further , scrub typhus and dengue cases were plotted against time along with average monthly temperatures and total monthly rainfall for the district ( Fig 1 ) . A total of 9 deaths were recorded during this study . Three patients had a diagnosis of scrub typhus ( 3/45 , 6 . 7% ) , two patients had bloodstream infections ( Talaromyces marneffei–previously Penicillium marneffei–and Haemophilus influenzae , both HIV positive cases ) , while 4 patients had unknown aetiologies ( 4/97 , 4 . 1% ) . The 9 patients consisted of 5 males and 4 females , with a median age of 44 ( IQR 40–51 ) , 7 worked in agriculture , none had received pre-admission antibiotics and all were treated with antibiotics upon admission with the exception of 1 patient who died soon after presentation . The median number of fever days prior to hospital admission was 4 ( IQR 3–6 ) and the number of days admitted was 3 ( IQR 2–4 ) . The majority were febrile on or after admission ( 8/9 , 88 . 9% ) and had neurologic ( 7/8 , 87 . 5% ) , respiratory ( 5/8 , 62 . 5% ) , gastrointestinal ( 5/8 , 62 . 5% ) or severe disease ( 5/8 , 62 . 5% ) . The median ( IQR ) CRP , PCT , WBC , neutrophil count values for this sub-group were 150 mg/L ( 149–150 ) , 37 . 2 ng/mL ( 2 . 0–59 . 7 ) , 10 . 2x103/mm3 ( 8 . 3–14 . 8 ) and 9 . 3x103/mm3 ( 6 . 9–12 . 2 ) , respectively . 169 out of 200 patients ( 84 . 5% ) received antibiotics during the study ( pre-admission and/or during admission ) . Of the 31 patients who did not receive any antibiotics , 6 had a viral infection ( exclusively dengue ) , 7 had a bacterial infection ( 5 scrub typhus , 1 leptospirosis , and 1 bacteraemia ) and 18 had an unknown aetiology . For monotherapy , ceftriaxone was the most commonly used antibiotic ( 131/169 , 77 . 5% ) followed by doxycycline ( 118/169 , 69 . 8% ) and chloramphenicol ( 26/169 , 15 . 4% ) . Use of combination antibiotic therapy was common and particularly applied to patients during their in-patient stay 105/168 ( 62 . 5% ) compared to those who received antibiotics prior to admission 9/34 ( 26 . 5% ) . Ceftriaxone and doxycycline was the most commonly used combination with 79/169 ( 46 . 7% ) patients receiving this therapy . Eighteen of twenty four patients ( 75% ) with a viral diagnosis received antibiotics while patients with a bacterial diagnosis and those with an unknown aetiology received antibiotics in 93 . 3% ( 70/75 ) and 92 . 9% ( 79/85 ) of cases , respectively . Among patients with a diagnosis of scrub or murine typhus , 82 . 4% ( 42/51 ) received anti-rickettsial antibiotics ( mainly doxycycline or chloramphenicol ) , which meant 17 . 6% of patients ( 9/51 ) received antibiotics ineffective against both diseases . In contrast , 93 . 3% of patients ( 14/15 ) with leptospirosis received appropriate treatment ( ceftriaxone +/- doxycycline ) . An overview of antibiotic use is shown below in Table 2 . Nevertheless , the strategy of combining a beta-lactam with doxycycline was often used , and 53% , 73% and 67% of patients with a rickettsiosis , leptospirosis and culture-attributed bacterial infection received an antimicrobial treatment regimen combining a third generation cephalosporin with a rickettsia-active drug , respectively . Ceftriaxone monotherapy was most commonly used for leptospirosis and bacterial causes , while doxycycline monotherapy was commonly used for the rickettsial/dengue subgroups . CRP on admission was a significant predictor variable for the viral aetiology group ( low CRP ) when analysing the whole AUF cohort . When the unknown group was excluded , it remained an important predictor for the viral ( low CRP ) and bacterial ( high CRP ) groups . 92% and 86% of bacterial cases had CRP levels above the pre-defined cut-offs of >20mg/L and >40mg/L , respectively . For the viral aetiology group , 73% and 86% of cases had CRP levels below these cut-offs , respectively . The >20mg/L and >40mg/L CRP cut-offs correctly identified 87 . 2% and 86 . 2% of bacterial and viral cases , respectively . The CRP and PCT results are summarised in Table 3 . The optimal CRP plasma level cut-off to accurately distinguish between bacterial and viral causes for fever in this study was calculated to be >36mg/L [sensitivity 88 . 9% ( 95%CI 79 . 3–95 . 1 ) and specificity 86 . 4% ( 95%CI 65 . 1–97 . 1 ) ] , with 88 . 3% of cases correctly identified . If we compare the choice of CRP cut-offs according to available CRP assays of 20mg/L and 40mg/L , then using the 40mg/L cut-off would provide an improved balance between sensitivity and specificity , with a higher specificity than the lower cut-off of 20mg/L . On excluding the unknown aetiology group , PCT was good at defining bacterial cases , but poor at selecting for viral aetiologies , which is reflected by the poor specificity values . The higher cut-off at 0 . 50ng/mL improved specificity from 40 . 9 to 63 . 6 when compared to 0 . 25ng/mL , and was accompanied with a moderate drop in sensitivity and a minor reduction in the proportion of correctly identified cases . If a higher cut-off of 0 . 7ng/mL for PCT was chosen , sensitivity will fall slightly while the specificity will increase , but with the same number of correctly identified cases [sensitivity 79 . 2% ( 68 . 0–87 . 8 ) , specificity 68 . 2% ( 45 . 1–86 . 1 ) , correctly identified cases 76 . 6%] . Receiver operating characteristic ( ROC ) curves were generated to visualise the performance of CRP ( S2 Fig , panel A ) and PCT ( S2 Fig , panel B ) in differentiating bacterial versus viral infections for specified cut-off values . The areas under the ROC curve were 0 . 91 ( 0 . 85–0 . 96 , 95% CI ) and 0 . 80 ( 0 . 72–0 . 88 , 95% CI ) for CRP and PCT , respectively .
Scrub typhus was the leading cause of AUF followed by dengue , leptospirosis , murine typhus , and bloodstream infections ( 22 . 5% , 11 . 5% , 7 . 5% , 3 . 5% and 3 . 5% , respectively ) in this study . The incidence of both scrub typhus and dengue exhibited pronounced seasonality and were more common in the rainy season through to early winter ( June to November ) . Similar to previous studies , the clinical finding of an eschar was strongly associated with the diagnosis of scrub typhus and represents a useful diagnostic clue [7 , 28 , 29] . However , eschars are not always present in scrub typhus patients , and their formation can be influenced by the degree of past exposure to various Orientia tsutsugamushi strains and the presence of strain-specific immunity [30] . Previous studies on paediatric scrub typhus in northern Thailand reported the presence of an eschar in approximately 70% of children [31 , 32] , while only 7% of children from Songkhla , southern Thailand and 7% of adults from Udon Thani , north-eastern Thailand with scrub typhus were reported to have an eschar [33 , 34] . Whether this represents the spectrum of regular re-exposure to circulating Orientia tsutsugamushi strains in these regions remains to be determined in longitudinal studies . Five patients presented with eschars but tested negative for scrub typhus . One patient had Staphylococcus aureus bacteraemia while the other four patients were in the unknown aetiology group . Additional testing of samples from one of these four patients revealed one 17kDa qPCR positive blood sample suggestive of Rickettsia spp . As such , alternative causes for febrile patients presenting with an eschar , such as spotted fever group rickettsial infections , should be considered . It is important to note that true eschars are completely painless–a central feature to distinguish them from eschar-like lesions including spider and ( manipulated ) insect bites which are typically itchy and/or painful [35] . In addition , we have shown that elevated hepatic enzymes ( ALP and AST ) were important predictors of scrub typhus in patients admitted with AUF on multivariate logistic regression and classification and regression tree ( CART ) analyses ( ALP>289IU/L and AST>88IU/L ) . Raised hepatic enzymes have previously been described in scrub typhus observational studies in northern Thailand and India although not in cause of fever studies [36–38] . The overall mortality rate in our study cohort was 4 . 5% ( total of 9 deaths ) of which a third were attributable to scrub typhus . The scrub typhus mortality rate of 6 . 7% ( 3/45 ) was comparable to previous reports of untreated disease , as summarised in a recent systematic review , but was lower than the previously reported mortality from northern Thailand of 13 . 1% from 2004–2010 , possibly reflecting better awareness and treatment decisions [36 , 39] . The majority of patients ( 84 . 5% ) received empirical antibiotic treatment after admission to the provincial hospital , and 82 . 4% of patients subsequently diagnosed with scrub or murine typhus received an anti-rickettsial regimen . Doxycycline and chloramphenicol were the two main anti-rickettsial antibiotics used during the study and the majority of scrub typhus patients in our study recovered , despite previous reports of doxycycline and chloramphenicol resistant strains of O . tsutsugamushi in Chiangrai [40] . Of the 3 patients who died with scrub typhus , one did not receive any effective antimicrobial , one had delayed administration of chloramphenicol , and another received doxycycline and chloramphenicol from admission onwards . Azithromycin has been shown to be an effective alternative treatment in scrub typhus patients and appears also to be effective against resistant Chiangrai isolates of O . tsutsugamushi [41–43] . However , azithromycin was not used during this study due to the unavailability of more cost-effective generic formulations at the time . Nevertheless , the fact that 53% and 73% of patients with rickettsioses or leptospirosis , respectively , received a combination of a third generation cephalosporin plus a rickettsia-active antibiotic , and that in the remaining patients ceftriaxone was most commonly used for leptospirosis or bacterial causes , while doxycycline was commonly used for the rickettsial/dengue subgroups , demonstrates a high level of clinical experience and awareness among medical staff in this endemic area ( Table 2 ) . Roxithromycin was used in 1 patient with scrub typhus in combination with doxycycline . There have been limited clinical studies into the effectiveness of roxithromycin in the treatment of scrub typhus [31 , 44] and none reported for murine typhus . One case series from Chiangrai reported low efficacy of roxithromycin when compared to doxycycline or chloramphenicol in 20 children with scrub typhus [31] . In vitro susceptibility testing to roxithromycin has not been reported for Orientia tsutsugamushi while Rickettsia typhi appears susceptible [45] . Two patients with scrub typhus received ciprofloxacin , one as the sole anti-rickettsial antibiotic and the other in combination with doxycycline and chloramphenicol . Fluoroquinolones have been shown to be moderately effective in in vitro susceptibility tests and in limited clinical studies against murine typhus [45–47] . However , Orientia tsutsugamushi may be intrinsically resistant to fluoroquinolones which may explain the poor efficacy reported in clinical studies [48–51] . In contrast to antibiotic use in the hospital setting , only 34/200 ( 17% ) of study patients received antibiotics prior to admission . Rickettsial infections make up 25% of patients presenting with AUF , and only 10 of 34 patients received pre-admission antibiotics with anti-rickettsial activity ( 5% of the study cohort ) –of these only 5 patients received effective treatment for scrub typhus ( 2 . 5% of the study cohort ) . Supporting this observation prescription data from primary care units from the central Mueang district of Chiangrai province ( 2015 ) revealed low utilisation of anti-rickettsial antibiotics and doxycycline use was absent altogether . This highlights the need for improving the availability of specific antibiotics—particularly doxycycline—in rural endemic areas and for providing effective diagnostics to guide appropriate management of febrile patients , as inappropriate use of antibiotics has led to the development of antibiotic resistance , particularly impacting regions where access to effective antimicrobials is already limited [52] . The paucity of diagnostically useful clinical symptoms and signs in AUF cases should spur the development of affordable and effective rapid diagnostic tests ( RDTs ) . Even at a provincial hospital in Thailand , it is unrealistic for costly and expertise-reliant tests such as IFAs , ELISAs and PCR assays to be performed routinely [53] . Previous studies have shown that robust and high-quality RDTs for common causes of AUF provide the best balance for diagnostic cost-effectiveness [15] . This though requires up-to-date and representative local epidemiological data–ideally based on fever surveillance studies . The provision of effective RDTs to diagnose scrub typhus , dengue and leptospirosis will cover 41% of cases of AUF presenting to the provincial hospital in Chiangrai . The use of algorithms incorporating both clinical findings with accurate RDTs +/- basic laboratory tests to guide early appropriate antibiotic management of patients presenting with AUF will likely improve this further . In high income countries , biomarkers such as C-reactive protein ( CRP ) and procalcitonin , have been shown to be safe , cost-effective , and improve correct antibiotic use in the management of respiratory tract infections in primary care settings [17 , 25 , 54] . In Southeast Asia , it has been demonstrated that CRP can discriminate between bacterial and viral infections in acutely febrile patients and reduce antibiotic use in patients with non-severe respiratory tract infections in the community [18 , 19] . Modelling the impact and cost-effectiveness of pathogen-specific RDTs and CRP tests using data from febrile outpatients in Laos revealed that tests for common prevalent bacterial infections ( scrub typhus in that setting ) and CRP levels were likely to be cost-effective for direct health benefits while tests for viral pathogens ( e . g . dengue ) were not [15] . This study demonstrated that low CRP and low WBC were significant predictors of a viral infection ( mainly dengue , CRP≤37mg/L and WBC≤7 . 9x103/mm3 on CART analysis ) . CRP was highly sensitive and very specific for defining bacterial infections ( AUROC curve 0 . 9059 ) , when directly comparing bacterial and viral groups , consistent with data from previous fever studies from Southeast Asia [18] . Currently , two CRP cut-offs are under investigation– 20mg/L and 40mg/L . The results in our study suggest that from a statistical point-of-view , choosing the higher cut-off improves specificity by almost 14% , thus reducing false positivity . However , this needs to be put into clinical context as the reduction of incorrectly treated viral cases from 6/22 ( 27 . 3% ) to 3/22 ( 13 . 6% ) is offset by “missing” 4/72 ( 5 . 6% ) of potentially severe bacterial cases–thus , reducing 3 cases with inappropriate antibiotic treatment comes at a cost of not treating 4 cases that would require antibiotics . If the test is employed at a community/primary care level , where monitoring facilities are limited , it could be argued that incorrectly treating an additional 3/22 ( 4 . 8% ) febrile patients with a viral aetiology may be acceptable if an additional 4/72 ( 5 . 6% ) patients with a potentially severe bacterial aetiology can be treated appropriately . When comparing viral and bacterial groups , high procalcitonin was sensitive for the detection of bacterial infections but low levels were poor at selecting viral infections leading to low specificity . In Laos , elevated WBC counts have been shown to be significantly associated with fevers of bacterial aetiologies [6] . Previous fever studies from Southeast Asia have not specifically reported any association between neutrophilia and bacterial infections [3 , 4 , 6 , 8 , 55] , although multiple reports have associated neutrophilia , lymphopaenia and elevated neutrophil-to-lymphocyte ratios with bacteraemic medical emergencies in high-income settings [56–58] . Our results suggest that simple laboratory tests such as full blood count and CRP could be beneficial in differentiating between bacterial and viral infections in acutely febrile patients at the hospital level , while a CRP-based POCT test ( at USD 0 . 5–2 . 0 per test ) is likely to be cost-effective in community settings in rural Southeast Asia . As Thailand expands its community health care system to fulfil one of five core priorities in partnership with the World Health Organization—this information is relevant to the development and commissioning of diagnostics in the community/district hospitals [59] . Although we were able to assign diagnoses to 51 . 5% of the febrile patient cohort , a large number of patients with unknown aetiologies demonstrated elevated laboratory markers described above and median CRP levels comparable to the diagnosed bacterial group–suggesting that a large proportion of potentially antibiotic treatable diseases go undiagnosed . The study has important limitations: i ) due to budget constraints all cultures were performed in the local microbiology laboratory at the discretion of the treating physician; ii ) there was an imbalance in the diagnostic investigations performed due to costs and limited access to high quality tests which may have led to bias ( i . e . diagnostics for scrub typhus included PCR , culture and serology , while for leptospirosis only RDTs and culture were performed ) ; and iii ) the sample size is relatively small and external validity is limited although some conclusions can be drawn when the results are taken in context of previously published dataset from other fever studies from the region . In conclusion , this study has highlighted the importance of scrub typhus and dengue in the aetiology of AUF in Chiangrai province , northern Thailand . It has provided more evidence for including anti-rickettsial antibiotics into empirical hospital treatment guidelines and management strategies of AUF in the community . It contributes to the mounting evidence that good quality , accurate , pathogen-specific RDTs are urgently needed , which together with biomarker POCTs such as CRP , may aid healthcare workers in the correct use of antibiotics as part of the wider focus on antimicrobial stewardship . Finally , it emphasises the need for further prospective studies into the causes of AUF in the community along with evaluations of CRP POCTs in improving disease management algorithms , diagnostic accuracy , patient safety and reducing inappropriate antibiotic use in the tropics .
|
Fever remains an important reason why people are hospitalised in Southeast Asia . We do not know the most common causes of fever in many regions of the tropics . This knowledge would help doctors decide on the most appropriate treatment in areas where access to diagnostics is difficult . Establishing diagnostic tests for all possible diseases in an area is expensive and often impractical . An alternative is to measure ‘marker’ chemicals in the blood which the body produces in response to infection . These are usually higher in patients with bacterial infections . Differentiating bacterial from viral infections will help reduce inappropriate antibiotic use , which can contribute to the development of antibiotic-resistant bacteria . In this study , we investigated the causes of fever in hospitalised patients in Chiangrai , northern Thailand , and assessed if two chemical markers ( CRP and procalcitonin ) could distinguish bacterial from viral infections . Scrub typhus , dengue and leptospirosis were the major causes of fever , and these were not always accurately diagnosed and managed . We also found that CRP was better than procalcitonin in differentiating bacterial from viral infections . These results should help improve the management of febrile patients and increase the awareness of these neglected tropical diseases that are potentially deadly if missed .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] |
[
"antimicrobials",
"dermatology",
"typhus",
"medicine",
"and",
"health",
"sciences",
"pathology",
"and",
"laboratory",
"medicine",
"viral",
"transmission",
"and",
"infection",
"drugs",
"tropical",
"diseases",
"microbiology",
"bacterial",
"diseases",
"signs",
"and",
"symptoms",
"antibiotics",
"neglected",
"tropical",
"diseases",
"pharmacology",
"murine",
"typhus",
"infectious",
"diseases",
"zoonoses",
"eschar",
"scrub",
"typhus",
"leptospirosis",
"diagnostic",
"medicine",
"fevers",
"virology",
"microbial",
"control",
"biology",
"and",
"life",
"sciences",
"etiology"
] |
2018
|
Causes of acute undifferentiated fever and the utility of biomarkers in Chiangrai, northern Thailand
|
The abundance of different SSU rRNA ( “16S” ) gene sequences in environmental samples is widely used in studies of microbial ecology as a measure of microbial community structure and diversity . However , the genomic copy number of the 16S gene varies greatly – from one in many species to up to 15 in some bacteria and to hundreds in some microbial eukaryotes . As a result of this variation the relative abundance of 16S genes in environmental samples can be attributed both to variation in the relative abundance of different organisms , and to variation in genomic 16S copy number among those organisms . Despite this fact , many studies assume that the abundance of 16S gene sequences is a surrogate measure of the relative abundance of the organisms containing those sequences . Here we present a method that uses data on sequences and genomic copy number of 16S genes along with phylogenetic placement and ancestral state estimation to estimate organismal abundances from environmental DNA sequence data . We use theory and simulations to demonstrate that 16S genomic copy number can be accurately estimated from the short reads typically obtained from high-throughput environmental sequencing of the 16S gene , and that organismal abundances in microbial communities are more strongly correlated with estimated abundances obtained from our method than with gene abundances . We re-analyze several published empirical data sets and demonstrate that the use of gene abundance versus estimated organismal abundance can lead to different inferences about community diversity and structure and the identity of the dominant taxa in microbial communities . Our approach will allow microbial ecologists to make more accurate inferences about microbial diversity and abundance based on 16S sequence data .
The SSU rRNA gene ( also known as the 16S rRNA gene , referred to as “16S” hereafter ) is widely used in studies of microbial ecology as a “barcode gene” [1] to quantify microbial community structure and diversity [2] , [3] . The widespread adoption of 16S as a microbial barcode gene has been driven by several desirable properties of the gene , including the fact that it is universal across bacteria and archaea , it can be easily amplified from a wide diversity of taxa at one time by the polymerase chain reaction ( PCR ) , it is phylogenetically informative , and it can be used to identify and phylotype sequences based on extensive databases of 16S sequences with associated taxonomic and phylogenetic information [2] , [4] . In 2011 there were 3 , 574 publications in the Web of Science database matching a search for the terms “16S and ( communit* or diversit* or abundance* ) ” . There are numerous advantages to using 16S as a microbial community barcode gene , but also numerous disadvantages including amplification and sequencing bias and error [5] , [6] , difficulty with the accurate taxonomic identification and binning of short sequences [7]–[10] , and a lack of benchmark studies to guide decisions about quality control , filtering , and analysis of 16S sequence data sets derived from novel sequencing technologies . Another disadvantage of the 16S gene that is particularly relevant to inferring microbial abundance from 16S gene sequence abundances is that genomic 16S copy number varies a great deal across the tree of life [11]–[13] . For example , among bacterial taxa with fully sequenced genomes , 16S copy number varies from a single copy in Erythrobacter litoralis to fifteen copies in Photobacterium profundum [14] , [15] . As a result of this variation in copy number , the variation in the relative abundance of 16S gene sequences in an environmental sample can be attributed both to variation in the relative abundance of different organisms , and to variation in genomic 16S copy number among those organisms ( [12] , [16]–[18]; Figure 1 ) . The use of a single-copy protein coding gene such as rpoB as a microbial barcode would avoid this problem [11] , [19] , [20] , but these genes are not as widely used as the 16S gene , and there are biases inherent in the use of every barcode gene and sequencing technology . Though metagenomic data will help in allowing the use of genes that have less variation in copy number [20] , PCR amplification of 16S genes is still the method of choice in many environmental surveys . The vast majority of such studies either explicitly or ( usually ) implicitly assume that the relative abundance of 16S gene sequences is an accurate measure of relative abundance of the organisms containing those sequences in analyses of community diversity and composition . The degree to which this assumption is warranted , and the effect of treating 16S gene abundance as a surrogate measure of organismal abundance on estimates of microbial community structure , is unknown . In this study we present a method for phylogenetic estimation of 16S copy number and organismal abundance that allows us to improve estimation of microbial abundance and community structure by accounting for copy number variation among taxa . Our specific objectives are threefold . First , we demonstrate that 16S gene abundance is a function of both organismal abundance and 16S gene copy number , and show how this relationship can influence the ability to estimate community structure and diversity from sequence data . Second , we develop a method that allows estimation of organismal 16S gene copy number and abundance as a function of 16S gene abundances in environmental samples , and assess the performance of this method with simulated data sets . Finally , we apply our method to several empirical data sets to illustrate the practical effects of treating 16S gene abundance as a measure of organismal abundance on measures of microbial community diversity , structure , and composition .
Our interest lies in relating the observed abundances of 16S genes in biological samples to the abundance of cells , or organisms , from which these genes arise . For any taxon i within a biological community , the relationship between the abundance of 16S genes from that taxon , Gi , and the abundance of organisms from that taxon , Ni , is determined by the genomic 16S copy number of that taxon , Ci , where Gi = NiCi . Defining the relative 16S gene abundance of taxon i , , and the relative organismal abundance of taxon i , , it follows that: ( 1 ) Here , the summation is across all taxa i within the biological community , and is thus a constant . Because = 1 , equation 1 shows that in communities where all taxa have 16S copy number equal to one , all sampled taxa will have identical 16S gene and organismal relative abundance . As the 16S copy number of one or more taxa increases , disparity between the 16S gene abundance and organismal abundance of individual taxa grows . We can also readily explore community-level patterns of microbial abundance . We characterize the taxa-gene distribution , P ( G ) , as the fraction of taxa in a biological sample with 16S gene abundance G . Similarly , we characterize the taxa-abundance distribution , P ( N ) , as the fraction of taxa with N organisms . These two distributions are related by: ( 2 ) Here , the summation is over all possible combinations of N and C with product equal to G , and P ( N , C ) is the joint probability of a taxon having an abundance N and copy number C . In the case where organismal abundance and copy number are independent of one another , this simplifies to: ( 3 ) where P ( C ) is the distribution of copy number across taxa within the biological community . To understand the potential differences between gene abundance distributions and the organismal abundance distributions from which they are derived , we used two approaches . First we qualitatively compared the shapes of the distributions of P ( N ) and P ( G ) . To model the taxa-abundance distribution , P ( N ) , we simulated biological communities assuming a zero-truncated Poisson lognormal distribution [21] . We chose the lognormal distribution for illustrative purposes because it is one of the most widely discussed taxa-abundance distributions in biology [22] , [23] . To model the distribution of genomic 16S copy number across taxa , P ( C ) , we simulated biological communities with a zero-truncated Poisson distribution . We chose the Poisson distribution because it approximated the empirical copy number distribution in our reference data set ( Supporting Figure S1 ) . For each simulated community we calculated the resulting taxa-gene distribution , P ( G ) , from equation 3 . Second , we examined how sampling from the simulated biological communities with corresponding distributions P ( N ) and P ( G ) resulted in different biodiversity estimates . Our motivation for this was to understand the differences expected when sampling genes versus individuals from communities . To do this we sampled a fixed number of genes , or individuals , from the simulated communities . We focused on a key attribute of the sample distributions: the numbers of taxa that are unobserved , or hidden behind the ‘veil line’ of the sampled taxa-abundance and taxa-gene distributions [22] . For each sample we used standard parametric tools to estimate the number of unobserved taxa for P ( N ) and P ( G ) ( reviewed in [24] ) . We tested whether estimating the total taxa richness based on P ( G ) versus P ( N ) could lead to different inferences about diversity using an ANOVA to compare predicted taxa richness using these two different distributions . Environmental sequencing studies that utilize the 16S gene as a barcode provide a measure of 16S gene relative abundance gi . Given the relationship between 16S gene relative abundance gi , copy number Ci , and organismal relative abundance ni outlined above ( Equation 1 ) , we can estimate ni given information on gi and Ci . But the genomic copy number Ci of the 16S gene ( referred to as “copy number” hereafter ) is usually not observed directly from environmental sequence data because the full genomes of the organisms containing the gene are not sequenced . Metagenomic studies could theoretically address this issue [19] , , but metagenomic sequencing generally provides insufficient sampling depth to provide full genome coverage for all of the organisms in diverse communities . To overcome this challenge , we use methods from comparative biology and leverage phylogenetic signal in copy number to estimate copy number and organismal abundance for organisms for which we observe only 16S gene abundances . The general approach we use to estimate copy number and organismal abundance from environmental 16S sequences is to place those sequences onto a reference phylogeny of organisms for which genomic 16S copy number is known ( Figure 2 ) . Using ancestral state reconstruction via phylogenetically independent contrasts [26] , [27] , we can then obtain an estimate of genomic 16S gene copy number , , for any taxon i . By combining the estimated copy number , and the observed relative gene abundance of taxon i , gi , we can obtain an estimate of the relative abundance of taxon i following Equation 1: ( 4 ) To illustrate the impact of variation in copy number on empirical estimates of microbial community structure and diversity , we reanalyzed data from two previously published studies: a survey of microbial communities along an oceanic depth gradient using Sanger sequencing [41] , and a survey of the skin , gut , and mouth microbiome of a human female using pyrosequencing ( subject F1-3 from [42] ) . For each data set , we estimated the relative abundance for each OTU using our copy number estimation pipeline . We then asked whether accounting for copy number variation influenced several commonly used measures of community structure and diversity for each data set . We estimated the fit of gi and abundance distributions from these data sets to a lognormal model of relative abundance distributions . We classified each sequence in the empirical data sets to the taxonomic order level using the RDP taxonomic classifier [43] and evaluated changes in the relative abundance of bacterial orders based on gi versus . We measured overall community dissimilarity among samples from each study using the weighted UniFrac phylogenetic distance metric [44] , based on the both the gi and values , and then performed a hierarchical clustering with complete linkage to evaluate the overall similarity of samples in each study .
Plots of simulated P ( N ) and P ( G ) abundance distributions ( Figure 3 ) indicated that the shape of these distributions are different . For the simulation parameters we considered , treating Gi as a measure of organismal abundance lead to an underestimation of the abundance of rare taxa and overestimation of the abundance of the most abundant taxa compared to the distribution of Ni ( Figures 3 and 4 ) . Estimates of total species pool richness fit using a parametric method [23] were significantly lower for Gi than for Ni ( ANOVA; all P<0 . 01; Figure 4 ) . Copy number variation can have substantial effects on inferences about numerous aspects of community diversity and structure including relative abundance distributions , the estimated abundance of different taxa , and the overall similarity of ecological communities . In both empirical data sets , rank abundance distribution plots of and gi revealed that failure to account for copy number variation resulted in gi underestimating the relative abundance of the most abundant taxa and overestimating the relative abundance of the rarest taxa relative to ( Figure 7 ) . The fit of empirical rank abundance distributions of and gi to a log-normal distribution model was much better for than for gi ( human microbiome: AIC ( gi ) = −200903 , AIC ( ) = −215791; ocean: AIC ( gi ) = −4573 . 7 , AIC ( ) = −4808 . 1 ) . In addition to changes in the overall shape of rank-abundance distributions , the relative abundance of several microbial taxa also changed substantially after accounting for copy number variation among taxa . In the human microbiome data set , these changes did not greatly modify the overall abundance structure of the community ( Figure 8 ) . However , in the ocean data set the relative abundance of several taxa differed greatly when based on gene versus organismal abundance estimates ( Figure 8 ) . For example , the relative abundance of sequences assigned to Cyanobacteria Family II nearly doubled and this taxonomic group went from being the ninth most abundant based on gene abundance ( gi = 0 . 04 ) to the second most abundant based on estimated organismal abundance ( = 0 . 09 ) . The use of organismal versus gene abundances did not have a major effect on the clustering of ocean communities based on their phylogenetic similarity , with samples tending to cluster together with other samples from similar depths regardless of whether gi or was used to calculate weighted UniFrac similarity of samples ( results not shown ) . However , for the human microbiome data set , using gi versus as the abundance measure changed the overall similarity of communities from different habitats as measured by hierarchical clustering of communities based on the weighted UniFrac phylogenetic distance metric ( Figure 9 ) . Based on gene abundances , microbial communities from the inner ear/earwax clustered with communities from the sole of the foot ( Figure 9A ) , but based on estimated organismal abundance the inner ear/earwax community formed a distinct cluster with communities from the nostril , and these two communities from relatively moist skin microhabitats were compositionally distinct from all other microbial communities on drier skin sites and the gut and mouth ( Figure 9B ) .
We have demonstrated how data on the sequence and abundance of 16S genes in environmental samples can be used to accurately estimate 16S gene copy number and improve estimates of organismal abundance in microbial communities . Using simulated and empirical data sets , we have shown that treating gene abundance as if it were equivalent to organismal abundance can lead to misleading inferences about microbial community structure and diversity . Our simulations indicate that genomic 16S copy number can be estimated accurately for environmental sequences through the use of phylogenetic reference data , and that failure to account for copy number variation among taxa in environmental samples can lead to the observed relative abundance of 16S sequences ( gi ) being weakly correlated with the true abundance of organisms in the community ( ni ) . Our findings have wide-ranging implications for studies treating 16S gene sequence abundances as a measure of organismal abundances in communities . In some simulations , less than 30% of the variance in true organismal abundance was explained by observed gene abundance ( Figure 6 ) . The weak correlations between observed 16S gene abundance and true organismal abundance suggest that estimation of organismal abundance from gene abundance and copy number should be a routine part of any 16S sequencing study , since it will reduce one of the numerous potential biases inherent to inferring microbial community structure from environmental sequencing data . Analyses of several empirical data sets indicated that copy number variation can affect numerous aspects of community structure that are commonly measured by studies using the 16S gene , including relative abundance distributions , estimates of the abundance of different taxa , and overall measures of community diversity and similarity . The effects of copy number variation on community structure will not be consistent across studies , as they will depend on the relative copy number of taxa in a particular community , and on the distribution of , and relationship between , Ni and Ci in that community . Our simulations of gene and organismal abundance distributions , P ( G ) and P ( N ) , indicate that these distributions can have different properties . Under the simulation parameters we explored , there was a tendency for P ( G ) to have lower abundances for the rarest species and higher abundances for the most abundant species in comparison with P ( N ) . Estimates of species richness based on gene abundances were also consistently lower than estimates based on organismal abundances . These differences are likely due both to the fact that P ( G ) is a function of P ( C ) and P ( N ) ( cf . Equations 2 and 3 ) leading to a difference in the shape of gene and organismal abundance distributions , and due to sampling depth being effectively lower for gene abundance distributions than for organismal abundance distributions for a given number of genes/individuals sampled , since multiple copies of the genes of each organism make up the pool of genes in the community . We simulated P ( N ) and P ( C ) as statistically independent distributions , but it is also possible to imagine situations where P ( N ) and P ( C ) are correlated ( e . g . where abundant taxa have consistently higher or lower 16S copy number ) , which could further obscure relationships between gene abundance and organismal abundance . In the abundance distributions for the empirical data sets we examined , we observed that gene abundances were generally higher for the rarest taxa and lower for the most abundant taxa compared to estimated organismal abundances , a pattern opposite that seen in our simulations . This discrepancy highlights the fact that relationships between gene and organismal abundances will vary depending on numerous factors including the distribution of organismal abundances and copy numbers as well as the relationship between organismal abundance and copy number in natural communities , and further highlights the need to estimate copy number and organismal abundance for empirical data sets . There was not always a large effect of using gene versus organismal abundance to measure community structure in the empirical data sets we examined , but we did see major impacts on our inferences about community structure in some data sets , including changes in estimates of the identity of the common and rare taxa within communities and the similarity of communities among different habitats . If there is not a consistent difference in copy number between abundant and rare taxa , there could be little effect of adjusting relative abundance to account for copy number , but the only way to assess differences in gene versus organismal abundances for a particular community will be to estimate copy number and organismal abundance for the taxa in that community . There is great interest in understanding the structure and dynamics of the “rare biosphere” , the rare microbial taxa whose detection in ecological communities was only possible with the advent of high-throughput sequencing technology and deep sequencing of environmental samples [45] . In our simulations and analyses of ecological communities , we found that estimates of the relative abundance of rare taxa were consistently affected by variation in copy number , likely due to the fact that the effects of copy number on detection probability and abundance estimation will be strongest for the rarest taxa in a community [46] . It will be useful to disentangle the effects of copy number variation versus ecological rarity per se on our perception of the ecology of the rare biosphere . The phylogenetic method for copy number estimation we present in this study could be applied to predict any microbial trait for which reference sequence and trait data are available , and will help to further develop a trait-based approach to microbial ecology [47] . Numerous hypotheses about the environmental distribution of microbial traits including genomic 16S copy number have been proposed [48] , [49] and it will be possible to test these hypotheses using estimation of the traits of microbial communities . This approach will complement metagenomic approaches to understanding the distribution of microbial traits and functions , since it could be applicable to phenotypic traits of microbes that cannot be directly measured from metagenomic data such as genomic copy number or ecological attributes of taxa such as growth rate or pathogenicity . Since uncertainty in copy number estimates depends on the branch length separating environmental sequences from reference sequences , there will be greater uncertainty in estimates of copy number for sequences from poorly known and unculturable bacterial clades lacking close relatives in reference genomic data sets . However , for the empirical data sets we analyzed , the largest standard error of copy number predictions was less than one copy per sequence , even for the environmental sequences distantly related to all taxa in the reference data set . Our ability to estimate copy number accurately will be improved as the genomes of greater numbers of uncultured and rare microorganisms continue to be sequenced . The method we present in this study can be used with any set of reference sequences , and as greater numbers of genomes from uncultured and phylogenetically diverse microbes are sequenced [28] , we expect that our ability to estimate copy number and abundance will become even more accurate . Understanding patterns of organismal abundance across space , time and environments lies at the core of microbial biodiversity and biogeography research . The ability to estimate copy number and abundance for microorganisms based on environmental sequences opens the door to the application of numerous ecological methods developed for estimating taxa richness , taxa range distributions , and community similarity while taking variation in detection probability into account . Future studies utilizing the copy number and abundance estimation approach we have developed will improve our understanding of the structure and dynamics of microbial communities .
|
Microbial ecologists cannot observe their study organisms directly , so they use molecular sequencing to measure the abundance of different microbes living in the wild . The most commonly used method for measuring the abundance of different microbes is to collect a DNA sample from an environment and sequence a particular gene , the 16S SSU rRNA gene ( “16S” ) from those samples . The abundance of 16S sequences from different microbes is then used as a surrogate measure of the abundance of the microbial taxa in the community . One problem with the use of the 16S gene as a measure of microbial abundance is that many microbes have multiple copies of the gene in their genome . Thus , variation in 16S gene abundances can be caused by both genomic copy number variation and variation in the abundance of organisms . In this study we present a computational method that allows estimation of the abundance and genomic 16S copy number of microbes based on environmental sequencing of the 16S gene . We use simulations and analysis of microbial community data sets to demonstrate that estimating the abundance of organisms from 16S data improves our ability to accurately measure the diversity and abundance of microbial communities .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"phylogenetics",
"ecology",
"biology",
"evolutionary",
"biology",
"microbiology",
"evolutionary",
"systematics",
"microbial",
"ecology"
] |
2012
|
Incorporating 16S Gene Copy Number Information Improves Estimates of Microbial Diversity and Abundance
|
Plasticity studies suggest that behavioral relevance can change the cortical processing of trained or conditioned sensory stimuli . However , whether this occurs in the context of natural communication , where stimulus significance is acquired through social interaction , has not been well investigated , perhaps because neural responses to species-specific vocalizations can be difficult to interpret within a systematic framework . The ultrasonic communication system between isolated mouse pups and adult females that either do or do not recognize the calls' significance provides an opportunity to explore this issue . We applied an information-based analysis to multi- and single unit data collected from anesthetized mothers and pup-naïve females to quantify how the communicative significance of pup calls affects their encoding in the auditory cortex . The timing and magnitude of information that cortical responses convey ( at a 2-ms resolution ) for pup call detection and discrimination was significantly improved in mothers compared to naïve females , most likely because of changes in call frequency encoding . This was not the case for a non-natural sound ensemble outside the mouse vocalization repertoire . The results demonstrate that a sensory cortical change in the timing code for communication sounds is correlated with the vocalizations' behavioral relevance , potentially enhancing functional processing by improving its signal to noise ratio .
A central question in neuroscience is how behaviorally relevant sensory signals are encoded by the brain . In the context of species-specific communication , this issue is complicated by the fact that many sounds with the same meaning are variable in their physical characteristics , such as speech phonemes spoken by different people [1 , 2] . In some cases , only the message itself is relevant , and just its detection over background noise is necessary; in other cases this variability discriminates between various speakers . What aspects of the neural code carry information for the detection and discrimination of such naturally varying sounds , and does their behavioral relevance affect their encoding ? These questions have not been fully addressed in mammals , despite a rich literature on the neural representation of communication sounds , particularly in auditory cortex . Most research has focused on the selectivity of cortical neurons for intraspecies communication calls of primates [3–7] , guinea pigs [8 , 9] , bats [10 , 11] , and cats [12] . Neurons have generally not been found to be call specific in their response [13] . This poor selectivity does not imply an absence of information that could be useful for detecting and discriminating calls: some neurons may be more informative than others , even if they are not call selective . Evaluating this possibility first requires a quantitative characterization of vocalization variability , as has been done for the marmoset [14] and bat [15] . In the latter , this has led to the conclusion that the average temporal pattern of neural responses helps discriminate categories of calls with very different acoustic structure [16] . These animal models do not reveal though whether the significance of the communication sound per se impacts neural coding , as some cross-species studies suggest [13 , 17] . Therefore , a model system is needed in which the encoding of variable vocalizations can be quantitatively compared between animals ( of the same species ) for which specific sounds either do or do not carry communicative significance . The mouse ultrasound communication system provides such an opportunity [18–20] . The emission of ultrasonic calls by isolated mouse pups acts as a communication signal to elicit a search and retrieval by mouse mothers [20 , 21] . The variability in the acoustic parameters of these calls has been extensively characterized [19] , laying the foundation for quantitative neural-coding studies . In two-alternative choice tests , mothers preferentially approach pup-like ultrasounds over a neutral sound not in the mouse vocal repertoire and can even discriminate ultrasounds based on frequency , duration , and bandwidth [22–24] . This preference is a clear indication that pup calls carry communicative significance for mothers , a significance that is not recognized by pup-naïve virgins , which do not favor these ultrasounds [25] . This contrast therefore supplies a natural control animal group for investigating whether and which aspects of the neural code correlate with communicative significance . We pursued this by recording auditory cortical spiking activity in response to natural mouse pup isolation calls from anesthetized mothers ( whose pups were weaned within one week prior to experiments ) and virgins ( which were never housed as adults with pups ) . The auditory cortex was chosen since immediate early gene expression ( c-Fos ) [26] and neuronal responsiveness to the call bout structure [27] hint that this area reflects the recognition of pup calls by mothers . Here , we introduce a novel methodology to evaluate the information that auditory cortical neurons carry for the detection and discrimination of pup calls and test whether differences in information encoding exist between mothers and virgins . Because behavioral preference is a consequence of sensory , motivational , decisional , and motor processing , it is not immediately obvious that the neural firing in a sensory cortical area will be correlated with communicative significance . We found that the timing of the information about pup calls in cortical responses of mothers is significantly different from pup-naïve virgins , resulting in improved detection and discrimination ability for behaviorally relevant communication sounds .
A typical pup call evoked a strong , time-locked neural response in the auditory cortex of anesthetized mice ( Figure 1A ) . Similar average spike counts were elicited from both mothers and pup-naïve virgin female mice ( Figure 1B ) . This might lead to the conclusion that auditory cortical processing is not sensitive to the behavioral significance of a communication sound . However , looking closely at the time course of the responses ( Figure 1C ) , important differences between animal groups appeared . The peristimulus time histogram ( PSTH ) generally peaked earlier in mothers , with a larger and narrower response . This was most prevalent for recording sites having characteristic frequencies ( CFs ) ( frequency of the lowest amplitude tone that elicits a response ) near the pup call frequency range ( 40–80 kHz ) , but was also seen for mid-CF sites ( 20–40 kHz ) . Thus , might the timing of neural responses carry information about behavioral relevance ? To investigate this , we focused on two behaviorally important functions in communication: detection ( “did a call occur ? ” ) and discrimination ( “is one call different from another ? ” ) . In practice , these tasks are complicated by the natural variability of communication calls . For example , both the median frequency and duration vary due to individual pup differences as well as age-related changes [19] . To test how neural responses encode this natural variation , we played back 18 different pup calls , two each chosen randomly from nine regions in the frequency-duration plane ( Figure 2 ) . This collection included both high and low probability calls that varied systematically in these two parameters ( rather than only higher probability calls that would have been chosen by a purely random selection strategy ) . We collected multiunit ( MU ) spike activity in response to these calls at a population of recording sites across the mouse auditory cortex of mothers and naïve females ( see Materials and Methods ) . The tonal CFs and thresholds of these sites were not significantly different between the two animal groups ( p > 0 . 05 ) ( Figure 3 ) . The 18 calls ( Figure 4 , left column ) elicited a variety of different responses . Figure 4A , 4C , 4E , and 4G ( 4B , 4D , 4F , and 4H ) show the raster of spike activity from four MU sites in mothers ( naïve females ) , along with their respective spontaneous activities ( top panels ) . These examples were selected to convey the range of strong and weak responses observed in both animal groups . The overall firing rate elicited by all calls ( bottom panels ) rose sharply just after sound onset for many sites . MU 482 responded selectively to some vocalizations ( such as numbers 1–3 ) and not others ( like numbers 7–9 ) , with slight shifts in latency ( compare numbers 10 and 15 ) . MU 528 responded only at the onset of nearly all the calls , albeit with different firing probabilities for different calls . Two well-isolated single units ( SUs ) from mothers ( SU A and SU B in Figure 5A and 5B , respectively ) showed similar response features: SU A was an onset responder to all calls , while SU B responded in a slightly more graded fashion to different calls . SU C from a naïve female responded weakly to pure tones ( unpublished data ) , but had an identifiable CF near 60 kHz . Its overall firing to pup calls , however , showed only a slight elevation during the calls relative to its spontaneous firing . To make quantitative statements about the processing of these communication sounds , we evaluated the information that neural responses conveyed for call detection and discrimination . In general , information about a stimulus s may be provided by the entire time course of a response r . We analyzed this response time course in 2-ms bins , treating each bin as independent . In principle , a pattern of spikes may convey more or less information than the contribution from each spike [28 , 29] . We did not integrate the information over time , so our approach ignored potential synergy and redundancy for spikes lying in different 2-ms bins . However , we could nevertheless reveal coding differences that were correlated with communicative significance , the main objective of this work . Intuitively , information about a stimulus is gained from a response if the latter reduces the uncertainty about what stimulus occurred . For detection , different calls within a category are equivalent , and the acoustic variability is immaterial . The uncertainty is only about whether any call occurred relative to silence . Thus , an ideal detector would generate the same spike response regardless of the call—a response different from its spontaneous firing . Formally , the mutual information between the stimulus possibilities ( s = “call” or “no call” ) and response possibilities ( spike count in a 2-ms bin ) quantifies how much the latter changes our uncertainty in the former ( see Materials and Methods ) . Figure 6A–6F illustrates this for MU 482 at a time bin t corresponding to the peak in the PSTH ( arrow in the bottom panel of Figure 4C ) . We grouped all 18 calls together into a single category , thereby ignoring the identity of individual vocalizations . Before observing the response r at t , both the “call” and “no call” possibilities were considered equally likely ( log2[2] = 1 bit of uncertainty ) , so that their probabilities ( diameter of the circular icons in Figure 6B–6F ) were the same . If no spikes were observed , the two stimulus possibilities were still about the same ( Figure 6C ) . However , if one , two , or three spikes were observed , it was progressively more likely that a pup call occurred ( Figure 6D–6F ) . Thus , the detection uncertainty was reduced by observing r , and information was gained . The total amount of information contributed by this time bin was 0 . 1 bits , defined as the average change in uncertainty from each response possibility ( zero , one , two , or three spikes ) , weighted by the probability of that response ( Figure 6A ) . This analysis therefore provided a quantitative measurement of the ability of this neural response to convey information for detecting the behaviorally important communication call . A similar analysis can be applied to quantify how well the neural response discriminates calls . In this case , differences between pup isolation calls are important , and the uncertainty is about which of the 18 calls occurred . An ideal discriminator would fire in a unique manner for each of the s = “call 1” to s = “call 18” calls . How much information do real auditory cortical responses provide for discriminating calls ? Figure 6G–6L illustrates this assessment for MU 482 , at the same time bin considered for detection , above . A priori , all calls were considered equally likely , resulting in log2[18] = 4 . 2 bits of uncertainty ( Figure 6H ) . Conditioning on the different possible spike count responses ( Figure 6G ) , some stimuli clearly became more likely . It was found that zero spikes did not markedly change the uncertainty ( Figure 6I ) . However , if one spike was observed , one of the lower frequency calls most likely elicited that response ( Figure 6J ) . The stimulus uncertainty was further reduced by two or three spikes ( Figure 6K–6L ) , since only five or two of the 18 calls , respectively , were likely . Overall , this time bin provided 0 . 6 bits of discrimination information for MU A , thereby quantifying the neural ability to tell these different communication calls apart . Using this methodology , we derived time courses for the information each independent time bin in the response conveyed for detection ( Figure 7A–7K ) and discrimination ( Figure 7L–7V ) , for the examples in Figures 4 and 5 . For comparison , we also randomized trials across the stimulus possibilities ( see Materials and Methods ) so that no information was in principle available . Since the noise in the spike counts from finite trials can cause both a bias and fluctuations in the information estimates [30 , 31] , the randomized estimate provided a baseline for comparing whether peaks in the information were significant . MU 482 exhibited large peaks above the randomized estimate ( gray lines ) in both detection and discrimination information soon after the stimulus onset , as expected from its consistent yet systematically varying responses to different calls . On the other hand , MU 528 showed appreciable detection but very little discrimination information , as expected from the onset nature of its responses . SU A exhibited an extended period of detection information lasting beyond the duration of the call ( Figure 7I ) . This matched the interval when calls suppressed spiking relative to the spontaneous activity ( Figure 4C ) and demonstrates that the absence of spikes can also be informative . Furthermore , the naïve female example , SU C ( Figure 7K and 7V ) , produced time courses that were fairly similar between the actual information and the randomized control information . This was not surprising , given the unit's poor response to individual calls ( Figure 5C ) . These examples demonstrate how our information analysis quantitatively summarized the complex neural coding of natural calls , yielding results that were consistent with our qualitative impressions . Next , the MU neural population was analyzed as described above to see whether mothers and naïve females coded pup calls differently . Since we were not very restrictive in selecting recording sites ( see Materials and Methods ) , many showed rather poor information; with peak information values ( time of maximum detection information ) during the calls that varied little from peak values long after the calls were presented ( i . e . , after the spike rate had returned to the spontaneous level ) . To avoid claiming that such sites carry significant information , we assessed the distribution of peak information during a very late period in the activity ( arbitrarily chosen at 365 to 430 ms after call onsets ) . The cumulative probability distributions of this peak information are shown by the dashed lines in Figure 8A and 8B for detection and discrimination , respectively . As expected , there was virtually no difference between mothers ( magenta ) and naïve females ( sage ) . This was a period when information about the calls should be minimal , and any residual information is likely dominated by noise or bias in our estimation procedure , which would be common to both animal groups . A further check of the cumulative distributions of the peaks in the randomized control information also showed no difference between mothers and naïve females ( unpublished data ) , providing further confidence that our estimation procedure did not artificially inflate the information values of one animal group over the other . In contrast , when the peak information during the response to the calls was compared between sites for mothers and naïve females ( peak between 5–70 ms after call onset , an interval equal to the longest call duration ) ( solid magenta and sage lines in Figure 8 , respectively ) , a clear difference was apparent . The cumulative probability distributions showed a sizeable gap between the two animal groups , with the mothers exhibiting a larger proportion of sites with higher detection and discrimination information ( see Figure 8 legend for further details ) . This was the first indication that the coding of pup calls differs between animals with and without exposure to and experience with pups . This was further evident in a comparison of the average ( over sites ) time course for detection and discrimination information ( Figures 9 and 10 ) . Restricting ourselves to only those sites that likely carried significant peak information ( to the right of the black threshold line in Figure 8 , see Materials and Methods ) , the neural responses in mothers ( Figure 9A ) conveyed more detection information on average than naïve females ( Figure 9B ) . Even when all sites were included , these conclusions were the same . In particular , mothers showed a strong , early peak that was lacking in naïve females ( empirically fit peak at 18 ms versus 29 ms , relative to the stimulus onset , respectively; see Figure 9 legend ) . This can also be seen by comparing the peaks of individual sites , as illustrated in Figure 9C . Plotted on a logarithmic scale for clarity , higher information sites had shorter latencies , especially for mothers . Indeed , the distribution of peak information latencies was shifted to shorter times ( max at 16 versus 28 ms ) in mothers compared to naïve females ( Figure 9D ) . These results suggest that neurons in mothers provide earlier and greater information for detecting pup calls . Discrimination improvement was even more striking . When all sites were considered , the average information time course peaked strongly for mothers , but barely changed from the randomized control for naïve females . When only the most significant sites were averaged together and the resulting peaks were numerically fit , mothers had a relative information peak around three times larger , and earlier ( 14 . 6 ms versus 21 ms ) , than naïve females ( Figure 10A versus 10B , see legend ) . The sites that contributed the greatest peak discrimination information were again clustered at the shortest latencies , particularly for mothers ( Figure 10C ) . The distribution of these peak information latencies was broad for both groups , but weighted toward shorter latencies in mothers ( Figure10D ) . Finally , there were significantly more sites in mothers that conveyed discrimination information beyond our threshold ( p < 0 . 05 , test of proportions ) . Taken together with the detection results , our study suggests a correlation between the communicative significance of a sound category to an animal and that animal's auditory cortical detection and discrimination processing of those sounds . To try to understand the origin of this improved information in mothers , we considered several additional analyses . Since the calls varied systematically in frequency and duration , we asked whether the responses discriminated between calls because they provided information specifically about these acoustic parameters . We reanalyzed the information by grouping calls first according to the three different frequency ranges from which they were selected ( frequency information ) and also according to the three different duration ranges ( duration information ) . This ignored all other acoustic differences between calls , such as amplitude envelope variations , and only considered how the responses informed about the consistent acoustic parameter—frequency or duration . Figure 11 plots the maximum ( over time ) information available for distinguishing call frequency and duration ( see Materials and Methods ) . Only those sites with significant call discrimination information are shown . For both mothers and naïve females , there was a tendency towards better frequency rather than duration information ( data lie mainly to the right of the diagonal ) . Furthermore , more sites in mothers had large frequency information , suggesting that the mothers' improved information for call discrimination may be related to a neural change in frequency sensitivity over this very narrow range of pup call frequencies . Indeed , for both animal groups , discrimination information was higher at sites with greater frequency information . Using the MATLAB analysis of covariance tool AOCTOOL , the linear regression slope of 0 . 98 was significantly different from 0 , p ≪ 0 . 05 , and the slopes for mothers and naïve females were not significantly different from each other , F ( 1 , 62 ) = 0 . 8 , not significant ( n . s . ) . Importantly , although frequency sensitivity appears to be an important factor in improving the discrimination information , the improvement was not restricted simply to units with higher CFs , which might be expected to have better frequency sensitivity in the pup call range . In fact , discrimination information was generally better in mothers across all CFs . After taking into account the CF dependence of the discrimination information , which was fitted at 0 . 0016 bits/kHz ( an analysis of covariance showed that the slopes were not significantly different between mothers and naïve females , F[1 , 57] = 1 . 67 , n . s . ) , a significant main effect due to animal group was found ( F[1 , 58] = 6 . 53 , p < 0 . 05 ) . We next wondered whether the coding difference between mothers and naïve females was specific to pup isolation calls , or whether it might generalize to a noncommunicative sound ensemble as well . Since the recognition of an ultrasound signal as a pup call by a mother can depend on spectral cues [22 , 24 , 32] , we frequency halved the natural pup call frequencies ( see Materials and Methods ) to generate a collection of behaviorally irrelevant but acoustically related sounds . Because of the logarithmic frequency scale of the basilar membrane , this ensemble spanned an extent along the cochlea comparable to that of the original calls . We presented the sounds at a random subset of recording sites and computed the detection and discrimination information for these translated calls . This ensemble excited the recording sites quite well since their ~35 kHz frequency was closer to the frequency of the minimum behavioral hearing thresholds for mice [33] . Therefore , it was not surprising to see that both mothers and naïve females showed strong detection information ( Figure 12A and 12B ) . However , the distribution of latencies to the peak detection information was not significantly different between the two groups ( Figure 12C ) . Therefore , although there may be some generalized improvement in sound detection information in mothers , it occurred without significantly changing the timing of the information . Furthermore , unlike the natural pup call case , there was virtually no information on average for discriminating these frequency-divided sounds in either mothers or naïve females ( Figure 12D and 12E ) . Peak information latencies were also not significantly different ( Figure 12F ) . Hence , the change from the naïve to maternal state did not appear to substantially affect the neural discrimination information for these noncommunicative signals . This is consistent with the idea that behavioral relevance is an important factor for altering the auditory cortical coding of a sound ensemble .
Our main new finding was that the behavioral relevance of an intraspecies communication call is correlated with changes in the timing of the auditory cortical spiking response . Specifically , our analysis revealed that the information neural responses convey for detecting and discriminating natural vocalizations reaches a larger and earlier peak in animals for which the calls have communicative significance . Moreover , the data suggest that sites conveying the most information do so with the shortest latencies , a property that may improve the synchronization of relevant neurons , as well as the signal to noise level at the input of downstream areas . Finally , better frequency encoding of calls , regardless of CF , appears to be primarily responsible for improving call discrimination information in the auditory cortical responses of mothers . These results provide evidence in a novel mammalian model that the timing of spikes , and not just the average spike count , is an important aspect of the neural code for communication sounds . Two caveats should be mentioned . First , this study was performed in anesthetized animals , where stimulus control and animal state can be straightforwardly controlled . Ultimately though , the coding of behaviorally relevant stimuli should be tested in nonanesthetized preparations as well . Second , our conclusions are primarily based on MU data , although they were apparently not very sensitive to the number of contributing neurons ( see Materials and Methods ) ; and our SU examples agreed with our findings . The idea that spike timing might be important for behaviorally relevant vocalization encoding within the auditory system has been implied in earlier work . A pioneering study in marmosets found that the auditory cortical discharge to a species-specific twitter call was much more synchronized across recording sites than would be predicted by the spectrographic representation of the sound , with many neurons firing earlier than expected [5] . Furthermore , a recent study in the zebra finch homologue of the inferior colliculus reported that neurons fired earlier , more precisely , and synchronously to natural bird songs than to behaviorally irrelevant modulation-limited noise [34] . Another study in the zebra finch homologue of auditory cortex found that finer ( 10 ms ) rather than coarser temporal resolutions were optimal for spike trains to discriminate different bird songs [35] . These studies looked only at animals for which the natural vocalizations were already behaviorally relevant; the results could therefore be due to evolutionary , developmental , and/or experience-dependent mechanisms [36] . To our knowledge , the current work demonstrates for the first time that the neural code for communication sounds in adult animals can change ( because of either experience or possibly hormonal mechanisms ) as the significance is acquired , and that this plasticity can quantitatively improve information processing for specific communicative functions . This goes beyond a parallel study in mice that looked at changes between naïve females and mothers in the cortical entrainment to sequences of identical pup calls [27] . That study only analyzed total spike counts and found that auditory cortical MUs in mothers but not naïve females could follow sequences of pup calls up to the naturally occurring pup call repetition rate of ~5 Hz . It explored neither the information encoding of single pup calls nor changes in the timing of spiking information within the response to each call . Moreover , while better entrainment could arise from changes in the duration of the long after-hyperpolarization potential following a spiking response , the mechanisms responsible for the fine time-scale changes in spiking during the call ( such as frequency sensitivity ) are probably different . Nevertheless , both these works utilized natural control groups ( virgins ) to explore the impact that behavioral significance has on the neural code for vocalizations . This natural paradigm has not been exploited before , perhaps because a progression in the significance of an intraspecies vocalization is difficult to trace through the life of an animal . An alternative approach to this would be to instrumentally train animals in specific behavioral tasks using unfamiliar vocalizations , such as from another species [13] . However , that may or may not generate the same type of plasticity as the natural context . On the one hand , training monkeys in a tactile discrimination task produced an earlier and larger pooled MU PSTH response for primary somatosensory cortical neurons after stimulation of the trained digit ( behaviorally relevant ) compared to an untrained digit ( not behaviorally relevant ) [37] . This is reminiscent of the differences between naïve females and mothers in the PSTH response auditory cortical MUs to a typical pup call ( Figure 1B ) . On the other hand though , there are reasonable arguments for why the plasticity may be different . First , training contexts usually familiarize an animal to only a small number of vocalization tokens , while communication sound learning likely involves a huge variety of exemplars from which statistical regularities are extracted [38] . Second , the reinforcement mechanisms in instrumental versus natural contexts could differ , depending on the nature and value of the reward . Supporting this , in the natural maternal context , young pups have been found to be more rewarding to a recent rat mother than cocaine [39] . Moreover , suckling pups stimulate a mother's dopaminergic reward system differently than cocaine does [40] . Hence , rewards derived from a social environment might produce different brain changes than food or water . In conjunction with this , an animal's hormonal state likely differs in natural and training contexts . This is relevant in light of a recent study showing that estrogen can modulate the auditory processing of behaviorally relevant song signals in the bird [41] . Indeed , training tasks have not always yielded the same kind of coding plasticity that is reportedly achieved naturally in a communication setting . For example , some marmoset auditory cortical neurons have a firing rate preference for the normal forward direction of a marmoset twitter call compared to its reverse [5] , but water-restricted ferrets trained to recognize those same marmoset tokens in a go/no-go task do not [13] . Interestingly though , the temporal response patterns from the cortical neurons of trained ferrets do carry more information for classifying real and reversed call tokens compared to untrained animals [13] . One interpretation of these findings is that the specific plastic changes that are induced depend on the behavioral task [42] . In communication , multiple tasks , such as detection , discrimination , and categorization , are sometimes simultaneously necessary . This might require a different encoding of the vocalizations than what results from training animals in a specific task on specific exemplars with a specific form of reward . This reinforces the need to use new methods , such as the one implemented here , to evaluate the contributions that neurons make towards useful communication processing tasks . Our conclusions were based on an information theoretic analysis of the responses to multiple vocalizations from within a single intraspecies communication sound category . Information theoretic analyses have been used previously to study the auditory cortical coding of sound location [43] , generic sound [44 , 45] , and animal vocalization [12 , 13] classification . Our focus though was on how natural acoustic variation is incorporated into communication processing . We considered two complementary psychophysical functions that must deal in different ways with this variability . Detection requires that neural responses be the same for different variations within the call category , while discrimination is best when they are reliably different . This dual approach goes beyond computing information just between responses and individual sounds , as has been done for vocalization ensembles with altered ( e . g . , time expanded or reversed ) calls [12 , 13] . Instead , to quantify the detection and discrimination of a communication category , different real calls that sample the category's known acoustic variability should be presented . Methodologically , we should point out that while calls were selected based on the distribution of acoustic parameters within our large library [19] , we do not know the “true” likelihood that a given animal actually encountered each type of call or how rare calls are in the natural setting . Changes in these probabilities would affect the a priori entropy for discrimination and detection , respectively . However , since the same probabilities were assumed for all animals , and our conclusions were based on comparisons between animal groups , we did not feel this was a serious limitation . Within such a paradigm , we found that sites could be better at conveying one type of information compared to another ( off-diagonal points in the upper right quadrant of Figure 8 ) . For example , onset responders tended to be better at detection than discrimination ( for example , MU 528 and SU A ) . It will be interesting to see in future experiments whether neurons carrying different information might be spatially clustered in the auditory cortex , perhaps forming functional modules . In support of this possibility , a recent guinea pig study found similarities within an auditory cortical column in the response to an intraspecies call and segregation of different response types across the cortex [9] . The cortical encoding changes we found are only a first step in fully characterizing the differences that are correlated with the communicative significance of pup calls . For example , if there is significant synergy or redundancy in the neural responses , the information we computed in independent 2-ms bins would not predict how the full time course of the response might detect or discriminate the pup calls . This requires many more trials in order to accurately estimate the Shannon information when more response possibilities can occur ( i . e . , more spikes in larger bins ) . An alternative approach has been to classify neural responses according to a specific decoding algorithm and then to calculate information between the actual and assigned stimuli [13 , 35 , 46 , 47] . By the data-processing inequality [48] , this would produce a lower bound on the true information between stimulus and response . Although we do not assume a decoder , our study is comparable in that it is also limited by the data-processing inequality due to the 2-ms binning . Nevertheless , even if the full detection or discrimination information is similar between mothers and naïve females , our results still suggest that changes occur in how this information is distributed across time . Finally , it is important to note that mothers and naïve females show behavioral differences in the recognition of pup calls , as inferred from their relative preferences to approach these sounds in two-alternative choice tests [25 , 26] . It is not known though whether the “perceptual qualities” of the calls differ for the two animal groups . In fact , a preferred approach could in principle result from a change in the motivation or decision to respond to a stimulus , without any sensory changes . Yet our findings demonstrate that the sensory cortical neural encoding of communication sounds can not only change , but actually might enhance the neural ability to detect and discriminate calls once they are preferred . In this particular system , such plasticity might be behaviorally advantageous for retrieval performance in natural settings . In general , sensory improvements that increase the signal over the background noise may be an important preprocessing step for decisions to act .
MU experiments on six recent mothers and six pup-naïve female CBA/CaJ mice ( 11–18 wk ) were carried out at the University of California at San Francisco ( UCSF ) ; additional SU studies in two mothers and one pup-naïve female were conducted at Emory University . The Institutional Animal Care and Use Committees of both UCSF and Emory approved all procedures . Animals were housed under a reversed light cycle . Details of the surgery and setup have been described elsewhere [27 , 49] . Briefly , mice were anesthetized with a combination of ketamine ( 100 mg/kg initial dose and 65 mg/kg maintenance ) and medetomidine ( 0 . 3 mg/kg ) and secured in a nose clamp for a craniotomy over the left auditory cortex [50] and recording . After surgery , animals were repositioned in front of a wide-bandwidth ribbon tweeter ( High Energy EMIT-B , Infinity , http://www . infinitysystems . com ) or a Tucker Davis Technology's ( TDT , http://www . tdt . com ) ES1 electrostatic speaker ( Emory ) in an anechoic chamber ( Industrial Acoustics , http://www . industrialacoustics . com ) . The sound delivery system was calibrated by TDT software using a Brüel and Kjær ( B&K , http://www . bksv . com ) free-field microphone coupled to a B&K 2669 preamp and 2690 amplifier . Stimuli were generated using TDT System 3 hardware and software ( sample rate of 195 , 312 . 5 samples per second via an RP2 . 1 digital-signal processing module at UCSF and 223 , 214 . 2857 samples per second via an RX6 module at Emory ) and presented through Brainware ( http://www . brainware . com ) , which also served to collect thresholded action potentials . Noise bursts and frequency sweeps were used as search sounds to locate auditory responses . We used 60-ms tone pips of varying amplitude and frequency to estimate the CF and threshold for each recording site ( details available in [27] ) . Recordings of pup calls were drawn from a large library of natural ultrasound vocalizations from CBA/CaJ mice [19] . Recording snippets were high-pass filtered in software ( 25 kHz corner , eight-order Butterworth filter , MATLAB ) , spectrally denoised [19] , multiplied by a 0 . 5-ms cos2 onset and offset function , and scaled to a target root-mean-square ( RMS ) amplitude to generate clean vocalizations for playback . Frequency-divided pup calls were generated in the same manner , except that a Hilbert transform was applied to extract the real call's phase function . This was multiplied by 0 . 5 before inverse transforming the signal with the original amplitude envelope to generate a frequency-halved call . For MU recordings , unless otherwise noted , twelve trials ( 600 ms long ) of each of the calls were presented in random order at all sites; during SU recordings the number of trials varied depending on how long units were held ( see Figure 4 for details ) . Recordings of adult CBA/CaJ calls and synthetic narrowband noise models of the typical pup call ( Figure 1 ) were also played back but were not analyzed in this work . The exposed cortical area was photographed to record penetration locations . Epoxylite-coated tungsten microelectrodes ( Fred Haer and Company , http://www . fh-co . com ) were introduced perpendicularly into the cortex and advanced 300–600 μm below the surface . For MU recordings , electrode impedances were typically 1–2 MΩ; 4–10 MΩ electrodes were used in SU experiments . Initial penetrations were directed towards the expected center of the auditory cortex . Subsequent penetrations ( usually along the rostral-caudal axis ) tried to locate the border between the primary ( A1 ) and anterior auditory fields ( AAF ) by searching for a reversal of the tonotopic gradient . Once identified , we next tried to target the ultrasound field ( UF ) or the secondary auditory field ( A2 ) and dorsal posterior ( DP ) fields , based on relative topography and response properties [27 , 50] . If neurons responded to tone frequencies above 20 kHz ( which happened most of the time ) , we presented the pup call stimuli ( at a few sites , only the typical pup call or the full pup call ensemble was played back , but not both ) . At a random subset of sites , frequency-halved pup calls were also presented . Some sites did not have a readily identifiable tuning curve , although they were driven by sounds . In total , MU responses to either the typical pup call ( Figure 1 ) or the pup call and frequency-divided pup call ensembles were collected from 112 sites in mothers and 106 sites in naïve females . The target RMS amplitude for the typical call was 65 dB sound pressure level ( dBSPL ) . For the call ensembles , the target was 74 dBSPL at most sites and 65 dBSPL at a few others . The inclusion of the 65 dBSPL data did not affect our conclusions , so data from both sound levels were combined . Since complete auditory cortical maps were not obtained , we were not always able to unambiguously determine the likely auditory field for each site . Our population in mothers ( naïve females ) included 54 ( 32 ) primary ( i . e . , A1 or AAF ) , 18 ( 12 ) UF , 5 ( 4 ) A2 , and 35 ( 58 ) other/uncertain sites ( classification as in [27] ) . Although Figures 5 and 6 illustrate detection and discrimination information in terms of the reduction in stimulus uncertainty , information was estimated in practice by applying Bayes' Theorem [48] and looking at response uncertainty . Thus , the Shannon information was defined as the difference between the response entropy H ( r ) = p ( r ) log2 p ( r ) and the response entropy conditioned on the stimulus H ( r|s ) = p ( r|s ) log2 p ( r|s ) , where p ( r ) is the probability of response r ( zero , one , two , . . . spikes in a 2-ms bin ) , and p ( r|s ) is the probability of response given the stimulus s . These quantities were estimated through data-size scaling procedures , wherein trials were considered first together and then randomly partitioned into two , three , and four groups [28] . For each data size ( 1 , 1/2 , 1/3 , and 1/4 ) , the entropies were calculated ( from the probabilities as described below ) and averaged together . These were then fit to a quadratic function of the inverse data size to extract an infinite data limit [51] . The difference between the fitted response and conditional entropies provided one estimate of the information . The procedure was then repeated 50× with different random data partitions , and the final information estimate averaged these . Although our methods can in principle be extended to bin sizes larger than 2 ms , this resolution was used to limit the possible responses so that probabilities could be more accurately estimated from the relatively small number of trials . For detection , p ( r|s = “call” ) at time bin t was found by grouping all trials for all pup calls together ( usually 12 trials × 18 calls = 216 effective trials ) , since the identity of individual calls was ignored . The probability for a specific spike count ( e . g . , one spike ) in that bin was defined as the number of trials having that spike count , divided by the total number of effective trials . A time-invariant p ( r|s = “no call” ) was estimated by drawing the same number ( e . g . , 216 ) of response time bins from random times in the spontaneous activity . The randomized detection information at time bin t was computed by randomly assigning the call and spontaneous trials to either the “call” or “no-call” stimulus . For discrimination , the probabilities were estimated from the trials to each individual call . The randomized discrimination information was computed by randomly assigning those trials to the different calls . For frequency ( duration ) information , the six calls lying within the same frequency ( duration ) range were grouped together , forming three different frequency ( duration ) ranges . It should be pointed out that when probability distributions must be inferred from finite trials , all information estimates are subject to bias [52] . Although techniques such as data-size scaling have been developed to minimize it , non-negligible biases can be present when there are large numbers of probable responses ( those with nonzero likelihood ) or a large number of stimuli , as in the case of call discrimination . This is probably why the examples in Figure 7L–7V show an offset in the time course of discrimination information long after the stimulus turned off ( for both the actual and randomized control information ) . Importantly , the biases implied by these late-period offsets were not necessarily the same as the biases at the time of the peak information , since bias is sensitive to the exact probability distribution of the responses . To convince ourselves that our conclusions based on the peak information values were not affected by this , we tested several different bias-correction methods . In addition to the data-size scaling procedure described above , we also checked our results using a data-size scaling procedure with ( 1 , 11/12 , . . . , 6/12 ) random partitions of the data ( again averaging together the estimates from 50 different randomizations ) . We also tried the so-called naïve-bias correction procedure described in [31] . By performing simulations that assumed the empirical response probability distribution from one of our recording sites was the true probability distribution , we found that the two data-size scaling procedures reasonably estimated the true discrimination information when this was above ~0 . 1 bits ( errors up to ~0 . 02 bits for true information near 0 . 1 bits and improving to <0 . 003 bits for progressively larger true information ) . The naïve-bias correction methods suffered greater systematic bias . We also applied the best-upper-bound ( BUB ) information estimate developed by Paninski [52] to our data but found this to yield generally much larger biases . Hence , we relied on the data-size scaling method and chose a threshold of 0 . 1 bits for the discrimination situation ( bias was not a concern in the other situations ) , above which we considered our information estimate to be valid . While higher thresholds could be chosen , we decided against this since it left too few sites in the naïve female group to yield statistically sound comparisons . It is important to note that even when the alternative bias-correction methods were applied to all our data , our conclusions were still the same . This is probably because our analysis was dependent not so much on the absolute value of the information but rather on the relative information values compared across the two animal groups . The differences between the two groups were apparently sufficiently large to emerge regardless of how we attempted to correct the bias . Our population comparisons were based on the MU recordings . To test whether our main conclusions would be sensitive to this , we looked at five sites in both mothers and naïve females that had the largest peak information ( considering detection and discrimination separately ) . Since the threshold level for detecting action potentials could affect how many units contributed to a recording , we systematically increased that level to reduce the overall spike counts for a site to approximately 50% , 25% , and 10% of the original recording . Thus , at the 10% level , we should only have been including the largest spikes at a site . When information was recalculated , the peak times in nearly all cases stayed the same , suggesting that the difference in timing observed between mothers and naïve females was not very sensitive to the MU nature of the data . In fact , the peak information times for SU A and SU B in Figures 9 and 10 agreed well with the MU distribution for mothers .
|
Like a student in a foreign country immersed in an unfamiliar language or a young mother trying to decipher her baby's cries , we all encounter initially meaningless sounds that in fact carry meaning . As these sounds gain significance , we become better at detecting and discriminating between them . How does this occur ? What happens in our brain to facilitate this improvement ? We explored these questions in a mouse model by measuring how neurons in the auditory cortex of female mice respond when the ultrasonic calls of mouse pups are played back to the animals . Earlier studies demonstrated that mothers , but not virgin females , recognize these calls as behaviorally significant . Our results indicate that the timing and magnitude of the auditory cortical responses to these communicative sounds differ between these two groups of female mice and that this difference may provide the auditory system in mothers with the capacity for detecting and discriminating pup calls . The results demonstrate that behavioral significance can be correlated with quantifiable functional improvements in the sensory cortical representation of a communication sound .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"computational",
"biology",
"mus",
"(mouse)",
"neuroscience"
] |
2007
|
Auditory Cortical Detection and Discrimination Correlates with Communicative Significance
|
Cortical networks , in-vitro as well as in-vivo , can spontaneously generate a variety of collective dynamical events such as network spikes , UP and DOWN states , global oscillations , and avalanches . Though each of them has been variously recognized in previous works as expression of the excitability of the cortical tissue and the associated nonlinear dynamics , a unified picture of the determinant factors ( dynamical and architectural ) is desirable and not yet available . Progress has also been partially hindered by the use of a variety of statistical measures to define the network events of interest . We propose here a common probabilistic definition of network events that , applied to the firing activity of cultured neural networks , highlights the co-occurrence of network spikes , power-law distributed avalanches , and exponentially distributed ‘quasi-orbits’ , which offer a third type of collective behavior . A rate model , including synaptic excitation and inhibition with no imposed topology , synaptic short-term depression , and finite-size noise , accounts for all these different , coexisting phenomena . We find that their emergence is largely regulated by the proximity to an oscillatory instability of the dynamics , where the non-linear excitable behavior leads to a self-amplification of activity fluctuations over a wide range of scales in space and time . In this sense , the cultured network dynamics is compatible with an excitation-inhibition balance corresponding to a slightly sub-critical regime . Finally , we propose and test a method to infer the characteristic time of the fatigue process , from the observed time course of the network’s firing rate . Unlike the model , possessing a single fatigue mechanism , the cultured network appears to show multiple time scales , signalling the possible coexistence of different fatigue mechanisms .
The spontaneous activity of excitable neuronal networks exhibits a spectrum of dynamic regimes ranging from quasi-linear , small fluctuations close to stationary activity , to dramatic events such as abrupt and transient synchronization . Understanding the underpinnings of such dynamic versatility is important , as different spontaneous modes may imply in general different state-dependent response properties to incoming stimuli and different information processing routes . Cultured neuronal networks offer a controllable experimental setting to open a window into the network excitability and its dynamics , and have been used intensively for the purpose . Neuronal cultures in early development phases naturally show alternating quasi-quiescent states and ‘network spikes’ ( NS ) of brief outbreaks of network activity [1–6] . In addition , recent observations in-vitro ( and later even in-vivo ) revealed a rich structure of network events ( ‘avalanches’ ) that attracted much attention because their spatial and temporal structure exhibited features ( power-law distributions ) reminiscent of what is observed in a ‘critical state’ of a physical system ( see e . g . [7 , 8] , and [9 , 10] and references therein ) . Generically , an avalanche is a cascade of neural activities clustered in time; while there persist ongoing debate on the relation between observed avalanches and whatever ‘criticality’ may mean for brain dynamics [11] , understanding their dynamical origin remains on the agenda . Quasi-synchronous NS , avalanches and small activity fluctuations are frequently coexisting elements of the network dynamics . Besides , as we will describe in the following , in certain conditions one can recognize network events which are clearly distinct from the mentioned network events , which we name here as ‘quasi-orbits’ . The abundant modeling literature on the above dynamical phenomena has been frequently focused on specific aspects of one of them [12 , 13]; on the other hand , getting a unified picture is made often difficult by different assumptions on the network’s structure and constitutive elements and , importantly , by different methods used to detect the dynamic events of interest . In the present paper we define a common probabilistic criterion to detect various coexisting dynamic events ( NS , avalanches and quasi-orbits ) and adopt it to analyze the spontaneous activity recorded from both cultured networks , and a computational rate model . Most theoretical models accounting for NS are based on an interplay between network self-excitation on one side , and on the other side some fatigue mechanism provoking the extinction of the network spike [12 , 13] . For such a mechanism two main options , up to details , have been considered: neural ‘spike-frequency adaptation’ [3 , 14] and synaptic ‘short-term depression’ ( STD ) [4 , 5 , 15–18] . Despite their differences , both mechanisms share the basic property of generating an activity-dependent self-inhibition in response to the upsurge of activity upon the generation of a NS , the more vigorously , the stronger the NS ( i . e . the higher the average firing rate ) . In this paper , we will mainly focus on STD , stressing the similarities of the two mechanisms , yet not denying their possibly different dynamic implications . While STD acts as an activity-dependent self-inhibition , the self-excitability of the network depends on the balance between synaptic excitation and inhibition; investigating how such balance , experimentally modifiable through pharmacology , influences the dynamics of spontaneous NSs is interesting and relevant as a step towards the identification of the ‘excitability working point’ in the experimental preparation . To study the factors governing the co-occurrence of different network events and their properties we adopt a rate model for the dynamics of the global network activity , that takes into accounts finite-size fluctuations and the synaptic interplay between one excitatory and one inhibitory population , with excitatory synapses being subject to STD . On purpose we implicitly exclude any spatial topology in the model , which is meant to describe the dynamics of a randomly connected , sparse network , since we intend to expose the exquisite implications of the balance between synaptic excitation and inhibition , and the activity-dependent self-inhibition due to STD . In doing this , we purposely leave out not only the known relevance of a topological organization [9 , 19 , 20] , but also the role of cliques of neurons which have been proposed to play a pivotal role in the the generation of NS as functional hubs [21] , as well as the putative role of ‘leader neurons’ . We perform a systematic numerical and analytical study of NSs for varying excitation/inhibition balance . The distance from an oscillatory instability of the mean-field dynamics ( in terms of the dominant eigenvalue of the linearized dynamics ) largely appears to be the sole variable governing the statistics of the inter-NS intervals , ranging from a very sparse , irregular bursting ( coefficient of variation CV ∼ 1 ) to a sustained , periodic one ( CV ∼ 0 ) . The intermediate , weakly synchronized regime ( CV ∼ 0 . 5 ) , in which the experimental cultures are often observed to operate , is found in a neighborhood of the instability that shrinks as the endogenous fluctuations in the network activity become smaller with increasing network size . Moreover , the model robustly shows the co-presence of avalanches with NS and quasi-orbits . The avalanche sizes are distributed according to a power-law over a wide region of the excitation-inhibition plane , although the crossing of the instability line is signaled by a bump in the large-size tail of the distribution; we compare such distributions and their modulation ( as well as the distributions of NS ) across the instability line with the experimental results from cortical neuronal cultures; again the results appear to confirm that neuronal cultures operate in close proximity of an instability line . Taking advantage of the fact that the sizes of both NS and quasi-orbits are found to be significantly correlated with the dynamic variable associated with STD ( available synaptic resources ) just before the onset of the event , we developed a simple optimization method to infer , from the recorded activity , the characteristic time-scales of the putative fatigue mechanism at work . We first tested the method on the model , and then applied it to in-vitro recordings; we could identify in several cases one or two long time-scales , ranging from few hundreds milliseconds to few seconds . Weak or no correlations were found instead between avalanche sizes and the STD dynamics; this suggests that avalanches originate from synaptic interaction which amplifies a wide spectrum of small fluctuations , and are mostly ineffective in eliciting a strong self-inhibition .
As originally described in [2] , cortical neurons were obtained from newborn rats within 24 hours after birth , following standard procedures . Briefly , the neurons were plated directly onto a substrate-integrated multielectrode array ( MEA ) . The cells were bathed in MEM supplemented with heat-inactivated horse serum ( 5% ) , glutamine ( 0 . 5 mM ) , glucose ( 20 mM ) , and gentamycin ( 10 μg/ml ) and were maintained in an atmosphere of 37°C , 5% CO2/95% air in a tissue culture incubator as well as during the recording phases . The data analyzed here was collected during the third week after plating , thus allowing functional and structural maturation of the neurons . MEAs of 60 Ti/Au/TiN electrodes , 30 μm in diameter , and spaced 200 μm from each other ( Multi Channel Systems , Reutlingen , Germany ) were used . The insulation layer ( silicon nitride ) was pretreated with poly-D-lysine . All experiments were conducted under a slow perfusion system with perfusion rates of ∼100 μl/h . A commercial 60-channel amplifier ( B-MEA-1060; Multi Channel Systems ) with frequency limits of 1–5000 Hz and a gain of 1024× was used . The B-MEA-1060 was connected to MCPPlus variable gain filter amplifiers ( Alpha Omega , Nazareth , Israel ) for additional amplification . Data was digitized using two parallel 5200a/526 analog-to-digital boards ( Microstar Laboratories , Bellevue , WA ) . Each channel was sampled at a frequency of 24000 Hz and prepared for analysis using the AlphaMap interface ( Alpha Omega ) . Thresholds ( 8× root mean square units; typically in the range of 10–20 μV ) were defined separately for each of the recording channels before the beginning of the experiment . The electrophysiological data is freely accessible for download at marom . net . technion . ac . il/neural-activity-data/ . A set of Wilson-Cowan-like equations [22] for the spike-rate of the excitatory ( νE ) and the inhibitory ( νI ) neuronal populations lies at the core of our dynamic mean-field model: τ E ν ˙ E = - ν E - Φ ( I E ) τ I ν ˙ I = - ν I - Φ ( I I ) , ( 1 ) where τE and τI represent two characteristic times ( of the order of few to few tens of ms ) , and Φ is the gain function of the input currents , IE and II , that in turn depend on νE , νI , and the synaptic efficacies . We chose Φ to be the transfer function of the leaky integrate-and-fire neuron under the assumptions of Gaussian , uncorrelated input of mean μ and infinitesimal variance σ2[23]: Φ [ μ , σ 2 ] ≡ [ π τ V ∫ V r e s e t − V r e s t − μ τ V σ 2 τ V V t h r e s h − V r e s t − μ τ V σ 2 τ V exp ( s 2 ) [ e r f ( s ) + 1 ] d s + τ r e f r a c t ] − 1 , ( 2 ) where τV is the membrane time constant , τrefract is a refractory period , and Vrest , Vreset , and Vthresh are respectively the rest , the post-firing reset , and the firing-threshold membrane potential of the neuron ( we assume the membrane resistance R = 1 ) . The model incorporates the non-instantaneous nature of synaptic transmission in its simplest form , by letting the νs being low-pass filtered by a single synaptic time-scale τ ˜: τ ˜ ν ˜ ˙ = ( ν − ν ˜ ) . ( 3 ) One can regard the variables ν ˜ s as the instantaneous firing rates as seen by post-synaptic neurons , after synaptic filtering . The form of Eq 3 and our choice of τ ˜ values ( see Table 1 ) implicitly neglects slow NMDA contributions and is restricted to AMPA and GABA synaptic currents . Thus , the input currents IE and II in Eq 1 will be functions of the rates νs through these filtered rates; with reference to Eq 2 , the model assumes the following form for the mean and the variance of the current IE ( the expressions for II are similarly defined ) : μE≡cnEν˜EwexcJEErE+cnIν˜IwinhJEI+νextJextσE2≡cnEν˜Ewexc2 ( JEE2+σJEE2 ) rE2+cnIν˜Iwinh2 ( JEI2+σJEI2 ) +νext ( Jext2+σJext2 ) , ( 4 ) where the nE and nI are the number of neurons in the excitatory and inhibitory population respectively; c is the probability of two neurons being synaptically connected; JEE ( JEI ) is the average synaptic efficacy from an excitatory ( inhibitory ) pre-synaptic neuron to an excitatory one , σ J 2 is the variance of the J-distribution; wexc and winh are dimensionless parameters that we will use in the following to independently rescale excitatory and inhibitory synapses respectively . Finally , an external current is assumed in the form of a Poisson train of spikes of rate νext driving the neurons in the network with average synaptic efficacy Jext . In Eq 4 rE ( t ) ( 0 < rE < 1 ) is the fraction of synaptic resources available at time t for the response of an excitatory synapse to a pre-synaptic spike; the evolution of rE evolves according to the following dynamics , which implements the effects of short-term depression ( STD ) [24 , 25] into the network dynamics: τ STD r ˙ E = ( 1 - r E ) - u STD r E τ STD ν ˜ E , ( 5 ) where 0 < uSTD < 1 represents the ( constant ) fraction of the available synaptic resources consumed by an excitatory postsynaptic potential , and τSTD is the recovery time for the synaptic resources . Finally , for a network of n neurons , we introduce finite-size noise by assuming that the signal the synapses integrate in Eq 3 is a random process νn of mean ν; in a time-bin dt , we expect the number of action potentials fired to be a Poisson variable of mean n ν ( t ) dt; Eq 3 will thus become: τ ˜ ν ˜ ˙ = ν n - ν ˜ ν n ≡ Poisson [ n ν dt ] n dt . ( 6 ) Putting all together , the noisy dynamic mean-field model is described by the following set of ( stochastic ) differential equations: τ E ν ˙ E = Φ μ E , σ E 2 - ν E τ I ν ˙ I = Φ μ I , σ I 2 - ν I τ ˜ E ν ˜ ˙ E = ν n E - ν ˜ E τ ˜ I ν ˜ ˙ I = ν n I - ν ˜ I τ STD r ˙ E = ( 1 - r E ) - u STD τ STD r E ν ˜ E ( 7 ) complemented by Eqs 2 , 4 and 6 . The values of all the fixed network parameters are shown in Table 1 . Since we will compare the dynamics of networks of different sizes , we scale the connectivity with network size in order to keep invariant the mean field equations: we hold the number of synaptic connection per neuron constant by rescaling , with reference to Eq 4 , the probability of connection c so that c nE and c nI are kept constant to the reference values that can be deduced from Table 1 . Spike-frequency adaptation ( SFA ) ( not present in simulations unless explicitly stated ) is introduced by subtracting a term to the instantaneous mean value of the IE current: μ E → μ E - g SFA c E ( t ) ( 8 ) proportional to the instantaneous value of the variable cE , that simply integrates νnE: τ SFA d c E d t = - c E + ν n E , ( 9 ) with a characteristic time τSFA . This additional term aims to model an after-hyperpolarization , Ca2+-dependent K+ current [26 , 27] . In this sense , cE can be interpreted as the cytoplasmic calcium concentration [Ca2+] ) , whose effects on the network dynamics are controlled by the value of the “conductance” gSFA . Simulations are performed by integrating the stochastic dynamics with a fixed time step dt = 0 . 25 ms . In the following , by “spike count” we will mean the quantity ν ( t ) n dt . For the detection of network events ( NSs , quasi-orbits , and avalanches ) we developed a unified approach based on Hidden Markov Models ( HMM ) [28] . Despite HMM have been widely used for temporal pattern recognition in many different fields , to our knowledge few attempts have been made to use them in the context of interest here [29 , 30] . For the purpose of the present description , we just remind that a HMM is a stochastic system that evolves according to Markov transitions between “hidden” , i . e . unobservable , states; at each step of the dynamics the visible output depends probabilistically on the current hidden state . Such models can be naturally adapted to the detection of network events , the observations being the number of detected spikes per time bin , and the underlying hidden states , between which the system spontaneously alternates , being associated with high or low network activity ( ‘network event—no network event’ ) . A standard optimization procedure adapts then the HMM to the recorded activity sample by determining the most probable sequence of hidden states given the observations . The two-step method we propose is based on HMM , has no user-defined parameters , and automatically adapts to different conditions . In the first step , the algorithm finds the parameters of the two-state HMM ( one low-activity state , representing the quasi-quiescent periods , and one high-activity state , associated with network events ) that best accounts for a given sequence of spike counts—the visible states in the HMM; such fitting is performed through the Baum-Welch algorithm [28] . In the second step , the most probable sequence for the two alternating hidden levels , given the sequence of spike counts and the fitted parameters , is found through the Viterbi algorithm . Network events are identified as the periods of dominance of the high activity hidden state . In order to retain only the most significant events a minimum event duration is imposed; such threshold is self-consistently determined as follows . The Viterbi algorithm is also applied to a “surrogate” time-series obtained by randomly shuffling the original one , thereby generating a set of “surrogate” events . The purpose is to determine the desired minimum event duration from the high duration tail of surrogate events ( which , by construction , come from a time-series with no real temporal structure ) . Since the high duration distribution tail is found to be roughly exponential , we fit such tail by considering only the surrogate events of duration larger than the 75th percentile . Then , from the fitted exponential , we compute the duration value such that the probability of durations greater than this value is P ( surrogate ) = 10−3 . In other words , we set the threshold on minimum duration of detected events to the duration of exceptionally long ( P < 10−3 ) surrogate events . As already remarked , we used essentially the same algorithm for detecting NS/quasi-orbits and avalanches . The only significant difference is that , in the case of avalanches , the emission probability of the low-activity hidden state is kept fixed during the Baum-Welch algorithm to p ( n ) ≃ δn0 ( δij is the Kronecker delta; p ( n ) is the probability of emitting n spikes in a time-bin ) . Thus the lower state is constrained to a negligible probability of outputting non-zero spike-counts , conforming to the intuition that in between avalanches the network is ( almost ) completely silent . More precisely , we set p ( 1 ) = 10−6 〈n〉 , where 〈n〉 is the average number of spikes that the network emits during a time-bin dt . After the modified Baum-Welch first step , avalanches are determined , as above , by applying the Viterbi algorithm; no threshold is applied in this case , neither to the avalanche duration nor to its size . The proposed procedures introduce three arbitrary parameters: the time bin dt , the probability P ( surrogate ) for network spikes and quasi-orbits , and the probability p ( 1 ) . To test the robustness of the algorithms , we varied these parameters over ample ranges: dt between 0 . 25 and 8 ms; P ( surrogate ) between 10−2 and 10−4; p ( 1 ) between 10−8 and 10−4 . We found that avalanche size distributions are virtually unaffected under variations of p ( 1 ) , and only mildly affected for the largest dt explored; higher values of P ( surrogate ) lead , as expected , to detect a larger number of small quasi-orbits , yet these additional events do not alter the overall shape of the size distribution predicted by the theory ( see next section ) ; on the other hand , a large number of very small quasi-orbits does have a detrimental effect on the correlation results reported in Section “Inferring the time-scales” . Simulations and data analysis have been performed using custom-written mixed C++/ MATLAB ( version R2013a , Mathworks , Natick , MA ) functions and scripts . The non-linear rate model described above can show a wide repertoire of dynamical patterns , as for example multiple stable fixed points and large , quasi-periodic oscillations . As we will show , for sufficiently excitable networks , a stable state of asynchronous activity ( fixed point ) is destabilized , in favor of stable global oscillations . Finite size noise probes differently network’s excitability at different distances from such instability . Before global oscillations become stable ( in the infinite network limit ) , the network’s highly non-linear reaction to its own fluctuations can ignite large , relatively stereotyped , “network spikes” . Also , in the proximity of the oscillatory ( Hopf ) instability , noise can promote “quasi orbits” , i . e . , transient departures from the fixed point which develop on time-scales dictated by the upcoming oscillatory instability , of which they are precursors . Under a linear approximation , the probability distribution of the amplitude l of these quasi-orbits can be explicitly derived as explained in the following . Consider a generic planar linear dynamics with noise: z ˙ = A z + σ ξ , ( 10 ) where A is 2 × 2 real matrix , and ξ = ( ξ ( t ) , 0 ) is a white noise with 〈ξ ( t ) ξ ( t′ ) 〉 = δ ( t − t′ ) . We here assume that the system is close to a Hopf bifurcation; in other words that the matrix A has complex-conjugated eigenvalues λ± = ℜλ + i ℑλ , with ℜλ < 0 and ∣ℜλ∣ ≪ ℑλ . By means of a linear transformation , the system can be rewritten as: x ˙ = ℜ λ x - ℑ λ y + σ x ξ y ˙ = ℑ λ x + ℜ λ y + σ y ξ , ( 11 ) with σx and σy constants determined by the coordinate transformation . Making use of Itō’s lemma to write: x 2 ˙ = 2 ℜ λ x 2 - 2 ℑ λ x y + σ x 2 + 2 x σ x ξ y 2 ˙ = 2 ℑ λ x y + 2 ℜ λ y 2 + σ y 2 + 2 y σ y ξ , and summing the previous two equations , we find for the square radius l2 ≡ x2 + y2 the dynamics: l 2 ˙ = 2 ℜ λ l 2 + σ ′ 2 + 2 σ x x + σ y y ξ , ( 12 ) with σ ′ 2≐σ x 2 + σ y 2 . As long as ℑλ ≫ ∣ℜλ∣ , it is physically sound to make the approximation: ( x ( t ) , y ( t ) ) = l ( 0 ) cos ( ℑ λ t + φ ) , sin ( ℑ λ t + φ ) , ( 13 ) for 0 ≤ t ≤ T = 2 π/ℑλ and then to average the variance of the noise over such period to get: l ( 0 ) 2 T ∫ 0 T σ x cos ( ℑ λ t + φ ) + σ y sin ( ℑ λ t + φ ) 2 d t = = l ( 0 ) 2 σ x 2 + σ y 2 2 = l ( 0 ) 2 σ ′ 2 2 . in order to rewrite Eq ( 12 ) as: l 2 ˙ = 2 ℜ λ l 2 + σ ′ 2 + 2 l σ ′ ξ . ( 14 ) Such stochastic differential equation is associated with the Fokker-Planck equation: ∂ t p ( l 2 , t ) = - ∂ l 2 [ 2 ℜ λ l 2 + σ ′ 2 ] p ( l 2 , t ) + + σ ′ 2 ∂ l 2 2 l 2 p ( l 2 , t ) ≡ L l 2 p ( l 2 , t ) ( 15 ) that admits an exponential distribution as stationary solution: p s s ( l 2 ) = 2 | ℜ λ | σ ′ 2 exp - 2 | ℜ λ | l 2 σ ′ 2 , ( 16 ) that is , a Rayleigh distribution for l: p s s ( l ) = 4 | ℜ λ | σ ′ 2 l exp - 2 | ℜ λ | l 2 σ ′ 2 . ( 17 ) On the other hand , we found a correlation between l ( the maximal departure from the low-activity fixed point ) and the duration of the quasi-orbit . Therefore the size of the quasi-orbit ( the ‘area’ below the firing rate time profile during the excursion from the fixed point ) is expected to scale as l2 , so that it should be exponentially distributed . For network spikes we do not have a theoretical argument to predict the shape of the size distribution , however empirically a ( left-truncated ) Gaussian distribution proved to be roughly adequate . Since we expect that quasi-orbits and NS contribute with different weights for varying excitatory/inhibitory balance , we adopted the following form for the overall distribution of network event size to fit experimental data: p ( x ) = p 0 τ 0 exp - ( x - x 0 ) τ 0 + 1 - p 0 2 π σ 1 exp - ( x - m 1 ) 2 2 σ 1 2 . ( 18 ) The parameters of the two distributions and their relative weight 0 ≤ p0 ≤ 1 are estimated by minimizing the log-likelihood on the data . A threshold for the event size is determined as the value having equal probability of being generated by either the exponential or the normal distribution . In the following , NSs are defined as events having size larger than this threshold . In those cases in which a threshold smaller than the peak of the normal distribution could not be determined , no threshold was set .
As the relative balance of excitation and inhibition is expected to be a major determinant of NS statistics we investigated first , for spontaneous NSs , how the inter-NS intervals ( INSI ) and their regularity ( as measured by the coefficient of variation , CVINSI ) depend on such balance . In Fig 2 we report the average INSI ( left panel ) and CVINSI ( right panel ) in the plane ( wexc , winh ) of the excitatory and inhibitory synaptic efficacies ( JEE → wE JEE , JIE → wE JIE , JEI → wI JEI , JII → wI JII , see Eq 4 ) . Starting from the center of this plane ( wexc = 1 , winh = 1 ) and moving along the horizontal axis , all the excitatory synapses of the network are multiplied by a factor wexc: moving right , the total excitation of the network increases ( wexc > 1 ) , toward left it decreases ( wexc < 1 ) . Along the vertical line , instead , all the inhibitory synapses are damped ( moving downward , winh < 1 ) or strengthened ( going upward , winh > 1 ) . It is clearly seen that both 〈INSI〉 and CVINSI are approximately distributed in the plane along almost straight lines of equal values: for a chosen 〈INSI〉 or CVINSI one can trade more excitation for less inhibition keeping the value constant , suggesting that , at this level of approximation , a measure of net synaptic excitation governs the NS statistics . Besides , not surprisingly , for high net excitation NSs are more frequent ( ∼ 1 Hz ) and quasi-periodic ( low CVINSI ) , due to the fact that the STD recovery time determines quasi-deterministically when the network is again in the condition of generating a new NS . Weak excitability , on the other hand , leads to rare NSs , approaching a Poisson statistics ( CVINSI ≃ 1 ) , since excitability is so low that fluctuations are essential for recruiting enough activation to elicit a NS , with STD playing little or no role at the ignition time; below an “excitation threshold” , NSs disappear . The solid lines in Fig 2 are derived from the linearization of the 5-dimensional dynamical system ( see Eq 7 ) , and are curves of iso-ℜλ , where λ is the dominant eigenvalue of the Jacobian: ℜλ = 0 Hz ( white line , signaling a Hopf bifurcation in the corresponding deterministic system ) , ℜλ = 3 . 5 Hz ( red line ) , and ℜλ = −3 . 5 Hz ( black line ) . Values of CV found in typical cultured networks are close to model results near the bifurcation line ℜλ = 0 Hz . We observe , furthermore , that such lines roughly follow iso-〈INSI〉 and iso-CVINSI curves , suggesting that a quasi one-dimensional representation might be extracted . We show in Fig 3 〈INSI〉 ( panel A ) and CVINSI ( panel B ) against ℜλ for the same networks ( circles ) of Fig 2 , and for a set of larger networks ( N = 8000 neurons , squares ) that are otherwise identical to the first ones , pointwise in the excitation-inhibition plane ( the average number of synaptic connections per neuron for the larger networks is kept constant to the value used in the original , smaller ones , as explained in Models and Analysis ) The difference in size amounts , for the new , larger networks , to weaker endogenous noise entering the stochastic dynamics of the populations’ firing rates ( see Eq 6 , second line ) . The points are seen to approximately collapse onto lines for both sets of networks , thus confirming ℜλ as the relevant control quantity for 〈INSI〉 and CVINSI . It is seen that , for the smaller networks , 〈INSI〉 and CVINSI depend smoothly on ℜλ , due to finite-size effects smearing the bifurcation . Also note the branch of points ( filled circles ) for which ℑλ = 0 and then no oscillatory component is present , corresponding to points in the extreme top-left region of the planes in Fig 2 . For the set of larger networks , the dependence of 〈INSI〉 and CVINSI on the ℜλ is much sharper , as expected given the much smaller finite-size effects; this shrinks the available region , around the instability line , allowing for intermediate , more biologically plausible values of CVINSI . We remark that NSs are highly non-linear and relatively stereotyped events , typical of an excitable non-linear system . The good predictive power of the linear analysis for the statistics of INSI signals that relatively small fluctuations around the system’s fixed point , described well by a linear analysis , can ignite a NS . Our mean-field , finite-size network is a non-linear excitable system which , to the left of the Hopf bifurcation line , and close to it , can express different types of excursions from the otherwise stable fixed point . Large ( almost stereotyped for high excitation ) NSs are exquisite manifestations of the non-linear excitable nature of the system , ignited by noise; the distribution of NS size ( number of spikes generated during the event ) is relatively narrow and approximately symmetric ( the Gaussian component of Eq 18 ) . Noise can also induce smaller , transient excursions from the fixed point which can be adequately described as quasi-orbits in a linear approximation . In fact , noise induces a probability distribution on the size of such events , which can be computed as explained in Methods and Analysis ( the exponential part in Eq 18 ) . Fig 4 , panel A , shows the activity of a simulated network ( blue line ) alongside with detected network events . We remark that the the different event types may not in general be easily distinguished on a single-event basis , while we argue that they are probabilistically distinguishable . From the best fit for the expected size distribution a threshold for the event size can be determined to separate events that are ( a-posteriori ) more probably quasi-orbits from the ones that are more probably NSs ( for details , see Models and Analysis ) . Following such classification , the green line in Fig 4 , panel A , marks the detection of two NSs ( first and third event ) and two quasi-orbits ( second and fourth event ) . We also emphasize that the existence of quasi-orbits is a specific consequence of the fact that in the whole excitation-inhibition plane explored for the model , the low-activity fixed point becomes unstable via a Hopf bifurcation . It is indeed known that for nonlinear systems in the proximity of a Hopf bifurcation , noise promotes precursors of the bifurcation , which appear as transient synchronization events ( see , e . g . , [31] ) . As one moves around the excitation-inhibition plane , to the left of the bifurcation line , the two types of events contribute differently to the overall distribution of network event sizes . Qualitatively , the farther from the bifurcation line , the higher the contribution of the small , “quasi-linear” events . This fact can be understood by noting that the average size of such events is expected to scale as 1/∣ℜλ∣ , where ℜλ is the real part of the dominant eigenvalue of the ( stable ) linearized dynamics ( see Models and Analysis , Eq 16 ) . The average size is furthermore expected to scale with the amount of noise affecting the dynamics , thus the contribution of quasi-linear events is also expected to vanish for larger networks . It has been previously reported that activity dynamics may be different from one network to the other , reflecting idiosyncrasies of composition and history-dependent processes ( [32] ) . Moreover , the dynamics of a given network , as well as its individual neurons , may shift over time ( minutes and hours ) between different modes of activity ( [32–34] ) . We therefore chose to demonstrate the efficacy of our analytical approach on two data sets of large-scale random cortical networks . In panels A-C of Fig 5 , we show the experimental distributions of event sizes for two cultured networks: panels A and B are ∼40-minute recordings taken from a very long recording for the same network; panel C is ∼1-hour recording from a different cultured network . By visual inspection , the distributions appear to be consistent with two components contributing with various weights , both for different periods of the same network , and for different networks . In the light of the above theoretical considerations , one is led to generate the hypothesis that the two components contributing to the overall distribution were associated with quasi-orbits and network spikes respectively; to test this hypothesis , we fitted ( solid lines in Fig 5 ) the experimental distributions with the sum of an exponential and a Gaussian distribution ( see Models and Analysis , Eq 18 ) , prepared to interpret a predominance of the exponential ( Gaussian ) component as a lesser ( greater ) excitability of the network . We remark that ( see panels A and B ) the relative weights of the two components appear to change over time for the same network , as if the excitability level would dynamically change; more on this at the end of this section . To substantiate the above interpretation of experimental results , we turned to long simulations ( about 5 . 5 hours ) of networks in different points in the excitation-inhibition plane ( Fig 2 ) , from which we extracted the distribution of network events and fitted them with Eq 18 as for experimental data ( see panels D-F in Fig 5 ) . Again , to the eye , the fits appear to be consistent with the two components variously contributing to the overall distribution , depending on the excitability of the network . If , however , the fits are subject to a Kolmogorov-Smirnov test , the test fails ( p < 0 . 01 ) for panels D and F . By inspecting the maximum distance between the cumulative distributions for simulation data and the fit , we found it at the lowest size bin for panel D , while the “Gaussian” part gives the greater mismatch for panel F . As for panel D , while the theoretical argument for the quasi-orbits clearly captures the shape of the size distributions , the way the test fails in the exponential part is interesting . In fact , network events cannot be detected with arbitrarily small size: in a way , the detection procedure imposes a soft threshold on the event size , below which the exponential distribution is not applicable . We can provide a rough estimate of such soft threshold as follows . A quasi-orbit duration is , to a first approximation , proportional to 1/ℑλ , which is of the order of few hundreds milliseconds not too far from the bifurcation line in the excitation-inhibition plane . Taking , for instance , 150 ms , an event will be detected if network activity within this time-span is larger than average ( typically few spikes per second per neuron; we take 3 for the present example ) : this leads to a soft threshold of about 100 spikes . This would be the lower limit of applicability of the exponential part of the distribution; this also explains the trough observed for very small sizes . As for the failure of the Kolmogorov-Smirnov test for the right part of the distribution in panel F , it should be remarked that the assumption of a Gaussian distribution for the size of network spikes , although generically plausible , is not grounded in a theoretical argument , and it’s not surprising that , on the order of 104 detected events , even a moderate skewness , as the one observed , can make the test fail . The fit for experimental data of panels A-C passed the Kolmogorov-Smirnov test ( p > 0 . 01 ) . As mentioned in the introduction , avalanches are cascades of neural activities clustered in time ( see Models and Analysis for our operational definition; examples of different methods used in the literature to detect avalanches can be found in [7 , 35–37] ) . Fig 4 , panel A and panel B , shows an example of the structure of the detected avalanches ( red lines ) in the model network . We extracted avalanches from simulated data , as well as from experimental data . For simulations , we choose data corresponding to three points in the ( wexc , winh ) plane of Fig 2 , with constant winh = 1 and increasing wexc , with the rightmost falling exactly over the instability line ( white solid line in Fig 2 ) . Three experimental data sets were extracted from different periods of a very long recording of spontaneous activity from a neural culture; each data set is a 40-minute recording . In Fig 6 we show ( in log-log scale ) the distribution of avalanche sizes for the three simulated networks ( top row ) and the three experimental ( bottom row ) data sets ( blue dots ) ; red lines are power-law fits [38] . From the panels in the top row we see that the distributions are well fitted , over a range of two orders of magnitude , by power-laws with exponents ranging from about 1 . 5 to about 2 . 2 , consistent with the results found in [7] . Note that in the cited paper the algorithm used for avalanche detection is quite different from ours , and the wide range of power-law exponents is related to their dependence on the time-window used to discretize data . In [39] ( adopting yet another algorithm for avalanche detection ) , both the shape of the avalanche distribution and the exponent vary depending on using pharmacology to manipulate synaptic transmission , over a range compatible with our model findings; notably , they find the slope of the power-law to be increasing with the excitability of the network , which is consistent with our modeling results . Panels B and C of Fig 6 clearly show the buildup of ‘bumps’ in the high-size tails , increasing with the self-excitation of the network; this is understood as reflecting the predominance of a contribution from NS and possibly quasi-orbits in that region of the distribution , on top of a persisting wide spectrum of avalanches . This feature also is consistent with the experimental findings of [39] , and has been previously shown in a theoretical model [40] for non-leaky integrate-and-fire neurons endowed with STD and synaptic facilitation . Turning to the plots in the bottom row of Fig 6 , we observe the following features: power-laws are again observed over two decades and more; in panels E and F , bumps are visible , similar to model findings; power-law exponents cover a smaller range just above 2 . While the sequence of plots in two rows ( modeling and experiment ) clearly shows similar features , we emphasize that experimental data were extracted from a unique long recording , with no intervening pharmacological manipulations affecting synaptic transmission; on the other hand , it has been suggested [41] that a dynamic modulation of the excitatory/inhibitory balance can indeed be observed in long recordings; although our model would be inherently unable to capture such effects , it is tempting to interpret the suggestive similarity between the theoretical and experimental distributions in Fig 6 as a manifestation of such changes of excitatory/inhibitory balance in time , of which the theoretical distributions would be a ‘static’ analog . To rule out the possibility that different behaviors in time could be due to intrinsic and global modifications in the experimental preparation , we checked ( see S1 Fig ) the waveforms of the recorded spikes across all MEA electrodes , comparing the earliest and latest used recordings ( about 40 minutes each , separated by about 34 hours ) . In most cases the waveforms for the two recordings are remarkably similar , and when they are not , no systematic trend in the differences is observed . If our interpretation is correct , the experimental preparation operates below , and close , to an oscillatory instability; on the other hand , contrary to NS , the appearance of avalanches does not seem to be exquisitely related to a Hopf bifurcation , rather they seem to generically reflect the non-linear amplification of spontaneous fluctuations around an almost unstable fixed point—a related point will be mentioned in the next section . We also remark that we obtain power-law distributed avalanches in a ( noisy ) mean-field rate model , by definition lacking any spatial structure; while the latter could well determine specific ( possibly repeating ) patterns of activations ( as observed in [19] ) , it is here suggested to be not necessary for power-law distributed avalanches . The avalanche size distribution for the same network as in Fig 5 , panel C , is sparser but qualitatively compatible with the distribution in Fig 6 , panel F ( see S2 Fig ) ; in particular , the distribution shows a prominent peak for high-size avalanches , consistently with the interpretation , given in connection with Fig 5 , of high excitability . We do not provide examples of avalanche and NS-quasi orbits size distributions in the super-critical region on the right of the Hopf bifurcation line in Fig 2; this is because the phenomenology in that region is relatively stereotyped and easy to guess/understand: the high excitability of the network generates , moving on the right of the bifurcation line , increasingly stereotyped network spikes , which dominate the size distribution of the network events ( see S3 Fig , panel A ) ; even though finite-size fluctuations blur the bifurcation line , quasi-orbits are expected to contribute very little in the supercritical region; the distribution of avalanche sizes is increasingly dominated by the high-size bump associated with network spikes ( see S3 Fig , panel B ) . The fatigue mechanism at work ( STD in our case ) is a key element of the transient network events , in its interplay with the excitability of the system . While the latter can be manipulated through pharmacology , STD itself ( or spike frequency adaptation , another neural fatigue mechanism ) cannot be directly modulated . It is therefore interesting to explore ways to infer relevant properties of such fatigue mechanisms from the experimentally accessible information , i . e . the firing activity of the network . We focus in the following on deriving the effective ( activity-dependent ) time scale of STD from the sampled firing history . The starting point is the expectation that the fatigue level just before a NS should affect the strength of the subsequent NS . We therefore measured the correlation between r ( fraction of available synaptic resources ) and the total number of spikes emitted during the NS ( NS size ) from simulations . We found that the average value of r just before a NS is an effective predictor of the NS size , the more so as the excitability of the network grows . Based on the r-NS size correlation , we took the above “experimental” point of view , that only the firing activity ν is directly observable , while r is not experimentally accessible . Furthermore , the success of the linear analysis for the inter-NS interval statistics ( due to the NS being a low-threshold very non-linear phenomenon ) , suggests that without assuming a specific form for the dynamics of the fatigue variable f , we may tentatively adopt for it a generic linear integrator form , of which we want to infer the characteristic time-scale τ*: f ˙ = - f τ * + ν ( t ) ( 19 ) To do this , first we reconstruct f ( t ) from ν ( t ) for a given τ* then we set up an optimization procedure to estimate τ optim * , based on the maximization of the ( negative ) f-NS size correlation ( a strategy inspired by a similar principle was adopted in [11] ) . Fig 7 , panel A , shows an illustrative example of how the correlation peaks around the optimal value . As a reference , the dotted line marks the value below which 95% of the correlations computed from surrogate data fall; surrogate data are obtained by shuffling the values of f at the beginning of each NS . We remark that in this analysis we use both NS and quasi-orbit events ( which are both related to the proximity to a Hopf bifurcation ) . This is reasonable since we expect to gain more information about the anti-correlation between f and NS size by including both types of large network events . Although the procedure successfully recovers a maximum in the correlation , the value of τ optim * ( 0 . 58 s ) reported in Fig 7 , panel A , is not close to the value of τSTD ( 0 . 8 s ) . Yet this is expected , since in Eq 19 , τ* will in general depend on τSTD and other parameters of the dynamics , but also on the point around which the dynamics is being linearized , more precisely on the average activity 〈ν〉 . Specifically , when the fatigue variable follows the Tsodyks-Markram model of STD ( which of course was actually the case in the simulations ) , linearizing the dynamics of r around a fixed point 〈r〉 ( 〈r〉 = 1/ ( 1 + uSTD 〈ν〉 τSTD ) ) , r behaves as a simple linear integrator with a time-constant: τ optim * = τ STD ⟨ r ⟩ = τ STD 1 + u STD ⟨ ν ⟩ τ STD ( 20 ) that depends on τSTD , uSTD , and 〈ν〉 . To test this relationship , we performed the optimization procedure for each point of the excitation-inhibition plane . The optimal τ* values across the excitation-inhibition plane against 〈ν〉 are plotted in Fig 7 , panel B ( dots ) . The solid line is the best fit of τSTD and uSTD from Eq 20 , which are consistent with the actual values used in simulations . This result is suggestive of the possibility of estimating from experiments the time-scale of an otherwise inaccessible fatigue variable , by modeling it as a generic linear integrator , with a “state dependent” time-constant . Fig 8 shows the outcome of the same inference procedure for two segments of experimental recordings . The plot in panel A is qualitatively similar to panel A in Fig 7: although the peak is broader and the maximum correlation ( in absolute value ) is smaller , the τ* peak is clearly identified and statistically significant ( with respect to surrogates , dotted line ) , thus suggesting a dominant time scale for the putative underlying , unobserved fatigue process . However , Fig 8 , panel B , clearly shows two significant peaks in the correlation plot; it would be natural to interpret this as two fatigue processes , with time scales differing by an order of magnitude , simultaneously active in the considered recording segment . To test the plausibility of this interpretation , we simulated networks with simultaneously active STD and spike-frequency adaptation ( SFA , see Models and Analysis ) . Fig 9 shows the results of time scale inference for two cases sharing the same time scale for STD ( 800 ms ) and time scale of SFA differing by a factor of 2 ( τSFA = 15 and 30 s respectively ) . In both cases the negative correlation peaks at around τ* ≃ 500 ms; this peak is plausibly related to the characteristic time of STD , consistently with Fig 7 . The peaks at higher τ*s , found respectively at 12 and 22 s , roughly preserve the ratio of the corresponding τSFA values . This analysis provides preliminary support to the above interpretation of the double peak in Fig 8 , right panel , in terms of two coexisting fatigue processes with different time scales . We also checked to what extent the avalanche sizes were influenced by the immediately preceding amount of available synaptic resources r , and we found weak or no correlations; this further supports the interpretation offered at the end on the previous section , that avalanches are a genuine manifestation of the network excitability which amplifies a wide spectrum of small fluctuations .
Several works recently advocated a key role of specific network connectivity topologies in generating ‘critical’ neural dynamics as manifested in power-law distributions of avalanches size and duration ( see [20 , 42] ) . Also , it has been suggested that ‘leader neurons’ , or selected coalitions of neurons , play a pivotal role in the onset of network events ( see e . g . [21 , 43–45] ) . While a role of network topology , or heterogeneity in neurons’ excitability , is all to be expected , we set out to investigate what repertoire of network events is accessible to a network with the simplest , randomly sparse , connectivity , over a wide range of excitation-inhibition balance , in the presence of STD as an activity-dependent self-inhibition . In the present work we showed that network spikes , avalanches and also large fluctuations we termed ‘quasi-orbits’ coexist in such networks , with various relative weights and statistical features depending on the excitation-inhibition balance , which we explored extensively , including the role of finite-size noise ( irregular synchronous regimes in balanced excitatory-inhibitory networks has been studied in [35] ) . We remark in passing that the occurrence of quasi-orbits is primarily related to the proximity to a Hopf bifurcation for the firing rate dynamics; on the other hand , the occurrence of NS and , presumably , avalanches , does not necessarily require this condition: for instance , NS can occur in the proximity of a saddle-node bifurcation , where the low-high-low activity transitions derive from the existence of two fixed points , the upper one getting destabilized by the fatigue mechanism ( see e . g . [46 , 47] ) ; notably , in [12] the authors find that , in a network of leaky integrate-and-fire neurons endowed with STD , when a saddle-node separates an up- and a down-state , the dynamics develops avalanches during up-state intervals only . We also remark that , with respect to the power-law distribution of avalanches , it is now widely recognized that while criticality implies power-law distributions , the converse is not true , which leaves open the problem of understanding what is actually in operation in the neural systems observed experimentally ( for a general discussion on the issues involved , see [48] ) . In the present work , we do not commit ourselves to the issue of whether avalanches could be considered as evidence of Self-Organized Criticality . In summary , the main contributions of the present work can be listed as follows . We present a low-dimensional network model , derived from the mean field theory for interacting leaky integrate-and-fire neurons with short-term depression , in which we include the effect of finite-size ( multiplicative ) noise . At the methodological level we introduce a probabilistic model for events detection , and a method for inferring the time-scale ( s ) of putative fatigue mechanisms . At the phenomenological level we recognize the existence of quasi-orbits as an additional type of network event , we show the coexistence of quasi-orbits , network spikes , and avalanches , and study their different mixing depending on the excitability of the network . We also offer a theoretical interpretation of the phenomenology , through a bifurcation analysis of the mean-field model , and a prediction on the effect of noise in the proximity of a Hopf bifurcation .
|
The spontaneous neural activity is the dynamic floor on which the cortex builds its response to incoming stimuli and organizes its information processing , thereby the importance of understanding its dynamical underpinnings . In-vitro preparations , as well as the intact cortex in deep sleep or anesthesia , display a variety of spontaneous collective events , including quasi-synchronous ‘network spikes’ and a complex spectrum of ‘avalanches’ , which has been considered suggestive of a ‘typically critical’ state . Light has been shed on selected aspects of such events; still , a unified picture stays elusive , also due to varying statistical definitions of network events . Our work aims to take a step in this direction . We first introduce a probabilistic definition of population events that naturally adapts to different scales of analysis; it reveals , in the activity of cultured networks , as well as in a simple rate model , the co-occurrence of network spikes , ‘quasi-orbits’ and avalanches . Model’s analysis suggests that their emergence is governed by a single parameter measuring the proximity to an oscillatory instability , where the network can amplify fluctuations on a wide range of scales in space and time . We also propose a procedure to infer from neural activity the slow underlying time-scales of the dynamics .
|
[
"Abstract",
"Introduction",
"Models",
"Results",
"Discussion"
] |
[] |
2015
|
Network Events on Multiple Space and Time Scales in Cultured Neural Networks and in a Stochastic Rate Model
|
The prevailing paradigm of T lymphocyte control of viral replication is that the protective capacity of virus-specific CD8+ T cells is directly proportional to the number of functions they can perform , with IL-2 production capacity considered critical . Having recently defined rapid perforin upregulation as a novel effector function of antigen-specific CD8+ T cells , here we sought to determine whether new perforin production is a component of polyfunctional CD8+ T cell responses that contributes to the control of several human viral infections: cytomegalovirus ( CMV ) , Epstein-Barr virus ( EBV ) , influenza ( flu ) , and adenovirus ( Ad ) . We stimulated normal human donor PBMC with synthetic peptides whose amino acid sequences correspond to defined CTL epitopes in the aforementioned viruses , and then used polychromatic flow cytometry to measure the functional capacity and the phenotype of the responding CD8+ T cells . While EBV and flu-specific CD8+ T cells rarely upregulate perforin , CMV-specific cells often do and Ad stimulates an exceptionally strong perforin response . The differential propensity of CD8+ T cells to produce either IL-2 or perforin is in part related to levels of CD28 and the transcription factor T-bet , as CD8+ T cells that rapidly upregulate perforin harbor high levels of T-bet and those producing IL-2 express high amounts of CD28 . Thus , “polyfunctional” profiling of antigen-specific CD8+ T cells must not be limited to simply the number of functions the cell can perform , or one particular memory phenotype , but should actually define which combinations of memory markers and functions are relevant in each pathogenic context .
Understanding the mechanisms by which human T cells provide effective control of pathogens is important for designing interventions against those that persist to cause severe morbidity and/or mortality . T cells generally limit the replication of Epstein Barr virus ( EBV ) [1] , [2] , Cytomegalovirus ( CMV ) [3] , [4] , [5] , and Hepatitis viruses B[6] , [7] , [8] and C[9] , [10] , but only rarely of the Human Immunodeficiency Virus ( HIV ) , as the majority of HIV infections inevitably result in progressive disease . Cytotoxic T lymphocytes ( CTL ) are thought to be a primary mediator of viral control , due in large part to their ability to recognize and eliminate virally infected autologous cells . Although CD8+ T cells respond to viral infection with a plethora of effector functions , the identification of a definite immune correlate of protection has not been forthcoming for any human pathogen . Recent strategies of assessing human antiviral T cell responses focus on the quality of the T cell response , defined by its polyfunctional nature . Briefly , the more effector functions that constitute the overall response , the more protective the response is considered[11] , [12] . Typically , the functions quantified simultaneously include upregulation of interferon gamma ( IFN-γ ) and interleukin-2 ( IL-2 ) [13] , [14] , [15] . A more elaborate assessment of the T cell response may include a measurement of tumour necrosis factor alpha ( TNF-α ) , a chemokine such as MIP-1β , and degranulation measured by CD107a exposure . A high frequency , multi-functional CD4+ T cell response composed of IFN-γ , IL-2 , and TNF-α provides protection against Leishmania major infection in mice[16] , however a similar correlation in humans for antiviral CD8+ T cells has not been formally proven . This is likely because none , or any combination , of these functions may directly inhibit pathogen replication . CTL clear virally infected target cells primarily via the exocytosis of cytotoxic granules containing granzymes and perforin[17] , [18] , [19] , [20] . The manifestations of genetic mutation or deletion of perforin are impaired cellular cytotoxicity and profound immunodeficiency[21] , [22] . We have recently shown that human CD8+ T cells can rapidly upregulate perforin de novo after antigen-specific stimulation[23] , which is immediately transported to the immunological synapse where it likely potentiates cytotoxicity[24] . The measurement of new perforin is different from that of pre-formed perforin stored in cytotoxic granules , in that it indicates the potential of the cell to rapidly reconstitute its cytotoxic nature . In contrast , the assessment of pre-formed perforin in granules indicates immediate killing potential , but likely does not predict the sustainability of the cytotoxic response . Thus , analyzing this novel aspect of T cell functionality could provide new insight into how CD8+ T cells mediate pathogenic control . Here we examine perforin upregulation ability in the context of polyfunctional CD8+ T cell responses to several common human viral pathogens: Cytomegalovirus ( CMV ) , Epstein-Barr virus ( EBV ) , Adenovirus ( Ad ) , or Influenza ( flu ) . Infection by EBV , CMV , flu , or Ad stimulates robust memory T cell responses that are associated with protection from viral pathogenesis . However , each course of infection is different: CMV establishes latency but remains lytically active , thereby creating a constant supply of antigen to the immune system , whereas EBV enters the lytic phase infrequently after establishing latency . Thus , EBV-specific CD8+ T cells likely only receive periodic restimulation . Primary Ad and flu infections are quickly resolved by the host immune response , but since Ad may become persistent , and there are many Ad serotypes whose sequences are highly conserved[25] , [26] , Ad-specific CD8+ T cells are likely repeatedly stimulated . In contrast , flu infections are seasonal and readily cleared , thus flu-specific CD8+ T cell restimulation is likely more intermittent than that for other viruses . We show that the measurement of perforin upregulation redefines our interpretation of polyfunctional CD8+ T cell responses and memory phenotypes that we associate with control of these pathogens , and represents a novel correlate of antiviral immunity that should be considered in assessments of human antiviral CD8+ T cell responses .
While IFN-γ production is commonly measured to identify virus-specific CD8+ T cells , it is unclear whether or not it represents a true correlate of immune protection for EBV , CMV , flu , or Ad . We therefore assessed the capacity of IFN-γ producing CD8+ T cells to perform other functions which might be associated with viral control , including rapid perforin upregulation , TNF-α , IL-2 , and degranulation . We recently characterized rapid perforin upregulation as a novel function of antigen-specific CD8+ T cells[24] , the measurement of which indicates the cells' potential to sustain cytotoxicity . Briefly , newly produced perforin can be detected by a specific anti-perforin antibody ( clone D48 ) which recognizes both pre-formed perforin stored in cytotoxic granules as well as new perforin that has been rapidly produced in response to antigenic stimulation . In contrast , a second perforin antibody ( clone δG9 ) primarily recognizes perforin within cytotoxic granules . By quantifying perforin with the D48 antibody together with another function such as IFN-γ , it is possible to discriminate new perforin from pre-formed granule-associated perforin in activated CD8+ T cells . As shown in Figure 2A ( top row ) , the activated cells that produced IFN-γ but failed to degranulate possessed the most perforin ( Q3 , green ) . Activated CD8+ T cells that both degranulated and upregulated IFN-γ harbored an intermediate amount of perforin ( Q2 , blue ) . Perforin was essentially absent in degranulating cells that failed to also produce IFN-γ ( Q1 , red ) . The Q1 and Q2 populations were both degranulating to the same degree , yet they represent responding cells that differentially upregulate perforin production . This may simply be an issue of kinetics , in that the cells that only degranulate may not have yet upregulated perforin , or signify a truly separate subpopulation that cannot upregulate perforin . Similarly , the IFN-γ producing responder population ( Q2 and Q3 ) is divided into distinct functional ( perforin ± CD107a ) subsets . In contrast , the δG9 antibody , specific only for granule-associated perforin , failed to detect perforin in any of the functional subpopulations ( Figure 2A , bottom row ) . Together , these data indicate that antigen-specific human CD8+ T cells are capable of upregulating perforin rapidly after stimulation , in the absence of cellular proliferation , and without the addition of exogenous cytokines or other co-factors . We next examined whether rapid perforin upregulation was characteristic of EBV , CMV , flu , and Ad-specific CD8+ T cell responses in our subject cohort . First , we determined what proportion of every antigen-specific IFN-γ response also upregulated perforin in each donor . A representative example is shown in Figure 2B ( left panel ) , where 87 . 2% of the IFN-γ producing cells concomitantly upregulated perforin . The cohort results are illustrated in the right panel of Figure 2B . Whereas nearly all IFN-γ+ Ad-specific CD8+ T cells upregulated perforin [orange group , median ( black bar ) = 73 . 6%] , those responding to EBV and flu displayed limited perforin upregulation [blue group , median = 4 . 95% and black group , median = 0% , respectively] . Only donor U mounted a substantial EBV-specific perforin response ( 58 . 3% of the IFN-γ+ CD8+ T cells ) . CMV-specific perforin upregulation was highly variable between subjects [red group , median ( black bar ) = 25 . 1% , range: 5 . 39%–78 . 0%]; while some donors exhibited strong perforin upregulation ( donors D = 78 . 0% and E = 60 . 9% ) , others were more limited [donor B , red open diamond: CMV peptide 20 = 5 . 39% , CMV peptide 23 = 12 . 8%] . Polyclonal SEB stimulation also resulted in varying degrees of perforin responsiveness [green group , median = 26 . 0% , range: 1 . 02%–62 . 8%] . Thus , immediate perforin upregulation is not an effector function common to all antigen-specific CD8+ T cells; rather it seems to be characteristic of CD8+ T cells specific for particular viral infections . Since perforin is typically expressed with other cytotoxic proteins[19] , [27] , [28] , we assayed for concomitant granzyme B and perforin upregulation in response to SEB stimulation in a small cohort of normal PBMC donors . As depicted in Figure 2C , perforin and granzyme B upregulation are tightly linked functions of responsive CD8+ T cells [r = 0 . 9135 , 95% C . I . = 0 . 5861 to 0 . 9845 , two tailed p value = 0 . 0015; Pearson correlation] . Thus , it is reasonable to infer CD8+ T cells that upregulate perforin are also producing new granzyme B . Next , we examined the capacity of virus-specific IFN-γ+ CD8+ T cells to also produce IL-2 ( Figure 3A ) . In contrast to perforin , IL-2 production was elevated in EBV- and flu-specific CD8+ T cells [EBV: blue group , median = 71 . 7% , flu: black group , median = 55 . 7% , respectively] , whereas in CMV and Ad it was much lower [CMV: red group , median ( black bar ) = 39 . 4% , Ad: orange group , median = 0 . 21% , respectively] . Finally , we analyzed degranulation capacity using the lysosomal and granule resident marker CD107a[29] , [30] , as well as TNF-α production [Figure 2B and 2C , respectively] . The CD107a response pattern across all the viral settings was similar to that of perforin: CMV and Ad stimuli induced strong CD8+ T cell degranulation [Figure 2B; CMV: red group , median ( black bar ) = 65 . 05% , Ad: orange group , median = 87 . 2%] , whereas EBV and flu did to a lesser extent [EBV: blue group , median = 37 . 2% , flu: black group , median = 25 . 8%] . Unlike both CD107a and perforin , TNF-α production was ubiquitously expressed , as nearly every virus-specific IFN-γ+ CD8+ T cell also produced TNF-α [Figure 2C; CMV: red group , median ( black bar ) = 94 . 6%; EBV: blue group , median ( black bar ) = 91 . 35%; Ad: orange group , median ( black bar ) = 83 . 0%; flu: black group , median ( black bar ) = 84 . 6%] . Taken together , these results suggest that there are substantially different CD8+ T cell functional profiles against CMV , EBV , Ad , and flu , and that no single function ( or pair of functions ) likely defines a universal correlate of immune protection for all of these viruses . We next characterized the polyfunctionality of the virus-specific CD8+ T cells from each donor to see if a particular response profile ( s ) was consistently detected in all viral contexts . We grouped donor responses according to viral specificity , and then assessed the average CD8+ T cell polyfunctional profile specific for that viral infection . As shown in Figure 4A , each viral antigen stimulated a unique functional profile consisting of varying degrees of polyfunctionality . Perforin production ( designated by purple arcs around the pies ) dominated the Ad-specific response profile compared to the other viral stimulations , and was highly expressed in a 4+ population ( orange pie slice ) together with CD107a , IFN-γ , and TNF-α . In the case of CMV , perforin upregulation was somewhat less dominant , but was similarly co-expressed with CD107a , IFN-γ , and TNF-α ( Figure 4B ) to form a substantial 4+ population ( Figure 4A , orange pie slice ) . EBV also generated a highly multi-functional response , however the 4+ population ( orange pie slice ) was composed entirely of an IL-2+CD107a+IFN-γ+TNF-α+ CD8+ T cell subset ( Figure 4B ) . IL-2 production also dominated the 4+ polyfunctional profile of flu , as the IL-2+CD107a+IFN-γ+TNF-α+ subset was again the principal multi-functional population ( Figure 4B ) . In fact , as depicted by the arcs around the pies in Figure 4A , it appears that IL-2 production ( black arcs ) and perforin upregulation ( purple arcs ) generally are not co-expressed within any polyfunctional population . Thus , while every virus stimulated a high frequency of CD8+ T cells capable of four effector functions simultaneously , CMV and Ad induced a perforin driven 4+ responder population , whereas EBV and flu preferentially stimulated an IL-2 dominated 4+ subset . Strikingly , none of the virus-specific CD8+ T cell response profiles included a 5+ subset , suggesting that responding CD8+ T cells rarely upregulate perforin and IL-2 simultaneously . To further explore this possibility , we plotted the proportion of antigen-specific IFN-γ+ cells producing either new perforin or IL-2 . As depicted in Figure 5A , a statistically significant inverse correlation exists between IL-2 and perforin positivity in virus-specific IFN-γ+ CD8+ T cells ( r = −0 . 5684 , 95% C . I . = −0 . 7604 to −0 . 2849 , p<0 . 0005; Pearson correlation ) . A strong correlation also results if only SEB-induced responses are considered ( Figure 5B; r = −0 . 6011 , 95% C . I . = −0 . 8244 to −0 . 2159 , p<0 . 0051; Pearson correlation ) . We performed our analysis on total CD8+ T cells , even though naïve cells preferentially produce IL-2 over IFN-γ and perforin . In our data set , however , the contribution of IL-2 from naïve CD8+ T cells in response to SEB stimulation is minimal compared to the antigen-experienced cells ( not shown ) , and does not change the relationship we observe between IL-2 and perforin . Thus , although not absolute , simultaneous production of IL-2 and perforin within the same CD8+ T cell , or within a virus-specific CD8+ T cell population , is exceptionally rare , suggesting a mutually exclusive relationship between these functions . There are several precedents characterizing the functional attributes of particular CD8+ T cell memory phenotypes in the context of specific viral infections[31] , [32] , [33] , [34] , however the basis for these differences remains unknown . We investigated the potential role of two cellular factors in determining the preferential expression of either perforin or IL-2 by virus-specific human effector CD8+ T cells: CD28 , a co-receptor whose signaling is critical for the induction of IL-2 production[35] , [36] , and T-bet , the T-box transcription factor associated with effector function[37] , [38] , [39] . As illustrated in Figure 6A , CD28 is commonly detected on IL-2 producing CD8+ T cells ( mean = 65 . 4% , SEM = 6 . 125 , 95% C . I . 52 . 2–78 . 5% ) , whereas its expression is significantly lower on those upregulating perforin ( mean = 19 . 7% , SEM = 5 . 047 , 95% C . I 8 . 85–30 . 5%; p = <0 . 0001 , Paired t-test ) . Within all subjects tested ( each represented by a unique symbol ) , CD28 expression was always higher on the antigen-specific CD8+ T cells ( each stimuli represented by a unique colour ) producing IL-2 than their counterparts upregulating perforin ( Figure 6A ) . A phenotypic evaluation of IL-2 producing and perforin upregulating cells reveals that the former cells bear relatively high levels of CD27 and CD28 but low levels of CD57 , whereas the latter cells are mostly CD27+/−CD28loCD57hi ( Supplementary Figure S1 ) . Thus , CD28 , which is important mechanistically for IL-2 production , is not commonly detected on CD8+ T cells that are rapidly upregulating perforin . Although T-bet has been linked to the development of TH1 responses and effector function in murine CD4+ and CD8+ T cells , respectively[38] , [39] , a similar relationship has yet to be formally demonstrated in humans . The only possible exception is a clinical study of ICOS-deficient sibling patients in whom impaired CD8+ T cell effector function and decreased development of memory T cell populations was indirectly linked to T-bet [40] . As illustrated in Figure 6B , we first examined the levels of T-bet in resting human CD8+ T cell memory subsets directly ex vivo . Effector cells ( CCR7−CD45RO− ) , and to a lesser degree effector memory cells ( CCR7−CD45RO+ ) , exhibited concordant levels of perforin and T-bet , whereas both factors were absent in naïve ( CCR7+CD45RO− ) and central memory cells ( CCR7+CD45RO+ ) . We then stimulated the PBMC from the same donor with SEB to activate all the functional subsets and assessed T-bet expression in the fraction of CD8+ T cells that rapidly upregulates perforin , compared to that producing IL-2 ( Figure 6C ) . T-bet expression was most pronounced in perforin-producing cells compared to both IL-2 producing cells and naïve CD8+ T cells , although IL-2 producing cells did harbor more T-bet than naïve CD8+ T cells . Overall , in 6 separate individuals we observed T-bet in a higher proportion of CD8+ T cells upregulating perforin than in those producing IL-2 ( Figure 6D: median = 91 . 75% vs . 58 . 1% , mean = 89 . 05±7 . 77% vs . 57 . 25±20 . 93% , p = 0 . 0068 , Paired t-test ) . Furthermore , CD8+ T cells that rapidly upregulate perforin express more T-bet on a per cell basis than those that produce IL-2 ( Figure 6E: median = 1476 vs . 893 . 5 , mean = 1491±205 . 9 vs . 967 . 2±248 . 7 median fluorescence intensity , p = 0 . 004 , Paired t-test ) . In conclusion , the preferential expression of the transcription factor T-bet in CD8+ T cells that rapidly upregulate perforin over those that produce IL-2 supports a significant role for T-bet in the differentiation of antigen-specific human CD8+ T cells into cytotoxic effector cells . Furthermore , the expression of CD28 co-receptor is correlated to IL-2 production .
What defines the “optimal” CD8+ T cell polyfunctional profile for viral infections in humans ? The data we have presented here suggest that based upon the characteristics of replication , latency , persistence , and antigen load , every virus will potentially stimulate multiple polyfunctional profiles distinct from those of other viral infections . Here we examined four different viral infections , each of which is controlled or eliminated at least in part by viral-specific CD8+ T cells , and for each of these viral specificities we have found unique polyfunctional profiles . At the simplest level , it appears that rapid perforin upregulation and IL-2 production define complementary functional CD8+ T cell subsets that bear unique phenotypic profiles and predominate according to antigenic burden . It has long been appreciated that CD8+ T cells play a pivotal role in the direct elimination of virally infected cells , and that perforin is a key mediator of this process through its distinct ability to enable the entry of apoptosis-inducing granzymes[17] , [20] . We previously demonstrated that virus-specific CD8+ T cells rapidly upregulate perforin after activation and then target the protein directly to the interface between the CTL and its target[24] . This sustained production and targeted release of new perforin after stimulation may allow the CD8+ T cell to recognize and kill additional targets after the initial release and depletion of pre-formed perforin stored in cytotoxic granules . The measurement of new perforin production is significant because it serves as a gauge of a CD8+ T cells' potential to repeatedly eliminate infected host cells and , hence , control viral pathogenesis . Here we show that rapid perforin upregulation is a highly specialized ability not common to all CD8+ T cells . Rather , it appears to be tied to the antigenic history of the cell . Whereas perforin upregulation and degranulation are commonly associated functions of CTL , there appears to be a mutually exclusive relationship between new perforin and IL-2 upregulation . Rapid perforin upregulation ability is not commonly observed against influenza and EBV . The fact that these pathogens establish latency ( EBV ) or are rapidly cleared ( flu ) suggests that antigen load or continual antigen exposure may in part maintain perforin upregulation ability and drive effector phenotype differentiation . For these viruses , proliferation and/or consequent differentiation of the EBV and flu-specific memory CD8+ T cells may be necessary to induce perforin upregulation . In contrast , perforin upregulation is more prominent in response to CMV and Adenovirus . CMV infection is characterized by continual low-level viral replication , and can induce massive expansions of CMV-specific effector CD8+ T cells[41] . Hence , it is not entirely surprising that CMV-specific CD8+ T cells should be capable of rapid perforin upregulation . The potent ability of Ad-specific CD8+ T cells to upregulate perforin , on the other hand , was unexpected since Ad , much like flu , should be rapidly cleared . However , there is evidence that Ad can become persistent[42] . Furthermore , there are at least 51 different Ad serotypes in circulation around the world , with differential levels of neutralizing antibodies against each serotype being present within a given individual[43] . As there is a high degree of sequence conservation between various Ad serotypes[26] , cross-reactive CD8+ T cells are quite common , even between distantly related Ad serotypes[25] . Therefore , it is likely that exposure to Ad antigen from persisting virus or exposure to different Ad serotypes repeatedly activates Ad-specific CD8+ T cells , thereby driving the maintenance of a stable Ad-specific effector population . In contrast , EBV- and flu-specific CD8+ T cells typically produce IL-2 and bear a central memory phenotype . Since antigen load in the chronic phase of these infections is low or absent , the responding CD8+ T cell populations have likely differentiated to a resting memory state , where immediate cytotoxic potential is not critical . An alternative interpretation of our data is that control of some viruses requires differential functional profiles: a polyfunctional response led by IL-2 is necessary for EBV and influenza , while CMV and Adenovirus may need to be controlled or cleared by a perforin-dominated response . Our phenotypic profiling of the perforin and IL-2 functional subsets as effector and central memory-like T cells , respectively , ( Supplementary Figure S2 ) is in agreement with previous work on CD8+ T cell maturation , which included the measurement of pre-formed perforin , to ascribe discrete functional attributes to specific stages of differentiation[31] , [33] , [44] , [45] . On this basis , several studies have related particular memory phenotypes to control of certain viral infections[32] , [34] , [46] , [47] , [48] . Our work elaborates on these earlier studies by correlating specific , complex functional profiles to immunity against different viral pathogens , irrespective of stage of differentiation . The D48 perforin antibody used here enabled the measurement of both pre-formed and new perforin , permitting a detailed characterization of the complete perforin compartment and a sharper definition of the mutually exclusive relationship between the perforin and IL-2 CD8+ T cell functional subsets . Given the cross-sectional nature of our study , it is not possible to ascertain whether new perforin and IL-2 dominated functional subsets represent stable CD8+ T cell populations that actually abrogate their respective viral burdens , or if they are subsets that result as a consequence to a waning antigenic presence . A longitudinal analysis of CD8+ T cells responding to the live yellow fever virus and smallpox vaccines recently showed that both vaccines generated a primary virus-specific CD8+ T cell response that passed through an obligate effector phase in which the cells abundantly expressed perforin and granzyme B[49] . The cells then differentiated into long-lived memory cells that maintained the ability to proliferate and secrete effector cytokines in response to antigen[49] . Thus , the perforin and IL-2 functional subsets we describe herein likely serve to mediate protective immunity at different stages of infection . What is responsible for the transition from a polyfunctional response highlighted by rapid perforin upregulation to an IL-2-dominated response ? What determines the array of functions a CD8+ T cell can perform ? Antigen sensitivity has recently been reported to be required for the development of a polyfunctional CD8+ T cell response[50] , but the mechanism behind this phenomenon remains to be elucidated . Our association between elevated CD28 levels and IL-2 production by antigen-specific CD8+ T cells confirms published findings describing a direct role for CD28 signaling in IL-2 induction[35] , [36] . Our observation that new perforin preferentially accumulates in human CD8+ T cells that express the transcription factor T-bet supports the role of T-bet as a ‘master regulator’ of effector CD8+ T cell responses[37] , [38] , [51] , [52] . Corollary , the relatively reduced levels of T-bet in the IL-2 producing CD8+ T cells supports data from mouse models of T cell differentiation which demonstrate that T-bet is also a transcriptional repressor of IL-2[53] , [54] . Furthermore , T-bet expression correlates with the development of short-lived effector cells in mice , whereas a moderate decrease in T-bet expression promotes long-lived memory [51] , [52] , [55] . Thus , our data suggest that T-bet is intimately involved in determining the functional capabilities of virus-specific CD8+ T cells , and provide an important premise in humans on which to explore the relationship between T-bet and the perforin gene . The interplay between IL-2 and perforin thus necessitates a re-evaluation of our current interpretation of CD8+ T cell polyfunctionality . The prevailing rationale is that antigen-specific polyfunctional CD8+ T cell responses containing IL-2 are most effective at controlling viral replication[13]; a premise that is driving current T cell based vaccine strategies . Our data suggest that we need to reclassify CD8+ T cell polyfunctionality into at least two distinct types: polyfunctional memory ( IL-2 + IFN-γ + other functions without perforin ) or polyfunctional effector ( perforin + IFN-γ + other functions without IL-2 ) , each profile being distinct and worthy of independent consideration . In reality , both functional subsets will likely be required for a protective immune response , each being instrumental at different stages of infection .
The University of Pennsylvania's Center for AIDS Research Human Immunology Core ( IRB# 705906 ) , The Wistar Institute ( IRB#2506215 ) , and Duke University ( IRB exempt ) obtained written , informed consent from every donor subject in order to collect PBMC samples and approved the methods employed in this study . PBMC were cryopreserved in fetal bovine serum ( FBS; ICS Hyclone , Logan , Utah ) containing 10% dimethyl sulfoxide ( DMSO; Fisher Scientific , Pittsburgh , Pennsylvania ) . Individual peptide stimuli were determined by prior epitope mapping by IFN-γ Elispot experiments . In subjects for whom epitopes were not identified , pools of peptides ( 15mers overlapping by 11 amino acids ) were used . Regarding the use of 15 versus 9 amino acid individual peptides , several studies have shown that although some variation in function and magnitude can be present between some epitopes , on average the magnitude and functionality of responses to CTL epitopes represented as a 9 mer or within a 15 mer peptide are generally equivalent . As a proof of concept , we stimulated Subject E with both an optimal and a 15 amino acid peptide containing the epitope TPRVTGGGA and quantified very similar responses ( Supplemental Figure S3 ) . Antibodies for surface staining included anti-CD4 PE Cy5-5 ( Invitrogen; Carlsbad , California ) , anti-CD107a FITC ( BD Biosciences; San Jose , California ) , anti-CD8 Qdot 655 ( custom ) or TRPE ( Invitrogen; Carlsbad , California ) , anti-CD14 Pac Blue ( BD Biosciences; San Jose , California ) , anti-CD16 Pac Blue ( BD Biosciences; San Jose , California ) , and anti-CD19 Pac Blue ( Invitrogen; Carlsbad , California ) , anti-CD57 Qdot 565 ( custom ) , anti-CD27 PE Cy5 ( Beckman Coulter , Inc; Fullerton , California ) or PerCP Cy5-5 ( Biolegend; San Diego , California ) , anti-CD28 ECD ( Beckman Coulter , Inc; Fullerton , California ) and anti-CD45RO Qdot 605/705 ( custom ) or ECD ( Beckman Coulter , Inc; Fullerton , California ) . Antibodies for intracellular staining included anti-CD3 Qdot 585 ( custom ) , anti-Granzyme B Texas Red PE ( BD Pharmingen; San Diego , California ) , anti-IFN-γ Alexa 700 ( BD Pharmingen; San Diego , California ) , anti-IL-2 APC ( BD Pharmingen; San Diego , California ) , anti-TNF-α PE Cy7 ( BD Biosciences; San Jose , California ) , and anti-T-bet ( Santa Cruz Biotechnology; Santa Cruz , California ) . Custom conjugations to Quantum ( Q ) dot nanocrystals were performed in our laboratory as previously described[56] , with reagents purchased from Invitrogen ( Carlsbad , California ) . Anti-human perforin antibodies were purchased from Tepnel ( clone D48 , Besancon , France ) and BD Biosciences ( clone δG9 , San Jose , California ) . Cryopreserved PBMC were thawed , and then rested overnight at 37°C , 5% CO2 in complete medium [RPMI ( Mediatech Inc; Manassas , Virginia ) supplemented with 10% FBS , 1% L-glutamine ( Mediatech Inc; Manassas , Virginia ) , and 1% penicillin-streptomycin ( Lonza; Walkersville , Maryland ) , sterile filtered] at a concentration of 2×106 cells per ml medium in 12-well plates . The next day , the cells were washed with complete medium and resuspended at a concentration of 1×106 cells/ml with costimulatory antibodies ( anti-CD28 and anti-CD49d; 1 µg/ml final concentration; BD Biosciences; San Jose , California ) , in the presence of monensin ( 0 . 7 µg/ml final concentration; BD Biosciences; San Jose , California ) and brefeldin A ( 1 µg/ml final concentration; Sigma-Aldrich; St . Louis , Missouri ) . Anti-CD107a was always added at the start of all stimulation periods , as described previously[29] . As a negative control , 5 µl of DMSO was added to the cells , an equivalent concentration compared to the peptide stimulus . SEB served as the positive control ( 1 µg/ml final concentration; Sigma-Aldrich; St . Louis , Missouri ) . Peptide stimulations were performed at a final concentration of 2 µM . Stimulation tubes were incubated at 37°C , 5% CO2 for six hours , after which cells were washed once with PBS and then stained for viability with Aqua amine-reactive viability dye ( Invitrogen; Carlsbad , California ) for ten minutes in the dark at room temperature . A cocktail of antibodies was then added to the cells to stain for surface markers for an additional twenty minutes . The cells were washed with PBS containing 1% bovine serum albumin ( BSA , Fisher Scientific; Pittsburgh , Pennsylvania ) and 0 . 1% sodium azide ( Fisher Scientific; Pittsburgh , Pennsylvania ) , and permeabilized using the Cytofix/Cytoperm kit ( BD Biosciences; San Jose , California ) according to the manufacturer's instructions . A cocktail of antibodies against intracellular markers was then added to the cells and allowed to incubate for one hour in the dark at room temperature . The cells were then washed once with Perm Wash buffer ( BD Biosciences; San Jose , California ) and fixed in PBS containing 1% paraformaldehyde ( Sigma-Aldrich; St . Louis , Missouri ) . Fixed cells were stored in the dark at 4°C until the time of collection . For each specimen , between 500 , 000 and 1 , 000 , 000 total events were acquired on a modified flow cytometer ( LSRII; BD Immunocytometry Systems; San Jose , California ) equipped for the detection of 18 fluorescent parameters . Antibody capture beads ( BD Biosciences; San Jose , California ) were used to prepare individual compensation tubes for each antibody used in the experiment . Data analysis was performed using FlowJo version 8 . 7 . 3 ( TreeStar , Ashland , Oregon ) . Reported data have been corrected for background . Canvas software , version 10 . 4 . 9 ( ACD Systems; Miami , Florida ) , and Prism software , version 5 . 0 ( Graphpad; La Jolla , California ) , were used to create the figures . Labels and boxes were added to raw data images in Canvas . The dots for Subject C in the bottom right panel of Figure 6 were enlarged in Canvas to facilitate visual identification and discrimination . Correlation between %IL-2 and %perforin of IFN-γ producing CD8+ T cells was determined by a two-tailed Pearson correlation test . A two-tailed Paired t-test was used to define statistically significant differences in CD28 and T-bet expression between IL-2 and perforin producing CD8+ T cells . Both analyses were performed using Prism software .
|
Although CD8+ T cells are thought to be largely responsible for the control of viral infections , exactly how they mediate protection is uncertain . One approach to assessing their protective capacity is to measure several of their functions simultaneously . Generally , it is believed the more functions a cell can perform , the better its potential to control viral replication . A multi-functional response including interleukin-2 ( IL-2 ) production is currently valued as the key correlate of protection . We recently characterized a novel CD8+ T cell function: rapid perforin upregulation , which serves to contribute to and sustain the killing of virally infected host cells . In this study , we show that new perforin is abundant during adenovirus and cytomegalovirus infections , but scarcely detected in the context of influenza and Epstein-Barr virus . Importantly , perforin and IL-2 are rarely co-expressed . The significance of this relationship is that we can no longer assume the more functions a CD8+ T cell performs in response to a virus the better . Thus , when considering vaccine design , no single functional profile will likely be protective across all pathogens . Rather , vaccine-induced T cell responses may need to be “pathogen-specific” , as different T cell functional responses will be important for controlling different viral infections .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"immunology/immunity",
"to",
"infections",
"immunology/immune",
"response"
] |
2010
|
Perforin and IL-2 Upregulation Define Qualitative Differences among Highly Functional Virus-Specific Human CD8+ T Cells
|
While it has been established that a number of microenvironment components can affect the likelihood of metastasis , the link between microenvironment and tumor cell phenotypes is poorly understood . Here we have examined microenvironment control over two different tumor cell motility phenotypes required for metastasis . By high-resolution multiphoton microscopy of mammary carcinoma in mice , we detected two phenotypes of motile tumor cells , different in locomotion speed . Only slower tumor cells exhibited protrusions with molecular , morphological , and functional characteristics associated with invadopodia . Each region in the primary tumor exhibited either fast- or slow-locomotion . To understand how the tumor microenvironment controls invadopodium formation and tumor cell locomotion , we systematically analyzed components of the microenvironment previously associated with cell invasion and migration . No single microenvironmental property was able to predict the locations of tumor cell phenotypes in the tumor if used in isolation or combined linearly . To solve this , we utilized the support vector machine ( SVM ) algorithm to classify phenotypes in a nonlinear fashion . This approach identified conditions that promoted either motility phenotype . We then demonstrated that varying one of the conditions may change tumor cell behavior only in a context-dependent manner . In addition , to establish the link between phenotypes and cell fates , we photoconverted and monitored the fate of tumor cells in different microenvironments , finding that only tumor cells in the invadopodium-rich microenvironments degraded extracellular matrix ( ECM ) and disseminated . The number of invadopodia positively correlated with degradation , while the inhibiting metalloproteases eliminated degradation and lung metastasis , consistent with a direct link among invadopodia , ECM degradation , and metastasis . We have detected and characterized two phenotypes of motile tumor cells in vivo , which occurred in spatially distinct microenvironments of primary tumors . We show how machine-learning analysis can classify heterogeneous microenvironments in vivo to enable prediction of motility phenotypes and tumor cell fate . The ability to predict the locations of tumor cell behavior leading to metastasis in breast cancer models may lead towards understanding the heterogeneity of response to treatment .
Statistics from 2013 show that more than 90% of breast cancer deaths are a consequence of metastatic disease and that median survival for metastatic breast cancer is 3 years [1] . Accordingly , a better understanding of the physiological conditions that help initiate or facilitate metastatic recurrence would be valuable for identifying and improving outcomes for patients with metastatic breast cancer . The process of metastasis [2] includes epithelial-mesenchymal transition , tumor cell locomotion through the interstitial matrix , dissemination from the primary tumor via vasculature ( intravasation and extravasation ) , and seeding of secondary organs [3] . The chemical signals that initiate and direct tumor cell motility are typically growth factors secreted either by host cells in the primary tumor microenvironment ( e . g . , macrophages , endothelial cells , fibroblasts , and pericytes ) or by tumor cells themselves [4] , and may be present either in soluble form or bound to the extracellular matrix ( ECM ) . Recently , increasing evidence has shown that mechanical signals in tumor microenvironments may also modify tumor cell behavior . Tumor cell locomotion has been shown to correlate with proximity to macrophages [5] and blood vessels [6] while increased cross-linking [7] , stiffness [8] , [9] , and alignment [10] of the ECM correlate with increased metastasis . In summary , many studies have investigated individual tumor cell–stromal cell or tumor cell–ECM interactions , revealing a number of possible sources of chemo- and hapto-tactic signals for tumor cells , as well as the high level of complexity in the tumor microenvironment network [4] . Moreover , a number of tumor microenvironment studies [11] have shown positive or negative effects of stromal cell types or individual mechanical parameters on tumor cell motility within the primary tumor and subsequent metastasis . However , an integrated view of the tumor microenvironment as a complex system , including how all relevant biological players in combination affect tumor cell motility and metastatic outcome , is still missing . In ECM that is soft , porous , and with low levels of cross-linking , single tumor cells can locomote quickly and in a metalloprotease ( MMP ) -independent manner [12] by utilizing forces generated by adhesion and actin polymerization [13] , [14] . Locomotory protrusions ( pseudopodia ) form at the front of the cell , followed by locomotion and translocation of the entire cell . Stiff , non-compliant ECM contains high concentrations of cross-linked collagen and/or basement membrane proteins . To enable movement , tumor cells remodel their immediate ECM , mainly via the degradative action of MMPs [12] , [15] , [16] . Studies in cell culture models have shown that , when grown on ECM in the presence of growth factors , metastatic cancer cells form invadopodia [17] . These are dynamic membrane protrusions that are enriched in actin , actin regulatory proteins , cortactin , tyrosine kinases , Tks5 , and proteases [18] , and are ECM-degrading structures . In vitro , invadopodia form in 10%–20% of tumor cells plated on ECM [19] , and their average size in the light microscope is approximately 1–3 µm in diameter and 2–12 µm in length , depending on ECM density and dimensionality ( 2-D versus 3-D ) [19]–[21] . The invadopodium compartment is simultaneously the site of actin polymerization , cell contact to ECM , and matrix proteolysis; thus , the invadopodial lifetime is on the scale of hours , significantly longer than any other type of membrane protrusions in tumor cells [19] , [22] . Although a large part of the invadopodial signaling network overlaps with that of other membrane structures such as lamellipodia , filopodia , and focal adhesions , the defining feature of invadopodia is their high proteolytic activity , which degrades and remodels ECM . This suggests that invadopodia may be a key mechanism for MMP-dependent tumor cell invasion [23] . Recent studies have gone beyond 2-D and into the 3-D realm of studying invadopodia [20] and podosomes [24] . Moreover , there are recent attempts to study these structures in situ , in their natural environment , including podosomes in mouse aorta [25] , invadopodia in dermis [26] , as well as protrusions similar to invadopodia in organogenesis models of Caenorhabditis elegans [27] and C . intestinalis [28] . Further , we have recently shown a direct link between the ability of tumor cells to assemble invadopodia and degrade matrix using MMPs in vitro and in situ and the ability to intravasate and metastasize in vivo [29] , [30] . Collective data from these and similar studies demonstrate that morphology and protein distributions of invadopodia and podosomes depend on the matrix dimensionality , architecture , and complexity as well as the cell line/type . Neither tumor cells nor the structures they assemble appear the same in 2-D or 3-D [20] culture , or within the fixed tissue [29] . Hence , to identify podosomes and/or invadopodia in different microenvironments and under new experimental conditions , a combination of defining features such as small size , structural components ( e . g . , Tks5 , cortactin ) and functionality ( MMP-dependent ECM-degradation ) may be used , classifying structures as podosomes when referring to structures in myelomonocytic , endothelial , and vascular smooth muscle cells [17] and invadopodia when referring to tumor cells . Although tumor cell locomotion in primary tumors has been observed in a number of studies [4] , [5] , [31]–[34] , the exact location and timing of invadopodium formation is not presently known . Combined results from numerous studies using preparations of collagen I ( with or without cross-linking ) , collagen IV , laminin , fibrin , chorioallantoic , and peritoneal membranes [12] , [35] converge on the conclusion that invadopodia have the ability to degrade and remodel different ECM components present in interstitial tissue and basement membranes of metastatic tumor models [29] . However , where and when invadopodia form in the primary tumor in vivo has yet to be determined . Here , we have characterized and quantified two motility phenotypes occurring in primary breast tumors in vivo: fast-locomotion and slow-locomotion associated with invadopodia . We show that invadopodia ( and slow-locomotion ) occur in regions of primary tumor that are spatially distinct from regions where fast-locomotion is the dominant phenotype . We further combined intravital multiphoton microscopy , image analysis , and multiparametric data classification , to analyze microenvironmental conditions under which tumor cells move via either phenotype . We found that the prediction of primary tumor locations where tumor cells locomote fast or form invadopodia and locomote slowly cannot be achieved by any individual tumor microenvironment parameters but only by their combination .
By recording 4-D stacks in orthotopic xenograft tumors formed from a metastatic , breast cancer cell line MDA-MB-231-Dendra2 , we confirmed the presence of a tumor cell behavior previously described in detail , where cells quickly locomote through the tissue either individually or in streams ( Figure 1A–1C; Movie S1a ) [4] , [31]–[34] . Subtraction of the initial frame ( 0′ ) from that taken at 30 min ( 30′ ) in a single z-section , detects the parts of the cell that moved over 30 min and the area the cell covered ( Δ30 cell tracks ) ( Figure 1B , blue ) . However , in some fields of view , tumor cells exhibited a different phenotype . Fast-locomotion was not detected ( Figure 1D and 1E; Movie S1b ) but was replaced by slow-locomotion . Imaging at high magnification revealed the presence of small protrusions ( Figure 1E , white arrowheads and red cell tracks; Movie S2 ) , often adjacent to collagen fibers ( Movie S2a and S2b ) , macrophages and blood vessels ( Movie S2c and S2d ) . Longer time-lapse imaging showed that tumor cells with small protrusions exhibited a slower locomotion phenotype Figure 1F and 1G , red ) in comparison to previously reported fast-locomotion ( Figure 1F and 1G , blue ) . Cells migrating with velocities of 22–250 µm/h were classified as fast-locomoting , while those with small protrusions and velocities 2–15 µm/h were classified as slow-locomoting ( Figure 1F ) . Systematic analysis of fields of view in ten animals has shown that the two motility phenotypes were usually mutually exclusive , i . e . , tumor cells within a specific field of view exhibited only one of the observed phenotypes at a time in 184/187 fields ( Figure 2A ) . As xenograft tumors consist of genetically similar tumor cells , we hypothesized that the presence of two motility phenotypes is controlled via the tumor microenvironment and may be distinguished by different levels of one or more microenvironmental factors , whose measurement can be then used as a predictor of tumor cell behavior . To investigate which microenvironmental conditions are amenable for either motility phenotype , we monitored tumor cell phenotype relative to the number of microenvironment parameters previously reported to influence tumor cell locomotion ( Figure 2B ) [5]–[10] . These included the density of collagen fibers [7]–[10] , tumor cells , and macrophages [5] as well as the number and size of blood vessels [6] present in the field of view . We tested each of the parameters individually for their correlation with either of the motility phenotypes , attempting to find values of a single parameter or combination of them that coincide with one of the two phenotypes . However , none of the commonly used statistical methods based on linear separation ( model selection , partial correlation , or causal modeling ) ( Figure S1A–S1C ) were able to provide a model capable of distinguishing between microenvironment parameters and the presence of different motility phenotypes . A possible reason for the lack of linear predictors is the existence of complex interactions among parameters , such as those present in the tumor microenvironment [4] , [11] . To test this , we employed support vector machine ( SVM ) , an algorithm suitable for segmentation of non-linearly separable data , such as the data in Figure S1A . Intuitively , the SVM algorithm “converts” a non-linearly separable dataset into linearly separable set by increasing the dimensionality of the hyperspace ( initially , N parameters create hyperspace of N dimensions ) . For illustrative purposes , see a one-dimensional example in Figure S11A . In order to link between microenvironmental parameters and tumor cell locomotion phenotypes , we used microenvironmental parameters as an “input” for the SVM classifier , with fast-locomotion ( blue dots ) or slow-locomotion ( red dots ) as the classifier's “output” ( Figure S1D ) . Classification was done on the basis of three , four , or five parameters with increasing accuracy , reaching maximum accuracy of 92% when all five parameters were used ( Figure S1D ) . Figure 2C shows a 3-D projection of the hyperspace with maximum accuracy classification . The high accuracy of the SVM classification ( and cross-validation ) means that it is possible to predict whether tumor cells will locomote fast or slow ( output ) on the basis of a given set of microenvironmental parameters ( input ) . By utilizing only the microenvironmental parameters within a field of view as input information , we get a prediction of the motility phenotype present in that same field of view , which is output information . Misclassification ( green dots ) probability of <8% ( and <5% when training includes the entire dataset ) is associated mainly with parameter values on the classification border between red space and blue space ( Figure 2C , red- and blue-shaded areas ) . None of these individual microenvironment parameters alone offers a sufficient predictor , as their contribution to motility phenotype depends on the context created by all other microenvironment parameters . Only a nonlinear transformation of all parameters could distinguish between microenvironments associated with either motility phenotype ( Figure S1 ) . In conclusion , tumor motility phenotypes can be distinguished only by a non-linear , multiparametric analysis such as SVM , as they are a result of balance among multiple signals within complex tumor microenvironments . To better understand the link between the tumor microenvironment and tumor cell motility phenotypes , we analyzed each of the motility phenotypes at single cell level ( Figure 3 ) . We characterized the relationship between individual tumor cells of either phenotype and landmarks within the tumor microenvironment ( Figure 3A ) . While directionality of fast-locomotion seemed to be controlled by the orientation of adjacent collagen fibers ( Figure 3A , blue pie charts ) , the small protrusions in slow-locomoting cells were commonly perpendicular to adjacent blood vessels and collagen fibers that surround blood vessels ( Figure 3A , red pie charts ) . To further analyze the relationship between blood vessels and small protrusions , we compared the distance of tumor cells with either motility phenotype to the nearest blood vessel and the size of that blood vessel ( Figure 3B ) . Our results show that fast-locomotion may occur far from blood vessels ( 21/132 of fast-locomoting cells were found>100 µm from the nearest blood vessel ) and is independent of the blood vessel size . However , the distances from the nearest vessel at which small protrusions formed were smaller ( 2/95 of slow-locomoting cells were found>100 µm from the nearest blood vessel ) and positively correlated with vessel diameter ( Figure 3B , red triangle ) , suggesting that small protrusions are most likely initiated by a chemical or mechanical gradient associated with the blood vessel . We also measured the amount of ECM components not detected by second harmonic generation ( Figure S2 ) , demonstrating that deposits of basement membrane proteins laminin ( Figure S2A ) and collagen IV ( Figure S2B ) increase with blood vessel size , suggesting that small protrusions are associated with larger blood vessels with this matrix composition . We next looked at the number of motile cells within fields of view where either fast- or slow-locomotion was detected . Both fast ( blue bars ) and slow ( red bars ) locomotion each occurred in approximately 15% of tumor cells in the field of view ( Figure 3C ) and only slow-locomoting cells exhibited small protrusions ( Figure 3D ) . Finally , small protrusions associated with slow-locomotion were found to be inhibited by tail-vein injection of MMP inhibitor GM6001 , suggesting MMP-dependence , while the fast-locomotion remained unaffected throughout the 3 h time lapse ( Figure 3E ) . Our phenomenological data , including morphology , relationship to ECM and blood vessels , as well as MMP-dependence of small protrusions led us to the hypothesis that the observed small protrusions are invadopodia in vivo . To test this hypothesis , we used previously established structural and functional markers of invadopodia . Molecularly , invadopodia have been shown to be enriched in active proteases and structural components including cortactin and Tks5 [35]–[39] . While cortactin is present both in invadopodia and at the leading edge of migrating cells in 2-D culture [40] , in 3-D conditions it is enriched primarily at the tip of invadopodial protrusions at the cell front [20] , [29] . By using MDA-MB-231-cortactin-GFP cells [20] , we were able to directly compare cortactin-enriched compartments in 3-D collagen I ( Figure S3A ) , in cryosections ( Figure S3B ) [29] and in vivo ( Figure 4 ) . Similarly to 3-D and cryosections , the small protrusions observed in vivo ( left and middle panels ) showed a peak of cortactin fluorescence at the protrusion tip ( Figure 4A , yellow lines in upper panels and associated line-scans in lower panels; Movie S4b ) . In contrast , fast-locomoting cells showed a homogeneous distribution of cortactin throughout the cell ( Figure 4A , right panels; Movie S4a ) . These results are consistent with the identification of the small protrusions as invadopodia in vivo . We next constructed cell lines where the Tks5 structural protein of invadopodia was knockdown ( KD ) with shRNA , which specifically targets invadopodia in MDA-MB-231 cells and tumors ( Figure S4 ) . Tks5 knockdown was previously shown to inhibit invadopodium maturation and degradation [39] , [41] in 2-D culture conditions [39] as well as to greatly reduce lung metastases in mouse models [41] . Our data confirmed that Tks5 KD1 and Tks5 KD2 cells ( Figure S4A ) did not assemble invadopodia in vitro , under culture conditions ( Figure S4B and S4C ) , as shown using Tks5/cortactin antibodies and fluorescent gelatin degradation . Similar results were confirmed in primary tumors , where knockdown efficiency was maintained ( Figure S4D ) , and Tks5-positive punctae which colocalized with collagen I degradation ( Figure S4E ) were present only in TKs5 CTRL tumors but not in knockdowns ( Figure S4F ) . In addition , Tks5 KD1 and Tks5 KD2 tumors in vivo did not exhibit small protrusions , while the fast locomotion behavior was only slightly affected ( Figure S4G ) , supporting our hypothesis that small protrusions are indeed invadopodia and they were selectively targeted by Tks5 knockdown . Finally , to directly test if the small protrusions function as invadopodia in vivo , we measured the ECM degradation activity associated with the small protrusions . The antibody against degraded collagen we used for immunohistofluorescence ( Figure S3 ) is unsuitable for in vivo use due to the inefficient delivery and labeling . Instead , we used the MMP-activated substrate MMPSense 680 ( Perkin Elmer ) for intravital imaging [42] . To validate this reporter , we compared ECM degradation as measured by MMPSense 680 solution ( cyan ) and a more commonly used substrate , DQ-collagen I gel ( red ) [14] , [43] in 3-D culture of cortactin-TagRFP cells ( green ) ( Figure S5A and S5B ) . Quantitation of ECM degradation area with or without MMP inhibitor GM6001 ( Figure S5C ) showed similar trends with both reporters . In vivo , MMPSense 680 ( cyan ) injected into MDA-MB-231-cortactin-GFP tumors ( green ) colocalized with cortactin-enriched protrusions after 3 h ( Figure 4B , left panels; yellow lines correspond to line-scans; Movie S5a ) . In contrast , fast-locomoting cells did not colocalize with MMPSense 680 ( Figure 4B , right panels; Movie S5b ) . Finally , we monitored the accumulation of MMPSense 680 signal in microenvironments where small protrusions were either present or absent , in microenvironments treated with GM6001 , as well as in Tks5 knockdown tumors . This experiment was done using photoconversion followed by MMPSense 680 injection and image collection from 0–96 h ( Figure S6 ) . We quantified MMPSense 680 signal ( Figure 4C ) at 24 h , when the signal in the tumor plateaus . In fields of view where small protrusions were present , we measured that 6% of image area contained MMPSense 680 fluorescence above threshold , reflecting the locations of MMP-based degradation and this was approximately 3-fold higher than in other areas , GM6001-treated areas or knockdown tumors . In addition , MMPSense 680 fluorescence was positively correlated with the number of small protrusions ( Figure 4D ) , and this correlation was eliminated by GM6001 treatment . As all our results are consistent with the hypothesis that the small protrusions are in fact invadopodia in vivo , we will henceforth refer to them as invadopodia . The results of SVM classification show that the two motility phenotypes , which we now recognized as fast-locomotion and invadopodia-associated slow-locomotion , exist in distinct yet neighboring conditions . Such a result suggests that a shift along one of the axes in the 3-D plot in Figure 2C may induce a change in the number of tumor cells exhibiting either motility phenotype , i . e . , induce a phenotype switch . The changes in tumor cell behavior are context-dependent , which means that the extent of the influence of one parameter over the phenotype , or its capability to cause a phenotype switch , depends on the context of other parameters ( Figure S7 ) . We hypothesized that a motility phenotype switch in primary tumors may occur , for example , when collagen fiber density changes over time . An increase in collagen fiber density may be initiated because of fibroblast deposition of collagen , while decreases may be a result of the invadopodium-mediated degradation of fibers . Such logic is strengthened by previous in vitro reports showing that both the speed of MMP-dependent 3-D migration [44] and the number of invadopodia in 2-D assays are controlled by the rigidity and cross-linking level in basement membrane extracts , collagen , and synthetic matrices [45] , [46] . We tested the effect of ECM rigidity/cross-linking by modulating ECM cross-linking levels and measuring the number of invadopodia , which are associated with slow-locomotion phenotype ( Figure 5 ) . In the control set of animals , we imaged the same fields of view ( using photoconversion to match fields over time ) at 0 , 24 , and 48 h , demonstrating that invadopodia are present over the entire period under control conditions ( Figure 5A ) . Collagen imaging confirmed that over a 48 h period , collagen I fibers remained stable with minor changes ( Figures 5E , purple bars , and S8A ) . A different set of animals was treated with L-ribose , which was shown to increase cross-linking and hence stiffness in collagen-based gels [7] . A considerable increase in the number of invadopodia ( Figure 5B ) was detected , accompanied by an increase in collagen I density ( Figures 5E , green bars , and S8B ) . The third set of animals was treated with the lysil oxidase ( LOX ) inhibitor β-aminopropionitrile ( BAPN ) , which reduces cross-linking and loosens the ECM [7] . Treatment with BAPN reduced and finally eliminated invadopodia ( Figure 5C ) . Figure 5D shows that invadopodium number follows the trends in collagen fiber density . Moreover , as a result of fibrillar collagen I density decrease ( Figures 5E , red bars , and S8C ) , in several fields of view , tumor cells switched from invadopodium-associated slow-locomotion at 0 h to fast-locomotion at 48 h in the same field of view ( Figure 5C , inset ) . Areas with fast-locomotion also showed a dramatic reduction of fast-locomoting cells following the L-ribose treatment ( Figure 5F , green bars ) accompanied by the appearance of invadopodia . Interestingly , BAPN treatments also induced a slight decrease in the number of fast-locomoting cells over 48 h ( Figure 5F , red bars ) . One of the possible explanations may be the lack of balance between adhesion and traction forces at low ECM stiffness levels , which was previously shown to affect tumor cell migration in 3-D environments [44] . The observation of phenotypic switching is consistent with the prediction of SVM classification ( Figure 2C ) , which although none of the microenvironment parameters individually are sufficient predictors of tumor cell motility phenotype , a change in one or more of them in the context of the others can modulate and even switch the tumor cell behavior . Therefore , a decrease in collagen fiber density , depending on the context of other parameters ( Figure S7 ) may transform the microenvironment that favors the invadopodial phenotype ( Figure 2C , red sphere ) into the microenvironment that favors fast-locomotion ( Figure 2C , blue sphere ) , in turn switching the motility phenotype; similarly , an increase in collagen fibers may induce a switch from fast- to slow-locomotion is worth noting that a similar phenotypic switch , from MMP-dependent to MMP-independent motility , has been previously demonstrated in vitro by inhibiting ROCK or MMPs in melanoma cells [47] , [48] and fibrosarcoma cells [49] . However , previous to our results reported here , it was unclear if such a phenotypic switch can occur in vivo and what collagen density would regulate this phenotypic switch . Finally , we hypothesized that the two motility phenotypes may lead to different tumor cell fates ( Figure 6 ) . To test this , we photoconverted subpopulations of tumor cells within microenvironments where either slow- ( Figure 6A ) or fast-locomotion ( Figure 6B ) was present . A 175×175 µm2 square was photoconverted within each of the four to 12 neighboring fields of view at 0 h ( see Figure S6 ) , enabling us to follow the tumor cell fate at subsequent timepoints ( Materials and Methods and [6] , [29] for experimental details ) . In the fields with invadopodia , there was a significant negative trend in the number of photoconverted ( red ) cells at 24–48 h ( Figure 6C , red bars ) , previously shown to be due to dissemination from the primary tumor [6] , [29] , [32] , [50] . In contrast , other fields that contained fast-locomoting cells had cell numbers that were mostly unchanged and sometimes increased in numbers of photoconverted cells , likely a result of the cell division ( Figure 6C , blue bars ) [6] . In addition , the number of invadopodia at 0 h shows very strong negative correlation ( p = 5 . 4×10−6 ) with the number of photoconverted cells at 24 h ( Figure 6D ) , suggesting that the presence of invadopodia is directly linked to tumor cell disappearance from the field of view and , possibly , tumor cell dissemination [3] . Next , GM6001 irreversibly eliminated invadopodia ( Figure 6E ) and inhibited the disappearance of red cells ( Figure 6C , grey bars ) , supporting the hypothesis that tumor cells with invadopodia disseminate from the primary tumor [29] . To confirm that the disappearance of red cells in the areas rich with invadopodia ( but not in areas with fast-locomoting cells ) is due to the intravasation , we tested alternative processes that may contribute to changes of red tumor cell numbers , such as apoptosis and failure to proliferate ( Figures S9 and S10 ) . By combining photoconversion in vivo with additional ex vivo labeling ( Figure S9 ) , it is possible to asses rates of proliferation ( Figure S9A and S9C ) and apoptosis ( Figure S9B and S9C ) in fields of view with either fast- or slow-locomoting cells . No significant differences were observed in the proliferation rates of fields of view associated with slow- and fast-locomotion and there was practically no apoptosis observed . To confirm such an observation , an injectable marker SR-FLIVO was used for direct apoptosis measurements in real time ( Figure S10 ) , combined with cisplatin treatment to induce apoptosis in the positive control group . This method also reported the absence of apoptosis in areas where fast- or slow- locomotion occurs , strengthening the link between the red cells disappearance and intravasation . Following disseminated red cells to the secondary organs , we monitored the number of tumor cells that had formed lung metastases at 5 days post-photoconversion ( Figure 6F and 6G ) . Some lung metastases contained red tumor cells ( photoconverted cells that arrived at the lung at 0–5 days from the photoconversion site in the primary tumor ) , orange and yellow tumor cells ( photoconverted cells where red/green Dendra2 ratio is reduced due to the cell division and synthesis of new green Dendra2 ) , and green tumor cells ( whose origin cannot be traced ) ( Figure 6F ) . When tumors were chemically treated with BAPN ( which selectively inhibits invadopodium formation by modifying the external microenvironment ) or GM6001 ( which inhibits ECM degradation by invadopodia ) starting at photoconversion time ( 0 h , see Materials and Methods ) , we observed a significant reduction in the number of metastases containing red tumor cells ( Figure 6G , panel i ) , pointing to onset of inhibition of dissemination at 0 h . Moreover , lung metastases were practically eliminated in Tks5-knockdown tumors ( Figure 6G , panel ii ) where invadopodia were removed , while the fast-locomotion phenotype remained largely unaffected ( Figure S4G ) . These data together support the hypotheses that invadopodia and ECM degradation by tumor cells are essential for dissemination and metastasis , and that metastasis does not occur without active ECM degradation by tumor cells .
In this study , we monitor two different phenotypes of tumor cell motility: fast-locomotion and invadopodium-associated slow-locomotion , which appear in spatially separate microenvironments in primary , orthotopic xenograft tumors . Such a separation of phenotypes among different fields was striking and suggests that the size of the imaging field is well below the average size of each phenotypic microenvironment . As both phenotypes can be seen in the same animal ( imaging window covers ∼50 mm2 ) but they did not commonly occur in the same ( ∼0 . 5 mm2 ) or adjacent fields , we estimate the phenotypic microenvironment size to be 1–50 mm2 . Although both motility phenotypes are dependent on the presence of vasculature , ECM , and macrophages , they do not show obvious dependency on any of these parameters , either individually or when ( linearly ) combined . However , by using a multiparametric , non-linear SVM classification , we were able to predictably classify microenvironmental parameters that favor each motility phenotype . The non-linear nature of this statistical model exposes the complexity of tumor microenvironment , a theme often addressed in reviews , but commonly avoided in experimental science , where the focus is on individual microenvironmental parameters and regard others as stable and homogeneous “background” . Here we show that more than one microenvironmental component is involved in phenotype determination in primary tumors and also , that small regions with genetically identical tumor cells may exhibit different phenotypes . We have analyzed five microenvironment parameters ( number of macrophages , density of collagen fibers and tumor cells , number and size of blood vessels ) , which all contribute to the determination of tumor cell motility phenotype to a smaller or larger extent . The extent of contribution of each parameter depends on the context of the other four parameters ( Figure S7 ) and can change through time . Hence , heterogeneity , complexity , and dynamical changes in the tumor microenvironment dominate tumor cell phenotype and should be considered in the interpretation of future intravital imaging studies and in the development of new diagnostics and treatments . Stepping away from the focus on single microenvironment parameter analysis towards the multiparametric analysis may not only contribute to generating a more comprehensive view of how tumor microenvironments determine cell behavior , but also help with the search for new therapeutic targets , which is currently conducted by analysis of individual molecules or specific protein interactions . SVM classification suggests the existence of a fine balance among chemical and mechanical signals produced by tumor cells , macrophages , blood vessels , and ECM , which determines the motility phenotype of tumor cells . Ongoing changes in one or more microenvironment parameters ( collagen fibrillogenesis , angiogenesis , macrophage recruitment , etc . ) can lead to a switch between fast-locomotion and invadopodium-associated slow-locomotion . It is likely that both motility phenotypes contribute to metastasis . Fast-locomotion is more efficient in translocation of tumor cells , which may transport them to regions where they stop due to ECM composition and stiffness , such as in regions adjacent to some blood vessels as shown here , and then shift to the invadopodium-associated phenotype that results in tumor cell dissemination . The presence of misclassifications ( Figure 2C , green dots ) on the borders between two phenotypes suggests that there may be microenvironmental conditions where the switch between motility phenotypes commonly takes place . In further support of this , we were able to modulate invadopodium frequency by changing one of the parameters , collagen fiber density ( Figure S6 ) . A decrease in collagen fiber density resulted in elimination of invadopodia and , in some areas , a switch to fast-locomotion . In theory , changing other parameters such as blood vessel size and macrophage number would achieve a similar result ( Figure S6 ) . To understand the mechanism of this phenotypic switching , further mathematical modeling and direct in vitro measurements are currently underway . We speculate that both motility phenotypes are likely to be exhibited by the same tumor cells and that there are no genetic mutations unique to the two motility phenotypes . In support of this , we see both phenotypes in a xenograft mouse model , based on genetically identical tumor cells . Interestingly , the fraction of cells exhibiting either of the phenotypes in a particular field of view is the same , 15% . Similar fractions of tumor cells were previously reported as capable of assembling invadopodia in situ , tumor sections , or in vitro , when plated on thin gelatin [29] . This leads to the suggestion that both fast- and slow-locomotion are phenotypes that arise when similar internal conditions are present ( motility cycle is active ) but the cell is exposed to different external conditions . We propose that within the same region , under the same external conditions , the other 85% cells are not capable for motility at the moment , and the reason for that may be that they are mitotic or bound by energetic and signaling constrains . Broadening this hypothesis into other motility phenotypes , similar fractions of cells were also reported in individually or collectively migrating cells in mammary carcinoma xenografts , where phenotype was dependent on intracellular transforming growth factor beta ( TGFβ ) expression [51] . We demonstrated that the slow-locomotion phenotype , which was not previously characterized in vivo , is associated with small tumor cell protrusions identified as invadopodia using morphological , molecular , and functional assays . This motility phenotype is linked to ECM degradation , tumor cell dissemination from the primary tumor and subsequent lung metastasis , and inhibited by MMP-inhibitor . It is worth noting that the treatment with GM6001 was previously shown in 2-D culture to minimally affect existing invadopodial core structures [52] , but their growth and extension were not tested . In fact , invadopodia in 3-D grow in cycles of extension-retraction [20] and degradation of ECM by MMPs followed by enlargement or elongation of invadopodia was suggested to be a part of a positive feedback mechanism [37] . We propose that by abolishing this cycle , GM6001 inhibits elongation , rendering protrusions too small to be detected in vivo . In summary , our results support the hypothesis of a positive feedback loop between products of MMP proteolysis and continued invadopodial extension [37] . Based on the evidence collected in vitro in 2-D and 3-D [12] , [15] , [53] , we hypothesize that membrane-tethered MT1-MMP1 is likely enriched in the actively degrading invadopodia compared to fast-locomoting cells . However , none of the available antibodies were able to give us a satisfactory signal . While experimentally , MMPs remained only broadly tested using GM6001 , we speculate that the key role of MMPs includes immobilization of MT1-MMP on invadopodia , leading to local activation of all MMPs present in the external microenvironment [54] .
All procedures involving animals were conducted in accordance with NIH regulations , and approved by the Albert Einstein College of Medicine IACUC , by 20101010 and 20130909 protocol numbers . A metastatic human breast cancer line MDA-MB-231 was cultured and maintained in DMEM media supplemented with 10% fetal bovine serum ( FBS ) and 50 U penicillin/50 µg streptomycin per ml . The Dendra2-MDA-MB-231 cell line [50] was created by electroporation of the Dendra2 cloning vector C1 with the geneticin selection marker [6] . We performed fluorescence-activated cell sorting ( FACS ) after 14–20 days of selection; after removing the highest-expressing top 5% , we kept the top 20% of the highest fluorescing cells and maintained them under selection using 500 µg/ml geneticin ( Invitrogen ) . No changes in cell morphology , viability , or proliferation were observed in the labeled MDA-MB-231 cells compared to the parental cell line . Cell lines labeled with cortactin-GFP [20] and cortactin-TagRFP [19] have been previously described . Tks5 knockdown cell lines Tks5 KD1 and Tks5 KD2 and knockdown control cell line Tks5 CTRL were all created by transduction of Dendra2-MDA-MB-231 cells with lentiviral particles ( five particles/cell ) , which contained shRNA in pLKO . 1 vector , targeting Tks5 ( knockdowns ) or non-targeting shRNA ( CTRL ) ( Sigma Aldrich MISSION library based on Broad Institute consortium ) . Tumors were formed by injecting 106 cells in 150 µl of 20% collagen I in PBS into the mammary fat pads of 5–7 week old SCID mice . Imaging experiments were done 9–12 weeks after injection . Multiphoton imaging was done on a custom-built system based on an inverted Olympus IX71 microscope and two Ti-Sapphire lasers , one of them equipped with an Optical Parametric Oscillator ( OPO ) extension [55] . The first laser was tuned to 880 nm for excitation of GFP-like fluorophores and Texas Red , while the second laser was tuned for photoconverted Dendra2 at 1035 nm and MMPSense 680 at 1250 nm [55] . For experiments involving simultaneous Dendra2 and MMPSense 680 imaging , MMPSense680 was excited using the 880 nm laser . The microscope system is equipped with four simultaneously acquiring detectors , which allowed simultaneous imaging of collagen fibers via second harmonic generation ( blue ) and fluorescence from GFP-like fluorophores ( green ) , Texas Red or photoconverted Dendra 2 ( red ) , and MMPSense ( far red ) . In the photoconversion-enabled fate mapping experiments , a commercially available microscope system with multiphoton and confocal modes ( Olympus FV1000MPE with ULTRA 25× , 1 . 05 NA water immersion objective ) was used for initial 4-D imaging over 30 min ( in multiphoton mode , with laser locked at 880 nm ) and subsequent photoconversion of Dendra2 ( in confocal mode , using 405- , 488- , and 546-nm laser lines ) . Stacks ( 512 µm×512 µm×100 µm×30 min ) were collected at 5-µm-thick z-sections at 3-min timepoints for time-resolved series . Imaging was done at 0 . 25 µm2/pixel . Multiphoton imaging of mammary tumors was done as described previously [56] . MDA-MB-231 cells labeled with Dendra2- or cortactin-GFP were injected into the mammary fat pads of 5–7 week old SCID mice . After 10–12 weeks , we injected Texas Red 70 kDa dextran ( 250 nmol in 100 µl PBS per injection ) [5] for macrophage labeling and performed skin flap surgery 2 h later on anesthetized animals . Exposed mammary tumors were positioned on top of a coverslip on an inverted microscope and imaged continuously using a custom-built two-laser multiphoton microscope for up to 3–5 h . In some experiments , additional tail-vein injections were done for blood vessel labeling using Texas Red 70 kDa dextran ( Invitrogen; 250 nmol in 100 µl PBS per injection ) , MMPSense 680 ( Perkin Elmer; 2 nmol in 150 µl PBS per injection ) , SR-FLIVO ( Immunochemistry Technologies; 1∶10 dilution , 2 h prior to imaging ) , or the pan-MMP inhibitor GM6001 ( Milipore; 1 µmol in 100 µl injection ) . A stock solution of GM6001 , 500 mM in DMSO , was diluted in sterile PBS before tail vein injection . Post-surgical injection assures that compounds are only present in tumor blood vessels that are intact , connected to tumor vasculature and flowing at the time of injection . As previously described [6] , [29] , [50] , [57]–[59] , the mammary imaging window was combined with photoconvertible Dendra2 as a cytoplasmic marker . Briefly , Dendra2-MDA-MB-231 cells were injected into the mammary fat pads of 5–7 week old SCID mice . After 9–11 weeks , a mammary imaging window was implanted on top of the growing tumor ( 5–7 mm in diameter ) . Following a 3-day recovery , mice were anesthetized and placed in the imaging chamber on an Olympus FV1000-MPE hybrid multiphoton-confocal system . In each area , ten z-sections were acquired ( total stack size is 512 µm×512 µm×100 µm ) over 30 min . One to three areas were imaged per animal . Subsequently , photoconversion was done in 175×175 µm areas , separated by 150 µm along both x and y axes ( Figure S6 ) . Photoconversion areas were scanned for ten to 20 cycles with a 405-nm laser in confocal mode . Further imaging was done using the two-laser multiphoton system , using 880 nm for imaging of green species of Dendra2 and 1 , 035 nm for red species of Dendra2 . 3-D stacks were collected at photoconverted sites at 0 h , 24 h , and 48 h post-photoconversion . The number of red ( photoconverted ) Dendra2 cells was counted from the RGB overlay on individual z-sections , using manual Cell Counter plugin in ImageJ . Reported numbers are sums of counts at five z-sections separated by 25 µm , for method details see [29] . Values were corrected for cell division by dividing by control values from areas where no motility was detected [6] . The dispersion of photoconverted cells is monitored as the surface embedding all red cells in the 0–100 µm maximum image projection [6] . In MMPSense 680-based degradation measurements ( Figure 4B–4D ) , in SR-FLIVO apoptosis measurements ( Figure S9 ) , and in microenvironment modulation experiments ( Figure 5 ) , photoconversion was used to locate the same microenvironments at 0–48 h . Intravenous injections of 150 µl MMPSense 680 were done at 0 h , intraperitoneal injections of 150 µl PBS containing either 1 g/kg/day L-ribose or 20 mg/kg/day BAPN were done at 0 h and 24 h , while intraperitoneal injections of 2 mg/kg/day cisplatin were done daily for 5 days . As previously described [55] , transdermal photoconversion of Dendra2 can be used for fate mapping in pulmonary metastases . Briefly , 12 weeks after injection of Dendra2-MDA-MB-231 cells , pairs of SCID mice were matched by tumor size from the same cage . In each pair , one mouse was injected intraperitoneally with 150 µl of sterile PBS ( control ) , while the other was injected with 1 µg GM6001 in 150 µl PBS ( GM6001-treated [31] ) . Six hours later , mice were anesthetized by isoflurane and hair was shaved off the top of the mammary tumor . The tumors were photoconverted transdermally , using a 405-nm LED photodiode array [55] . Treatments with PBS or GM6001 were repeated daily , and mice were sacrificed at 5 days . Lungs were taken out and metastatic colonies ( groups containing 1+ tumor cells ) were counted in 50 fields of view through the ocular of the microscope , using a 25× objective . Colonies containing 1+ red tumor cells were classified as red , although most also contained green cells owing to the cell division over 5 days . Representative images ( Figure 6F and 6G ) were taken using an Olympus multiphoton setup , using the same imaging parameters as for intravital imaging . Counted colonies were analyzed by contingency table and compared for their average values ( Figure 6H ) , resulting in significant differences between control and GM6001-treated red metastases ( χ2 = 20 . 04 , p = 0 . 048 ) . Labeling of invadopodia in cell culture was done as previously described [19] , [20] , [29] . Briefly , 50 k MDA-MB-231-Dendra2 cells ( Tks5 CTRL , Tks5 KD1 , and Tks5 KD2 ) were plated in triplicate for 6 h on Mattek dishes coated with Alexa405 gelatin . After this time , they were fixed for 15 min in 3 . 7% paraformaldehyde ( PFA ) , washed three times in PBS , and permeabilized for 5 min in 0 . 1% Triton X-100 , blocked for 2 h in 1% BSA and 1% FBS , and incubated for 1 h in primary antibodies against cortactin ( Abcam , ab33333 ) and Tks5 ( Santa Cruz , M-300 ) and 1 h in secondary Alexa antibodies . Labeling of invadopodia in tissue sections was done as previously described [29] . MDA-MB-231 tumors ( either -cortactin-GFP or -Dendra2 ) were excised and fixed overnight in 3 . 7% PFA , washed for 1 h in cold PBS and dehydrated overnight in 30% sucrose . They were embedded in Optimal Cutting Temperature ( OCT ) compound , cut to 6-µm sections , permeabilized for 10 min with 0 . 1% Triton X-100 , blocked for 2 h in 1% BSA and 1% FBS , and finally incubated for 2 h with primary antibodies against degraded collagen ( Ibex Pharmaceuticals , Col2¾C short antibody , 1∶100 or immunoGlobe Col1-3/4C , 1∶100 ) and Tks5 ( Sigma , mouse anti-SH3PXD2A , 1∶100 ) . Secondary antibodies were conjugated to AlexaFluor-543 ( Invitrogen , 1∶300 ) and AlexaFluor-633 ( Invitrogen , 1∶300 ) and mixed with DAPI ( Invitrogen , 1∶1 , 000 ) . Alternatively , sections were permeabilized for 10 min in cold acetone and incubated with primary antibodies against endomucin ( V . 7C7 , Santa Cruz Biotechnologies , 1∶200 ) as well as tumor ECM components such as laminin ( ab11575 , Abcam , 1∶100 ) , collagen IV ( 2150-1470 , AbD Serotec , 1∶100 ) , collagen I ( ab90395 , Abcam , 1∶100 ) , or fibronectin ( ab6328 , Abcam , 1∶100 ) . Secondary antibodies were conjugated to AlexaFluor-488 or -543 ( Invitrogen , 1∶300 ) . Cells were imaged with a Leica SP5 confocal microscope ( Figures S3 and S4 ) . Labeling of tumor cell proliferation and apoptosis post-photoconversion was done by cryosectioning of the primary tumor regions under the mammary imaging window ( top 150 µm , [60] ) . Neighboring sections were labeled by Ki67 ( DAKO ) and cleaved caspase 3 ( Milipore ) , using Alexa633 as secondary antibodies . Blood vessel coverage by ECM components , collagen I and IV , laminin and fibronectin was measured using ImageJ ( Figure S2 ) . Briefly , images of blood vessels and ECM proteins were thresholded and tested for colocalization using Manders' M2 coefficient [61] . Blood vessel edges were traced using the green channel , with a 5-µm-thick band generated to encapsulate the area of ECM proteins . The amount of ECM components deposited around blood vessels was measured and normalized to the surface area . MMPSense 680 ( Perkin Elmer ) is a polymer with quenched VT680 chromophore and is weakly fluorescent prior to exposure to MMPs . Proteolytic cleavage of the peptide that links the copolymer probe can be executed by a wide range of MMPs and produces a signal in the near-infrared range ( with emission peak at 680 nm ) [62] . As MMPSense 680 is soluble , it can be added either to culture media or injected directly into the animal vasculature . To validate MMPSense 680 as a reporter of MMP-driven proteolytic activity at single cell resolution ( Figure S5 ) , a comparison was made to an established marker of matrix degradation , FITC-DQ-collagen I ( Invitrogen ) in 3-D culture based on rat-tail collagen I ( BD Biosciences ) . A 2–4-µm-thick layer of DQ-collagen I ( 10 µg/ml ) and collagen I ( 2 mg/ml ) mixture ( 1∶100 , DQ-unlabeled collagen ) in cold , phenylalanine-free DMEM/10% FBS serum was established on the bottom of a Mattek dish . After 20 min at 37°C , 5×104 MDA-MB-231-cortactin-TagRFP cells were seeded on this layer and covered with additional collagen I . After 60 min at 37°C , we added 50 nM MMPSense 680 and cultured the cells for 6–24 h . Image stacks were collected at 0 , 6 , and 24 h using a Leica SP5 confocal microscope , utilizing 488- , 543- , and 633-nm laser lines and setting the prism for detection at 500–525 , 555–610 , and 670–740 nm . Image processing and quantification of the amount of degraded ECM , measured as percent of total image area after threshold , was done in ImageJ . Briefly , 3-D stacks of degraded DQ collagen or MMPSense 680 were combined into a maximum intensity projection and processed using a 1-pixel median filter . Thresholding of images taken at 24 h ( Figure S5A ) was done using average values of images taken at 0 h . The workflow of our intravital systems microscopy approach is outlined in Figure 2A and 2B . Each 4-D image stack was collected in three or four channels . After a brief quality screening of each z-section , two adjacent z-sections were combined into a maximum intensity projection ( 10 µm thick ) and the recorded stack was separated into four neighboring fields of view ( 256×256 µm ) for easier processing . All images were passed through smoothing filters and thresholded to remove background fluorescence . Individual microenvironment time lapses were passed through the StackReg/TurboReg plugin [63] , available at http://bigwww . epfl . ch/thevenaz/stackreg/ . These Java classes efficiently remove minor drifts in the xy plane resulting from breathing and tissue settling . Time-lapse movies of individual z-sections in each field of view ( 30′–60′ duration ) were first visually scored for morphological determinants of tumor cell fast locomotion , as previously described [4]–[6] , [29] , [31]–[34] . Briefly , the fast locomoting cell translocates by the extension of the cell front , movement of the center and the contraction of the rear , with an average speed of 1 µm [34] . To account for the number of fast locomoting cells the frame taken at 0′ was subtracted from the frame taken at 30′ , both in the green channel , yielding an image of all pixels translocated during this time interval . Particle analysis and Region of Interest ( ROI ) Manager in ImageJ were then used to count number of locomoting cells and overlay them with the original 0′ frame to show initial tumor cell position and the trace left by the tumor cell migration over 30′ ( Figure 1B , green for tumor cells and blue for the tumor cell tracks ) . Some tumor cells moved out of the analyzed z-section during the 30′ period and were traced in the neighboring z-sections . In such cases , cell tracks were visualized via maximum projection , accounting for the thickness of the z-slice ( 5 µm ) . Next , time-lapse movies were viewed at 2–4× zoom , exposing small protrusions characterized by ( a ) finger-like morphology , ( b ) width of 1–3 µm , and ( c ) dynamics . Frame subtraction and particle analysis were used to visualize the dynamics of small protrusions over 30′ ( tip movement and change in size due to the cycles of extension and retraction ) as well as the number of small protrusions per cell ( Figures 1F , red line , and 2D ) . In order to detect migration of the tumor cell body ( front , center , and rear ) in cells with small protrusions , movement of time-lapse movies were extended to 5 h . The following five microenvironment parameters were extracted from the multichannel 3-D stacks: ( a ) density of collagen fibers , defined as percent area above threshold in the blue channel , ( b ) density of tumor cells , defined as percent area in the green channel , ( c ) number of macrophages , defined as macrophages labeled with 70 kDa dextran [5] , ( d ) number of flowing blood vessels in the field of view , and ( e ) diameter of the largest flowing blood vessel visible in the field of view . For ( d ) and ( e ) , we made the assumption that all blood vessels flowing at the time of post-surgical injection by Texas Red were labeled . As macrophages and blood vessels were collected in the red channel , prior to measurements they were separated on the basis of size and morphology , using the Analyze Particles plugin in ImageJ . Microenvironment parameters and the motility phenotype ( i . e . , fast- or slow-locomotion ) form a multiparametric matrix that we used for SVM classification [64] . Different types of kernels and other parameters were tested systematically in R software . Classification was finally done using Gaussian radial basis kernel , using the “tune . svm” procedure in the R-package “e1071 , ” starting from the default γ- and cost-parameters and iteratively optimizing them to increase accuracy , which we tested using 10-fold cross-validation . The dataset was divided into randomly grouped ten equal-size sets , training on nine while testing on the left-out set . Starting from SVM based on three-parameter combinations , we were able to achieve 78%–84% accuracy ( Figure S1D ) . By adding all five parameters measured , we ultimately achieved 92% accuracy . Without cross-validation ( i . e . , using all data ) classification accuracy is 95% . This means that we were able to predict motility phenotype on the basis of measurements of all five microenvironment parameters in 92/100 fields of view and that removing any of the parameters reduces the predictive power of SVM . However , for purposes of presentation in 3-D space , we have used a 3-D projection that contains parameters with the highest influence: density of collagen fibers , number of macrophages , and diameter of blood vessels ( Figure 2C ) . Misclassifications ( Figure 2C , 5/100 , green dots ) are present in boundary areas . Note that blue- and red-shaded areas in Figure 2C are provided only as an illustration . Collected image stacks were processed by a combination of the ImageJ [65] and Imaris software packages . ImageJ was supplemented with custom-written plugins based on our work and work of others for image browsing , color correction , and migration trajectories . We used Imaris 7 . 4 for 3-D reconstructions . Data was directly exported to Microsoft Excel for statistical analysis and plotting; the R package was used for trend analyses , variance trend analysis , SVM classifications , correlations , linear regression , and calculation of p-values , while Matlab was used for SVM representation . All numerical values used in figure graphs are included in Data S1 file . To test the influence of the blood vessel distance and diameter on tumor cell motility phenotypes ( Figure 3B ) , we measured the shortest distance of the tumor cell front to the nearest blood vessel . We hypothesized that in case the blood vessel is the source of a chemical or mechanical gradient , then the increase in blood vessel diameter will allow motility at larger distances , thus increasing the spread of distances ( variance ) of motile tumor cells from the blood vessel . To test this hypothesis , we first calculated the residuals that result from regressing the observed distance of the cell front on the vessel diameter . In turn , the log square residuals were regressed on the vessel diameter . As hypothesized , an increase in blood vessel diameter was associated with a significant increase in residuals in cells with small protrusions ( R2 = 0 . 3 , p = 7×10−8 ) indicating that in the fields of view where only microvessels are present , small protrusions form only next to the blood vessel , while in the presence of large-diameter blood vessels , small protrusions can be both close and far from the blood vessel . Fast-locomoting cells exhibit no such behavior ( R2 = 0 . 07 , p = 10−3 ) . Note the small R2 value . Red-shaded area in Figure 3B illustrates variance trend as blood vessel diameter increases .
|
A large proportion of cancer deaths are due to metastasis—the spread of cancer from the primary tumor to other parts of the body . Movement of cells may require the formation of protrusions called invadopodia , which degrade extracellular matrix . Although some studies have reported on locomotion in primary tumors , the presence of invadopodia was not tested . Here , we show that single cells from mouse mammary carcinoma can move using a fast- or slow-locomotion mode depending on different levels of cues present in the tumor microenvironment . Using multiphoton microscopy in vivo combined with a machine-learning algorithm we show how manipulation of microenvironmental conditions can induce predictable changes in the number of locomoting cells or switch between the two locomotion modes . We also demonstrate that only the slower moving cells are associated with invadopodia in vivo , and that only tumor cells from regions rich in invadopodia degrade the surrounding extracellular matrix and disseminate . Specific targeting of invadopodia results in inhibition of lung metastasis . This work proposes a systems biology view of how tumor microenvironments regulate tumor progression and presents insight into the heterogeneity of the treatment response . The ability to define and predict conditions under which tumor cells disseminate offers potential therapeutic benefits in regulating tumor progression .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"oncology",
"systems",
"biology",
"medicine",
"and",
"health",
"sciences",
"cell",
"biology",
"molecular",
"biology",
"biology",
"and",
"life",
"sciences",
"computational",
"biology",
"biomechanics"
] |
2014
|
Multiparametric Classification Links Tumor Microenvironments with Tumor Cell Phenotype
|
Hepatitis C virus ( HCV ) infection remains a major cause of hepatic inflammation and liver disease . HCV triggers NLRP3 inflammasome activation and interleukin-1β ( IL-1β ) production from hepatic macrophages , or Kupffer cells , to drive the hepatic inflammatory response . Here we examined HCV activation of the NLRP3 inflammasome signaling cascade in primary human monocyte derived macrophages and THP-1 cell models of hepatic macrophages to define the HCV-specific agonist and cellular processes of inflammasome activation . We identified the HCV core protein as a virion-specific factor of inflammasome activation . The core protein was both necessary and sufficient for IL-1β production from macrophages exposed to HCV or soluble core protein alone . NLRP3 inflammasome activation by the HCV core protein required calcium mobilization linked with phospholipase-C activation . Our findings reveal a molecular basis of hepatic inflammasome activation and IL-1β release triggered by HCV core protein .
HCV continues as a global health problem causing chronic and progressive liver disease [1–5] . HCV is a major risk factor for hepatocellular carcinoma , and infection is a consistent cause of liver transplants . HCV is a small , enveloped , single-stranded RNA virus that belongs to the Flaviviridae family [6] . It is transmitted through parenteral routes and replicates primarily in the liver . Most often , exposure to HCV leads to chronic infection , which is characterized by persistent hepatic inflammation . The hallmark of chronic HCV infection is dysregulated and persistent inflammatory responses that are thought to serve as a platform for ongoing liver damage and the onset of cirrhosis and hepatocellular carcinoma [7] . While currently no vaccine for HCV is available for clinical use , the advent of direct acting antivirals ( DAAs ) has revolutionized patient care and these drugs are proven to be effective treatment options for HCV infected individuals beyond interferon ( IFN ) -based therapy [8 , 9] . DAAs are oral regimens , well-tolerated and most patients achieve 80–90% sustained virologic responses ( SVRs , defined as the absence of HCV RNA detection after cessation of treatment with DAAs ) . However , with DAAs there is a concern of the emergence of drug resistant HCV variants , the unknown effects of drug-to-drug interactions , and the expensive nature of these drugs [10 , 11] . Most importantly , further prospective studies are needed to assess the effects of treatment with DAAs on preventing liver fibrosis and mitigating HCV-induced severe liver disease such as HCC [12 , 13] . Therefore , understanding the complete molecular mechanism of HCV-induced hepatic inflammation is essential to design the best therapeutic regimen to treat hepatic inflammation and to reduce liver damage resulting from chronic HCV infection . HCV replicates in hepatocytes , the chief parenchymal cell of the liver . During infection HCV also interacts with hepatic macrophages such as the liver-resident Kupffer cells ( KCs ) , which make up 15–20% of the hepatic non-parenchymal cells [14] . KCs are highly phagocytic and play an important dual role within the hepatic microenvironment . They maintain hepatic homeostasis during immune responses to liver injury and also function as central mediators of hepatic inflammation induced in response to microbial-derived products [14–16] . The inflammatory cascade within the liver is initiated and propagated by KCs upon recognition of danger-associated molecular patterns ( DAMPs ) such as HMGB1 and pathogen-associated molecular patterns ( PAMPs ) such as viral RNA and/or viral proteins [17 , 18] . Activated KCs produce and secrete a diverse array of chemokines and cytokines leading to leukocyte recruitment to the liver . One of the key intrahepatic inflammatory soluble factors produced by KCs in response to DAMP or PAMP interaction is interleukin-1β ( IL-1β ) [19] . IL-1β is a potent proinflammatory cytokine that induces the production of chemokines and cytokines such as CXCL4 , TNF and IL-6 . IL-1β production by hepatic macrophages leads to the recruitment and activation of myeloid cells and lymphocytes in the liver [20–23] . Importantly , IL-1β plays a pivotal role to modulate the immune response during both acute and chronic virus infection [24–27] . While IL-1α , a closely related cytokine to IL-1β , signals through the same receptor [28] , it is widely expressed by many cells in direct response to stimuli that activate NF-κB [29] . In contrast , IL-1β production is tightly regulated through a two-step process of IL-1β expression and inflammasome activation . The production of a bioactive IL-1β cytokine requires the assembly of a cytoplasmic multiprotein complex called the inflammasome [30 , 31] . This multimeric complex is typically composed of at least three proteins: a nucleotide-binding oligomerization domain-like receptor ( NOD-like receptor ) such as NLRP3 , the adaptor protein ASC and the effector protease caspase-1 . IL-1β production requires a priming signal or ( signal-one ) initiated by DAMP and/or PAMP recognition and signaling by the responding cell to drive the production of inactive , proIL-1β protein . An inflammasome-activating signal or ( signal-two ) is then required to recruit and assemble the inflammasome components with ASC and procaspase-1 leading to caspase-1 activation , cleavage of pro-IL-1β to mature form and the release of active IL-1β protein . The NLRP3 inflammasome is one of the well-studied inflammasomes that is activated by diverse stimuli including RNA viruses [32 , 33] . Assembly of the NLRP3 inflammasome is triggered and governed by integrating diverse activating signals such as calcium mobilization and influx , potassium efflux , reactive oxygen species and/or by interaction with cellular factors such as NEK7 [34–36] . Elevated serum levels of IL-1β and IL-18 are prevalent in patients infected with HCV [27 , 37] . Furthermore , IL-1β is expressed exclusively within the liver of patients with cirrhosis , but not within a normal liver or an HCV-infected liver exhibiting no fibrosis/disease [27 , 37] . These findings provide strong evidence linking IL-1β with HCV-induced hepatic inflammation and disease . HCV interaction with macrophages triggers IL-1β production and release through NLRP3 inflammasome activation [19 , 27 , 33] . In these studies , HCV was shown to stimulate both immature and mature IL-1β production , indicating that one or more of the HCV virion components provides the necessary signals to stimulate NLRP3 inflammasome from within macrophages . HCV induces signal-one of NLRP3 inflammasome activation through viral RNA triggering and signaling through Toll-like receptor ( TLR ) 7 [27 , 33] , but how it imparts signal-two to drive inflammasome assembly and activation are not known . In this current study , we sought to identify the HCV-specific component ( s ) that stimulate NLRP3 inflammasome activation and the production of IL-1β to define the molecular mechanism of hepatic inflammation directed by IL-1β and induced by HCV . Our study identifies virion core protein as the major specific NLRP3 inflammasome agonist that drives inflammasome assembly leading to the production/release of bioactive IL-1β from macrophages . We show that the viral core protein directs intracellular calcium mobilization to impart NLRP3 inflammasome assembly through activation and signaling of phospholipase-C . Our study reveals a pivotal role of virion-associated and soluble/circulating core protein in HCV-induced hepatic inflammation , underscoring the contribution of the viral core protein in HCV pathogenesis and liver disease [38–43] .
HCV is a potent inducer of IL-1β production in macrophages and our studies have shown that HCV itself contains all the factors needed to trigger both signal-one and signal-two of NLRP3 inflammasome activation [27 , 33] . To determine which of the HCV virion component ( s ) is ( are ) essential for stimulating NLRP3 inflammasome activation , we prepared HCV and subjected it to ultraviolet ( uv ) light , resulting in inactivated HCV ( uv-HCV ) ( Fig 1B ) . Uv-HCV retains the ability to bind and enter cells but is unable to replicate [27] . THP-1 macrophages produce active IL-1β after exposure to uv-HCV ( Fig 1A and 1C ) coincident with the rapid processing and activation of caspase-1 ( Fig 1D ) and the formation of ASC- specks indicative of inflammasome activation ( Fig 1E ) . These results collectively reveal that the components of the inactivated HCV virion , containing the viral RNA and structural proteins , but not the viral non-structural protein ( s ) [44] , serve as the NLRP3 inflammasome agonist to drive caspase-1 processing , ASC-speck formation and IL-1β release from macrophages exposed to HCV . Thus , while viral replication is not required for inflammasome activation , components of the incoming virion must deliver the necessary signals , both to prime ( signal-one ) and activate the NLRP3 inflammasome ( signal-two ) . Ten proteins are encoded by the HCV virion wherein three of these proteins , Core , E1 and E2 , are the viral structural proteins [44] . HCV core protein encapsulates the viral RNA and it is essential for viral assembly whereas the viral glycoproteins , E1 and E2 , are the envelope proteins involved in entry [45] . E1 and E2 co-exist as a heterodimer via non-covalent interaction in which E2 is essential for proper function of E1 . In addition to the structural proteins , HCV encodes a small ion-channel protein p7 and six non-structural proteins ( NS2 , NS3 , NS4A , NS4B , NS5A and NS5B ) . Although p7 protein is essential for both virion assembly and egress in hepatocytes , it is unclear to this date if p7 is a true component of the mature HCV virion [46] . We evaluated the role of each structural protein and p7 in NLRP3 inflammasome activation using the NLRP3 inflammasome reconstitution [47] system in U2OS ( human osteosarcoma ) cells . In this cell system , the NLRP3 inflammasome components of NLRP3 , ASC , procaspase-1 , and proIL-1β are ectopically co-expressed , thus bypassing the requirement for signal-one to drive the expression of inflammasome components but are able to respond to a signal-two stimulus for inflammasome assembly and activity . We first co-expressed each HCV structural protein or p7 with the NLRP3 inflammasome components in U2OS cells and examined the ability of each viral protein to drive inflammasome activation marked by induction of mature IL-1β release as compared to cells co-expressing vector control . We found that the HCV core protein , and to a lesser extent the viral p7 protein , but not the envelope glycoproteins E1-E2 , were able to stimulate NLRP3 inflammasome activation ( Fig 2A ) . To further assess a possible role for E1-E2 in inflammasome activation , we evaluated the ability of HCV pseudoparticles containing E1-E2 ( HCVpp ) [48] to activate the NLRP3 inflammasome in THP-1 macrophages . HCVpp are viable and able to infect Huh7 [49] hepatoma cells as measured by luciferase activity encoded by the HCVpp ( Fig 2B upper panel ) . THP-1 cells treated with HCVpp did not release IL-1β , but cells treated with vesicular stomatitis virus G protein pseudoparticles ( VSVpp; control ) produced and released IL-1β , consistent with VSV-mediated inflammasome activation [26] ( Fig 2B lower panel ) . The HCV ion channel p7 , is a small transmembrane protein [46] . Although it remains to be determined if p7 is truly a component of the HCV virion , p7 has been shown to stimulate a level of NLRP3 inflammasome activity , consistent with our observations and others ( see Fig 2A and [50] ) . The differential magnitude of IL-1β production induced by the p7 and core protein suggests that p7 might not be the primary or sole activator of the NLRP3 inflammasome by the HCV virion if indeed it is a virion component . To further examine the role of p7 and its activity as an ion channel in inflammasome activation , we treated THP1 cells with uv-HCV in the presence of a small molecule inhibitor of the p7 ion channel , JK3/32 [51] . As a parallel control , we also treated cells with an inactive compound analog , R-21 , that does not inhibit p7 . Treatment with either compound had no effect on IL-1β production and release from cells in response to uv-HCV ( S1 Fig ) . We confirmed that the THP1 cells responded to LPS/Nigericin ( Ng ) treatment to stimulate IL-1β production and release in the presence of each compound ( S1A Fig ) , though p7 inhibitor treatment suppressed HCV infection in Huh7 cells ( S1B Fig ) . Thus , while we confirm that p7 can independently direct a level of inflammasome activation in the reconstituted system , in the context of the HCV virion p7 activity does not play a primary role in inflammasome activation in macrophages . Together , our results reveal the viral core protein as the major component of the HCV virion that induces NLRP3 inflammasome activation leading to the processing and release of IL-1β . We examined the ability of HCV core protein to stimulate IL-1β production and ASC-speck formation in the reconstituted system . We found the HCV core protein , as similar to the influenza A virus M2 protein ( Flu-M2 ) - a known NLRP3 inflammasome signal-two viral protein stimulus [25] , triggers inflammasome activation in reconstituted U2OS cells . HCV core induced the production of IL-1β to similar or greater extent as Flu-M2 ( Fig 3A ) . HCV core induced IL-1β release in a dose-dependent manner ( Fig 3B ) and supported high level of IL-1β release when cells received increasing input of proIL-1β cDNA construct ( Fig 3C ) . The presence of HCV core protein caused the appearance of ASC-specs in the reconstituted cells , but not in neighboring nonreconstituted cells ( Fig 3D ) . These results demonstrate that HCV core protein stimulates canonical NLRP3 inflammasome activation to mediate ASC-spec formation and processing and release of IL-1β . To determine if the HCV core protein is a signal-two stimulus for NLRP3 inflammasome activation as occurs in macrophages exposed to HCV [19 , 33] , we assessed the ability of core protein to induce inflammasome activation in THP-1 macrophages . We generated THP-1 cells stably expressing either vector only , HCV core or Flu-M2 ( control ) . When primed with TNF-α as a signal-one agonist ( S2 Fig ) , each THP-1 macrophage cell line released IL-1β ( Fig 4A ) . TNF-α priming triggered a small and non-significant level of IL-1β in vector-expressing cells as compared to unprimed cells . On the other hand , in core-expressing cells we observed a significant enhancement of IL-1β release as compared to unprimed cells or the control primed-vector-expressing cells . Treatment of cells with poly-U/UC RNA from the HCV genome , a known PAMP and activator of innate immune signaling through retinoic acid inducible gene-I ( RIG-I ) [52] , also primed the cells to facilitate a significant level of IL-1β release from cells expressing core but not vector alone ( Fig 4B ) . Virion-free soluble HCV core protein is readily detected at high concentration in the plasma of HCV-infected individuals [53 , 54] . We therefore tested the ability of purified , recombinant HCV core protein ( rHCV-core ) to stimulate NLRP3 activation in THP-1 cells . We assessed the formation of ASC-specks in cells treated with rHCV-core or purified recombinant green fluorescence protein ( rGFP , control ) . As phagocytic cells , THP-1 macrophages readily internalize cell-free macromolecules in a fashion similar to KCs [27 , 55] , and immunofluorescence staining of recombinant core or fluorescence analysis of rGFP in cells demonstrated the uptake of each protein by treated THP-1 macrophages . We found that rHCV-core but not rGFP triggered ASC-speck formation after cell uptake ( Fig 4C ) . Treatment of THP-1 cells with rHCV-core following signal-one priming with TNF-α stimulated robust IL-1β production and processing as did ATP , a known signal-two agonist of the NLRP3 inflammasome ( Fig 4D ) . We next confirmed these findings in primary human monocyte-derived macrophages . As in THP-1 cells , TNF-α-primed macrophages produce IL-1β when stimulated with rHCV-core ( Fig 4E ) as the control cells stimulated with nigericin ( Ng ) , a known activator of the NLRP3 inflammasome [56] . These observations demonstrate that HCV core protein uptake by macrophages undergoing signal-one priming stimulates inflammasome activation to trigger IL-1β production . Further , we confirmed that rHCV-core-induced IL-1β release requires NLRP3 inflammasome-dependent signaling ( Fig 4F and 4G ) . Taken together , these observations imply that both cell-free HCV core protein and virion-associated HCV core protein serve as viral protein agonists of NLRP3 inflammasome activation in macrophages . The HCV core protein is expressed from the viral polyprotein as a 191-amino acid polypeptide [57 , 58] . The full-length core protein is composed of major domain-I and domain-II . Domain-I comprises the basic N-terminus region that binds viral RNA and is involved in self-oligomerization [57 , 59] . Domain-II comprises the hydrophobic region of the protein that mediates association with membrane and lipid droplets . The minor domain-III comprises the C-terminal region and the signal sequence for cellular localization of E1 wherein it is cleaved off by signal peptide peptidase during polyprotein processing [57] . To define the domain of HCV core that is required for NLRP3 inflammasome activation , we tested a panel of core truncation mutants for their ability to induce NLRP3 inflammasome activation in the reconstituted U2OS cell system ( Fig 5A ) . HCV core constructs lacking Domain-II and domain-III retained the ability to stimulate IL-1β processing and release similar to full-length core protein , revealing that NLRP3 activation is directed by core protein domain-I ( Fig 5B ) . Truncation of domain-I to a 26 aa N-terminal portion resulted in a loss of NLRP3 activation . Overexpression of the 26 aa N-terminal protein resulted in no processing of IL-1β similar to vector control ( Fig 5B , lower panel ) , which is evident by the presence of more proIL-1β band in the western blot . Therefore , the full domain-I region is necessary and sufficient for induction of NLRP3 inflammasome activation by HCV core protein . We next tested the ability of four patient-derived HCV core protein constructs to trigger NLRP3 inflammasome activation . Each of the clinical isolates of the core-coding region of HCV was recovered from sera of patients undergoing acute HCV infection with HCV genotype 1 prior to antiviral therapy , with each patient progressing to chronic infection [60] . For patient-4 , sera samples were collected at two time points , at month-0 and month-2 post-HCV infection . HCV core protein from each patient stimulated NLRP3 inflammasome activation for IL-1β processing and release when expressed in reconstituted U2OS cells ( Fig 6A and 6B ) . Interestingly , patient 4 core protein sequence displays variable activity to induce the NLRP3 inflammasome , and this core protein lacked the ability to significantly trigger IL-1β release over vector control while core sequence 4–1 , isolated two months later , triggered significant IL-1β release . Examination of the sequence of each core protein construct from patient 4 revealed the acquisition of an amino acid substitution in domain-I within core 4–1 that links with inflammasome activation ( [60] and S3 Fig ) . Thus , sequence diversity among core domain-I might impact NLRP3 activation potential of the HCV core protein . To determine the mechanism by which HCV core protein triggers IL-1β production , we first assessed the subcellular distribution of the core protein after virion uptake of HCV-treated THP-1 macrophages . We found that within one hour following exposure to HCV , the core protein was present within the cell cytoplasm . HCV core protein was also punctuated around the plasma membrane , likely representing cell surface-bound virion [27] ( Fig 7A ) . We examined the possibility of core interacting with any of the NLRP3 inflammasome components as a mechanism of core-induced IL-1β production , but we observed no interaction between core and any of the NLRP3 inflammasome components ( S4A Fig ) . Among other actions of the HCV core protein , cytoplasmic core protein modulates calcium flux and calcium-dependent signaling in infected hepatocytes [61] . Ion flux , including calcium and potassium , are major contributors of NLRP3 inflammasome activation [62–67] . We therefore measured calcium mobilization in macrophages treated with purified rHCV-core as a potential mechanism of NLRP3 inflammasome activation . Treatment with rHCV-core but not rGFP ( control ) induced a rapid increase of intracellular calcium both in primary monocytes-derived macrophages and THP-1 macrophages ( Figs 7B and S4B ) . As phospholipase C is a major regulator of intracellular calcium mobilization , we then treated THP-1 macrophages with DMSO ( vehicle control ) , the phospholipase C inhibitor ( u-73122 ) [68] or an inactive inhibitor analog ( u-73343 ) in the presence of HCV or rHCV-core . HCV triggered caspase-1 activation and IL-1β processing in cells treated with DMSO or u-73343 , but treatment with the active phospholipase inhibitor u-73122 prevented caspase-1 activation and IL-1β processing ( Fig 7C ) . Importantly , we found IL-1β production induced by rHCV-core treatment of THP-1 macrophages is specifically suppressed by treatment of cells with u-73122 phospholipase C inhibitor ( Figs 7D , 7E and S4C ) as treatment with the control Ng . As shown above ( see S2 Fig ) , treatment with TNF-α stimulates signal-one activation , and we tested if inhibition of phospholipase C by u-71322 treatment did not block signal-one priming of the NLRP3 inflammasome . When we reversed the treatment sequence by stimulating with TNF-α first then treating with the phospholipase C inhibitor , IL-1β production was similarly suppressed ( S4D Fig ) . These results collectively indicate that the HCV core protein imparts an activating signal for IL-1β production and release from macrophages through induction of phospholipase C-mediated calcium mobilization that activates the NLRP3 inflammasome .
Our findings defines the molecular mechanism by which HCV triggers IL-1β production in macrophages ( Fig 8 ) . Hepatic macrophages effectively phagocytose macromolecules in the liver to continually survey the hepatic environment and respond to microbial threats [16] . Within the HCV-infected liver , virus-derived antigens , viral RNA and/or viral protein and other inflammatory mediators such as TNF-α can serve as priming factors in hepatic macrophages leading to NFκB activation and proIL-1β production [27 , 69] . Our observations demonstrate that the resulting signal-one primed macrophages are then responsive to HCV core protein both within the virion and as a cell-free protein present in patient blood [70] . HCV virion or core protein uptake then deposits the viral core protein in the cell cytoplasm where it induces phospholipase C-mediated calcium flux leading to NLRP3 inflammasome activation , thus establishing the hepatic inflammatory environment . Calcium signaling is a complex event that is important for many cellular processes including activation of the NLRP3 inflammasome [62 , 65 , 67 , 71] . It has been shown that the dynamics of potassium efflux and calcium influx , and subsequent increase in intracellular calcium , leads to mitochondria damage , which releases reactive oxygen species that ultimately activate NLRP3 inflammasome [65] . Our finding that HCV , through the action of its virion core protein within macrophages , modulates intracellular calcium flux/mobilization through a phospholipase C-dependent process , suggests that the HCV core protein may modulate the expression of calcium-sensing receptor ( CaSR ) or other components within a phospholipase C-regulated pathway [72 , 73] . CaSR is expressed in macrophages , including THP-1 cells and its expression is enhanced upon calcium binding wherein it may promote further accumulation of intracellular calcium following stimulation by the core protein [71–73] . Alternatively , the subcellular localization within cytoplasmic compartments of the macrophage might dictate the activity of HCV core to impart inflammasome activation . For example , the HCV core protein has been shown to associate with cytosolic membrane compartments including the endoplasmic reticulum membrane in contact with lipid droplets . Core insertion within these membranes might also lead to calcium "leakage" or depletion to evoke CRAC channel activity and Ca influx [72] . While further studies need to delineate the role of other pathways that mediate intracellular calcium increase in core-induced IL-1β production , our observations indicate that phospholipase C-mediated calcium production/mobilization directed by the viral core protein is an essential process of HCV-induced NLPR3 inflammasome activation following virion or core protein uptake by macrophages . Our studies show that KCs , THP-1 macrophages , and primary macrophages produce and secrete inflammatory IL-1β following exposure to HCV [27] or soluble viral core protein ( this study ) . Although macrophages do not support HCV replication , the sensing of viral-derived antigens could occur at multiple places initiated by the process of virion uptake by macrophages such as phagocytosis-mediated engulfment . Our studies show that the HCV virion itself comprises the mediators to drive both signal-one and signal-two of NLRP3 inflammasome activation . Just as inflammatory cytokines , such as TNF-α within the blood and hepatic environment , can mediate signal-one triggering in responsive macrophages , engagement of TLR7 by the viral RNA within the internalized HCV virion serves as a potent signal-one inducer [27] . Moreover , our current findings indicate that poly-U/UC RNA of the major PAMP of HCV recognized by RIG-I [52 , 74] can trigger signal-one to prime macrophages . Thus , multiple stimuli can prime the macrophage for a response to HCV core protein in order to trigger NLRP3 inflammasome activation . As a result , hepatic macrophages produce and release IL-1β and propagate the hepatic inflammatory response that underlies liver inflammation and disease in chronic HCV infection . p7 could also play an important role in hepatic inflammation . As previously reported [19 , 50] , and consistent with our data sets ( see Fig 2A ) p7 can impart NLRP3 inflammasome activation . p7 is a viroporin that has ion channel activity [75] , wherein p7-mediated ion-flux might impart NLRP3 inflammasome signal-two . However , our results argue against p7 activity having a major role in inflammasome activation induced by the incoming HCV virion , as treatment with a p7 ionophore inhibitor had no effect upon virion-induced NLRP3 activation . Instead , we propose that within an infected liver , macrophage engulfment of infected hepatocytes undergoing active HCV replication might be more likely to result in p7-directed inflammasome activation . HCV employs many innate immune evasion strategies to mediate persistent infection linked with chronic liver disease [76] . In infected hepatocytes , HCV blocks viral poly-U/UC PAMP/RIG-I-mediated production of the type 1 and 3 interferon ( IFN ) through the action of the viral NS3/4A protease , which targets and cleaves MAVS to abrogate intracellular antiviral defenses [77] . Furthermore , the HCV core protein is known to antagonize IFN signaling pathways preventing the expression of antiviral effector genes [78] . Moreover , IFN negatively regulates the NLRP3 inflammasome through anti-inflammatory IL-10 [79] . These processes collectively dampen the establishment of an effective antiviral state and provide a suitable platform for persistent infection , IL-1β production , and hepatic inflammation that contribute to ongoing immune-mediated liver injury and hepatitis . Our study underscores a central role of macrophages in HCV pathogenesis . IL-1β is a major mediator of hepatic inflammatory cytokine production from liver cells responding directly to IL-1β signaling [80] . Blockade of IL-1β production with anti-IL-1β agents may offer an attractive therapeutic option for HCV infected individuals . While treatment with DAAs present a cure for HCV infection and can restore hepatic innate immune and inflammatory homeostasis in the liver [81 , 82] , studies show that hepatic inflammation and altered inflammatory cytokine levels still persists in some patients successfully treated with DAAs who display ongoing liver injury [12 , 83–85] . These studies highlight sustained inflammatory cytokines such as IL-1β and IL-1β responsive cytokines as components linked to residual sustained liver disease from hepatic injury after HCV infection . The IL-1β /NLRP3 inflammasome cascade [86] may therefore present an attractive target for treating liver disease induced by HCV and other causes of liver inflammation .
Antibodies: monoclonal anti-HCV core antibody ( ThermoFisher ) , anti-influenza-M2 and anti-Caspase-1 ( Santa Cruz Technologies ) , anti-Flag ( Sigma ) , anti-actin ( Millipore ) , anti-IL-1β ( Cell Signaling , detects only cleaved IL-1β protein ) , goat-anti-mouse or goat-anti-rabbit secondary antibodies ( ThermoFisher ) . HCV E2 antibody ( Austral Biologicals ) . Other reagents: Recombinant human TNF and m-CSF were purchased from Peprotech and recombinant HCV core protein ( rHCV-core ) ( Meridian Life Science , Inc ) . The recombinant core protein is produced in Pichia pastoris and was used at 10-30ug/ml . Nigericin , ATP , D609 , u-73343 , u-73122 , DMSO , and PMA ( Sigma ) . Indo-I-AM calcium flux detection kit was purchased from Calbiochem and LPS from Addipogen . For polyU/UC RNA transfection and/or DNA transfection , Mirus Trans-IT mRNA transfection ( for RNA transfection ) and Mirus low-toxicity ( LT1 ) ( for DNA transfection ) were used . The polyU/UC RNA was made as previously described [52] . The p7 inhibitor ( JK3/32 ) , which is the active analog , and ( R-21 ) , which is the inactive analog , were dissolved in DMSO and used at 1-10uM with final DMSO content of 0 . 25% . For all kits used , the manufacturer’s protocol was followed and if a modification is made , it is noted . THP-1 cells were obtained from ATCC , U2OS ( osteosarcoma cells ) were obtained from ATCC , and Huh7 K2040 cells and Huh7 . 5 cells were described previously [87] ) . NLRP3 knockout cells or non-targeting controls in THP-1 cells were made using CRISPR ( Kindly generated by Andrey Shuvarikov , University of Washington ) . Cells were maintained in RPMI-1640 for THP-1 and DMEM for U2OS cells . All media were supplemented with 10% FBS , 0 . 01M Hepes , 1mM sodium pyruvate , 2mM L-glutamine , antibiotics and non-essential amino acid at 1x . All cells were maintained at 37°C under 5% CO2 . For differentiating THP-1 cells into macrophages , THP-1 monocytes were treated with 40nM of PMA for 24hrs . The following day the cells were washed with PBS and then cultured in fresh media for another 24hrs for resting . Stimulation of THP-1 cells took place after resting ( 48hrs after initial PMA treatment ) . HCV was propagated and tittered as described previously [88] . Briefly , Huh7 . 5 [49] were inoculated with HCV . Virus was removed after one hour of inoculation and then replaced with fresh media . 48hrs post infection , culture media containing virus was removed and replaced with fresh media . Infected cells were then cultured for 3–5 days . At the day of harvest , the infectious supernatant was filtered through 0 . 1um to remove cell debris and concentrated the virus using the Millipore virus concentrating centricons . After concentrating , the virus stock was aliquoted and stored at -80°C . The virus titration analysis was performed on Huh7 . 5 using the NS5A monoclonal antibody ( kindly provided by Dr . Charles Rice , The Rockefeller University ) . The HCV pseudoparticles ( HCVpp ) and VSVpp ( kindly provided by Dr . Jane McKeating , University of Oxford ) . For ultraviolet inactivation of HCV , the virus was crosslinked using the uv stratalinker ( 1200mJ/cm2 over 1800 seconds ) . For measuring JFH-1 inhibition by JK3/32 , Huh7 cells were cultured and propagated as described previously [51] . Secreted infectivity was measured as described [89] . Briefly , 1ug of linearized HCV construct pJFH-1 was used as a template for in vitro transcription ( RiboMax express , promega ) as per manufacturer’s instructions . RNA was then purified by acid phono/chloroform and isopropanol precipitation . 4x106 Huh7 cells were then electroporated as described previously [51] . Electroporated cells were seeded at 2 . 5x104 cells/well in 100ul volume in 96 well plates and incubated for 4hrs prior to addition of inhibitor . Compounds were prepared at 400x in DMSO , diluted 1:20 into media in an intermediate plate , and 1:20 into the final cell plate to yield final 0 . 25% ( v/v ) DMSO . All compound treatments were dosed in duplicate . Treated cells were incubated 72hrs before performing 1:4 dilution ( 50ul ) of virus-containing supernatant onto a plate of naïve Huh7 cells ( 150uls ) , seeded at 8x103cells/well 6hrs previously . Infected Huh7 cells were incubated 48hrs before washing 3x in PBS and fixing in 4% PFA . Fixed cells were washed in PBS and permeabilized using 0 . 1% triton X-100 ( v/v ) in PBS for 10min , room temperature . Following PBS wash , cells were immune-stained for NS5A to quantify infected cells . Sheep anti-NS5A antibody [90] was used at 1:2000 in PBS supplemented with 10% FBS , 16hrs at 4°C . Following 3x PBS washes , Alexafluor594 nm Donkey anti-Sheep antibody ( Invitrogen ) was added at 1:500 , 2hrs in the dark . Cells were washed twice in PBS and imaged using phase and red channels ( IncuCyte ZOOM ) . Infected cells positive for NS5A expression were quantified using IncuCyte parameters previously described [89] , normalized to DMSO controls . For the inflammasome components , the cDNA of human proIL-1β , procaspase-1 , NLRP3 and ASC were purchased from Origene . These cDNAs were subcloned into pEF vector which contains the human elongation factor-1 alpha ( EF1α ) promotor to drive ectopic gene expression . The inflammasome components are Flag-tagged for easy detection . Viral proteins were subcloned and expression was detected either by Flag antibody or gene-specific antibody . To amplify the core region from the pSJ [88] ( HCV JFH-1 sequence , genotype-2a ) plasmid , the following primers were ordered from IDT: Core-forward primer: 5’gcgcgcggccgcatgagcacaaatcct3’ , Core-reverse-primer: 5’gcgcggatccagcagagaccggaacggt3’ . For p7 ( genotype-2a ) and M2 , a minigene was ordered and cloned into either the pEF vector or the polycistronic pRRL-MND-vector ( kindly provided by Dr . Daniel Stetson , University of Washington ) . For influenza M2 ion channel protein , the sequence of influenza A virus ( A/Wyoming/03/2003 ( H3N2 ) , accession number: DQ849011 . 1 ) was obtained from NCBI . The JFH-1-E1E2 polyprotein expressing plasmid ( generously provided by Dr . Jane McKeating , University of Oxford ) . The patient core expressing constructs ( kindly provided by Dr . Stephen Polyak , University of Washington ) . To achieve inducible inflammasome response in the U2OS-reconstituted system the following amounts of plasmids were co-transfected: 1ng of ASC , 1ng of procaspase-1 , 100ng of NLRP3 and 100ng of proIL-1β . All plasmids including the viral protein expressing constructs or vector control were transfected at the same time . Samples were collected at either 24 or 48hrs post-transfection for ELISA and immunoblot analyses . For overexpression of HCV core or Flu-M2 in THP-1 cells , Lentiviral transduction method was employed . To generate Lentiviruses expressing HCV core or Flu-M2 or empty vector , the pRRL-MND-2A-eGFP-2A-Puro Lentiviral vector was used . The Lentiviruses were generated in 293 by transfecting the Lentiviral vector expressing gene of interest , pVSV expressing envelope and packaging vector pSPAX2 . 24hrs post transfection , fresh media was added and then collected lentiviral particle containing supernatant at 48hr post media change . The Lentiviral particles were filtered to remove cell debris before proceeding with transduction . To generate stable cells expressing the desired transgene , 5-10x10^6 THP-1 cells were transduced in 10cm dish and then selected with puromycin . Plasmids and associated materials can be provided upon request . For western blot , cell pellets were collected and washed once with 1x PBS . Then cells were lysed in modified RIPA buffer ( 50mM Tris-HCl , pH-7 . 5 , 150mM NaCl , 0 . 1 SDS , 1% triton-x/NP-40 , 1% Na-deoxycholate , 5mM EDTA ) supplemented with protease inhibitor cocktail . Protein from cell lysates were resolved by SDS-PAGE with either commercially purchased precast gels ( Biorad ) or in lab made-15% or 12 . 5% gels and then transferred onto PVDF membrane . Probing for the indicated protein was performed using a specific antibody or anti-Flag and actin was used as a loading control . For immunoprecipitation ( IP ) , cells were lysed in RIPA buffer . Cell lysate were incubated with anti-Core antibody or IgG control overnight at 4°C . The following day , IP was performed using protein G dynabeads and then immunoblot was performed as described . RNA from cells was extracted using the RNeasy mini kit ( Qiagen ) . After obtaining RNA , cDNA was made using the Biorad Iscrip Select cDNA synthesis kit . Then real time-PCR ( qRT-PCR ) was performed using SYBR green PCR master mix . IL-1β and GAPDH ( housekeeping gene ) primers were used ( both purchased from Qiagen ) . Cells were seeded on cover slips . After stimulation is complete , cells were washed with 1xPBS and then fixed with either 3% paraformaldehyde or ice-cold methanol . After fixation , cells were permeabilized with 0 . 2% of triton-x . Non-specific binding was blocked with 3% non-fat incarnation milk in 1x PBS . Staining with primary antibody was performed for 30min at room temperature . Alexa-Fluor-594 or alexa-Fluor-488 were used as secondary antibodies . The nucleus is stained with DAPI and wheat-germ agglutinin-594 ( ThermoFisher ) marks the plasma membrane . Imaging was performed using Nikon Eclipse Ti confocal microscope . Images were taken with 60X oil immersion objective . All images were mounted using the prolong gold antifade reagent . Images were acquired and processed using the NIS advanced research software . Calcium flux was measured in differentiated THP-1 cells . 2-5x10^6 cells were first stained with Indo-I-AM at 1:500 in 200ul-RPMI for 45min at 37°C . after staining with Indo-I , cells were washed with RPMI and either left untreated or stimulated with recombinant core for 10min-1hr . Samples were run on LSRII under the following settings: SS-355 , FS-231 , DAPI-A-352 and Hoechst-red-A-368 . For baseline calcium levels , untreated cells were acquired for 10–30 seconds then stimulated cells were acquired for 2-3min . Calcium flux is represented as a ratio of DAPI to Hoechst-red . Calcium flux represents a shift from 530/30-uv-Hoechst-red ( Blue , Ca-unbound ) to 405/20-uv-DAPI ( Violet , Ca-bound ) . Peripheral blood mononuclear cells ( PBMCs ) of healthy donors were obtained from TRIMA leukoreduction ( LRS ) chambers ( Bloodworks Northwest , WA ) . Total mononuclear cells were obtained by centrifugation using ficoll-paque plus ( followed manufacturer’s protocol ) . Briefly , the whole blood was diluted 1:2 with 1xPBS then 35ml of the diluted blood was layered on top of 15ml of Ficoll-paque plus in 50ml conical tube . Centrifugation was performed at room temperature at 300xg for 30-45min with no brake . After centrifugation is complete , 5–10 million total PBMCs were cultured in complete-RPMI containing 40-100ng/ml of human m-CSF . Two days after culturing , non-adherent cells were removed and fresh RPMI containing 40-100ng/ml of m-CSF was added . After five days of culturing , cells were collected by scrapping gently in RPMI then seeded 0 . 3–0 . 8 million cells per well of a 12-well plate . The following day cells were treated with different stimuli . The human blood samples were obtained from healthy donors . The samples were purchased from Bloodworks Northwest ( Washington ) . The samples are delivered in TRIMA leukoreduction ( LRS ) chambers and then processed to isolate PBMCs . Samples were non-identifiable . For ELISA , supernatant from both non-stimulated and stimulated cells were collected . To remove cell debris , cells were spun for 5 minutes at medium speed . ELISA was performed according to the manufacturer’s manual ( Biolegend , Inc . ) . For statistical analysis , student t-test or one-way Anova ( for multiple comparisons ) was performed . Any p value less than 0 . 05 was considered statistically significant . Graphs are represented as the means +/- SD .
|
This study deciphers the molecular mechanism of Hepatitis C virus ( HCV ) -induced hepatic inflammation . HCV triggers NLRP3 inflammasome activation and IL-1β release from hepatic macrophages , thus driving liver inflammation . Using biochemical , virological , and genetic approaches we identified the HCV core protein as the specific viral stimulus that triggers intracellular calcium signaling linked with phospholipase-C activation to drive NLRP3 inflammasome activation and IL-1β release in macrophages .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"method"
] |
[
"blood",
"cells",
"medicine",
"and",
"health",
"sciences",
"immune",
"cells",
"enzyme-linked",
"immunoassays",
"pathology",
"and",
"laboratory",
"medicine",
"hepacivirus",
"pathogens",
"immunology",
"microbiology",
"viral",
"structure",
"immune",
"receptor",
"signaling",
"viruses",
"rna",
"viruses",
"membrane",
"receptor",
"signaling",
"inflammasomes",
"immunologic",
"techniques",
"calcium",
"signaling",
"research",
"and",
"analysis",
"methods",
"immune",
"system",
"proteins",
"white",
"blood",
"cells",
"viral",
"core",
"animal",
"cells",
"proteins",
"medical",
"microbiology",
"microbial",
"pathogens",
"immunoassays",
"hepatitis",
"c",
"virus",
"hepatitis",
"viruses",
"virions",
"biochemistry",
"signal",
"transduction",
"cell",
"biology",
"flaviviruses",
"virology",
"viral",
"pathogens",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"macrophages",
"cell",
"signaling",
"organisms"
] |
2019
|
Modulation of calcium signaling pathway by hepatitis C virus core protein stimulates NLRP3 inflammasome activation
|
This study focused on the savannah tsetse species Glossina swynnertoni and G . morsitans centralis , both efficient vectors of human and animal trypanosomiasis in , respectively , East and Central Africa . The aim was to develop long-lasting , practical and cost-effective visually attractive devices that induce the strongest landing responses in these two species for use as insecticide-impregnated tools in population suppression . Trials were conducted in different seasons and years in Tanzania ( G . swynnertoni ) and in Angola and the Democratic Republic of the Congo ( DRC , G . m . centralis ) to measure the performance of traps ( pyramidal and epsilon ) and targets of different sizes , shapes and colours , with and without chemical baits , at different population densities and under different environmental conditions . Adhesive film was used to catch flies landing on devices at the remote locations to compare tsetse-landing efficiencies . Landing rates by G . m . centralis in both Angola and the DRC were highest on blue-black 1 m2 oblong and 0 . 5 m2 square and oblong targets but were not significantly different from landings on the pyramidal trap . Landings by G . swynnertoni on 0 . 5 m2 blue-black oblong targets were likewise not significantly lower than on equivalent 1 m2 square targets . The length of target horizontal edge was closely correlated with landing rate . Blue-black 0 . 5 m2 targets performed better than equivalents in all-blue for both G . swynnertoni and G . m . centralis , although not consistently . Baiting with chemicals increased the proportion of G . m . centralis entering pyramidal traps . This study confirms earlier findings on G . swynnertoni that smaller visual targets , down to 0 . 5 m2 , would be as efficient as using 1 m2 targets for population management of this species . This is also the case for G . m . centralis . An insecticide-impregnated pyramidal trap would also constitute an effective control device for G . m . centralis .
Diseases transmitted by tsetse flies , notably human African trypanosomiasis ( HAT or sleeping sickness ) and African animal trypanosomosis ( AAT or Nagana ) , are caused by the transmission of trypanosomes , and are still a serious health and economic burden in sub-Saharan Africa [1 , 2] . After a resurgence in cases in the 1990s [3] , increased treatment and vector control reduced the reported incidence of HAT from over 30 , 000 per year to below 3 , 000 per year in 2015 [1] . However , many more cases still go untreated , with an estimated 30 , 000 unreported cases in 2012 [3] , and recalcitrant HAT foci remain across the continent [4] . Despite recent improvements , the economic and social cost of AAT continues to be a major burden in rural areas , where it is a significant cause of poverty and malnutrition [5] . This study focuses on Glossina swynnertoni Austen ( Diptera , Glossinidae ) and G . morsitans centralis Machado , two closely related savannah or Morsitans group tsetse [6] . Important information on G . pallidipes was also collected and is reported . Both G . swynnertoni and G . m . centralis are efficient vectors of human and animal trypanosomiasis [7 , 8] and HAT foci persist within the geographic ranges of these species [4] . Historically , in northern Tanzania , G . swynnertoni was found to have a higher trypanosome infection rate than G . pallidipes [7] , but confirmation of infection with T . brucei required the use of special techniques [9] . Both species have since been the focus of several studies in the context of HAT cases in the Serengeti [10 , 11 , 12] . The trypanosome transmission capacity of G . m . centralis is equal to or greater than that of G . pallidipes , depending on the trypanosome species [8 , 13] . G . swynnertoni is restricted to north-west Tanzania and south west Kenya [14 , 15] , whereas G . m . centralis has a much more extensive distribution extending across the southern tsetse belt from western Tanzania and southern Uganda westwards through Zambia across the south-east of the Democratic Republic of the Congo to the eastern limits of Angola , with an isolated pocket in central Angola [16] ( Fig 1 ) . The former population in northern Botswana centred on the Okavango Delta , an important pastoral and conservation region , has been successfully eradicated following a concerted control programme of aerial spraying with insecticides and use of insecticide-impregnated visual targets in 2001 and 2002 [17] . The region has been tsetse-free for over 10 years [18] . Both G . swynnertoni and G . m . centralis are abundant in and around conservation areas [19] which are an important source of revenue particularly in Kenya , Tanzania [20] and parts of Zambia [21] , but where effective vector management can be a particular challenge as an abundant wildlife reservoir means tsetse populations can reach high densities [22] . The transmission risk to neighbouring pastoralists and their livestock is very high , notwithstanding concerns for tourists , park staff [23] , and even conservation programmes for endangered species such as the black rhinoceros Diceros bicornis [24 , 25] . Visually-attractive control devices such as insecticide-impregnated traps [26] and targets [27 , 28] have been widely used to control savannah tsetse since the 1980s [29] , including G . m . centralis [30] and G . swynnertoni [31 , 32] , although their use has been sporadic and often on a small scale [32] . The deployment of insecticide-impregnated targets alone has been successful in eliminating tsetse from geographically isolated pockets , such as the Lambwe Valley in Kenya [33] . They are a suitable environmentally friendly technique to use in joint efforts in and around game reserves [31] and have also been widely used to create barriers to prevent tsetse re-invading cleared areas [17 , 30] . Large targets , up to 1 . 5–1 . 8 m wide , have been traditionally used in eastern and southern Africa to manage savannah tsetse populations [33 , 34] . Some research has advocated the use of all-black targets for use against savannah tsetse [28 , 35] , but blue-black or all blue targets have been shown to be most effective against Morsitans group tsetse [36–38] and are now the most commonly advocated [34 , 39] . Very large numbers of insecticide-impregnated targets need to be deployed and maintained to clear an area of tsetse and to create effective-barriers to prevent re-invasion . In the Tanzanian National Parks alone , over 20 , 000 targets were deployed between 2007 and 2010 [19] . The cost of materials , deployment and maintenance are major outlays and the traditionally large targets are also prone to wind damage in sandy soils and theft can also be a problem [30] . Recent research on riverine or Palpalis group tsetse has shown that much smaller targets ( 0 . 25–0 . 5 m2 ) can capture more flies per m2 than larger targets and would be more cost-effective in programmes targeting species such as Glossina fuscipes fuscipes and G . palpalis palpalis [40–43] . In contrast , for savannah species such as G . morsitans morsitans and G . pallidipes , research in Zimbabwe has shown that they would not be as effective as larger targets [28] . However , it appears that G . swynnertoni may respond differently to other savannah tsetse . Field trials by Mramba et al . [44] in the Serengeti made between 2009 and 2012 , which were part of a pan-African WHO-TDR initiative on maximising the efficiency of visual baits for tsetse , showed that smaller sized horizontal ( wider than high ) 0 . 47 m2 blue leg panels and 0 . 5 m2 horizontal blue and blue-black-blue targets are equally efficient at inducing landing by G . swynnertoni as 1 . 5 m2 and 1 m2 blue-black or blue-black-blue targets previously used in East Africa . In these trials targets of 0 . 25 m2 were less efficient . Following on from these trials , we set out to identify the most appropriate reduced target shape and design for use as a visual control device for G . swynnertoni . Our aim was to maximise the efficiency and cost-effectiveness of these devices . The trials were repeated with G . m . centralis , a close relative of G . swynnertoni , to see whether this savannah species shared the same behavioural responses or was more akin to G . m . morsitans . Information on G . pallidipes is reported here where this species was also present at field sites . Additional trials were also conducted with G . m . centralis to measure the performance of pyramidal and epsilon traps , which are still widely used to control and monitor this species . The relative performance of these traps was compared to targets with and without a chemical bait for G . m . centralis . Such information had already been collected for G . swynnertoni in an earlier set of trials [44] .
Studies on G . swynnertoni and G . pallidipes were conducted at one site in 2013 in Tanzania . Studies on G . m . centralis were undertaken at two sites in central Angola ( at one site in 2010 and at another in 2014 ) and at one site in the Democratic Republic of the Congo in 2014 . A brief description of each site is given below . The field trials were made either on public land or on lands where owners/residents gave permission for the field trials to be conducted . In all three countries a series of 1 m2 and 0 . 5 m2 rectangular and square targets made of equal vertical rectangles of blue and black or all-blue cloth were tested ( Table 1 ) . Rectangular targets , termed here horizontal oblongs , had their long sides set up horizontal to the ground . A selection of different dimensions and designs was chosen to assess the influence of target shape , size and colour on fly landing rates . In addition , pyramidal traps [45] were included in the 2014 Angolan and Congolese trials and epsilon traps [46] were used in the 2010 Angolan trials . Catches and landing rates for pyramidal traps were compared with landing rates on targets in trials carried out earlier at the same site in Tanzania [44] and so were not repeated . All devices were set in the open , 30 cm above ground , and vegetation was removed within several metres of each device . In all devices except the epsilon trap , two fabrics were used: C180 Azur 623 phthalogen blue 100% cotton ( 180 g/m2 , TDV , Laval , France ) with a reflectance peak at 460 nm as measured with a Datacolor Check Spectrophotometer ( Datacolor AG , Dietlikon , Switzerland ) and a 100% polyester black ( 225 g/m2 , Q15093 Sunflag , Nairobi ) . The epsilon trap was made of blue polyester ( PermaNet , Vestergaard Fransen , Denmark ) , also with a reflectance peak at 460 nm ( see Supporting Information S1 Fig for spectral reflectance curves ) . To enumerate flies landing on the devices , one-sided adhesive film ( 30 cm wide rolls , Rentokil FE45 , Liverpool , UK ) was stitched onto both sides of the targets and onto the cloth panels of the pyramidal traps . These fly catches permitted measurement of tsetse landing rates on the different devices , the essential behavioural response underlying the use of insecticide-impregnated visual control devices for tsetse . The adhesive film does not affect spectral reflectance except in the ultra-violet spectrum , absorbing virtually all UV wavelengths below 380 nm . This is due to the addition of a UV absorber in the glue . In addition , spectrophotometric measurements of light reflected from adhesive film applied onto the same fabrics used in this study indicate that all wavelengths in the UV range were mostly absorbed by the fabrics [43] . In the 2010 Angolan trial , a 1 x 1 m square of adhesive film alone ( without any cloth backing ) was compared to cloth targets with adhesive film attached to both sides to ascertain whether adhesive film in itself attracts flies . A 1:4:8 mixture of 3-n-propylphenol ( P ) , 1-octen-3-ol ( O ) , and p-cresol ( C ) ( Ubichem Research LTD , Budapest/Hungary , global purity of 98% ) with acetone ( A ) was used as an attractant for experiments comparing performance ranking of baited devices based on its efficacy for several tsetse species . This combination is termed POCA bait and was made up as per Torr et al . [47] . Sachets of 4 cm x 5 cm 500 gauge / 0 . 125 mm polyethylene containing 3 g of the 1:4:8 mixture were placed below the visual devices , 10 cm above ground , alongside a 250 ml bottle buried up to the shoulders containing acetone with a 2 mm aperture in the stopper . In all field trials randomization was set up using design . lsd in the package agricolae [48] R version 3 . 01 [49] . Data were analysed using a linear model including the following additional packages: MASS [50] and multcomp [51] . Analysis was performed on log ( x+1 ) transformed data including day and position as additional explanatory parameters . Position had no significant effect in any field trial ( P > 0 . 05 , F-test ) and running the model separately for replicates also revealed no significant effect in any of the field trials ( P > 0 . 3 , F-test ) . Tukey contrasts were calculated to compare treatments . The Wilcoxon paired test was used to compare fly landings on the blue and black portions of targets .
The largest target tested induced the highest number of G . swynnertoni and G . pallidipes to land on it , but this was not significantly greater than the daily landing rates for the 0 . 5 m2 blue-black oblong target for both species ( P > 0 . 05; Table 2 & Fig 2 ) . Landings on the 0 . 5 m2 blue/black oblong were particularly high for G . swynnertoni ( 90% of the daily landing rate on the 1 m2 square target ) . For G . m . centralis , daily landing rates were highest on the 1 and 0 . 5 m2 blue-black oblong targets but were not significantly different ( P > 0 . 05 ) from the sticky pyramidal trap and 0 . 5 m2 square blue-black target ( Table 2 & Fig 3 ) . The largest square blue-black targets ( 1 m2 ) showed the highest landing rates for G . swynnertoni and G . pallidipes; landings were reduced by around 50% on the equivalent 0 . 5 m2 square targets ( Table 2 & Fig 2 ) . For both species , landing rates were higher on the 0 . 5 m2 blue-black oblong targets than on the equivalent square targets , most noticeably for G . swynnertoni ( 62% more ) but this difference was not significant ( P > 0 . 05 ) . The daily landing rates for G . m . centralis in the DRC were nearly the same on the 1 and 0 . 5 m2 blue-black oblong targets ( 17 . 9 and 19 . 1 flies per day , respectively ) , which were 20–25% more than on the equivalent 0 . 5 m2 square target ( Table 2 & Fig 3 ) . Very similar trends were observed in the smaller landing rates recorded in Angola . None of these differences were significant ( P > 0 . 05 ) . The blue-black targets performed better than their equivalents in all-blue for both G . swynnertoni and G . m . centralis , with landings significantly lower for G . swynnertoni on the all-blue square target and for G . m . centralis in the DRC on the all-blue oblong target ( P < 0 . 05 ) . In contrast , there was no significant difference between daily landing rates for G . pallidipes on the blue-black and equivalent all-blue 0 . 5 m2 oblong and square targets ( Table 2 ) . Landing rates relative to colour / size / shape were also equivalent for the experiment conducted in Angola at low numbers of G . m . centralis . When the daily landing rates are adjusted to a uniform size of 1 m2 for the targets of various shapes and sizes , the optimal landing rates were recorded on the 0 . 5 m2 blue-black oblong targets for all three species ( Table 3 ) . Glossina pallidipes was the only species with similar landing rates on the 0 . 5 m2 all-blue oblong ( Table 3 ) . Although landing rates per m2 were approximately double those of the 1 m2 targets for G . swynnertoni and G . m . centralis , landing rates per m2 were only slightly higher ( ~10% ) for G . pallidipes , and were actually lower for the same shaped smaller target . The rank order in performance of devices was the same in the baited and unabaited experiments ( Table 4 ) . Addition of POCA bait had no influence on the proportion of G . m . centralis flies entering the epsilon trap relative to landings on the blue-black cloth target , with only slightly fewer flies entering the baited trap compared to that recorded in the unbaited experiment ( Table 4 ) . In contrast , the addition of the POCA bait increased the proportion of flies entering the pyramidal trap compared to landings on the cloth target by over 60% ( Table 4 ) .
One of the objectives of the present study was to quantify the performance of pyramidal traps relative to targets for several savannah tsetse , as this trap ( or similar monoconical traps such as the Vavoua ) is often used as a generic tsetse sampling device in areas with many species [42 , 43 , 44 , 52] . In Tanzania , pyramidal traps are often used for sampling G . swynnertoni populations ( following the early work of Muangirwa [53] ) , but landing efficiency is about 50% lower than for a target , and trapping efficiency only about a quarter [44] . Despite this , pyramidal traps continue to be used for monitoring for practical reasons , although their large-scale use in the control of G . swynnertoni is not recommended . We have no explanation for the difference in trapping efficiency of the pyramidal trap for G . m . centralis estimated at 25% in Angola and 68% in the DRC other than to note that the pertinent field trial was made in the dry season in Angola and in the wet season in the DRC . In Angola , where G . m . centralis is present , insecticide-impregnated pyramidal traps rather than targets are widely-used for tsetse control [cf . 42] . Here , we show that the number of G . m . centralis landing on pyramidal traps covered with adhesive film is similar to , but somewhat lower than , the numbers landing on the best blue-black target ( 68–95% of the best target in two experiments ) . Also , no flies were captured in the cages of the sticky traps in these trials . This tsetse species [54] , like most savannah species [44 , 52] , seems to have a very low propensity to enter a trap cage without first landing on the cloth , unlike some riverine species such as G . palpalis palpalis [42] . This behavioural trait combined with the relative attractiveness of the pyramidal trap means that insecticide-impregnated pyramidal traps are sufficiently effective fly-killing devices to support their continued deployment for the control of G . m . centralis [42] . In countries such as Angola , hanging traps from bushes and stems of trees is a typical deployment strategy in wooded savannah ( Fig 4 ) where it has proven to be more practical and economical than implanting supports for targets in the ground [55] . This current study was carried out following on from other target/trap comparisons we have made across Africa [42 , 43 , 44 , 52] , and hence our trials focused on optimizing lessons learned in previous work , particularly the unexpected finding of the high performance of small targets for savannah species in Tanzania and Kenya [44] . In our current trials in three countries for three related savannah species , the highest landing rates were most frequently recorded on the 1 m2 blue-black target that we adopted as a standard for coordinated experiments . This design was expected to be highly-attractive , even when unbaited , based on a large body of previous work by many researchers [34] . However , our results , consistent with Mramba et al . ’s earlier findings on G . swynnertoni [44] , also show that some smaller 0 . 5 m2 targets can be just as efficient for other savannah species ( highest efficiency index in terms of fly landings per m2 of cloth with statistically equivalent total landings relative to a 1 m2 blue-black target ) and should be considered as sampling/control devices . These consistent findings for several species in different countries contrast with results of a test of a “tiny” target ( 0 . 06 m2 ) for two savannah species in Zimbabwe ( G . pallidipes and G . m . morsitans ) , where Torr et al . [28] found that very few tsetse were attracted to or landed on a 0 . 25 x 0 . 25 m square , all-black target ( with and without flanking e-nets and/or baits ) . The simplest explanation for such dramatically different results among experiments in different countries ( given that G . pallidipes is represented in both sets of trials ) is that the Zimbabwe trials tested only small targets that were all-black , i . e . without a blue element . In the key studies leading up to modern targets for savannah tsetse , Vale [56] concluded that bicoloured blue/black panels would make the best targets , and there are many examples of the importance of blue in tsetse vision [57 , 58] . The presence of a contrasting blue element may be critical for attracting certain tsetse to small versus large targets [40] . Any interpretation of the importance of blue cannot necessarily be inferred from the preferential landing by tsetse on the blue portions of devices tested here ( Supporting Information S2 Table ) . In earlier studies designed to assess the effect of adding adhesive film to visual devices , blue-black 1 m2 targets with no adhesive film applied , and similar targets covered on both sides by adhesive film with the sticky side facing inwards ( i . e . with the shiny plastic base facing outwards ) , were placed within electric grids designed to kill alighting flies [42 , 52] . These experiments showed that addition of a specular component to the light reflecting from cloth significantly reduced landings on the black but not the blue portion of targets for other tsetse species ( G . p . palpalis , G . tachinoides and G . gambiensis ) . As the Rentokil film is also selectively UV-absorbing , these effects could have been due to the fact that the appearance of matt-finished phthalogen blue cotton and black polyester fabrics was also altered in terms of UV reflectance . High UV reflectance is typically assumed to negatively affect tsetse responses to objects independent of visible reflectance based on statistical trends in tests of a wide variety of materials . However , spectrophotometric measurements of light reflected from adhesive film applied onto the phthalogen blue cotton and black polyester fabrics as on tsetse visual devices indicate that all wavelengths in the UV range were in any case mostly absorbed by the fabrics [43] . Also , the adhesive film served to increase landings by G . palpalis gambiensis on the blue portion of targets [52] . This suggests that other complex fly visual phenomena may be at play [59] and serves to underline that colour preferences using this sticky method of enumeration should be interpreted with caution [42 , 52] . Also , as noted by Vale [56] , blue-black targets generally perform better for savannah species , including G . pallidipes and G . m . morsitans , than all-black targets [36 , 38] . As a relevant example , Knols et al . [25] gradually replaced 1 . 8 m wide x 1 m tall all-black targets of the Zimbabwe design with bicoloured blue-black targets for the control of G . m . centralis in Zambia . Lastly , the presence of a blue element of the correct spectral characteristics ( including ultraviolet reflectance [58] ) has been shown to be important in the optimization of small targets for riverine tsetse [40 , 43 , 60] . The use of very small targets ( i . e . 0 . 25 m2 or smaller ) as proposed for some riverine tsetse [40 , 41] may not prove to be suitable for savannah tsetse , given our previous results for 0 . 25 m2 targets for G . swynnertoni [44] , and the poor results for all-black “tiny” targets from Zimbabwe cited above [28] . Nevertheless , the cost benefit and other practical implications of deploying targets somewhat smaller than 1–1 . 5 m2 warrant serious consideration . Control campaigns and the establishment of barriers against re-invasion require thousands to tens of thousands of visual targets , hence “size matters” [17 , 31] . In addition , in regions where wind damage and implanting supports for targets can be difficult ( very hard ground or loose sandy soils [30] ) deploying smaller targets is a practical solution , provided they remain efficient at inducing landing when left in situ for long periods of time . We therefore continue below with a more focused discussion of the performance of the 0 . 5 m2 target designs for the three savannah species studied here . In Tanzania , equal vertical rectangles of blue-black-blue have been traditionally used as targets for tsetse following the initial recommendations of Vale [56] in Zimbabwe . However , as we previously found no difference in the performance of blue-black-blue and blue-black targets in phthalogen blue cloth for inducing landing by G . swynnertoni [44] , we used the simpler blue-black configuration for further tests of smaller targets . For G . swynnertoni , and G . m . centralis ( in the DRC ) , landings on the blue-black targets were 55–75% higher than on the all-blue devices . A black portion was therefore an essential element for inducing landing in these two species and its contribution would probably have been more significant in absence of the adhesive film ( see above ) . This is in contrast to G . pallidipes , where all-blue targets were found to be as efficient as blue-black targets . Glossina m . centralis is genetically closer to G . swynnertoni than to G . m . morsitans [6] and this may be a reason why its landing behaviour is more similar to the former . The higher preference for the blue-black over the all-blue devices by G . swynnertoni is greater than the 30% increase recorded earlier by Mramba et al . [44] at the same site , when flies were more evenly distributed between blue and black . This may be a seasonal difference as revealed in the early work of Vale [56] . Horizontal oblong targets appear to be better at inducing landing than square targets for certain tsetse species , such as the riverine species G . tachinoides [43 , 61] and G . f . fuscipes [40 , 43] , especially for smaller targets . Increasing target width has also been found to increase landing rates by certain savannah species , such as G . austeni [62] and G . m . morsitans and G . pallidipes [55] . We therefore tested whether shape was a factor affecting landing efficiency for the three savannah tsetse species studied here using carefully-matched small target designs . We found that in all countries , irrespective of savannah species or season , horizontal oblong targets were better at inducing landing than an equivalent size square target , confirming our initial supposition . Our results have shown a close predictive correlation between the length of horizontal edge and tsetse landing rates independent of colour for all three species , in the target size ranges investigated . The exploitation of edge or border effects through the incorporation of simple geometric shapes/borders is a relatively unexplored area of research for improving targets for tsetse [63] as is the colour/background contrasts in targets [64] . In the extensive literature on tsetse visual responses , only a few researchers have systematically examined how tsetse land on different parts of large targets in relation to potential edge and colour/contrast effects [56 , 61] . Since Vale established that most tsetse land in the centre of targets [56] it could be that visual targets with a longer horizontal aspect with respect to ground are better at accommodating landings by fast-flying tsetse . In the laboratory , a preference for the edges of objects by G . m . morsitans is a particularly interesting finding [65] . A similar landing preference for the horizontal edge of targets by G . f . fuscipes was observed in the field by Oloo et al [43] . Vreysen et al [61] tested targets with horizontal or diagonal arrangements of solid blocks of colour to discern the impact on total landings by G . austeni and found a strong preference for the bottom corner edge . If the three species studied here truly have similar innate behavioural responses to a horizontal edge phenomenon , this could explain why the horizontal oblong , with a higher edge/surface area ratio than the square , and with longer upper and lower edges than the square , was more efficient at inducing landings . The use of the POCA bait has been shown to increase trap entry by flies for several savannah tsetse , notably G . m . morsitans and G . pallidipes [66] . Trials in Kenya and Tanzania on G . swynnertoni [44] showed that POCA could double pyramidal trap entry relative to landing on blue-black targets , but this increase was inconsistent and was not recorded in all circumstances . In this study in Angola , in a single trial with G . m . centralis , the addition of POCA increased pyramidal trap entry by 60% compared to an unbaited trap , relative to landings on the target . In contrast the POCA bait did not influence on entry into the epsilon trap . Earlier work has already shown that the epsilon trap catches fewer G . swynnertoni than conical trap designs [53] and would appear unsuitable in these countries as a monitoring device for these two species . Its single entry point and the fact that it is placed lower on the ground where it is more easily hidden by tall grass may also be contributory factors . However , in contrast , the epsilon trap has proved to be a satisfactory tsetse trapping/monitoring device in southern Africa ( e . g . in Botswana and Zimbabwe ) [67] . Considering the very modest improvements in trap entry by G . m . centralis with POCA , which are similar to earlier results with G . swynnertoni [44] , and previous failures to substantially improve G . swynnertoni catches with chemical baits ( i . e . double or more ) [54 , 68 , 69] , there appears to be little benefit in deploying and maintaining baits for controlling these species . Simply increasing the deployment of smaller targets may be a more cost-effective strategy . This study has confirmed earlier findings on G . swynnertoni that smaller visual targets , down to 0 . 5 m2 , would be as efficient as using 1 m2 targets visual targets for this species . This is also the case for G . m . centralis . To maximise the efficiency of smaller targets , horizontal rectangles with respect to ground that have both a black and phthalogen blue element appear to be best . These two features induced the highest landing response . All-blue devices were as efficient as blue-black devices for G . pallidipes . Adhesive film was used as a convenient alternative to other techniques to catch flies landing on visual devices at the remote locations to compare tsetse-landing efficiencies . However , because of interpretation difficulties inherent to the use of adhesive film to catch flies that land on visual targets , further studies with other techniques for intercepting flies landing on or circling targets ( e . g . electric grids and targets with netting panels ) are nevertheless still needed to better define the most economical and practical target for the control of all three species . Insecticide-impregnated pyramidal traps are also effective devices for the control of G . m . centralis as they induce a strong landing response and hence would achieve the desired end-point of killing flies . Although they are not as economical as smaller targets , their continued use would be appropriate where hanging traps from tree branches would be less problematic than the implantation of supports for targets in the ground ( e . g . in wooded savannah ) .
|
Glossina swynnertoni is restricted to open savannah in northwestern Tanzania and southwestern Kenya whereas G . morsitans centralis has a much wider distribution from western Tanzania/southern Uganda westwards through Zambia and southeast of the Democratic Republic of the Congo ( DRC ) to Angola . Both are savannah tsetse and are efficient vectors of human and animal trypanosomiasis . In comparison to other tsetse species , relatively little work has been done to test the efficacy of traps and targets for controlling G . swynnertoni and G . m . centralis . To determine the most visually-attractive and practical objects we conducted field tests with devices of various shapes , sizes and colours in Tanzania , DRC and Angola in different years , seasons , environmental conditions and at different population densities . The strongest landing responses were on 0 . 5 m2 horizontal rectangular targets with respect to ground that had both black and phthalogen blue elements with fly landing rates not significantly lower than on equivalent 1 m2 targets used till now for both species . The pyramidal trap proved efficient as a landing stimulus as targets of either size for G . m . centralis . Insecticide-impregnated blue-black 0 . 5 m2 cloth targets show promise as cost-effective devices for management of G . swynnertoni and G . m . centralis populations .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] |
[
"invertebrates",
"livestock",
"medicine",
"and",
"health",
"sciences",
"african",
"trypanosomiasis",
"angola",
"geographical",
"locations",
"tropical",
"diseases",
"parasitic",
"diseases",
"animals",
"parasitic",
"protozoans",
"glossina",
"research",
"design",
"protozoans",
"tanzania",
"materials",
"science",
"neglected",
"tropical",
"diseases",
"field",
"trials",
"africa",
"research",
"and",
"analysis",
"methods",
"infectious",
"diseases",
"zoonoses",
"protozoan",
"infections",
"trypanosomiasis",
"insects",
"adhesives",
"agriculture",
"arthropoda",
"people",
"and",
"places",
"trypanosoma",
"eukaryota",
"biology",
"and",
"life",
"sciences",
"materials",
"physical",
"sciences",
"organisms"
] |
2018
|
Standardising visual control devices for Tsetse: East and Central African Savannah species Glossina swynnertoni, Glossina morsitans centralis and Glossina pallidipes
|
An accurate early warning system to predict impending epidemics enhances the effectiveness of preventive measures against dengue fever . The aim of this study was to develop and validate a forecasting model that could predict dengue cases and provide timely early warning in Singapore . We developed a time series Poisson multivariate regression model using weekly mean temperature and cumulative rainfall over the period 2000–2010 . Weather data were modeled using piecewise linear spline functions . We analyzed various lag times between dengue and weather variables to identify the optimal dengue forecasting period . Autoregression , seasonality and trend were considered in the model . We validated the model by forecasting dengue cases for week 1 of 2011 up to week 16 of 2012 using weather data alone . Model selection and validation were based on Akaike's Information Criterion , standardized Root Mean Square Error , and residuals diagnoses . A Receiver Operating Characteristics curve was used to analyze the sensitivity of the forecast of epidemics . The optimal period for dengue forecast was 16 weeks . Our model forecasted correctly with errors of 0 . 3 and 0 . 32 of the standard deviation of reported cases during the model training and validation periods , respectively . It was sensitive enough to distinguish between outbreak and non-outbreak to a 96% ( CI = 93–98% ) in 2004–2010 and 98% ( CI = 95%–100% ) in 2011 . The model predicted the outbreak in 2011 accurately with less than 3% possibility of false alarm . We have developed a weather-based dengue forecasting model that allows warning 16 weeks in advance of dengue epidemics with high sensitivity and specificity . We demonstrate that models using temperature and rainfall could be simple , precise , and low cost tools for dengue forecasting which could be used to enhance decision making on the timing , scale of vector control operations , and utilization of limited resources .
Previous study by Hii et al . ( 2009 ) has shown that elevated weekly mean temperature and cumulative rainfall influence the risks of dengue cases in Singapore at lag times up to 20 weeks with higher relative risks of dengue cases at time lag of 3–4 months [15] . Also , a recent study by Hii et al . ( 2012 ) has suggested that a dengue early warning issues about 3 months in advance could provide sufficient time for an effective mitigation [27] . Based on previous findings , this study aims to develop a simple , precise , and low cost early warning model to enhance dengue surveillance and control in Singapore . Hence , our objectives were first to develop a weather-based dengue forecasting model to project dengue cases or potential outbreak that would allow sufficient time for local authorities to implement preventive measures and second to validate and report the performance of the forecast .
Singapore is a highly urbanized island state nation situated at 1° . 17′N and 103° . 50′E of the equator with a land size of about 700 km2 . As of 2011 , the island accommodates a population of around 5 . 2 million with about 93% of the population residing in either government or private high rise residential buildings [28] . As a tropical country , Singapore experiences high temperature , rainfall , and humidity year round . Weather in Singapore is influenced by the monsoon rain-belt with highest rainfall between December and early March [29] . Weekly dengue cases from 2000 to 2011 were obtained from the weekly infectious diseases bulletins of Communicable Diseases Division , Ministry of Health ( MOH ) Singapore [30] . The Infectious Diseases Act in Singapore stipulates mandatory disease notification within 24 hours of diagnosis by all medical clinics and laboratories . Daily mean temperature and rainfall recorded by the Changi Airport meteorological , southeast of Singapore , for the period of 2000–2011 were extracted from the National Climatic Data Centre , National Oceanic and Atmospheric Administration ( NOAA ) , USA [31] . Weather data were provided to the NOAA by the Meteorological Department of National Environment Agency , Singapore under the regional data collaboration agreement . The daily mean temperature was based on 24 hours average temperature; while daily rainfall was the summation of 24 hours rainfall collected using rain gauges . We developed a dengue forecasting model using time series Poisson multivariate regression that allowed over-dispersion of data . Mean weekly predicted cases were estimated through regression on multiple independent variables that include retrospective dengue cases , weekly mean temperature , weekly cumulative rainfall , trend , epidemic cycles and seasonal factors . The forecasting model was developed using three processes: 1 ) model construction and training using data from 2000–2010; 2 ) model validation by forecasting cases in 2011–2012; and 3 ) sensitivity tests on outbreak diagnoses . Our statistical analysis was conducted using R [32] and STATA 11 ( 2009 StataCorp LP , Texas ) based on 95% confidence interval . We modeled dengue distribution patterns using retrospective data and then extrapolated the patterns several weeks ahead . We developed dengue forecasting models based on assumption that the past dengue distribution patterns will , to a large extent , continue in the future [33] . Bivariate equation ( Dx ) for each independent variable was first formulated using quasi Poisson regression and subsequently combined to form a multivariate model that takes multiple factors into consideration . where represents weekly average number of predicted dengue cases as a function of independent variable x . One characteristic of infectious disease is the influence of past cases on the number of current cases . Therefore , autoregression was included in the model to account for the serial relationship between past and current cases . We derived possible lag time of serial correlation through data analysis using Autocorrelation Function ( ACF ) , Partial Autocorrelation Function ( PACF ) , and prior knowledge on dengue transmission . ACF analysis on dengue data showed gradual decreasing spikes that indicated strong autocorrelation between past and current cases; whereas , PACF cut off after the 4th spike suggesting a lag time of 4 weeks . However , previous studies have shown possible autocorrelation of dengue cases for longer period due to complex reasons that influence the dynamic of dengue transmission [34] . Thus , we examined lag times ranging from 4–12 weeks and selected the optimal lag order using model selection and validation tests . We denote DAR as the autoregression of dengue cases k weeks before forecast in week t . The effects of autoregression on dengue cases are computed as: ( 1 ) where = dengue cases at lag week k , = the constant number of dengue cases , = parameter of autoregression at lag week k . We examined the time gap between exposure to weather conditions and subsequent occurrence of dengue cases using cross correlation function and literature reviews . Correlation between temperature and dengue showed sine wave oscillating at about 24-weeks cycle or interval with stronger positive association between lag week 9 and 17 . While correlation between rainfall and dengue revealed different length of time cycles with a negative relationship from week 0 to 22 . It is possible for dengue transmission to occur several months after favorable weather conditions as mosquito eggs can withstand desiccation for several months with an average egg survival time of 18 . 3 weeks for Aedes aegypti [35] . We identified the optimal lag term and weather time cycle for forecasting by testing lag terms 1–20 weeks with various data cycle periods of weather variables ranging from 12 to 24 weeks . Piecewise regression was used to consider a non-linear relationship between weather and dengue cases . Thus , we partitioned weather data into 4 equally spaced percentiles with knots at 25th , 50th , and 75th percentiles using spline function . The impact of weekly weather on dengue cases is estimated as follows: Let depicts the number of dengue cases as a function of weekly mean temperature: ( 2 ) where is the baseline number of dengue cases; = parameter of mean temperature at lag term f in p range of mean temperature; f = t - ( L+n ) ; t = week; L = lag term in week; n = data cycle period of weekly mean temperature; p = temp11 to temp14 derived from piecewise spline function . Let denotes number of dengue cases as a function of weekly cumulative rainfall: ( 3 ) where β0 is the baseline number of dengue cases; = parameter of rain at lag term g in q range of weekly cumulative rain; g = t - ( L+m ) ; t = week; L = lag term in week; m = data cycle period of weekly cumulative rainfall; q = rain11 to rain14 derived from piecewise spline function . To account for non-climatic factors such as vector control , circulating serotypes of dengue virus , and other factors that influence the number of dengue cases , we performed graphical examination on the trend , cycle , and seasonal distribution patterns of dengue cases over the period 2000–2010 . The trend of dengue cases increased with cyclic variation from 2000 to peak at 2005 before declining thereafter . Increases in dengue cases were generally observed in the second half of each year; while major epidemics occurred in 2004–5 and 2007 . We included a curvilinear or parabola and sine function to account for trend , epidemic cycle and seasonal influence on dengue cases during the study period , respectively . Let represents dengue cases influenced by trend over the study period: ( 4 ) whereas , = constant , = parameter measures the trend , t = week , = point in time where maximal impact of trend is reached . Let depicts cyclical and seasonal impacts on dengue cases: ( 5 ) where = constant or baseline contribution of cycle and season , = parameter that gives rise to cyclical and seasonal effects , t = week . Dengue cases are subject to interactions of multiple complex factors . Thus , we composed a Poisson multivariate regression model by combining equations ( 1 ) to ( 5 ) to account for influences of multiple factors on dengue cases . We also adjusted our findings for population change by offsetting midyear population ( offset = log ( pop ) ) during the study period . Now we summarize our model as follows:and ( 6 ) where is the average predicted dengue cases at week t , is the constant derived from multivariate model , and if all the independent variables remain constant . Model selection was based on lowest Akaike's Information Criterion ( AIC ) or Bayesian Information Criterion ( BIC ) and standardized Root Mean Square Errors ( SRMSE ) of prediction . Residuals diagnoses were performed to examine and validate a good fit of the model using sequence plots to ensure sufficiency of model and constant variation of errors , and residual normality plots to examine normal distribution of errors . Furthermore , plots of fitted versus reported dengue cases were also examined for good fit of the model . Upon selection of a model that best described the data based on 2000–2010 dengue cases , we used the model to forecast cases for years 2011 and 2012 . In the first 16 weeks of 2011 , we used data in the last quarter of 2010 to forecast dengue incidence from January–April 2011 . Subsequently , we input only weather data for January–December 2011 and prompt our model to forecast dengue cases from week 17 of 2011 to week 16 of 2012 . Only weather data that were known at the time of issuing the 16 weeks forecast were used . Forecasted dengue cases in each period were then computed as autoregression for subsequent 16-week forecast . The forecast was repeated iteratively over time to generate the forecast for 2011–2012 . Finally , we analyzed forecast precision by comparing forecasted cases against real-time clinical and laboratory-confirmed dengue cases ( external data ) reported by the MOH in each week . We also performed sensitivity tests on these data . An effective dengue forecast provides accurate information and minimizes false alarms so as to reduce unnecessary wastage of limited resources . We therefore further identified the optimal model using C-statistics or a Receiver Operating Characteristics ( ROC ) curve to evaluate and compare the sensitivity of the selected model in detecting true dengue outbreaks during both the model development and forecasting periods . The ROC curve analyzes the sensitivity or true positive rate of a model to predict outbreaks versus the false positive rate ( 1-specificity ) . The area of the ROC curve is the proportion of accurate prediction and this measures overall ability of a model to distinguish between a true outbreak and non-outbreak . We obtained annual outbreak or epidemic thresholds that were available for 2004–2011 from epidemiological reports published by the MOH Singapore . The local authorities computed warning level and epidemic threshold annually and dengue epidemic would be declared if total weekly cases exceed the epidemic threshold . We computed binary outcome of positive or negative outbreaks in each year based on given epidemic threshold values .
During the forecast for 2011–2012 , the optimal model forecasted cases versus actual clinical reported dengue cases gave an average error of 0 . 32 of the standard deviation of reported cases . As shown in Figure 3 , the model forecasted cases with lower errors against actual reported cases in the 2nd half of the year . In 2011 , reported clinical cases exceeded the epidemic threshold for 5 consecutive weeks between weeks 27 and 31 . Our model forecasted all the cases above the epidemic threshold with one false positive case at week 32 . We have matched our forecast against external data or the real-time reported weekly cases from MOH up to week 12 of 2012; thus far , the model forecasted dengue incidence within the estimated range of errors . ROC analysis suggested that our model performed with sensitivity ranging from 98–99% during outbreaks in 2004 , 2005 , and 2007 . Estimated ROC areas for the period 2004–2010 indicated that the selected model performed at about 96% ( CI = 93%–98% ) sensitivity in distinguishing between outbreaks and non-outbreaks ( Figure 4: Graph A ) , and in 2011 forecast with 98% ( CI = 95%–100% ) sensitivity in detecting a true outbreak ( Figure 4: Graph B ) . ROC curves as shown in Figure 4 suggest a sensitivity for diagnosing true outbreaks between 90% and 98% during years 2004–2010 corresponding with a 10% to 20% risk of false alarm; whereas , in 2011 the forecasting model showed 100% sensitivity with less than 3% risk of false positive . Overall , the ROC suggested that the selected model performed consistently at above 90% during both model development and forecast periods .
Our model forecasted dengue cases up to 16 weeks ahead using retrospective weekly mean temperature and cumulative rainfall . It showed a consistent ability to separate weeks and years with epidemic and non-epidemic transmission in the training data , as well as outside the training time period in 2011 . Based on lagged weather data and dengue counts the model predicted 5 out of the 5 epidemic weeks in 2011 correctly , using a 16 week lead time , thus , allowing sufficient time to prepare and potentially curb the epidemic . During the forecasting period in 2011 , forecast precision based on prediction error ( SRMSE ) and sensitivity ( ROC ) tests suggested that the model forecast cases with high sensitivity for detecting outbreaks with a low risk of false alarms . The tests results during both training and forecast periods showed small discrepancy in SRMSE with absence of over fitting; thus demonstrating the stability of the model since the forecast in 2011 was performed without using actual reported cases as autoregression . In recent years , the ability to predict local and regional weather in terms of accuracy and lead times has rapidly been improved due to advances in technology . This had allowed a better understanding of the interaction between climate and the temporal-spatial distribution of infectious diseases as well as stimulating research interest on epidemic prediction modeling [36] . We developed the weather-based dengue forecasting model based on scientific evidence that temperature and rainfall has significant influence on vectors and dengue viruses [21] , [22] , [23] , [24] , [35] , [37] , [38] . Dengue cases are influenced by complex interactions of ecology , environment , human , vectors , and virus factors . The lag time between weather and dengue cases could be partly accounted for by the impact of weather conditions on the biological development of the mosquito vector including long egg hatching periods and high possibility of Aedes' eggs to survive waterless for several months [21] , [22] , [23] , [24] , [35] . Several studies have documented relationship between weather variables and dengue cases in Singapore . In the late 90 s , a study that examined the links between dengue cases and Aedes mosquito population as well as weather conditions in Singapore shows that escalating temperature precedes rising dengue incidence by 8–20 weeks [16] . A recent study on the association between weather variables and dengue cases in Singapore using data from 2000–2007 has suggested that minimum and maximum temperature are strong weather predictors for the increase of dengue cases; whereas , rainfall and relative humidity have little correlation with dengue cases [39] . Using a different approach in study design , Hii et al . ( 2009 ) have quantified the effects of weekly mean temperature and cumulative rainfall on the risks of dengue cases across lag times up to 20 weeks [15] , [27] . In their study they considered lag relationship between weather and dengue cases , impact of previous outbreaks on current number of cases , and influences of non-climatic factors . In addition , they applied smoothing functions to allow non-linear relationship between exposures ( mean temperature and rainfall ) and responses ( risk of dengue cases ) as well as adopted quasi-Poisson to permit over dispersion of data . Their findings show impacts of mean temperature and cumulative rainfall on risks of dengue cases vary according to each unit change in weather predictors in different lag windows ( 1–20 weeks ) . Overall , higher relative risks of dengue cases were identified at lag weeks 9–16 . Evidence that weather is also a driver of dengue epidemics and trends of dengue has recently been confirmed by Descloux et al . ( 2012 ) in a study in New Caledonia [40] . It therefore seemed reasonable to assume that weather would be a precipitating factor in dengue epidemics in Singapore . This study demonstrates that weather variables could be important factors for the development of a simple , precise , and low cost functional dengue early warning . A weather-based dengue early warning system could benefit local vector surveillance and control in several ways . First , an early warning system enhances efforts of dengue control to reduce the size of an outbreak which in turn decreases disease transmission , averts possible mortality , and lowers healthcare burden and operating costs incurred during an outbreak . Second , the use of publicly available weather variables removes the necessity for financial investment in weather-based predictive methods and further allows vector control units to focus their operations on high risk period; thus , maximizing limited vector control resources . Third , reports and study have suggested that local authorities require a maximum 3 months to curb a localized dengue outbreak [7] , [27] . The forecast window of 16 weeks shown in this model offers ample time for local authorities to mitigate a potential outbreak effectively . Finally , high precision and sensitivity of a forecast minimizes the use of resources and prevents unnecessary vector control operations triggered by false alarms . Vector control can be resource and capital intensive; hence , high operating costs and unnecessary psychosocial stress in the population subsequent to false alarms could possibly hamper the decision to adopt a dengue early warning . Thresholds for true or false positive rates could vary according to scale of operational complexity and its consequences . We recommend an economic study on cost-effectiveness analysis to identify thresholds of true and false positive rates of forecast to serve as yardstick for decision making as well as to evaluate the long term benefits of an early warning against operating costs . Nevertheless , a dengue forecasting model faces the challenge of long term sustainability of forecast precision since it assumes that a historic distribution pattern will be repeated in the future; while dengue epidemiology is influenced by a combination of factors which are dynamic and possibly evolving over time . Implementation of a new vector control policy could exert direct impact on the size of the vector population and dengue incidence rate in the locality . These changes are likely to influence the trend and epidemic cycle in the long run . Though changes of dengue distribution in the long term are inevitable due to the dynamics of disease transmission and changes of relevant policy , forecast errors can be minimized by making appropriate adjustment of the model through anticipating 1 ) changes in risk factors and 2 ) changes in related fields that will eventually influence the disease transmission . Therefore , current knowledge of factors influencing dengue distribution patterns can be used to re-calibrate the model in the future to maintain long term forecast precision . A weather-based dengue forecast is often geographically fixed due to variability of local weather conditions . Likewise , the dynamics of dengue disease transmission in a community can be influenced by risk factors unique to that local context . Therefore , a locality based dengue forecast is usually applicable only to a specific study area . Nevertheless , the methodology of a weather-based dengue forecasting model could be extrapolated to other geographical areas . Partly due to an exponential growth of regional travels and trades , the Asia Pacific region has experienced an upsurge of dengue incidence in recent years . This suggests that a dengue endemic nation such as Singapore will no longer be able to curb or eliminate dengue without wider regional efforts . A regional dengue early warning system could signal risk of epidemic to all neighboring countries and help to prevent the regional chain effects of dengue outbreaks and so reduce the burden of dengue disease in neighboring countries . Therefore , a regional dengue forecast using weather anomaly such as El Nino index or sea surface temperature will inevitably complement and enhance the success of both national and regional dengue control . In recent years , local authorities in Singapore heighten alert for the risk of increase in dengue cases as ambient temperature increases . Our study results demonstrate that a weather-based dengue forecasting model could provide more precise information on occurrence , timing , and size of dengue epidemics . A forecast that diagnoses outbreaks accurately and simultaneously gives about a four months window for implementing control measures could be invaluable in making control or even elimination of the cyclical dengue epidemic in Singapore a feasible possibility . We recommend a further study to analyze the possibility of incorporating a weather-based dengue early warning into the national dengue surveillance system . Further studies to improve long term sustainability of forecast precision will help to maintain the performance of a forecasting model . Moreover , a research to transform the forecasting model into a user-friendly or non-technical operational instrument comprehensible by users without specialist knowledge would encourage widespread adoption of such a dengue early warning system .
|
Without effective drugs or a vaccine , vector control remains the only method of controlling dengue fever outbreaks in Singapore . Based on our previous findings on the effects of weather on dengue cases and optimal timing for issuing dengue early warning in Singapore , the purpose of this study was to develop a dengue forecasting model that would provide early warning of a dengue outbreak several months in advance to allow sufficient time for effective control to be implemented . We constructed a statistical model using weekly mean temperature and rainfall . This involved 1 ) identifying the optimal lag period for forecasting dengue cases; 2 ) developing the model that described past dengue distribution patterns; 3 ) performing sensitivity tests to analyze whether the selected model could detect actual outbreaks . Finally , we used the selected model to forecast dengue cases from 2011–2012 week16 using weather data alone . Our model forecasted for a period of 16 weeks with high sensitivity in distinguishing between an outbreak and a non-outbreak . We conclude that weather can be an important factor for providing early warning of dengue epidemics , long term sustainability of forecast precision is challenging considering the complex dynamics of disease transmission .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"infectious",
"diseases",
"public",
"health",
"and",
"epidemiology",
"epidemiology",
"global",
"health",
"public",
"health"
] |
2012
|
Forecast of Dengue Incidence Using Temperature and Rainfall
|
The World Health Organization currently recommends combined streptomycin and rifampicin antibiotic treatment as first-line therapy for Mycobacterium ulcerans infections . Alternatives are needed when these are not tolerated or accepted by patients , contraindicated , or neither accessible nor affordable . Despite in vitro effectiveness , clinical evidence for fluoroquinolone antibiotic use against Mycobacterium ulcerans is lacking . We describe outcomes and tolerability of fluoroquinolone-containing antibiotic regimens for Mycobacterium ulcerans in south-eastern Australia . Analysis was performed of prospectively collected data including all primary Mycobacterium ulcerans infections treated at Barwon Health between 1998 and 2010 . Medical treatment involved antibiotic use for more than 7 days; surgical treatment involved surgical excision of a lesion . Treatment success was defined as complete lesion healing without recurrence at 12 months follow-up . A complication was defined as an adverse event attributed to an antibiotic that required its cessation . A total of 133 patients with 137 lesions were studied . Median age was 62 years ( range 3–94 years ) . 47 ( 34% ) had surgical treatment alone , and 90 ( 66% ) had combined surgical and medical treatment . Rifampicin and ciprofloxacin comprised 61% and rifampicin and clarithromycin 23% of first-line antibiotic regimens . 13/47 ( 30% ) treated with surgery alone failed treatment compared to 0/90 ( 0% ) of those treated with combination medical and surgical treatment ( p<0 . 0001 ) . There was no difference in treatment success rate for antibiotic combinations containing a fluoroquinolone ( 61/61 cases; 100% ) compared with those not containing a fluoroquinolone ( 29/29 cases; 100% ) . Complication rates were similar between ciprofloxacin and rifampicin ( 31% ) and rifampicin and clarithromycin ( 33% ) regimens ( OR 0 . 89 , 95% CI 0 . 27–2 . 99 ) . Paradoxical reactions during treatment were observed in 8 ( 9% ) of antibiotic treated cases . Antibiotics combined with surgery may significantly increase treatment success for Mycobacterium ulcerans infections , and fluoroquinolone combined with rifampicin-containing antibiotic regimens can provide an effective and safe oral treatment option .
In recent years the treatment of Mycobacterium ulcerans ( M . ulcerans ) has radically changed , evolving from a predominantly surgically [1] , [2] to a predominantly medically treated disease [3] . This resulted from clinical experience supported by scientific data showing superior outcomes when antibiotics were used alone [4] , [5] , or combined with surgery [6] , [7] . It is also supported by in vitro data [8]–[10] of antibiotic effectiveness against M . ulcerans . The World Health Organization now recommends combined streptomycin and rifampicin antibiotic treatment as first-line therapy for M . ulcerans infections , with surgery reserved mainly to remove necrotic tissue , cover skin defects , and correct deformities [11] . The Bellarine Peninsula in south-eastern Australia has been experiencing an epidemic of M . ulcerans since 1998 . It affects local residents , but also visitors from outside the endemic region , with cases in those living locally managed at the local referral health service , Barwon Health . In our region , despite recommendations at the time against their use , adjunctive antibiotic treatment of M . ulcerans was initiated from 1998 in response to severe disease causing significant morbidity and requiring reconstructive surgery , and disease recurrences despite surgery . Fluoroquinolones ( FQ ) were introduced into antibiotic regimens in 2003 [6] , [12] in response to perceived treatment failures and excess toxicity with other antibiotics , as well as their potential advantages in treating M . ulcerans; documented in vitro evidence of activity [9] , [10] , [13]–[15] , good bioavailability [16] , and excellent bone and tissue penetration [17] . Since their introduction they have commonly been employed in the antibiotic regimens used at Barwon Health . FQ antibiotics offer the possibility of completely oral antibiotic regimens when combined with another active oral antibiotic , usually rifampicin . This can be especially useful where other recommended antibiotics are not tolerated , accepted , accessible , or affordable . However , although there is evidence of good activity against M . ulcerans in the laboratory [9]–[10] , [13] , in mouse footpad models [8] , [14] , [18] , and small numbers of clinical cases [6] , [12] , [19]–[21] , clinical evidence of FQ efficacy is lacking . Therefore we undertook a study during the current epidemic in the Bellarine Peninsula to describe the use of FQ antibiotics in M . ulcerans treatment and to compare their outcomes and tolerability with other antibiotics used .
Analysis was performed using prospectively collected data from an electronic database containing information on all cases of M . ulcerans infection treated at Barwon Health between 1st March 1998 and 31st May 2010 . Data collected includes epidemiological details , diagnostic tests , clinical features , treatment , and outcomes . Only first lesions on initial presentation were analyzed to avoid potential confounders when analyzing recurrent cases . Data was analyzed anonymously . A case of M . ulcerans was defined as the presence of a lesion clinically suggestive of M . ulcerans plus any of ( 1 ) a culture of M . ulcerans from the lesion , ( 2 ) a positive PCR from a swab or biopsy of the lesion [22] , or ( 3 ) histopathology of an excised lesion showing a necrotic granulomatous ulcer with the presence of acid-fast bacilli consistent with acute M . ulcerans [23] . Surgical treatment was defined as the surgical excision of a lesion . Due to the paucity of cases managed without surgery , only those undergoing surgery were included . Major surgery involved the use of a split thickness skin graft or a vascularized tissue flap to close the tissue defect , whereas minor surgery involved excision plus primary closure of the defect . A positive margin was defined on histology as granulomatous inflammation or necrotic tissue extending to the margins of the excised lesion . Medical treatment was defined as the use of antibiotics for more than 7 days , and first-line regimens were the initial antibiotic regimens used . A complication was an adverse event attributed to an antibiotic that required the cessation of that medication . Drug dosages for adults included ciprofloxacin 500 mg twice daily , moxifloxacin 400 mg daily , rifampicin 10 mg/kg/day ( up to a maximum of 600 mg daily ) , clarithromycin 500 mg twice daily , and amikacin 15 mg/kg/day . Treatment success was defined as complete healing of the lesion without recurrence 12 months after treatment commencement . Treatment failure was defined as those who had a recurrence with in at least 12 months of follow-up . Recurrence was defined as a new lesion appearing either in the wound , locally , or in another part of the body after the surgical excision of the initial lesion that met the case definition for M . ulcerans . Paradoxical reactions were not considered a recurrence and were defined as: initial clinical improvement followed by the clinical deterioration of the treated lesion , or the appearance of a new lesion , either locally or in a distant body site , that on histopathology showed evidence of an intense immunological reaction consistent with an immune-mediated paradoxical reaction [19] . There was no standardized treatment protocol for M . ulcerans followed in Barwon Health during the study period . The role of surgery and the use of antibiotics were determined by individual specialist practitioners involved in M . ulcerans treatment . Patients were followed up according to routine clinical practice and observed antibiotic complications recorded in clinical notes when they occurred . This study was approved by the Barwon Health Research and Ethics Committee . Data were collected and analysed with Epi-Info 6 ( Centers for Disease Control , Atlanta ) . Statistical significance was determined using the 2-tailed Fisher exact test for 2×2 tables for each of the categorical values . A non-parametric cumulative failure graph using the Kaplan-Meier method and the endpoint of antibiotic cessation was plotted using the statistics package Minitab ( version 15 ) to model the proportion of antibiotics ceased over time due to complications .
One hundred and forty seven patients with M . ulcerans were diagnosed and treated at Barwon Health over the study period 1st March 1998 to 31st May 2010 . Fourteen were excluded from further analysis: 1 had no surgery , 2 died before the completion of follow-up ( 1 from a cerebrovascular accident 52 days and 1 of sepsis secondary to the M . ulcerans lesion 5 days post treatment commencement ) , 1 was lost to follow-up 85 days post treatment commencement , and for 10 treatment and follow-up was ongoing . Therefore a total of 133 patients with 137 lesions ( 4 patients had 2 lesions at initial presentation ) were included in the analysis . Median age of patients was 62 years ( range 3–94 years ) ; 7 ( 5% ) were <15 years . Sixty-seven ( 50% ) were male . Associated co-morbidities included diabetes mellitus ( 11 ) , malignancy ( 5 ) , connective tissue disease ( 4 ) , and immunosuppressive treatment ( 4 ) . For 122 cases where the clinical type of lesion was recorded , 106 ( 87% ) were ulcers , 9 ( 7% ) were nodules , and 7 ( 6% ) were oedematous lesions . Diagnosis was made by PCR in 116 ( 87% ) , positive culture in 22 ( 17% ) , and consistent histopathology in 54 ( 41% ) cases . Eighteen of 24 ( 75% ) PCR positive cases where culture was performed were culture positive , but no cases were culture positive and PCR negative . One case was PCR negative but positive on histopathology . Forty-seven ( 34% ) lesions were treated with surgical excision alone , and 90 ( 66% ) had surgical excision and adjunctive antibiotic therapy . To close the skin defect after excision of the lesion , 65 ( 47% ) required a split thickness skin graft and 16 ( 12% ) required a vascularized tissue flap . The proportion of cases receiving antibiotics significantly increased pre 2005 compared to post 2005 , rising from 45% to 74% [<2005 18/40 , ≥2005 72/97 , OR 3 . 52 ( 1 . 52–8 . 20 ) ] ( Figure 1 ) . The most common initial antibiotic regimens were rifampicin and ciprofloxacin ( 61% ) and rifampicin and clarithromycin ( 23% ) . Four patients received ciprofloxacin and clarithromycin , and 2 patients received rifampicin and moxifloxacin regimens ( Table 1 ) . FQ antibiotics were used in 3 of 5 children aged <15 years who received antibiotics . Antibiotics were given for a median duration of 76 days ( range 12–155 days ) ; 21 ( 23% ) between 12–30 days , 18 ( 20% ) 31–60 days , 30 ( 33% ) 61–90 days , 14 ( 16% ) 91–120 days , and 7 ( 8% ) 121–155 days . FQ antibiotics were not used in this study until 2004 , but from then 82% of regimens were FQ containing . Fourteen of 47 ( 30% ) of those treated with surgery alone failed treatment compared to 0/90 ( 0% ) of those treated with a combination of medical and surgical treatment ( p<0 . 0001 ) . The risk of treatment failure increased significantly with no antibiotics compared to those treated with antibiotics for major surgery ( p<0 . 0001 ) , minor surgery ( p = 0 . 01 ) , positive margins ( p<0 . 0001 ) , and negative margins ( p = 0 . 05 ) ( Table 2 ) . If minor surgery and negative margins were present , 1/22 ( 5% ) failed treatment . If only regimens containing an FQ ( n = 64 ) are compared with surgery alone , the risk of treatment failure remained significantly increased when no antibiotics were used overall ( p<0 . 0001 ) and for major surgery ( p<0 . 0001 ) and positive margins ( p<0 . 0001 ) ( Table 3 ) . There was no difference in treatment success rate for antibiotic combinations containing an FQ ( 61/61 cases; 100% ) compared with those not containing an FQ ( 29/29 cases; 100% ) . Treatment success rates with antibiotics were also similar pre-2004 when no FQs were used ( 11/11; 100% ) compared with post 2004 ( 79/79; 100% ) . All four cases treated with ciprofloxacin and clarithromycin combined with surgery experienced treatment success . For those failing treatment , the recurrences occurred a median 90 days post surgery ( range 14–300 days ) . In 9 ( 64% ) patients this was local and in 6 ( 43% ) patients it was distant ( 1 had both ) . Paradoxical reactions occurred in 8/90 ( 9% ) of cases given antibiotics after a median duration of 48 days ( range 14–85 days ) . Fifty-eight ( 64% ) patients had antibiotics prior to surgery for a median duration of 8 days ( range 1–36 days ) . Of these , mycobacterial cultures were performed on 28 excised specimens . Cultures were positive for M . ulcerans in 11/20 ( 55% ) of those who received ≤14 days of antibiotics prior to surgery , and in 1/8 ( 12 . 5% ) of those who received >14 days of antibiotics prior to surgery ( Table 4 ) . All cases with positive cultures were associated with successful outcomes after a median antibiotic duration of 87 days ( range 30–155 days ) . Rifampicin was associated with a complication in 19/85 ( 22% ) cases occurring at a median 27 days ( range 6–94 days ) and involved gastrointestinal intolerance in 15 , hepatitis 4 , rash 3 , and hypoglycemia in 1 case . Ciprofloxacin was associated with a complication in 13/63 ( 21% ) cases occurring at a median 24 days ( range 6–90 days ) and involved gastrointestinal intolerance in 10 , joint or tendon effects in 3 , rash in 2 , and hallucinations in 1 case . Clarithromycin was associated with a complication in 10/38 ( 26% ) cases occurring at a median 25 days ( range 2–60 days ) and involved gastrointestinal intolerance in 9 , hepatitis in 1 , and palpitations in 1 case . By 120 days on treatment , the proportion of cases in which rifampicin [28 . 6% ( 95% CI 16 . 0 , 41 . 2 ) ] , clarithromycin [29 . 4% ( 95% CI 13 . 8 , 45 . 0 ) ] , and ciprofloxacin [24 . 9% ( 95% CI 12 . 3 , 37 . 5 ) ] were ceased were similar ( Figure 2 ) , and complication rates were similar between ciprofloxacin and rifampicin 17/55 ( 31% ) and rifampicin and clarithromycin 7/21 ( 33% ) regimens ( OR 0 . 89 , 95% CI 0 . 27–2 . 99 ) .
Our study demonstrates that , combined with surgical excision of M . ulcerans lesions , antibiotics appear highly effective at preventing disease recurrences; we describe a reduction of recurrence rates from more than one quarter of cases with surgery alone to none if antibiotics are used . This includes 64 cases treated with FQ-containing regimens ( the majority involving ciprofloxacin ) which show similar efficacy to non-FQ containing regimens . Recent studies have provided strong evidence that non-FQ-containing antibiotic regimens were effective in M . ulcerans treatment; 8-week combinations of rifampicin and streptomycin cured 96% of cases in a randomized controlled trial in Africa [5] , and 8 weeks of rifampicin and clarithromycin were 100% effective in an uncontrolled trial in Benin [4] . Although we have previously published small numbers of M . ulcerans cases treated with FQ-containing regimens [6] , [12] , [19] , this is the first study large enough to provide significant evidence of the clinical effectiveness of FQ antibiotics combined mainly with rifampicin in M . ulcerans treatment . Ciprofloxacin has been shown to have good in vitro activity against M . ulcerans with minimal inhibitory concentrations ( MIC ) of between 0 . 5 and 1 mg/l in two published studies [9] , [15]; in the same studies these MICs compared favorably against the MICs of currently recommended antibiotics ( rifampicin 1–2 mg/l , amikacin 1 mg/l , clarithromycin 1 mg/l ) . Ciprofloxacin has also been shown to have rapid bactericidal activity in humans against M . tuberculosis [24]–[26] , and has been used to successfully treat other non-tuberculous mycobacterial infections including M . marinum [27] , the species most closely related to M . ulcerans . Moxifloxacin similarly has good in vitro activity ( MIC 0 . 25 ) [15] . In the mouse footpad model , moxifloxacin has bactericidal activity against M . ulcerans and , when combined with rifampicin , is as effective as combinations of rifampicin/streptomycin , rifampicin/amikacin and rifampicin/clarithromycin [8] , [18] . Furthermore , other factors that favor the use of FQs include their high oral bioavailability ( 78% ) [16] and excellent bone and tissue penetration [17] . Nevertheless , we would caution that fluoroquinolones should not be used as monotherapy for M . ulcerans treatment as there is the potential for the development of resistance , as has been shown to occur when FQs are used as monotherapy for M . tuberculosis [25] . Moxifloxacin may be favoured over ciprofloxacin due to slightly better published MICs against M . ulcerans , the evidence from the mouse footpad models of which there is no similar published data for ciprofloxacin , greater potential barrier to resistance if effects are similar to that in M . tuberculosis [25] , and once-daily administration . A constraint at present is its significant increased cost compared to ciprofloxacin , with a cost of $671 compared with $32 Australian dollars for an 8-week treatment course at our institution . In mouse footpad models , rifampicin has the greatest bactericidal activity against M . ulcerans [8] , [18] , and thus is assumed to be the most active and important antibiotic in vivo , though there are no human studies of rifampicin monotherapy to confirm this . It is possible that the second agent , including the FQs , act only as bacteriostatic agents preventing the emergence of rifampicin resistance . Therefore we recommend rifampicin as the first antibiotic choice , to be combined with another agent . Our data indicate that FQs are an appropriate combination choice , especially in cases where other antibiotics such as clarithromycin or streptomycin are not tolerated , accepted by patients , or accessible , or are contraindicated . FQs , greatest advantage may be their potential to be combined with other oral antibiotics such as rifampicin to provide completely orally administered regimens in endemic settings . This allows outpatient care to be provided close to patient homes and avoids daily intramuscular injections , potentially increasing patient willingness to present early for diagnosis and increase adherence to treatment . Furthermore , ciprofloxacin is generically produced , reducing its cost , and is readily available in many resource-limited settings . Antibiotics were most commonly given for between 1 and 3 months ( 53% ) in our cohort . A proportion of cases ( 21; 23% ) had successful outcomes with less than 30 days of treatment , although 55% of cases given less than 2 weeks of antibiotics remained culture positive . One case remained positive after 36 days of antibiotics but still achieved cure , reinforcing findings from previous studies that a small proportion of cases may remain culture positive after 1–2 months of antibiotics but still achieve cure with at least 8 weeks of antibiotics [4] , [5] . The proportion of patients receiving antibiotic treatment significantly increased from 2005 as it became apparent in patients treated at Barwon Health that antibiotics were associated with a reduction in disease recurrences and permitted more conservative surgery to be performed [6] . The efficacy of treatment is also determined by the tolerance of the antibiotic regimens . In our study there were no significant differences in the tolerability of the 3 main oral antibiotic choices of rifampicin , clarithromycin , and ciprofloxacin . In addition , the complications mainly involved gastrointestinal intolerance with no significant sequelae . This differs from the significant incidence of serious adverse events previously described for alternative antibiotics such as streptomycin [5] , [28] and amikacin [6] . It is important to note that the complication rates were higher in our study compared to studies from African populations [4] , [5] , [7] . This is likely due to the older patient population in our cohort where these medications are less well tolerated , especially from a gastro-intestinal viewpoint , and there is more potential for drug interactions . Nevertheless it underlines the importance of having a number of oral antibiotic combinations available if first-line antimicrobials need to be substituted on account of intolerance . Finally , paradoxical reactions occurred in 9% of antibiotic treated cases . To our knowledge this is the first published rate of paradoxical reactions in a cohort of patients treated for M . ulcerans . In our study the reactions occurred as early as 2 weeks and as late as 3 months after antibiotics were commenced . Recently , paradoxical reactions occurring after the cessation of antibiotics have been described [29] . It is important that paradoxical reactions are considered and recognized during treatment , and distinguished from treatment failures , to avoid unnecessary antibiotic regimens changes or further surgery [19] . There are a number of limitations to our study . First , it is an observational cohort and thus there is the potential for unknown confounders to have affected the results . Despite this possibility , the treatment outcome results are highly significant . Second , some of the infections acquired from the local endemic region occurred in visitors to the region and were not managed by our health service ( Barwon Health ) but in the health services where they live . Although this may have introduced a selection bias , we feel it is unlikely that this would have changed the findings of our study . Third , all cases underwent surgical excision and therefore the outcomes may not be applicable to cases where antibiotics alone are used . We advocate for further randomized studies using FQ-containing antibiotic regimens without curative surgery be performed to provide further information . In summary , antibiotics in combination with surgery may significantly increase treatment success for M . ulcerans infections . In addition , antibiotic regimens containing oral FQs combined with rifampicin can provide an effective and safe treatment option and should be further studied in the treatment of M . ulcerans .
|
Buruli ulcer is a necrotizing infection of skin and subcutaneous tissue caused by Mycobacterium ulcerans and is the third most common mycobacterial disease worldwide ( after tuberculosis and leprosy ) . In recent years its treatment has radically changed , evolving from a predominantly surgically to a predominantly medically treated disease . The World Health Organization now recommends combined streptomycin and rifampicin antibiotic treatment as first-line therapy for Mycobacterium ulcerans infections . However , alternatives are needed where recommended antibiotics are not tolerated or accepted by patients , contraindicated , or not accessible nor affordable . This study describes the use of antibiotics , including oral fluoroquinolones , in the treatment of Mycobacterium ulcerans in south-eastern Australia . It demonstrates that antibiotics combined with surgery are highly effective in the treatment of Mycobacterium ulcerans . In addition , oral fluoroquinolone-containing antibiotic combinations are shown to be as effective and well tolerated as other recommended antibiotic combinations . Fluoroquinolone antibiotics therefore offer the potential to provide an alternative oral antibiotic to be combined with rifampicin for Mycobacterium ulcerans treatment , allowing more accessible and acceptable , less toxic , and less expensive treatment regimens to be available , especially in resource-limited settings where the disease burden is greatest .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"infectious",
"diseases",
"neglected",
"tropical",
"diseases"
] |
2012
|
Successful Outcomes with Oral Fluoroquinolones Combined with Rifampicin in the Treatment of Mycobacterium ulcerans: An Observational Cohort Study
|
H9N2 subtype influenza viruses have been detected in different species of wild birds and domestic poultry in many countries for several decades . Because these viruses are of low pathogenicity in poultry , their eradication is not a priority for animal disease control in many countries , which has allowed them to continue to evolve and spread . Here , we characterized the genetic variation , receptor-binding specificity , replication capability , and transmission in mammals of a series of H9N2 influenza viruses that were detected in live poultry markets in southern China between 2009 and 2013 . Thirty-five viruses represented 17 genotypes on the basis of genomic diversity , and one specific “internal-gene-combination” predominated among the H9N2 viruses . This gene combination was also present in the H7N9 and H10N8 viruses that have infected humans in China . All of the 35 viruses preferentially bound to the human-like receptor , although two also retained the ability to bind to the avian-like receptor . Six of nine viruses tested were transmissible in ferrets by respiratory droplet; two were highly transmissible . Some H9N2 viruses readily acquired the 627K or 701N mutation in their PB2 gene upon infection of ferrets , further enhancing their virulence and transmission in mammals . Our study indicates that the widespread dissemination of H9N2 viruses poses a threat to human health not only because of the potential of these viruses to cause an influenza pandemic , but also because they can function as “vehicles” to deliver different subtypes of influenza viruses from avian species to humans .
Avian influenza viruses of several subtypes continue to present challenges to human health . The H5N1 influenza viruses have caused 380 fatal cases among the 641 documented human infections in 16 countries [1] , and several studies have documented the transmission potential of H5N1 mutants or reassortants [2]–[4] . Eighty-seven human cases of H7N7 influenza virus infection were confirmed in the Netherlands in 2003 , one of which was fatal [5] , [6] . H6 viruses can infect and cause illness in mice and ferrets , and are transmissible in guinea pigs [7]–[9]; an H6N1 virus was isolated from a human with influenza-like symptoms in Taiwan in 2013 [10] . An approximately 30% mortality rate is associated with the 400 human infections with the newly emerged H7N9 viruses in China reported by the end of March , 2014 [11] . H10N8 virus caused three human infections in China in 2013 , two of which were fatal [12] . These facts emphasize that it is not only the highly pathogenic H5N1 and H7N7 influenza viruses that pose a severe threat to human health , but also that the nonlethal influenza viruses circulating in avian species can cause disease and even death in humans . During the last several decades , H9N2 influenza viruses have been isolated worldwide from wild and domestic avian species [13] , [14] . These viruses have also been detected in pigs [15]–[17] . Many studies have been performed to evaluate the pandemic potential of the H9N2 influenza viruses . The viruses have been shown to replicate in mice without pre-adaptation [18]–[21] , and some strains from poultry in Asia have human virus-like receptor specificity [22] . Sorrell et al . reported that following adaptation in the ferret , a reassortant carrying the surface proteins of an avian H9N2 in a human H3N2 backbone could transmit efficiently via respiratory droplet [23] . Other studies have reported that H9N2 reassortants bearing genes from the 2009 H1N1 pandemic virus exhibited increased virulence in mice [24] or transmissibility in ferrets [25] . Wan et al . found that two of five wild-type H9N2 viruses isolated from different avian species between 1988 and 2003 transmitted to direct contact ferrets [26] . However , none of the naturally isolated H9N2 viruses has been reported to transmit to ferrets via respiratory droplet . Although the H9N2 viruses have been detected in chickens and ducks in many provinces in China since 1993 [18] , [27] , their low pathogenic nature to poultry has made them a low priority for animal disease control . However , the H9N2 viruses caused human infections in China in 1999 , 2003 , and 2013 [28]–[30] , and some poultry workers in China , India , Cambodia , Romania , America , Nigeria , and Vietnam were reportedly serologically positive for H9N2 viruses [31]–[38] , implying a substantial threat to public health . Recent studies indicated that the H9N2 viruses contributed the six internal genes to the newly emerged H7N9 virus in southern China and to the H10N8 virus that caused three human infections in Jiangxi province , China [12] , [39] , [40] . These facts prompted us to assess the biologic properties and pandemic potential of H9N2 influenza viruses circulating in poultry .
To investigate the genetic relationship of the viruses from different times and places , we sequenced the genomes of 35 viruses that were collected from 2009 to 2013 from 12 provinces in southern China ( Figure S1 ) . The amino acid motif at the cleavage site of the hemagglutinin ( HA ) of these isolates is –RSSR- , which is a characteristic of viruses of low pathogenicity in chickens . The HA gene of the 35 viruses shared 87 . 8%–99 . 7% identity at the nucleotide level , and they formed five phylogenetic groups ( Figure 1A ) . The neuraminidase ( NA ) genes of these viruses shared 85 . 4%–99 . 6% identity at the nucleotide level and formed four phylogenetic groups ( Figure S2A ) . The 31 viruses in groups 1 , 2 and 3 have a 3-amino acid deletion in the NA stalk ( residues 62–64 ) , whereas the four isolates in group 4 have no such deletion . Several amino acid changes related to the increased replication or virulence of avian influenza viruses in mammals [41] were detected in these H9N2 viruses ( Table S1 ) . The amino acid changes R207K , H436Y , and M677T in basic polymerase 1 ( PB1 ) [42] , [43] , A515T in acidic polymerase ( PA ) [42] , N30D and T215A in matrix protein ( M1 ) [44] , and P42S in nonstructural protein 1 ( NS1 ) were conserved in all strains [45] , whereas the virulence-related mutation I368V in PB1[3] , T139A in M1[46] , and the amantadine and rimantadine resistance-conferring mutation S31N/G in M2 were detected in some of the strains ( Table S1 ) [47] . Two amino acid changes in basic polymerase 2 ( PB2 ) , glutamic acid to lysine at position 627 ( E627K ) and aspartic acid to asparagine at position 701 ( D701N ) , are important for the virulence and transmission of H5N1 viruses in mammals [48]–[50] , and are also frequently presented in the H7N9 viruses isolated from humans [40] , [51] . A detailed comparison of the amino acid differences among the H9N2 viruses showed that all 35 of the H9N2 viruses have the amino acid combination of 627E/701D in their PB2 . The six internal genes of the H9N2 viruses showed distinct diversity , with PB2 , PB1 , PA , nucleoprotein ( NP ) , M , NS genes of the 35 viruses sharing 84 . 8%–99 . 4% , 86 . 4%–99 . 6% , 87 . 9%–99 . 5% , 92 . 5%–99 . 6% , 94 . 1%–99 . 9% , and 91 . 3%–99 . 6% identity , respectively , at the nucleotide level . The PB2 genes formed six groups in their polygenic trees ( Figure S2B ) , and the PB1 genes formed five groups in their polygenic trees ( Figure S2C ) . The PA and M genes each formed four groups in their polygenic trees ( Figure S2D and Figure S2F ) , whereas the NP and NS genes each formed two groups in their polygenic trees ( Figure S2E and Figure S2G ) . On the basis of this genomic diversity , the viruses examined in this study were divided into 17 genotypes ( Figure 1B ) . Of note , 19 viruses in genotypes 1 , 2 , 3 , 4 , and 16 have a similar combination of their six internal genes ( the DK/ZJ/C1036/09-like combination ) . Receptor-binding preference has important implications for influenza virus replication and transmission [3] , [4] . The change of receptor-binding preference from α-2 , 3-linked sialic acids ( Sias ) ( avian-type receptors ) to α-2 , 6-linked Sias ( human-type receptors ) is thought to be a prerequisite for an avian influenza virus to transmit from human to human . By using a solid-phase binding assay as described previously [4] , [7] , we tested the receptor-binding specificity of 41 H9N2 viruses , the 35 viruses described above and six "early" viruses that were isolated from poultry in China between 1996 and 2001 [18] ( Figure 2 , Figure S3 ) , to two different glycopolymers: the α-2 , 3-siaylglycopolymer [Neu5Acα2-3Galβ1-4GlcNAcβ1-pAP ( para-aminophenyl ) -alpha-polyglutamic acid ( α-PGA ) ] and the α-2 , 6-sialylglycopolymer [Neu5Acα2-6Galβ1-4GlcNAcβ1-pAP ( para-aminophenyl ) -alpha-polyglutamic acid ( α-PGA ) ] . All 41 viruses were able to bind to the α-2 , 6-siaylglycopolymer , although eight , including the six "early" viruses , also bound to the α-2 , 3-siaylglycopolymer with moderate to high affinity ( Figure 2 , Figure S3 , and Table S2 ) . These results indicate that H9N2 viruses isolated naturally from poultry have acquired the ability to preferentially bind to the human-type receptor , similar to the widely circulating human influenza viruses . The amino acid change Q226L is reported to contribute to the human-type receptor binding of H9N2 virus [26] , but this mutation was not detected in the six "early" isolates in our study ( Table S1 ) , which were able to bind to the α-2 , 6-siaylglycopolymer ( Figure S3 ) , suggesting that some other amino acid ( s ) may contribute to this phenotype . As shown in Table S1 , three more amino acid changes ( I155T , H183N , and A190V ) that have been reported to affect the receptor binding preference of other subtypes of influenza viruses were also detected in these H9N2 viruses . Since the amino acid at position 190 of HA was not conserved , and this amino acid has not been linked to the receptor-binding phenotype observed , we investigated the contributions of only I155T and H183N to the human-type receptor binding of H9N2 virus . By using plasmid-based reverse genetics , we constructed a reassortant virus containing the HA and NA genes of the "early" H9N2 isolate A/chicken/Guangxi/9/99 ( CK/GX/9/99 ) and the six internal genes of the A/Puerto Rico/8/1934 ( H1N1 ) ( PR8 ) virus and designated it as rCK/GX/9/99 . Then , we introduced the avian influenza virus-like amino acids I and H into the HA gene at positions 155 and 183 , respectively , to create the mutants we designated as rCK/GX/9/99-HAT155I and rCK/GX/9/99-HAN183H , respectively . Receptor binding analysis indicated that , similar to the wild-type CK/GX/9/99 virus ( Figure S3 ) , the rCK/GX/9/99 and rCK/GX/9/99-HAN183H viruses bound to both the α-2 , 3-siaylglycopolymer and α-2 , 6-siaylglycopolymer ( Figure 2J and L ) ; however , the rCK/GX/9/99-HAT155I variant only maintained the ability to bind to the α-2 , 3-siaylglycopolymer and lost its ability to bind to the α-2 , 6-siaylglycopolymer ( Figure 2K ) . These results indicate that , in addition to the Q226L mutation , the I155T mutation in HA also plays an important role in the binding of H9N2 virus to the human-type receptor . We selected 26 H9N2 influenza viruses , to include one virus from each genotype from each year , and evaluated their replication and virulence in BALB/c mice . All 26 viruses replicated in lungs of mice , with titers ranging from 1 . 8 to 6 . 8log10TCID50 , 22 viruses were also detected in the nasal turbinates of mice , with titers ranging from 0 . 7 to 5 . 3log10TCID50 ( Figure 3 ) . Virus was not detected in the spleen , kidneys , or brain of any mice . Mice infected with these viruses showed diverse body weight changes during the observation period: fourteen viruses caused 1 . 5% to 17 . 5% body weight loss in mice , whereas mice gained body weight despite inoculation with the other twelve viruses ( Figure 3 ) . All mice survived during the observation period . These results indicate that , unlike the H7N9 viruses isolated from poultry , which did not cause any disease in mice [40] , the H9N2 viruses isolated from poultry show a range of virulence in mice . As shown in Figure 1B and Table S3 , the viruses of genotypes 1–4 , 13 , and 16 were detected from multiple provinces and in different years , and we therefore selected one or two viruses from these genotypes and tested their replication and transmission in ferrets . However , the viruses of genotypes 5–12 , 14 , 15 , and 17 were only isolated from individual provinces; therefore , we speculated that those strains were unlikely to be widespread and , accordingly , only one strain from genotype 6 was selected and tested in ferrets ( Table S3 ) . Two ferrets were inoculated i . n . with 106 EID50 of each virus , and the nasal turbinates , tonsils , trachea , various lung lobes , brain , spleen , kidneys , and liver from each ferret were collected on day 4 p . i . for virus titration in MDCK cells . All of the nine viruses replicated in the nasal turbinates of ferrets , with titers ranging from 2 . 8–7 . 3log10TCID50 ( Figure 4 ) . Virus replication in trachea was detected in ferrets inoculated by eight viruses , but not in the CK/ZJ/C1219/10 virus-infected ferrets ( Figure 4E ) . Virus was detected in all lobes of lungs of ferrets inoculated with CK/ZJ/SC324/13 and CK/HuB/C4196/09 ( Figure 4D and H ) , but was not detectable in some of lobes of the lungs of ferrets inoculated with the other seven viruses ( Figure 4 ) . Virus was detected in the spleen of one ferret inoculated with CK/JS/C4258/12 and two ferrets inoculated with CK/CQ/C1258/11 ( Figure 4F and G ) , but was not detected in the brain , kidney , or liver of any ferret . Pathological studies were performed on lung samples from the virus-infected ferrets . Most of the lungs showed mild damage after infection with CK/ZJ/C1219/10 or DK/ZJ/C2046/12 ( Figure 5A , Figure S4A ) . By contrast , the lungs of the other seven virus-infected ferrets showed severe bronchopneumonia ( Figure 5 B and C , and Figure S4B–F ) . To investigate respiratory droplet transmission , we inoculated three ferrets i . n . with 106 . 0 EID50 of test virus and then housed them separately in solid stainless-steel cages within an isolator . Twenty-four hours later , three naïve ferrets were placed in adjacent cages . Each pair of animals was separated by a double-layered net divider as described previously [2] , [40] . Nasal washes were collected every 2 days from all of the animals beginning 2 days p . i . [1 day post-exposure ( p . e . ) ] for the detection of virus shedding . Sera were collected from all animals on day 21 p . i . for hemagglutinin inhibition ( HI ) antibody detection . Respiratory droplet transmission was confirmed when virus was detected in the nasal washes or by seroconversion of the naïve exposed animals at the end of the 3-week observation period . Virus was detected in all of the directly infected animals ( Figure 6A–L ) . However , virus was not detected in any of the animals exposed to the CK/CQ/C1258/11- , CK/HuB/C4196/09- , and CK/HuN/C4136/10-inoculated ferrets ( Figure 6G , H , and I ) . Virus was detected in one ferret exposed to the ferrets that had been inoculated with the CK/SH/SC197/13 , DK/JS/C2046/12 , CK/ZJ/SC324/13 , and CK/ZJ/C1219/10 viruses ( Figure 6B , C , D , and E ) . Virus was detected in all three ferrets exposed to the ferrets that had been inoculated with CK/GX/C1435/12 and CK/JS/C4258/12 ( Figure 6A and F ) . Because the efficient transmission of naturally isolated H9N2 influenza viruses has never been reported before , we repeated this respiratory droplet transmission study with the CK/JS/C4258/12 virus in ferrets and found the results to be reproducible ( Figure 6J ) . The ferrets that were inoculated with these nine viruses experienced a 1 . 4% to 7 . 9% body weight loss , and the body weight loss of the exposed ferrets was up to 8 . 8% ( Table 1 and Table S4 ) . Seroconversion occurred in all of the virus-inoculated animals and in all exposed animals that were virus-positive ( Table 1 ) . These results indicate that six of the nine H9N2 viruses tested can transmit between ferrets , and two of them transmit highly efficiently via respiratory droplet . Previous studies showed that the H7N9 viruses easily acquire the 627K or 701N mutations in PB2 during their replication in humans [40] , [52] , [53] . To investigate whether the H9N2 viruses similarly acquire such mutations during replication in ferrets , we sequenced the PB2 gene of ten randomly selected clones of each sample recovered at different time points from each ferret . The 627K or 701N mutations in PB2 were detected from the samples recovered from the ferrets that were inoculated or exposed to CK/GX/C1435/12 , CK/ZJ/C1219/10 , CK/CQ/C1258/11 , CK/JS/C4258/12 , CK/HuB/C4196/09 , and CK/HuN/C4136/10 , but were not detected in the samples recovered from the ferrets that were inoculated with or exposed to CK/SH/SC197/13 , DK/ZJ/C2046/12 , and CK/ZJ/SC324/13 viruses ( Table 2 ) . To investigate whether these mutations existed in the inoculums , the PB2 gene of all nine viruses tested in ferrets was checked by using a deep sequencing approach; the 627K and 701N mutations in PB2 were not detected in any of the viral stocks ( Table 2 , Table S5 ) . We also deep sequenced the PB2 of nine selected samples that were recovered from the virus-inoculated ferrets , and found that the ratios of the PB2 627K and 701N mutations were comparable to our previous sequencing results presented in Table 2 ( Table S5 ) . To investigate whether the 627K or 701N mutations in PB2 could increase the virulence and transmissibility of the H9N2 viruses in mammals , we plaque-purified two mutants , CK/ZJ/C1219/10-PB2/627K and CK/ZJ/C1219/10-PB2/701N , and tested them in mice and ferrets . The viral titers in the nasal turbinates and lungs of the mutant-infected mice were significantly higher than that of the CK/ZJ/C1219/10 virus-inoculated mice ( Figure 3 ) . The CK/ZJ/C1219/10-PB2/627K virus killed all five mice by day 6 p . i . , whereas the mice inoculated with the CK/ZJ/C1219/10-PB2/701N virus experienced a 9 . 2% body weight loss and survived the infection for the observation period ( Figure 3 ) . Viral titers in the nasal turbinates and lungs of the two mutant-inoculated ferrets were notably higher than that of the CK/ZJ/C1219/10-inoculated ferrets , and virus was also detected in the spleen of one ferret inoculated with the CK/ZJ/C1219/10-PB2/627K virus ( Figure 4 , J and K ) . The lung damage of the two mutant-inoculated ferrets was much more severe than that of the CK/ZJ/C1219/10-inoculated ferrets ( Figure 5 , D and E ) . Both mutants transmitted to two of three ferrets via respiratory droplet ( Figure 6 , K and L , Table 1 ) . The CK/ZJ/C1219/10-PB2/627K-inoculated and -exposed ferrets experienced 6 . 4% and 4 . 5% weight loss , respectively , and the body weight loss of the CK/ZJ/C1219/10-PB2/701N-inoculated and -exposed ferrets was 7 . 2% and 4 . 4% , respectively ( Table 1 , Table S4 ) . These results indicate that the 627K and 701N mutations in PB2 increase the virulence and further promote the transmissibility of H9N2 viruses in mammals .
Receptor-binding preference has important implications for influenza virus replication and transmission [3] , [4] , [54] , [55] . Generally , it is believed that the HA of human infective influenza subtypes preferentially recognizes α-2 , 6- linked Sias , whereas the HA of avian influenza subtypes preferentially recognizes α-2 , 3-linked Sias [3] , [56] . Some naturally isolated avian influenza viruses of the H5 and H6 subtypes have been reported to bind to α-2 , 6-linked Sias [2] , [7] , [50] , but their affinity to the α-2 , 3-linked Sias was much higher than that to α-2 , 6-linked Sias . The newly emerged H7N9 viruses isolated from both avian species and humans bind to α-2 , 6-linked Sias with high affinity and to α-2 , 3-linked Sias with affinity that varies among strains [40] , [57] . However , most of the H9N2 viruses circulating in China bind exclusively to α-2 , 6-linked Sias , as is observed with human influenza viruses . Thus influenza viruses can acquire the ability to bind human-type receptors during their circulation in avian species , and mammalian intermediate hosts , for example pigs , are not necessarily needed for this process . The molecular determinants of the receptor binding preference of influenza viruses are not fully understood . Several amino acid changes in HA , including I155T , H183N , A190V , Q226L , and G228S , have been reported to promote the affinity of avian influenza viruses for human-type receptors [22] , [50] , [54] , [58] , [59] . Although 155T and 183N were conserved in all of the H9N2 viruses ( Table S1 ) , our mutagenesis study indicated that 155T , but not 183N , is necessary for the H9N2 virus to bind to the human-type receptor . The amino acid at 190 was not conserved in these strains , and the avian influenza virus-like 190A was present in several viruses that exclusively bound to the α-2 , 6-linked Sias; the amino acid change G228S was not detected in any of our H9N2 viruses . Therefore , the H183N , A190V , and G228S mutations in HA are not necessary for an H9N2 virus to bind to the human-type receptor , whereas the mutations I155T and Q226L play important roles in H9N2 virus binding to the human-type receptor . Avian influenza viruses can acquire different mutations that confer increased receptor-binding ability , virulence , or transmissibility during their replication in mammalian hosts , and some of these mutations have been detected in the H9N2 viruses ( Table S1 ) , although we were not able to find a strong relationship between these changes and the observed virulence or transmissibility in mammals of our H9N2 viruses . The 627K and 701N mutations in PB2 were detected in some viruses recovered from both inoculated and exposed ferrets , indicating that certain H9N2 viruses are predisposed to acquiring the 627K or 701N mutation in their PB2 gene when they replicate in mammals . However , the absence of the 627K and 701N mutations in the PB2 of some of the viruses recovered from the exposed animals ( Table 2 , Table S5 ) suggests that transmission of H9N2 viruses in ferrets may be independent of these changes , although such changes in PB2 could further increase their virulence and transmission , as was seen with the H5N1 and H7N9 viruses [48]–[50] , [52] , [60] . In addition to the receptor-binding preference conferred by HA , internal gene combinations also play a determinative role in virus transmissibility in mammals [2] . Similar to other avian influenza viruses circulating in poultry in Southern China [7] , [61]–[63] , the H9N2 viruses formed multiple genotypes . The DK/ZJ/C1036/09-like internal gene combination was detected in the H9N2 viruses with different groups of HA and NA genes that were isolated between 2009 and 2013 in nine of the 12 provinces investigated , and was also detected in the H7N9 and H10N8 viruses that have infected humans [12] , [40] ( Figure 1B ) , suggesting that this predominant internal gene combination is more stable and compatible with different surface genes . This internal gene combination also functions in transmission because it was present in all six of the transmissible viruses . Therefore , the H9N2 viruses pose a threat to human health not only because they will likely cause new influenza pandemic , but also because they can transfer different subtypes of influenza viruses from avian species to humans .
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Ministry of Science and Technology of the People's Republic of China . The protocols for animal studies were approved by the Committee on the Ethics of Animal Experiments of the Harbin Veterinary Research Institute ( HVRI ) of the Chinese Academy of Agricultural Sciences ( CAAS ) ( approval numbers BRDW-XBS–12 for mice and BRDW-XD–12 for ferrets ) . All experiments with live H9N2 viruses were conducted within the enhanced animal biosafety level 2+ ( ABSL2+ ) facility in the HVRI of the CAAS . The animal isolators in the facility are hyper-filtered . The researchers who work with mice and ferrets wear N95 masks and disposable overalls; they shower on exiting the facility . The H9N2 viruses used in this study were isolated from poultry in different regions of China between 2009 and 2013 . Virus stocks were grown in specific pathogen-free ( SPF ) chicken eggs . Madin-Darby canine kidney ( MDCK ) cells used for virus titration were cultured in DMEM ( CORNING , Cellgro ) medium with 4% fetal bovine serum ( FBS ) . 293T cells were cultured in DMEM medium with 10% FBS . All cells were incubated at 37°C with 5% CO2 . Viral gene amplification and sequencing was carried out as described previously [18] , [61] . Sequence data were compiled with the SEQMAN program ( DNASTAR , Madison , WI ) , and phylogenetic analyses were carried out with the PHYLIP program of MEGA 5 . 0 software using the neighbor-joining algorithm . Bootstrap values of 1 , 000 were used , and 95% sequence identity cutoffs were used to categorize each gene segment in the phylogenetic trees . Receptor specificity was analyzed by use of a solid-phase direct binding assay as described previously with modified using two different glycopolymers: α-2 , 3-siaylglycopolymer [Neu5Acα2-3Galβ1-4GlcNAcβ1-pAP ( para-aminophenyl ) -alpha-polyglutamic acid ( α-PGA ) ] and the α-2 , 6-sialylglycopolymer [Neu5Acα2-6Galβ1-4GlcNAcβ1-pAP ( para-aminophenyl ) -alpha-polyglutamic acid ( α-PGA ) ] [4] , [7] . Briefly , viruses were grown in eggs , clarified by low-speed centrifugation , laid over a cushion of 30% sucrose in phosphate buffered saline ( PBS ) , and ultracentrifuged at 28 , 000 r . p . m for 2 h at 4°C . Virus stocks were aliquoted and stored at −80°C until use . Virus concentrations were determined by using haemagglutination assays with 0 . 5% cRBCs . Microtitre plates ( Nunc ) were incubated with serial two-fold dilutions of sodium salts of sialyglycopolymers in PBS at 4°C for 30 min . Then the plates were exposed to UV light ( 254 nm ) for 10 min . After the glycopolymer solution was removed , the plates were washed three times with PBS . Then 50 µl of virus suspensions diluted with PBST ( PBS containing 0 . 1% Twen-20 ) was added to the wells and the plate was incubated at 4°C for 2 to 3 h . After being washed five times with 250 µl of PBST , the plates were fixed with 10% formalin in PBST for 30 min . After the plates were again washed five times with PBST , 50 µl of chicken antiserum against the CK/JS/C4258/12 ( H9N2 ) virus diluted with PBST was added to the wells and incubated at 37°C for 1 h . After being washed for a further five times with PBST , the plates were incubated with a horseradish peroxidase ( HRP ) -conjugated goat-anti-chicken antibody ( Sigma-Aldrich , St . Louis , MO , USA ) for 1 h at 37°C . The plates were then washed again and incubated with O-phenylenediamine ( Sigma-Aldrich , St . Louis , MO , USA ) in PBS containing 0 . 01% H2O2 for 10 min at room temperature . The reaction was stopped with 0 . 05 ml of 0 . 5 M H2SO4 . The optical density at 490 nm was determined in a microplate reader ( BIO-RAD ) . Dose-response curves of virus binding to the glycopolymers were analyzed by using a single site binding algorithm and curve fitting by GraphPad Prism to determine the association constant values ( Ka ) . Each value presented is the mean ± SD of three experiments , which were each performed in triplicate . The HA and NA segments of CK/GX/9/99 were inserted into the bidirectional transcription vector pBD as described previously [49] . The QuikChange Lighting Site-Directed Mutagenesis Kit ( Stratagene , http://www . agilent . com ) was used to create specific mutations in the HA gene by using the following primers: forward: 5′AGCATGAGATGGTTGATTCAA AAGGACAAC , reverse: 5′AGCGTTGTCCTTTTGAATCAACCATCTCAT ( for HAT155I mutation ) ; and forward: 5′ TTCATGTGGGGCATACATCACCCACCCACC , reverse: 5′TCGGTGGGTGGGTGA TGTATGCCCCACATG ( for HAN183H mutation ) . The HA ( with or without the mutations ) and NA genes of the CK/GX/9/99 virus and the six internal genes of the PR8 virus were used to generate the viruses as previously described [49] . All HA and NA genes of the constructs and rescued viruses were completely sequenced to ensure the absence of unwanted mutations . The gene sequences of the CK/GX/9/99 virus used in this study have been reported previously [18] . Groups of six-week-old female BALB/c mice ( Beijing Vital River Laboratories , Beijing , China ) were anesthetized with CO2 and inoculated intranasally ( i . n . ) with 106 . 0 EID50 of test viruses in a volume of 50 µl . Three mice were euthanized on day 3 p . i . , and the nasal turbinates , lungs , kidneys , spleens , and brains were collected for virus titration in MDCK cells . The remaining five mice in each group were monitored daily for 14 days for weight loss and survival . Four-month-old female ferrets ( Wuxi Cay Ferret Farm , Jiangsu , China ) that were serologically negative for influenza viruses were used in these studies . The animals were anesthetized via intramuscular injection of ketamine ( 20 mg/kg ) and xylazine ( 1 mg/kg ) . To examine virus replication , groups of two ferrets were anesthetized and inoculated i . n . with 106 . 0 EID50 of test viruses in a 500 µl volume ( 250 µl per nostril ) . The ferrets were euthanized on day 4 p . i . and the nasal turbinates , tonsils , trachea , lung , spleen , kidneys , liver , and brain were collected for virus titration in MDCK cells . The lung tissue was also collected for histologic study as described previously [64] . For the respiratory droplet transmission studies , groups of three ferrets were inoculated i . n . with 106 . 0 EID50 of test virus and housed in specially designed cages inside an isolator as described previously [40] . Twenty-four hours later , three naïve animals were placed in an adjacent cage . Nasal washes were collected at 2-day intervals , beginning on day 2 p . i . ( 1 day post-exposure ) and titrated in MDCK cells . The ambient conditions for these studies were set at 20–22°C and 30%–40% relative humidity . The airflow in the isolator was horizontal with a speed of 0 . 1 m/s; the airflow direction was from the inoculated animals to the exposed animals . Viral RNA was extracted and converted to cDNA by use primer 5′AGC RAA AGC AGG . Specific amplification of a 1000-nucleotide PB2 fragment covering codons 627 to 701 was applied by use a pair of specific primers ( Forward: 5′GCAACRGCTATYYTRAGGAAAGC; Reverse 5′ AGTAGAAACAAGGTCGTTTTTAAA ) . PCR fragments of each virus were pooled in equal concentrations , and libraries were created for each virus by using the Ion Xpress Plus Fragment Library Kit ( Life Technologies ) . Sequencing runs were performed by using the Ion Torrent personal genome machine ( PGM , Life Technologies ) . Sequence reads were sorted by the Ion Xpress Barcode Adaptors 1–18 ( Life Technologies ) . Reads were aligned to the PB2 reference sequence of each virus by using CLC Genomics Workbench 5 . 0 . 1 . The threshold for mutation detection was manually set at 1% . The genome sequences of the 35 viruses reported in this study are available in GenBank with the access numbers of KM113042 – KM113321 .
|
Avian influenza viruses continue to present challenges to human health . Recently the H7N9 and H10N8 viruses that are of low pathogenicity for poultry have caused human infections and deaths in China . H9N2 influenza virus have been isolated worldwide from wild and domestic avian species for several decades , and their low pathogenic nature to poultry made them a low priority for animal disease control , which has allowed them to continue to evolve and spread . Here , we investigated a series of H9N2 influenza viruses that were detected in live poultry markets in southern China . We found that these viruses are able to preferentially bind to the human-type receptor , and some of them can cause disease and transmit between ferrets by respiratory droplet . All the transmissible H9N2 viruses have a similar internal gene constellation , which was also present in the H7N9 and H10N8 viruses . Our study indicates that the widespread dissemination of H9N2 viruses poses a threat to human health not only because of the potential of these viruses to cause an influenza pandemic , but also because they can function as “vehicles” to deliver different subtypes of influenza viruses from avian species to humans .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"organismal",
"evolution",
"viral",
"transmission",
"and",
"infection",
"microbial",
"evolution",
"virology",
"biology",
"and",
"life",
"sciences",
"microbiology",
"evolutionary",
"biology",
"viral",
"evolution"
] |
2014
|
Genetics, Receptor Binding Property, and Transmissibility in Mammals of Naturally Isolated H9N2 Avian Influenza Viruses
|
The cornerstone of current schistosomiasis control programmes is delivery of praziquantel to at-risk populations . Such preventive chemotherapy requires accurate information on the geographic distribution of infection , yet the performance of alternative survey designs for estimating prevalence and converting this into treatment decisions has not been thoroughly evaluated . We used baseline schistosomiasis mapping surveys from three countries ( Malawi , Côte d’Ivoire and Liberia ) to generate spatially realistic gold standard datasets , against which we tested alternative two-stage cluster survey designs . We assessed how sampling different numbers of schools per district ( 2–20 ) and children per school ( 10–50 ) influences the accuracy of prevalence estimates and treatment class assignment , and we compared survey cost-efficiency using data from Malawi . Due to the focal nature of schistosomiasis , up to 53% simulated surveys involving 2–5 schools per district failed to detect schistosomiasis in low endemicity areas ( 1–10% prevalence ) . Increasing the number of schools surveyed per district improved treatment class assignment far more than increasing the number of children sampled per school . For Malawi , surveys of 15 schools per district and 20–30 children per school reliably detected endemic schistosomiasis and maximised cost-efficiency . In sensitivity analyses where treatment costs and the country considered were varied , optimal survey size was remarkably consistent , with cost-efficiency maximised at 15–20 schools per district . Among two-stage cluster surveys for schistosomiasis , our simulations indicated that surveying 15–20 schools per district and 20–30 children per school optimised cost-efficiency and minimised the risk of under-treatment , with surveys involving more schools of greater cost-efficiency as treatment costs rose .
Schistosomiasis is a major global health problem , and is estimated to infect 230 million people , cause at least 11 , 000 deaths per year [1 , 2] and 3 . 3 million Disability Adjusted Life Years in 2010 [3] . While various tools are used to control the disease , population-level preventive chemotherapy ( PC ) with praziquantel is currently the cornerstone of schistosomiasis control , and PC-based programmes are scaling up across Africa [4 , 5] . Since schistosomiasis shows high spatial heterogeneity in its geographic distribution , and school-age children ( SAC , aged 5–14 years ) are an important high-risk group [6] , PC is targeted to those areas at greatest risk of infection and focuses on this age group . Treating all SAC in such areas remains more cost-effective than individual-level test-and-treat strategies [7 , 8] and is achieved by assigning geographic areas to WHO-recommended treatment classes based on infection prevalence ( Table 1 ) [6 , 9] . Thus , to decide treatment classes , epidemiological data must first be collected to ascertain prevalence . Published WHO guidelines do not provide specific guidance on survey design for this purpose , but recommend collecting prevalence data according to ‘ecological zones’ [6] . Ecological zones are not well defined , however , and it is unclear how to convert ecological zone-based prevalence data into treatment decisions , which are commonly applied across implementation units , such as health or educational districts . As a result , a variety of survey designs have been adopted [10–14] and the performance of alternative designs has not been thoroughly evaluated . Computerized sampling simulations have proved useful for assessing alternative survey approaches for various tropical diseases , including soil-transmitted helminthiases , trachoma and intestinal schistosomiasis [15–17] . These entail generating ‘‘gold standard” datasets that maintain the spatial heterogeneity observed in empirical datasets , and then comparing how well alternative sampling approaches perform in estimating known parameters from this spatially realistic simulated data . Such simulations allow multiple approaches to be tested with far greater replication than would be possible in field tests . A previous simulation study on schistosomiasis [15] compared the performance of two methods to estimate school level infection prevalence , lot quality assurance sampling ( LQAS ) and a spatial grid-based survey combined with spatial interpolation . Here , we focus on comparing simple two-stage cluster survey designs that aim to estimate schistosomiasis prevalence among SAC for a set of implementation units , such as districts . This simple survey design is often used when more complex ( e . g . spatially informed ) survey designs are not possible , when the spatial distribution of infection within districts is thought to be relatively homogeneous , or when a simple , easy to analyse survey is desired . For example , the design of spatially stratified surveys requires reliable information on the likely distribution of disease , as well as accurate location data for all potential sampling locations . Such information may not available for the entire area to be surveyed . Two-stage cluster surveys involving a simple random sample of sites from a list in each implementation unit , have therefore been used to generate implementable PC plans in a number of countries [18–21] . We apply sampling simulations to data from three African countries in order to assess what number of sampling sites , and number of individuals screened at each site , maximises survey accuracy and cost-efficiency respectively . Survey accuracy is expected to increase asymptotically with survey size , with diminishing returns as surveys get larger . However , a curved relationship between survey size and cost-efficiency is expected , with the optimum reflecting a balance between survey accuracy and cost . Cost-efficiency should increase with survey size initially , but then decrease as the costs of very large surveys outweigh their accuracy benefits . The key question then , is at what point this curve turns , i . e . what survey size leads to the most accurate treatment decisions per unit cost ?
Ethical approval for the surveys analysed here ( including the consent process and all methodology ) was obtained from Imperial College London Research Ethics Committee ( ICREC_8_2_2 ) . Surveys were performed as part of national schistosomiasis and soil transmitted helminth control programmes in all countries , overseen and approved by the Ministry of Health . In Malawi , all participants provided written consent . In Côte d’Ivoire and Liberia , as participants were under the age of 18 years the Ministry of Health required written consent be provided by adults . As literacy levels were very low , written consent from every parent or guardian was not possible , and each head teacher provided written consent for the survey and all individual participants provided oral consent . In Liberia , headteachers also received oral consent from parents at Parent Teacher Association ( PTA ) meetings before surveys began , with parent presence at PTA meetings documented on an attendance register . Only children who consented either in writing ( in Malawi ) or orally ( in Côte d’Ivoire and Liberia ) took part in the surveys . All analyses were performed in R v3 . 1 . 0 . Data from baseline mapping surveys in Malawi , Côte d’Ivoire and Liberia were used to characterise spatial heterogeneity in schistosomiasis prevalence ( Table 2; Fig 1 ) . All surveys adopted the same two-stage cluster survey protocol . This involved randomly selecting 15–20 primary schools in each implementation unit ( health or educational district ) , then selecting 30 children per school ( with an even gender ratio ) to be tested for Schistosoma haematobium using a single urine filtration and S . mansoni using duplicate Kato-Katz slides from a single stool . At each school , children aged 10–14 were eligible for participation . Where age was difficult to determine , grades corresponding to the targeted age group were used . Children were selected systematically at each school , by assembling a line each of eligible boys and girls , and using a sampling interval to select the required 15 children of each gender . Survey sample size was originally decided based on precision-based sample size calculations ( S1 Text ) . We used semivariograms to characterise spatial heterogeneity in the prevalence of schistosomiasis ( infection by either S . haematobium , S . mansoni , or either species ) , in each country . Before creating semivariograms , logistic regression models were performed to remove large-scale spatial trends in prevalence . For each country , individual level infection ( 1/0 ) was modelled as a function of longitude and latitude , with school as a random factor , using a binomial mixed model in the R package lme4 . Where either longitude or latitude did not explain significant variation in prevalence ( i . e . p>0 . 05 in likelihood ratio tests ) , these terms were removed from the final model . Using the R package geoR , omnidirectional semivariograms were then fitted for each country to the school level random effects , with weighted least squares fits of exponential , spherical , and Gaussian models . To further characterise spatial heterogeneity in schistosomiasis prevalence and provide parameters useful in future sample size calculations , we calculated the intra-cluster correlation coefficient ( ICC ) for schistosomiasis prevalence using the iccbin function in R package aod [22] . The ICC is a measure of the relatedness or similarity of clustered data , with values ranging from 0 to 1 . As the ICC increases , the more individuals within clusters , and the less individuals in different clusters , resemble one another . In the context of our surveys , the higher the district-level ICC , the more similar children in the same school are in their infection status , compared to children in different schools . In sample size calculations , ICC determines the design effect; higher ICC values increase the design effect and the sample size required to obtain a given level of precision for parameter estimates such as prevalence . Spatially realistic sampling locations were required to create gold standard datasets . Since geo-referenced data on primary school locations was only available for one of the three countries ( Malawi ) , locations were simulated . We used PPS ( “probability proportional to size” ) sampling according to population density ( data from Worldpop , http://www . worldpop . org . uk/ ) in package raster [23] to assign primary schools to locations in each country ( Table A in S2 Text ) . Geo-referenced school datasets from Malawi and Kenya were used to validate this approach . Good correspondence was observed between simulated and actual school locations across both countries , which was improved by assigning schools separately to large cities ( population >1m; Lilongwe , Blantyre and Nairobi ) and to the rest of the country ( Fig A and Fig B in S3 Text ) . We focused simulations on survey designs estimating pooled S . haematobium and S . mansoni prevalence , since treatment guidelines are based on pooled schistosomiasis prevalence ( Table 1 ) . Conditional simulation was performed using semivariogram parameters to generate 1000 different realizations per country that reproduced the global characteristics of the source data . For each country , the semivariogram model for pooled schistosome infection with the lowest sum of squares was chosen as the best fit for conditional simulation . For Malawi and Liberia , conditionally simulated random effect values were added to predicted trend surfaces ( predictions using latitude and longitude ) on the log odds scale before being back transformed to prevalence . For Côte d’Ivoire , where no spatial autocorrelation in schistosomiasis was detected ( and therefore conditional simulation could not be performed ) random effect values were assigned to each school by sampling from the distribution of observed school random effect values . To generate a distribution of random effect values to sample from , the distribution of observed random effects was smoothed using kernel smoothing . Overall , we created 1000 different schistosomiasis prevalence realizations for 5 , 239 primary school locations in Malawi , 11 , 429 in Côte d’Ivoire and 2 , 785 in Liberia ( Table A in S2 Text ) . Based on data provided by government ministries or enrolment figures collected during mapping surveys ( Table A in S2 Text ) , we assumed a population of school-age children ( SAC , defined by the WHO as children aged 5–14 ) per school of 400 for Malawi , 250 for Côte d’Ivoire and 150 for Liberia , and thus calculated the number of SAC that were infected and uninfected with schistosomiasis at each school in each realization . We simulated two-stage cluster surveys involving a random sample of 2 , 5 , 10 , 15 or 20 schools per district , and a random sample of 10 , 20 , 30 , 40 or 50 SAC per school . For each realization , the survey package in R [24] was used to calculate district prevalence and 95% confidence intervals with the ‘beta’ method for proportions . Districts were classified into WHO endemicity and associated treatment classes according to point prevalence estimates ( Table 1 ) [6] . We assessed the accuracy of each survey design across 1000 realizations using four key metrics: ( 1 ) the proportion of times that treatable levels of infection ( ≥1% prevalence ) were not detected ( 2 ) prevalence estimate precision , as reflected by the width of the 95% confidence interval ( 3 ) the proportion of times districts were misclassified ( assigned to the wrong treatment class ) and ( 4 ) the proportion of times districts were under-classified ( assigned to a treatment class below their true class ) . We also explored how three alternative district assignment rules performed , wherein districts were placed into the next highest treatment class if their prevalence estimate was within either 2 or 5 percentage points of the 10% or 50% threshold ( “boosting rules" ) , or if the upper 95% confidence limit overlapped a higher treatment class threshold ( “upper CL rule” ) . To ensure comparability across all survey designs , districts with fewer than 20 schools were excluded from analyses . This resulted in each survey design being assessed 1000 times across a total of 143 districts from the three countries ( 26 districts in Malawi , 78 in Côte d’Ivoire and 39 in Liberia , Table A in S2 Text ) . We further explored how survey accuracy according to these four metrics varied with district size . The cost of each survey design was estimated for Malawi using an ingredients based approach [25] , using itemized cost data from a 2014 schistosomiasis/STH mapping survey conducted in Malawi . We assumed each survey covered the whole country and monitored S . haematobium and S . mansoni . Only financial costs were estimated , with some costs fixed ( invariant of sampling strategy ) and others variable ( depending on survey design; Table 3 ) . Capital item costs were annualized over their useful lifespan ( Table B in S2 Text ) and a wastage factor of 10% was applied to all relevant items [26] . Based on field experience in Malawi , we assumed three survey teams of constant size worked in parallel , and that survey duration ( and hence staff salary costs ) changed with survey design ( Table C in S2 Text ) . Survey design-specific travel distances for fuel costs were estimated in qGIS: for each district and for each number of schools , a single random selection of schools was taken from the geo-referenced school database and the shortest path linking them was used as an estimate of average distance travelled per district . All costs were adjusted to US dollars ( US$ ) using the July 1st , 2014 exchange rates of 399 . 216 Malawian Kwacha ( MWK ) and 0 . 586 UK pounds to 1 US dollar ( www . oanda . com/convert/classic ) . Per-capita treatment cost was assumed to be $0 . 30 , based on financial costs of programme delivery and purchased praziquantel costs in Malawi for 2014 , but was varied from $0 . 10 to $0 . 60 in sensitivity analyses [27–29] . The total cost for each design was calculated by summing survey costs and the cost of treating the entire Malawi SAC population ( 4 . 3 million according to Ministry of Education figures from February 2014 ) over the subsequent six years , in accordance with survey results . Six years was chosen as WHO recommends re-assessment surveys after 5–6 years [6] . Using WHO guidelines , we assumed low endemicity districts would be treated every three years ( such that SAC are treated twice during primary schooling ) , moderate endemicity districts would be treated biennially and high endemicity districts would be treated annually , while districts with prevalence below 1% would not receive treatment . Under the assumption that treating districts more frequently than required is not harmful but treating less frequently is , we define districts as receiving “adequate” treatment when they are assigned to either their correct or a higher treatment class . The annual cost per district adequately treated , c , incorporates the assumptions described and was calculated using Eq 1 . We converted c to a percentage of the annual cost of blanket treatment per district ( cprop ) , using Eq 2 . Cost-efficiency is optimised when cprop is minimised . Parameters are defined in Table 4 . The sensitivity of survey costs to variation in three inputs was assessed: 1 ) the number of survey teams , 2 ) survey staff salaries and 3 ) capital item lifespan . We also examined how variation in per capita treatment cost affected cost-efficiency , as per-capita treatment costs are likely to vary according to country-specific delivery costs and economies of scale [28 , 30] . Finally , to explore how conclusions might be affected by country-specific differences in geography and epidemiology , we examined cost-efficiency using simulation results for either Côte d’Ivoire or Liberia , combined with Malawi’s survey cost estimates and SAC population size .
Countries varied in overall schistosomiasis prevalence , species composition and the spatial heterogeneity of prevalence ( Table 2 , Fig C in S3 Text ) . Spatial autocorrelation in prevalence was detected for S . haematobium in all three countries , for pooled schistosomiasis in Malawi and Liberia but not Côte d’Ivoire , and for S . mansoni only in Liberia ( Fig 2 ) . Range values indicated that no spatial correlation was present beyond 20–65km , with the exact distance within this range depending on the country and species considered . Values of the district-level intra-cluster correlation coefficient ( ICC ) ranged from 0 to 0 . 774 across the three country dataset ( Table 5 , Fig C in S3 Text ) , indicating that districts varied widely in the extent to which prevalence clustered by school . Districts of a given endemicity class ( according to the point estimate of prevalence ) , contained schools with widely varying schistosomiasis prevalence , particularly in moderate and highly endemic districts ( Fig D in S3 Text ) . A saturating relationship was seen between survey-estimated district prevalence and the proportion of schools positive for schistosomiasis; the proportion of endemic schools increased steeply up to ~10% prevalence , but once district prevalence exceeded 30% it was rare to find schools entirely free of infection ( Fig 3 ) . District prevalence estimates converged on true prevalence as the number of schools surveyed increased ( Fig 4 ) , as expected . The probability of failing to detect endemic schistosomiasis at treatable levels ( ≥1% prevalence ) declined as more schools were sampled per district ( though with diminishing returns , particularly beyond 10 schools per district ) , while increasing the number of children tested per school led to very small improvements ( Fig 5A and 5B , Fig E in S3 Text and Table D in S2 Text ) . Failure to detect treatable levels of schistosomiasis with smaller surveys was most acute in districts with low prevalence . For example , surveying 2–5 schools in low endemic districts led to detection failure 10–50% of the time ( Fig 5A , Table D in S2 Text ) . Surveying more schools clearly improved the precision of prevalence estimates and the accuracy of treatment class assignment ( Fig 5C , 5E and 5G , Fig E in S3 Text ) , again with diminishing returns above 10 schools per district . In contrast , increasing the number of children sampled per school led to negligible improvements in precision and assignment accuracy ( Fig 5D , 5F and 5H , Fig E in S3 Text ) . For a given total sample size , sampling fewer children in more schools rather than many children in few schools minimised detection failure and maximised the accuracy of prevalence estimates and treatment class assignment ( Fig F in S3 Text ) . The use of more lenient district assignment rules ( boosting rules ) reduced the probability that districts were classified into a treatment class below their true class ( Fig G in S3 Text and Table E in S2 Text ) . Use of boosting rules also allowed smaller surveys to be used while still achieving a specified maximum allowable probability of under-classification . For example , if one wanted the risk of district under-classification to not exceed 7% , this could be achieved by either surveying 20 schools and using point estimate district assignment , or surveying 10 schools and using a 2% boosting rule . District size had an influence on some measures of survey accuracy , with the probability of wrongly classifying and under-classifying districts being greater in large districts , for a given survey design ( Fig H in S3 Text ) . However , district size did not appreciably alter the shape of the relationship between survey size and measures of accuracy ( Fig H in S3 Text ) . Although surveying more schools per district improved the accuracy of district-level classification , schools still showed great variation in prevalence within any given district . If the same WHO thresholds were applied at a school level , under most survey designs around 50% of schools would have been wrongly classified and 10–30% under-classified , with only very modest improvement as more schools were sampled per district ( Fig I in S3 Text ) . This is due to the underlying large heterogeneity in prevalence among schools within a district ( Fig D in S3 Text ) . Based on a range of 2–20 schools surveyed per district and 10–50 children sampled per school , the estimated cost of a nationwide schistosomiasis survey in Malawi varied between $22 , 482 and $135 , 033 ( Table F in S2 Text ) . In sensitivity analyses , survey costs were most sensitive to variation in staff salaries ( per diems ) , while the number of survey teams and capital item lifespan had minimal effects ( Fig J in S3 Text ) . The number of schools surveyed strongly influenced both the absolute cost of surveys and cost-efficiency , while varying the number of children screened per school had smaller cost implications , particularly in terms of cost efficiency ( Fig 6 , Fig K in S3 Text ) . Under parameters expected for Malawi ( per capita treatment cost $0 . 30 , 4 . 3 million SAC nationwide ) , surveying 15 schools per district was the most cost-efficient survey size , with assignment using the point prevalence estimate or boosting rules performing similarly ( Fig 7A ) . Use of 95% confidence intervals to assign treatment classes was cost inefficient , particularly with small surveys where confidence intervals were wide ( Fig 7A ) . While the use of boosting rules was not always the most cost-efficient strategy , they notably improved the proportion of districts adequately treated ( Fig 7B ) for some additional cost ( Fig 7C ) . These cost-efficiency results for Malawi were robust to variation in staff salaries ( Fig L in S3 Text ) . Furthermore , when we halved or double the assumed per capita treatment cost , and assessed how cost-efficiency might be altered when simulation results from either Côte d’Ivoire or Liberia were used instead of those from Malawi , results were remarkably consistent: surveys of 15–20 schools per district maximised cost-efficiency in all cases ( Fig 8 ) . As treatment costs increased , the cost-efficiency of surveying more schools per district increased , and at $0 . 60 per treatment a survey of 20 schools per district was often more cost efficient than 15 ( Fig 8 ) . In these sensitivity analyses , treatment class assignment using either the point prevalence estimate or the two percentage point boosting rule was most cost efficient ( depending on the country considered ) , with worse cost-efficiency for the larger five percentage point boosting rule , and worse still when using 95% confidence intervals ( Fig 8 ) .
Schistosomiasis is often described as a focal disease , with prevalence varying widely even from one village to the next . Our results from Malawi , Côte d’Ivoire and Liberia illustrate this well . Prevalence varied very widely across schools in moderate or high endemicity districts ( Fig D S3 Text ) , and spatial autocorrelation in prevalence was detected across short distances ( 20–65km ) , similar to previous findings from East and West Africa [15 , 31] . Moreover , spatial autocorrelation was not detected in all cases , indicating no spatial clustering or clustering at scales too small to detect with our data . District-level intracluster correlation coefficients ( ICC ) were also highly variable , ranging from 0 to 0 . 775 with a mean of 0 . 116 . These findings indicate greater within-district spatial heterogeneity in schistosomiasis prevalence than reported for STH [16] . Our simulations clearly demonstrated that increasing the number of schools surveyed provided much greater gains in the accuracy of prevalence estimates and treatment class assignments than increasing the number of children tested per school . Surveying too few schools per district risked failing to detect treatable levels of schistosomiasis , particularly in low endemic districts . For instance , simulated surveys of five schools per district led to low endemic districts being classified as non-endemic between 2% and 23% of the time . Since district prevalence will usually not be re-assessed for 5–6 years [6] , it is important that enough schools are surveyed initially such that districts are not erroneously classified as non-endemic , missing treatment for several years . Using cost data from Malawi , cost-efficiency improved notably as surveys increased in size from 2 to 15 schools per district , with only a very small decline in cost-efficiency at 20 schools per district . Conversely , the number of children sampled per school did not greatly affect cost-efficiency . Decisions about sample size per school could , therefore , be made in light of practical considerations , such as to maximise the number of schools visited per day , provide reasonably accurate prevalence estimates to feed back to communities , or acquire data for operational research needs . Sensitivity analyses showed that as treatment costs increased , larger surveys of 20 schools per district became most cost-efficient; essentially , the higher the cost of nationwide treatment , the greater the benefit of larger surveys to enable accurate geographic targeting of drug delivery . This may apply even more so in suspected high endemicity areas where treatment of at-risk adults as well as SAC is to be performed . In all sensitivity analyses , surveys involving 10 or fewer schools per district showed inferior cost-efficiency compared to those involving 15–20 schools . Thus , although surveys of five schools per district have been suggested to be cost efficient for soil-transmitted helminths [16] , schistosomiasis surveys need to be larger [10] . Two-stage cluster surveys aiming to provide treatment guidelines for both types of infection should therefore optimise sample size for schistosomiasis , and STH prevalence estimates of sufficient accuracy will follow . The latest draft of WHO guidelines on schistosomiasis mapping [32] suggest that districts can be mapped with as few as five schools per district , where the distribution of infection is thought to be homogeneous . Our results suggest that this may not be optimal , as this sample size led to some districts with treatable levels of infection going undetected , and poorer cost-efficiency than surveys involving 15 or 20 schools per district . Thus , what might be perceived as quite large surveys ( 15–20 schools per district ) are shown here to pay off for schistosomiasis mapping in terms of cost-efficiency . The draft guidelines also suggest surveying 50 children per school . Our results indicate that a sample size of 20–30 children per school produced very similar accuracy and cost-efficiency results to those involving 50 per school ( Fig 4 , Fig F in S3 Text ) , such that this recommended within-school sample size could be reduced , particularly when paired with surveying a larger number of schools ( e . g . 15–20 ) per district . We found that using alternative rules to convert prevalence estimates into treatment class assignments could be beneficial in some situations . The use of boosting rules , where districts were boosted into the next highest treatment class when close to a threshold , in combination with small surveys ( 2–5 schools per district ) was associated with a reduced risk of under-treatment , and comparable or improved cost-efficiency compared to assignment using point prevalence estimates . However , with larger surveys ( 10–20 schools ) this pattern reversed and while boosting rules reduced under-treatment they also reduced cost-efficiency . Therefore , when small surveys ( <10 schools per district ) are unavoidable due to logistical , budgetary or time constraints , boosting rules might prove useful to avoid under-treatment . Use of 95% confidence intervals in assignment of districts to treatment classes was never cost-efficient due to a high degree of over-treatment . This was particularly true when used in combination with small survey sizes , suggesting this practice should be avoided . By using data from three different African countries , our simulations allowed us to identify a ball-park sample size and design for optimising district-level schistosomiasis surveys across similar settings . However , several limitations to this study warrant discussion . First , we only compared simple two-stage cluster surveys , frequently adopted where there is limited reliable information about schistosomiasis distribution that could be used to design more complex , potentially more powerful surveys , or when a survey design that is easy to implement and analyse is desired . If information about the likely distribution of infection is available and is of sufficient quality and geographic scope , other survey designs could prove superior . For example , where schistosomiasis prevalence is strongly associated with known geographic risk factors ( such as lake proximity for S . mansoni in Uganda and Burundi [33] ) , geographic survey stratification can improve the accuracy of district-level prevalence estimates , while guarding against random samples that happen not to capture schools in high risk areas . Re-assessment surveys may often be able to incorporate such stratification , using prevalence information from prior surveys to guide design . Where geographic risk factors drive strong within-district variation in prevalence , survey and treatment strategies at a sub-district level are also likely to be fruitful , where feasible . If geographic coordinates of potential sampling locations are available , explicitly spatial survey designs may also be an option [15 , 34] and geospatial modeling can be used to convert survey data into PC implementation maps [34–36] . However , both stratified designs and geospatial approaches require increased expertise for appropriate survey implementation and data analysis . Studies examining how best to optimise other surveys designs , for example those stratifying for water body proximity for S . mansoni , would be valuable . Second , we have not thoroughly examined how differences in factors such as district size , population density and urbanization , all of which can vary widely across countries , might affect optimal survey design . We show that the accuracy of district assignment to treatment classes was lower for larger districts ( Fig H in S3 Text ) . However , since survey costs can also be expected to scale with the size of districts , it is unclear exactly how district size might alter survey cost-efficiency . With cost data from only one country we were unable to explore this . Future studies that do so , using matched survey and cost data from a range of countries with pronounced variation in factors like district size , would be valuable . Finally , while our results concern the use of urine filtration and Kato-Katz for diagnosing schistosome infection , use of the urine-based circulating cathodic antigen ( CCA ) test for S . mansoni surveys is increasing [37 , 38] . It is possible that use of this alternative diagnostic could alter optimal survey size , for example if increased test sensitivity leads to lower observed spatial heterogeneity in infection . Thus , our findings may not be directly applicable to the use of different diagnostic tests , and further work is needed to understand how the CCA test alters survey results and optimal design . Although our aim here was to assess survey performance for generating district-level treatment decisions , our data highlight the broader issue that schools within a district often vary widely in prevalence , particularly in high prevalence areas ( Fig D in S3 Text ) . If WHO treatment recommendations were applied at the school level , our simulations showed that using district level prevalence estimates derived from two-stage cluster surveys would lead to 10–30% schools being under-treated . Increasing survey size made very limited improvements here , and achieving this would require a fundamental shift in approach , for example towards either surveys at finer spatial scale or school-level surveillance . Currently , the cost and complexity of schistosomiasis diagnostic tests means surveillance is often impractical , particularly for S . mansoni where a test that could be easily administrated by teachers is not currently available . In areas where only S . haematobium is endemic , however , surveillance strategies may be fruitful and could for example utilise blood-in-urine questionnaires or urine dipsticks [39] delivered to schools through widespread vaccination or STH ( Albendazole ) control programmes , such that they could self-perform and report an LQAS assessment for schistosomiasis . Alternatively , geospatial modeling could be used to predict endemicity class at non-surveyed schools using data from surveyed schools [15] . Such school-level approaches could ameliorate the problem highlighted here that district-level averages mask wide variation in infection risk among sites , though use of such approaches would require new accompanying guidelines for converting site-specific prevalence levels into treatment practice .
Based on our findings , we suggest that among simple two-stage cluster survey designs , sampling 15–20 schools per district and 20–30 children sampled per school is suitable for estimating district-level schistosomiasis prevalence and placing districts into treatment classes . This design fulfilled three key requirements: ( 1 ) it minimized the risk of failing to detect treatable infection levels in a district ( 2 ) it generated reasonably accurate prevalence estimates and treatment decisions and ( 3 ) it was cost-efficient , minimizing the cost per district adequately treated . Surveys towards the upper end of this range are likely to be more cost-efficient where treatment costs are high . The data presented here provides an initial evidence base for sample size in schistosomasis surveys , which can be built upon by future work assessing optimal survey design in the context of disease control . Our approach can also be adapted to inform on optimal survey design for other NTDs where two-stage cluster surveys are appropriate .
|
Many countries are currently scaling up efforts to control schistosomiasis , a helminthic disease for which preventive chemotherapy with praziquantel is the main control tool . In order to apply WHO guidelines on how frequently to treat a given district or similar geographic unit for schistosomiasis , survey-based estimates of infection prevalence are required . However , the optimal size and design of survey for generating such data is not clear , and there is a clear trade-off between accuracy and cost–larger surveys provide more accurate information with which to target treatment , but cost more to carry out . Here , we systematically assess what size and design of simple 2-stage cluster survey ( where primary school children are tested for infection ) , might best enable control programmes to implement WHO treatment guidelines . We use empirical data on schistosomiasis distribution from three African countries together with computer simulations to compare survey performance , in terms of accuracy and cost-efficiency–the ability of a survey to accurately determine treatment frequency , per unit cost . We show that although small surveys of around 5 schools per district are frequently adopted for mapping schistosomiasis , such small surveys are prone to miss endemic schistosomiasis fairly often , and are also not cost efficient . Our results suggest that among the designs tested , surveys involving 15–20 schools per district optimise cost-efficiency , providing the most accurate treatment decisions per dollar spent . These findings have important implications for the schistosomiasis control community , and provide the first evidence-based suggestion of a simple survey design for mapping schistosomiasis in endemic countries .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion",
"Conclusion"
] |
[
"schistosoma",
"invertebrates",
"schistosoma",
"mansoni",
"medicine",
"and",
"health",
"sciences",
"education",
"helminths",
"sociology",
"tropical",
"diseases",
"geographical",
"locations",
"social",
"sciences",
"malawi",
"parasitic",
"diseases",
"animals",
"health",
"care",
"mathematics",
"statistics",
"(mathematics)",
"neglected",
"tropical",
"diseases",
"africa",
"schistosoma",
"haematobium",
"treatment",
"guidelines",
"schools",
"people",
"and",
"places",
"helminth",
"infections",
"schistosomiasis",
"confidence",
"intervals",
"liberia",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"health",
"care",
"policy",
"organisms"
] |
2017
|
Optimising cluster survey design for planning schistosomiasis preventive chemotherapy
|
We have previously reported that Vivax Malaria Protein 001 ( VMP001 ) , a vaccine candidate based on the circumsporozoite protein of Plasmodium vivax , is immunogenic in mice and rhesus monkeys in the presence of various adjuvants . In the present study , we evaluated the immunogenicity and efficacy of VMP001 formulated with a TLR9 agonist in a water-in-oil emulsion . Following immunization , the vaccine efficacy was assessed by challenging Aotus nancymaae monkeys with P . vivax sporozoites . Monkeys from both the low- and high-dose vaccine groups generated strong humoral immune responses to the vaccine ( peak median titers of 291 , 622 ) , and its subunits ( peak median titers to the N-term , central repeat and C-term regions of 22 , 188; 66 , 120 and 179 , 947 , respectively ) . 66 . 7% of vaccinated monkeys demonstrated sterile protection following challenge . Protection was associated with antibodies directed against the central repeat region . The protected monkeys had a median anti-repeat titer of 97 , 841 compared to 14 , 822 in the non-protected monkeys . This is the first report demonstrating P . vivax CSP vaccine-induced protection of Aotus monkeys challenged with P . vivax sporozoites .
The range of Plasmodium vivax transmission spans 95 countries putting 2 . 86 billion people at risk for this malaria parasite [1] and causes an estimated 132–391 million infections each year [2] . In addition to its widespread distribution , P . vivax also has the propensity to form dormant hypnozoites in the liver , which reactivate periodically and result in recurring relapse infections . Currently , the only treatment for these intrahepatic hypnozoites is the 8-aminoquinoline , primaquine ( PQ ) , which is contraindicated in a variable proportion of populations due to a risk of hemolysis in individuals with G6PD deficiency [3] or during pregnancy . More recently , Bennett and colleagues reported an association between decreased activity of the CYP2D6 isoenzyme and reduced metabolism of PQ resulting in treatment failure [4] . This further reduces the pool of individuals who may be treated with PQ , reinforcing the need to develop a vaccine to prevent P . vivax malaria . However , resources for vivax research remain limited , with only 5% of malaria funds specifically directed toward P . vivax between 2007 and 2011 ( PATH Malaria R&D Report , 2013 ) . In addition , funding initiatives such as the U . S . government's President's Malaria Initiative ( PMI ) have strictly limited assistance , mostly to select countries in Africa , leaving little room for funding vivax malaria control or research [5] . Due to the unpredictability of hypnozoite reactivations that cause relapse infections , an intervention based on a preerythroctyic stage antigen is even more imperative for P . vivax to prevent primary infection and subsequent relapse infections . The circumsporozoite protein ( CSP ) is the major protein present on the surface of sporozoites and is involved in hepatocyte binding and invasion and as such is the lead vaccine candidate for malaria . Presence of CSP on hypnozoites [6] makes it an attractive target against both the sporozoite , and intrahepatic parasites . We have designed and produced a vaccine based on the CSP of P . vivax and demonstrated its antigenicity and immunogenicity [7] [8] . Rodents serve as a platform for the initial screening of malaria vaccine candidates . However , non-human primates , being closer to humans , are more suitable models to assess vaccines . A limited number of studies have been performed to analyze immunogenicity , and even fewer to assess the efficacy , of candidate vaccines for P . vivax in non-human primates . In the late 1980s and early 1990s studies were performed with recombinant P . vivax CS proteins produced in yeast [9] and E . coli ( WRAIR-SKB ) , which gave little to no protection in immunized Saimiri monkeys [10] . Subsequently , multiple antigen constructs were used to develop epitope-based vaccines using the vivax repeat motif . Protection was observed in Saimiri monkeys [11] [12] , but the lack of a control group makes it difficult to conclusively interpret the data in these studies . An Aotus monkey model was used to assess immunogenicity of CS multiple antigen peptides ( MAP ) and long synthetic peptides ( LSP ) and cells from immunized monkeys were used to delineate T-cell epitopes on the protein [13] [14] . No challenge studies were performed with the MAP or LSP immunized monkeys to correlate the immunogenicity to protection . In this study , we evaluated the immunogenicity and protective efficacy of VMP001 , a soluble P . vivax CSP-based vaccine , co-formulated with a TLR-9 agonist and a water-in-oil emulsion . Immunized A . nancymaae monkeys demonstrated high antibody titers and high efficacy following challenge with P . vivax sporozoites . Vaccine efficacy was shown to be associated with antibodies to the repeat region of P . vivax CSP . To our knowledge , this is the first report of an efficacy study of a recombinant CSP-based vaccine in Aotus nancymaae monkeys , and the first to conclusively correlate this protection to antibodies generated against the repeat region of the CSP following immunization .
The vaccine antigen , designated Vivax Malaria Protein 001 ( VMP001 ) , is based on the P . vivax CSP and has been described previously [7] , [8] . Briefly , the gene encoding a chimeric central repeat region containing the nonameric repeat motifs from both VK210 ( nine copies ) and VK247 ( one copy ) parasites and flanked by the N- and C-terminal regions of the P . vivax CSP was cloned into a plasmid under a T5 promoter , expressed in Escherichia coli , and purified under cGMP . The purified vaccine antigen did not have detectable endotoxin , both by an in vitro assay , as well as by in vivo pyrogenicity testing in rabbits [8] . P . falciparum Liver Stage Antigen 1 ( PfLSA-1 ) , kindly provided by Dr . David Lanar [15] , was used as an unrelated malaria antigen for immunizing the control group of animals . Recombinant proteins encoding the N-terminal and C-terminal regions of P . vivax CSP were produced in E . coli for use as plate antigens in humoral assays . A 36-mer peptide representing four copies of the classical Type 1 repeat motif was used to detect antibodies to this repeat region . The research using non-human primates ( Aotus nancymaae and Aotus lemuirinus grisiemembra ) reported in this manuscript was conducted under animal protocol number 1588BARMONB and was approved by the Institutional Animal Care and Use Committee ( IACUC ) of the Centers for Disease Control and Prevention . The non-human primate ( NHP ) research was conducted in compliance with the Animal Welfare Act and other U . S . federal government statutes and regulations relating to animals and research involving animals All NHP research adhered to the principles stated in the Guide for the Care and Use of Laboratory Animals , National Research Council . The New World monkeys used in this study were either captive bred or imported for research from Peru under a program administered through PAHO and NIH for research purposes and were socially pair-housed in appropriately spacious stainless steel cages in a BSL2-facility at the CDC vivarium facilities in Atlanta , Georgia , USA under controlled conditions of humidity , temperature , and light ( 12-hour light/12-hour dark cycles ) . This facility is fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care International and has an approved OLAW Assurance #A4365-01 . The animals were provided with commercial high protein food biscuits and supplemented with appropriate fresh vegetables , fruits and treats . Drinking water was provided ad libitum . Enrichment was provided in the form of hollow pieces of large tubing to simulate tree trunk cavities , mirrors , food puzzles and perches . Animals were monitored daily for health and discomfort . A large experienced staff is available including full time veterinarians and a pathologist . All steps were taken to ameliorate the welfare and to avoid any suffering of the animals in accordance with the recommendations of the Weatherall report for the use of nonhuman primates in research . Major surgeries ( splenectomy ) were performed with full anesthesia in aseptic surgical suites by an experienced surgical veterinarian . Monkeys were not euthanized during this research . Vivax malaria infections were treated with standard antimalarial drugs; quinine , chloroquine and primaquine as noted previously . Male and female monkeys were randomly assigned to three groups of eight animals each . Two test groups received either 15 µg or 50 µg of VMP001 per immunization . A third group of monkeys , which served as controls , was immunized with 50 µg P . falciparum LSA-1 . All antigens were formulated in Montanide ISA 720 ( Seppic , NJ ) ( antigen to adjuvant ratio of 30∶70 ) plus 200 µg of CpG 10104 ( Coley Pharma ) per immunization . Animals were immunized intramuscularly in the thigh muscles , alternating legs for each immunization , with approximately 300 µl of each vaccine formulation on day 0 , 28 and 114 . Monkeys were bled two weeks prior to immunization and at two-week intervals following vaccinations for the collection of serum . Serum was separated by centrifugation , aliquoted and frozen at −70°C until use . Sporozoites were isolated from salivary glands of Anopheles dirus that were infected with the Brazil VII strain of P . vivax after feeding on an infected A . lemurinus grisemembra monkey ( Barnwell et al . manuscript in preparation ) . The Brazil VII strain of P . vivax carries a type I ( VK210 ) csp gene ( J . Barnwell , unpublished data ) . Each monkey was challenged intravenously with 10 , 000 sporozoites six weeks after the third immunization . Two to three days following sporozoite challenge all the monkeys were splenectomized under general anesthesia and sterile operating conditions . Splenectomies were performed to allow for the establishment of a synchronous blood stage infection resulting in a more uniform , and shorter , prepatent period . Blood stage parasitemia was monitored in thick and thin Giemsa stained blood smears starting approximately 14 days post-challenge . Monkeys that demonstrated patent blood stage infection were treated with 50 mg/kg of quinine for seven days . Parasitemia was monitored on a daily basis until the conclusion of second phase of the study ( described below ) . Kaplan-Meier survival curves were generated for each group individually , as well as for pooled ( low- and high-dose ) vaccinated groups , and plotted as percentage of uninfected monkeys over time . Survival curve comparisons were performed using Log-rank ( Mantel-Cox ) test . Overall vaccine efficacy ( VE ) was calculated as 1-R×100 , where R = Iv/Ic i . e . the ratio of Incidence of malaria in vaccinated group ( Iv ) to the Incidence of malaria in the control group ( Ic ) . At eight weeks post primary challenge all surviving monkeys were rechallenged intravenoulsy with a higher dose of 15 , 000 P . vivax Brazil VII sporozoites . Monkeys that demonstrated patent blood infection were treated with quinine as above . Sera from monkeys were tested for the presence of CSP antibodies following immunization . Immulon 2HB plates ( Dynatech , VA ) were coated with 0 . 4 µg/ml VMP001 , N- and C-terminal regions of P . vivax CSP or 1 µg/ml of Type 1 peptide . Sera from a selected time point were also screened against PfLSA1 to determine if the control group generated antibodies against the transgene , and to ensure that there was no cross reactivity between the control and vaccine groups . Plates were blocked , incubated with serially diluted serum , followed by 1∶6000 dilution of a custom prepared goat anti-Aotus IgG ( CDC , Atlanta ) labeled with horseradish peroxidase ( HRP ) . The reaction was developed with ABTS and read after 60 min at A414 . ELISA titers are defined as the serum dilution that gives an optical density ( OD414 ) of 1 . 0 . GraphPad Prism version 5 . 0 was used for statistical analysis . Tests used for each analysis are mentioned along with the corresponding data .
Monkeys were immunized with either 15 µg ( low dose ) or 50 µg ( high dose ) of VMP001formulated with Montanide ISA 720 in combination with 200 µg of CpG 10104 , a TLR-9 agonist . This study was designed with an ultimate goal of assessing the efficacy of the VMP001 vaccine formulation . Adjuvants are known to induce innate immune responses which can affect the outcome of a challenge . Therefore , in order to rule out the role of non-specific immune responses we immunized a group of monkeys with P . falciparum LSA-1 that served as an unrelated malaria antigen , formulated in adjuvant . Monkeys were immunized three times , with a one month interval between the 1st and 2nd immunization , and a three month interval between the 2nd and 3rd immunizations . Animals were monitored following each immunization and no reactogenicity or adverse events were observed . None of the control monkeys that received PfLSA-1 generated any antibodies against VMP001 following immunization , however , robust anti-PfLSA1responses were detected in this group , while none of the VMP001 immunized sera showed any anti-PfLSA1 reactivity ( data not shown ) . Anti-VMP001 antibodies were detected in all monkeys immunized with VMP001 two weeks following the 1st immunization ( Figure 1 ) . The median titers at 2 weeks post 1st immunization were 14 , 581 and 67 , 695 for the low and high dose groups , respectively . These titers were boosted significantly at 2 weeks post 2nd immunization for both the low dose ( 144 , 302; p = . 02 ) and high dose ( 198 , 024; p = . 02 ) groups . Antibody titers declined over time and the titers were significantly lower in both immunized groups at 12 weeks after the 2nd immunization compared to the peak titers obtained two weeks after the 2nd immunization ( p = . 0006 and . 015 for the low and high dose groups , respectively ) . Titers showed a boost following the 3rd immunization and peaked 4 weeks post 3rd immunization . There was a significant difference between the titers at 12 weeks post 2nd immunization ( 28 , 940 ) and 4 weeks post 3rd immunization ( 140 , 367; p = 0 . 006 ) for the low dose group and the high dose group ( 78 , 163 at 12 weeks post 2nd immunization and 291 , 622 at 4 weeks post-3rd immunization; p = . 005 ) . The peak titers post the 2nd and 3rd immunizations were not significantly different ( Figure 1 ) . The antibody titers for both the low and high dose groups were similar at 2 weeks post each immunization ( Figure 2 ) . The median titers ranged from a low of 14 , 581 at 2weeks post 1st immunization for the low dose group , to a high of 198 , 024 for the high dose group at 2 weeks post 2nd immunization . There were no statistical differences in antibody titers between the groups at each time point indicating that a strong adjuvant overcomes antigen dose effects . Following intravenous inoculation of sporozoites , six of the eight ( 75% ) PfLSA-1 control monkeys became parasitemic , with a tight pre-patent period ranging from day 14 to 18 post-challenge ( Figure 3A ) . In fact , five of the six infected monkeys had demonstrated patent infection by day 16 . The remaining monkey became positive on day 18 . Six of the eight monkeys in each of the two vaccinated groups did not demonstrate patent infection . In both the low and high dose groups 25% monkeys in each group ( 2/8 ) became parasitemic with a prepatent period ( day 17 and 28 for the low dose group; day 16 and 17 for the high dose group ) that was delayed in one animal compared to the controls . The median prepatent period for the control group was 15 . 5 days , 22 . 5 days for the low dose group and 16 . 5 days for the high dose group . Total vaccine efficacy , calculated by comparing the ratio of infected monkeys in the vaccinated vs . the control group , for each group was 66 . 7%; and the p values , based on the Mantel-Cox Log-rank test , for the time to parasitemia were . 02 and . 03 for the low and high dose groups respectively . Following treatment , all surviving monkeys were rechallenged intravenously eight weeks post primary challenge with a higher dose of 15 , 000 P . vivax Brazil VII sporozoites . This allowed for an approximately two-week interval between the clearance of any primary infections before the rechallenge . Seven of seven control monkeys became parasitemic between days 11 and 15 ( Median day 13 ) ( Figure 3B ) . Six of six rechallenged monkeys in the high dose group became parasitemic between days 12 and 22 ( Median day 14 . 5 ) and seven of eight of the low dose group became parasitemic between day 12 and 18 ( Median day 13 . 5 ) . One monkey in the low dose group did not become parasitemic during the 33 day post-rechallenge observation period . As expected , the control animals , immunized with an unrelated malaria antigen , PfLSA-1 , did not induce antibodies to VMP001 ( Figure 4 ) after the 3rd immunization . However , following primary challenge , these animals generated antibodies to CSP , as demonstrated by the presence of antibodies to VMP001 ( Median titer: 797 ) . The titers were boosted approximately 10-fold following rechallenge to a median titer of 7 , 670 . The differences between the prechallenge and post-challenge anti-PvCSP ( VMP001 ) titers in the PfLSA-1 group were statistically significant ( Figure 4 ) . Since the prechallenge anti-VMP001 titers in the vaccinated monkeys were already high , the differences in titers post-challenge were not significant . Serum from immunized monkeys was further analyzed to assess the breadth of antibody responses generated . At 4 weeks post 3rd immunization ( 2 weeks pre-challenge ) all monkeys demonstrated presence of antibodies to the N- , Central Repeat as well as the C-terminal region of P . vivax CSP . As with VMP001 ( Median titers of 140 , 367 and 291 , 622 for the low and high dose groups respectively ) there was no significant difference between the antibody titers against the N-term ( Median 7 , 277 vs . 22 , 188 ) , Type 1 repeat peptide ( Median 66 , 120 vs . 25 , 297 ) and C-term regions ( Median 77 , 111 vs . 179 , 947 ) . The data was reanalyzed by pooling the protected ( n = 12 ) and non-protected ( n = 4 ) monkeys from both the high and low dose groups ( Figure 5 ) . There were no statistical differences between the VMP001 ( Median 214 , 659 vs . 156 , 247 ) , N-term ( Median 14 , 979 vs . 11 , 952 ) and C-term ( Median 116 , 291 vs . 202 , 126 ) titers . The titers to the Type 1 repeat peptide were significantly different between the protected ( Median 97 , 841 ) and non-protected ( Median 21 , 517 ) groups ( p = 0 . 03; 2-tailed Mann-Whitney U test ) , indicating that high anti-repeat antibodies may be a correlate of protection . This is in accordance with data from P . falciparum vaccine studies where anti-repeat antibodies have been shown to be the strongest correlate of protection in humans [16] . As stated earlier , all but one of the vaccinated monkeys that were protected following primary challenge ( Figure 6A ) lost their protection upon rechallenge ( Figure 6B ) . Nevertheless , when data from the low- and high-dose groups were combined following rechallenge and compared to the controls , there was still a significant difference between the immunized and control groups ( p = 0 . 036 ) ( Figure 6B ) . In comparison , following primary challenge the combined vaccinated groups had a p-value of 0 . 004 compared to the controls ( Figure 6A ) . We therefore compared the antibody titers in the prechallenge sera from the monkeys that were protected prior to the primary challenge to the titers in sera from those that were not protected following rechallenge . Sera were drawn two weeks before the primary- , or re-challenge . Here too , the only significant differences were found between the anti-repeat antibodies . As shown in Figure 6C the median anti-Type 1 repeat antibody titer in the protected monkeys was 97 , 841 at the time of primary challenge compared to 14 , 822 ( p = 0 . 02 ) in the monkeys that were not protected following rechallenge . Thus , the loss of protection in the previously protected monkeys could be attributed to waning of the anti-repeat antibodies . Antibodies to VMP001 also declined ( Median 214 , 569 vs . 97 , 296 ) but the decrease was not statistically significant . There was no association between antibody titers and time to patency in the monkeys that became parasitemic .
Due to financial , logistical , and scientific challenges , efforts to develop vaccines for P . vivax have been disproportionately slower compared to that for P . falciparum . In addition to the limited number of preclinical studies in non-human primates that we describe above , a recent review by Reyes-Sandoval and Bachman [17] highlights the meager clinical portfolio of P . vivax vaccines and the current status of vivax research and funding . In sharp contrast to P . falciparum , so far , only one efficacy study has been conducted in humans ( clinical trials registration number NCT01157897 , Bennett et . al . in preparation ) . There is a need to test products to down-select candidate ( s ) that can be advanced to the clinic . Towards this end , we produced a recombinant CSP-based vaccine and assessed its immunogenicity in rhesus monkeys using two different TLR4 agonists as adjuvants . In a study using a synthetic TLR4 agonist , glucopyranosyl lipid A ( GLA ) , formulated in oil ( GLA-SE ) , we demonstrated the safety and immunogenicity of VMP001 . The vaccine induced potent humoral responses directed to the different regions of the molecule , and a cellular response predominantly directed to the C-terminal region of the protein [18] . In another study , rhesus monkeys were immunized with soluble recombinant protein ( VMP001 ) , as well as its particulate counterpart CSV-S , S which is a fusion between VMP001 and the hepatitis B surface antigen to compare the immunogenicity of the two formulations in AS01 , an adjuvant approved for human use [19] . Both the soluble and particulate vaccines induced strong humoral and cellular immune responses , but with some qualitative and quantitative differences between the two formulations . The present study was undertaken to test the immunogenicity , as well as efficacy , of VMP001 in Aotus monkeys . We immunized monkeys with two different doses of antigen formulated in Montanide ISA 720 and CpG 10104 . Montanide ISA 720 is a water-in-oil emulsion constituted of a nonmineral metabolizable oil and a mannide monooleate surfactant [20] . CpG 10104 is a B-class 24-mer synthetic oligonucleotide with a fully phosphorothioate backbone containing 5 CpG motifs ( Coley Pharma ) . These unmethylated CpG DNA motifs are recognized by mammalian TLR9 and are known to activate dendritic cells , macrophages and B cells [21] . Both Montanide ISA 720 and CpG have been used in clinical studies in humans with malaria antigens ( www . clinicatrials . gov ) and thus , have a possible path forward for the clinical evaluation of vaccine candidates . Therefore , we chose to evaluate our vaccine in the presence of these adjuvants to take advantage of their synergistic roles in inducing a functionally-relevant immune response . Due to the possible effect of the innate immune responses induced by this combination of adjuvants on the outcome of a parasite challenge , we included a control arm in which the monkeys were injected with the adjuvant formulated with an unrelated P . falciparum antigen that does not exist in P . vivax . The final vaccine efficacy was estimated by comparing it to the outcome of the control group . Due to the fact that 25% of the control monkeys did not become parasitemic , our calculated vaccine efficacy for both the low and high dose groups singly , and in combination , following the primary challenge was 66 . 7% with a total of 12 out of 16 immunized monkeys remaining free of blood stage parasites . Both the low- and high-dose groups induced robust immune responses to the whole recombinant protein ( VMP001 ) , as well as the N-term , repeat- and C-term regions of the protein . However , only anti-repeat antibodies showed a statistical association with protection . The twelve protected monkeys had statistically significant antibody response that was 4 . 5-fold higher compared to the response in the non-protected monkeys . In contrast , the protected and non-protected monkeys had differences of 1 . 7-fold , or less , in anti-VMP001 , N- and C-term antibody titers . While similar comparisons with the various components ( the entire immunogen , or the C-term region ) have not been published for RTS , S the only subunit malaria vaccine to show consistent protection in humans , evaluation of anti-repeat antibodies has been shown to be a strong correlate of protection [16] . In a previous study , passive transfer of a mouse monoclonal antibody directed against the repeat region of the P . vivax CSP was shown to protect 4/6 monkeys [22] . However , immunization with PvCSP recombinant proteins failed to induce protective anti-repeat antibodies [22] [10] . The strength of our protective correlate to the anti-repeat titers is further bolstered by the results of the rechallenge experiment . Since only one monkey stayed protected after the rechallenge , it was not possible to compare the protected vs . non-protected groups post-rechallenge . Therefore , we compared the anti-Type 1 repeat antibody titers in the monkeys that were protected post-primary challenge vs . those that were not protected post-rechallenge . The antibody titers showed a significant decline from a median of 97 , 841 pre-primary challenge to 14 , 822 pre-rechallenge . This 6 . 6-fold drop in titers was statistically significant ( p = . 02 ) and explains the loss of protection in the previously protected monkeys . We would be remiss in not stating , as pointed out by one of the reviewers , that T cell responses were not evaluated in this study . In addition , this study was not designed to test effectiveness against in situ hypnozoites and it is possible that this vaccine may or may not be effective against established hypnozoites , and thus may need to be reformulated and associated with other antigens . In summary , we demonstrate VMP001 vaccine-induced protection in the presence of strong immunomodulators , Montanide ISA 720 and a TLR9 agonist . The presence of one , or both , of these components may have caused a qualitative shift in the immune response to drive the production of high-titers of anti-repeat antibodies to a level that resulted in protection . The results of this study—using the nonhuman primate model of the P . vivax Brazil VII strain in A . nancymaae—can help serve as a benchmark for down-selection of adjuvant formulations in future studies with the current vaccine antigen , or while designing new antigens to serve as a vaccine .
|
Plasmodium vivax is responsible for causing malaria in large parts of the globe , including regions with temperate climates not suited for the transmission of other Plasmodium species . In addition , P . vivax has the propensity to form dormant forms , known as hypnozoites , that can remain latent for weeks to months and reactive periodically to cause recurrent infections . Prevention of P . vivax malaria , more than any other form , will require a vaccine-based intervention due to limitations in treatment options . To this end , we tested the efficacy in non-human primates , of a vaccine based on circumsporozoite protein , a preerythrocytic stage antigen , of P . vivax . Aotus monkeys were immunized with clinical-grade antigen , combined with two immunomodulators , and then challenged with P . vivax sporozoites . Following challenge 66 . 7% of monkeys were protected . Analysis of serum samples indicated that protection was associated with antibodies to the central repeat region of the molecule , and that protection was lost upon waning of these antibodies . This is the first report demonstrating that active immunization with a recombinant protein can lead to complete protection in monkeys following sporozoite challenge , while also demonstrating a protective associate . Our data can help serve as a benchmark for down-selection of future vaccine formulations for P . vivax .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"humoral",
"immunity",
"infectious",
"disease",
"immunology",
"clinical",
"immunology",
"immunity",
"biology",
"and",
"life",
"sciences",
"immunology",
"humoral",
"immune",
"response"
] |
2014
|
Protective Efficacy of a Plasmodium vivax Circumsporozoite Protein-Based Vaccine in Aotus nancymaae Is Associated with Antibodies to the Repeat Region
|
Neurons receive excitatory or sensory inputs through their dendrites , which often branch extensively to form unique neuron-specific structures . How neurons regulate the formation of their particular arbor is only partially understood . In genetic screens using the multidendritic arbor of PVD somatosensory neurons in the nematode Caenorhabditis elegans , we identified a mutation in the ER stress sensor IRE-1/Ire1 ( inositol requiring enzyme 1 ) as crucial for proper PVD dendrite arborization in vivo . We further found that regulation of dendrite growth in cultured rat hippocampal neurons depends on Ire1 function , showing an evolutionarily conserved role for IRE-1/Ire1 in dendrite patterning . PVD neurons of nematodes lacking ire-1 display reduced arbor complexity , whereas mutations in genes encoding other ER stress sensors displayed normal PVD dendrites , specifying IRE-1 as a selective ER stress sensor that is essential for PVD dendrite morphogenesis . Although structure function analyses indicated that IRE-1’s nuclease activity is necessary for its role in dendrite morphogenesis , mutations in xbp-1 , the best-known target of non-canonical splicing by IRE-1/Ire1 , do not exhibit PVD phenotypes . We further determined that secretion and distal localization to dendrites of the DMA-1/leucine rich transmembrane receptor ( DMA-1/LRR-TM ) is defective in ire-1 but not xbp-1 mutants , suggesting a block in the secretory pathway . Interestingly , reducing Insulin/IGF1 signaling can bypass the secretory block and restore normal targeting of DMA-1 , and consequently normal PVD arborization even in the complete absence of functional IRE-1 . This bypass of ire-1 requires the DAF-16/FOXO transcription factor . In sum , our work identifies a conserved role for ire-1 in neuronal branching , which is independent of xbp-1 , and suggests that arborization defects associated with neuronal pathologies may be overcome by reducing Insulin/IGF signaling and improving ER homeostasis and function .
During their development neurons can form complex dendritic branching patterns . The specific arbor morphologies of different neuron types are thought to have evolved to mediate the acquisition and processing of distinct inputs [1] . Defective arbor morphologies in brain neurons are a common cellular symptom in many neuropsychiatric and neurodegenerative diseases [2–4] . Dendritic arbor growth requires the accurate orchestration of numerous cellular events that occur concomitantly at a distance from the neuronal cell body and integrate dramatic membrane extension , local protein translation and processing , vesicular transport , shifts in cytoskeleton dynamics and elevated metabolic activity . How neurons control these various processes at the genetic and molecular level remains only partially understood [5–7] . Our understanding of dendrite arbor morphogenesis has advanced significantly through the study of peripheral mechanosensory arbor development in the fly Drosophila melanogaster and the nematode Caenorhabditis elegans . In Drosophila , larval da ( dendrite arborization ) neurons are grouped into four classes according to the degree of arbor complexity [6 , 8] . Screens for da dendrite defects have identified many genes that control arborization , such as transcription factors , membrane receptors and their ligands , integrins , vesicular transport factors and cell adhesion molecules [5 , 6 , 8] . Recently , the polymodal sensory neuron PVD in C . elegans with its characteristic multidendritic arbor has become a model neuron for the study of dendrite morphogenesis [9 , 10] . Genetic work on the formation of the repetitive PVD menorah-shaped dendritic units has identified several genes not implicated before in dendrite morphogenesis , including roles for fusogens [11] , the LRR-type receptor dma-1 [12] , the fam151 family member mnr-1/menorin [13 , 14] , the secreted leukocyte cell-derived chemotaxin 2 lect-2/chondromodulin II [15 , 16] , and the furin-like protease kpc-1 [17–19] . The endoplasmic reticulum ( ER ) is the primary cellular site for secretory protein and lipid biosynthesis , both of which are essential for proper cellular function . In agreement , disruption of ER homeostasis is associated with pathologies such as neurodegenerative disorders [20–22] . To prevent deleterious outcomes of perturbed ER homeostasis , a cellular program called the Unfolded Protein Response ( UPR ) is triggered at times of increased load on the ER ( i . e . ER stress ) to ensure that ER homeostasis is retained regardless of the dynamic nature of cellular demand [23] . In mammalian cells ( as well as in C . elegans ) , the UPR is composed of three pathways that are initiated by distinct ER stress sensors: inositol-requiring enzyme 1 ( IRE1 ) , protein kinase RNA ( PKR ) -like ER kinase ( PERK ) and activating transcription factor-6 ( ATF6 ) . IRE1 is the most ancient of the UPR sensors , being conserved from yeast to humans , and bears both kinase and ribonuclease activities [24] . Upon its activation , IRE1 undergoes autophosphorylation and oligomerization into multimers [25 , 26] . In its oligomeric state it removes an intron from xbp-1 ( X-box binding protein-1 ) mRNA through unconventional splicing allowing the translation of an activated form of the XBP-1 transcription factor . This activated transcription factor induces the expression of chaperones , ERAD components and other ER auxiliary factors to rebalance ER capacity [27 , 28] . The UPR , and specifically the ire-1/xbp-1 arm of the UPR , is important even under normal physiological conditions ( i . e . in the absence of induced ER stress ) , as perturbations in this pathway impair secretory protein metabolism [29] . Additional xbp-1 independent functions of ire-1 are also known . These include activation of the cell death machinery [30–32] , induction of autophagosomes [33] , induction of a cellular anti-oxidant response [34] and degradation of ER-localized mRNAs that encode secreted and membrane proteins through the RIDD ( regulated Ire1-dependent decay ) pathway [35] . Recent in vitro work using the yeast Ire1 has suggested that RIDD activity can be mediated by IRE1 even in its monomeric state [36] . Here , we demonstrate that IRE1’s role in dendrite arborization is conserved during evolution from C . elegans to mammals . We show that in C . elegans ire-1 deficiency elicits a secretory block in the PVD neuron that interferes with the targeting of the DMA-1 receptor to the plasma membrane , strengthening similar results by Wei et al . [37] . We further reveal that this trafficking block , which does not occur in xbp-1 mutants , can be bypassed by reducing insulin/IGF1 signaling to restore normal arbor architecture . Altogether , this work assigns a conserved role for IRE-1 function in neuronal development and demonstrates that activation of alternative ER homeostasis-promoting pathways can counteract and prevent the deleterious consequences of compromised ER homeostasis on neuronal development .
The dendrites of the polymodal somatosensory PVD neurons are stereotypically patterned , by the consecutive branching of secondary , tertiary , and quaternary branches from primary dendrites that exit the PVD cell body on either side both in an anterior and in a posterior direction ( Fig 1A ) . In concordance with a recent report [37] , we isolated a mutant allele of ire-1 , which encodes the C . elegans homolog of the inositol requiring enzyme 1 ( IRE1 ) in a screen for genes required for PVD morphogenesis [13] . The ire-1 ( dz176 ) allele changes Glycine 708 , a residue that is located in an alpha helix of the kinase domain and conserved from yeast to humans ( Fig 1A and 1B ) . The PVD phenotypes were shared with another missense allele ( zc14 ) , which also changed a perfectly conserved G723 in the kinase domain , as well as the deletion allele ok799 ( Fig 1A–1C ) . Mutant phenotypes were transgenically rescued by both a wild-type copy of ire-1 ( S1A–S1C Fig ) and expression of a cDNA pan-neuronally or in PVD neurons , but not in the intestine or hypodermis ( skin ) ( S1D Fig ) . These findings complement previous mosaic studies [37] , and together strongly argue for a cell-autonomous function of ire-1 . As reported previously , ire-1 mutants formed dendrites with quaternary branches only in the area proximal to the PVD cell body , and gradually became less developed as their distance from the cell body increased both anteriorly and posteriorly ( Fig 1A ) [37] . We extended these observations in morphometric analyses , which showed a reduction both in the number and aggregate length of secondary , tertiary and quaternary branches in the presumptive ire-1 ( ok799 ) null mutant ( Fig 1D and 1E; S1E Fig ) . In addition , we discovered a self-avoidance defect in ire-1 mutants , where adjacent tertiary dendrites failed to retract upon mutual contact , thereby eliminating the characteristic gaps between them ( Fig 1F ) . Since the role of IRE1 in dendrite patterning in mammals had never been addressed before , we investigated whether IRE1 serves an evolutionarily conserved function during dendrite patterning in mammals . We studied dendrite morphogenesis in dissociated rat hippocampal cultures and measured changes in dendrite length and complexity after 8 and 12 days in vitro ( DIV ) , a time period during which dendrites undergo dynamic growth ( Fig 2A ) . During this time window , neurons in culture were either treated with vehicle or the IRE1-specific inhibitor 4μ8C [38] . Vehicle-treated neurons showed the expected developmental increase in total dendritic branch length from 8 to 12DIV ( Fig 2B; 8DIV , 897 +/- 50 μm , n = 42 vs . 12DIV+veh , 1308 +/- 75 μm , n = 42; p<0 . 0001 ) . In contrast , neurons treated from 8DIV onwards with 50 μM of the specific IRE1 RNAse inhibitor 4μ8C did not show this developmental increase in total dendritic branch length ( Fig 2B , 8DIV vs . 12DIV+4μ8C , 893 +/- 66 μm , n = 41; p = 0 . 99 ) . In neurons treated with the IRE1 inhibitor , there was a trend towards fewer dendrite tips as compared with vehicle treated neurons ( Fig 2C; 12DIV+veh , 22 +/- 1 . 1 tips vs . 12DIV+4μ8C , 18 +/- 1 . 6 tips; p = 0 . 078 ) , consistent with the correlation of shorter total dendritic branch length with fewer dendritic tips [39] . IRE1 inhibition did not restrict the developmental increase in average dendritic branch length , supporting the notion that this aspect of dendrite differentiation was not impaired ( Fig 2D; 8DIV , 26 +/- 1 . 1 μm vs . 12DIV+veh , 36 +/- 1 . 6 μm; p<0 . 0001; 8DIV vs . 12DIV+4μ8C , 33 +/- 2 . 3 μm , p<0 . 001 ) . Importantly , the effect of IRE1 inhibition was specific to higher order branches and did not alter the number of primary branches ( Fig 2E ) , similar to the effects observed in PVD dendrites of ire-1 mutants in C . elegans ( Fig 1D and 1E ) . We conclude that IRE-1 serves a conserved role in dendritic dendrite morphogenesis under normal physiological conditions , and in the absence of external induction of ER stress . Thus , our studies in rats and C . elegans provide the first example for a conserved developmental function of the ire1 stress sensor in neural development . This adds to a growing body of literature , based on knockout approaches in mice , that show functions for the unfolded protein response during liver development [40–42] , as well as in the development of antibody-producing B cells [43] and secretory cells of the pancreas [44] . The IRE-1 protein is composed of a luminal unfolded protein sensor domain and a cytosolic bifunctional active site , comprising a kinase and a ribonuclease domain . Our mutants in the kinase domain , as well as mutants identified by Wei et al . [37] suggested that both domains may be important for ire-1 function . To further investigate this notion , we generated mutant versions of IRE-1 , defective in each of these domains , and conducted rescue experiments in ire-1 mutants . We found that expression of a mutant where the ER luminal domain , thought to serve as an unfolded protein receptor [45] , had been replaced by red fluorescent protein ( mCherry ) , rescued PVD morphology in ire-1 mutant animals , although not as efficiently as the full length transgene ( Fig 3D ) . In contrast , expression of the ribonuclease-deficient mutant version IRE-1K853A , affecting a highly conserved residue in the putative nuclease active site [46] and completely devoid of any detectable xbp-1 splicing activity ( Fig 3E ) , failed to rescue PVD arbor morphology in ire-1 mutant animals ( Fig 3D ) . This implies that IRE-1 nuclease activity is necessary for dendrite morphogenesis . In addition , we expressed IRE-1L589G , an IRE-1 transgene harboring a mutation analogous to the yeast ire1p mutation L745G , which alters the specificity of the ATP binding site in the kinase domain of the protein [47] . In contrast to the yeast studies , the ribonuclease activity of IRE-1L589G appeared intact in an xbp-1 splicing assay ( Fig 3E ) . We found that expression of IRE-1L589G also rescued the arborization defects in PVD sensory dendrites ( Fig 3D ) . Collectively , our rescue studies show that PVD development requires the ribonuclease activity of IRE-1 . This conclusion is consistent with the defective PVD arborization phenotype previously observed in ire-1 ( wy762 ) mutants , in which a conserved residue in the endoribonuclease domain of the protein has been altered [37] . In addition , kinase activity is likely required , because three mutant alleles of ire-1 ( dz176 , zc14 , this study; wy782 , [37] ) that result in substitutions of distinct conserved residues in the kinase domain of the protein , display a defective PVD arborization phenotype . In addition to IRE-1 , metazoans have at least two distinct additional sensors of ER stress , the pek-1/PERK kinase and the atf-6/ATF6 transcription factor [48] . Interestingly , PVD development proceeds normally in pek-1/PERK and atf-6/ATF6 mutants , demonstrating that they do not individually serve a critical role in PVD dendrite morphogenesis , and pointing at a unique function of IRE-1 ( Fig 3A and 3C ) [37] . To gain insight in the downstream effectors of IRE-1 signaling , we focused on the processing of xbp-1 mRNA through unconventional splicing by IRE-1 , the best known activity of IRE-1 [27 , 28] . Interestingly , two different xbp-1 mutant alleles , zc12 and tm2457 displayed a PVD arbor that was indistinguishable from wild type animals ( Fig 3B and 3C ) , suggesting that IRE-1 can function through xbp-1-independent activities in patterning PVD dendrites . Known xbp-1-independent functions of ire-1 include activation of the TRAF2 and JNK kinase signaling cascade [30 , 49] , and degradation of ER-localized mRNAs that encode secreted and membrane proteins through the RIDD ( regulated Ire1-dependent decay ) pathway [35] . We found that PVD arborization remained normal upon depletion of the C . elegans TRAF2 homolog trf-1 or concomitant depletion of all three C . elegans jnk-1-related kinases ( Fig 3F ) suggesting that neither pathway plays non-redundant roles in PVD morphogenesis . To directly explore whether RIDD is the mechanism by which IRE-1 controls PVD arborization , we sought another way to maintain RIDD activity in ire-1 mutants while compromising xbp-1 splicing activity . A mutation in the yeast yIre1 protein , R1087D , uncouples the two nuclease activities of ire1p in yeast by impairing xbp-1 splicing while leaving RIDD activity intact [36] . The analogous mutation in worms , IRE-1R882D , failed to rescue PVD architecture ( Fig 3D ) . Since xbp-1 function is not required for PVD morphogenesis , we suggest that C . elegans IRE-1R882D mutant protein may not discriminate between xbp-1-related and unrelated nuclease activities . Thus , among the known xbp-1 independent activities of IRE-1 , RIDD remains the most likely to mediate PVD dendrite arborization . This conclusion supports experiments where mosaic knock out of an essential xrn-1 RNA endonuclease , believed to be part of the RIDD pathway , produced low penetrance defects in PVD neurons [37] . Recently , it was shown that even under normal growth conditions ( i . e . without artificially-induced ER stress ) ire-1 mutants display defects in the metabolism of secretory proteins [29] . One central protein located on the cell membrane of PVD and essential for proper dendrite branching is the DMA-1 leucine rich repeat transmembrane receptor [12] . In concordance with a recent report [37] we found that a DMA-1::GFP reporter primarily localized to the cell body of ire-1 mutant animals ( Fig 4B and 4E ) . In contrast , in wild-type animals , the DMA-1::GFP reporter localized both to the cell body as well as to the entire PVD arbor , throughout the primary , secondary , tertiary and quaternary branches ( Fig 4A and 4E ) . Importantly , although the primary branch of the PVD dendrite is always present and extends along the body of ire-1-deficient animals ( Fig 1A ) , DMA-1::GFP expression was restricted to the cell body and was not detected on the plasma membrane of the primary branch of PVD ( Fig 4B ) . This suggests that DMA-1 is specifically required for patterning of the secondary , tertiary and quaternary branches in ire-1-deficient animals . This further suggests that the DMA-1::GFP localization defect in ire-1 mutants precedes the PVD patterning defect . Collectively , these observations suggest that DMA-1 fails to shuttle properly through the secretory pathway , resulting in patterning defects of higher order branches of the PVD dendrite . Intriguingly , xbp-1 mutants displayed completely normal DMA-1::GFP staining of the entire PVD dendrite similar to wild-type animals ( Fig 4C ) , and did not accumulate DMA-1::GFP in the soma like ire-1 mutants ( Fig 4E ) . The proper localization of DMA-1::GFP in xbp-1 mutants contrasts with its mislocalization in ire-1 mutants ( Fig 4A–4C and 4E ) , but is consistent with the PVD arborization architecture seen in the respective mutants ( Fig 3A–3C ) . This finding was surprising , given that loss of xbp-1 has been shown to perturb ER homeostasis and interfere with secretory protein metabolism [29] . Based on the trafficking defects in ire-1 mutant animals , and the finding that the basal activity of the UPR in PVD itself is largely dependent on DMA-1 expression , it has been suggested that the failure of DMA-1::GFP to reach the plasma membrane is a consequence of a folding challenge of DMA-1 itself [37] . However , we point out that although in many cases functional xbp-1 is also required for the trafficking and maturation of other secreted and transmembrane proteins [29 , 50] , it was not required for DMA-1 trafficking to the plasma membrane , and PVD morphogenesis is normal in xbp-1 mutants . Thus , DMA-1 can traffic to the plasma membrane and support PVD dendrite morphogenesis even under the unfavorable proteostatic conditions in the ER of xbp-1-deficient animals . This suggests that DMA-1 may not have an intrinsic tendency to fold improperly and that the DMA-1 trafficking defect is more likely a reflection of a general overload and perturbed function of the ER in the PVD neuron that lacks ire-1 . A possible explanation for the differences between ire-1 mutants and xbp-1 mutants is that ER homeostasis and function is less compromised in xbp-1-deficient animals compared to ire-1-deficient animals [29] . Thus , the DMA-1-dependent activation of the UPR in PVD suggested by Wei et al . [37] may be an indirect consequence of DMA-1 promoting dendrite morphogenesis and expansion , both of which require the synthesis of membranes proteins and lipids and impose a significant biosynthetic load on the ER . This ‘capacity model’ is also consistent with the observation that overexpression of spliced xbp-1 or its target , the ER-localized heat shock protein HSP-4/BiP/grp78 can bypass the requirement for ire-1 and rescue the morphological defects and the DMA-1::GFP secretion defects in ire-1 mutants [37] . Altogether , our report adds to a growing number of recent works delineating an xbp-1-independent branch of the ire-1 pathway [29 , 35 , 40 , 51 , 52] . If the failure to form menorahs in ire-1 mutants is a result of a block in the secretory pathway in the PVD neuron then conditions that release the secretory block in ire-1 mutants should restore PVD arborization . One way to overcome the secretory block in ire-1 mutants is by activating the FOXO transcription factor DAF-16 , which is inhibited by the insulin/IGF-1 signaling ( IIS ) pathway [52] . Indeed , reducing IIS in animals through a mutation in their daf-2 gene , the only insulin-like growth factor receptor in C . elegans , resulted in reduced accumulation of DMA-1::GFP in the cell body and redistribution to the plasma membrane of PVD in ire-1 mutants ( Fig 4D and 4E ) . Consistent with the restored localization of DMA-1::GFP expression pattern in the PVD neuron , we found that PVD dendrite morphogenesis defects in ire-1; daf-2 double mutants were completely reversed and PVD arbors of double mutants were indistinguishable from wild type animals ( Fig 4F ) . This finding was further corroborated by morphometric analyses . We found that the reduced length of secondary , tertiary and quaternary branches in ire-1 mutants was suppressed upon reduced DAF-2/IIS signaling ( Fig 4G ) . Moreover , this suppression was largely ( although not completely ) dma-1-dependent , because dma-1 appeared epistatic in a dma-1; ire-1; daf-2 triple mutant ( Fig 4G ) . The physiological consequence of reduced DAF-2/IIS signaling , including improving ER homeostasis in ire-1-deficient animals [52] , in many cases depends on the activation of the transcription factor DAF-16/FOXO . We found that daf-16; ire-1; daf-2 triple mutant animals showed the same frequency of PVD defects as ire-1 single mutants , indicating that the suppression of defects in ire-1 mutants by loss of daf-2 insulin signaling was entirely dependent on daf-16 activation ( Fig 4F ) . In other words , the defects in dendrite morphogenesis of ire-1 mutants can be rescued by compromising DAF-2/IIS signaling in a daf-16/FOXO-dependent manner . Our finding that trafficking of a DMA-1::GFP reporter is restored in daf-2/IIS mutants suggests that ( 1 ) different approaches can be used to relieve the secretory block in ire-1 mutants , and ( 2 ) are consistent with previous observations that attenuation of IIS can result in favorable effects on proteostasis , ER homeostasis , organismal health and survival in C . elegans , as well as other organisms [53 , 54] . Similarly , activation of the IIS regulated transcription factor DAF-16/FOXO3A in ire-1-deficient cells can bypass the requirement of the canonical ire-1/xbp-1 pathway for the maintenance of ER homeostasis , and improve both ER homeostasis and restoration of normal secretory protein trafficking in worms and mammalian cells [52] . Thus , our findings may provide a mechanistic explanation for observations in several studies showing that neurons grow and function better under reduced IIS conditions [55–57] , and expands this notion to include dendritic arbor morphogenesis . Since the improvement on DMA-1::GFP trafficking and dendrite morphology were dependent on activation of the DAF-16/FOXO transcription factor , the activation of this pathway by alternative cues including starvation as well as a variety of cytotoxic stresses ( e . g . heat-shock and oxidative stresses ) , which directly or indirectly activate DAF-16 , hold the potential to recover PVD dendrite morphogenesis in the absence of a properly functioning UPR . In summary , our results establish that the function of the IRE-1 UPR sensor in neuronal patterning is conserved from invertebrates to mammals . Our findings demonstrate that promoting ER homeostasis , e . g . by reducing IIS , can overcome morphological defects in neuronal patterning . This underscores the importance of discovering and investigating new approaches that can bypass excessive ER stress . Given the conservation of the role of the UPR in dendrite branching and morphogenesis from C . elegans to mammals , as well as the conservation of the proteostasis-promoting effects of the IIS pathway , these findings may offer novel approaches for treatment of neurodegenerative disorders .
Worms were grown on OP50 Escherichia coli-seeded nematode growth medium plates at 20°C . Strains used in this work include: N2 ( wild type reference ) , ire-1 ( dz176 ) , ire-1 ( ok799 ) , ire-1 ( zc14 ) , xbp-1 ( tm2457 ) , xbp-1 ( zc12 ) , pek-1 ( ok275 ) , atf-6 ( ok551 ) , daf-2 ( e1370 ) , daf-16 ( mu86 ) , trf-1 ( nr2014 ) , kgb-1 ( um3 ) kgb-2 ( gk361 ) jnk-1 ( gk7 ) . PVD neurons were visualized by the integrated transgene wdIs52 ( Is[F49H12 . 4::GFP] ) . Transgenic strains for cell-specific rescue were established by injecting the respective plasmids at 5–10 ng/μl together with rol-6 ( su1006 ) or Pttx-3::mCherry ( labeling the interneuron AIY ) at 50 ng/μl as an injection marker into ire-1 ( ok799 ) ; wdIs52 . The PVD::DMA-1::GFP translational fusion was a kind gift of K . Shen ( Stanford U , California ) . For a complete strain list see Supporting Information . The ire-1 cDNA was amplified with gene specific primers from a N2 mixed stage cDNA sample and cloned KpnI/SphI downstream of the Pttx-3promB regulatory element [58] . For the cell specific heterologous rescue the ire-1 cDNA was placed under control of the Pdpy-7 ( hypodermis-specific ) , Pmyo-3 ( muscle ) , Pges-1 ( intestine ) , Prgef-1 ( pan-neuronal ) or Pser-2prom3 promoter ( PVD/OLL specific ) . For further details see Supporting Information . On day 1 of adulthood , animals were collected for RNA extraction , purification and reverse transcription , using random 9-mers and standard protocol . A primers set encompassing the noncanonical intron of the xbp-1 transcript was used , giving rise to two PCR products of amplified spliced and unspliced xbp-1 transcript ( primers: 5’- TCCGCTTGGGCTCTTGAGATGTTC-3’ and 5’-TGTCGTCGTCGGAGGAGAGGATCG- 3’ ) . PCR products were visualized on a 2% agarose gel stained with ethidium bromide . Images of immobilized animals ( 1–5 mM levamisol , Sigma ) were captured using either a Zeiss Axioimager Z1 Apotome at 40X , where Z stacks were collected and maximum projections were used for imaging of dendrites , or with a CCD digital camera using a Nikon 90i fluorescence microscope at 20X magnification . For DMA-1::GFP signal quantification the NIS element software was used to quantify sum and mean fluorescence intensity as measured by intensity of each pixel in the selected area . Hippocampal neurons were prepared from rats at E18 as previously described [59] with modifications . In brief , dissected hippocampi were incubated in 0 . 05% trypsin at 37°C for 20 minutes ( Invitrogen 25300054 ) and plated at a density of 60 , 000 cells per 12 mm coverslip coated with poly-l-lysine ( Sigma P1274 ) . Cells were incubated in a cell culture incubator maintained at 37°C with 5 . 0% CO2 . Cytosine arabinoside ( ara-c , Sigma C1768 ) was added at a final concentration of 2 μM at 2 days in vitro ( DIV ) to prevent glia cell overgrowth before being replaced with Neurobasal without ara-c at 4DIV . Neurons were transfected at 5-6DIV using Lipofectamine LTX and Plus Reagent ( ThermoFisher ) . For cytoplasmic labeling of neurons to visualize dendrites , 60 , 000 cells were transfected with 0 . 25 μg pCAGGS-mCherry [60] . Neurons were treated at 8DIV with IRE-1 RNAse inhibitor 4μ8C ( EMD Millipore 412512 ) . 4μ8C was first dissolved in DMSO ( Invitrogen ) and diluted in supplemented Neurobasal . Diluted 4μ8C at 100 μM was added to cultures at 1:1 with conditioned neural media with final concentrations of 50 μM 4μ8C and 0 . 5% DMSO . Additional 4μ8C was added to neurons at 10DIV resulting in a final concentrations of 37 . 5 μM 4μ8C and 0 . 5% DMSO . Vehicle neurons were treated in identical ways using media containing DMSO only . At 8 or 12 DIV coverslips with neurons were quickly washed two times with PBS , followed by fixation for 15 min with 4% PFA / 4% sucrose in PBS at RT . Cells were blocked and permeabilized with 3% horse serum and 0 . 05% Triton X-100 in PBS for 1 h at RT . Cells were incubated with antibodies against mCherry ( Rockland 600-401-379 ) at 4°C overnight . The next day cells were washed three times with PBS and labeled with AlexaFluor-conjugated secondary antibodies ( Invitrogen; 1:500 ) for 1 hr at RT . Cells were washed three times with PBS , stained with DAPI , and mounted on slides Aqua-Mount mounting media ( Thermo Scientific ) . Tiled images of dendritic arbors were acquired using a Keyence BZ-X710 Fluorescence Microscope equipped with a Nikon 60X oil-immersion 1 . 40 NA objective . Merged composite images of the individually acquired tiled images were generated using Keyence software . Dendritic arbors were traced using the NeuronJ plugin for ImageJ [61] . Total dendritic branch length was calculated as the sum of the length of all dendrites . Average dendrite branch length is average length of each dendritic branch excluding primary dendrites , as primary dendrite lengths are highly variable across neurons . All images were acquired and all analysis was performed with the experimenter blind to conditions . Data analysis was performed using GraphPad Prism 6 . Error bars represent the standard error of the mean ( SEM ) of at least 3 independent experiments . P values were calculated using the unpaired Student's t test , or one-way ANOVA with the Tukey correction for multiple comparisons ( GraphPad Prism 6 ) .
|
Sensory neurons sample their environment through highly branched structures termed dendritic arbors or trees . The precise patterning of dendritic arbors is important for the proper functioning of the nervous system , and evidence indicates an involvement of sensory neurons in neuropsychiatric disease such as autism spectrum disorders . The unfolded protein response is a cellular process that ensures and maintains a functional protein-folding environment in the cell’s endoplasmic reticulum , and is compromised in a number of neurodegenerative conditions . Recently , this process has also been implicated in dendrite patterning . We discovered that the function of the unfolded protein response in dendrite patterning is evolutionarily conserved from the roundworm C . elegans to mammals . Specifically , dendrites in both worms and mammals require the function of a conserved enzyme with both kinase and ribonuclease activity , which acts as a sensor for the unfolded protein response . Importantly , we find that loss of this enzyme can be bypassed by reducing the signaling through the insulin-like growth factor receptor . Our findings reveal a new way of bypassing defects in the unfolded protein response during dendrite development .
|
[
"Abstract",
"Introduction",
"Results",
"and",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"invertebrates",
"medicine",
"and",
"health",
"sciences",
"caenorhabditis",
"gene",
"regulation",
"regulatory",
"proteins",
"endoplasmic",
"reticulum",
"cell",
"processes",
"dna-binding",
"proteins",
"neuroscience",
"animals",
"animal",
"models",
"physiological",
"processes",
"developmental",
"biology",
"caenorhabditis",
"elegans",
"model",
"organisms",
"homeostasis",
"experimental",
"organism",
"systems",
"transcription",
"factors",
"morphogenesis",
"neuronal",
"dendrites",
"cellular",
"structures",
"and",
"organelles",
"research",
"and",
"analysis",
"methods",
"animal",
"cells",
"proteins",
"gene",
"expression",
"biochemistry",
"cellular",
"neuroscience",
"neuronal",
"morphology",
"cell",
"biology",
"secretory",
"pathway",
"physiology",
"neurons",
"genetics",
"nematoda",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"organisms"
] |
2017
|
Reduced Insulin/Insulin-Like Growth Factor Receptor Signaling Mitigates Defective Dendrite Morphogenesis in Mutants of the ER Stress Sensor IRE-1
|
The circadian oscillator , an internal time-keeping device found in most organisms , enables timely regulation of daily biological activities by maintaining synchrony with the external environment . The mechanistic basis underlying the adjustment of circadian rhythms to changing external conditions , however , has yet to be clearly elucidated . We explored the mechanism of action of nicotinamide in Arabidopsis thaliana , a metabolite that lengthens the period of circadian rhythms , to understand the regulation of circadian period . To identify the key mechanisms involved in the circadian response to nicotinamide , we developed a systematic and practical modeling framework based on the identification and comparison of gene regulatory dynamics . Our mathematical predictions , confirmed by experimentation , identified key transcriptional regulatory mechanisms of circadian period and uncovered the role of blue light in the response of the circadian oscillator to nicotinamide . We suggest that our methodology could be adapted to predict mechanisms of drug action in complex biological systems .
The synchronization of physiological rhythms with the external environment is important for nearly all organisms . Circadian oscillators are internal timing devices that produce rhythms with a period of about 24 hours to regulate a wide range of biological processes . Circadian rhythms maintain synchrony with the daily timing of light and dark cycles resulting from Earth’s rotation by constantly integrating environmental signals . This process of synchronization is called entrainment . Studying the mechanisms that dynamically adjust circadian period and phase , therefore , is critical to understand the control of daily biological activities . In Arabidopsis thaliana , the circadian oscillator consists of a complex circuit of highly connected transcriptional regulators . Together , they coordinate global transcript accumulation and diverse biological processes , such as photosynthesis , hormone signaling , hypocotyl elongation and plant-pathogen interactions [1 , 2 , 3 , 4 , 5] . The light perception of the circadian oscillator is conferred by a suite of photoreceptors . The photoreceptors are split into two classes: phytochromes ( principally PHYA and PHYB ) , that primarily sense the red portion of the spectrum [6] and cryptochromes ( CRY1 and CRY2 ) that are sensitive to blue light [7 , 8 , 9] . Recent studies have demonstrated a role for metabolism in regulating and entraining the circadian oscillator of Arabidopsis thaliana . The primary metabolite sucrose accelerates the circadian oscillator ( i . e . , reduces its period ) through regulation of the morning expressed gene PSEUDO RESPONSE REGULATOR ( PRR ) 7 [10] , while GIGANTEA ( GI ) has been identified as a necessary sucrose-signaling mediator in the dark [11] . Another metabolite , nicotinamide ( NAM ) , a breakdown product of nicotinamide adenine dinucleotide ( NAD ) , causes long period of the circadian oscillator in all organisms tested [12 , 13] . The mode of action of NAM is uncertain: various mechanisms having been proposed , including inhibition of the production of the Ca2+-agonist cyclic adenosine diphosphate ribose ( cADPR ) , inhibition of polyADP ribose polymerases and histone modifications [12 , 13 , 14] . The goal of this study was to use NAM as a tool to identify the processes responsible for a change in circadian period , which might be required for circadian entrainment and homeostatic adjustment [15 , 16 , 17] . The discovery of drug modes of action , however , is still a costly and inefficient process , which often requires considerable prior knowledge of a biological system and/or a vast amount of data in several experimental condition ( e . g . mutations ) . A major difficulty is the complex ripple effect of treatments affecting transcriptional networks . Large sections of the transcriptome can be differentially expressed , despite not being directly affected by the treatment ( off-targets ) ( Fig 1 ) . Due to the large number of feedback loops involved in a complex and relatively small Gene Regulatory Network ( GRN ) , such as the circadian clock , this effect is particularly significant as a perturbation anywhere in the network typically strongly affects all molecular concentrations . Additionally , as the perturbations induced by NAM in the circadian clock are intrinsically related to changes in circadian period , a large part of the transcripts will typically be differentially expressed . Thus , Differential Expression ( DE ) analysis , the traditional approach used to identify the mechanisms that alter biological behavior in response to drugs , environmental signals or genetic lesions [20] , will usually fail to identify the small number of genes central to the biological perturbation . The main reason is that DE only performs statistical analysis of changes in gene expression levels [21 , 22] . As an alternative to the DE analysis , we devised a modeling framework that identifies and characterizes differentiated regulatory dynamics between genes to capture key mechanisms involved in NAM-induced perturbations in the circadian system of Arabidopsis . The rationale behind this approach is that not only genes , but also their interactions , are affected by a drug . This reasoning is further supported by [18 , 23 , 24] , which highlight the fact that drugs and diseases mechanisms should be regarded as network instead of gene-centric perturbations . We designed our modelling strategy so that it could be applied to scarce data without the need to cover extensive experiments or to make prior assumptions of network dynamics . In particular , we consider only gene expression data with and without NAM . On one hand , complex nonlinear models have the potential to capture the dynamical relationships between genes with great precision . A successful application of Michaelis-Menten dynamics to reverse engineer network topology from multiple experiments , circadian time-series data is presented and compared to state-of-the-art strategies in [25] . Alternatively , a community-driven comparison of ( non ) linear approaches ( e . g . mutual information-based , Bayesian networks , random forests , neural-networks , etc . ) for the inference of ( non-circadian ) gene regulatory networks has been achieved in [26 , 27] . On the other hand , high model complexity can lead to overfitting ( fitting the noise instead of the dynamics ) without sufficient data or detailed knowledge such as network topology , types of nonlinear interactions , or potentially some of the model parameters ( e . g . Hill coefficients ) . As for non-model based methods , such as [28 , 29 , 30 , 31] , it is not clear how they can be used to compare subtle changes in dynamics caused by perturbations , and pinpoint the source of those perturbation . We developed , therefore , a systematic and scalable dynamical modeling framework named Dynamical Differential Expression ( DyDE ) . DyDE uses a black box-type modeling approach to reverse-engineer simple yet consistent and comparable gene regulatory dynamics from time-series data . In addition , it does not use any prior information and , hence , it is unbiased towards prior knowledge of network topology and dynamics . By comparing changes in both topology and subtle dynamic modifications of regulatory mechanisms , we were able to considerably narrow down potential targets of NAM in the circadian clock . The findings predicted by DyDE are experimentally tested and demonstrate the role of the circadian gene PRR7 as a key regulator of dynamics adjustment of the circadian clock . In addition , TIMING OF CAB EXPRESSION 1 ( TOC1 ) and the interplay between PRR7 and PSEUDO RESPONSE REGULATOR 9 ( PRR9 ) are identified as the main mediators of the circadian system response to NAM . The modeling insights also identified alterations in CRY2 dynamics resulting from the NAM treatment . Therefore , we also investigated the role of blue light in the circadian period change of NAM-treated plants . In particular , we found that blue light regulates circadian oscillations of [Ca2+]cyt through a NAM-sensitive pathway . These new perspectives contribute to the understanding of the mechanistic details underlying the regulation of period of circadian oscillators . Overall , the results suggest that DyDE is a useful tool to generate reliable hypothesis from time-series data for the identification of drug targets in complex biological systems .
To assess whether genes are regulated by the circadian oscillator , most methods take advantage that circadian regulation of transcript abundance resemble a sinusoid . To estimate circadian period of the regulation of a particular transcript , the main idea is to find the sinusoid that most closely matches its abundance over time [32 , 33] . However , in NAM-treated plants the changes in abundance of circadian-regulated transcripts have a considerable number of non-sinusoidal profiles ( S1 Fig ) . To overcome this problem , we devised a learning approach based on pseudo-sinusoidal functions to properly assess the rhythmicity and the corresponding circadian period of signals from gcRMA normalized microarray data of NAM treated plants . To infer period , phase and amplitude , linear trends are eliminated by removing the best straight-line fit and pseudo-sinusoidal functions are fitted to each signal to minimize the 2-norm error . Pseudo-sinusoidal functions account for many signals that are periodic but not sinusoidal . Pseudo-sinusoidal functions are constructed by joining together two sinusoids with different periods . Hence , a complete oscillation of a pseudo-sinusoidal function consists of the first sinusoid ( of period p1 ) in the first half-oscillation , and the second sinusoid ( of period p2 ) in the second half-oscillation ( Fig 2A ) . The resulting period of the pseudo-sinusoidal function is defined as p=p1+p22 . This can be expressed by: S={A*sin ( 2πp1*t+φ1 ) , t∈[0 , p12]A*sin ( 2πp2* ( t−p12+p22 ) +φ1 ) , t∈[p12 , p12+p22] where A is a scaling factor that accounts for the amplitude of the signal and φ1 is the phase of the signal . The algorithm searches possible combinations of p1 and p2 to minimize the least square distance between pseudo-sinusoidal functions and the data . We allowed periods p1 and p2 to vary between 12 and 36 hours . A perfect sinusoid gave a high fit for the wild-type background dataset . We found that three periodic signals were highly represented in the dataset . In particular , those with p1 , p2 equal to: p/2 , p/2 ( pure sinusoid ) ; p/2+3 . 8 , p/2–3 . 8 ( p1 is greater than p2 ) ; and p/2–7 . 3 , p/2+7 . 3 ( p1 is smaller than p2 ) ( Fig 2A ) . We used a logistic regression framework to generate a probabilistic discriminative model that estimates the probability of a gene to be rhythmic given its time course data . In this case , the classification problem only contains two classes: rhythmic ( C1 ) and arrhythmic ( C2 ) . For each transcript , a set of 8 features x = {X1 , X2 , … , X8} is computed and empirically believed to be crucial to distinguish between rhythmic and arrhythmic transcripts . The features were computed from 2 signals: the first signal ( A ) corresponds to the average of replicates and ( B ) being a single replicate for which the L2-norm error with the best fitted pseudo-sinusoidal function is lower than for the other replicate . The following features were computed: ratio of power in the 18–32 hours frequency range ( of ( A ) and ( B ) ) , L2-norm of the error to the best fit of pseudo sinusoidal function ( of ( A ) and ( B ) ) , the variance of the power spectrum ( of ( A ) and ( B ) ) and the amplitude of the best fitted pseudo-sinusoidal function ( of ( A ) and ( B ) ) . The log of the ratio of probabilities between the two classes , also known as the log odds , is given by [34]: ln ( p ( rhythmic|x ) p ( arrhythmic|x ) ) =ln ( p ( C1|x ) p ( C2|x ) ) =ln ( σ1−σ ) =logit ( σ ) The goal of the logistic regression is to estimate σ for a linear combination of the Xn features such that: logit ( σ ) =b0+b1X1+b2X2+⋯+bnXn The weights bi of the independent variables Xi were estimated using the mnrfit function in MATLAB . The algorithm is initially trained with a mix of 100 rhythmic and 100 arrhythmic transcripts randomly chosen from the dataset and visually inspected to show clear ( ar ) rhythmicity . Finally , the decision boundary was set so that if p ( C1|x ) >0 . 5 , the gene was classified as rhythmic , and vice versa . Our approach , therefore , is inspired by the patterns observed in the dataset but not strictly constrained to pure cosine shapes . With the inclusion of the S function , we allow the search for asymmetric signals , which represent a large part of the transcriptome . A main distinction with the previously introduced algorithms , therefore , is the data-specific , learning approach devised to allow for a wider range of periodic signals . However , this offers additional advantages such as a dedicated way to handle noise between replicates , or the information in the frequency domain of the signal , which are both learned from the data . Comparison of performances with standard periodicity assessment tools is shown on S2 Fig . Like most biological systems , circadian clocks have a large number of feedback loops . Hence , a perturbation anywhere in the network typically affects all nodes ( in this case , their molecular concentration and time profiles ) , which makes the problem of inferring the entry point of a perturbation hard using DE analysis . We proposed , instead , that key mechanisms involved in NAM-induced perturbations in the circadian system of Arabidopsis can be captured by identification and comparison of regulatory dynamics before and after the perturbation occurred . Assume that a perturbation , such as NAM , changes the regulatory dynamics between two genes ( e . g . by binding to a transcription factor ) while leaving intact the rest of the system . Due to feedback interconnections , all the clock genes would change their expression , which , in turn , would change the expression of all circadian genes . While thousands of genes change their expression , only one regulatory link was actually affected . Our goal is to find this link ( or links , in case of multiple perturbation entry points ) . To achieve this , we developed DyDE that looks for changes in links , instead of nodes . DyDE uses Linear Time-Invariant ( LTI ) models , a type of black box model , to systematically capture the dynamics underlying the biochemical mechanisms of circadian gene regulation , without relying on a priori knowledge of the system or extensive database . Such models benefit from a rich theory and a well-established collection of tools that makes the analysis of its dynamical properties straightforward , as contrast to detailed mechanistic models . In addition , the estimation of the parameters of such models is reliable and computationally efficient . The description of biological mechanisms of the Arabidopsis circadian clock from time-series data by LTI models has been studied in [35] . More recently , the performances of such linear modelling approach to reverse engineer the clock topology were compared for two extensively used Arabidopsis oscillator models [36] and confirmed that the majority of oscillator links can be represented by simple linear dynamics . However , the use of LTI models to detect dynamical perturbation in the gene regulatory network resulting from chemical treatments is novel . The first step of DyDE consists of uncovering dependencies and quantifying dynamics between genes with LTI models . Our mathematical framework estimates a collection of Single Input-Single Output ( SISO ) models between pairs of genes to characterize the system dynamics . The limited number of available time points restricted the modelling of SISO systems to first and second order models . Overall , second order systems did not improve significantly the fitness of models and resulted in a considerable increase of false positives ( overfitting ) . Hence , in this analysis of the circadian system , we restrict the model order to one . Mathematically , the dynamics between two genes can be represented as: dy ( t ) dt=au ( t ) −by ( t ) +c where u ( t ) and y ( t ) represent the time series of the regulatory gene and the regulated gene , respectively . In addition , b y ( t ) corresponds to the degradation rate of gene y , a u ( t ) corresponds to the influence of u ( t ) on the rate of y ( t ) and c is a constant offset . System identification is performed using the function ‘pem’ implemented in MATLAB to minimize the prediction error [37] . The model has a total of three parameters ( a , b , and c ) , leading to efficient solutions . We chose a subspace initialization algorithm since it performed similarly as randomizing initial conditions–for the vast majority of models ( 99% ) , the final solution was identical with either method . This suggests that the chances of being trapped into a local minimum are negligible . The estimation of parameters requires low computational time: a single system between a pair of genes is typically identified within few seconds ( Intel Core i5 ) . This modeling is independently repeated for all available pairwise genes , where each gene takes its turn as being an input and then an output to another gene . This modeling approach , therefore , generates a large amount of SISO LTI models ( n2−n models , where n corresponds to the amount of genes , and self-regulation is not considered ) to describe the system . Each potential link between two genes is validated if the corresponding model reproduces the dynamics involved with a sufficient degree of precision , which is characterized by a high goodness of fit , defined as: fitness=100* ( 1−∑k=1N ( y−y^k ) 2∑k=1N ( y−y¯ ) 2 ) where y is the validation data , y¯ is the average value of the validation data , and y^k is the estimated output . MATLAB function compare can be used to compute the fitness of the model . A fitness equal to 100% corresponds to a perfect identification . The choice of such metric is motivated by the dependency of noise towards the abundance of gene expression . When the distance of the true data points towards the mean is large ( represented by the denominator in the above equation ) , the fitness conveniently penalizes less the error term , which lies in regions where the intrinsic noise involved in the gene expression is potentially the largest . The second step consists in identifying the effect of a treatment , NAM in our case , on the biological network . While a treatment might affect the abundance of many transcripts , only a few links are affected , as depicted in red in Fig 1 . Hence , checking whether links are affected before and after perturbation can potentially lead to the finding of the entry point of the treatment . For this purpose , two cases are of particular interest . First , a link between two genes is validated in the untreated system alone ( i . e . it is not possible to find a combination of a , b and c so that the model in the treated system provides a good match with the data anymore ) . Second , a link is validated in both systems , but the way one gene regulates the other may change; this is a much subtler change in the dynamics of the link . The latter case requires us to compare the dynamics between both links . Here , we use a rigorous and well-established tool from engineering known as the nu-gap [38] . Originally developed to address the stability properties of closed loops systems defined in the same feedback loop , the nu-gap essentially measures the distance , from a perturbation point of view , between linear models . This property is particularly relevant in the context of circadian clock networks , which consist in regulatory networks with several feedback loops . This then facilitates us to determine the significance of the dynamical change of a link between experimental conditions . The nu-gap returns a value between 0 to 1 , quantifying whether the models are similar or very different , respectively . [39] have suggested that values above ~0 . 2 could be used to infer the main target of a perturbation . The nu-gap is computed using the gapmetric function in MATLAB . It should be applied to all models that have been estimated in both networks . If the signals are concentrated around a particular range of frequencies ( such as oscillating signals ) , the gap should be measured ‘locally’ around that range of frequencies only , since they dominated the model estimation in Step 1 . Next , we explain the key ideas behind DyDE through a small number of genes in the Arabidopsis circadian oscillator . For example , the following model considers TOC1 as an input and PRR9 as an output . d[PRR9]tdt=a[TOC1]t−b[PRR9]t+c where b represents the strength of activation or repression induced by TOC1 on the expression rate of PRR9 , and a corresponds to the degradation rate of PRR9 . These parameters are estimated by minimizing the prediction error from the untreated time-series for both TOC1 and PRR9 . In this case , we found a model in good agreement with the data ( 57% fitness ) , suggesting that indeed TOC1 regulates PRR9 ( Fig 3A ) . Moreover , the model demonstrates that the rate of change of the concentration of PRR9 is proportional to the concentration of TOC1 . Note that the other way around ( i . e . , PRR9 regulating TOC1 ) could not be established since the respective model has a low goodness of fit ( 16% , Fig 3B ) . These results are consistent with the literature [40] . Hence , we would then establish a link from TOC1 to PRR9 , but not the other way around ( Fig 3C ) . Then , a model is estimated between TOC1 and PRR9 from the NAM-treated time-series . From the untreated and treated time-series alone , it is unclear whether the link dynamics have changed ( Fig 3D ) . The optimal model parameters , however , have significantly changed . A nu-gap of ~0 . 5 confirms that indeed the link has been affected . This result indicates that there is large perturbation in the regulatory dynamics that links TOC1 to PRR9 , which , therefore , should be considered as a strong candidate for being an entry point for NAM in the system . If true , knocking down either TOC1 or PPR9 would therefore lead to NAM no longer affecting the clock . This analysis is then repeated for all common links between untreated and treated plants .
We considered a total of 17 known clock genes: CCA1 , LHY , PRR9 , PRR7 , PRR5 , RVE8 , GI , TOC1 , ZTL , ELF4 , ELF3 , PHYA , PHYB , CRY1 , CRY2 , CHE and PRR3 . However , the core oscillator genes ZTL , ELF3 , PHYB , CRY1 , PRR3 and CHE were identified as non-rhythmic in the presence of NAM , which was confirmed by visual inspection ( S1 Fig ) . Hence , these genes are excluded from the modeling of NAM targets as they cannot be contributing to the rhythmic dynamics of the remaining oscillator components that are measured in the presence of NAM . As a first step , we computed models for all available pairs of the clock genes for both conditions , totaling 220 SISO models ( 110 in untreated and 110 in NAM ) . We kept only those models with good agreement with the data , i . e . above a fitness threshold . On one hand , the user-defined threshold has to be set large enough to reliably capture the dynamics involved between genes , and provide the nu-gap analysis with comparable models . On the other hand , the threshold has to be set sufficiently low to consider enough gene-to-gene relationships to detect a dynamical perturbation in the network . Here , the fitness threshold was set to 46% as we noted that below this threshold , the amount of unknown regulations dramatically raised ( S3 Fig; S3 Table ) . In total , 70 regulatory links were retained for untreated plants and 55 links for NAM-treated plants between the 11 clock genes . The untreated models describe 70% of the known regulatory pathways among these 11 genes ( S3 Table; S3 and S4 Figs [40] ) . 64% of which , had the expected activation or inhibition effect . These numbers are remarkable , taking into account the model simplicity , and confirms that the majority of clock links can be represented by simple linear dynamics [35 , 41 , 42] . In particular , 28 links were present in the untreated samples but not in the NAM-treated samples . These 28 links form a network from now on referred to as “regulation loss” network , which captures the links abolished by NAM . In addition , 42 links are present in both conditions which form a network , so called “common” network that is common to both treated and untreated plants ( S3 Table ) . We used the nu-gap to identify those links among the common network whose dynamics were significantly affected by NAM . Fig 4A and S4 Table depict the comparison of the dynamics of each link with the nu-gap . All regulatory interactions are somehow affected by the treatment , which is expected from the interconnected circadian network . Let us then consider the highest nu-gap values , which are associated with the following links: TOC1 to PRR9 ( 0 . 5 ) , those originating from CRY2 to ELF4 ( 0 . 47 ) , LHY ( 0 . 42 ) and RVE8 ( 0 . 37 ) and PRR9 to CRY2 ( 0 . 35 ) . Interestingly , the only inferred interaction originating from CRY2 that does not seem affected connects to TOC1 ( nu-gap of 0 . 06 ) . These results suggest that a major dynamical change is induced to CRY2 in the dynamical response of the circadian clock to NAM . In addition , the largest nu-gap value suggests that the causality within the time course data of TOC1 and PRR9 has changed significantly differently towards the treatment , as compared to the other parts of the circadian network . We then used a standard network topology metric to identify the genes that are central to the drastic changes in dynamics captured by the regulation loss network . This topology metric accounts for the connectivity of a gene , i . e . the number of its incoming and outgoing links . This measure is estimated for each gene of the regulation loss network . As an example , PRR7 has six incoming links and nine outgoing links for untreated plants . The connectivity of PRR7 in untreated plants is then equal to 15 . Among those , only six of were present in NAM-treated plants . PRR7 , therefore , has a connectivity of nine in the regulation loss network , which correspond to a loss of 60% of its connectivity from untreated to NAM treated plants . As a result , CCA1 ( 61% ) , PRR7 ( 60% ) , TOC1 ( 57% ) exhibit the highest connectivity drop ( Fig 4B; S5 Table ) . This result identifies the biological functions of CCA1 , PRR7 and TOC1 as being highly affected by NAM in the regulation of the circadian clock . DyDE , therefore , identifies the regulatory dynamics of TOC1-CRY2-CCA1-PRR7 as being predominately affected by NAM as a result of both nu-gap and connectivity analysis . Accordingly , the strong emergence of the blue light receptor CRY2 in the nu-gap analysis suggests that nicotinamide alters the regulation of the interactions between light signaling and the circadian oscillator . These findings are further examined through mutant analysis and single wavelength light experiments . To test the predictions that TOC1 , CRY2 , CCA1 and PRR7 are associated with the effect of NAM on the circadian oscillator , we experimentally investigated the sensitivity of circadian mutants to NAM . All mutants responded to NAM with increased circadian periods , with the exception of two independent lines of the same T-DNA insertion allele of PRR7 , which were insensitive ( prr7-3 p > 0 . 95; prr7-11 p > 0 . 95 Fig 5A and 5B; S5 Fig; S6 Table ) . The insensitivity of prr7-11 to NAM was confirmed by measuring circadian rhythms of leaf movement ( S6 Fig ) . prr7-11 was not affected by NAM at any tested concentration , contrasting with a dose-dependent effect of NAM on circadian period in other prr mutants and associated backgrounds ( R2 > 0 . 9; Fig 5C ) . In contrast , toc1-2 and TOC1-ox had significantly greater responses to NAM than wild type ( Fig 5A; S6 Table ) . These results support our predictions that NAM induces dynamical changes specifically to PRR7 and TOC1 . No dramatic changes of period , however , were observed for cry2-1 and cca1-11 , suggesting that these might not contribute directly to the response to NAM . Finally , derived from the nu-gap analysis , the possible change in the dynamical behavior of PRR9 in mediating the effect of NAM on the clock was evaluated with a prr7-3 and prr9-10 double mutant . prr7-3 and prr9-10 had an epistatic interaction , with the double mutant responding to NAM by a 5 . 3 ± 1 . 6 h increase of period , more than either the insensitive prr7-3 or the oversensitive prr9-10 alone ( Fig 5A ) . The epistasis of prr9-10 to prr7-3 was confirmed at all concentrations of NAM tested ( Fig 5C ) . The mutant analysis did not confirm the modeling dynamical perturbation of CRY2 in the response to NAM . However , CRY2 is one of a pair of cryptochrome blue light photoreceptors and so mutant analysis might not be the most appropriate tool to investigate the role of the blue light photoreceptor . To investigate further we also investigated the role of blue light in the response to NAM using monochromatic light conditions . High frequency measurements of the circadian promoter fusions PRR9:LUC , PRR7:LUC , TOC1:LUC , CCA1:LUC , LHY:LUC and GI:LUC were collected in the presence or absence of 20 mM nicotinamide under constant blue or red light ( S7 Fig ) . In the absence of blue light , NAM was without effect on the circadian period or amplitude of CCA1:LUC ( Fig 6A ) and other promoter:luciferase fusions ( Fig 6B ) . This demonstrates that input pathways associated with blue light are sensitive to NAM . Under blue light exposure , all promoter:luciferase fusions considered had an increase in period in the presence of NAM ( Fig 6B ) . Under red light exposure , the period response was either negligible ( PRR9:LUC , CCA1:LUC , LHY:LUC , GI:LUC ) or negative ( PRR7:LUC , TOC1:LUC ) . These results suggest that blue light increase the response of circadian period to NAM , while red light decrease its responsiveness . Having previously proposed that the effects of NAM on the circadian system are due to the inhibition of the production of the Ca2+-agonist cADPR [12] , we tested if the response to NAM of prr7-11 is due to altered Ca2+ signaling . We investigated , therefore , the inhibitory effects of NAM on circadian [Ca2+]cyt oscillations in prr7-11 and in light signaling mutants in red and blue light . 20 mM NAM was equally effective in abolishing circadian rhythms of [Ca2+]cyt in both Col-0 , prr7-11 and prr7-3 prr9-10 ( Fig 6C ) . This suggests either that there are multiple sites of action of NAM or that PRR7 is downstream of the effects of NAM on [Ca2+]cyt . In constant blue light , there were robust oscillations of [Ca2+]cyt in plants with functional CRY1 photoreceptors , being abolished in cry1 and , cry1cry2 but unaffected by cry2 , phototropins and Phy loss-of-function mutants ( Fig 6D , S8 Fig ) . Under blue light , NAM abolished [Ca2+]cyt oscillations but did not reduce oscillations further in cry1 or cry1cry2 ( Fig 6D , S9 Fig ) . High amplitude oscillations of [Ca2+]cyt were dependent on blue light because in constant red light , [Ca2+]cyt increased early in each cycle but without a subsequent decrease ( Fig 6E; S8 Fig ) . This red light-induced increase in [Ca2+]cyt was dependent on PHYB ( S8 Fig ) . To examine the role of PHYB further we measured [Ca2+]cyt in PhyB-ox ( Fig 6F , S8 Fig ) and determined that in these plants [Ca2+]cyt was rhythmic with a sinusoidal period of 25 . 0 ± 0 . 5 h in constant red light ( Fig 6F ) . NAM was without effect on [Ca2+]cyt in constant red light , even in the PHYB-ox background ( Fig 6E and 6F , S6 and S7 Figs ) demonstrating that blue light regulates circadian oscillations of [Ca2+]cyt through a NAM-sensitive pathway . This pathway appears to be required for the major oscillatory dynamics of [Ca2+]cyt . DyDE was further adapted to explore the rhythmic genome for additional targets for NAM and novel clock genes . For this purpose , models were computed between each pair of the 988 genes that were scored rhythmic in both untreated and NAM treated conditions , resulting in 2 million models corresponding to potential interactions . We selected the models that exhibit the highest goodness of fit ( over 80% ) in both untreated and NAM-treated plants to minimize the identification of erroneous interactions and computed their nu-gap value to investigate dynamics affected by NAM . As a result , out of ten , two models only were retained with a nu-gap > 0 . 2 . These models identified the regulation of AT5G35970 ( P-loop containing nucleoside triphosphate hydrolases superfamily protein ) by AT2G21860 ( violaxanthin de-epoxidase-like protein ) and the regulation of ATG21660 ( GRP7/CCR2 ) by AT1G78600 ( LZF1/BBX22 ) as being altered by NAM . The regulation of AT5G35970 by AT2G21860 may be important as AT5G35970 is identified by DyDE as being a hub regulated by four circadian oscillator genes ( S8 and S9 Tables ) . The second link is easier to explain because GRP7 along with GRP8 forms a slave oscillator driven by the circadian clock that regulates ABA responses [43] . GRP7 is an RNA binding protein regulated by ADP ribosylation [44] . As the enzymes that perform ADP ribosylation are inhibited by NAD , this could suggest a role for nicotinamide inhibiting ADP ribosylation of an oscillator or slave oscillator component . Then , the fitness threshold was released to 60% to further investigate novel clock components . For this purpose , we searched for those genes for which models can be computed from/to clock components ( S8 Table ) . Models with a nu-gap value above 0 . 2 were discarded as a consistency criterion . Finally , candidates were ranked according to their connectivity with the clock . As a result , 20 high potential genes were isolated ( S9 Table ) . The whole genome analysis of clock input and output hubs and the nu-gap analysis suggest interesting roles for previously characterized genes , including AT3G47500 ( CYCLING DOF FACTOR3 ) [45] , AT4G38960 ( BBX19 ) [46] , AT1G78600 ( BBX22 ) [47 , 48] , AT3G22840 ( CRY3 ) [49] , AT1G28330 ( DRM1 ) , AT2G33830 ( DRM2 ) [50 , 51] and uncharacterized genes including AT5G35970 .
|
Circadian oscillators are internal timing devices that produce rhythms with a period of about 24 hours to regulate a wide range of biological processes for nearly all organisms . Circadian oscillators adjust phase and period in response to external cues such as light and temperatures and internal cues such as metabolites to maintain synchrony with the daily timing of light and dark cycles resulting from Earth’s rotation . We investigated the dynamical response of the Arabidopsis thaliana circadian network to nicotinamide , a metabolite that lengthens the period of circadian rhythms , to uncover key transcriptional mechanisms required for the adjustment of daily biological activity . The identification of the mechanisms of drug response is challenging , as a complex cascading effect causes large section of the transcriptome to be differentially expressed , despite not being directly affected by the drug . To identify the source of the change in circadian period , we introduce a modelling strategy based on the identification and comparison of gene regulatory dynamics before and after the perturbation occurred . The Dynamical Differential Expression ( DyDE ) methodology uses a reverse engineering approach that favours both the identification of unknown Gene Regulatory Network ( GRN ) topology and the interpretation of the possible dynamical changes , without the need to cover extensive experiments or to make prior assumptions of network dynamics . Subsequently , we show that our methodology can reliably identify the source of a perturbation in complex regulatory systems such as the circadian network . The proposed mathematical framework is scalable and flexible , so that it can be applied to large datasets with scarce sampling .
|
[
"Abstract",
"Introduction",
"Methods",
"Results"
] |
[
"genetic",
"networks",
"gene",
"regulation",
"brassica",
"light",
"electromagnetic",
"radiation",
"circadian",
"oscillators",
"model",
"organisms",
"network",
"analysis",
"experimental",
"organism",
"systems",
"chronobiology",
"plants",
"arabidopsis",
"thaliana",
"research",
"and",
"analysis",
"methods",
"computer",
"and",
"information",
"sciences",
"animal",
"studies",
"gene",
"expression",
"physics",
"biochemistry",
"circadian",
"rhythms",
"genetic",
"oscillators",
"eukaryota",
"plant",
"and",
"algal",
"models",
"gene",
"identification",
"and",
"analysis",
"genetics",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"organisms"
] |
2019
|
Dynamical differential expression (DyDE) reveals the period control mechanisms of the Arabidopsis circadian oscillator
|
Recent computational and behavioral studies suggest that motor adaptation results from the update of multiple memories with different timescales . Here , we designed a model-based functional magnetic resonance imaging ( fMRI ) experiment in which subjects adapted to two opposing visuomotor rotations . A computational model of motor adaptation with multiple memories was fitted to the behavioral data to generate time-varying regressors of brain activity . We identified regional specificity to timescales: in particular , the activity in the inferior parietal region and in the anterior-medial cerebellum was associated with memories for intermediate and long timescales , respectively . A sparse singular value decomposition analysis of variability in specificities to timescales over the brain identified four components , two fast , one middle , and one slow , each associated with different brain networks . Finally , a multivariate decoding analysis showed that activity patterns in the anterior-medial cerebellum progressively represented the two rotations . Our results support the existence of brain regions associated with multiple timescales in adaptation and a role of the cerebellum in storing multiple internal models .
Behavioral and computational modeling studies , on the one hand , and neuroimaging studies , on the other hand , have greatly advanced our understanding of motor adaptation . In particular , recent behavioral and computational modeling studies have shed light on the temporal structure of motor adaptation by showing that motor behavior is well accounted for by the sum of multiple motor memory states with different timescales . For instance , models with two time constants can reproduce a number of adaptation phenomena such as anterograde interference , spontaneous recovery , and savings [1–4] . A model with a larger number of time constants can account for adaptation occurring at multiple timescales , e . g . , fatigue and aging [5] . In contrast , neuroimaging studies , using either functional magnetic resonance imaging ( fMRI ) [6–8] or positron emission tomography ( PET ) [9–12] , have investigated the spatial distribution of the neural correlates and plastic changes across different brain regions at specific times during and after adaptation , with the prefrontal cortex ( PFC ) , the posterior parietal cortex ( PPC ) , and the cerebellum consistently showing activation . The PFC mostly contributes to the early , but not the late , stage of adaptation , which is consistent with its role in spatial working memory and in attention and arousal at the onsets of target presentation [13 , 14] . The PPC is also important in the early stage of motor adaptation [6 , 9 , 15] , here again consistent with its role in working memory [13 , 14] , planning movements and early adaptation to a new visuomotor transformation [9 , 16 , 17] . The activity of the cerebellum increases in a later stage of visuomotor adaptation [6 , 15 , 18] and correlates with the degree of savings at transfer of learning [19] . Such activation is consistent with cerebellum learning from errors [20 , 21] , building internal models [22–24] , and storing multiple motor skills [25] . However , these modeling and neuroimaging studies have been conducted independently of each other . As a result , little is known about the neural correlates of the latent ( i . e . , nondirectly observable from the behavioral data ) motor memories at different timescales suggested by computational models . In particular , it is unclear whether the multiple motor memories proposed by the models reside within a single system that contains a distribution of possible timescales or in a finite set of qualitatively distinguishable neural systems [1 , 26] . In addition , because experimental behavioral data can be well accounted for by models with different number of time constants , it is unclear how many distinct memories the brain actually updates during a specific type of motor adaptation . Finally , it is unclear whether the neural substrates identified in the early and late phases of adaptation in previous fMRI studies map onto putative “fast” and “slow” processes suggested by the computational models . Here , we combined modeling and imaging approaches via a model-based fMRI study of the spatial and temporal distribution of multiple motor memories during adaptation . Subjects adapted to two opposing visuomotor rotations in short alternating blocks . We estimated the multiple memories via a multiple-timescale adaptation model that generalizes the two-state models to multiple states with a logarithmic distribution of time scales from seconds to hours , as in a previous study [5] . A model-based approach based on regression analyses of brain activity would have insufficient power to dissociate multiple memories that are highly correlated with each other . We therefore propose a novel two-step approach . In a first step , we conducted exploratory multiple single regressions with individual memories , which avoids the problem of multicollinearity . In a second step , we performed a sparse singular value decomposition ( SVD ) on the voxels identified in the first step , in order to select a small number of orthogonal components . As a result , we identified four characteristic networks , each associated with formation of different time-scales of memories .
The overall mean adaptation level of 21 subjects showed that fast adaptation occurred within task blocks , and slow adaptation occurred across the task blocks for each task ( Fig 2A ) . Forgetting across blocks , which can be observed by comparing the last trial of a task block and the first trial of the next block of the same task , gradually decreased across blocks . Thus , visual inspection of the behavioral data suggests the existence of multiple timescales in motor memory , initially dominated by faster memories and eventually dominated by slower memories . We modeled adaptation as the sum of multiple memory states with different time constants . As in a previous model [5] , a continuous distribution of timescales was approximated by multiple time constants . We defined 30 time constants , ranging from 2 s to ~92 . 6 h , in a logarithmic scale ( S1 Table ) . In addition , the contextual cue ( here the target color ) selects the states for a specific task , as in a previous study [2] . Thus , the motor output at each trial n is given by y ( n ) =∑k=130xk ( n ) Tc ( n ) , ( 1 ) where xk = [xk , 1 xk , 2] with time constant τk ( k = 1 , … , 30 ) . The contextual cue vector is c = [1 0]T for Task 1 and c = [0 1]T for Task 2 . This model thus assumes no interference between two tasks , i . e . , perfect switching ( please refer to S1 Text for the rationale of this choice of a model with no interference ) . The states were updated by the error feedback , e = f − y , where f is , at each trial , one of the two visuomotor rotations , 40° or −40° . Because the intertrial interval ( ITI ) was random ( see details in Materials and Methods ) , we modeled time decay as an exponential function of time [27] . The update equation from trial n to n+1 for the state of motor memory with time constant τk is thus given by xk ( n+1 ) =xk ( n ) e−T ( n ) /τk+βk⋅e ( n ) ⋅c ( n ) , ( 2 ) where T ( n ) is the ITI following trial n and βk is the learning rate . The learning rates depend inversely on the time constant τk , as follows: βk=rτkq , ( 3 ) where r and q are strictly positive free parameters . As a result of this relationship , states with smaller time constants decay faster but are more rapidly updated [5 , 28] . Using averaged data of 21 subjects in actual adaptation to Tasks 1 and 2 ( see details in Materials and Methods ) , the best-fitted parameters of the proposed model were r = 0 . 0333 and q = 0 . 201 . The model fit is shown in Fig 2A , and the 30 states of Task 1 used in the fit are shown in Fig 2B . The overall fit to the averaged data was excellent ( root mean squared error = 4 . 96° , R2 = 0 . 981 ) . In addition , the fit to individual subjects’ data was satisfactory overall , although the fit was only modest for some subjects ( mean ± SEM across subjects: R2 = 0 . 832 ± 0 . 189 ) . We thus used the regressors calculated from the averaged adaptation data for the subsequent fMRI analysis . Because the multiple memory states are highly correlated with each other , especially for larger time constants ( for instance , the correlation coefficient between the state with τ21 = 2 . 2 h and the state with τ30 = 92 . 6 h was 0 . 994 ) , we first entered the states as single regressors in independent univariate analyses of blood-oxygen-level dependent ( BOLD ) signal ( see Materials and Methods ) . Overall , this univariate model-based regression analysis revealed distinct patterns of regions for states with increasing time constants ( Fig 3A ) . The faster states ( τk , ranging from 2 . 0 to 4 . 6 s; k = 1 , 2 , and 3 ) correlated mainly with activity in large regions in the frontal and parietal cortices and with activity in regions in the posterior-lateral cerebellum ( see below ) . By contrast , the states with intermediate time constants ( τk , ranging from 2 . 1 to 87 . 9 min; k = 11 , … , 20 ) correlated with activity in a restricted area in the right anterior region of the inferior parietal lobe ( aIPL , indicated by a blue circle in Fig 3A ) , which is the most anterior of the intraparietal sulcus . Note that the aIPL activity was prominently found in the right hemisphere , contralateral to the left hand used to perform the task but that weak activity was found in the left hemisphere when the threshold was lowered ( see Discussion ) . The slower states ( τk , ranging from 2 . 2 to 92 . 6 h: k = 21 , … , 30 ) primarily correlated with activity in the anterior-medial cerebellum ( see below ) . Supplementary videos show the patterns of correlated regions in both hemispheres for all time constants and corresponding regressors ( S1 and S2 Videos ) . These patterns for each time constant were similar in both the 40° and the −40° condition . Table 1 summarizes prominent clusters for the small ( k = 1 ) , middle ( k = 16 ) , and large ( k = 30 ) time constants for Task 1 . Significant clusters were found to be related to the fastest state ( k = 1 ) in the parietal , frontal , and cerebellar regions when thresholded at p < 0 . 05 corrected for multiple comparisons ( family-wise error rate [FWER] ) throughout the brain at cluster level . Activity related to the middle state ( k = 16 ) was significant in the aIPL ( p < 0 . 05 ) when we applied small volume correction for the superior and inferior lobules , in line with previous studies reporting parietal contribution to visuomotor adaptation [9 , 15] . Activity related to the slowest state ( k = 30 ) was significant in the left and right cerebellum at p < 0 . 05 corrected for the entire brain at cluster level . Fig 3B shows regional difference in the cerebellum between the fast ( k = 1 ) and the slow ( k = 30 ) states . Regions related to the fast state ( blue ) distribute in the posterior-lateral parts of the cerebellar cortex . These parts are mainly in the left and right crus 1 ( see Table 1 ) , which is connected with the prefrontal-parietal networks [29] . Note that activity was found in crus 1 not only for the fastest ( k = 1 ) but also for relatively fast ( k = 2 , … , 7 ) states ( S1 Fig ) . By contrast , regions related to the slow state ( red ) distribute in the anterior-medial parts ( mainly lobule 6 and partly lobule 8 ) , which are connected with cerebral sensorimotor networks [30] . Results from the univariate regression analysis with the 30-state model provided highly redundant but rich information on possible brain regions related to formation of multiple motor memories . For comparison , we checked if the standard two-state model [1 , 28] could explain our behavioral data and brain activity ( see Materials and Methods ) . We found that the two-state model was a subset of the 30-state model . That is , the time constants of the fast process and the slow process in the two-state model were respectively 47 . 9 s and 1 . 5 h . These are very close to two time constants of the 30-state model , k = 9 ( 55 . 2 s ) and k = 21 ( 2 . 22 h ) , with correlation coefficients of the states between the two models being respectively 0 . 993 and 0 . 991 . Brain activity identified by the two-state model was essentially the same as that found by the states of k = 9 and k = 21 in the 30-state model ( S2 Fig ) . A possible concern with these univariate regression analyses is that the high-pass filtering with 128 s cut-off frequency , which is used in preprocessing to remove low-frequency noise due to scanner drift , could filter out the lower-frequency components in the cerebellum . An analysis of the frequency components of high-pass filtered BOLD signals using Fast Fourier Transform ( FFT ) shows that the frequency components in the cerebellum remained large enough to be correlated with the regressors with the slower time constants ( see S3 Fig ) . A second possible concern with this analysis is that the identified activity is not related to memory states , but to errors . This is an especially valid concern for the faster states , because the memory states of the fast components correlate with the errors used for updating adaptation , with high values at the initial stages and low values at the late stages of adaptation . In our regression analysis of memory states , we included parametric regressors associated with error as those of no interest , i . e . , as nuisance regressors for the decreasing effects of error-related activity on the estimation of memory-related activity . We verified that these error-associated nuisance regressors appropriately explained away the error in the regression analyses for the faster states ( see S2 Text and S4 Fig ) . The T-map shown in Fig 3 is redundant because of highly correlated regressors . We thus applied the sparse SVD to the T-value profiles of voxels as function of time constants to extract principal components ( see Materials and Methods for details ) . All the voxels that survived the voxel-level threshold p < 0 . 001 for at least one time constant in the initial exploratory regression analysis were included in this sparse SVD analysis . The SVD analysis decomposed the data matrix into the following three matrices: eigenvariates , eigenvalues , and eigenimages . For both Tasks 1 and 2 , a sparse SVD model with four components was selected via the model selection method of Bayesian Information Criterion ( BIC ) [31] . The contributions of these four components to the variance in the matrix of T-values were ( 54 . 26% , 38 . 94% , 6 . 32% , 0 . 47% ) and ( 51 . 38% , 39 . 72% , 8 . 35% , 0 . 55% ) for Tasks 1 and 2 , respectively . The first and second eigenvariates correspond to the fast states ( the second eigenvariate has notably large values for small time constants that rapidly approach zero around 2 min ) , while the third and fourth eigenvariates represent the slow and middle states , respectively ( Fig 4A and 4B ) . Because a similar pattern was observed in corresponding eigenimages for Tasks 1 and 2 ( see S5 Fig and S1 and S2 Videos ) , their overlap is presented in Fig 4C after each image was thresholded so that the top 10% of voxels with the highest values are included in the image . The first component is located mainly around the junction between the supplementary motor area and superior frontal gyrus ( SMA/SFG ) , and in medial occipitoparietal regions ( MOP ) . The second is located mainly in the posterior region of the intraparietal sulcus ( pIPS ) and partly in the posterior cerebellum ( pCBL: crura 1 and 2 ) . The third is mainly in the anterior-medial part of the cerebellum ( a-mCBL: lobules 6 and 8 ) , and the right temporoparietal junction ( TPJ ) . The fourth is in the anterior part of the intraparietal sulcus ( aIPS ) , which includes the aIPL region for the middle component ( e . g . , k = 16 ) , the middle temporal and inferior temporal gyri ( M/ITG ) , and the inferior frontal gyrus ( IFG ) . More detailed quantification of eigenimages for Tasks 1 and 2 are provided in S2–S9 Tables . We then verified whether a reduced model with four states derived from the SVD analysis , instead of the full model with 30 states , could account for the behavioral adaption . From the eigenvariates of the T-map ( Fig 4A and 4B ) , we constructed a reduced four-state model consisting of "eigenstates , " each of which was estimated as a linear combination of the 30 regressors weighted by the corresponding eigenvalues . The variance explained by the four-state model was as high as the original 30-state model ( mean squared error = 4 . 96° , R2 = 0 . 981 ) . Thus , the model reduced from 30 to four states based on the neural activities well explains the behavioral data . The above model-based regression analysis indicated contributions of the parietal and the anterior cerebellar regions to the middle and slow states , respectively . To exclude the possibility that these regression results are due to spurious correlations , we then conducted a decoding analysis to test whether the regional brain activity could be used to classify the two rotations ( 40° and −40° ) . If classification accuracy varies across the three sessions , this would indicate that activity in these regions changes with dynamics similar to the dynamics of medium or slower states . We thus applied a multivoxel pattern analysis ( MVPA ) to parietal regions in which BOLD signals significantly correlated with at least one of the intermediate components ( k = 11 , … , 20 ) , and in cerebellar regions in which signals were significantly correlated with at least one of the slow components ( k = 21 , … , 30 ) ( Fig 5A , see Materials and Methods ) . The MVPA revealed significantly above chance accuracy ( 50% ) in the right aIPL as well as in the cerebellum for all sessions , with averaged accuracy across subjects ranging from 60% to 70% ( Fig 5B ) . This above-chance classification is not surprising because the direction of hand movements changed depending on the rotation types . However , two-way analysis of variance ( ANOVA ) with regions of interest ( ROIs ) and sessions as a within-subject factor revealed a significant interaction between the two factors ( F ( 2 , 40 ) = 3 . 504 , p < 0 . 05 ) . A simple main effect analysis revealed significant increase of accuracy across sessions in the cerebellum ( F ( 2 , 80 ) = 10 . 16 , p < 0 . 001 ) , but no significant difference in the right aIPL ( F ( 2 , 80 ) = 0 . 736 , p = 0 . 482 ) . These results thus indicate that specificity of activity patterns to the task ( 40 or −40° rotations ) increased with sessions in the cerebellum but did not change in the parietal regions . We then verified that the increase in classification accuracy across sessions observed in the cerebellum was unlikely due to behavioral confounds during adaptation ( see S3 Text ) .
We investigated the spatiotemporal neural correlates of motor memory involved in visuomotor adaptation via estimation of the latent memory states derived from a model with multiple states with different time constants . A univariate regression analysis , which correlated the model states with brain activity during the whole adaptation process , first located the neural substrates related to formation of the multiple motor memories . Then , a sparse SVD analysis showed four characteristic networks , associated with a specific profile of correlation with different time constants . Finally , a classification analysis showed that specific activity patterns to the rotation type were acquired in the cerebellum as adaptation proceeded . We organize the following discussion of our results from the faster to the slower time constants . For the first few fastest constants ( 2 s to 4 . 6 s ) , various brain regions were activated , including frontal and parietal lobes , as well as the visual cortex , the temporal cortex , and regions in the posterior part of the cerebellum , specifically in crus 1 . It is known that regions in the crus 1 are connected with prefrontal and parietal cerebral regions according to studies on cerebro-cerebellar connections in monkeys [29] and humans [30] . For slower but still relatively fast time constants up to k = 6 ( 15 . 9 s ) , the widespread activated regions became more localized into the PPC . A possible reason for activation of parietal activity is mental rotation , which has been known to contribute to early adaptation [7 , 32 , 33] and has been consistently localized in the superior parietal regions ( e . g . , [34 , 35] ) . In line with these studies , the subjects in our study who showed larger reaction times tended to perform the task better with lower directional errors ( R2 = 0 . 229 , p = 0 . 028 ) . For intermediate time constants around 16 . 7 min , we found a characteristic region of activation in the right aIPL ( blue circle in Fig 3 ) . This is consistent with a prism-adaptation study [6] finding activity contralateral to the reaching hand in the lateral bank of the intraparietal sulcus , close to the aIPL activity in our study ( note that subjects used the left hand in the current experiment ) . It has been suggested that the left parietal regions are critical for visuomotor rotation , because patients with left parietal damage show a deficit in adaptation [36 , 37] . We also found activity in the left aIPL correlated with intermediate time constants if the threshold is lowered ( p < 0 . 05 uncorrected ) , but the activity in the right aIPL was more significant than that in the left aIPL . Muhta and colleagues showed the importance of the left parietal region for construction of visuomotor mapping based on online correction of error [36] . In contrast , we only provided terminal feedback after the end of joystick movement , and the role of online correction was relatively small . Thus , although further studies are needed to understand the contradiction in parietal laterality between our study and Mutha et al . ’s , the above difference in error feedback may explain this difference . For slower time constants , the number of correlated voxels in the aIPL decreased , and the number in the anterior-medial cerebellum increased . With time constants longer than 1 h , the main activities were identified in the anterior-medial cerebellum; this result is in line with previous studies [6 , 15 , 18] . Regions related to the slow states distribute in the anterior-medial parts and mainly in lobule 6 . These cerebellar regions are connected with cerebral sensorimotor networks , including the primary motor and sensory cortices , the premotor cortex , and the supplementary motor area [30] . This suggests that the cerebellar slow states directly contribute to sensorimotor control without help from cognitive processes ( prefrontal-parietal functions ) probably corresponding to an “autonomous” stage [38] by constructing internal models [18 , 25 , 39] . A recent study reported that transcranial direct current stimulation ( tDCS ) over the cerebellum induced faster adaptation during training but did not affect retention after training [40] . Because tDCS is likely to affect neural activity in the posterior part of the cerebellum to a greater extent than in the anterior-medial part , our findings of the fast components of motor memory in the posterior part ( Fig 3 and S1 Fig ) are consistent with this previous study . The existence of slow components in lobule 6 is consistent with a study reporting that patients with focal degeneration in this lobule have difficulty in adapting to a visuomotor rotation [41] . In addition , activity related to kinematic errors that drive visuomotor rotation has been found in cerebellar regions , including lobule 6 [20] . The SVD analysis of variability in specificities to timescales over the brain identified four components , two fast , one middle , and one slow , each associated with different brain networks . Three groups of components ( fast , middle , and slow ) are consistent with a recently proposed three-component model of visuomotor adaptation [42] . The first and second SVD components are subset regions of the fastest component delineated by the regression analysis and indicate the existence of two types of fast components: one related to SMA/SFG and MOP , and the other related to posterior IPS and cerebellum . Our previous study [43] indicated that medial parietal regions ( including MOP ) are related to switching of internal models based on contextual cues and that posterior-lateral regions of the IPS are related to switching based on sensorimotor feedback . Analogy of the current results to our previous study suggests that the first SVD component corresponds to association between visual cue ( target color ) and responses for tasks , while the second one corresponds to fast adaptation ( or rearrangement of haptic directions ) based on sensorimotor feedback . The third component confirmed the slow component in the anterior-medial cerebellum . The fourth component in the aIPS completely includes aIPL found for the middle time constant . This component was also found in the IFG and M/ITG . Neurons in the IFG are activated when monkeys [44]and humans [45 , 46] observe goal-directed hand actions and when humans imagine hand actions [47] . The IFG has been suggested to contain sensorimotor memory representation related to hand movement [48] . The M/ITG is known to be involved in visual motion analysis [49] , but specific interpretation of this region in the visuomotor adaptation is unknown , at least to our knowledge . Our decoding analysis indicated that the activity pattern in the cerebellum became more specific to rotation type as adaptation proceeded . A previous study by one of us [25] showed that cerebellar activities correlated with learning to control two different cursors ( rotation and velocity ) were spatially segregated , supporting modular organization of internal models , thus suggesting that overlapped regions represent common properties of learning two tasks . In contrast , the results of the present study show no significant regional difference of activities in either the parietal or the cerebellar regions between visuomotor rotations of 40° and −40° ( S6 Fig ) . We surmise that the overlap arises because the two visuomotor transformations are identical except for the rotation angle . Within this common cerebellar region , as well as in the PPC , the MVPA discriminated the representation of the opposing rotations with higher decoding accuracy than chance level . As mentioned earlier , this result is not surprising because the direction of hand movements changed depending on the rotational types and could contribute to successful decoding . Importantly , however , we observed significant increase of the decoding accuracy across sessions in the cerebellar regions related to the slow states ( lobules 6 and 8 ) , but not in the parietal regions related to middle states . The increase in classification accuracy in the cerebellum , especially from the second to the third session , does not appear to be due to changes in performance , because we could not identify significant difference in performance between the second and third sessions . While the primary motor cortex ( M1 ) has been involved in the late stage of adaptation in previous studies [8 , 40 , 50 , 51] , we found no significantly correlated activities in M1 . In addition , in a recent study [51] , we found that MVPA could classify opposite rotational types ( 90° or −90° ) from activity patterns in sensorimotor cortex including M1 , suggesting separate representation for dual visuomotor adaptation . However , the classification was based on fMRI activity measured after intensive training on a continuous tracking task for more than 3 d and a total of 160 min . Therefore , M1 may be correlated with even slower timescales beyond the range of our study , inducing structural changes [8] , although further studies would be necessary to confirm this correlation . Unlike conventional fMRI regression analysis , model-based fMRI regression analysis allows the study of the underlying latent variables generating the behavior in motor adaptation . In previous neuroimaging studies of motor adaptation , the observable behavioral variables of interest were used to define contrasts or regressors for analysis of brain activity . However , the multiple motor memories and related activity that drive the behavior are internal to the subject undergoing adaptation and thus cannot be measured directly ( although they can , in theory , be manipulated by experimental conditions such as task schedules , e . g . , [52] ) . Here , as in a number of reinforcement learning studies ( e . g . , [53] for review ) , we circumvented this difficulty by first estimating internal memory states via computational modeling and then by using these internal variables in the regression analysis to detect neural representations related to formation of motor memories at multiple timescales . Note that we carefully designed our regression models by including possible confounding variables , notably type of task , hand movement , error , and reaction time in each trial . However , when confounding variables are correlated with memory states , such as errors with fast memory states , regression cannot completely dissociate activity related to these confounding variables from activity directly related to memory itself . In addition , our regression models did not include additional behavioral and physical quantities that may correlate with memory states in the model and that may or may not be related to formation of multiple memories , such as attention , eye movements , and repetition of the task . Thus , our results revealed the neural substrates related to formation of the multiple memories at multiple time scales , but not necessarily the neural substrates of multiple memories at multiple time scales per se . Experiments in which nuisance parameters ( such as error ) are varied while adaptation is constant would allow the effects of these confounds to be dissociated . Recent studies have suggested that motor adaptation is a multifaceted process . In particular , behavior during adaptation is not only updated by error-based learning mechanisms , as we have assumed with our model , but also presumably updated by reward-based and use-dependent mechanisms [54–56] . Each of these processes likely operates at multiple time constants as well . In addition , explicit and implicit aspects of motor adaptation have been recently shown to have fast and slow dynamics , respectively [57] . Thus , although the interpretation of what “memory states” represent varies between adaptation studies , our experiment of the neural correlates related to formation of motor memories at multiple time scales is , we believe , highly relevant . Note , however , that our study does not provide a clear picture of the connectivity and spatial arrangements of the multiple neural representations involved . In particular , in line with a previous study supporting a parallel architecture of motor memories over a serial architecture [2] , we have assumed a parallel architecture in our model in which all memories were updated by a common error signal ( see Eq 2 ) . Our results , however , cannot provide evidence for such parallel architecture . To further clarify the actual neural mechanism , model-based fMRI regression can be complemented by functional connectivity analysis [58] or causal and interfering manipulation of neural function via transcranial magnetic stimulation ( TMS ) or tDCS [59–62] .
Twenty-one right-handed and neurologically healthy volunteers participated in the study ( 20–50 y old , mean age of 27 . 3 y , six females ) . Handedness was assessed by a modified version of the Edinburgh Handedness Inventory [63] . Written informed consent was obtained from all subjects in accordance with the Declaration of Helsinki . The experimental protocol received approval from the local ethics committee at the Advanced Telecommunications Research Institute International . We designed a dual-task adaptation experiment with two opposing visuomotor rotations . At the beginning of each trial , a white cross appeared at the center of screen; this cross served both as the fixation point and as the initial cursor position . A round colored target of 0 . 7 cm radius appeared on the top of the screen 8 cm from the center . Subjects were instructed to manipulate an fMRI compatible joystick to move the cursor to the target within 1 . 5 s ( maximum movement time ) ; otherwise , the trial was considered a missed trial , and the data was not analyzed . After the maximum movement time , the cursor appeared in the direction of the joystick movement at 8 cm from the center for 500 ms to provide angular error feedback . To encourage subjects to respond faster , the color of the feedback cursor turned yellow if the subject did not move within 800 ms . ITIs were randomly generated from 4 to 14 s from an exponential distribution , in 2 s increments . For each trial , we calculated the angular error between the target direction and the final cursor direction from the center of the screen . Note that the size of the target was equivalent to 10° in visual angle , allowing up to ±5° of error to “hit” the target . There were three different tasks: a control task and the two different visuomotor tasks , Tasks 1 , 2 , in which the cursor movement was rotated 40° and −40° , respectively . In the control task , the cursor movement was not rotated . The experiment was divided in three sessions , each session lasting about 11 min , with a 1-min break between sessions . Each session consisted of 99 trials , with 27 trials for the control task and 36 trials for each of Tasks 1 and 2 . Three different target colors , red , blue , and green , were used to distinguish the different tasks . Each task was presented in blocks of nine trials , with blocks presented according to schedules such as C1212C2121C2121C1212C , where C , 1 , and 2 indicate a block of nine trials for the Control , Task 1 , and Task 2 . There were two possible schedules starting with either Task 1 ( 11 subjects ) or 2 ( 10 subjects ) , because we counterbalanced the sequence of Tasks 1 and 2 across subjects and sessions to eliminate any confounding effects due to schedule . Similarly , target colors were counterbalanced across subjects . Before the experiment , the participants performed a familiarization session of 150 trials of the control task . Stimuli were presented on a liquid crystal display and projected onto a custom-made viewing screen . Subjects laid in a supine position in the scanner , viewed the screen via a mirror , and were unable to see their hand throughout this task . They were instructed to use their left thumb and index/middle pair fingers to control the joystick with the left upper arm immobilized using foam pads to minimize body motions . We used the MATLAB fmincon function to estimate the value of the two parameters r and q that minimize the mean squared error between the actual adaptations of subjects for Tasks 1 and 2 and model predictions , y ( n ) ( see Eqs 1 and 3 ) . The adaptation data used for the model fit were calculated by averaging the observed adaptations of 21 subjects , excluding missed trials and trials with large ( >40° ) overshoot . Less than 1% of the total number of trials was excluded . Because of the task sequence counterbalancing , the average was computed after inverting the sign of adaptation for ten subjects starting with Task 2 . Using the estimated parameters , we simulated the time series of the 30 states of memory for each task , xk every 1 . 8 s , corresponding to scanner repetition time ( TR ) . Because the model equation ( Eq 2 ) updates the states at each trial , we interpolated and resampled the states of memory at the time of image acquisitions , i . e . , multiples of TR by calculating the decay depending on time constants following trial n . The 30 simulated memory traces for each task were used as regressors for the univariate fMRI analysis . For additional modeling with the standard two-state model [1 , 28] , we fitted four free parameters , time constants and learning rates for the fast and the slow processes , using the same method described above . It is notable that the 30 time constants in the proposed model were predetermined with a logarithmic scale and the 30 learning rates were calculated by two free parameters ( Eq 3 ) . The estimated states of the fast and the slow processes were compared with those of the proposed model ( see Results and S2 Fig ) . A 3-T Siemens Trio scanner ( Erlangen , Germany ) with a 12-channel head coil was used to perform T2*-weighted echo planar imaging ( EPI ) . A total of 368 scans were acquired for each session with a gradient echo EPI sequence , and each subject underwent three sessions . The first five scans were discarded to allow for T1 equilibration . Scanning parameters were repetition time ( TR ) , 1 , 800 ms; echo time ( TE ) , 30 ms; flip angle ( FA ) , 70°; field of view ( FOV ) , 192 × 192 mm; matrix , 64 × 64; 30 axial slices; and slice thickness , 5 mm without gap . T1-weighted anatomical imaging with an MP-RAGE sequence was performed with the following parameters: TR , 2 , 250 ms; TE , 3 . 06 ms; FA , 9°; FOV , 256 × 256 mm; matrix , 256 × 256; 192 axial slices; and slice thickness , 1 mm without gap . Image preprocessing was performed using SPM8 software ( Wellcome Trust Centre for Neuroimaging , http://www . fil . ion . ucl . ac . uk/spm ) . All functional images were first realigned to adjust for motion-related artifacts . The realigned images were then spatially normalized with the Montreal Neurological Institute ( MNI ) template and resampled into 2-mm-cube voxels with sinc interpolation . All images were spatially smoothed using a Gaussian kernel of 8 × 8 × 8 mm full width at half maximum . The smoothing was not performed for multivoxel pattern analysis ( see below ) , as this could blur fine-grained information contained in multivoxel activity [64] . We first conducted a model-based regression analysis of fMRI data . For each of Task 1 and Task 2 , the 30 memory traces with different time constants , which were estimated with the previous behavioral modeling , were used as explanatory variables ( i . e . , regressors ) using the general linear model ( GLM ) . To accommodate the problem of multicollinearity due to similarity of regressors between adjacent time constants , we separately estimated 30 regression models corresponding to individual time constants: S=α1xk , 1+α2xk , 2+ ( Effects of no interests ) +ε , ( 4 ) Here , S is a time series of the BOLD signal at each voxel . The regressors ( xk , 1 and xk , 2 ) in each model are time series of motor memories corresponding to one of the 30 time constants ( k = 1 , … , 30; see Eq 1 in behavioral results and modeling ) for Tasks 1 ( 40° ) and 2 ( −40° ) . Each of them was resampled at scan timings of brain activity and orthogonalized using an SPM function ( spm_orth . m ) . We included the following regressors as “effects of no interests” in the analysis . First , pulse functions that were assigned 1 at every onset of joystick movement and 0 otherwise were included to model hand movements . We assumed that convolution of these functions with a canonical hemodynamic function can model the hand movements with a short movement time in our task ( mean ± SEM: 296 ± 2 . 50 ms ) , with a separate model for each trial type ( Tasks 1 and 2 and the Control ) . In addition , two parametric regressors were included to model the effect of directional errors and reaction times ( “parametric modulation” in SPM ) . In each trial , these parametric regressors also used the pulse functions at onset of joystick movement , but their amplitudes were modulated by directional error and reaction time . We included three boxcar functions , each of which modeled a session effect . Therefore , 12 regressors in total ( 3 [hand movement , error and reaction time] x 3 [tasks] + 3 [sessions] ) were included as effects of no interests . Low-frequency noise was removed using a high-pass filter with a cut-off period of 128 s , and serial correlations among scans were estimated with an autoregressive model implemented in SPM8 . Contrast images of each subject , generated using a fixed-effects model , were taken into the group analysis using a random-effects model of a one-sample t-test . Because the secondary purpose of the model-based regression analysis was to recruit possible regions related to many ( 30 ) states of motor memory for the sparse singular value decomposition analysis ( see below ) , activation was reported with a lenient threshold of p < 0 . 001 uncorrected for multiple comparisons at the voxel level . In further analyses , we applied a stricter inclusion criterion , a cluster-level correction based on the FWER , to representative activations such as those related to the fast , middle , or slow states . The univariate analysis with a p-value cutoff resulted in a grand total of 23 , 413 and 24 , 676 selected voxels associated with 30 memory states for Tasks 1 and 2 , respectively . Thus , the univariate analysis provided two matrices X of T-values for these selected voxels , each for one of the two tasks . We applied the sparse SVD [65] to each matrix X . The sparse SVD was implemented in a refined way with the orthogonality constraints . Specifically , we adopted the regularized estimator ( D^ , U^ , V^ ) that minimizes the sum of squared Frobenius norm of the difference between X and UDVT and a sparsity-inducing regularization term on matrices D , UD , and VD , subject to the orthogonality constraints that both matrices U ( eigenvariates ) and V ( eigenimages ) are orthonormal , where D ( eigenvalues ) is a diagonal matrix . We employed the entry-wise L1 norm , which is the sum of all absolute entries of a matrix , multiplied by a regularization parameter to regularize the three matrices D , UD , and VD . Each regularization parameter was chosen in a decreasing grid of 20 values ranging from 200 to 0 . 1 ( equally spaced in the logarithmic scale ) . For each set of regularization parameters , we obtained a sparse SVD model ( D^ , U^ , V^ ) in which the number of nonzero singular values in D^ gives the rank of the matrix decomposition and singular vectors in U^ , V^ can be sparse with some entries being zero . This produced a sequence of SVD models with sparse singular values and vectors . We then employed the standard BIC model selection criterion [31] to select the sparse SVD model . We additionally conducted an MVPA to test if the regional brain activity could be used to classify the two rotational types ( 40° and −40° ) . The ROIs include the right parietal lobe and the cerebellum . Our previous model-based regression analysis suggested that these regions are related to the middle ( the parietal regions ) and the slow ( the cerebellum ) states ( blue circles in Fig 3A ) . The ROI of the right parietal region was the superior and the inferior parietal lobes according to anatomical map in Pick Atlas ( http://fmri . wfubmc . edu/software/PickAtlas ) . The cerebellar ROI was anatomically defined bilaterally . Fig 5A shows the parietal and cerebellar regions enclosed by red and cyan curves , respectively . Within the parietal ROI , we applied MVPA to BOLD signals of voxels that were significantly correlated with at least one of the intermediate states ( τk , ranging from 2 . 1 to 87 . 9 min: k = 11 , … , 20 ) in the model-based regression analysis . In the cerebellum , MVPA was applied to signals that were significantly correlated with at least one of the slow states ( τk , ranging from 2 . 2 to 92 . 6 h: k = 21 , … , 30 ) . The voxels selected in Tasks 1 and 2 were jointly used by taking a union . To conduct the classification , we first modeled all 297 trials as separate pulse regressors at the onset of movement , which were convolved with a canonical hemodynamic response function . This analysis yielded 297 independently estimated parameters ( beta values ) for each individual voxel . The 198 trials with rotational conditions ( 40° or −40° ) were subsequently used as inputs for the MVPA . The classification was performed with a linear support vector machine ( SVM ) implemented in LIBSVM ( http://www . csie . ntu . edu . tw/~cjlin/libsvm/ ) , with default parameters ( a fixed regularization parameter C = 1 ) . The separate training and testing datasets were generated with a pseudo-random half split of all the samples . Cross validation was then conducted for 1 , 000 times for each subject , and the average classification accuracy was estimated . The two-way ANOVA with ROIs and sessions as an intrasubject factor was used to test the differences in classification accuracies .
|
Motor adaptation , a form of motor learning in which motor commands are modified to compensate for disturbances in the external environment , usually proceeds at a rapid pace initially and is then followed by more gradual adjustments . This suggests that at least two learning processes are involved , but little is known about how many distinct memories the brain actually updates during motor adaptation . In addition , it is unclear whether these putative multiple motor memories reside within a single neural system that encompasses different timescales or in qualitatively distinct neural systems . We addressed these issues using a model-based functional imaging approach in which we first used behavioral data to derive a large number of possible memory “states , ” each with different dynamics , and then correlated these memory states with neural activities . We identified four components: two fast , one intermediate , and one slow , each associated with different brain networks . In particular , areas in the prefrontal and parietal lobes and the posterior part of the cerebellum were associated with formation of memories for short timescales . By contrast , the inferior parietal region and the anterior-medial cerebellum were associated with formation of memories for intermediate and long timescales , respectively .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2015
|
Neural Substrates Related to Motor Memory with Multiple Timescales in Sensorimotor Adaptation
|
The Sec system is responsible for protein insertion , translocation and secretion across membranes in all cells . The bacterial actin homolog MreB controls various processes , including cell wall synthesis , membrane organization and polarity establishment . Here we show that the two systems genetically interact and that components of the Sec system , especially the SecA motor protein , are essential for spatiotemporal organization of MreB in E . coli , as evidenced by the accumulation of MreB at irregular sites in Sec-impaired cells . MreB mislocalization in SecA-defective cells significantly affects MreB-coordinated processes , such as cell wall synthesis , and induce formation of membrane invaginations enriched in high fluidity domains . Additionally , MreB is not recruited to the FtsZ ring in secA mutant cells , contributing to division arrest and cell filamentation . Our results show that all these faults are due to improper targeting of MreB to the membrane in the absence of SecA . Thus , when we reroute RodZ , MreB membrane-anchor , by fusing it to a SecA-independent integral membrane protein and overproducing it , MreB localization is restored and the defect in cell division is corrected . Notably , the RodZ moiety is not properly inserted into the membrane , strongly suggesting that it only serves as a bait for placing MreB around the cell circumference . Finally , we show that MreB localization depends on SecA also in C . crescentus , suggesting that regulation of MreB by the Sec system is conserved in bacteria . Taken together , our data reveal that the secretion system plays an important role in determining the organization and functioning of the cytoskeletal system in bacteria .
Internal organization of bacterial cells is a complex process coordinated spatiotemporally by several molecular machineries . In most rod-shaped bacteria , the actin-homolog , MreB , functions as an intracellular organizer controlling cell wall synthesis , cell shape maintenance , cell polarity , cell division and more [1 , 2 , 3] . Consequently , disruption of the MreB cytoskeletal system leads to pleiotropic phenotypes , including disorganized cell wall synthesis , loss of rod shape , mislocalization of proteins and aberrant chromosome organization [2] . The subcellular organization of the MreB filaments themselves has been at the center of an ongoing debate . Initially , MreB was suggested to form continuous helical filaments along the long axis of the cell underneath the cell surface [4] . However , later high-resolution studies suggested that MreB forms discrete , short patches of filaments that move perpendicular to the long axis of the cell via force generated by cell wall synthesis [5 , 6 , 7] . Then again , more recent reports documented the existence of extended MreB filaments , which are sometimes helical [8] , leaving the issue of the exact subcellular organization of MreB still in doubt [1] . Dynamic physical interaction between MreB and cell wall biosynthetic enzymes support a model by which MreB serves as a scaffold for cell wall synthesis [9] , whereas the process of cell wall synthesis drives MreB circumferential motions [5 , 6 , 7] . Recently , RodZ was also demonstrated to participate in MreB rotation by assisting in coupling it to the cell wall biosynthetic enzymes [10] . In addition to its role in lateral cell wall synthesis and cell elongation , MreB is also implicated in septal cell wall synthesis through its interaction with FtsZ , the bacterial tubulin homolog [11 , 12] . Thus , as a result of MreB-FtsZ interaction , penicillin-binding proteins ( PBPs ) , the major peptidoglycan ( PG ) synthetic enzymes that are bound to MreB , were suggested to be recruited to the septum without the requirement for other cell division proteins [11 , 12] . Thus , the MreB cytoskeletal system and the cell wall synthetic machinery are interlinked and disruption of either of these systems affects the organization and function of the other . MreB filaments are positioned at the inner surface of the cytoplasmic membrane , whereas MreC and MreD , its cytoskeletal partners , are integral membrane proteins , with MreC having a large periplasmic domain and MreD being largely membrane-embedded [13] . Association of MreB with the membrane requires its membrane insertion loop and , in some bacteria , it is further assisted by an N-terminal amphipathic helix [14] . Recent studies indicated that lipid-linked PG precursors are also important for the association of MreB with the membrane [15] . In vivo observations using specific lipid-binding dyes showed that the assembly of MreB filaments with the membrane generates fluid lipid domains and promotes movement of membrane proteins and lipids [16] , similar to actin cortical cytoskeleton of eukaryotes [17] . While the association of MreB with the cell membrane has been broadly studied [14 , 15 , 18] , the possible involvement of membrane-organizing systems in MreB localization and function is largely unexplored . The Sec protein translocation pathway is involved in biogenesis of a large number of membrane-bound and secreted proteins in most bacteria ( reviewed in [19] and [20] ) . The Sec system is comprised of the membrane-embedded SecYEG translocon , which forms the pore through which polypeptides are translocated in unfolded conformation [21] , the SecA ATPase , which functions as the motor protein driving protein translocation [22] and the SecB chaperone , which maintains the newly synthesized proteins in an unfolded conformation[23] . Depending on the type of protein cargo that needs to be transported , the Sec system also cooperates with the Signal Recognition Particle ( SRP ) pathway [24] . The substrates of the Sec system generally encompass an N-terminal signal sequence , which gets proteolytically cleaved by the signal peptidase during translocation [25] . The Sec system has been extensively studied for its role in membrane protein targeting and secretion , with few studies suggesting that it is involved in targeting membrane or secreted proteins specifically to the poles [26 , 27] . Although MreB is not an integral membrane protein and does not have a Sec-type signal sequence , three types of data encouraged us to investigate the relationship between the main bacterial membrane translocation machinery and the MreB cytoskeleton . First , a high-throughput survey of protein interactions in E . coli suggested that SecA and MreB are interaction partners [28] . Second , in E . coli cells depleted for SecE , MreB was found to be enriched in the cytoplasm [29] . Finally , in yeast cells , disruption of the Sec system was shown to affect organization of the MreB-structural homolog , actin [30] . Here we show that SecA and MreB interact genetically and that the organization and function of MreB is regulated by the Sec system . Upon inactivation or depletion of components of the Sec machinery , in particular SecA , MreB changes its localization pattern and accumulates mainly at polar or sub-polar sites . MreB mislocalization in secA mutant cells results in disordered cell wall formation and generation of multilayer membrane regions , which are enriched with high fluidity domains . Furthermore , the mislocalized MreB in secA mutant cells is not efficiently recruited to the Z-ring resulting in incomplete cell division and filamentation . We demonstrate that the above defects are due to RodZ , MreB partner protein , not getting to the membrane in the absence of active SecA , because when we reroute and overexpress RodZ through a SecA-independent pathway , MreB localization and the division defect of SecA-inactivated cells are partially corrected . Finally , we show that the SecA-dependent mechanism for MreB localization exists also in Caulobacter crescentus , suggesting that it is conserved across diverged gram negative bacteria . Taken together , our findings underscore a previously unrecognized capability of the membrane translocation system in organizing bacterial cells on both sides of the membrane .
In order to test if the Sec system and MreB cooperate with each other , we first asked whether the genes encoding the Sec proteins and MreB interact genetically . For this purpose , we compared the growth of cells defective in either the sec genes or mreB to that of cells defective in both . We initiated this survey by spotting serial dilutions of wild-type E . coli and secA51 mutant , which carries a temperature sensitive mutation that renders the cell highly defective in Sec functions [31] , on LB agar plates containing or not containing a sub-inhibitory concentration of A22 , which inhibits MreB polymerization by directly binding to its ATP-binding domain [32] . In addition , we also spotted serial dilutions of ΔmreBCD cells , which lack the entire MreBCD complex , ( enabled by the presence of suppressor mutations , see [33] ) and of secA51ΔmreBCD double mutant cells , which are defective in both systems , on LB agar plates . The plates were incubated at the permissive temperature ( 30°C ) or at a semi-restrictive temperature ( 37°C ) . The results , compiled in Fig 1A , show that already at the permissive temperature , complete depletion of MreB from secA51 cells via a deletion ( the double mutant ) attenuated cell growth , as evidenced by reduction in colony forming ability , raising the possibility of synthetic inhibition upon manipulation of the two genes ( treatment by a sub-inhibitory concentration of A22 had only a minor effect ) . The results at the semi-restrictive temperature confirmed this possibility , since both cultures of secA51 , in which MreB had been either inhibited by A22 or absent due to a deletion , exhibited at least 20 fold reduction in their colony forming ability compared to secA51 cells in which the MreB level or activity have not been manipulated ( Fig 1A , compare growth in the two boxes bordered by red dashed lines to growth in the box bordered by a green dashed line ) . The reduction in the ability to form colonies is more likely to be a synthetic effect , rather than an additive effect of two mutations , since treatment with sub-inhibitory concentration of A22 specifically affected the secA51 cells and not the wild-type . Next , we compared the sensitivity of wild-type cells and of secA204 cells , which carry the prlD21 mutation ( prlD is the former name of secA ) that renders them resistant to sodium azide ( NaN3 ) , to the MreB inhibitor A22 . To this end , we spotted serial dilutions of the two strains on LB agar plates lacking or containing 0 . 1 mM NaN3 or 1 mM NaN3 , supplemented or not with a sub-inhibitory concentration of A22 , and incubated the plates overnight at 37°C . The results , compiled in Fig 1B , show that both strains grew similarly on LB agar lacking or containing low concentration of NaN3 ( 0 . 1 mM ) , whereas only secA204 cells , which are resistant to sodium azide , formed colonies on LB agar containing 1 mM NaN3 , as expected ( Fig 1B , left panel ) . When A22 was included in the plate , growth of wild-type cells was not affected in the absence of NaN3 , but was severely inhibited in the presence of only 0 . 1 mM NaN3 . In contrast , growth of secA204 cells was drastically inhibited in the presence of A22 , and the inhibition became even more severe in the presence of NaN3 ( Fig 1B , right panel ) . Since the secA204 cells are known to be defective in secretion , apparently because the mutation slows the ATPase activity of SecA [34] , it is most likely that their A22 hypersensitivity phenotype of is due to the secretion defect . The possibility of ineffective efflux system contributing to the increased sensitivity of SecA-defective cells to A22 cannot be completely ruled out . However , our observation that the secA51ΔmreBCD double mutant cells exhibit a pronounced growth defect at the semi-restrictive temperature suggests that A22 hypersensitivity of secA51 cells is less likely to be caused by an inefficient efflux system . Finally , we tested whether mutations in other sec genes , namely secE15 and secY39 , both of which defective in secretion at low temperatures , also exhibit growth defect in the presence of a sub-inhibitory concentration of A22 . To test this , we spotted serial dilutions of overnight-grown cultures of wild-type , secE15 and secY39 on LB agar plates , supplemented or not with A22 , and incubated them at the permissive ( 37°C ) , semi-restrictive ( 30°C ) or restrictive ( 23°C ) temperatures . The results , compiled in S1 Fig , show that , unlike SecA-defective cells , SecE- or SecY-defective cells did not show altered sensitivity to A22 at the permissive ( 37°C ) or semi-restrictive ( 30°C ) temperatures , while their growth was completely inhibited at the restrictive temperature ( 23°C ) . Taken together , the data presented above shows that co-inhibition of SecA and MreB largely compromise cell growth , compared to inhibition of only one of them , strongly suggesting that the genes encoding them interact genetically . Having observed that SecA and MreB interact genetically , we asked whether localization of the two proteins depend upon each other . First , we asked whether SecA affects the subcellular organization of the MreB cytoskeletal protein in E . coli . For this purpose , we monitored the localization of an MreB tagged with monomeric superfolder green fluorescent protein ( MreB-msfGFPSW a sandwich fusion shown to be functional in vivo [10] ) , expressed from its native chromosomal locus , in wild-type and in secA51 cells , grown at the permissive ( 30°C ) , semi-restrictive ( 37°C ) and restrictive ( 42°C ) temperatures . In wild-type cells , MreB-msfGFPSW was unaffected at all temperatures and was observed as puncta distributed along the cell periphery ( Fig 2A , left panels , and Fig 2E ) . However , whereas the pattern of MreB-msfGFPSW localization in the secA51 mutant cells grown at the permissive temperature ( 30°C ) was similar to that observed in wild-type cells , in cells grown at the semi-restrictive ( 37°C ) or the restrictive ( 42°C ) temperatures , MreB-msfGFPSW exhibited an entirely different distribution pattern ( Fig 2A , right panels , and Fig 2E ) . Under these conditions , the secA51 cells became elongated , placing a considerable amount of the MreB-msfGFPSW molecules ( 24 ±7% of total cellular MreB , n = 50 cells ) in foci at polar and sub-polar regions ( Fig 2A , right panels ) . Line scan analysis on both sides of the membrane along the cell length indicated that the intensity profile of MreB in secA51 and in wild-type cells is clearly different ( Fig 2B ) . Further quantification of the fluorescence intensities revealed that the average fluorescence intensity , which estimates the relative amount of MreB , is somewhat lower in secA51 cells compared to the wild-type cells ( Fig 2C , left ) , whereas the variance in fluorescence distribution , which provides information on non-homogeneous distribution of MreB , is much higher in secA51 cells compared to wild-type , reflecting the presence of aberrantly distributed MreB clusters in these cells ( Fig 2C , right ) . The observed changes in MreB-msfGFPSW localization are not due to cleavage of the chimeric protein that releases GFP , since Western blot analysis of wild-type and secA51 cells expressing MreB-msfGFPSW using α-GFP antibodies revealed that the MreB-msfGFPSW fusion protein is relatively stable and the negligible amount of what might be free GFP is the same in wild-type and secA51 cells grown at the restrictive ( 42°C ) temperature ( S2 Fig ) . Notably , we obtained similar results when we used MreB–red fluorescent protein sandwich fusion ( MreB-RFPSW ) [35] as a reporter for MreB localization ( S3A Fig ) . Next , we followed MreB localization upon SecA inactivation by time-lapse microscopy . When wild-type cells expressing MreB-msfGFPSW were grown at the restrictive temperature , no defect in MreB localization was observed ( Fig 2D , upper panels , and S1 Movie ) . However , two hours after shifting the secA51 cells to 42°C , a considerable amount of the MreB-msfGFPSW molecules accumulated in clusters located mainly near the cell poles . As growth of the secA51 cells at the non-permissive temperature continued , cells became filamented and the MreB-msfGFPSW continued to concentrate mainly at polar or sub-polar sites ( Fig 2D , lower panels , and S2 Movie ) . Evidently , the secA51 cells did not become spherical when grown at the restrictive temperature , probably due to the fraction of MreB molecules that localized along the cell membrane and remained dynamic ( S3 Movie ) . Importantly , mislocalized MreB-msfGFPSW in secA51 cells grown at the non-permissive temperature ( 42°C ) could be reversed to localize as in wild-type cells when shifted back to the permissive temperature ( 30°C ) ( S3B Fig and S4 Movie ) . This result indicates that MreB localization is greatly affected by the protein secretion status of the cells and that secA51 cells in which MreB is mislocalized are not dead . The ability of MreB to resume its normal localization pattern when shifted back to non-restrictive conditions is not surprising in light of the demonstrated spatial plasticity of MreB . Thus , during the cell cycle , MreB condenses to mid-cell to co-localize with the division plane-associated ring and expands back after cell division [11 , 12 , 36] . Still , the possibility that de novo MreB synthesis during the experiment contributes to the reversal cannot be ruled out . To confirm that the Sec system in the secA51 mutant is defective under the conditions used , we tested the localization pattern of super-folder GFP fused to MalE , which is a known substrate of the Sec system and which has been shown to localize to the cell poles [37 , 38] . The results in S3C Fig demonstrate that polar localization of MalE-sfGFP was abolished in secA51 cells , but not in wild-type cells , grown at the restrictive temperature , indicating that the Sec system is indeed impaired in secA51 under the conditions used . The wild-type-like localization pattern of MreB-msfGFPSW was restored in secA51 cells upon ectopic expression of SecA ( S3D Fig , left panels , and Fig 2E ) , but not upon expression of an unrelated protein , LacZ ( S3D Fig , right panels , and Fig 2E ) . Mislocalization of MreB-RFPSW was also observed when SecA was depleted from DRH729 , a strain that expresses secA from an inducible promoter ( S3E Fig ) . In wild-type cells , approximately half of the cellular SecA content is associated with the inner membrane , whereas the other half is soluble [39] . Since SecA is an interaction partner of MreB [28] , we hypothesized that when it is overexpressed , the amount of SecA in the cytoplasm will increase , and a significant fraction of it would be available to interact with MreB , keeping it in the cytoplasm . Such a scenario is expected to affect cell shape , since it will reduce the pool of membrane-associated MreB , which is involved in cell wall synthesis . To test this hypothesis , we overexpressed SecA in cells expressing MreB-msfGFPSW as the sole source of MreB . The results , presented in Fig 2F upper panels , show that upon overexpression of SecA , the mass of 51% of the cells ( n = 555 ) increased significantly and they occupied a lemon shape with their central part inflated , whereas the MreB-msfGFPSW appeared largely diffused throughout the cytoplasm . In contrast , when we overexpressed a control protein , LacZ , neither the localization of MreB-msfGFPSW nor the shape of the cells were affected ( Fig 2F , lower panels ) . These results suggest that indeed excess SecA is titrating MreB away from the membrane . However , an alternative explanation could be that high levels of SecA change the stoichiometry of the Sec system , thus exerting a dominant negative effect on the secretion machinery , which might indirectly affect MreB localization . However , the fact that MreB localization appeared diffuse , rather than membrane localized upon SecA overexpression , suggests that this effect is largely direct . The results presented thus far indicate that the Sec system is important for proper localization of the MreB cytoskeleton . Is the reverse also true , that is , is MreB important for SecA localization ? To address this question , we observed the subcellular localization of SecA fused to YFP and expressed from its native chromosomal locus in cells also expressing MreB-RFPSW . Cells were either treated with different concentrations of A22 or were not treated by A22 , and images were acquired to monitor the localization of MreB and SecA at different time points after treatment . The results show that , in A22-untreated cells , SecA-YFP molecules localized mainly along the membrane ( S4 Fig , upper panels ) . Notably , even in the A22 treated cells , SecA-YFP remained unaffected , whereas MreB-RFPSW localization was clearly affected by A22 , in a concentration dependent manner ( S4 Fig , center and lower panels ) . Of note , the effect of A22 on the localization of MreB-msfGFPSW was also similar to that of MreB-RFPSW ( S4 Fig , compare right panels to left panels ) . Our observations are in agreement with a previous report , which suggested that SecA localization in B . subtilis cells is independent of MreB [40] . Taken together , our results suggest that SecA is a morphogenetic protein that affects MreB localization , as well as cell shape . To check if other components of the Sec system also influence MreB localization , we tested the localization of MreB-msfGFPSW in secY39 and in secE15 mutant cells . The results show that , while the localization of MreB-msfGFPSW was largely unaffected in wild-type cells grown at the at the permissive ( 37°C ) or restrictive ( 23°C ) temperatures ( Fig 3A and 3D ) , a significant fraction of MreB-msfGFPSW was mislocalized , accumulating in intracellular foci , in secY39 mutant cells grown at the restrictive temperature ( 23°C ) , but not at the permissive temperature ( 37°C ) ( Fig 3B and 3D ) . Similarly , MreB-RFPSW was mislocalized in secE15 mutant cells grown at the restrictive temperature ( 23°C ) , but not at the permissive temperature ( 37°C ) ( Fig 3C and 3D ) . Consistent with the defect in MreB localization , the shape of SecY- and SecE-impaired cells also appeared abnormal . Of note , unlike secA51 cells , which exhibited a defect in MreB localization when grown at the semi-permissive temperature ( 37°C ) , the secY39 and secE15 mutant cells displayed defects in MreB localization only at the restrictive temperature ( 23°C ) and not at the semi-restrictive temperature ( 30°C ) ( Fig 3D ) . This apparently explains why secY39 and secE15 cells were not sensitive A22 when grown at the semi-restrictive temperature ( S1 Fig ) . The results thus far indicate that components of the Sec system are important for spatial organization of MreB in E . coli cells . Since spatial organization of cell wall synthesis is regulated by MreB , we tested whether inactivation of SecA , shown above to affect MreB localization , also affects cell wall synthesis . For this purpose , we stained the wild-type and secA51 cells , which were grown under the restrictive temperature , using fluorescent HCC-amino-D-alanine ( HADA ) [41] . The results , presented in Fig 4A , show that the localization pattern of the HADA label appeared roughly similar in secA51 mutant cells and in wild-type cells . This was expected , since SecA-defective cells contain cell wall , as evidenced by their rod shape rigidity . Still , when we incubated wild-type and SecA-defective cells , grown at the restrictive temperature , with various concentrations of HADA and quantified their average fluorescence intensities after 30 minutes , we found that , the average fluorescence of HADA was significantly lower in SecA-defective cells compared to wild-type cells in all HADA concentrations tested ( Fig 4B ) . This data suggests that inactivation of SecA results in defective cell wall synthesis in these cells . A recent study has shown that artificial localization of E . coli MreB to polar regions results in polar cell wall synthesis and formation of ectopic poles [42] . Although ectopic poles are not observed in secA51 cells , we asked whether the mislocalized MreB in these cells , which is mainly localized at polar or sub-polar regions , performs polar cell wall synthesis . For this purpose , we stained the wild-type and secA51 mutant cells , grown at the restrictive temperature with HADA and performed pulse-chase time-lapse microscopy to visualize the sites of new cell wall synthesis . Notably , the sites at which new cell wall synthesis occur can be identified as regions at which the HADA label disappears . The results show that , in wild-type cell , the HADA label disappeared mainly from mid-cell during 30 minutes of chase at the restrictive temperature and was gradually concentrating in the poles ( Fig 4C , left panel ) . On the other hand , in secA51 cells , the HADA label disappeared from random sites along the cell axis and was not accumulating in the poles during the 30 minutes chase ( Fig 4C , right panel ) . This indicates that the mislocalized MreB in SecA-defective cells does not mediate polar cell wall synthesis . In light of the defects in cell wall synthesis in secA51 cells , we speculated that these mutant cells would exhibit altered sensitivity to cell wall-targeting antibiotics compared to wild-type cells . To test this hypothesis , we performed antibiotic-induced cell lysis assay using ampicillin and cefotaxime , which are cell wall-targeting antibiotics . As a control antibiotic that does not target the cell wall , we treated cells with rifampicin , which is a transcription inhibitor . All antibiotics were added in sub-inhibitory concentrations . When incubated with ampicillin or cefotaxime at a semi-restrictive temperature ( 37°C ) , the OD600 of a secA51 culture dropped rapidly indicating cell lysis ( Fig 4D , left and middle panels ) . In contrast , untreated secA51 cells , antibiotic-treated or untreated wild-type cells and rifampicin treated cells continued to grow during the course of the experiment ( Fig 4D , right panel ) , suggesting that the secA51 cells are hypersensitive to cell wall targeting antibiotics . To directly visualize the effect of ampicillin on the secA51 cells , we imaged wild-type and secA51 cells grown in the absence or in the presence of a sub-inhibitory concentration of ampicillin at the semi-restrictive temperature . Consistent with the growth inhibition of the secA51 cells observed in Fig 4D , lysis of secA51 cells , but not of the wild-type cells , was observed within 2 hours after the antibiotic additions ( S5 Fig ) . Together , these results indicate that the synthesis and the physiological properties of the cell wall are severely affected upon disruption of the Sec system . MreB was suggested to influence membrane organization and membrane protein dynamics [16] . To examine whether the mislocalized MreB in secA51 cells affects their membrane organization , we first stained the membrane of wild-type and secA51 cells , both expressing MreB-RFPSW and grown at the restrictive temperature , with Mito-tracker green ( MTG ) , which is a lipid-permeable dye . The results in Fig 5A show that , unlike the even staining of the membrane of wild-type cells by MTG , which exhibited the expected punctate pattern of MreB-RFPSW along the cell periphery ( Fig 5A , upper panels ) , the MTG in secA51 cells concentrated at polar and sub-polar regions , which are also the sites of MreB-RFPSW accumulation ( Fig 5A , lower panels ) . Next , we asked if the sites at which the MTG stain concentrated in secA51 cells are regions of increased fluidity ( RIFs ) , which are specialized lipid domains formed at the sites of MreB assembly [16] . To check this , we stained the membrane of wild-type and secA51 cells , grown at the restrictive temperature , with the membrane fluidity-sensitive dye Nile Red [43] . The results in Fig 5B show that , although the Nile Red was hardly detected at the membrane of wild-type cells under the staining conditions used , it formed bright foci in secA51 cells . Of note , unlike B . subtilis , which can be stained by Nile Red under the conditions used here , wild-type E . coli cells cannot be stained by the dye ( S6 Fig ) , although the secA51 cells were stained ( Fig 5B ) . Moreover , the Nile Red foci co-localized with MreB-msfGFPSW , expressed from the native locus , in secA51 cells grown at the restrictive temperature ( Fig 5C ) , suggesting that the brightly stained RIFs observed in the secA51 cells were formed due to MreB accumulation . To test whether MreB is the cause for the formation of the RIFs in secA51 cells , we stained secA51ΔmreBCD double mutant and ΔmreBCD mutant cells grown under sec-restrictive conditions with Nile Red and calculated the percentage of cells exhibiting Nile Red staining . The results in Fig 5D and 5E show that the percentage of secA51ΔmreBCD double mutant cells stained by Nile Red was 3 times lower compared to cells carrying only the secA51 mutation ( Fig 5E ) . These results suggest that the RIFs in secA51 cells are largely formed in an MreB-dependent manner . Because Nile Red is known to be a substrate for the pmf-driven multidrug efflux in E . coli , an alternative explanation for the poor staining of wild-type cells could be that they are well energized [44] . However , SecA-defective cells are unlikely to be compromised for their efflux function , since increased staining with Nile Red is reversed when SecA is inactivated in the absence of MreB . To take a closer look at the membrane distortion in secA51 cells , we performed TEM analysis on wild-type and secA51 mutant cells that were grown at the restrictive temperature . Representative images , shown in the Fig 5Fb , reveal extensive multilayer membrane regions near the poles or at other regions of secA51 cells , which were not observed in wild-type cells ( Fig 5Fa ) , suggesting that the intensely stained membrane foci observed in secA51 cells by light microscopy at the sites of MreB accumulation are membrane involutions composed of many layers of membrane . Again , these multilayers of membranes are formed in an MreB-dependent manner , since they were not detected in ΔmreBCD mutant or secA51ΔmreBCD double mutant cells grown under sec-restrictive conditions ( Fig 5Fc and 5Fd ) . Together , these results suggest that MreB induces the formation of RIF-rich multilayer membrane regions in cells defective for SecA . The secA51 cells are filamentous at the restrictive temperature , indicating that their cell division is inhibited . In light of the finding that recruitment of MreB to the septum and its interaction with FtsZ are essential for cell division , together with the demonstration that cells expressing mutant MreB that cannot interact with FtsZ are defective in cell division and form filaments [11] , we asked if the mislocalized MreB in secA51 cells also fails to be recruited to the Z-ring , thus causing stalling of cell division and cell filamentation . To answer this question , we followed the localization of MreB-RFPSW together with ZapA-GFP in both wild-type and secA51 cells grown at the restrictive temperature by time-lapse microscopy . In agreement with previous findings [11] , our results show that MreB was detected as co-localizing with the Z-ring in almost 60% of wild-type cells for a brief period of time ( less than 3 minutes ) prior to being redistributed in the cell periphery ( Fig 6A and 6C ) . In contrast , the sub-polarly accumulated MreB-RFPSW in secA51 cells remained mostly static and was detected as co-localizing with the Z-ring in only 25% of the cells ( Fig 6B and 6C ) . Based on these results , we suggest that the inefficient recruitment of MreB to the Z-ring , which affects the interaction of MreB and FtsZ , contributes to the defect in cell division that the secA51 cells show . Having observed that reducing the amount of active SecA precludes MreB-Z-ring association and in light of the known interaction of SecA with FtsZ , we asked if increasing the amount of SecA would also affect Z-ring formation and/or distribution . The results presented in Fig 6D show that , upon overexpression of SecA , ZapA-GFP appears completely diffused in the cytoplasm ( Fig 6D , upper panels ) , whereas overexpression of a control protein , LacZ , did not affect Z-ring distribution ( Fig 6D , lower panels ) . In light of the effects of SecA overexpression on ZapA and MreB localization , we tested if SecA overexpression affects growth rate . The results in Fig 6E show that the growth rate of cells overexpressing SecA was dramatically reduced compared to cells overexpressing LacZ or not overexpressing any of these proteins ( Fig 6E ) . All in all , the above results suggest that inactivation of SecA affects cell division by perturbing MreB-Z-ring association , while overproduction of SecA affects cell division by leading to Z-ring dispersal . Our findings , which imply that the Sec system is important for the localization and functioning of MreB , raise the question of what is the molecular basis of MreB mislocalization in secA51 cells . RodZ , which is important for localizing MreB near the membrane and linking it to the cell wall synthesizing proteins [35 , 45 , 46] , was recently shown to be inserted into the membrane by the Sec system [47] and could account for MreB mislocalization in cells with impaired Sec system . To validate that RodZ is not targeted to the membrane in Sec-deficient cells , we took advantage of RodZ membrane topology in the following way . Since the N-terminus ( N-ter ) of RodZ is in the cytoplasm , whereas its C-terminus ( C-ter ) is in the periplasm , only GFP fused to its N-ter is expected to be fluorescent , because regular GFP does not fluoresce in the periplasm [38 , 48] . We therefore fused GFP to the N-ter or C-ter of RodZ and , after confirming the ability of both fusions to complement RodZ-deficient cells , monitored their fluorescence/localization in wild-type and secA51 cells grown at the restrictive temperature . GFP fused to RodZ N-ter ( GFP-RodZ ) formed a typical spotty helix-like distribution pattern when expressed in wild-type cells ( Fig 7A , left panel ) , whereas in secA51 cells it exhibited a mixture of the typical spotty helix-like distribution , as well as clusters of aberrantly localized protein ( Fig 7A , right panel ) . However , GFP fused to RodZ C-ter ( RodZ-GFP ) was not fluorescent at all when expressed in wild-type cells , ( Fig 7B , left panel ) , consistent with the presence of GFP in the periplasmic space , not enabling its proper folding . When expressed in secA51 cells grown at the restrictive conditions , RodZ-GFP was fluorescent and formed a spotty localization pattern ( Fig 7B , right panel ) , indicating that RodZ is not inserted into the membrane and its C-ter does not reach the periplasm . These results confirm that RodZ requires a functional Sec system for its membrane insertion . Of note , RodZ and MreB exhibit different distribution patterns in SecA-inactivated cells , as indicated by imaging RodZ-GFP together with MreB-RFPSW in secA51 cells at the restrictive temperature ( Fig 7C , upper panels ) , suggesting that their association is affected in the absence of functional SecA . Previous studies have shown that partial suppression of translation , which reduces the level of proteins that need to be secreted , can rescue the secretion defect in secA51 cells [49] . To confirm that the defect in protein secretion is the reason for RodZ and MreB mislocalization in secA51 cells , we asked if the localization pattern of RodZ-GFP and MreB-RFPSW can be restored in secA51 cells grown at the restrictive temperature , by treating them with a sub-inhibitory concentration of the translation inhibitor chloramphenicol . The results in Fig 7C , lower panels , demonstrate that wild-type-like patterns were indeed restored for both MreB-RFPSW and RodZ-GFP in secA51 cells grown at the restrictive temperature after treating them with CM , Proper localization of RodZ in these cells is evident from the lack of GFP signal . Taken together , both MreB and RodZ fail to localize in secA51 cells , but when the burden on the Sec system is partially relieved , localization of both proteins is restored . Next , we asked if the defect in MreB localization in SecA-deficient cells can be restored by rerouting RodZ to the membrane in a SecA-independent manner . For this purpose , we constructed a RodZ hybrid protein , which is fused at its C-ter to GFP and at its N-ter to the β-glucoside permease BglF , an integral membrane protein that lacks a signal sequence , as predicted by SPOCTOPUS [50] and SignalP 4 . 1 [51] ( which predict the Sec-type signal-anchor within a protein sequences ) . Importantly , BglF-like permeases of the phosphotransferase system ( PTS ) were shown to be inserted into the membrane independently of SecA [52] . We then tested if this chimeric protein can recruit MreB to the membrane in SecA-impaired cells . The results in Fig 7D demonstrate that the C-ter of RodZ , which contains its membrane-insertion domain ( a single α helix ) and is fused to GFP , although brought by BglF to the cell periphery , does not reach the periplasm , but , rather , remains in the cytoplasm , as indicated by the fluorescence of the BglF-RodZ-GFP fusion expressed from the tight Ara promoter ( Fig 7D , bottom panels ) . Nevertheless , upon overexpression of the BglF-RodZ-GFP fusion by the addition of 0 . 1% arabinose or more , the localization pattern of the chromosomally expressed MreB-RFPSW in secA51 cells grown at 42°C was restored , as opposed to its mislocalization when expression of the BglF-RodZ-GFP fusion in these cells was not induced ( Fig 7D , upper panels ) . These results imply that the overproduced RodZ , which was brought near the membrane by BglF , although not properly inserted into it , restored the wild-type-like localization of MreB in the absence of SecA . Notably , MreB has been shown to interact with the cytoplasmic domain of RodZ [35 , 53] , explaining why the improperly membrane-inserted RodZ fusion protein can still interact with MreB . Intriguingly , not only was localization of MreB restored in these cells , but the filamented phenotype of the secA51 cells was significantly reduced in a BglF-RodZ-GFP concentration dependent manner ( Fig 7E ) . While overexpression of BglF-RodZ-GFP ameliorated the cell division defect of secA51 cells , without significantly affecting the wild-type cells ( Fig 7E ) , it did not restore their viability at the non-permissive temperature ( S7 Fig ) , indicating that partial restoration of MreB localization did not solve the secretion defect . To determine whether RodZ alone contributes to MreB mislocalization in SecA-defective cells , we constructed a ΔrodZ secA51 strain , which expresses MreB-RFPSW , and monitored the localization of MreB under sec-restrictive and permissive conditions . Interestingly , inactivation of SecA in ΔrodZ cells still affected MreB localization , as evidenced by the presence of aberrant MreB clusters in these cells ( Fig 7F ) . These aberrant MreB clusters are not formed due to the absence of RodZ , since MreB localization appeared similarly in ΔrodZ secA51 cells , grown under permissive condition , with and without sub-inhibitory concentration of chloramphenicol , or grown under Sec depletion condition , with sub-inhibitory concentration of chloramphenicol ( Fig 7F ) . This implies that SecA affects MreB localization via RodZ-dependent and RodZ-independent pathways . Both the Sec system and the MreB cytoskeletal system are highly conserved across bacterial species , with MreB being largely conserved in rod-shaped bacterial cells . To test whether SecA-controlled localization of MreB is a conserved mechanism in bacteria , we examined the localization of MreB in Caulobacter cresentus cells defective in functional SecA . Of note , C . crescentus also has a RodZ protein which is important for the localization of MreB [46] . To visualize MreB , we integrated a plasmid that expresses MreB-GFP under the control of xylose promoter into the chromosome of LS107 , a wild-type C . crescentus strain , and of LS416 , a secAts mutant C . crescentus strain , which is defective in SecA activity when grown at non-permissive temperature ( 37°C ) . When wild-type or secAts mutant cells were grown under the permissive temperature ( 30°C ) , localization of MreB-GFP appeared normal in both cell types , as indicated by the formation of spotty or medial localization pattern ( Fig 8 , left panels ) . However , when grown at the non-permissive temperature ( 37°C ) , localization of GFP-MreB , which remains unaffected in wild-type cells , was severely affected in secAts mutant cells ( Fig 8 , right panels ) . Thus , rather than forming a wild-type-like localization patterns , MreB-GFP formed multiple distinct puncta along the secAts mutant cells . Of note , in accordance with a previous report [54] , the secAts mutant cells that were grown at the non-permissive temperature were elongated , due to inhibition of cell division , indicating that SecA is inactivated in these cells . Hence , our data suggest that SecA-dependent localization of MreB is conserved in gram negative bacteria that diverged one billion years ago .
Central systems are known to cooperate in cell organization , e . g . , in rod-shaped bacteria , interaction of the actin homolog MreB and the tubulin homolog FtsZ is important for cell division [11] . In this study , we uncovered a previously uncharacterized genetic linkage between the genes encoding SecA and MreB , as well as cooperation between the Sec and the MreB cytoskeletal system in bacterial cell organization . Our data indicate that the Sec system is important for proper filament organization and function of MreB . Thus , in secA51 cells , which produce non-functional SecA protein at the restrictive temperature , an average of 25% of the total cellular MreB molecules mislocalize as large clusters near the cell poles . The mislocalized MreB clusters either remain static or are repositioned to sub-polar regions . Mislocalization of MreB was also observed in C . crescentus cells defective in SecA , indicating that the linkage between the two systems is conserved across bacterial species . Still , MreB mislocalization in C . crescentus cells does not appear similar to that of E . coli cells possibly due to the different patterns of MreB localization in these two widely diverged gram negative bacterial species [1 , 13 , 36] . We further show that mislocalization of MreB in Sec-defective E . coli cells largely compromises its activity in cell wall synthesis , cell division and membrane organization . Finally , we present evidence that mislocalization of MreB in Sec-defective cells is due to improper positioning of MreB around the cell circumference . One of the important functions of MreB is spatial coordination of cell wall synthesis . It does so by transiently interacting with and relocating the PG synthesizing proteins around the cell circumference , which result in cell elongation ( reviewed in [55 , 56] ) ; at the same time the physical force generated by the insertion of new cell wall segments drives MreB movement [5 , 6 , 7] . Our results show that cell wall synthesis is affected in SecA-inactivated cells , as implied by the reduced incorporation of the fluorescent D-amino acid HADA and by the formation of envelope bulges at the regions where MreB accumulates , which were shown to form as a result of disorganized PG synthesis [57 , 58] . The hypersensitivity of Sec-defective cells to cell wall targeting antibiotics suggest that in addition to impairment of the synthesis of cell wall , its physiological properties are also affected upon SecA-inactivation . One question that arises is why the secA51 cells , grown at the restrictive temperature , which has a moderate growth defect , possibly due to reduced cell wall synthesis , do not adopt a round morphology . The explanation might be provided by the fraction of cellular MreB that retain wild-type-like localization patterns and dynamics , which we observed in SecA-inactivated cells . This fraction might be composed of MreB molecules that were synthesized and localized prior to the inactivation of the Sec machinery . In any case , this fraction may drive basal cell wall synthesis , which , in turn , would account for rod shape maintenance in secA mutant cells . For nearly three decades , inactivation of the Sec system was known to affect membrane biogenesis and induce the formation of intra-cytoplasmic multilayer membranes [59 , 60] . However , the molecular mechanism/s that underlie this process remained unclear . Our results point at mislocalization of the MreB cytoskeleton as the cause for the formation of such multilayer membranes regions in Sec-impaired cells , since they were not formed in secA51ΔmreBCD cells grown at sec-restrictive conditions . Intriguingly , depletion or overexpression of MreB were previously shown to correlate with the formation of intra-cytoplasmic membrane-bound compartments apparently due to loss of correlation between membrane and cell wall synthesis , which leads to the generation of excess membrane that folds inward [14 , 33 , 61] . Hence , our finding that mislocalization of MreB due to Sec inactivation , which results in accumulation of multilayer membrane regions , might very well be also due to the disruption of the balance between the rate of membrane and cell wall synthesis . Of note , MreB-dependent formation of multilayer membrane regions in Sec-defective cells does not contradict the occurrence of other membrane defects in these cells that do not depend on MreB . Our observations are in agreement with a recent study , which suggested that the MreB is a membrane organizer , since aberrant localization of MreB distorts the membrane [16] . The same study also documented association between MreB and RIFs , which are specialized membrane domains with increased fluidity , consistent with our observation that RIFs are formed at regions of MreB accumulation in Sec-defective cells . Our results on MreB mislocalization in sec-defective cells also shed light on the cell division defect of these cells , which leads to their filamentation . In general , inhibition of the Sec system can affect cell division by blocking membrane translocation of Sec-dependent divisome proteins . Potential candidates include FtsQ and EnvC , which were shown to be inserted into the membrane and transported to the periplasm by the Sec system , respectively [62 , 63] and FtsK , which contains a potential signal sequence ( as predicted by SPOCTOPUS [50] ) . However , our results point at MreB , which has been shown to be recruited to the forming septum in E . coli by FtsZ during the initial stages of cell division to enable Z-ring constriction , divisome maturation and septal PG synthesis [11] , as an additional and early contributor of cell division defect in these cells . Noticeably , we show that by bypassing the dependence of MreB on the Sec system for membrane localization , the cell division defect of SecA-defective cells is partially corrected , suggesting that SecA-MreB cooperation is important for cell division . Our observation that overexpression of SecA affects MreB subcellular distribution , as well as Z-ring formation , suggests that SecA is a morphogenetic modulatory protein that interacts with central morphogenetic components of the cell , in this case the MreB cytoskeleton or components of the divisome , and affect their organization and thereby cell shape . In line with its unexplored morphogenetic function , a recent study has identified SecA as an important factor necessary for membrane targeting of DivIVA , a B . subtilis polarity-establishing protein [27] . Our data indicate that SecA-mediated targeting of MreB occurs via RodZ-dependent as well as RodZ-independent pathways . Currently , the mechanism involved in RodZ-independent targeting of MreB is not known . Yet , this pathway appears to be Sec-dependent , since partial suppression of translation , which relieves the secretion defect in SecA-inactivated cells , rescues MreB localization in the absence of RodZ . Having said that , the possibility of direct involvement of SecA in MreB localization cannot be completely ruled out . Finally , the fact that the cellular concentration of SecA is non-proportionally high , compared to that SecYEG ( approximately 13 , 000 compared to 500 copies per cell , respectively ) [22] , together with the observation that proteins without a typical signal sequence , such as MreB , FtsZ and DivIVA , interact with SecA and depend on it for their proper localization , supports the idea that SecA has novel , previously unknown functions in cell organization .
Strains and plasmids used in this study are listed in supplementary S1 Table . Overnight E . coli cultures were grown in LB or M9 glycerol , depending on the experiment , supplemented with appropriate antibiotics . C . crescentus cultures were grown in PYE medium . Unless the strain was cold sensitive , overnight cultures were grown at 30°C . When appropriate , antibiotics for E . coli cultures were added at the following concentrations: ampicillin ( 100 μg/ml ) , kanamycin ( 30 μg/ml ) , chloramphenicol ( 25 μg/ml ) or tetracycline ( 20 μg/ml ) ( Sigma-Aldrich ) . For C . crescentus , kanamycin was added at a concentration of 5 μg/ml ( for liquid media ) or 25 μg/ml ( for solid media ) . Unless indicated , M9 media supplemented with glycerol ( 0 . 2% ) was used for all microscopy experiments . Fluorescence microscopy was carried out as described previously [64] . In brief , 0 . 5 ml cells were centrifuged , washed with 1X phosphate buffered saline ( PBS ) and finally resuspended in 10–100 μl of PBS . Cell suspensions were placed on 1% M9 glycerol agarose pads with uncoated cover-slips or on poly-lysine coated coverslips . C . crescentus cells were placed on 1% PYE agarose pads with uncoated cover-slips . The membrane was stained with Mito Tracker Green ( MTG; Molecular Probes , Invitrogen ) at a final concentration of 10 μM . For Nile Red staining , cells were washed and resuspended in 1X PBS that contained 2 μg/ml of Nile Red and incubated at 37°C for 2 minutes . Cells were washed twice before microscopic examination . For staining the cell wall , fluorescent HCC-amino-D-alanine ( HADA ) was used as described in Supplementary Experimental Procedures ( S1 Text ) . Cells were visualized and photographed using an Axiovert 200M ( Zeiss ) inverted microscope equipped with CoolSnap HQ camera ( Photometrics , Roper Scientific ) or Nikon Eclipse Ti-E inverted microscope equipped with Perfect Focus System ( PFS ) and ORCA Flash 4 camera ( Hamamatsu photonics ) . Time-lapse imaging was performed using Nikon Eclipse Ti-E equipped with OKOLAB cage incubator . Unless indicated , cells were spotted in 1% M9 glycerol pads which had been pre-equilibrated to the appropriate temperature and imaged by time-lapse microscopy at the respective temperature . Images were processed using Metamorph ( Molecular devices ) or NIS Elements-AR software . Statistical analyses were performed using GraphPad Prism .
|
The notion that bacterial cells have intricate spatial organization , which affects many vital processes , is relatively new and , hence , the underlying mechanisms are largely unknown . The general secretion system and the cytoskeleton are central systems , each known to organize functions associated with certain cellular domains , in both eukaryotes and prokaryotes . While the role of the Sec system in membrane protein translocation and secretion has been largely explored , not much in known about its role in inner cell organization . We show that the Sec system is important for the localization pattern and functionality of the bacterial cytoskeletal system , which controls cell shape , cell division and polarity . Our findings highlight the Sec system as a central coordinator that controls cellular functions on both sides of the membrane .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"cell",
"walls",
"fluorescence",
"imaging",
"cell",
"cycle",
"and",
"cell",
"division",
"cell",
"processes",
"membrane",
"staining",
"light",
"microscopy",
"membrane",
"proteins",
"microscopy",
"cellular",
"structures",
"and",
"organelles",
"cytoskeleton",
"research",
"and",
"analysis",
"methods",
"specimen",
"preparation",
"and",
"treatment",
"staining",
"imaging",
"techniques",
"fluorescence",
"microscopy",
"cell",
"membranes",
"cell",
"staining",
"cell",
"biology",
"integral",
"membrane",
"proteins",
"biology",
"and",
"life",
"sciences"
] |
2017
|
The bacterial Sec system is required for the organization and function of the MreB cytoskeleton
|
Centromeres are critically important for chromosome stability and integrity . Most eukaryotes have regional centromeres that include long tracts of repetitive DNA packaged into pericentric heterochromatin . Neocentromeres , new sites of functional kinetochore assembly , can form at ectopic loci because no DNA sequence is strictly required for assembly of a functional kinetochore . In humans , neocentromeres often arise in cells with gross chromosome rearrangements that rescue an acentric chromosome . Here , we studied the properties of centromeres in Candida albicans , the most prevalent fungal pathogen of humans , which has small regional centromeres that lack pericentric heterochromatin . We functionally delimited centromere DNA on Chromosome 5 ( CEN5 ) and then replaced the entire region with the counter-selectable URA3 gene or other marker genes . All of the resulting cen5Δ::URA3 transformants stably retained both copies of Chr5 , indicating that a functional neocentromere had assembled efficiently on the homolog lacking CEN5 DNA . Strains selected to maintain only the cen5Δ::URA3 homolog and no wild-type Chr5 homolog also grew well , indicating that neocentromere function is independent of the presence of any wild-type CEN5 DNA . Two classes of neocentromere ( neoCEN ) strains were distinguishable: “proximal neoCEN” and “distal neoCEN” strains . Neocentromeres in the distal neoCEN strains formed at loci about 200–450 kb from cen5Δ::URA3 on either chromosome arm , as detected by massively parallel sequencing of DNA isolated by CENP-ACse4p chromatin immunoprecipitation ( ChIP ) . In the proximal neoCEN strains , the neocentromeres formed directly adjacent to cen5Δ::URA3 and moved onto the URA3 DNA , resulting in silencing of its expression . Functional neocentromeres form efficiently at several possible loci that share properties of low gene density and flanking repeated DNA sequences . Subsequently , neocentromeres can move locally , which can be detected by silencing of an adjacent URA3 gene , or can relocate to entirely different regions of the chromosome . The ability to select for neocentromere formation and movement in C . albicans permits mechanistic analysis of the assembly and maintenance of a regional centromere .
Centromeres , the DNA regions at which microtubules attach to and segregate daughter chromosomes , are essential for genome integrity . Point centromeres have been characterized extensively because of their small size and relative simplicity , especially in Saccharomyces cerevisiae . They are composed of one centromere-specific nucleosome that spans <200 bp of DNA , organized in a tripartite structure that includes a specific DNA binding site necessary for centromere function ( reviewed in [1] , [2] ) . In contrast , regional centromeres are found in most eukaryotes , including fungi other than a subgroup of the Saccharomycotina [3] . They span very large DNA domains ( 10′s to 1000′s of kb ) and are organized into core DNA regions associated with centromere-specific nucleosomes and flanked by highly repetitive DNA packaged into pericentric heterochromatin ( reviewed in [4] ) . Regional centromere function is epigenetic in character [5]–[8]: rather than being dependent upon a specific DNA sequence , the presence of CENP-A , the centromere-specific histone H3 variant , defines the position of a functional centromere . It is thought that CENP-A is regulated by its stabilization at functional centromeres/kinetochores , because at non-centromeric loci CENP-A is removed and proteolyzed [9] , ( reviewed in [7] ) . Neocentromeres , defined as functional kinetochores that assemble at ectopic positions , usually appear together with other chromosome rearrangement events ( reviewed in [6] ) . Neocentromeres have the properties of active centromeres , and , by definition , they associate with CENP-A [10]–[12] . Over 90 examples of human neocentromeres have been documented , most involving the formation of supernumerary chromosomes and often associated with developmental disabilities or specific cancers [6] . Many more neocentromeres likely escape detection because they are eliminated during development [6] , [8] , [13] . It is not known if neocentromere formation occurs first , followed by mutation of the natural centromere or , conversely , if mutation of the natural centromere leads to neocentromere formation . The chromosomal position of neocentromeres is different in different organisms . In Drosophila , they have been found only adjacent to chromosome breaks that inactivated the original centromere [14] , [15] . In S . pombe they appear only at telomeric loci [16] . In contrast , human neocentromeres exhibit flexible adaptation to changes in chromosome structure , often appearing far from the site of the original centromere at either terminal or submetacentric loci ( reviewed in [6] ) . No specific DNA sequence properties necessary for functional regional centromere assembly have been identified . Regional centromeres in humans , flies , plants and fungi are composed of long tracts of repetitive DNA , yet repeat tracts are not absolutely required for centromere function or for the formation of neocentromeres ( reviewed in [8] , [17]–[19] ) . In S . pombe , pericentric heterochromatin formation is necessary for efficient de novo assembly of a functional kinetochore [20] , as well as for formation of telocentric neocentromeres [16] . C . albicans , an opportunistic fungal pathogen that resides as a commensal in its human host , possess regional centromeres that are much smaller and simpler than other regional centromeres [3] , [21] . Each of its 8 diploid chromosomes has a centromere that is regional based on its size ( ∼3–4 . 5 kb that specifically associates with CENP-ACse4p [22] , [23] ) , the lack of tripartite point-centromere DNA structure , the presence of several orthologs of proteins found only at regional centromeres , and the absence of orthologs of proteins found only at point centromeres [3] . Most notably , C . albicans CENs lack pericentric heterochromatin: CENP-A associated core sequences are not embedded in long tracts of repetitive DNA [21] , [22]; there are no clear orthologs of either heterochromatin protein 1 ( HP1 ) or of enzymes necessary for the methylation of histone H3 lysine 9; and there is no homolog of CENP-V , a protein that regulates the extent of pericentric chromatin in human cells [24] . Consistent with a lack of pericentric heterochromatin , genes near the centromeres are transcribed at levels close to the average level of transcription across the genome ( K . E . S . Tang and JB , data not shown ) . Furthermore periodic nucleosome spacing is seen at inactive centromere DNA and not at active CENs , suggesting that nucleosomes at active CENs do not associate tightly with a specific DNA sequence [21] . Naked CEN7 DNA used to transform C . albicans did not permit de novo assembly of centromere function [21] , [22] . Taken together , these observations suggest that , like other regional CENs , the assembly of a centromere and the inheritance of centromere function in C . albicans requires epigenetic properties conferred by the association of CENP-A and other kinetochore proteins , rather than by a specific DNA sequence . The stoichiometry of microtubules and centromere-specific nucleosomes differs in different organisms . S . cerevisiae has only one CENP-A nucleosome and one microtubule per centromere , while S . pombe has ∼2–3 CENP-A nucleosomes [25] and ∼2–4 microtubules per centromere [26] . This suggests that one microtubule is attached to kinetochore proteins assembled at each CENP-A nucleosome [25] . In humans , the number of CENP-A nucleosomes is thought to be far larger than the number of microtubule attachments [4] , [27] . In C . albicans there are ∼8 CENP-ACse4p molecules per centromere , presumed to be organized into 4 centromere-specific nucleosomes , and only one microtubule per centromere [25] . This suggests that only one of the four CENP-A nucleosomes at each centromere assembles a kinetochore structure that binds a microtubule . Thus , the prevailing model is that at C . albicans CENs , as at human centromeres , some CENP-A-containing nucleosomes bind microtubules while others do not . Seven of the eight C . albicans centromeres are near short repeats; only the centromere of Chr7 ( CEN7 ) does not have obvious repeat sequences nearby [22] . Yet most analysis of C . albicans centromere function has been performed with CEN7 DNA , which is necessary for Chr7 stability [23] . On Chrs 2 , 3 and 6 there are direct repeats within ∼3 kb of the centromere core sequence . On Chrs 1 , 4 , 5 and R the CENP-A bound centromere core DNA is flanked by a short inverted repeat ( IR ) . The palindromic structure of the four centromeres with a flanking IR is most reminiscent of the structure of S . pombe and other regional centromeres . Here , we studied the properties of C . albicans CEN5 , a centromere with a palindromic structure , by replacing the CEN5 DNA and the flanking IR with a URA3 marker . We found that it behaved like a regional centromere: the resulting cen5Δ::URA3 Chr5 derivatives were stably maintained through mitosis by efficient formation of a neocentromere at one of several non-centromeric loci . Loss of the wild-type chromosome 5 homolog and homozygosis of the neoCEN5 homolog demonstrated that cells can survive in the absence of any CEN5 DNA . Some of the cen5Δ transformants formed a “proximal neoCEN” near cen5Δ::URA3 , which subsequently moved onto , and silenced the URA3 gene . Other cen5Δ transformants formed “distal neoCENs” at several different loci far from the deleted CEN5 locus . Thus , neocentromere formation does not require a specific DNA sequence and can occur at several different chromosomal loci .
We previously identified isochromosome derivatives of Chr5 , in which the derived chromosomes replaced either the right arm with left arm information ( i ( 5L ) ) , or replaced the left arm with right arm information ( i ( 5R ) ) [28] , [29] . The only DNA common to the i ( 5L ) and i ( 5R ) isochromosomes is the predicted Chr5 centromere ( CEN5 ) , including an inverted repeat ( IR ) flanking the central core ( CC5 ) ( Figure 1A ) . This CC5+IR structure resembles a simplified version of the central portion of typical regional centromeres ( e . g . , central core+innermost repeats in S . pombe centromeres ) . Telomere truncation constructs [30] replacing either the complete Chr5L arm [29] or the Chr5R arm ( Table S1 ) yielded chromosome fragments ( Figure 1A ) similar in mitotic stability to that of the isochromosomes ( <10−3 loss/division ) . Thus , the CC5+IR DNA is associated with centromere function in Chr5 and all its stable derivatives , implying that it provides centromere function . To ask if CEN5 DNA is necessary for centromere function , we replaced CEN5 ( CC5+IR ) on one of the two Chr5 homologs with a PCR product that included a URA3 gene flanked by homology to regions just outside the IR ( Figure 1B ) . In C . albicans laboratory strains , the two Chr5 homologs differ in at least two ways . First , the mating type-like locus ( MTL ) is heterozygous ( MTLa or MTLalpha ) . Second , there are two LTR repeats present in the ‘long’ allele of CEN5 ( one in the central core region and one to the left of the left side of the IR ) ( Figure 1B ) . In addition , the Major Repeat Sequence ( MRS ) on Chr5R can vary in size by >50 kb [31] , [32] . Initially , we found lower transformation frequencies for cen5Δ::URA3 constructs than for control myo1Δ::URA3 , constructs that replaced a gene 100 . 2 kb from CEN5 . For example , there were 3–5-fold fewer transformants per microgram of DNA for cen5Δ::URA3 than for the control when colonies were counted 3–4 days after transformation . The proportion of transformants that carried bona fide CEN5 deletions , detected by PCR amplification , was also 4–5 fold lower for the cen5Δ::URA3 than for the myo1::URA3 strains . However , for the cen5Δ transformation ( and not for the control transformants ) , new colonies continued to appear for up to ∼9 days after transformation . After 9 days , similar numbers of Ura+ transformants were obtained for the cen5Δ and the control transformations ( Figure 1B ) . When the transformant colonies that appeared later were analyzed , the proportion of transformants that carried bona fide CEN5 deletions , as detected by PCR amplification and confirmed by Southern analysis ( Figures 1C and S1A ) , was similar to , or slightly less than , that of a myo1Δ::URA3 control strain ( Figure 1B ) . We also obtained transformants in which CEN5 was replaced with NAT1 . Southern blots and hybridization to CEN5 were used to verify the correct insertion of these markers ( Figure S1B–D ) . The growth rates of the cen5Δ::URA3 strains were indistinguishable from the growth rates of wild-type strains and strains carrying URA3 at non-centromeric loci ( Table 1 , Figure 2 ) . Similar levels of stability and growth rates were also seen in the cen5Δ::NAT1 transformants ( Table 1 ) . This result is different from what was seen with CEN7: deletion of the Chr7 CENP-ACse4p binding region was reported to yield only highly unstable chromosomes [23] . Importantly , all of the correct cen5Δ::URA3 transformants exhibited relatively high levels of mitotic stability ( Table 1 ) , measured as the proportion of cells that retain the ability to grow in the absence of uridine . Transformation of C . albicans sometimes causes unintended aneuploidies [33] . To investigate this possibility , we analyzed the cen5Δ::URA3 transformants using comparative genome hybridization ( CGH ) and CHEF karyotype gels . CGH array analysis did not detect any copy number changes in any of the strains: like the parent , all were disomic for all 8 chromosomes ( Figure 1D and data not shown ) . In general , CHEF karyotypes for all but strain YJB9907 were indistinguishable from the parent strain for Chr5 when stained with ethidium bromide ( Figure 1E left panel ) . Only a Chr5-sized band was detected with a URA3 probe as well as probes for CEN5 , Chr5L and Chr5R ( Figure 1E and data not shown ) , indicating that no Chr5 rearrangements occurred . In one strain , ( YJB9907 ) the two Chr5 homologs became separable and the two separable Chr7 homologs became inseparable ( Figure 1E , left panel , lane 2 ) , most likely because of recombination between the MRS repeats on these chromosomes [34] . The CGH and CHEF gel analysis together indicate that no gross chromosomal rearrangements were evident in the cen5Δ strains . Thus , the cen5Δ strains retained two intact copies of Chr5 and genome integrity was maintained . The relative stability of strains lacking CEN5 implies that a neocentromere , an ectopic assembly of a functional kinetochore onto non-centromeric DNA [35] , formed on the Chr5 homologs carrying cen5Δ alleles . Importantly , these neocentromeres apparently formed in all recovered transformants in which CEN5 had been correctly replaced with a selectable marker . Because C . albicans is diploid , all of the cen5Δ::URA3 strains retained one wild-type copy of CEN5 . While centromeres generally function only in cis , phenomena that act in trans , such as transvection in Drosophila ( reviewed in [36] ) , raised the possibility that the cen5Δ::URA3 homolog was able to segregate through some interaction with the wild-type CEN5 homolog of Chr5 . To determine if cells could survive and divide in the absence of any CEN5 DNA , we utilized sorbose growth , which selects for cells that lose one copy of Chr5 [37] . Strains were plated to sorbose medium lacking uridine to select for Ura+ cells carrying only one Chr5 homolog ( Figure 3A , middle panel ) . The colonies that appeared on sorbose-uri medium were then streaked to rich medium ( YPAD ) and larger colonies ( those that likely reduplicated the single copy of Chr5 [37] , Figure 3A , lower panel ) were streaked to SDC-uri to ensure that they retained the URA3 marker . PCR analysis confirmed retention of the cen5Δ::URA3 and failed to detect wild-type CEN5 or MTLa ( Figure 3B ) in several isolates , consistent with the idea that the wild-type Chr5/CEN5 homolog had been lost . CHEF analysis indicated that the karyotype of these strains was unaltered ( Figure 3C ) . In the case of YJB9907 , the parental Chr5 homologs had distinct mobilities , and the sorbose-selected derivatives ( YJB9907-3s and YJB9907-6s ) carried only the smaller homolog , which hybridized to URA3 and not to CEN5 ( Figure 3C , lanes 5 and 8 ) . CGH analysis indicated that the remaining Chr5 homolog was present in two copies ( Figure 3D ) following growth on SDC-uri . For YJB9929 , the two derivatives ( 9929-1s and 9929-2s ) showed similar results to YJB9907 and its derivatives . Both PCR and Southern analysis confirmed that the original CEN5 DNA was absent from both strains ( Figure 3B&C ) . The four strains lacking any CEN5 DNA exhibited growth rates that were slightly slower than the growth rates of wild-type and cen5Δ::URA3/CEN5 strains ( Figure S1E , Table 1 ) . This reduced fitness could be because the cells lacked CEN5 DNA completely or it could be due to the homozygosis of all markers on Chr5 that occurred as a consequence of sorbose selection and subsequent growth on rich medium ( Figure 3A ) . To address this question , we performed growth rate experiments with strain YJB9726 , which maintained and then reduplicated one CEN5 homolog , yet was homozygous for all genes on Chr5 because it had been propagated on sorbose ( Table 1 , Figure S1E ) . Importantly , the growth rates of strains with and without intact CEN5 were similar , indicating that CEN5 DNA is not necessary for chromosome propagation and stable strain maintenance . Rather , the slower growth is likely due to one or more genes that , when homozygous , cause reduced fitness . This supports the idea that neocentromeres formed in the cen5Δ::URA3 strains have no obvious deleterious effect on growth rate or fitness in standard laboratory growth conditions . Two different classes of cen5Δ::URA3 transformants were distinguishable based on the rate at which cells acquire the ability to grow on 5-FOA ( FOAR cells ) , a compound that is toxic to Ura+ cells [38] . This assay is generally used as a proxy for mitotic loss rates and quantified using fluctuation analysis to identify median FOAR rates [39] , [40] . The Class A transformants , termed ‘proximal neoCENs’ based on data described below , became resistant to 5-FOA at a rate of ∼10−2–10−3/cell division and the phenotype was reversible . The Class B transformants ( termed ‘distal neoCENs’ based on data described below ) became irreversibly resistant to 5-FOA at a rate of ∼10−5–10−6/cell division , similar to the rate of 5-FOA resistance of myo1Δ::URA3 strains ( Figure 4A ) . Disruption of either Chr5 homolog ( short or long , Figure 1B ) yielded both Class A and Class B transformants , indicating that the class of transformant was not due to disruption of a specific CEN5 allele . Southern analysis of the Class B FOAR strains indicated that , like myo1Δ::URA3 transformants , when these cells became FOAR , they had lost the URA3 marker , ( Figure 4B ) . Thus , rare FOAR derivatives of Class B cen5Δ::URA3 transformants have either undergone a non-disjunction event to lose the entire chromosome carrying cen5Δ::URA3 or have lost the URA3 via recombination . To distinguish between these possibilities , we followed the segregation of markers on both arms of Chr5 ( HIS1/his1Δ::NAT1 near the Chr5R telomere and MTL on Chr5L ) . At least 98% of the FOAR Class B derivatives had lost heterozygosity at HIS1 ( on Chr5R ) and , among those , 100% of them also lost heterozygosity at MTL ( on Chr5L ) ( Figure 5 ) . Thus , there is strong genetic support for the idea that whole chromosome loss was the cause of the FOAR phenotype in Class B transformants ( Figure 5 ) . Several lines of evidence indicated that the higher rate of FOAR in Class A transformants was due to transcriptional silencing and not loss of URA3 DNA . First , the FOAR phenotype was reversible: transformants initially selected on SDC-uri gave rise to cells that subsequently grew on SD+5-FOA and these colonies could subsequently give rise to cells that grew on SDC-uri ( Figure 4A ) . In each transfer ( from SDC-uri ( U ) to SD+FOA ( UF ) , then to SDC-uri ( UFU ) , then to SD+FOA ( UFUF ) and so on ) , only a subpopulation of colonies appeared ( ∼1/1000 cells ) . Importantly , the reversible FOAR phenotype was unique to the cen5Δ::URA3 strains; when URA3 was inserted adjacent to an intact CEN5 ( e . g . , in the 5L-URA3-TEL strain ) , it was not reversibly silenced ( Figure 4A ) . Second , strains grown on SD+FOA retained URA3 DNA on Chr5 , as detected by CHEF Southern ( Figure 4B ) and PCR ( data not shown ) . Third , URA3 DNA sequence was unaltered in any of the FOAR Class A transformants ( YJB9909 , YJB9915 , YJB9916 and YJB9926 , data not shown ) , indicating URA3 did not acquire point mutations . Fourth , genetic analysis of Class A FOAR transformants carrying markers on Chr5L and Chr5R did not lose markers from either arm ( Figure 5 ) , reinforcing the idea that both Chr5 homologs were retained on 5-FOA . Finally , real time PCR detected ∼16 fold lower levels of URA3 mRNA in cells grown on SD+FOA relative to those grown on SDC-uri . This indicates that URA3 expression was silenced at the transcriptional level in the Class A FOAR cells ( Figure 4C ) . Taken together , these results indicate that URA3 DNA was silenced at the transcriptional level in most Class A FOAR transformants and therefore , that the cen5Δ::URA3 Chr5 homolog was stable to a similar degree in these cells and in the Class B cells . Thus , cen5Δ chromosomes are very stable , far in excess of the stability of minichromosomes in either S . cerevisiae or S . pombe that bear bona fide centromeres [41] , [42] . This provides more evidence that the cen5Δ::URA3 homologs in both classes of transformants must carry a functional neocentromere . Active regional centromeres and neocentromeres are , by definition , associated with CENP-ACse4p [10]–[12] , [43] , [44] . To identify the position of neocentromeres on cen5Δ::URA3 homologs , we performed chromatin immunoprecipitation ( ChIP ) experiments using antibody raised against an N-terminal peptide from C . albicans CENP-ACse4p ( see Materials and Methods ) . We first analyzed Class A transformants to investigate the mechanisms linking neocentromere function and the growth phenotypes on 5-FOA or medium lacking uridine in these strains . As expected , anti-CENP-ACse4p specifically recognized DNA in CEN4 , CEN5 and CEN6 central core regions , but did not preferentially precipitate non-centromeric sequences such as the TAC1 gene on Chr5L ( Figure 6A ) . In addition , in Class A transformants grown on SD+5-FOA , CENP-ACse4p preferentially associated with URA3 , while when these strains were grown on SDC-uri , CENP-ACse4p associated with URA3 to a lesser degree ( Figure 6B ) . This implies that the reversible silencing is due to reversible association of CENP-ACse4p with URA3 . In contrast , in Class B Ura+ cells , CENP-ACse4p was not associated with URA3 , and in FOAR cells the URA3 DNA had been lost from the strain and was not detectable in the lysate ( Figure 6B ) . To more precisely localize the active neocentromeres in Class A transformants , we analyzed DNA from anti-CENP-ACse4p ChIP experiments using PCR amplification of contiguous 400 bp DNA segments spanning a ∼15 kb region including and surrounding CEN5 . Consistent with previous studies [22] , [23] , CENP-ACse4p was associated with the CC5 sequence and not with the IR or flanking Ch5L or Ch5R sequences ( Figure 7A ) in two different parental strains that each carried intact copies of both CEN5 homologs . Furthermore , in six independent cen5Δ::URA3 transformants , the wild-type CC5 DNA on the intact , unmodified homolog ( which could be distinguished based on the presence or absence of the LTR insertion , Figure 1B ) was specifically associated with CENP-ACse4p ( Figure S2A and B ) . Thus , in all cen5Δ::URA3 strains , the centromere on the wild-type Chr5 homolog was not perturbed by the loss of CEN5 on the other homolog . In addition , CENP-ACse4p was associated with DNA ∼1 kb to the left of URA3 in all Class A transformants growing on SDC-uri . A similar pattern was seen in four independent transformants , including those carrying deletions of either CEN5 homolog ( with and without the LTR ) ( Figures 7B and S2C ) . In contrast , when Class A transformants were grown on 5-FOA ( URA3 gene silenced ) , CENP-ACse4p always associated with URA3 in all four independent Class A transformants , irrespective of the CEN5 homolog that had been disrupted ( Figures 7B , S2C and data not shown ) . Importantly , when strains were cycled from SDC-uri to 5-FOA and then back to SDC-uri for two additional cycles ( UFUFU ) and then to SDC+FOA again ( UFUFUF ) , the same patterns were retained: in cells grown on 5-FOA the CENP-ACse4p was associated mostly with URA3; in cells grown on SDC-Uri CENP-ACse4p enrichment was adjacent to the URA3 gene ( Figure 7C ) . This suggests that the neocentromere occupied different positions when Class A cells were grown under different conditions and that its average position had shifted position by ∼1–2 kb . Analysis of Class A transformants in which the short vs long CEN5 homologs were disrupted revealed that the episemon LTR was only associated with CENP-ACse4p when the LTR was linked to the cen5Δ::URA3 homolog ( Figure S2C ) . In addition , the LTR appeared to be a preferred site for CENP-ACse4p binding . When it was located on the same homolog as cen5Δ::URA3 , the lateral spread of CENP-ACse4p from URA3 into the adjacent DNA was reduced . Thus , CENP-ACse4p associated with DNA flanking the left side of the IR only in cis and not in trans . In the proximal neoCEN transformants , we detected the association of CENP-ACse4p with URA3 because of the FOAR counter-selectable phenotype of Ura− cells . Quantitative RT PCR indicated that the 16-fold silencing of URA3 occurred by affecting levels of mRNA ( Figure 4C ) . To ask if association with CENP-ACse4p resulted in transcriptional silencing of genes other than URA3 , we measured the expression levels of two genes located just left of cen5Δ::URA3 in Class A/proximal neoCEN strains . These two genes ( orf19 . 3161 and orf19 . 3163 ) were expressed at higher levels in FOAR cells ( when CENP-ACse4p was not associated with them ) relative to their expression in Ura+ cells ( when CENP-ACse4p was associated with them ) ( Figure 4C ) . This is consistent with the idea that association of a gene with CENP-ACse4p results in decreased transcription from that gene . In the Class B strains , CENP-ACse4p associated with CC5 on the wild-type homolog and was not associated with other DNA fragments near cen5Δ::URA3 ( Figure S2D ) . This suggested that the neocentromeres in Class B strains were located beyond the 15 kb region analyzed by ChIP , and thus are located too far from cen5Δ::URA3 to silence it . To locate the position ( s ) of neocentromeres in Class B transformants , we performed ChIP followed by massively parallel high-throughput sequencing ( ChIP-SEQ ) of one wild-type and two Class B transformants ( YJB9907 and YJB9929 ) . As expected , in each strain , the major region associated with CENP-ACse4p on each wild-type chromosome overlapped with the region previously described by Carbon and co-workers as the centromere [22] , [23] , although it was smaller , ranging from 2–3 . 3 kb ( Figure S3 ) . The difference in sizes predicted is likely due to the improved resolution from the smaller average fragment size used in the ChIP pulldowns . In all three strains , the degree to which each centromere , other than CEN5 , associated with CENP-ACse4p was similar relative to the other centromeres ( Figure 8A ) . In both of the cen5Δ::URA3/CEN5 strains , CENP-ACse4p associated with the remaining CEN5 copy at a level slightly higher than half the relative signal present on two copies of CEN5 in wild-type cells ( Figure 8A ) . Analysis of the ChIP-SEQ data also identified new CENP-ACse4p-associated DNA sequences on Chr5 in both Class B neocentromere strains ( Figure 8B ) . In YJB9907 CENP-ACse4p was associated with a ∼4 kb region near the left telomere . This peak ( present in one copy , termed neoCEN-1 ) included 36% of the ChIP-SEQ reads relative to the average centromere ( present in two copies ) . This is ∼half as much CENP-ACse4p as was present in the CEN5 peak from the other Chr5 homolog . In strain YJB9929 , two different regions of the genome exhibited small peaks of CENP-ACse4p association: ∼1 . 2 kb around orf19 . 1978 on Chr5L ( termed neoCEN-2 ) and a ∼1 kb region overlapping orf19 . 6678 and it's 5′ upstream region on Chr5R ( termed neoCEN-3 ) ( 6% and 8% of the signal seen in an average centromere , respectively ) . This suggests that the neoCENs associate with less CENP-ACse4p than does a wild-type centromere . The positions of the predicted neocentromeres in the two Class B strains , as well as in wild-type strains , were then analyzed by standard ChIP analysis , using primers that amplified fragments across the regions identified by ChIP-SEQ ( Figure S4 ) . As predicted from the ChIP-SEQ analysis , in YJB9907 CENP-ACse4p was associated with unique , telomere-adjacent sequence within and 5′ to orf19 . 5698 ( neoCEN-1 ) , and in YJB9929 CENP-ACse4p was associated with neoCEN-2 and neoCEN-3 DNA . Thus , neocentromeres can form at positions far from the deleted cen5Δ::URA3 locus . For this reason , we term Class B transformants ‘distal neoCEN’ transformants as compared to the Class A ‘proximal neoCEN’ transformants . Importantly , the neocentromere peaks were strain-specific . They never appeared in the wild-type strain or in the other neocentromere strain ( Figure 8B ) . ChIP extracts of YJB9907 analyzed by PCR confirmed the presence of CENP-ACse4p near the left telomere and its absence in the neoCEN-2 or neoCEN-3 regions . Similarly , in YJB9929 , CENP-ACse4p was associated with neoCEN-2 and neoCEN-3 and not with the neoCEN-1 region . Furthermore , on all other chromosomes , telomeric regions were not associated with CENP-ACse4p ( data not shown ) . Thus , one unambiguous neocentromere is formed in YJB9907 , while in YJB9929 at least two different DNA regions are associated with CENP-ACse4p and presumably function as neocentromeres . Because dicentric chromosomes are highly unstable [45] , [46] and because YJB9929 exhibited a growth rate similar to that of YJB9907 and wild-type strains ( Table 1 , Figure 2 ) , we hypothesized that the YJB9929 ChIP extract was prepared from a mixed population of cells carrying at least two different neocentromeres . To test this possibility , we analyzed 4 independent colonies from the YJB9929 culture by ChIP-SEQ . Despite the presence of a mixed population , we detect CENP-ACse4p peaks only at the neoCEN-3 position ( as well as at the original CEN5 position , which is expected because this strain is heterozygous for CEN5/cen5Δ::URA3 ( Figure 8B ) ) . PCR of these ChIP extracts confirmed that CENP-ACse4p is located at the neoCEN-3 position in 3 of the 4 colonies ( Figure S4 ) . This is consistent with the stronger signal for neoCEN-3 relative to neoCEN-2 in the original YJB9929 isolate ( Figure 8B ) . Thus , in different strains , neocentromeres formed at different loci and on both arms of Chr5 . We next determined the position of the neocentromeres in the sorbose-derived YJB9907s and YJB9929s strains lacking both wild-type CEN5 homologs . CENP-ACse4p ChIP-SEQ of strains YJB9907-3s and YJB9907-6s detected CENP-ACse4p associated with neoCEN-1 , near the Chr5L telomere ( Figure 8B ) , although the signal peak appeared to shift to the telomere-distal portion of the peak . PCR analysis of the two strains confirmed that CENP-ACse4p remained associated with this telomere-adjacent region . Furthermore , no major peak was detected at any other Chr5 locus tested ( e . g , neoCEN-2 or neoCEN-3 , Figure S4 ) and , consistent with the loss of CEN5 in these strains , no CEN5 DNA was associated with CENP-ACse4p . Thus , despite the fact that sorbose growth causes stresses that can result in chromosome loss , the neocentromere in strain YJB9907 was sufficiently stable to be maintained in the same general region during selection for Chr5 loss and reduplication/non-disjunction . Nonetheless , the shift in the peak position suggests that the neoCEN in YJB9907s strains may have undergone a local shift in its position . In contrast , ChIP of YJB9929-1s and YJB9929-2s did not detect CENP-ACse4p associated with neoCEN-1 , neoCEN-2 or neoCEN-3 ( Figure S4 ) . ChIP-SEQ of the two YJB9929-s strains identified a new , strong CENP-ACse4p signal associated with sequences ∼170 kb from the Chr5L telomere ( Figure 8B , neoCEN-4 ) . PCR of ChIP samples from the two independent derivatives ( Figure S4 ) confirmed that CENP-ACse4p was present at this locus . Thus , in strain YJB9929 exposed to the stress of growth in sorbose , the original neoCEN locations were lost and a new active neocentromere ( neoCEN-4 ) arose at a distance of ∼640 kb from the neoCEN-3 . It has been suggested that analysis of neocentromere DNA will identify features of DNA sequence necessary for centromere function [11] . The only characteristic common to all C . albicans centromeres is that they are found in very long intergenic regions ( average 7 . 5 kb , range 3 . 8–17 . 4 kb ) ( Figure 8C and [23] ) . In addition , seven of the eight C . albicans centromeres are flanked by repeats; four ( including CEN5 ) are oriented as inverted repeats while three are oriented as direct repeats . Like other regional centromeres , C . albicans centromeres have no common primary DNA sequence feature [23] . Consistent with this , no sequence motifs common to centromeres and/or neocentromere regions ( including flanking sequences upstream and downstream of the centromeres and neocentromeres ) were identified using stringent search criteria and no obvious inverted repeats , palindromes or common sequence motifs were found at all centromeres and neocentromeres ( using EMBOSS ( http://pro . genomics . purdue . edu/emboss/ ) [47] , JSTRING ( http://bioinf . dms . med . uniroma1 . it/JSTRING ) and MEME [48] ) . Furthermore , neither the centromeres nor neocentromeres were enriched for any nucleotide or dinucleotide combination relative to total genomic DNA . Importantly , the C . albicans neocentromeres identified here all shared two features seen at bona fide centromeres . First , for each of the neocentromeres , at least one intergenic region within the CENP-A binding peak was considerably larger than the average size of intergenic regions on Chr5 ( 902 bp ) or across the entire genome ( 853 bp ) ( Figure 8C ) . Furthermore , all of the neocentromeres were found in close proximity to repeat sequences ( Figure 8D ) . NeoCEN-1 is immediately next to the telomere-adjacent repeated TLO11 and the rho-5a LTR repeat and within 4 kb of the telomere repeats . NeoCEN-2 is centered on GIT2 , with adjacent copies of GIT1 and GIT4 , which encode predicted glycerophosphoinositol permeases that share 75% and 73% identity with GIT2 at the DNA level ( determined by BLAST [49] ) . NeoCEN-3 is flanked by an inverted repeat that includes orf19 . 6676 and orf19 . 6678 on the left and orf19 . 6681 and orf19 . 1124 . 2 on the right . These gene pairs are 99% ( orf19 . 6676 and orf19 . 1124 . 2 ) and 70% ( orf19 . 6678 and orf196681 ) identical . In addition , the intergenic region adjacent to orf19 . 6678 includes an omicron-5a LTR element . Finally , neoCEN-4 is within a large ( >4 . 4 kb ) intergenic region that includes an LTR element ( chi-5a ) and is flanked by orf19 . 570 and orf19 . 575 , which are >75% identical at the DNA sequence level and are organized in a direct repeat orientation . Thus , all four neoCENs include a large intergenic region as well as inverted or direct repeats . Furthermore , three of the four neoCENs also feature an LTR repeat in close proximity .
Chromosomes carrying a neocentromere exhibited loss rates of ∼10−4–10−5 per division . This is much more stable than minichromosomes in S . cerevisiae ( reviewed in [42] ) or in S . pombe [50] and resembles the high levels of neocentromere stability in humans ( reviewed in [6] ) . Deletion of CEN5 did not affect cell growth rates unless all Chr5 markers were homozygous , which grew at least as well as CEN5-containing strains that were homozygous for all Chr5 markers ( Table 1 , Figure S1E ) . Thus , slow growth was due to homozygosis of one or more genes on Chr5 and not to the loss of CEN5 . CEN5 deletion occurred with ∼equal frequency on the two Chr5 homologs , suggesting that C . albicans requires two different alleles at one or more Chr5 loci to maintain optimal fitness . The high level of stability in cen5Δ strains is consistent with the idea that C . albicans has regional centromeres . Yet deletion of CEN7 was reported previously to result in high levels of Chr7 instability ( >50% loss per division ) [23] . CEN7 is the only C . albicans centromere lacking any flanking repetitive DNA [22] . The role of repeat DNA in CEN5 structure , and whether or not it affects centromere function , remains to be determined . However , the presence of repeat DNA at each of the neocentromeres ( Figure 8D ) suggests that CEN7 may behave differently from all other C . albicans centromeres . A notable property of C . albicans neocentromeres is that they form in all bona fide transformants . This result is remarkable for two reasons . First , it suggests that when a centromere is perturbed , it induces the formation of neocentromeres . While we cannot rule out that events can occur in the opposite order , our results clearly suggest that loss of a functional centromere stimulates the formation of a new one . Second , it indicates that , when induced to form , neocentromere formation is efficient , since replacement of CEN5 yields similar numbers of transformants as replacement of a non-essential gene on Chr5 . This high level of neocentromere efficiency is unprecedented . Despite its efficiency , the establishment of neocentromeres was not always rapid . Some transformants appeared early while others continued to appear over time , with high levels of correct transformants appearing as small colonies 7–9 days after transformation . In contrast , when a control gene on Chr5 was disrupted , most colonies appeared within a few days and no new small colonies appeared after several days . Importantly , C . albicans cen5Δ strains that initially appeared as small colonies subsequently grew with wild-type growth rate kinetics ( Figure 2 ) . This suggests that deletion of a wild-type centromere led to the epigenetic establishment of neocentromeres at new positions and that neocentromere assembly occurred over the first ∼1 week after transformation . Since we isolated similar proportions of proximal and distal neocentromeres from cen5Δ::URA3 colonies that appeared early or late following transformation , the time at which transformants initially appeared was not a reflection of neocentromere position ( e . g . , proximal vs distal neoCEN ) . This slow appearance of transformants is reminiscent of the epigenetic nature of centromere establishment on minichromosomes in S . pombe . In this case , cells carrying minichromosomes with functional centromeres were selected because they grew more rapidly than cells carrying minichromosomes that had not yet formed functional centromeres [5] . During the course of this work , Ishii et al . ( 2008 ) demonstrated that excision of an S . pombe centromere results in several different cell fates . In a large proportion ( ∼80% ) of the cells , the chromosome became unstable , leading to chromosome missegregation and cell death . Cells are rescued at a frequency of <10−3 [16] by two types of events: telomere-telomere fusion of the acentric chromosome with one of the other chromosomes or formation of a telocentric neocentromere . Such telomere-telomere fusions were not detected in any of the strains we studied here , although we have documented a telomere-telomere fusion event between Chr5 and an isochromosome ( 5L ) in at least one drug-treated strain [28] . Thus , while telomere-telomere fusions can occur in C . albicans , they do not appear to be a major mechanism of chromosome rescue following loss of wild-type centromere DNA . In S . pombe , neocentromere formation was almost completely dependent upon the ability to form heterochromatin . The proportion of rescued cells carrying neocentromeres dropped from ∼75% in wild-type cells to <10% in cells lacking Swi6 ( an HP1 homolog ) , Clr4 ( a histone H3 methyl transferase ) or Dcr1 ( the dicer homolog required for RNAi-dependent heterochromatin formation ) [16] . Clearly , C . albicans CEN5 exhibits epigenetic properties , such as neocentromere formation and movement , in the absence of classic pericentric heterochromatin . Accordingly , we were able to obtain cen5Δ::URA3 transformants in strains lacking the histone deacetylase ortholog of S . cerevisiae Sir2p . The efficiency of transformation and the proportion of Class A to Class B transformants were similar in sir2Δ/sir2Δ and isogenic wild-type ( SIR2/SIR2 ) C . albicans strains ( H . W . , unpublished data ) . This result is consistent with the idea that canonical heterochromatin is not necessary for neocentromere formation . We propose that , rather than being required for kinetochore assembly per se , pericentric heterochromatin in larger regional centromeres may be required for the higher order centromere structure necessary to support the organization of multiple microtubule attachment sites per centromere . Because C . albicans has only one microtubule attachment per centromere/kinetochore , this type of centromere structure may be dispensable . The neocentromeres formed in this study had two common features: large intergenic regions and repeated DNA ( Figure 8C&D ) . Similarly , in organisms with larger regional centromeres , centromeres are positioned in ‘gene deserts’ . Interestingly , 14 newly evolved centromeres in primate species formed in regions significantly devoid of genes in other primate genomes . [51] . Furthermore , fine structure analysis ( at the level of 10–20 kb regions within the ∼2 Mb centromeric domians ) of rice centromeres indicates that the very few genes found within a centromeric domain were located within subdomains associated with histone H3 rather than with CENP-A/CEN-H3 histones , suggesting that centromeres evolve from gene-poor regions [51] , [52] . Here we found that association with CENP-ACse4p silences URA3 in C . albicans ( Figures 4 & 7 ) . Furthermore , active transcription through centromeric DNA inactivates kinetochore function in S . cerevisiae [53] as well as on a human minichromosome [54] . In S . pombe , CENP-ACnp1 silences ura4+ in the centromeric central core in a concentration-dependent manner [55] , [56] . We propose that the incompatibility of CENP-A nucleosomes and active transcription is a general feature of centromeres . The second structural feature of neocentromeres is proximity to repeated DNA sequences . At seven of the eight native centromeres , inverted or direct repeats are found within a few kb of the CENP-A enriched core . Importantly , while a single LTR sequence is sometimes seen within the region , these repeats are different from pericentric repeats flanking larger regional centromeres in other organisms because they are often unique to the individual centromere . For example , the inverted repeat flanking CEN5 is composed of DNA found only on the two sides of the CEN5 central core . Similarly , the GIT1 , GIT2 and GIT4 genes are found only in the neoCEN-2 region and orf . 6676 and orf19 . 1124 . 2 are found only in the neoCEN-3 region ( Figure 8D ) . We propose that a small amount of repeat DNA assists in the assembly of a functional kinetochore in a heterochromatin-independent manner . The reversible nature of the change in URA3 expression status in Class A/proximal neoCEN transformants and the correlated change in CENP-ACse4p position relative to URA3 suggests that neocentromeres can move locally , at least in a subset of cells ( 1/1000 ) in the population . During selection for chromosome loss on sorbose medium , a poor carbon source , the general telocentric position of the neocentromere in YJB9907 was maintained , although it may have shifted a few kb distal to the telomere . Telocentric chromosomes are common in mice , and C . albicans CEN6 is telocentric , located ∼50 kb from the Chr6R telomere , indicating that telomeres are one preferred site of centromere assembly , perhaps because they are generally devoid of essential genes and terminate in repetitive DNA . In strain YJB9929 it appears that the neocentromere ( s ) were less stable . The presence of two different neocentromere positions YJB9929 , which was originally isolated from a single colony , is also consistent with the idea that neocentromere positions are not fixed . Analysis of single colonies suggested that each cell most likely had only one neocentromere on Chr5 ( Figure 8B ) . The smaller size of the neocentromeric DNA region associated with CENP-ACse4p may result in a smaller number of centromeric nucleosomes than at wild-type centromeres . Human neocentromeres are generally similar in size to normal centromere regions when the size of the constriction is measured by electron microscopy [6] . However , levels of CENP-A at two human neocentromeres , measured with a GFP-CENP-A , are ∼1/3 the level seen at most other centromeres [57] , which was proposed to be due to less efficient loading of CENP-A at human neocentromeres [57] . We propose that , in C . albicans , the neocentromeres remained functional despite the smaller amount of CENP-A because only one CENP-A nucleosome is required to attach to a single microtubule on each sister chromatid [25] . In contrast to the stable neocentromere in YJB9907s , in two independent YJB9929s strains obtained by sorbose selection for loss of one Chr5 homolog and reduplication of the remaining copy , the neocentromere moved to an entirely new position ( neoCEN-4 ) . No CENP-ACse4p was detected at the neoCEN-4 locus in any of the YJB9907 isolates or in any of the YJB9929 isolates prior to sorbose selection . Thus , upon sorbose selection in YJB9929 a neocentromere formed ∼40 kb from neoCEN-2 on the same chromosome arm or ∼640 kb from neoCEN-3 to the other chromosome arm . We do not know if neocentromeres are repositioned by sliding along the DNA or by disassembling and reassembling at a new location . We propose that proximal neoCENs reposition by sliding locally ( 1–2 kb at URA3 ) because they tend to remain close by , resulting in reversible silencing when a neocentromere is within a few kb of URA3 and not when it is >200 kb away . Furthermore , similar local movements occur at distal neoCENs as evidenced by the slight shift of the neoCEN-1 peak . In contrast , because CENP-A is normally removed from non-centromeric DNA regions [9] , it is tempting to speculate that an assembly/disassembly mechanism ( Henikoff and Dalal 2005 ) contributes to distal neoCEN formation as well as to the movement of CENP-A from neoCEN-3 to neoCEN-4 during sorbose selection of strain YJB9929 . C . albicans centromeres are unique in that neocentromere formation at multiple positions is efficiently induced by deletion of a centromere . As in humans and Drosophila , chromosome breakage and healing ( mediated by homologous recombination between URA3 and the CEN5 flanking DNA ) accompanies neocentromere formation events in C . albicans cen5Δ transformants . This suggests that , while similar mechanisms operate in organisms that have pericentric heterochromatin , the process is simpler and much more efficient in C . albicans because the complexity of heterochromatin does not come into play . Neocentromeres can form close to the site of the original centromere ( as in Drosophila [15] ) , at distal loci with no sequence conservation ( as in humans ( reviewed in [6] ) or near a telomere ( as in S . pombe [16] ) ) . In humans , there are a small number of cases in which neocentromeres appear at ectopic positions and are not associated with any obvious chromosome rearrangement ( reviewed in [6] , [58] ) . While the formation of neocentromeres in C . albicans followed deletion of the wild-type centromere , we did not detect new aneuploidies or translocations in any of these strains . This is especially interesting in light of the high levels of chromosome rearrangements that can occur in C . albicans strains [59] . Another unique aspect of C . albicans centromeres is that the movement of proximal neoCENs can be monitored by selection and counter-selection for URA3 . This provides an extraordinarily powerful tool for direct studies of the mechanisms of local neocentromere movement and demonstrates that , while rare , short range neocentromere movement ( several kb ) is more frequent than movement over much longer distances . There is much that we do not understand about regional centromeres . In part , this is because S . cerevisiae point centromeres cannot be used to study epigenetic phenomena , such as neocentromere formation and movement , that occur only at regional centromeres . The small size and simplicity of C . albicans regional centromeres greatly facilitates studies of the core functions of regional centromeres without the complications associated with long , complex repeat tracts and pericentric heterochromatin . In this initial study , we found that C . albicans centromeres form neocentromeres , despite having no pericentric heterochromatin . Furthermore , neocentromeres can form at multiple chromosomal loci including loci close to the deleted centromere , telomeric loci and other distal loci on both chromosome arms . Finally , neocentromeres can move over short or long distances to new loci and can silence a gene associated with it . Additional work will be needed to identify the environmental and genetic conditions that influence the rate of formation as well as the structural features that modulate the location and movement of neocentromeres .
Strains are listed in Table S1 and were constructed using PCR-mediated gene deletion with 70 nucleotides of homology to genomic DNA with gene disruption primers listed in Table S2 and a lithium acetate protocol [60] . CEN5 was deleted by amplification of URA3 from pGEM-URA3 [60] , which was inserted at CEN5 in both orientations . As an alternate approach , CEN5 was replaced with NAT1 from plasmid pMG2120 using primers 3161 and 3162 . Strain YJB9779 ( 5L-URA3-Tel ) was constructed by transforming strain YJB7617 with URA3-TEL amplified from pMG2192 using primers 2249 and 1489 to insert it 360 bp to the right of the IR ( 37 bp centromere proximal to the orf19 . 4219 coding sequence ) . pMG2192 was constructed by digesting pMM100 [61] with Not1/Xho1 and 1000 bp CA7 telomere repeat was ligated to Xho1/Not1 cut pYPB1-ADH1 [62] . Strain YJB9858 ( TEL-NAT1-5R ) was constructed by transforming YJB7617 with TEL-NAT1 amplified from pMG2194 using a forward primer ( 2313 ) that produces a fragment that can insert 1 kb to the left of the IR ( 660 bp centromere proximal to the orf19 . 3161 coding sequence ) and a reverse primer ( 1489 ) with homology to telomere sequence in the plamid . pMG2194 was constructed by digesting pMM100 [61] with ApaI and the 900 bp fragment containing the telomeric repeat was cloned into pMG2171 [29] . C . albicans transformants were isolated after incubation on SDC-uri or YPAD+Nat at 30°C . Transformation frequencies and the proportion of bona fide transformants was similar for cen5Δ and control strains . The frequency of insertion into the long and short CEN5 homologs was also similar when either earlier or later appearing transformants were analyzed . For example , for a cen5Δ::URA3 transformation , of the 61 transformants , 8 were correct insertions and 4 deleted CEN5 long and 4 deleted CEN5 short . Transformants were screened for those with bona fide insertions by PCR using primers ( Table S2 ) flanking the insertion site together with primers within the inserted DNA as illustrated in Figure S1A . The CEN5 homolog that was disrupted with URA3 was identified by Southern analysis of conventional and CHEF karyotype gels using methods previously described [28] . Probes were amplified with primers listed in Table S2 . C . albicans karyotypes were analyzed by CHEF gel electrophoresis as described previously [33] using conditions that optimize visualization of chromosomes 2–7 . Comparative genome hybridization ( CGH ) was performed as described [33] and was plotted to the C . albicans genetic map using an updated version of the Chromosome_Map [33] , based on Assembly 21 coordinates [63] , [64] . DNA was transferred from agarose gels to Magnacharge nylon membranes ( GE Osmonics , Minnetonka , MN ) [65] . Membranes were probed overnight at 42°C and detected with anti-Digoxigenin-Alkaline Phosphatase and CDPstar essentially as described [66] . Probes were prepared by PCR amplification using DIG-labelled nucleotides according to manufacturer's instructions ( Roche , Indianapolis , IN ) . Strains carrying cen5Δ::URA3 were grown on sorbose medium without added uridine for 7–9 days . Colonies that appeared were streaked to YPAD media and large colonies that appeared early were then streaked to SDC-uri medium to ensure that they had retained the URA3 marker . PCR analysis was performed to identify derivative strains in which the cen5Δ::URA3 fragment was retained , the wild-type CEN5 was no longer detectable , and that only one of the two MTL loci on Chr5 was retained , using primers listed in Table S2 . Growth rates were determined using overnight stationary phase cultures diluted 1∶105 in YPAD media covered with mineral oil in a 96-well plate ( Corning , NY ) . Cultures were incubated in a microplate spectrophotometer ( Sunrise , Tecan , San José , CA ) at 30°C with constant linear shaking . OD readings were taken every 15 min . Doubling time was calculated based on the exponential constant of the fitted exponential curve of each well using software kindly provided by Sven Bergmann and Gil Hornung . Mitotic stability was determined by comparing the proportion of cells containing the relevant marker ( URA3 or NAT1 ) relative to the total number of cells in the population following growth under conditions selective for the marker [67] . Fluctuation analysis of loss rates was performed as described [39] using the method of the median [40] . Briefly , strains were streaked for single colonies and grown on SDC-uri for 2 days at 30° . Twenty independent colonies per strain were inoculated into 5 ml liquid non-selective medium ( YPAD ) and grown overnight at 30°C with shaking . Cultures were harvested by centrifugation and washed twice in 1 ml of sterile water . Appropriate dilutions were plated onto nonselective YPAD for total cell counts and selective media ( SD+FOA ) for selection of Ura- colonies . Plates were incubated at 30°C for 2–3 days , and colony counts were used to calculate the rate of FOAR/cell division [39] . To distinguish loss of the URA3 marker from silencing , we analyzed FOA resistant colonies identified in loss rate fluctuation analyses done with strain YJB10064 ( his1::NAT1/HIS1 ∼38 kb from the Chr5R telomere ) in which either CEN5 or MYO1 had been replaced with URA3 . The resulting strains [YJB10169 ( myo1Δ::URA3 ) , YJB10233 ( cen5Δ::URA3 Class A ) , and YJB10234 ( cen5Δ::URA3 Class B ) ] were replica-plated to SDC-his , SDC-uri , YPAD+Nat , and YPAD . FOAR colonies that re-grew when replica plated to SDC-uri were candidates for URA3 silencing . Images of the SDC-his , YPAD+Nat , and YPAD replica plates were overlaid in order to classify colonies as NatRHis+ , NatSHis+ , or NatRHis− , using the YPAD replica plate as a control for colony transfer . From each colony class obtained per strain , five colonies were analyzed by PCR amplification of the MTL loci using primers 1193 , 1194 , 1195 and 1197 ( Table S2 ) . Rabbit polyclonal anti-CENP-ACse4p antibodies were raised against an N-terminal peptide ( amino acids 1–18 ( Ac-MARLSGQSSGRQTGQGTSC-amide ) ) of Ca CENP-ACse4p and affinity purified ( Quality Controlled Biochemicals; Hopkinton , MA ) following the protocol of Sanyal and co-workers [68] . ChIP was performed using extraction protocols modified from Weber et al . [69] and Carbon and Co-workers [23] . Overnight cultures were diluted to an OD600 of ∼0 . 3 and harvested at an OD600 of ∼1 . 0 . Cells were fixed in 1% formaldehyde for 15 min at 30°C and then treated with zymolyase ( 0 . 2 mg/ml+0 . 5 mM AEBSF ) until 80–90% of cells were spheroplasts . Reactions were stopped with 1 . 2 M Sorbitol , 1 mM MgCl2 , 20 mM PIPES pH 6 . 8 , and sonicated to yield DNA fragments of 500–600 bp . Input lysate was collected after insoluble matter was precipitated twice by centrifugation . The lysate was incubated at 4°C for ∼16 h with or without 4 µg/ml anti-CENP-ACse4p antibody . Antibody and associated DNA was incubated with Protein A-agarose beads at 4°C for ≥3 hrs , washed several times and then eluted twice at 65°C . Crosslinks were reversed overnight at 65°C followed by RNase A and proteinase K treatments . Proteins were removed by phenol-chloroform/isoamyl alcohol/chloroform extraction . Samples were ethanol precipitated , washed in 70% ethanol and resuspended in 80 ml TE . Each Class A ChIP experiment was performed with the same isolate grown in SDC-uri and in SD+FOA . ChIP experiments were repeated with 4 different Class A strains ( YJB9909 , YJB9915 , YJB9916 and YJB9926 ) analyzing 3 independent colonies from YJB9909 and YJB9915 and one colony each from strains YJB9916 and YJB9926 . Two independent colonies from each of two different Class B strains ( YJB9907 and YJB9861 ) also were analyzed . Contiguous fragments spanning ∼15 kb across CEN5 were amplified using primer pairs listed in Table S2 . Dilution PCRs were used to ensure that the amplification of each PCR product was in the linear range . PCR products were run on an 1 . 4% agarose gel and band intensities were quantified using ImageJ software ( NIH ) with a background correction macro . Intensities were calculated ( +Ab/lysate ) and normalized to the mean intensity of CEN5 central core PCR products in all strains except cen5Δ/cen5Δ strains , in which case it was normalized to the mean intensity of CEN4 central core products . “No-antibody” samples were run in parallel to ensure quality of the samples and reactions ( e . g . , Figure 6 ) . Correlation coefficients for groups of samples were performed using the Excell ‘Correlation’ function [70] . Chromatin lysates and samples immunoprecipiated with CENP-ACse4p antibody ( as described above ) were processed for massively parallel high throughput sequencing using the Illumina Genome Analyzer Classic following manufacturer's instructions . Samples were nebulized to an average ∼200 bp fragment size . The ends were repaired and an “A” base was added to the 3′ ends . Illumina adapter oligos were ligated to the fragments and these were purified on a 2 . 0% agarose gel by excising a region of the gel corresponding to 200 base pairs . After purification using a Qiagen Gel Extraction kit , ( Qiagen , Valencia , CA ) , the adapter modified DNA fragments were enriched by PCR . The enriched fragments were purified using a Minelute PCR purification kit from Qiagen and quantified using the Quant-iT dsDNA HS Assay Kit from Invitrogen ( Carlsbad , CA ) . The samples were then diluted to give a final concentration of 2 pM and cluster generation was performed on the Illumina Cluster Station ( Illumina , San Diego , CA ) following manufacturer's instructions . The resulting flow cell was sequenced on the Illumina GA Classic for 36 cycles . The sequences were extracted from . tiff image files using Firecrest and Bustard tools of the Illumina Genome Analyzer pipeline . These sequence reads were aligned to the human genome as reference sequence ( NCBI v36 . 49 from ensembl ) using Illumina's ELAND tool . Both 36 bp and 25 bp reads were used , along with unmasked and repeat-masked [71] human genome . Peaks in the read coverage from uniquely aligned reads were identified using the FindPeaks tool [72] . Release version 3 . 1 of the tool was used with default parameters for binding site search , except the values for subpeaks as 0 . 2 and trim as 0 . 2 . For non-overlapping sliding window bins of chromosome 5 data , we calculated the ratio of reads mapped ( aligned ) in the bin , to the total number of reads mapped on the chromosome . The difference of this ratio in the sample compared to control was calculated to identify enriched bins along the chromosome for that sample . RNA was extracted from cells grown to log phase in SDC-uri or SD+FOA using the Epicentre MasterPure yeast RNA purification kit ( Epicentre Biotechnologies , Madison , WI ) according to manufacturer's instructions . cDNA was prepared with the BioRad iScript cDNA Synthesis Kit ( Bio-Rad Laboratories , Hercules , CA ) with a combination of random hexamers and oligo ( dT ) primers . Real-time PCR was performed using the Roche LightCycler FastStart DNA MasterPLUS SYBR Green I kit ( Roche Applied Science , Indianapolis , IN ) . Reactions were run on the Eppendorf Mastercycler ep realplex machine ( Eppendorf , Westbury , NY ) . Primers used are listed in Table S2 . A no RT negative control and melt curve analysis were performed on each run to ensure that no contaminating DNA or second products were amplified . Duplicate wells and 3 technical repeats were performed on cDNA from cen5Δ::URA3 Class A strains YJB9909 and YJB9916 . ΔCt values were determined from the mean of the results of all technical and biological replicates ( Figure 4C ) . Error bars show standard error of the mean ( SEM ) .
|
Centromere function is essential for proper chromosomal segregation . Most organisms , including humans , have regional centromeres in which centromere function is not strictly dependent on DNA sequence . Upon alteration of chromosomes , new functional centromeres ( neocentromeres ) can form at ectopic positions . The mechanisms of neocentromere formation are not understood , primarily because neocentromere formation is rarely detected . Here . we show that C . albicans , an important fungal pathogen of humans , has small regional centromeres and can form neocentromeres very efficiently when normal centromere DNA is deleted , and the resulting chromosomes are stably propagated . Neocentromeres can form either very close to the position of the deleted centromere or at other positions along the chromosome arms , including at the telomeres . Subsequently , neocentromeres can move to new chromosomal positions , and this movement can be detected by silencing of a counterselectable gene . The features common to sites of neocentromere formation are longer-than-average intergenic regions and the proximity of inverted or direct repeat sequences . The ability to select for neocentromere formation and movement in C . albicans permits mechanistic analysis of the assembly and maintenance of a regional centromere .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"cell",
"biology/microbial",
"growth",
"and",
"development",
"microbiology/microbial",
"evolution",
"and",
"genomics",
"genetics",
"and",
"genomics/chromosome",
"biology",
"molecular",
"biology/centromeres"
] |
2009
|
Neocentromeres Form Efficiently at Multiple Possible Loci in Candida albicans
|
Spermatogenesis consists broadly of three phases: proliferation of diploid germ cells , meiosis , and finally extensive differentiation of the haploid cells into effective delivery vehicles for the paternal genome . Despite detailed characterization of many haploid developmental steps leading to sperm , only fragmentary information exists on the control of gene expression underlying these processes . Here we report that the RFX2 transcription factor is a master regulator of genes required for the haploid phase . A targeted mutation of Rfx2 was created in mice . Rfx2-/- mice are perfectly viable but show complete male sterility . Spermatogenesis appears to progress unperturbed through meiosis . However , haploid cells undergo a complete arrest in spermatid development just prior to spermatid elongation . Arrested cells show altered Golgi apparatus organization , leading to a deficit in the generation of a spreading acrosomal cap from proacrosomal vesicles . Arrested cells ultimately merge to form giant multinucleated cells released to the epididymis . Spermatids also completely fail to form the flagellar axoneme . RNA-Seq analysis and ChIP-Seq analysis identified 139 genes directly controlled by RFX2 during spermiogenesis . Gene ontology analysis revealed that genes required for cilium function are specifically enriched in down- and upregulated genes showing that RFX2 allows precise temporal expression of ciliary genes . Several genes required for cell adhesion and cytoskeleton remodeling are also downregulated . Comparison of RFX2-regulated genes with those controlled by other major transcriptional regulators of spermiogenesis showed that each controls independent gene sets . Altogether , these observations show that RFX2 plays a major and specific function in spermiogenesis .
Reproductive failure affects 10–15% of couples worldwide , with responsibility distributed about equally between males and females [1 , 2] . Failure of spermatogenesis is a common cause of male infertility . A large fraction of such cases is believed to result from genetic causes . The advent of forward genetics has led to the identification of over 400 genes associated with male spermatogenic defects [1] . However , the precise etiology of most clinical cases of male infertility remains unknown . The process of sperm production in the testis is usually described in terms of three phases: the multiplication of diploid spermatogonia , the reduction of chromosome number during meiosis in spermatocytes , and the morphological conversion of round , immotile haploid spermatids into nearly mature sperm by the process of spermiogenesis . The overall process is controlled by master external regulators , including retinoic acid , the pituitary gonadotropins , and testosterone [3] . In contrast , details of the process depend in part on local biochemical communications with closely associated Sertoli cells , and a dynamic gene expression program within the germ cells that involves both transcriptional and post-transcriptional regulation [4 , 5] . The current study addresses the role of a member of the Regulatory Factor X ( RFX ) family of transcriptional regulators . Early studies led to description of the X-box as a DNA sequence motif conserved in the promoters of genes encoding MHC class II antigen-presenting proteins [6] . Search for the factor that activates MHC class II genes through this motif led progressively to the identification of a family of RFX transcription factors ( TFs ) that now numbers 8 members in mammals [7 , 8] and has its evolutionary origins traced to microorganisms . The most fundamental shared feature is a variant of the winged helix DNA binding domain [9] . In the invertebrate metazoan C . elegans there is single RFX relative ( DAF-19 ) , which controls genes important for cilia development and function [10] . In vertebrates several Rfx genes have also been shown to control ciliogenesis ( for review see [11 , 12] ) . The family can be divided into two groups based on their ability to form homo and heterodimers: RFX 1–4 , 6 and 8 are known or predicted to have this property , while RFX5 and 7 do not . This capacity to bind as heterodimers to the same DNA motif makes unambiguous assignment of functional roles to individual RFX proteins difficult when multiple members are present in the same cell . Mutational analysis has established critical roles for most Rfx genes . Rfx1 deletion in mice was found to be embryonic lethal , in keeping with reports of a wide variety of potential roles [13] . Mouse Rfx3 is critical for cilia development and function , and its absence leads to a variety of severe defects in left/right body patterning , CNS development and differentiation of endocrine cells in the pancreas [14–16] . Disruption of mouse Rfx4 causes failure in the development of dorsal midline brain structures . It is also critical for formation of the subcommissural organ [17–19] . Rfx5 was the object of the original search for the X box regulatory factor , and is strictly required for MHC class II gene expression in humans and mice [20] . Genetic ablation of Rfx6 leads to failure of pancreatic islet development and diabetes in zebra fish and humans [21 , 22] . Rfx7 deletion was recently shown to affect ciliogenesis in the developing neural tube in Xenopus , where it functions developmentally upstream of Rfx4 [23] . Rfx8 has only been identified by genomic similarity , and functional studies have not been reported ( http://www . uniprot . org/uniprot/D3YU81 ) . In non-mammalian vertebrates Rfx2 is crucial for the differentiation of cells carrying motile cilia and for the development of left/right asymmetry [24–27] . In contrast , the function for Rfx2 in mammals is not well established . Initial expression profiles established in mice demonstrated extremely high transcript levels in testis compared to other organs [28] . This was amply confirmed by multiple genome wide studies ( http://germonline . org/ ) . Developmental studies revealed an initial increase in Rfx2 mRNA during the meiotic phase of spermatogenesis [29–31] . This was confirmed at the protein level by immunohistology , showing that RFX2 is restricted to the germ line and not detected in Sertoli cells [32] . RFX2 has also been implicated in the upregulation of several genes in that period [33 , 34] . Regarding its own regulation , the Rfx2 promoter region contains multiple binding sites for the MYB family of transcriptional regulators [33] , of which A-Myb ( Mybl1 ) is critically required for spermatocytes to complete meiosis [35 , 36] . Because RFX2 expression is greatly reduced in A-Myb deficient mice , it was predicted that Rfx2 is controlled by A-MYB and in turn controls an unknown population of downstream genes important for spermatogenesis [37] . To test this last prediction we inactivated the Rfx2 gene in mice . Rfx2-/- mice develop normally and are healthy , but males are sterile whereas fertility of female mice is not affected . We observed that spermatogenesis proceeded essentially normally through meiosis , but that all spermatids failed to progress past the round cell stage and were shed from the germinal epithelium at approximately step 7 of development . We observed a general defect in Golgi organization and acrosome formation associated with complete failure of axoneme elongation . RNA-seq based transcription-profiling performed with Rfx2-deficient testes at P21 and P30 identified over 100 and 600 genes , respectively , that were downregulated greater than two-fold . ChIP-Seq analysis at P21 identified many of these as being direct RFX2 targets , with a substantial number of them sharing a developmental expression pattern similar to that of Rfx2 . Interestingly , this large group of downregulated genes shares almost no overlap with genes regulated by other transcription factors required for spermiogenesis [38–40] . These results clearly place RFX2 among the major transcriptional regulators of the haploid phase of sperm formation .
A null allele was created for Rfx2 by flanking exon 7 with loxP sites , creating targeted ES cells and then chimeric mice that were subsequently bred to obtain germline transmission . Subsequent Cre-induced recombination resulted in the excision of exon 7 ( see Materials and Methods and Fig 1 ) . Removal of exon 7 deletes most of the coding region for the DNA binding domain , and is predicted to change the reading frame and introduce premature out-of-frame stop codons if either exon 5 or 6 is spliced to exon 8 , 9 or 10 . Interbreeding of heterozygous exon 7 deleted mice generated the three expected genotypes for both sexes in the expected Mendelian ratios . Exon 7 was shown to be absent by RT-PCR of testis RNA from Rfx2-/- mice ( Fig 1E ) . RNA-Seq experiments ( see below ) confirmed the deletion as no reads were obtained for exon 7 . RFX2 protein was undetectable by Western blotting ( Fig 1E ) and immunohistology ( Fig 2A and 2B ) . Rfx2-/- pups grow normally and show no developmental defects . Hence Rfx2-/- mice do not exhibit the ciliopathy hallmarks observed in Rfx3-/- mice or other phenotypes observed for Rfx-deficient mice [13 , 14 , 17–19 , 21] . Rfx2-/- males are sterile , but Rfx2-/- females have no obvious reproductive defects , as Rfx2-/- females ( n = 7 ) produced litters that were comparable to Rfx2+/- females ( average litter size was 8 for both genotypes ) . Initial morphological characterization showed that young adult Rfx2-/- males had slightly smaller testes but that the remainder of the male reproductive tract and accessory glands , such as seminal vesicles , appeared normal . Microscopic examination ( Fig 2E–2H ) revealed that the epididymis contained no sperm but instead large numbers of degenerating small cells and cell debris . The overall microscopic architecture of the testis was normal , but there was a complete block of spermatogenesis prior to the point where round spermatids should begin to elongate and develop characteristic features of spermatozoa ( Fig 2C and 2D ) . Cells at this stage instead formed multinucleated cells with condensed nuclei , also called symplasts in some studies [41] , and were released from the tubules to flow into the epididymis ( white arrowhead , Fig 2B ) . One of the most readily observed changes between the end of meiosis and the beginning of spermatid elongation is formation of the acrosome , which is characteristically stained by the periodic acid-Schiff ( PAS ) reagent due to its high glycoprotein content . Light microscopic examination of testes from 30-day-old Rfx2-/- mice , in which the first wave of developing germ cells has only recently encountered the point of arrest , showed that Golgi-derived proacrosomic vesicles appear as expected . However , they frequently fail to attach to the nucleus , and the acrosome becomes increasingly disorganized in appearance ( Fig 3A–3H ) . This can be seen in greater detail using fluorescently tagged peanut agglutinin to label acrosomes ( Fig 3I–3P ) . Almost no spermatids showing a typical cup-shaped acrosome spread over the nucleus , as observed in wild type ( WT ) step 7 spermatids , can be visualized in Rfx2-/- testis sections . At approximately step 6–7 the mutant spermatids begin to fuse into multinucleate giant cells . WT and Rfx2-/- testis sections were stained by the TUNEL technique to ascertain if the arrest in spermatid development was associated with increased apoptosis . No difference in the distribution or number of TUNEL positive cells was detected ( S1A and S1B Fig ) . Differences in the patterns of phospho-H2AX staining in round spermatids or the condensed nuclei in multi-nucleated bodies were also not observed ( S1C Fig ) . Nuclear condensation in multinucleated cells is thus not due to induction of apoptosis . We next examined whether the appearance of smaller nuclei among the multi-nucleate bodies could be due to premature or aberrant synthesis of proteins involved in the normal histone to protamine transition . This stepwise process involves multiple proteins including transition proteins TNP1 and TNP2 , major players in this process [42] . mRNAs for TNP1 and 2 are not reduced in the knockout mice , as revealed by our RNA-Seq analysis described below . Like many transcripts present at the end of the round spermatid stage , these mRNAs are translationally repressed until the appropriate time . However , TNP1 was not detected by Western blotting at all developmental stages ( S2A Fig ) , whereas TNP2 was present in Rfx2-/- testis even though spermatid development does not reach the stage ( steps 9–10 ) at which it first appears in WT animals ( S2A Fig ) . When sections were stained for TNP2 it was detected largely over the cytoplasm of terminal stage Rfx2-/- spermatids , in contrast with its nuclear localization in Rfx2+/+ mice ( S2B and S2C Fig ) . This suggests that disorganized development may proceed to the point where translational repression of TNP2 mRNA is released , allowing accumulation of TNP2 protein , but that the latter fails to be transported efficiently into the nucleus . These results suggest that premature nuclear accumulation of transition proteins is not likely to be responsible for nuclear condensation in multinucleated cells . We cannot exclude that altered protamine levels could be involved in nuclei condensation , but this would likely be an indirect consequences of earlier nuclear defects as protamines are produced after the spermatid arrest observed in Rfx2-/- testes . EM analysis was performed to characterize in more detail the cellular defects observed in Rfx2-/- spermatids . Shortly after meiosis in WT spermatids , cell polarity is established such that the Golgi apparatus is oriented with the trans-Golgi facing the nucleus . This favors the transport of budding vesicles to the nuclear surface , where they attach to the nuclear lamina [43 , 44] . The migration of Golgi vesicles to the nucleus is guided by microtubule tracks and mediated by molecular motors and RAB adapter proteins [45–48] . At the opposite pole of the cell , a single axoneme develops near the plasma membrane from one of the two centrioles and gradually elongates as the basal body migrates inwards toward the nucleus [43] . These processes do not occur correctly in Rfx2-/- mice . The Golgi frequently appears disoriented such that dense-cored proacrosomic vesicles migrate away from the nucleus ( Fig 4G–4M ) . These vesicles fuse to form large dense-cored discs that do not attach to the nucleus ( Fig 4G , 4H , 4I and 4J ) . In other cases , vesicles have fused with the nucleus to form a partial acrosomal cap that only rarely spreads to cover half of the anterior end of the nucleus , as is the case for WT mice ( compare Fig 4B–4D with 4H–4J ) . Occasionally , highly atypical structures are found , such as the development of an acrosome having apparently engulfed multiple vesicles containing cytoplasm ( Fig 4I ) . Nuclei more often show protruding aneurysm-like ruptures in Rfx2-/- cells ( Fig 4L ) . No signs of developing axonemes are found , although these are evident in WT mice ( compare Fig 4E and 4F with 4M and 4N and insets ) . This is in agreement with the absence of acetylated tubulin stained flagella in the lumen of seminiferous tubules ( S3 Fig ) . Interestingly , centriole pairs are found , and can occur in clusters in the giant multinucleated cells ( Fig 4N–4R ) . In WT spermatids , once the acrosome cap has covered slightly more than half the nucleus , the nucleus and associated acrosome move to contact the plasma membrane , which becomes the site of new specialized junctions , the apical ectoplasmic specialization , between the germ cells and Sertoli cells [49 , 50] . In WT mice , these will control Sertoli cell—germ cell connections until the normal release of mature spermatids at spermiation . In Rfx2-/- spermatids the nucleus/acrosome is not observed to merge with the plasma membrane , suggesting that these junctions do not form . Coincident with this failure , mutant spermatids undergo fusion to form large multinucleate cells . Prior to this point , the mutant spermatids , like all differentiating male germ cells , maintain open cytoplasmic connections via so-called cytoplasmic bridges characterized by well-defined boundaries [51] ( Fig 4K ) . Breakdown of the supporting boundaries of these bridges presumably accompanies the formation of giant cells , but this could not be formally demonstrated from the ultrastructure images , suggesting that once begun this is a rapid process . With the failure to form and maintain the ectoplasmic specialization junctions with Sertoli cells , the spermatids , now largely or entirely in the form of giant cells , are released to pass out of the testis and into the epididymis . Another prominent ultrastructural feature of the maturing round spermatid is formation of the manchette , a dense arrangement of microtubules attached to the outer edge of the nucleus in a ring just below the border of the acrosome . The manchette is believed to be important for the nuclear shaping that marks the end of the round phase of spermatid differentiation [52 , 53] . Although no manchette is observed in Rfx2-/- spermatids , dense microtubule clusters are seen in the multinucleated cells , perhaps representing ectopic manchette formation ( Fig 4S ) . Staining of testis sections for the cis-Golgi marker GM130 revealed a severely altered GM130 distribution in Rfx2-/- mice ( Fig 5 ) . Whereas GM130 is mainly concentrated at one pole of WT spermatids , it is also diffusely distributed around nuclei in Rfx2-/- spermatids , illustrating an altered Golgi re-distribution at the onset of the haploid stage . RNA-Seq based transcriptome profiling of testis cells from Rfx2+/+ and Rfx2-/- mice was performed to assess the consequences of the loss of RFX2 function . Animals were chosen at postnatal days 21 ( P21 ) and 29–30 , henceforth referred to as ( P30 ) as there are no significant changes in germ cell populations in this time window . At P21 , the most advanced tubules contain only very early round spermatids while pachytene spermatocytes are plentiful . At this stage , defects in spermatogenesis are not yet histologically evident in Rfx2-/- mice . At P30 , the first wave of germ cells has largely reached the point of arrest observed in Rfx2-/- mice , but potential long term cumulative effects on seminiferous tubules should be minimal . Global representations of RNA-Seq experiments comparing the transcriptomes of WT and Rfx2-/- testes at days P21 and P30 are shown in S4 Fig . At P21 , 106 genes were down regulated ( p < 0 . 01 ) greater than 2-fold , of which 47 were decreased more than 10-fold ( Fig 6A , S1 Table ) . By P30 , the number of 2-fold downregulated genes increased to 640 , with 151 being decreased 10-fold or more ( Fig 6A , S1 Table ) . Most genes downregulated at P21 were also downregulated at P30: 95 genes were affected at both time points ( Fig 6B and 6C ) . Markedly fewer genes were upregulated more than 2-fold: 67 genes at P21 and 128 genes at P30 ( Fig 6A , S1 Table ) . Only 9 genes were upregulated significantly at both time points ( Fig 6A , S1 Table ) . To identify genes that are downregulated in Rfx2-/- testis and exhibit a developmental expression program consistent with activation by RFX2 in WT testis , we used the WT expression data described in a recent RNA-Seq analysis of mouse spermatogenesis [29] . Rfx2 expression first increases significantly by day 14 , as cells reach early-mid pachytene , and then increases dramatically between days 17 and 21 , as the leading cells complete meiosis ( Fig 7 ) . Among genes that are downregulated in Rfx2-/- testes and exhibit statistically reliable expression profiles in the data of Laiho et al [29] , the majority ( 47/55 at P21 and 226/281 at P31 ) exhibits expression profiles during spermatogenesis consistent with activation by Rfx2 , ( Fig 7 , S2 Table ) . Searches for transcription factor binding sites in the promoters of differentially expressed genes , revealed that RFX binding motifs are highly enriched in genes that are upregulated at P21 or downregulated at P21 and/or P30 ( S5 Fig ) . Similarly , an unbiased motif discovery approach identified a 14 bp inverted repeat exhibiting strong homology to the consensus RFX binding site ( X box ) in the promoter regions ( -500 to + 50 bp relative to the predicted transcription start site , TSS ) of genes that are upregulated at P21 and downregulated at P30 in Rfx2-/- testis ( Fig 6D ) . Overall , 159 differentially expressed genes contain an X-box in their promoter regions ( S1 Table ) . These observations suggest that RFX2 directly regulates these genes . To identify genes that are regulated directly by RFX2 in spermatocytes and early spermatids , we carried out chromatin immunoprecipitation followed by high throughput sequencing ( ChIP-Seq ) , using dissociated germ cell preparations from P21 WT mice . We identified nearly 3 , 000 reproducible and statistically-significant peaks corresponding to binding of RFX2 ( Fig 8 ) . Of these peaks , about 1/3 ( 977 ) were located within presumptive promoter regions ( -500 to- +50 ) , whereas the remaining were distributed in introns , exons and intergenic regions ( Fig 8A ) . The most robust ( lowest p-value ) peaks were preferentially enriched in promoter regions ( Fig 8B ) . Representative promoter peaks are shown in S6 Fig . The distribution of promoter peaks showed a marked concentration near the TSS ( Fig 8C ) . Genes with promoter peaks will henceforth be designated RFX2 targets . An unbiased search for overrepresented sequence motifs in the ChIP peaks again identified a 14 nucleotide motif exhibiting an almost perfect match to consensus binding sites for RFX1 and RFX2 ( Fig 8D ) , illustrating that the RFX2 binding motif found in spermatogenesis genes is not different from previously established X-box motifs . As expected , the RFX motifs are mostly located near the center of the ChIP peaks ( Fig 8E ) . In summary , these data clearly show that RFX2 binding sites are located very close to the TSS in the majority of target genes . Among RFX2 target genes identified by ChIP-seq , 138 were differentially expressed in Rfx2-/- testes ( Fig 8F and S1 Table ) , 111 being downregulated and 27 being upregulated . This pinpoints these genes as primary RFX2 target genes in the mouse testis . A substantial fraction of these target genes exhibit expression patterns during spermatogenesis [29] consistent with activation by RFX2 ( compare upper and lower left-hand panels in Fig 9A and 9B ) . These genes are therefore likely to be directly regulated by RFX2 in the testis . Gene ontology analysis ( see Supplemental methods ) revealed a specific and significant enrichment ( p <0 . 001 ) in genes associated with GO terms related to ciliogenesis—such as cilium morphogenesis ( GO:0060271 ) , cilium organisation ( GO:0044782 ) , cilium assembly ( GO:0042384 ) or cell projection GO:0042995—for both the set of direct RFX2 targets identified by ChIP-seq and the set of differentially expressed genes in Rfx2-/- testis ( S7 and S8 Figs , S3 Table ) . We therefore examined the list of downregulated genes for those related to cilia using two reference lists . The first is the database “CilDB” [54] . The second is the “gold standard” list established by the Syscilia consortium [55] . CilDB provides a score corresponding to the number of times a gene is found with low , medium or high confidence in various studies aiming at identifying cilia-associated genes . Among genes that are downregulated in Rfx2-/- testis , 52/106 ( 49% ) at P21 and 231/640 ( 36% ) at P30 are potentially involved in ciliogenesis , as they are found with high confidence in at least one cilia-related study documented in the CilDB database and/or in the Syscilia gold standard list ( Fig 10A , S1 Table ) . Of these , 43/52 at P21 ( 82% ) and 148/231 at P30 ( 64% ) are down regulated more than 3-fold ( Fig 10A , S1 Table ) . Many of these downregulated cilia-related genes are direct targets of RFX2 ( Fig 10B , S1 Table ) . Among genes that are both differentially expressed in Rfx2-/- testis and direct targets of RFX2 , 80/138 ( 57% ) are found with high confidence in at least one cilia-related study documented in the CilDB database and/or in the Syscilia gold standard list ( S1 Table ) . Most of these genes exhibit developmental expression profiles [29] , similar to Rfx2 ( Fig 9A and 9B right-hand panels ) . Examination of our ChIP-Seq data for a selection of cilia-associated target genes downregulated at P21 and P30 clearly showed peaks in their promoter regions ( see S6 Fig for examples ) . Finally , among 28 cilia-related genes that are upregulated at P21 , 11 are direct targets of RFX2 ( S9A Fig ) . These results clearly indicate that genes involved in ciliogenesis are primary targets of RFX2 in the mouse testis . To understand how RFX2 deficiency could lead to specific Golgi-associated and acrosome-formation defects , we examined all GO terms associated with differentially expressed genes at P21 or P30 . Many genes deregulated in Rfx2-/- testis were found to match to GO terms associated with intracellular transport , microtubule associated processes or small GTPase regulation , even though these terms were not significantly enriched . For example , of genes that are downregulated in Rfx2-/- testis , 35 belong to gene sets corresponding to the GO terms “microtubule” or “acrosome” , and a substantial fraction of these genes are direct RFX2 targets ( Fig 10C ) . Impaired expression of these genes could contribute to defective cell polarization , proacrosomic vesicle migration and acrosome formation during the first steps of spermatid differentiation . GO terms that are enriched with a p-value below threshold ( 0 . 01<p<0 . 001 ) , include terms linked to cell-adhesion or cytoskeleton organization . Genes associated with these terms could account for some of the spermatid defects observed in Rfx2-/- mice , in which spermatids lose their contacts with adjacent Sertoli cells and form giant multinucleated cells . These genes include Fndc3c1 , which is a paralogue of Fndc3a . The latter is mutated in the mouse sys strain ( S4 Table ) , which has a phenotype strikingly similar to that of Rfx2-/- mice in that it is characterized by the formation of giant multinucleated cells referred to as symplasts [56] . Another interesting candidate is Fascin 1 , which is down regulated both at P21 and P30 , and is known to be involved in the ectoplasmic specialization required for spermatid-Sertoli cell contacts in the rat testis [57] .
Genome wide analyses of their effects on transcriptional programs in testis have been performed for mutations in Crem and Taf7l [39 , 40] . Crem occupies a special place among genes encoding transcriptional regulators of spermatogenesis . In the testis it produces a specific splice variant encoding CREM-tau , a unique isoform of the cAMP-responsive element modulator ( CREM ) , which comprises a family of isoforms that bind to the cAMP response element ( CRE ) . While most of the splice variants produce inhibitory factors , CREM-tau is stimulatory and appears late in meiosis [62] . Crem was one of the earliest targets of engineered mutations [60 , 61] and proved to be essential for expression of a large number of classic genes expressed in the haploid phase of spermatogenesis . A recent genome wide study identified 627 genes that were downregulated greater than 2-fold in Crem-/- testis [39] . Of these CREM-regulated genes , 277 were shown by ChIP-Seq experiments to be direct targets of CREM in male germ cells [63] . Only 34 of the CREM-regulated genes and 9 of the CREM-occupied genes are downregulated greater than 2-fold in Rfx2-deficient testis ( S10A Fig ) . Furthermore , RFX2 and CREM regulated gene sets exhibit largely distinct patterns of expression during spermatogenesis ( S10B Fig ) . Direct RFX2 target genes tend to parallel the expression pattern of RFX2 itself , starting from little or no significant expression at P7 and rising to half maximal by or before P21 . In contrast , direct CREM targets have extremely variable expression at P7 but tend to share large increases after P21 ( S10B Fig ) . Genes that depend directly on RFX2 or CREM thus constitute nearly distinct sets . Spermatogenesis exhibits a striking dependence on variants of the Pol II GTFs [64] , which include TBL1 ( TRF2 ) [59] , TAF4b [65] , TAF7l [40] , and GTF2a1l ( ALF ) [66] . TAF7l , a variant of TAF7 , is a component of TFIID . A targeted mutation of Taf7l led to an arrest in spermatogenesis at the end of the round spermatid stage . Transcriptome analysis of Taf7l-deficient and WT testis [40] identified some 1 , 440 genes that were down regulated by more than two-fold , and 726 by over three-fold , in the mutant mice . Furthermore TBL1 , whose ablation also leads to a developmental arrest of round spermatids [59] , was found to co-occupy active promoters with TAF7L , and the two GTFs were suggested to function together on a subset of postmeiotic genes . Remarkably , none of the genes found to be downregulated in Taf7l-deficient testis are also deregulated in RFX2 or CREM deficient testis . This is striking support for the model that there are relatively few master transcriptional regulators for the postmeiotic phase , and that each one controls largely separate groups of genes . Gtf2a1l ( ALF ) is yet another GTF expressed during spermatogenesis [66 , 67] . Targeted mutations have not been reported for this gene . Gtf2a1l was shown to have a binding site for RFX factors [34] , although it was not identified by our ChIP-Seq experiment as being a direct RFX2 target . It’s expression is however reduced slightly more than 2-fold in the Rfx2-/- mouse . This reduction may contribute to the broad spectrum of indirectly downregulated genes . A strikingly large number of Rfx2-regulated genes is associated with the formation and function of cilia and flagella , which are complex organelles that are highly conserved among eukaryotes . In mammals , cilia occur as single non-motile cilia on a wide range of cells , motile monocilia on the embryonic node and multiple motile cilia on various epithelia , whereas flagella are found on sperm cells . The regulation of cilia formation and function by RFX factors traces back to metazoans [10] . As the family expanded through evolution , multiple family members remained associated with this function , including RFX3 and RFX4 in mammals , as well as RFX2 and RFX7 in non-mammalian vertebrates [14 , 17 , 18 , 23 , 24 , 26 , 27] . This study underlines a central role for mouse RFX2 in the regulation of cilia associated genes during spermatogenesis . RFX proteins control specific subsets of core genes required for cilia formation from C . elegans to mammals ( for review see [11 , 12] ) . In particular , RFX factors regulate genes encoding intraflagellar transport ( IFT ) components and Bardet-Biedl syndrome ( BBS ) proteins required for building cilia . RFX factors were also shown to regulate transition zone components in C . elegans , Drosophila , Xenopus and mice [11 , 12 , 68 , 69] . Transition zone proteins form a specialized structure at the base of the axoneme that connects the distal end of the basal body to the plasma membrane ( for review see [70 , 71] ) . Whereas many of the IFT genes are indeed bound by RFX2 , none are downregulated in Rfx2-/- testes at either time point ( S1 Table ) . Many of the Bbs genes are also bound by RFX2 , but none are downregulated more than 2-fold in Rfx2-/- testes . However , expression of some of these genes is significantly decreased by slightly less than the arbitrary 2-fold threshold applied for inclusion in S1 Table . As for IFT and BBS genes , most of the transition zone associated genes are also bound by RFX2 in the testis ( Rpgrip1L , B9D1 , D2 , Tctn1 , Tctn2 , Tctn3 , Tmem216 , Tmem138 , MKS1 , Tmem17 , Tmem231 , Tmem237 , Cep290 , Sdccag8 ) . Only two of these are upregulated at P21 ( Tcnt1 ) or downregulated at P30 ( Tmem231 ) . Nphp3 , Nphp4 , and Nek8 are downregulated in Rfx2-/- testis but do not appear to be direct RFX2 targets . Collectively , these results show that RFX2 binds to the promoters of many core genes required for building cilia , but does not seem to play a dominant role in controlling their expression . This implies that other TFs are implicated in regulating genes encoding IFT , BBS and transition zone components in the testis . Other RFX family members expressed in the testis , such as RFX1 or RFX3 , are good candidates , as they could function redundantly with RFX2 to regulate core ciliary genes . Surprisingly , several IFT ( IFT74 , IFT81 ) and cilia associated genes are upregulated at P21 in Rfx2-/- testes ( S9A Fig , S1 Table ) . This may imply that RFX2 represses these genes during spermatogenesis . However , since RFX factors can function as homo or heterodimers , an alternative hypothesis is that a controlled balance between different dimeric RFX complexes is required for precise tuning of ciliary gene expression . Perturbing this balance by ablating RFX2 may thus generate new RFX complexes capable of activating these upregulated genes . In contrast to the core ciliary genes , many genes required for ciliary motility are downregulated in Rfx2-/- testes . Among 32 genes in which defects lead to primary ciliary dyskinesia , six ( Dnaic2 , Ccdc40 , Armc4 , Ccdc164/DRC1 , Rsph9 , Ccdc11 ) are strongly downregulated in Rfx2-/- testes at both time points , and three others only at P30 ( Ccdc65 , Zmynd10 , Ccno ) ( for review see [72] ) . Many genes coding for structural components of flagella or centrioles are also downregulated ( S1 Table ) . RFX2 is thus only critical for the expression of a specific subset of ciliary genes in the testis . In zebra fish and Xenopus , RFX2 has been implicated in the motile cilia differentiation program required for the establishment of left/right asymmetry during embryonic development and the formation of motile cilia in various epithelia [24 , 26] . We did not observe any major defects in left/right asymmetry or multiciliated epithelia in Rfx2-/- mice in two different genetic backgrounds . This is unexpected as Rfx2 is expressed strongly in the mouse embryonic node and multiciliated epithelia [25 , 27] . The absence of aberrant left/right or ciliated-epithelia phenotypes in Rfx2-/- mice could be due to partial redundancy between RFX transcription factors . As Rfx3-/- mice display left/right asymmetry defects , it is possible that RFX3 ( or other RFX factors ) can fully compensate for the deficiency in RFX2 , whereas RFX2 cannot compensate efficiently for the loss of RFX3 . The opposite seems to be true in the testes , as the few Rfx3-/- males that survive past birth do not show defects in sperm or flagella production ( B . Durand , personal communication ) , showing that RFX2 and RFX3 do not have redundant functions in this cell type . In summary , our study implies that RFX TFs have evolved to regulate specific sets of cilia-related genes in different tissues . RFX2 appears to have become specialized for regulating a specific subset of cilia/flagella-related genes required for spermatid development . Double mutants will need to be studied to further understand how different RFX factors exert their respective functions in different mouse tissues . Furthermore , the complex interplay between members of the RFX family , notably Rfx2 and Rfx3 , in the regulation of cilia-related genes in different tissues can now be sorted out genetically using the available conditional mutants . Several mutant mice having defects in RNA biogenesis or small-RNA regulated processes show a developmental arrest in spermatogenesis similar to that of Rfx2-/- mice ( S4 Table ) . The genes mutated in these mouse models are not regulated by RFX2 . However , the similar phenotype prompted us to examine whether other genes associated with RNA processing and small RNA pathways could be regulated by RFX2 . A recent proteomic analysis identified 88 proteins associated with the Chromatoid Body , which is generally believed to be the site of post-nuclear RNA processing and may also contain translationally suppressed mRNAs [73] . Of these proteins , TEKT4 ( tectin 4 ) is encoded by a gene downregulated in Rfx2-/- testis at P30 . Another recent study demonstrated that many genes required for acrosome or flagella function are downregulated in the maelstrom mutant , which carries a mutation in the gene encoding MAEL , a conserved core component of the piRNA pathway [74] . This suggests that RFX2 and MAEL regulated pathways could converge on common genes , and that this might underlie the comparable phenotypes observed in Rfx2-deficient mice and piRNA biogenesis pathway mutants . However , we cannot exclude that RFX2 could regulate the expression of small RNAs that our RNA-seq analysis could not reveal . The transcription factor A-MYB has also been shown to regulate piRNA expression in mouse germ cells and many A-MYB target genes have been identified by ChIP-Seq experiments [75] . The binding motif identified in promoters of A-MYB target genes , including piRNA clusters , does not overlap with the well-defined RFX binding motif [75] . In addition , more than 90% of A-MYB targets identified by ChIP-Seq are unique to A-MYB and do not overlap with RFX2 targets . Thus , although A-MYB could regulate RFX2 [33] , the two transcription factors do not regulate the same sets of target genes . Early in normal spermatid development , a crucial polarity develops such that the acrosome will form and define the anterior end of the sperm , while at the opposite cellular pole , the centriole pair will move to the plasma membrane and initiate growth of the axoneme , which will eventually grow to form the interior elements of the flagellum [43 , 44] . Failure of normal acrosome cap formation is a characteristic of polarity loss . The mechanisms that set up this polarity are not well understood in spermatids . The PAR3-PAR6-aPKC complex was shown to be required to establish cell polarity [76] , but we did not observed any defects of the distribution of these markers ( S3 Fig ) , suggesting that only downstream actors of this complex are likely to be affected in Rfx2-/- mice . Cell polarity in many systems is set up by cell junctions with surrounding tissue and involves a set of highly conserved genes . Spermatids are connected to Sertoli cells by adherens-like junctions at this stage [50] . Mutation of junctional adhesion molecule-C ( JAM-C ) resulted in general polarization failure of spermatids and acrosomic failure similar to that observed in Rfx2-/- mice [77] . Around 20 genes required for cell adhesion or cell junctions are downregulated in Rfx2-/- testes , which could account for the observed spermatid defect . Microtubules and associated motors responsible for movement of vesicles and macromolecules are effectors of cell polarization , and required for Golgi organization and acrosome formation in spermatids [47 , 48 , 78 , 79] . 24 genes mapping to Golgi GO terms are downregulated in Rfx2-/- testes ( S9B Fig ) and this may contribute to disruption of overall Golgi organization and acrosome formation in Rfx2-/- spermatids . For example , RAB27a/b is known to function as an adapter for acrosome-bound vesicles [46] , and Rab27b is down regulated ~3-fold . 20 other downregulated genes also have GTPase regulator activity . Furthermore , some 28 microtubule-related genes are downregulated in Rfx2-/- testis , which could also contribute to overall polarity and vesicular transport failure ( Fig 10C ) . Hence , defective spermatogenesis in Rfx2-/- mice is likely to result from alterations in several different cellular pathways .
The conditional targeting vector ( Fig 1A ) carried loxP sites inserted into the introns flanking exon 7 , which encodes a large segment of the DNA binding domain . Deletion of exon 7 is predicted to alter the reading frame if either exon 5 or 6 is spliced to exon 8 ( exon 6 is variably included in the mature transcript ) . The vector also contained a FRT-flanked neomycin selection cassette . The vector was electroporated into R1 ES cells ( strain 129/Sv ) by the Mouse Biology Program at University of California , Davis . Targeted clones were identified using PCR and primer sets with outside members located both upstream and downstream of the homology region of the vector ( Fig 1B and 1C ) . Correctly targeted clones were introduced into C57BL/6 blastocysts to generate chimeras . These were bred to C57Bl/6 mice and offspring tested for transmission by PCR analysis of tail biopsy extracts ( Fig 1D ) using primers described in S5 Table . For initial experiments exon 7 was removed by breeding to a universal Cre-expressing strain , FVB/N-Tg ( ACTB-cre ) 2Mrt/3 , and deletion of exon 7 in offspring was verified by PCR ( Fig 1D ) . Heterozygous mice of either gender were fertile and were mated together to generate Rfx2-/-neo+ mice , which were maintained as a mixed background line . Age-matched littermates were used whenever possible . A second line , without the neo cassette , was developed by breeding to a flipase expressing strain , B6;SJL-Tg ( ACTFLPe ) 9205Dym/J [80] . Offspring were selected for loss of the neo cassette and flipase transgene . This strain was backcrossed onto a C57BL/6 genetic background . The phenotype of Rfx2-/- animals appears identical with respect to testis morphology for animals from either the C57BL/6 or mixed background lines . Mice were treated in accordance with the American Veterinary Medicine Association ( AVMA ) Guidelines for Euthanasia of Laboratory Animals and with French Institutional guidelines in compliance with French and European animal welfare regulations ( authorization n° 562 , French Ministry of Education and Research ) . All animal studies were approved by the Institutional Animal Care and Use Committee ( IACUC ) of the University of South Carolina ( AUP No 1818 ) and by the Ethics Committee of the University of Lyon-1 ( BH 2008–01 ) . Animals were sacrificed by CO2 anesthesia . Testes were fixed in Bouin’s solution and epididymis in PBS containing 4% paraformaldehyde and processed to produce paraffin sections . Dewaxed sections were stained with hematoxylin and eosin or by PAS using Light Green SF Yellowish as counterstain . SDS extraction of testes and Western blotting were carried out as previously described [81] . Separation was in an 8% ( RFX2 ) or 15% gel ( TNP1 , TNP2 ) . Primary antisera to TNP1 ( 1:5000 ) and TNP2 ( 1:5000 ) [82] , RFX2 ( 1:250 , C-15 , Santa Cruz Biotechnology , Santa Cruz , CA ) and actin ( 1:1000 , A-2066 , Sigma-Aldrich , St Louis , MO ) were detected with HRP-conjugated donkey anti-rabbit IgG and anti-goat IgG ( 1:1000 , Jackson ImmunoResearch Laboratories , West Grove , PA ) and chemiluminescence [83] . For semi-quantitative RT-PCR , total RNA ( 1 μg ) was reverse transcribed using the Superscript III First Strand Synthesis Kit ( Invitrogen/Life Technologies ) and random hexamer primers according to the manufacturer’s directions in a final volume of 20 μl . After a 1 to 10 dilution , 1 μl was used for a 50 μl PCR reaction . Control reactions lacking reverse transcriptase were done in parallel . Mice ( P25 to P40 ) were anesthetized by i . p . injection of sodium pentobarbital ( 0 . 1 ml/100g b . w . —CEVA santé Animale , France ) and fixed by intra-cardiac injection ( adapted from [84] ) with 2% glutaraldehyde and 0 . 5% paraformaldehyde in 0 . 1 M Sodium cacodylate , pH 7 . 35 . Testes were cut into cubes of 1 mm3 in fresh fixative solution and left for 2 days at 4°C . After extensive washing in 0 . 15 M sodium cacodylate ( 4 x 1h at RT and o/n at 4°C ) samples were postfixed in 0 . 15M sodium cacodylate containing 1% OsO4 for 1 hour at RT , briefly rinsed in distilled water and dehydrated through a graded series of ethanol solutions ( from 30° to 100° ) and two baths of propylene oxide for 15 min each . After substitution and impregnation , small blocks were embedded in epoxy resin in flat silicon molds and polymerized at 56°C for 48h . Ultrathin sections were cut with a UC7 Leica ultramicrotome . Ultrathin sections were contrasted in aqueous uranyl-acetate and lead citrate solutions using a Leica ultrostainer . Sections were observed with a Philips CM120 transmission electron microscope at 120Kv . Testes were isolated from Rfx2+/+ and Rfx2-/- mice ( P30 to P40 ) , fixed o/n in 4% PFA , embedded and frozen in 15% sucrose , 7 . 5% gelatin . Immunofluorescence and apoptosis experiments were performed with 14 μm cryosections . Briefly , samples were permeabilized in 0 . 1% Triton for 5 min , and nonspecific antibody binding was blocked with 3% BSA , 10% FBS . Primary antibodies were incubated o/n at 4°C . Secondary antibodies ( anti-rabbit Alexa 488 or anti-mouse Alexa 594 or Alexa 555 , Invitrogen , both at 1:500 ) were incubated for 2 h at RT . Nuclei were counterstained with DRAQ5 . Fluorescence was observed under a LSM 510 Confocal microscope . Primary antibodies were: anti mouse H2AX ( 1/500 , Millipore ) , Acetylated-tubulin ( 1/1000 , T6793 Sigma ) , Beta-catenin ( 1/50 , C2206 Sigma ) , ZO-1 ( 1/25 , PA5-28858 Zymed ) , PAR3 ( 1/100 , 07–330 Millipore ) , PAR6 ( 1/50 , sc-67393 Santa Cruz Biotechnology ) , aPKC ( 1/400 , sc-216 Santa Cruz Biotechnology ) . PAR6 and aPKC antibodies were kind gifts from M . Montcouquiol .
|
Failure of spermatogenesis , which is presumed to often result from genetic defects , is a common cause of male sterility . Although numerous genes associated with defects in male spermatogenesis have been identified , numerous cases of genetic male infertility remain unelucidated . We report here that the transcription factor RFX2 is a master regulator of gene expression programs required for progression through the haploid phase of spermatogenesis . Male RFX2-deficient mice are completely sterile . Spermatogenesis progresses through meiosis , but haploid cells undergo a complete block in development just prior to spermatid elongation . Gene expression profiling and ChIP-Seq analysis revealed that RFX2 controls key pathways implicated in cilium/flagellum formation , as well as genes implicated in microtubule and vesicle associated transport . The set of genes activated by RFX2 in spermatids exhibits virtually no overlap with those controlled by other known transcriptional regulators of spermiogenesis , establishing RFX2 as an essential new player in this developmental process . RFX2-deficient mice should therefore represent a valuable new model for deciphering the regulatory networks that direct sperm formation , and thereby contribute to the identification of causes of human male infertility .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2015
|
RFX2 Is a Major Transcriptional Regulator of Spermiogenesis
|
Mexicans are a recent admixture of Amerindians , Europeans , and Africans . We performed local ancestry analysis of Mexican samples from two genome-wide association studies obtained from dbGaP , and discovered that at the MHC region Mexicans have excessive African ancestral alleles compared to the rest of the genome , which is the hallmark of recent selection for admixed samples . The estimated selection coefficients are 0 . 05 and 0 . 07 for two datasets , which put our finding among the strongest known selections observed in humans , namely , lactase selection in northern Europeans and sickle-cell trait in Africans . Using inaccurate Amerindian training samples was a major concern for the credibility of previously reported selection signals in Latinos . Taking advantage of the flexibility of our statistical model , we devised a model fitting technique that can learn Amerindian ancestral haplotype from the admixed samples , which allows us to infer local ancestries for Mexicans using only European and African training samples . The strong selection signal at the MHC remains without Amerindian training samples . Finally , we note that medical history studies suggest such a strong selection at MHC is plausible in Mexicans .
In 1492 Columbus discovered America . Europeans , led by the Spaniards , and armed with horses , wheels , germs , and steel , rapidly conquered the New World [1] , and promptly Africans were brought there as slave labor . During the past 500 or so years , three populations—Amerindians , Europeans , and Africans—have occupied the same space and time , albeit asymmetrically , and were genetically admixing . Twenty generations later , the majority of the people inhabiting Central America , Caribbean Islands , and South America , such as Mexicans , Puerto Ricans , and Columbians have become an admixture of the three continental ancestral populations . These recently admixed populations are of great interest for modern genetic studies [2] . In 2007 , Tang and colleagues analyzed a small cohort of Puerto Rican samples and reported three regions that are under strong recent selection [3] . Using their then state-of-the-art local ancestry inference software Saber [4] , Tang and colleagues discovered in Puerto Rican samples genomic regions whose mean local ancestries ( averaged over individuals ) significantly deviated from the genome-wide average—a hallmark of recent selection for admixed samples . Price and colleagues cautioned that the strong selection discovered by Tang and colleagues might be artifacts and they provided three arguments [5] . First , Saber only models linkage disequilibrium ( LD ) , the non-independence of genetic markers in a population , between adjacently markers and thus may produce unreliable local ancestry estimates in regions that harbor long-range LD . It was noted that all three loci under selection that Tang and colleagues reported are within the long-range LD regions . Second , the Amerindian training samples used by Tang and colleagues , which are Maya and Pima samples from human genetic diversity panel ( HGDP ) [6] , is an inaccurate ancestral population for Puerto Ricans , which might produce artifacts in local ancestry inference . Third , Price and colleagues analyzed a larger sample using their software AncestryMap [7] and did not discover the deviation of local ancestry reported by Tang and colleagues . We would like to make the following comments . First , the AncestryMap uses the so called ancestry informative markers ( AIMs ) to infer local ancestry; because that AIMs are sparse and that ancestry informative haplotypes may not contain sufficient number of AIMs , the statistical method underlying AncestryMap is evidently under-powered in detecting local ancestry compared to those that attempt to model haplotypes , particularly more recent model-based methods such as HapMix [8] and ELAI [9] . Therefore , negative results from AncestryMap cannot convincingly refute positive findings by Tang and colleagues . Second , the long-range LD , if properly modeled , will benefit the local ancestry inference , because in regions that harbor long-range LD there are more markers in sync to define population specific haplotypes . Although Saber [4] has difficulty with long-range LD , more recent model-based methods , such as ELAI [9] , can benefit from long-range LD . Third , inaccurate Amerindian training samples is a challenge in studying local ancestry of Latinos . Amerinidan training samples are rarely found in the public domain; the ones that are available , such as Maya and Pima samples from HGDP [6] , have small sample sizes and many samples have non-neglegible European ancestries [10] . In this study we analyzed two datasets whose subjects are of Mexican descent , which we obtained from the database of genotype and phenotype ( dbGaP ) . Our primary motivation is to follow up with selection findings in an early study [9] , which discovered signatures of recent selection in HapMap3 [11] Mexican samples based on a departure of local ancestry from the global average . Our second motivation is to report a method that can overcome the technical challenge presented by inaccurate Amerindian training samples when analyzing local ancestry of Latinos . We devised a novel method to infer local ancestry which allows us to discard Amerindian samples and instead learn Amerindian haplotypes from Mexican samples . The strong selection in the MHC region in Mexicans was confirmed in our study .
In VIVA the global ancestry proportions ( that is , the admixture proportions ) for Amerindian , European , and African components are 0 . 484 , 0 . 452 , and 0 . 064 respectively . In Lipid the numbers are 0 . 552 , 0 . 409 , and 0 . 039 . Compared to Viva , Lipid has a higher Amerindian ancestry proportion and lower European and African ancestry proportions . The sampling location is likely to account for the difference: participants in Lipid were recruited in Mexico City , Mexico , whereas participants in Viva were recruited in Houston , Texas . For each ancestry component , there are substantial variations among individuals ( see two triangular plots in Fig 1 ) . For both datasets , the topological resemblance between the triangular plot and the principal component ( PC ) plot is remarkable . The relative positions of the Mexican outlier individuals are well matched , and an African American individual accidentally recruited in Viva is rather obvious . This suggests that ELAI estimates are sensible , and that using PC to derive admixture proportions has some merits [14] . It is believed that using East Asians as additional proxy to Amerindian training samples may improve the local ancestry inference of Latinos , because Amerindians are genetically more similar to East Asians . Our experience suggests , however , that this practice has little impact , and the PC plots , in which Chinese separate from Amerindians inconsistently in two datasets , seem to corroborate our experience . We computed at each marker the average dosages separately for each ancestral component by averaging that component over all individuals . The average ancestry dosages were computed differently for Viva to account for relatedness in the sample ( see Materials and Methods ) . Fig 2 shows variation of African average dosages along each autosome . ( S1A Fig has average dosages for all ancestries . ) The spikes on chromosome 6 in both datasets are rather striking . For Viva , the sample standard deviation ( ssd ) of average dosages for Amerindian , European , and African components are 0 . 046 , 0 . 043 , and 0 . 024 respectively . The largest deviations , measured by the ssd of average dosages for each ancestry , are 5 . 4 , 4 . 8 , and 9 . 9 . The locus whose African average dosage is 9 . 9 ssd above the mean is inside the MHC region , and under the normal approximation , a 9 . 9 ssd corresponds to a p-value of 2 × 10−23 , which surpasses any reasonable significant threshold for a genome-wide analysis ( in GWAS such a significant threshold is 5 × 10−8 ) . The same region inside MHC was again identified as significant in Lipid; the largest deviation of African average dosages is 14 . 8 ssd above the mean , which corresponds to a p-value of 3 × 10−49 . The region identified in MHC is the same region identified by analyzing HapMap3 Mexican samples [9] . In that study , a region on chromosome 8 was also identified as border-line significant in Amerindian average dosages . In both Viva and Lipid , however , this region was not replicated . We used HapMap3 Utah Residents with Northern and Western European Ancestry ( CEU ) as European training samples; Yoruba in Ibadan , Nigeria ( YRI ) , from west Africa , as African training samples; and Maya and Pima from HGDP [6] ( MAYA ) as Amerindian training samples . To test the robustness of our results against different choices of training samples , we first investigated European and African training samples as they both have alternative choices in HapMap3 . We used Tuscani in Italia ( TSI ) , from south Europe , as an alternative to CEU , and Maasai in Kinyawa , Kenya ( MKK ) , from east Africa , as an alternative to YRI , and these produced four combinations: CEU−YRI−MAYA , CEU−MKK−MAYA , TSI−YRI−MAYA , and TSI−MKK−MAYA . We also combined all training samples to perform inference ( CEU+TSI−YRI+MKK−MAYA ) . The genome-wide pattern of local ancestry is consistent for different sets of training samples ( S1 Table and S1B and S1C Fig ) . We thus focus on the MHC region shown in Fig 3 ( a ) and 3 ( b ) . We made the following observations: 1 ) Using TSI to replace CEU produced a less significant deviation at the MHC region . 2 ) Using MKK to replace YRI produced a more significant deviation at MHC . 3 ) Combining all training samples produced a significant deviation at MHC , and the significant level is intermediate among other combinations . 4 ) Outside the MHC region , different combinations of training samples produced congruent results . Fig 3 ( c ) shows the difference in inferred European average dosages between two European training samples ( average difference between TSI−YRI−MAYA vs CEU−YRI−MAYA and TSI-MKK-MAYA vs CEU-MKK-MAYA ) . Interestingly , the highest peak contains HLA-B and HLA-C loci . We naturally suspect that TSI has more genetic diversity than CEU at the MHC , because more genetically diverse European training samples tend to produce higher estimates of European ancestry dosages . Amerindian average dosages are congruent between choices of CEU and TSI training samples ( S2 Fig ) , and the deficiency in African average dosages when using TSI as training samples are compensated for by sufficiency of European average dosages . We extracted 8679 SNPs in the extended MHC region , 25–35Mb on chromosome 6 , from European and African training samples , and ran ELAI using two upper clusters without specifying the population label , which is essentially haplotype-based structure analysis [9] . One admixture component was arbitrarily chosen to make comparison , and the admixture component was averaged over 10 EM runs ( after adjusting for label-switching across EM runs ) . The violin plots in Fig 3 ( d ) show that TSI is indeed more diverse than CEU at the MHC , MKK is more diverse than YRI , and MKK is the most diverse among four non-admixed populations , which agrees with the theory of east African origin of modern humans [15] . Recently admixed African Americans ( ASW ) were included for sanity check of the haplotype-based structure inference . Next we turn to Amerindian training samples . The 1000 Genomes admixture analysis group used a collection of Amerindian samples [16] different from the Maya and Pima from HGDP that we used , but we had difficulty in obtaining that data . Moreover , a practical concern is that any specific choice of Amerindian training samples will be subject to suspicion of inaccuracy . To test the robustness of our inference against different Amerindian training samples , we elected to remove Amerindian training samples and used only European and African training samples to perform inference—but of course we kept the setting of three ancestral populations . ELAI can function with the absence of one training population as long as there are enough genetic components of that ancestry in the cohort samples . Because Mexicans have a large Amerindian ancestry proportion , when Amerindian training samples are missing , ELAI is still able to learn Amerindian ancestral haplotypes relatively easily from Mexican samples as long as the sample size is large . The same is true for European training samples , but it becomes more difficult if African training samples are missing . To borrow an analogy from next-generation sequencing , a large number of Mexican samples and a high ancestry proportion to local ancestry inference is analogous to a high coverage of sequencing reads to variant call . The recommended practice in an early version of ELAI is to split a large dataset into small subsets . Doing so not only improves computational efficiency on a computer cluster , but also allows ELAI to jointly fit training and cohort datasets . It is evident [17 , 18] that a cluster model becomes less fit to the training samples in the presence of an overwhelmingly large number of cohort samples , which undermines the performance of local ancestry inference ( or imputation ) . Recall that removing Amerindian training samples requires a large number of cohort samples jointly fitting the model with training samples—we are seemingly in a quandary . The solution is rather simple . In parameter estimation of the two-layer model underlying ELAI [9] , we can arbitrarily adjust relative weights between cohort and training samples without changing the expected ancestral allele ( haplotype ) frequency estimates . In other words , we can take an arbitrarily large number of cohort samples and down weight their contribution to parameter estimation . When the training samples are available , the weighting ensures the model fits to training samples sufficiently; otherwise , the ancestral alleles are estimated exclusively by cohort samples , and the weight cancels out in the parameter estimation as long as we assign equal weight to all cohort samples . ( The technical details can be found in Materials and Methods . ) Thus , the weighting allows us to take the extreme measure of removing Amerindian training samples . We implemented the weighting scheme and applied it to both datasets . We combined CEU and TSI as European training samples and YRI and MKK as African training samples . Fig 4 demonstrates , using both Viva and Lipid datasets , the difference , or lack of it , in the estimated African average dosages with and without Amerindian training samples . Comparing the Amerindian average dosages , however , the estimates without Amerindian training samples are higher than that with . The mean differences are 0 . 09 for Viva and 0 . 08 for Lipid . This is not too surprising considering 1 ) Maya and Pima samples have some European ancestral components ( PC plots in Fig 1 ) ; and 2 ) Maya and Pima samples may be imperfect representatives of the Amerindian source populations for Mexicans , and learning Amerindian ancestry components from a large number of cohort samples may provide a better fit . Our results shall eliminate concerns of possible artifacts caused by inaccurate Amerindian training samples . If purely by chance , it is very unlikely that Amerindians share more alleles with Africans at MHC than the rest of the genome at such a significant level; that the pathogens from the Old world are often lethal to the native inhabitants of the New World seems to argue against such a peculiar sharing . The effect of the population bottleneck and the drift do not distinguish the MHC from the rest of the genome [19] . If selection happened in Africans before admixture , one would expect to see such selection signals in African Americans , which are not there [20] . Therefore , it is safe to assume that the African average dosages in Mexicans rose from the genome-wide mean p0 , which is a proxy dosage before selection at MHC , to the inferred value of p1 at MHC in the past 20 generations , and it is selection at work . A selection coefficient s can be computed via a simple model p1 = p0 × ( 1+s ) 20 , which provides a lowerbound estimate of s compared to recursion formula for both dominance and additive models ( see Materials and Methods ) . Table 1 summarizes the estimates of selection coefficient under different models; the lower-bound estimates are s = 0 . 05 for Viva and s = 0 . 07 for Lipid . Both estimates indicate a very strong selection , on par with the lactase selection in northern Europeans ( 0 . 09–0 . 19 ) [21] and the sickle-cell trait in Africans ( 0 . 05–0 . 18 ) [22] . To understand how many SNPs have contributed to the selection signal in MHC , we assigned a phenotypic value to each individual based on their African ancestry dosage at the identified region in MHC ( detailed in Materials and Methods ) , regressed out six leading principal component and admixture proportions , and performed the single-SNP association test using BIMBAM [17] . At a very liberal threshold of log10 Bayesfactor > 10 , we discovered 1700 SNPs in the extended MHC region to be genome-wide significant ( S3 Fig ) . Considering the high correlation among SNPs in the region , we next performed multi-SNP analysis using a Bayesian variable selection regression procedure implemented in the software piMASS [23] . piMASS implements a Markov chain Monte Carlo ( MCMC ) procedure to sample the posterior distribution of model space ( SNP sets ) under sparse and shrinkage priors . The output contains posterior probability of association ( PPA ) for each SNP , which roughly reflects how often the SNP is being selected in an additive model . We ran piMASS using all markers from chromosome 6 of Lipid with 10 , 000 burn-in steps and 1 million sampling steps . Two independent runs were conducted . In both runs , the proportion of variation explained ( the narrow sense heritability ) estimates had the same posterior mean of 0 . 88 , with ssd of 0 . 015 and 0 . 017 respectively . The posterior mean model sizes ( the number of SNPs in the model sampled ) were 93±10 . 7 and 83±7 . 1 respectively ( mean ± ssd ) . The two runs had 126 and 116 SNPs with PPA >0 . 1; among them , 60 SNPs overlapped , and the union contained 182 SNPs . We removed these 182 SNPs and reran local ancestry inference of chromosome 6 . The pattern of the local ancestry was essentially unaffected . These exercises suggest that the observed selection signal is driven by a large number of SNPs and their constitutional haplotypes .
In this paper we analyzed two existing GWAS datasets of Mexican subjects and demonstrated that the MHC region is under strong recent selection in Mexicans . Because Viva contains related individuals , we split individuals into non-overlapping subsets , each containing 40–50 unrelated individuals; performed local ancestry inference separately for each subset; and aggregated them to compute the average dosages . This practice produced congruent results as our combined analysis . In Lipid , samples were assigned case-control labels according to their triglyceride levels . The results presented in the paper ignored the case-control status . We analyzed cases and controls separately , and the results were highly congruent to that of the combined analysis . We also analyzed African American samples in HapMap3 and did not find any region under selection , which agrees with a recent study [20] . This serves as a negative control for ELAI . We devised a model fitting technique to introduce weighting into parameter estimation , which makes it possible to infer local ancestry of Mexicans using only European and African training samples . This rids us of the concern that the detected selection signals in Mexicans are artifacts produced by inaccurate Amerindian training samples . A previous study detected selection in 1000 genomes Mexican samples through local ancestry analysis [9] . Bhatia and colleagues questioned the plausibility of that finding; they argued that if signals were there , the 1000 genomes admixture analysis group would have found it [20] . We took this opportunity to investigate why the 1000 genomes admixture analysis group failed to detect the strong selection at the MHC region in Mexicans . We simulated genotypes using a demographic model that mimic the out-of-Africa migration events [24] , performed forward simulations to mimic admixture and selection at three linked loci ( details in Materials and Methods ) , and inferred local ancestry . The 1000 genomes used consensus call from four programs: HapMix [8] , LAMP-LD [25] , RFMix [26] , and MultiMix [27] . The publicly available version of HapMix was designed exclusively for two-way admixture , and the extended version used to analyze the 1000 Genomes data was not available to us [28] . Thus it was excluded from our analysis . MultiMix performed poorly despite our best effort and was excluded as well . For both LAMP-LD and RFMix we used the same parameter settings as those used in the 1000 Genomes admixture analysis group [28] . Both LAMP-LD and RFMix require phased training samples , and RFMix also requires phased cohort samples . ( ELAI works with diplotypes . ) When supplied with true phasing , both LAMP-LD and RFMix works well , on par with ELAI . We then introduced 2% switch-errors into cohort haplotypes and training haplotypes that mimic Amerindians , 1% switch-errors into European and African training samples . LAMP-LD is robust to switch-errors , but RFMix under-performs ( S4A and S4B Fig ) . It is worthwhile to note that MHC is notoriously hard to phase , and phasing for admixed samples at MHC is even more challenging as it requires the phasing algorithm to correctly identify local ancestry—a catch-22 for RFMix . We were surprised at the worse-than-the-expected performance of RFMix in the presence of switch-errors ( S4C Fig ) . Further investigation revealed that its window size parameter has a sweet-spot ( S4D Fig ) . When using the best window size RFMix performed on par with ELAI ( S4E Fig ) . Going back to the question why the 1000 genomes admixture analysis group failed to detect the signal , our simulation studies suggested that the democratic strategy adopted by 1000 genomes admixture analysis group , which used consensus calls from four methods to identify local ancestry , was perhaps not optimal . The simulation studies prompted us to use LAMP-LD and RFMix to analyze chromosome 6 of Viva and Lipid data . We phased the Maya and Pima samples from HGDP using SHAPEIT [29] , which were used in combination with CEU and YRI haplotypes as training datasets . LAMP-LD was then applied to infer local ancestry of Viva and Lipid datasets . We then phased the Viva and Lipid datasets , and RFMix was applied to infer their local ancestry . Reassuringly , both LAMP-LD and RFMix discovered the signal of selection at MHC ( S5 Fig ) . The MHC region influences susceptibility and resistance to a broad range of infectious agents such as viruses , bacteria , and parasites . It is sensible to observe more alleles of African ancestry at MHC in Mexicans if those alleles confer selective advantages in the presence of certain infectious agents . The European conquerors brought to America European and African diseases such as smallpox , measles , and typhus . Spaniards imposed an urbanized life style and farming practice on native people . A sudden increase in local population concentration , displacement , social upheaval , food shortages , and stress made them much vulnerable to infectious diseases . An estimated 5–8 million native people perished in a smallpox epidemic alone in early 1500s [30] . Nevertheless , after “difficult struggles of the formative period , ” the acceptance and enthusiasm of the new life emerged from the persistence of the old; for a brief period a “fusion of European and Mesoamerican cultures seemed ready to emerge” [31] . But severe drought hit and lethal pandemic broke out [30 , 31] . The epidemic , called “huey cocoliztli , ” was symptomatically different from those imported from the Old World; some medical historians suspect it was a hemorrhagic fever caused by arenavirus carried by rodents [31] . It first broke out in 1545 and lingered until 1815 [31 , 32] . The epidemic selectively targeted native people , and 90% of the population perished in a few generations [30 , 32] . This sustained epidemic harbors plenty of opportunities for strong selection at MHC , which fits our analysis . Once again history left its mark in genomes for posterity [33] .
The first dataset , Viva La Familia obesity-diabetes familial risk study ( dbGaP Study Accession: phs000616 . v1 . p1 ) , contains 858 genotyped individuals [12] . Among them , 815 Mexicans children from 261 families were genotyped with Illumina HumanOmni 1-v1 . 0 BeadChips , and the remaining 43 children were genotyped on HumanOmni 2 . 5–8v1 BeadChips . We chose to analyze the 815 samples that were typed on the same chip . Study participants in Viva La Familiar study were recruited in Houston , Texas . The second dataset , Mexican hypertriglyceridemia study ( dbGaP Study Accession: phs000618 . v1 . p1 ) , contains 2229 samples with 1117 cases and 1112 controls , where the case–control status was ascertained based on an individual’s serum triglyceride level [13] . Note that although there were 4350 study samples reported in the paper , the dbGaP contains only 2229 that were genotyped with Illumina Human610-Quad BeadChips—stage 1 of the GWAS . The rest samples were only typed on selected 1200 SNPs—stage 2 . Study participants in this study were recruited in Mexico City . We call the first dataset Viva and the second Lipid . We removed all A/T , C/G SNPs whose potential allele flipping between different datasets cannot be identified without additional information . A SNP was removed if it was missing in one of the datasets , either training or cohort . We also removed SNPs whose missing proportion was larger than 5% . Although we realized that the Hardy-Weinburg disequilibrium test is not appropriate for admixed samples , we used it anyway to remove SNPs whose HWD test p-values <10−6 . It is understood that this practice errs toward the safe side by eliminating possibly good SNPs . Finally , we obtained the cluster plots for each SNP , devised a simple algorithm to assign quality scores to each SNP cluster plot , and visually inspected those SNPs whose score indicated low quality . We removed those SNPs that contained a fourth cluster , or whose clusters were not distinct ( examples of such cluster plots can be found in [34] ) . We were particularly stringent to conduct such SNP quality control at the MHC region . Of the two GWAS datasets we obtained from dbGaP , Viva contains SNP cluster information , but Lipid does not . In the end , we had 352 , 754 SNPs from Viva and 479 , 757 SNPs from Lipid . The low number of SNPs in Viva reflected small number of overlapping SNPs between the Illumina HumanOmni 1-v1 . 0 and the Illumina 650Y arrays , the latter of which was used by the HGDP study that generates the Maya and Pima genotypes used as Amerindian training samples . We used ELAI [9] for local ancestry inference , which has been demonstrated to outperform competing methods such as HapMix [8] and LAMP-LD [25] . ELAI implements a two-layer cluster model and the model is fitted via the EM algorithm . The upper-layer clusters are parameterized to represent haplotypes from ancestral populations , and the lower-layer clusters contemporary haplotypes . The two-layer model was motivated by approximating the coalescent with recombination . It directly applies to diplotypes and automatically integrates out phase uncertainty . It can also estimate the recombination rates between markers , and hence doesn’t require recombination map as an input . Thus , the requirement for running ELAI is minimal—just genotypes and marker positions . To run ELAI , one needs to provide training samples . We used European and African samples from HapMap3 and Maya and Pima samples from HGDP as default training samples ( or reference panels , or source populations ) . ELAI is a cluster-based model and we wanted to specify numbers of clusters . The number of upper-layer clusters represents the number of source populations and we set it as 3; the number of lower-layer clusters was set as 15 . Extensive simulations demonstrated that this parameter setting performs well [9] . Lastly , we needed to specify number of admixing generations and we used 20 . All ELAI results were averaged over 10 independent EM runs of 20 steps each , unless noted . Lipid data contains unrelated individuals , and we treated an individual as unit and the computation is straightforward . Viva data contains 261 unrelated families . Each family contains 1–8 children , with majority of families ( 242 ) having 2–4 children . To account for relatedness in Viva data , we treat a family instead of an individual as unit , and computed the average dosages in the following manner: first we obtained family ancestral dosages by averaging over family members , and then we averaged over families to obtain overall average dosages . The two-layer model and the details of model fitting using EM algorithm can be found in [9] . Here we show how to estimate θ , the allele frequency associated with the cluster which emits the observed data . To simplify notation and presentation , we assume observing haplotypes instead of diplotypes . The weighting scheme can be applied to mixed sample that contains both haplotypes and diplotypes . To update parameters in each EM step , we take derivative of the expected full data log likelihood with respect to a parameter we want to update , say x ∈ ξ , d d x E Z ( 1 ) , … , Z ( n ) | h ( 1 ) , … , h ( n ) , ξ * log p ( h ( 1 ) , … , h ( n ) , Z ( 1 ) , … , Z ( n ) | ξ ) = 0 , ( 1 ) and solve for x to obtain updates . Z ( i ) is the latent state of haplotype h ( i ) , which contains two components , one for each layer of clusters . The expectation in Eq ( 1 ) is with respect to the posterior probability of latent states , conditioning on ξ* , which is the collection of parameters of the two-layer model estimated from the previous iteration , and ξ is the collection of parameters to be estimated . At marker m , write q i j = ∑ s p ( Z m = ( s , j ) | h m ( i ) , ξ * ) , which is the marginal posterior probability of h m ( i ) emitted from cluster j . Let T k = { i : h m ( i ) = k } for k = 0 , 1 . Take the derivative with respect to θmj , which is the allele frequency associated with cluster j , to get - 1 1 - t j ∑ i ∈ T 0 q i j + 1 t j ∑ i ∈ T 1 q i j = 0 , ( 2 ) and solve to get t j = ∑ i ∈ T 1 q i j ∑ i ∈ T 0 q i j + ∑ i ∈ T 1 q i j , ( 3 ) which can be thought as estimates of θmj with equal weight 1 . To apply differential weights , we split Tk into training sample T k ( t ) and cohort sample T k ( c ) . For training sample we assign a weight wt and for cohort sample wc . Eq ( 3 ) is generalized to t j = w t ∑ i ∈ T 1 t q i j + w c ∑ i ∈ T 1 c q i j w t ∑ i ∈ T 0 t q i j + ∑ i ∈ T 1 t q i j + w c ∑ i ∈ T 0 c q i j + ∑ i ∈ T 1 c q i j . ( 4 ) Let wt ≫ wc , then cohort samples contribute very little to tj when training samples are present . This is often desirable because the qij estimates of training samples are more reliable , which is especially true in the context of imputation [17] . When training samples are missing , the first terms of both nominator and denominator on the right hand side which involve wt disappear and Eq ( 4 ) reduces to Eq ( 3 ) . Using simulated data ( described below ) , we fit the ELAI model using two training samples of European and African , discarding the Amerindian training samples . The African ancestral dosages were used to compare the inferred values and the truth . The results demonstrated that the weighting samples works well for selection coefficients of 0 . 02 and 0 . 05 , and showed a bias for selection coefficient of 0 . 10 , but the biased estimates were conservative for the purpose of detecting selection ( S6 Fig ) . We defined a marker set A that contained markers whose African average dosages were greater than 0 . 30 . This threshold was 13 sample standard deviations away from the mean ( in Lipid dataset ) , and the resulting markers formed a consecutive region within MHC . We assigned each individual a phenotypic value obtained by averaging African ancestry dosages over markers in A . Let s be the selection coefficient , and fn ( s ) denote allele frequency at the n-th generation which is a function of s . Here the allele is referred to as a class of population specific alleles . Assume that the population size is constant but infinite so that we have a deterministic model . For dominance model where both heterozygous individual and homozygous individual of advantageous alleles has the same fitness 1+s , we have recursion f n + 1 ( s ) = f n ( s ) ( 1 + s ) 1 + ( 2 - f n ( s ) ) f n ( s ) s . For additive model where a heterozygous individual has fitness 1+s and a homozygous individual of advantageous alleles has fitness 1+2s , we have recursion f n + 1 ( s ) = f n ( s ) ( 1 + s + f n ( s ) s ) 1 + 2 f n ( s ) s . Let n = 20; we know the values of f0 ( s ) and f20 ( s ) and we want to find s . Because fn ( s ) is a monotone function of s , we perform interval-bisection search to numerically solve for s . We start with an interval [a , b] , such that f20 ( a ) <f20 ( s ) <f20 ( b ) , we evaluate y = f 20 ( a + b 2 ) , if y > f20 ( s ) , we set b = y; otherwise we set a = y . We repeat this procedure until y−f20 ( s ) ∈ ( −ϵ , ϵ ) for a small ϵ . Note that to apply the recursion formulae , the input f0 ( s ) and f20 ( s ) have to be allele frequencies , which are half of the allele dosages for humans . We call the model defined by recursion fn+1 ( s ) = fn ( s ) ( 1+s ) the simple model . It is easy to check that for dominance model we have f n + 1 ( s ) = f n ( s ) ( 1 + s ) 1 + ( 2 - f n ( s ) ) f n ( s ) s < f n ( s ) ( 1 + s ) ; and for additive model we have f n + 1 ( s ) = f n ( s ) ( 1 + s + f n ( s ) s ) 1 + 2 f n ( s ) s < f n ( s ) ( 1 + s ) . Therefore the simple model produces a lower-bound estimate of s for both dominance and additive models . Let f0 ( s ) = p0 and f20 ( s ) = p1 , we have p1 = p0 ( 1+s ) 20 , and therefore the simple model estimate of selection coefficient is s = exp ( log ( p1/p0 ) /20 ) − 1 . We used a population genetics model that mimics the out-of-Africa migration events to simulate a 3 Mb region of three source populations that mimic Amerindian , European , and African [24] . After setting aside 200 haplotypes from each source population as training haplotypes , we used the remaining haplotypes to simulate three-way admixed individuals by a one-pulse model [35] . Specifically , we randomly selected 50 , 000 haplotypes from the three source populations using proportions of 50% , 45% , and 5% , mimicking the admixture proportion of Mexicans . We split 3 Mb into three segments , and assigned at two splitting points recombination hotspots . At each hotspot , we assumed equal recombination probability of 0 . 1 , 0 . 2 , and 0 . 5 per generation . We sampled two haplotypes with replacement and introduced possible crossover events at hotspots to produce two new haplotypes . We repeated the pairing and crossover 25 , 000 times to produce 50 , 000 haplotypes for the next generation . The admixture simulation was done for 20 generations . To simulate selection , we designated the mid-section as the locus under selection , and assumed selection coefficients of 0 . 02 , 0 . 05 , and 0 . 10 per generation . The alleles under positive selection were those 5% from the source population that mimicked Africans . After 20 generations , we randomly chose 2 , 000 admixed haplotypes , pairing them to form 1 , 000 diplotypes as cohort samples . We used two sizes of mid-section: 0 . 5 Mb and 1 Mb . A small mid-section produces a more challenging problem . To investigate how switch-errors affected local ancestry inference for different methods , in addition to perfect phasing situation , we also introduce 2% phasing errors into Amerindian training samples and the cohort samples , and 1% phasing errors to European and African training samples . To do so , at randomly selected heterozygous marker , from left to right we crossed-over two haplotypes .
|
Whether or not there exists recent selection since admixture in Latinos has been a subject of debate . To detect selection signal , a method uniquely applicable to recently admixed samples is local ancestry analysis . We infer local ancestry of admixed samples ( in our study , Mexicans ) , and look for regions where the average ancestry of one ancestry component significantly deviates from its genome-wide average . Inferring local ancestry requires training samples that represent the genuine ancestral source populations . One major concern for previously detected selection signals in Latinos via local ancestry analysis is the inaccuracy of Amerindian training samples . This is partly due to large genetic differences among Amerindian tribes and partly due to the difficulty in obtaining Amerindian training samples . We developed a new method which allows us to learn Amerindian ancestral haplotypes from Mexican cohorts in the absence of Amerindian training samples . Our work demonstrates the existence of recent strong selection at MHC in Mexicans .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"genome-wide",
"association",
"studies",
"medicine",
"and",
"health",
"sciences",
"population",
"genetics",
"immunology",
"ethnicities",
"clinical",
"medicine",
"genome",
"analysis",
"population",
"biology",
"chromosome",
"6",
"genomic",
"signal",
"processing",
"lipids",
"major",
"histocompatibility",
"complex",
"chromosome",
"biology",
"chromosome",
"pairs",
"people",
"and",
"places",
"biochemistry",
"signal",
"transduction",
"haplotypes",
"cell",
"biology",
"clinical",
"immunology",
"genetics",
"mexican",
"people",
"population",
"groupings",
"biology",
"and",
"life",
"sciences",
"genomics",
"evolutionary",
"biology",
"cell",
"signaling",
"computational",
"biology",
"chromosomes",
"human",
"genetics"
] |
2016
|
Strong Selection at MHC in Mexicans since Admixture
|
Modern agriculture favours the selection and spread of novel plant diseases . Furthermore , crop genetic resistance against pathogens is often rendered ineffective within a few years of its commercial deployment . Leptosphaeria maculans , the cause of phoma stem canker of oilseed rape , develops gene-for-gene interactions with its host plant , and has a high evolutionary potential to render ineffective novel sources of resistance in crops . Here , we established a four-year field experiment to monitor the evolution of populations confronted with the newly released Rlm7 resistance and to investigate the nature of the mutations responsible for virulence against Rlm7 . A total of 2551 fungal isolates were collected from experimental crops of a Rlm7 cultivar or a cultivar without Rlm7 . All isolates were phenotyped for virulence and a subset was genotyped with neutral genetic markers . Virulent isolates were investigated for molecular events at the AvrLm4-7 locus . Whilst virulent isolates were not found in neighbouring crops , their frequency had reached 36% in the experimental field after four years . An extreme diversity of independent molecular events leading to virulence was identified in populations , with large-scale Repeat Induced Point mutations or complete deletion of AvrLm4-7 being the most frequent . Our data suggest that increased mutability of fungal genes involved in the interactions with plants is directly related to their genomic environment and reproductive system . Thus , rapid allelic diversification of avirulence genes can be generated in L . maculans populations in a single field provided that large population sizes and sexual reproduction are favoured by agricultural practices .
Fungi are the most important pathogens of cultivated plants , with significant economic , food security and environmental impacts , the latter being due to the large quantities of fungicides used to control plant diseases [1] . In contrast to fungicide use , genetic resistance against pathogens in crops is an environmentally friendly strategy to control diseases . Frequently , effective resistance has been provided by the introduction of major resistance ( R ) genes into crop genotypes [2] . Unfortunately , fungal pathogens have an incredible plasticity with which they can respond to their environment . Furthermore , their ability to adapt to changes in their environment and to disseminate these adaptations makes them very successful in countering crop defenses and control methods [1] . The rapid emergence of new strains able to render ineffective new R genes is thus a common feature of fungal phytopathogens [3] . Durable disease resistance has always been a goal of plant breeding programs since it is cost-effective and environmentally friendly whilst promoting the conservation of rare genetic resources [4] . Durability of resistance is largely dependent on the biology of the pathogen and the evolutionary potential of the pathogen population [3] . In addition , cropping practices such as crop rotation and stubble management may directly affect the evolutionary potential of the pathogen by reducing its population size or dispersal or by interfering with its reproductive regime [5] . Major gene resistance depends on the “gene-for-gene” concept in which the host R protein interacts with a corresponding pathogen avirulence ( Avr ) protein to initiate plant disease responses and resistance [6] . Avr proteins are now known to be effectors involved in plant pathogenesis that were recognised by the plant surveillance machinery in the course of plant-pathogen co-evolution [7] . Thus , avoidance of recognition by host R proteins often involves pathogen effector gene inactivation , which may result in a fitness penalty for the pathogen [4] . An additional level of complexity arises because many fungal or oomycete pathogens have their effector genes in non-conventional , adaptable , regions of their genomes [8] . It was recently suggested that the location of effector genes of Leptosphaeria maculans , the cause of “phoma stem canker of oilseed rape” , in AT-rich , transposable element ( TE ) -rich blocks of the genome has major implications for gene mutability , resulting in either allelic diversification or gene inactivation under R gene selection [9] . For example , it was suggested that Repeat Induced Point mutation ( RIP ) , a fungal-specific mechanism of inactivation of repeated sequences in genomes , was acting on single-copy genes embedded in TE-rich blocks of the genome to favour the allelic diversification of such genes [9] , [10] , [11] . The cloning and characterization of the molecular determinants of resistance from the plants ( R genes ) and pathogens ( Avr genes ) has not only improved understanding of their functions and interactions but also provided the molecular information for detecting mutations in these genes . Fungal Avr genes have been cloned from only seven species , and the mechanisms by which fungal Avr genes evolve to evade host recognition is only documented for few Avr genes , mostly in Cladosporium fulvum , Magnaporthe oryzae , Rhynchosporium secalis and L . maculans [12] , [13] . Moreover the link between laboratory and field studies has often been missing , and when field populations of pathogens were investigated for mutations in Avr genes they have often been isolated from uncharacterized crop genotypes or have been collected and analysed a long time after the corresponding R gene had been commercially deployed in crops ( e . g . [14] ) . There is currently no example of identification of the initial events leading to virulence when a fungal pathogen is exposed for the first time to an R gene deployed under field conditions . L . maculans is a pathogen with a high evolutionary potential combining large population size , mixed reproduction regime and high dispersal ability . Following sexual reproduction taking place on stem debris , leaf infections by ascospores ( in autumn in western Europe ) cause phoma leaf spots , supporting asexual multiplication . The life cycle of the pathogen is completed by a lengthy symptomless colonisation phase when the pathogen grows from the leaf lesions along the petiole to the stem , where cankers develop to cause lodging and yield losses in crops at the end of the growing season ( spring and early summer in western Europe ) . Sexual mating between the numerous isolates that colonised the stem tissues then takes place [12] . Major gene resistance against L . maculans has been widely used in oilseed rape [15] , [16] , but was rendered ineffective in only a few growing seasons [17] , [18] . Three L . maculans Avr genes , AvrLm1 , AvrLm6 and AvrLm4-7 have been cloned [19]–[21] . All encode Small Secreted Proteins ( SSPs ) embedded in large TE-rich blocks of the genome termed AT-isochores . AvrLm4-7 is located within a 96-kb AT-isochore which only contains two other genes located 26 and 28 kb away . These two genes are also predicted to encode SSPs . AvrLm4-7 is recognised by two distinct R genes , Rlm4 and Rlm7 and escape from recognition by Rlm4 is due to a single-base non-synonymous mutation resulting in a Gly120Arg change in the protein . This change does not alter recognition by Rlm7 [21] . Before 2003 , L . maculans has not been exposed to the Rlm7 selection in Europe and a population survey done in 2000–2001 in France identified only one virulent isolate out of 1787 isolates ( 0 . 05% ) [22] . Rlm7 has been introduced in commercial cultivars in 2003 in France , with only 1% of the hectarages cropped with Rlm7 cultivars until 2005 ( X . Pinochet , CETIOM , personal communication ) , thus providing us with the opportunity to survey emergence of virulent L . maculans isolates at the time of initial selection pressure and to identify the first molecular events responsible for the overwhelming of the resistance gene . Here , we established a four-year field experiment ( Figure S1 in Text S1 , Figure S2 in Text S1 ) and combined molecular genetic and population genetic approaches to evaluate speed and patterns of mutations in the AvrLm4-7 gene responsible for the loss of the AvrLm7 specificity in L . maculans populations exposed to Rlm7 selection . Molecular analysis of events leading to the virulent phenotype in a single field revealed a tremendous diversity of mutation events , and confirmed the importance of the genomic environment in gene mutability .
We established a four-year ( 2004–2005 to 2007–2008 growing seasons ) field experiment at Grignon , France , during which Rlm7 and rlm7 cultivars were grown alongside each other ( with relative ca . 2/3 of the area cropped with the Rlm7 cultivar and 1/3 cropped with the rlm7 cultivar ) ( Figure S1 in Text S1 , Figure S2 in Text S1 ) . Exposure of the L . maculans population to Rlm7 was maximised because there was no crop rotation or ploughing in crop debris . Before the start of the experiment at Grignon ( i . e . between 2000 and 2004 ) no Rlm7 cv . had been grown and the frequency of virulent avrLm7 isolates was minimal at both Grignon and another crop located 12 km away at Versailles ( Figure S1A in Text S1 ) , with estimated frequencies of avrLm7 isolates ranging from 0 . 006% to 1 . 3% at Versailles and from 0 to 1 . 3% at Grignon ( Figure 1 , Table 1 , Table S1 in Text S1 ) . In the experimental field , the phoma stem canker was not severe in the summer of 2007 or 2008 . The G2 disease indices , a disease severity index summarising the proportions of plants observed within six canker severity classes and ranging between 0 ( all plants healthy ) to 9 ( all plants with severe canker ) [4] , were 1 . 53 ( rlm7 cv . ) and 0 . 86 ( Rlm7 cv . ) in 2007 . However , the G2 index had increased to 2 . 34 on the Rlm7 cultivar by 2008 , reflecting a localized overwhelming of the Rlm7 resistance during the course of the experiment . In the course of the four-year experiment , we collected 2551 isolates either from the Rlm7 cv . Exagone or from the rlm7 genotype Campala ( Grignon and Versailles; Table S2 in Text S1 ) . Of these , 1987 isolates were characterized for their interactions with Rlm4 and Rlm7 plant genotypes ( Table S2 in Text S1 ) . The number of virulent avrLm7 isolates remained low or undetectable in the control plot at Versailles on the susceptible cv . Campala . At Grignon , the frequency of virulent avrLm7 isolates on the susceptible rlm7 cv . grown next to the Rlm7 cultivar steadily increased to 36 . 2% of the population , whilst avrLm7 isolates were not detected in crops located less than 600 m from the experimental plot ( Figure 1 , Table 1 , Figure S1 in Text S1 , Table S2 in Text S1 ) . Consistent with analyses of L . maculans ascospore dispersal indicating that most spores are deposited within 500 meters from the source and especially in the first 100 meters [23] , [24] , this suggests that the observed increase in avrLm7 isolate frequency in the field experiment was due to the recurrent local use of Rlm7 and not to an increase of its frequency at the regional level . Nucleotide polymorphisms in AvrLm4-7 were analysed in 169 AvrLm7 isolates from the sample . A very low level of sequence polymorphism was found within the gene with only five polymorphic nucleotides , all corresponding to non-synonymous mutations in the protein ( Table 2 , Figure 2B ) . The combination of these polymorphic sites generated five haplotypes ( Table 2 ) . In accordance with previous data [21] , one haplotype included those of the isolates that showed the AvrLm4 specificity and differed from the other haplotypes by the presence of a guanine at base 358 resulting in a glycine at amino acid 120 ( Table 2 ) . All other mutations were found in isolates that had lost the AvrLm4 specificity but maintained the AvrLm7 avirulence , thus defining four distinct alleles for avrLm4-AvrLm7 isolates ( Table 2 ) . Of the 808 virulent avrLm7 isolates obtained from Grignon , 769 were characterized for molecular events responsible for virulence . Numerous mutational events responsible for loss of the avirulence function were identified , single-or di-nucleotide deletions , single nucleotide polymorphisms ( SNPs ) , wide degeneracy due to RIP , complete or partial deletion of the gene , major chromosomal rearrangements and alteration in gene expression . Only a few isolates had SNPs , single-nucleotide deletions , or under-expression of the gene whereas gene deletions and RIP mutations were present in 62 . 8% and 24 . 1% of the virulent isolates , respectively ( Figure 2A ) . In most cases the mutational events resulted in lack of protein production or production of a severely truncated protein , while in a few other cases the 3-D structure of the protein was probably modified ( e . g . mutation in cysteine residues involved in disulfide bridge formation ) ( Figure 2B , Table 3 ) . Lastly , in a few cases , the sequence of the gene was unaltered but its expression in planta was impaired ( Figure S3 in Text S1 ) suggesting mutation of regulatory elements outside of the gene sequence . In 2000 , one avrLm7 isolate was collected during a large-scale survey of French L . maculans populations [22] . In this isolate , the virulent phenotype was due to the insertion of a complete LTR of RLC-Pholy at base 6 of the coding sequence , resulting in the production of a 30 amino acid protein mainly corresponding to part of the TE ( data not shown ) . This event was not identified in the 2006–2008 sampling . Major genome rearrangements leading to complete or partial gene deletion and RIP mutations were the two main events responsible for virulence toward Rlm7 . However , our survey indicated a contrasting sequence of events ( Figure 5 ) . The proportion of the population affected by minor events ( SNP , single or di-nucleotide frequency , unaltered gene sequence ) , or major rearrangements leaving internal part of the gene unaltered remained stable over the three years ( Figure 5 ) . In contrast , a significant change in frequency was observed for the two most common events ( Pearson's approximate χ2 test , with 20 , 000 random samplings , P = 0 . 022 ) . Whereas RIP mutation was the most frequent event leading to virulence in the first year of the survey , at a time when only very few virulent isolates could be found ( Table S2 in Text S1 ) , complete gene deletion became more common in years 2 and 3 , while the frequency of RIP mutations decreased in the second year and then remained stable at ca . 20% ( Figure 5 ) . When analysing the intensity of RIP mutations , there was no significant increase in the mean number of mutated sites per isolate as a function of the year of isolation ( Kruskal-Wallis test; P = 0 . 371 , Figure 6 ) with for example , the same proportion of alleles with the least number of RIP mutations found in autumn 2006 as in summer 2008 ( Figure 2 , Figure 6 ) , or in contrast , alleles with the greatest number of RIP mutations found in autumn 2006 ( isolate G06-436 in Figure 2B ) . Effective population size was firstly estimated using Approximate Bayesian Computation ( ABC ) methods from the minisatellite polymorphism data obtained for the population sampled in Grignon on the susceptible cv . Campala . Effective population size ( Ne ) estimated at the field scale was 11 , 500 ( CI95% 3 , 490–29 , 900 ) . To investigate the origin of avrLm7 isolates and how their rapid increase in frequency influenced the genetic structure of L . maculans populations , a subset of 161 avrLm7 and 161 AvrLm7 isolates collected from the field experiment were genotyped using seven minisatellite ( MS ) markers located on different chromosomes . All seven markers were polymorphic and a total of 80 alleles were found in the collection . The average number of alleles over loci ranged between 6 . 86 and 8 . 86 ( Table 4 ) and revealed no significant difference in allelic variability between virulent and avirulent isolates sampled ( Kruskal-Wallis test; P = 0 . 95 ) . Over the three years , the average gene diversity was similar amongst avirulent and virulent isolates ( two-sided permutation test , 15 , 000 permutations , P = 0 . 39 ) . The AvrLm7 and avrLm7 isolates collected between autumn 2006 and summer 2008 showed very considerable genotypic diversity and the number of genotypes was identical or very close to the number of isolates ( Table 4 ) . Overall , a total of 313 multilocus genotypes ( MLG ) were differentiated , of which six were shared by two to four isolates and 307 isolates ( 95 . 3% ) had unique genotypes . Taking into account mating type alleles , AvrLm4-7 alleles and one additional MS marker located 70 kb away from AvrLm4-7 , the isolates with identical MLGs could be differentiated and shown to be unique genotypes ( data not shown ) . Both mating types were present in all isolate samples and occurred in equal frequencies for most of them , except for the sample comprising avrLm7 isolates collected in autumn 2006 , in which a significant deviation ( P = 0 . 012 ) from a 1∶1 ratio was detected ( Table 4 ) . This biased frequency may be attributable to a resampling bias or a bias linked to the scarcity of virulent isolates at the beginning of the experiment . The multilocus linkage disequilibrium values ( rd ) obtained for all the samples were close to zero and did not deviate significantly from the expectations under the null hypothesis of random mating in all samples ( Table 4 ) . These results suggest that there was no genotypic disequilibrium in the samples studied and that recombination regularly generates new AvrLm7 and avrLm7 genotypes in the population . To estimate differentiation between AvrLm7 and avrLm7 isolate samples , we firstly tested each pair of samples for heterogeneity in allele frequencies using the Fisher exact test ( data not shown ) . These estimates were consistent with the hypothesis that there was no genetic differentiation between the samples . F-statistics also showed no significant population differentiation between them for all loci and overall . Accordingly , the mean FST was not significantly different from zero ( P = 0 . 51 ) over loci between AvrLm7 and avrLm7 samples . Pair-wise levels of genetic differentiation estimated between all pairs of AvrLm7 and avrLm7 samples gave FST-estimates which were not significantly different from zero ( data not shown ) . Lastly , hierarchical AMOVA on all samples confirmed the absence of genetic differentiation between AvrLm7 and avrLm7 isolates and showed that only 0 . 04% of the total variance was distributed among populations , with 99 . 96% within populations ( ΦST = 0 . 0004 ; P = 0 . 14 ) .
Rapid adaptation of microbes to control methods ( drugs such as antibiotics and fungicides or plant disease resistance ) is a very common phenomenon driven by mutation and selection , along with reproduction regime and gene flow that amplify and disperse the new character in the pathogen population [3] . Resistance to drugs can be ascribed to various mechanisms ( reduced permeability or enhanced efflux , enzymatic inactivation , alteration or over-expression of the target gene ) [26] . In contrast , in simpler gene-for-gene systems representative of numerous plant-pathogen interactions , modification of a target ( i . e . the avirulence gene product ) allows the pathogen to escape the resistance gene-mediated plant defense responses [27 , this study] . Since avirulence gene products are pathogen effectors , how easy it is for the pathogen to modify or delete the target gene depends on the fitness deficit linked with loss or attenuation of its effector function [4] , [28] , [29] . In addition , the ability of a pathogen to render host resistance ineffective is a function of biological traits that contribute to its “evolutionary potential” , including its reproduction regime , size of populations and dispersal ability [1] , [3] . Using a fungal pathogen known to have a high evolutionary potential and a dedicated field experiment , we investigated the “breakdown” of the new resistance gene Rlm7 , corresponding to the AvrLm4-7 effector which makes an important contribution to fungal fitness [28] , [29] . The establishment of this experiment aimed to address two questions for which little or no information is currently available: ( i ) what are the initial mutations responsible for rendering ineffective a plant resistance gene at the scale of a single field ? ( ii ) how ( and how rapidly ) are these mutations generated ? Our data suggest that we have captured all of the possible mutational events existing very early in the process of selection and show that adaptation to selection occurs rapidly through numerous diverse mutational events at the AvrLm4-7 locus . Almost all mutations lead to gene inactivation or production of a non-functional effector protein . Unexpectedly , in a single 0 . 25-hectare field we observed all previously reported molecular events ( and more ) leading to loss of fungal avirulence in world-wide collections of isolates . Most of these mutational events were even observed during the first year of the experiment . These included complete or partial deletion of the gene [10] , [14] , [27] , [30]–[33] , amino acid substitutions [14] , [27] , [34] , point deletions and production of truncated proteins [27] , and “insertion” of a transposon [33] , [35] , [36] . In addition , the RIP mutations that commonly occurred , have not been previously reported as an inactivation mechanism for effector genes for pathogens other than L . maculans [10] , [31] . Three other new phenomena observed were common deletion of an AA dinucleotide , alteration of the gene expression and gene duplication possibly favouring ( or responsible for ? ) RIP mutations . The speed of “generation” and diversity of mutational events and increased ratio of virulent isolates in the population then raises questions about how these events were generated and dispersed . The first postulate was that strong selection in a large local population would have allowed the emergence of numerous mutational variants . The estimate of Ne obtained here using ABC approaches indeed indicated large effective population size in the field and was comparable to what is described for the few sexually reproducing phytopathogenic fungi for which similar analyses were performed [37] , [38] . We then investigated the sequence of mutation events found at the AvrLm4-7 locus following selection and showed that two of the many possible types of mutations were favored over others and varied in frequency over time . RIP was the prevalent mutation pattern at the beginning of the sampling at a time when only few mutants could be found , then followed by large-scale deletions . RIP was initially described in the model ascomycete fungus Neurospora crassa as a premeiotic process that efficiently detects and mutates duplicated sequences [39] . In L . maculans , the embedding of effector genes in mosaics of RIP-altered TEs and the presence of RIP signatures in the sequence of effector genes indicated that RIP could act on unduplicated sequences to promote gene diversification and that it could “leak” from the neighbouring RIP-affected sequences to generate mutations in single-copy genes [9] . Consistently , the 3′ part of the gene , directly bordered by TEs , is more affected by RIP than its 5′ part and promoter . This hypothesis , however , has to be reconsidered in view of our finding that part of the isolates with RIPped alleles of AvrLm4-7 probably has two copies of the gene . This might be more consistent with a canonical RIP mechanism , indicating that , at least in part of the cases , gene duplication precedes the action of RIP , thus acting on truly duplicated sequences and may be followed by segregation and deletion of one ( or two ) copies of the gene . In contrast with this finding , inactivation of the avirulence gene AvrLm6 by RIP mutations [10] , [11] was not associated with duplication of the gene in field isolates , thus substantiating the ‘leaking from neighbouring RIPped regions’ hypothesis [11] . Either leaking from neighbouring RIPped TEs or acting on truly duplicated sequences , RIP is an extraordinary efficient mutation mechanism that affects up to 30% of the G:C pairs of duplicated genes in a single sexual cycle of N . crassa [39] . This suggests that RIP mutations can be generated at a very high rate at each sexual cycle of L . maculans in the field , i . e . at the beginning of each growing season . In addition , frequency of RIP mutation is increased by the embedding of effector genes in TE-rich blocks of the genome , allowing action of RIP on single-copy genes . Both these data substantiate the importance of genome environment and sexual reproduction to promote an accelerated mutation rate of effector genes ( including AvrLm4-7 ) due to RIP [9] , [11] , which is likely to correspond to the most rapid adaptation to selection . In the second and third years of sampling , large-scale deletions became more common than RIP mutations as an inactivation mechanism illustrating a dynamic process in which many possible virulence alleles are generated , but only a small number eventually survive . While some large-scale deletions may in fact correspond to extremely RIPped alleles as found here for isolate NzT-4 , this finding is reminiscent of what we observed when analysing the avrLm1 locus in French populations of the fungus many years after the large-scale use of the Rlm1 resistance in the field: more than 90% of the virulent isolates had a 260-kb deletion of the gene and its TE-rich environment while only 0 . 7% of the isolates had an allele with RIP signatures [31] . In L . maculans , the presence of four widely expanded TE families representing 25 . 2% of the L . maculans genome [9] provides many targets for mis-pairing between sister chromatids during meiosis and suggests that unequal crossovers lead to production of large deletion/insertion events and production of chromosomes of novel sizes in the progeny [40] , [41] . Consistent with clustering of these four families of TEs in AT-isochores containing all currently known avirulence genes of L . maculans , large-size deletions were described as the main event leading to virulence at the AvrLm1 and AvrLm6 loci [10] , [31] , as for the AvrLm4-7 locus ( this study ) . Analysis of surrounding populations obtained from susceptible rlm7 cvs . indicated an increase in frequency of virulent avrLm7 isolates next to the resistant Rlm7 plot , but not in plots located less than 600 meters away , reflecting the rapid selection of virulent isolates expected in such experimental conditions . The cropping practices used in the field experiment are likely to have two consequences: ( i ) a large increase in size of the local population due to the lack of rotation and the close contact of the crop with unburied infected residues from the previous years and ( ii ) an increased rate of sexual reproduction because infected debris were left on the soil surface . RIP mutations , gene duplications and gene deletions , the most common modes of loss of the avirulence at the AvrLm7 locus directly depend upon the ability of the pathogen to undergo sexual mating . Favouring this part of its life cycle along with increase in population size directly impact the ability to generate a large number of virulent progeny and the probability they will be selected for by the resistant cultivars , eventually improving the opportunities for mating between two virulent parent isolates . The ( i ) wide diversity of RIPped alleles and the lack of increase in the mean number of mutations per allele during the 3 years of the experiment , ( ii ) the fact that virulent isolates could not be detected in local populations and ( iii ) the genetic similarities between the virulent and avirulent populations , indicating that virulent and avirulent isolates are part of the same genetic population in which the virulence allele is independently assorting with respect to all of the other genes , all suggest that at least part of the mutations at the AvrLm4-7 locus selected in the experimental field are generated locally within a short-time period and as a result of the large population size and meiotic recombination .
The field experiment was established at Grignon , France ( 48° 50′ 28 . 40″ N latitude; 1° 56′ 13 . 83″ E longitude ) ( Figure S1A in Text S1 ) . This location had been used in previous studies to describe L . maculans population race structure [22] . The field was a right-angled triangle with a 50 m long base and a 100 m long side . The experiment was started in autumn 2004 and was cropped for four growing seasons ( 2004–2005 to 2007–2008 ) as a monoculture of oilseed rape with minimum tillage ( chiselling ) . In normal agronomic practice ( e . g . at Grignon before the start of the field experiment ) , oilseed rape is grown as part of a farm rotation and rarely returns to the same field more than one year in three ( Figure S1B in Text S1 ) . Monoculture of oilseed rape without ploughing to bury infected stubble was chosen to increase the amount of annual sexual reproduction and the local population size of L . maculans , partly mimicking minimum tillage practices in which infected stubble are left at soil surface . No fungicides were used . The plot was sown with cultivars with Rlm7 resistance ( Roxet in 2004–2005 , and Exagone in 2005 to 2008 ) and was bordered by a 10-meter wide strip cropped with a cultivar without Rlm7 ( Campala ) that was used as a trap cultivar [22] ( Figure S1B in Text S1 ) . No Rlm7 cv . had been grown in Grignon fields before the start of the experiment . For comparison purposes , control plots were also assessed for disease severity and occurrence of virulent avrLm7 isolates of L . maculans . These control plots included a series of plots at Grignon cropped with a susceptible cultivar between 2000 and the start of sampling of the experiment ( autumn 2006 ) ( Figure S1B in Text S1 , Table S1 in Text S1 ) . Other control plots were cropped at Versailles ( 48° 48′ 27 . 59″ N latitude; 2° 5′ 12 . 30″ E longitude ) , ca . 12 km away from Grignon ( Figure S1A in Text S1 , Tables S1 and S2 in Text S1 ) with susceptible cultivars ( 2000–2007 ) or with the Rlm7 cultivar Exagone in autumn 2006 and 2007 . These control fields were cropped with the usual agronomical practices of French farmers with ploughing and rotation . Data collected included severity of stem canker , evaluated at crop maturity using the G2 disease index [42] on 160 randomly chosen plants per cultivar . Due to the life cycle of L . maculans in which ascospores are the origin of leaf lesions in autumn , which in turn initiate the systemic colonisation of plants eventually causing the stem canker in the following summer , populations collected from stem canker in the summer of year n , following meiosis , were considered similar to those collected from leaf lesions in the autumn of year n . Isolates were sampled either from leaf lesions ( single pycnidial isolates ) or from stem cankers ( single-ascospore isolates ) using methods described by Balesdent et al . [22] and West et al . [43] , respectively , during three continuous years corresponding to two cultural cycles ( 2006–2007; 2007–2008 ) , but to three generations of the fungus ( Figure S2 in Text S1 ) . The number of leaves with lesions collected on the Rlm7 cultivar ranged between 20 and 200 . Typical leaf spot lesions caused by L . maculans on the Rlm7 genotype were rare at the start of the experiment due to the scarcity or absence of virulent isolates ( Table S2 in Text S1 ) . Only 24 leaves with phoma leaf spots were found in the whole Grignon plot of Exagone in autumn 2006 and all of them were sampled , with sometimes more than one isolate per leaf sampled ( Table 1 , Table S2 in Text S1 ) . When more leaf lesions were present ( subsequent growing seasons for Exagone , all years for Campala ) , infected leaves were randomly collected with only one isolate obtained from each individual plant . As the frequency of virulent ( avrLm7 ) isolates on Exagone was expected to be small in the first sampling year , 500 distinct leaf lesions were collected from cv . Campala in autumn 2006 ( Table S2 in Text S1 ) , so that a frequency ≥0 . 5% of virulent isolates could be detected with a 95% confidence interval . This number was then decreased to 200 leaves for the second year of the experiment ( autumn 2007 ) . Similarly , 100 to 200 infected stems were collected each year from each cultivar . Induction of pseudothecial maturation and isolation of ascospores from infected stems was as described by West et al . [43] . Due to the scarcity of mature pseudothecia , more than one ( average three ) ejected ascospores were used for isolation from a single stem . Of these , only three isolates or less per stem were analysed using molecular markers ( see below ) . The frequency of virulent avrLm7 isolates before the establishment of the field experiment and in the vicinity of the field experiment during the course of the experiment was estimated using a collection of 1233 isolates recovered from leaf lesions or from ascospores on stems of susceptible cultivars including around 100 isolates collected in Grignon in autumn 2006 from two oilseed rape fields located less than 600 m from the field experiment ( Table 1 , Tables S1 and S2 in Text S1 ) . The reference isolates v23 . 1 . 3 ( AvrLm4-AvrLm7; double avirulent ) whose genome sequence is available [9] , v23 . 1 . 2 ( avrLm4-AvrLm7; avirulent towards Rlm7 only ) , and Nz-T4 ( avrLm4-avrLm7; double virulent ) [44] , [45] were used as controls for inoculation tests , and as sequence reference for the AvrLm4-AvrLm7 and avrLm4-AvrLm7 alleles of the AvrLm4-7 gene [21] . Isolate M3 . 2 , the first avrLm7 isolate collected in France in 2000 was also included to determine the type of mutation it harbours at the AvrLm4-7 allele [22] . All fungal cultures were maintained on V8-juice agar and conidia were collected from 12–15 day-old cultures according to the procedure described by Ansan-Melayah et al . [46] . AvrLm4 and AvrLm7 avirulence/virulence phenotypes were determined following inoculation of 15-days-old B . napus cotyledons with 10 µL of 107 mL−1 conidia suspension as described by Balesdent et al . [47] . Genomic DNA was extracted from conidia suspensions using the DNeasy 96 Plant Kit and the QIAGEN BioRobot 3000 in accordance with the manufacturer's recommendations . PCR primers used for mating type amplification , AvrLm4-7 analyses and minisatellite analyses were designed with PRIMER 3 [48] and are described in Supplementary Table S4 . AvrLm4-7 was amplified using ( i ) “external” primers spanning part of the promoter region and part of the 3′ UTR and generating a 1434 bp fragment and ( ii ) “internal” primers located within the coding sequence , between the ATG and intron and spanning 478 bp ( Figure 3A ) . Standard PCR were performed in an Eppendorf Mastercycler EP Gradient thermocycler ( Eppendorf , Le Pecq , France ) , with 30 cycles of 94°C for 30 s , 30 s of hybridization with variable hybridization temperatures , 72°C for 80 s , with a final extension at 72°C of variable duration ( Table S4 in Text S1 ) . Sequencing was performed on PCR products using a Beckman Coulter CEQ 8000 automated sequencer ( Beckman Coulter , Fullerton , CA , USA ) according to the manufacturer's instructions . Primers for sequencing ( Figure 3A ) were chosen so that all bases of the gene , including the 5′ UTR and 128 bp of the promoter region were independently read two or three times . Sequences were compared following sequence alignment using MULTALIN and CLUSTALX [49] , [50] . Automated analysis of RIP in AvrLm4-7 alleles was done using RIPCAL ( http://www . sourceforge . net/projects/ripcal ) , a software tool that computes RIP indexes and performs alignment-based analyses [25] . For high-throughput identification of AvrLm4-7 allelic variants before allele sequencing ( when relevant ) , High Resolution Melting PCR ( HRM PCR ) was used as an alternative to sequencing , using qPCR 7500 Fast Real-Time PCR equipment ( Applied Biosystems ) . A set of control isolates with known AvrLm4-7 sequences was included in each HRM-PCR run . Melting curves were analysed with High Resolution Melting software v2 . 0 ( Applied Biosystems ) . To recover DNA sequences flanking AvrLm4-7 , Thermal Asymmetric Interlaced PCR ( TAIL-PCR ) was used , following the design of nested AvrLm4-7 sequence-specific primers ( Table S4 in Text S1 ) . Arbitrary Degenerated ( AD ) primers used in association with AvrLm4-7 primers were AD1 , AD2 and AD3 [51] ( Table S4 in Text S1 ) , with AD2 used for the two subsequent rounds of PCR amplification . First and second rounds of TAIL PCR were done as described by Liu & Whittier [51] . Secondary TAIL-PCR products were purified using the Nucleospin Extract II purification Kit ( Macherey-Nagel , Hoerd , Fr ) and were used either for the third round of TAIL PCR , whenever the amount of amplified product was insufficient , or as template for DNA sequencing using the specific tertiary border primer Tail-GD3 as a sequencing primer ( Figure 3A ) . To validate the TAIL-PCR results , primers were designed with PRIMER 3 from the TAIL-PCR sequence product and used to PCR-amplify the corresponding sequence from genomic DNA of the corresponding isolate , with v23 . 1 . 3 as a negative control . For Southern blots , mycelia from the isolates grown in liquid Fries medium for two weeks were harvested by filtration , freeze-dried , ground to a fine powder , and DNA extraction was done as described by Balesdent et al . [52] . Southern blot analysis was performed on XbaI or SpeI or HpaI restricted genomic DNA ( 10 µg ) , size-fractionated on 0 . 8% w/v agarose gels and transferred to positively charged nylon membrane ( Qbiogen ) according to standard protocols . A 459-bp probe was generated by PCR using the AvrLm4-7Int-F and AvrLm4-7ext-R primers ( Table S4 in Text S1 ) and gel purified after electrophoresis using the NucleoSpin Plasmid QuickPure kit ( Machery-Nagel ) . Preparation of a [α-32P]dCTP probe was performed using the random priming Ready-To-Go DNA labelling beads kit ( GE Healthcare ) . High stringency hybridization ( 65°C ) was done using standard protocols . For Quantitative RT-PCR , cotyledons of cv . Westar ( susceptible control ) were inoculated and sampled 7 days after inoculation . Total RNA extraction and single-strand cDNA synthesis were performed as described by Fudal et al . [19] . Inoculation and RNA extraction were repeated twice . Water and RNA from plants inoculated with isolate G07-E441 , lacking the AvrLm4-7 gene , were used as negative controls . Primers for AvrLm4-7 amplification were as described by Parlange et al . [21] . qRT-PCR was performed using 7700 real-time PCR equipment ( Applied Biosystems , Foster City , CA , USA ) and ABsolute SYBR Green ROX dUTP Mix ( ABgene , Courtaboeuf , France ) , as described by Fudal et al . [19] . Actin was used as a constitutive reference gene . A multiplex PCR was used to characterize the distribution of the Mat1-1 and Mat1-2 alleles in the collection of isolates [53] . Seven genetically independent minisatellite markers ( Table S4 in Text S1 ) were used for population genetic analyses [54] . For each isolate , the allele sizes were determined using quantity one 1-D Analysis software ( BioRad , Marnes-la-Coquette , France ) by comparison with band sizes of the 1-kb+ ladder ( Invitrogen , Cergy Pontoise , France ) and internal control with known allele size and known number of repeats of the core motif ( i . e . , the sequenced reference v23 . 1 . 3 isolate ) . Data were scored as the number of repeat units for each isolate and each minisatellite locus . ABC methods implemented in the DIYABC program v1 . 0 . 4 . 39 [55] were used to estimate the effective population size ( Ne ) of the L . maculans populations sampled in 2006 , 2007 and 2008 from the susceptible cv . Campala . For the genetic parameters , the Generalized Stepwise Mutation ( GSM ) model was used to simulate mutations at the minisatellite loci and the prior interval specifications for the mean mutation rate were as described in Dilmaghani et al . [54] . A total of 1 , 000 , 000 data sets were simulated to generate a reference table . This reference table comprised summary statistics ( e . g . genetic diversity per sample and genetic distance between samples ) that enable estimation of the posterior distributions of the demographic parameters , under a given scenario , using comparisons between simulated and observed data sets . We used DIYABC to estimate effective population size under a simple scenario corresponding to one population , from which several samples had been taken over three consecutive generations ( three events of sexual reproduction ) . The summary statistics used were mean number of alleles per locus , mean genetic diversity [56] , mean variance in allele size , genetic differentiation between pairwise groups ( FST , [57] ) and genetic distances ( σμ ) 2 [58] . A local linear regression on the 1% simulated data sets closest to the observed data sets was then used to estimate the posterior distribution of the parameters . A generation time of 1 year was assumed , based on biological and epidemiological studies [24] , [59] . The frequency of each molecular event leading to virulence ( avrLm7 ) was calculated in the different isolate samples . Where more than one isolate was recovered from a single plant , as with isolates obtained from stem residues , a sub-sample was obtained by random selection of a single isolate from each individual plant . The molecular event for each isolate sampled was then recorded and the frequency of each molecular event was calculated on the basis of this sub-sampling . This randomized sub-sampling was repeated 10 , 000 times , and the mean frequency for each molecular event calculated . The resulting frequencies of each type of event were compared over the three years using the Approximate Pearson's Chi squared test with 20 , 000 randomizations as implemented in XLSTAT v2010 . 5 . 01 . To analyse minisatellite variability , the software FSTAT version 2 . 9 [60] was used to compute allele frequencies , number of alleles per minisatellite ( A ) , number of private alleles and Nei's gene diversity ( H ) [61] at each locus and over all loci , within and over the samples . Tests for differences between groups of samples comprising AvrLm7 or avrLm7 isolates for polymorphism statistics were based on two-sided permutation tests ( 15 , 000 permutations ) and performed using FSTAT . Linkage disequilibrium was evaluated using two different approaches . First , minisatellites were tested pairwise within and across samples using the genotypic disequilibrium test in Genepop [62] . The statistical significance of each pairwise test of linkage disequilibrium was tested by Fisher's exact test . The associations of alleles among different loci were also estimated with the standardized version of the index of association rd , using MULTILOCUS [63] . The significance of rd was established by comparing the observed value to the distribution obtained from 1000 randomizations with alleles at each locus being resampled without replacement to simulate the effect of random mating . The hypothesis of random mating was tested as follows: the distribution of mating types was compared to the 1∶1 ratio expected under random mating for a haploid fungus using χ2 tests . Genetic structure was analysed with standard FST coefficients of population differentiation , which were calculated and tested for significance using 1000 permutations using FSTAT . To further analyse population differentiation , heterogeneity of allele frequencies among samples was tested for each locus using the Fisher exact test in the GENEPOP program [63] . Genetic differentiation amongst samples was examined using an analysis of molecular variance ( AMOVA ) in Arlequin v3 . 5 [64] .
|
Plant disease resistance often relies on simple “gene-for-gene” systems and , in the pathogen , a mutation in a single “avirulence” gene matching the plant resistance gene is sufficient to render the resistance ineffective . In agricultural systems , breeding for resistance is challenged by both the high evolutionary potential of the pathogen and the large scale of crop production; together , these factors encourage “breakdown” of novel sources of resistance soon after their deployment . Here , we established a four-year field experiment to evaluate the mechanisms and speed with which a fungal pathogen of oilseed rape , Leptosphaeria maculans renders ineffective the novel resistance gene Rlm7 . The pathogen showed a very high evolutionary potential; the proportion of isolates in the population that were virulent against Rlm7 increased from 0 to 36% in four years . The experiment demonstrated that an extremely diverse range of molecular events leading to virulence , from more or less extensive nucleotide mutations or deletions to complete gene deletion , can occur in a single field . These results suggest that the genomic environment of the avirulence gene and the reproductive regime of the pathogen promote mutability at a single locus to produce virulence . Cropping practices that promote large pathogen populations and encourage sexual reproduction therefore favour rapid adaptation of the pathogen to the novel resistance .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"mutation",
"mycology",
"fungi",
"plant",
"microbiology",
"evolutionary",
"biology",
"plant",
"biology",
"crops",
"population",
"genetics",
"crop",
"diseases",
"biology",
"microbiology",
"host-pathogen",
"interaction",
"agriculture"
] |
2012
|
Genome Structure and Reproductive Behaviour Influence the Evolutionary Potential of a Fungal Phytopathogen
|
The mechanisms generating stably differentiated cell-types from multipotent precursors are key to understanding normal development and have implications for treatment of cancer and the therapeutic use of stem cells . Pigment cells are a major derivative of neural crest stem cells and a key model cell-type for our understanding of the genetics of cell differentiation . Several factors driving melanocyte fate specification have been identified , including the transcription factor and master regulator of melanocyte development , Mitf , and Wnt signalling and the multipotency and fate specification factor , Sox10 , which drive mitf expression . While these factors together drive multipotent neural crest cells to become specified melanoblasts , the mechanisms stabilising melanocyte differentiation remain unclear . Furthermore , there is controversy over whether Sox10 has an ongoing role in melanocyte differentiation . Here we use zebrafish to explore in vivo the gene regulatory network ( GRN ) underlying melanocyte specification and differentiation . We use an iterative process of mathematical modelling and experimental observation to explore methodically the core melanocyte GRN we have defined . We show that Sox10 is not required for ongoing differentiation and expression is downregulated in differentiating cells , in response to Mitfa and Hdac1 . Unexpectedly , we find that Sox10 represses Mitf-dependent expression of melanocyte differentiation genes . Our systems biology approach allowed us to predict two novel features of the melanocyte GRN , which we then validate experimentally . Specifically , we show that maintenance of mitfa expression is Mitfa-dependent , and identify Sox9b as providing an Mitfa-independent input to melanocyte differentiation . Our data supports our previous suggestion that Sox10 only functions transiently in regulation of mitfa and cannot be responsible for long-term maintenance of mitfa expression; indeed , Sox10 is likely to slow melanocyte differentiation in the zebrafish embryo . More generally , this novel approach to understanding melanocyte differentiation provides a basis for systematic modelling of differentiation in this and other cell-types .
Understanding the mechanisms of generation of differentiated cell-types from multipotent precursors is a fundamental aspect of development , with profound implications for the therapeutic use of stem cells . Whilst numerous transcription factors mediating fate choice from stem cells have been characterised , we still lack a robust understanding of how these factors and their target differentiation genes interact to form the gene regulatory networks ( GRNs ) that result in stable differentiation . At the time of fate specification , a multipotent cell's GRN is configured so as to allow multiple fates to be chosen; after specification this GRN must shift to a new stable state to establish commitment to , and full differentiation of , a specific fate . Tour de force studies of the early development of the sea urchin embryo have become perhaps the most completely understood example [1] . These studies , amongst others , have identified two key themes of fate specification , that the adopted fate becomes stabilized by factors initiating positive feedback loops and that these then are reinforced by activation of repressors of alternative fates [2] . Increasingly it is becoming clear that mathematical modelling of these proposed networks is very informative for a rigorous understanding of their properties [3]–[5] , but this remains rare , especially for vertebrate systems . Vertebrate melanocytes ( melanophores in fish , amphibians and reptiles ) are critical for body pigmentation and play roles , for example , in mate recognition and protection against UV light . Numerous diseases result from failures of melanocyte specification ( e . g . Waardenburg syndromes ) , differentiation ( albinism ) , survival ( vitiligo ) or control of proliferation ( melanoma ) [6] . Melanocytes are genetically amongst the best characterised cell-types , with a long history of genetic analysis in mammals [7] , but so far these data have not been used to generate mathematical models of melanocyte differentiation . Embryonic melanocytes are derived from the neural crest [8]–[10] and in the adult are renewed from dormant melanocyte stem cells [11] . Melanocyte specification centers on the transcriptional activation of Mitf , a bHLH-LZ transcription factor that is a master regulator of melanocyte differentiation [12] , [13] . Key target genes of Mitf include those encoding the melanogenic enzymes Dopachrome tautomerase ( Dct ) , Tyrosinase ( Tyr ) and Tyrosinase-related protein 1 ( Tyrp1 ) and the melanosome structural protein Silver ( Si ) . The Sox transcription factor Sox10 is also crucial for melanocyte development , where it contributes to melanocyte fate-specification by transcriptional activation of Mitf , consistent with the association of SOX10 with Waardenburg syndrome in humans [14]–[20] . Given that both MITF and SOX10 are frequently mutated in melanoma [21] and that MITF itself is considered to be a lineage addiction oncogene [22] , understanding the melanocyte GRN is of crucial importance . However , controversy surrounds the precise role of Sox10 , with in vivo data from zebrafish arguing that an ongoing role in melanocyte differentiation is not required in this organism [16] , while in vitro data from mouse indicates that Sox10 may contribute to expression of melanocyte differentiation genes , Dct and Tyr [23]–[27] . We here combine experimental and mathematical modelling approaches to examine this issue in more detail in zebrafish . We document the rapid loss of Sox10 from differentiating melanocytes in zebrafish embryos . We adapt a simple GRN model of sympathetic neuron development to the melanocyte case and assess its validity both experimentally and by mathematical modelling . This melanocyte model predicts that Sox10 represses expression of melanocyte differentiation genes , and that in this way Sox10 antagonizes Mitfa-mediated differentiation . Our analysis of gene expression patterns in zebrafish sox10 and mitfa mutants provides strong support for this , and overexpression studies in zebrafish embryos confirm the repressive action of Sox10 on Mitfa-mediated transcription . The model also predicts that the turning off of sox10 expression in differentiating melanocytes results from Mitfa-dependent repression of sox10 transcription . We provide evidence that Mitfa can regulate sox10 expression and that in vivo this effect is likely to be repressive and dependent upon Hdac1 function . We use simple mathematical modelling of this GRN , in conjunction with our previous experimental data , to establish that it is insufficient to explain stable melanocyte differentiation . We show that addition of further features , including a Sox10-independent positive feedback loop regulating mitfa , and a Sox10-independent weak activator of melanocyte differentiation gene expression , are sufficient to alter the GRN behaviour to allow stable differentiation of this cell-type and to explain our in vivo observations . Finally , we provide genetic evidence that Sox9b contributes to the second of these factors . The mathematical modelling of the melanocyte GRN proposed here provides the first such model for this important and well-characterised cell-type and provides the basis for future qualitative and quantitative refinement of our understanding of melanocyte differentiation . Our data supports the previous suggestion that Sox10 only functions transiently in mitfa expression and cannot be responsible for long-term maintenance of mitfa expression in zebrafish; indeed , Sox10 is likely to slow melanocyte differentiation in the embryo . This work has clear implications for the proposed model of sympathetic neuron differentiation , but also more broadly for our understanding of commitment to specific fates . Furthermore , these studies emphasize the importance of robust mathematical modelling of proposed GRNs to test their behaviour in a rigorous and quantitative manner .
Our previous studies have shown that sox10 mRNA expression is rapidly lost from differentiating sensory neurons [28] . We asked whether this pattern was seen for sox10 expression in differentiating melanocytes too . We used both whole-mount in situ hybridisation and immunofluorescence using a Sox10 antibody ( kind gift of B . Appel ) to evaluate the temporal persistence of Sox10 expression throughout a time-course ( Figure 1 ) . Melanocytes were selected at random from all dorso-ventral positions between the edge of the yolk and the end of the yolk sac extension . Expression was scored as the percentage of melanised cells showing detectable signal . The earliest signs of melanisation in trunk melanocytes in wild-type embryos are seen around 27 hpf [29] . At 30 hpf , almost all melanocytes showed detectable sox10 and Sox10 expression , but this rapidly decreased , so that by c . 50 hpf , signal was not detected in any cells . This contrasts with the continuing expression of mitfa ( data not shown and see Figure S3 ) . We note that at this stage , melanocyte differentiation and melanisation is still incomplete , and we conclude that expression of Sox10 is rapidly downregulated in differentiating melanocytes in zebrafish . Studies in mouse have not documented the temporal changes in Sox10 expression in vivo , but in adult human melanocytes there is evidence that SOX9 expression may partially replace SOX10 and is necessary for maintenance of melanocyte differentiation [30] . Strikingly , studies of cultured differentiating human melanoblasts show that SOX10 expression is lost in differentiating melanocytes , but that SOX9 expression is upregulated [31] . Neither of the zebrafish orthologues , sox9a and sox9b , have been reported as expressed in melanocytes [32] , [33] , [34] . To assess directly whether a similar shift from sox10 to sox9 expression might occur in zebrafish melanocytes , we used whole-mount in situ hybridisation to assess sox9a and sox9b expression in zebrafish embryos , but found no evidence for such expression between 24 hpf and 72 hpf ( Figure S1; data not shown ) . We conclude that in zebrafish embryos , sox10 expression is lost from differentiating melanocytes , but this is not replaced by sox9 gene expression . This pattern of sox10 expression attenuation during neural crest differentiation has also been described for the sympathetic neurons in mouse ( Figure 2A; [35] ) . These authors suggested a model whereby Sox10-mediated activation of MASH-1 and Phox2B drives sympathetic neuron specification , whilst initially feed-forward repression by Sox10 delays sympathetic neuron differentiation; subsequently negative feedback by MASH-1/Phox2B turns off Sox10 and differentiation can now proceed . In melanocyte development , Sox10 drives mitfa expression; we have shown in zebrafish that the interaction is direct and identified some of the relevant Sox10-binding sites in the mitfa promoter [16] . We asked to what extent the Kim et al . model could be generalised to another neural crest derivative . We proposed an analogous initial model of melanocyte differentiation in which Sox10 drives fate specification by activating mitfa expression , but perhaps delayed melanocyte differentiation by a feed-forward repression ( Figure 2B ) . Based on this analogy , we made two predictions . Firstly , melanocyte differentiation genes might be derepressed in sox10 mutants , just as , in Sox10 mutant mice , Phox2A expression is seen in the absence of MASH-1/Phox2B . Secondly , that sox10 repression would be directly or indirectly dependent upon mitfa expression . Here we explore these predictions experimentally . We had previously observed residual melanin in dorsal positions of 3 dpf zebrafish sox10 mutants , but had not examined this trait in detail [16] . Surprisingly , we had shown genetically that this residual melanin was independent of mitfa function; thus , it was consistent with possible derepression of melanocyte differentiation genes in sox10 mutants . We examined three series of sox10 mutant embryos , documenting the timing and appearance of these cells ( Figure S2 ) . Melanisation in these mutants is substantially delayed compared with wild-type siblings . In contrast to wild-type siblings which showed faint melanin from c . 25 hpf , we were unable to detect melanin before 36 hpf in any of 29 embryos followed ( Figure S2C ) . As in wild-types , numbers of melanised cells increased with developmental age , and tended to form in an anterior-posterior progression ( data not shown ) . Melanised cells were scored for their position with respect to the trunk and tail segments defined by the myotome . The numbers were very variable , with occasional embryos developing melanised cells in up to 21 segments ( n = 1 ) , whereas others never showed any ( n = 2 ) , and they were usually confined to the trunk and anterior-most tail , and never seen in the posterior-most tail ( Figure S2C and data not shown ) . As noted before , melanin is very faint in these cells , but it undergoes a dynamic change in appearance from initially rather diffuse to later more compacted , forming a tiny but dense spot ( Figure S2B ) . In summary , it seems that melanisation is highly residual and strongly delayed compared with wild-type siblings , consistent with low level derepression of melanogenic genes . From our model , we predicted that derepression of melanogenic genes would be detected as increased expression in sox10 mutants compared with mitfa mutants , and would be independent of Mitfa function i . e . would persist in sox10; mitfa double mutants . We had previously observed residual dct expression in dorsally-located cells in sox10 mutants [36] , but had not compared mitfa mutants . To assess whether dct and other key melanocyte differentiation genes were derepressed in sox10 mutants compared with mitfa mutants , we performed a series of parallel in situ hybridisation studies using four melanogenic genes , dct , tyrp1b , tyr and silva , on sox10 and mitfa mutant embryos ( Figure 3 ) . Pilot experiments showed that expression in mitfa mutants was extremely weak and was undetectable in fish older than 36 hpf , so careful comparisons were made at stages between 24 and 36 hpf ( Table 1 ) . In all cases , marker expression in wild-types was very strong , but to test for low level expression in mutants the in situs were stained longer , resulting in higher background than normal . Mutant embryos were developed in parallel with the same probe under identical conditions; expression of all markers in the pigmented retinal epithelium ( PRE ) was unaffected in each mutant and was used as an internal control for the procedure on each embryo . We saw a consistent pattern for all genes examined , with sox10 mutants showing slightly more elevated and more consistently-detectable expression ( i . e . a higher proportion of embryos showed a signal ) and a longer duration ( from 24 to 48+ hpf in sox10 mutants , but from 24 to 30+ hpf in mitfa mutants ) of detectable expression than mitfa mutants ( Figure 3 and Table 1 ) . The differential expression of dct , tyr and silva between the two mutants was striking; in contrast effects on tyrp1b were subtle , with very little detectable expression being seen ( Figure 3 ) , although this residual expression was more consistent and more prolonged ( Table 1 ) in sox10 mutants . As an independent confirmation of these data , we used quantitative real-time PCR on embryos at 30 , 36 and 72 hpf ( Figure S3 ) . As expected , expression levels of mitfa , dct and tyrp1b are all much reduced in both mutants compared with wild-types . However , consistent with our in situ hybridisation data , at 30 hpf , but not at later stages , the expression levels of dct , and to a much lesser extent tyrp1b , are significantly higher in sox10 mutants compared with mitfa mutants , confirming the weak and transient derepression of melanogenic genes in the sox10 mutant embryos . We had previously shown that residual melanin in sox10 mutants was not due to low level expression of Mitfa , since sox10;mitfa double mutants also showed residual melanisation . To assess whether the low level derepression of melanocyte differentiation genes was also independent of Mitfa , we repeated our in situ hybridisation studies on sox10;mitfa double mutants generated by crossing sox10+/t3;mitfaw2/w2 parents , so that all embryos were mitfa homozygotes , and 25% were double homozygotes . We focused on the 36 hpf stage , when mitfa mutants have consistently lost expression , but sox10 mutants show detectable levels ( Table 1 ) ; thus , if derepression of melanocyte gene expression was independent of Mitfa , we expected that 25% of embryos would show ‘rescue’ of differentiation gene expression . We found detectable expression of the markers in nearly 25% of embryos ( 12/60 , dct; 9/48 , silva; 9/49 , tyr; 7/45 , tyrp1b ) from this cross ( Figure 3 ) , and interpret these data as showing derepression in most sox10;mitfa double mutants . We conclude that melanocyte differentiation gene derepression is independent of mitfa . To test experimentally the conclusions from this loss of function analysis we performed overexpression experiments in early zebrafish embryos . Embryos were injected at 1-cell stage with 115 pg sox10 or 35 pg mitfa ( initial trials showed 115 pg of mitfa to induce severe lethality ) sense RNA; as controls we used 115 pg of sox10m618 or mitfaw2 RNA respectively which encode the loss of function mutant forms . Injected embryos were examined for induced gene expression by whole-mount in situ hybridisation at an early ( 6 hpf ) or later ( 10 . 5 hpf ) time-point; note that each of these times is prior to endogenous expression of any of the genes assessed . Mitfa expression might be expected to drive expression of most melanocyte differentiation genes , although data from mouse studies might suggest that Mitf alone may be insufficient for some genes , perhaps especially tyrosinase [24] . In contrast , our Sox10-mediated repression model predicts that Mitfa alone will be sufficient , but that whilst Sox10 alone would drive mitfa , Mitfa-dependent expression of other melanocyte differentiation genes ( with the likely exception of tyrp1b ) would be repressed by the presence of Sox10 . In all cases , the negative control RNAs induced no gene expression . We observed a clear-cut distinction between the effects of Sox10 and Mitfa overexpression ( Figure 4 ) . Overexpression of wild-type mitfa mRNA resulted in strong expression of all melanocyte differentiation genes by 6 hpf . In contrast , wild-type sox10 induced mitfa , but no melanocyte differentiation genes , by 6 hpf; by 10 . 5 hpf , tyrp1b was also induced , whereas dct , tyr and silva were not . That this tyrp1b expression was Mitfa-dependent was shown by injecting embryos from a cross of homozygous mitfaw2 mutants with sox10; whilst mitfa transcription was induced by 6 hpf , tyrp1b expression was never seen at 10 . 5 hpf ( data not shown ) . Our results were fully-consistent with our Sox10-mediated repression model with the modification that tyrp1b is insensitive to Sox10: Mitfa expression led to melanocyte differentiation gene expression by the early time-point , yet , whilst sox10 expression induced robust mitfa by the early time-point , even at the later one only tyrp1b was expressed . As a further test of our model , we asked whether co-injection of both mitfa and sox10 RNA would give a Sox10-like pattern of induction , but at the early time-point . Embryos were injected at 1-cell stage with 115 pg sox10 and 35 pg mitfa sense RNA; control embryos were injected with 115 pg of both sox10m618 and mitfaw2 RNA . Again the result was clear-cut; tyrp1b expression was readily detected at 6 hpf , whereas dct , silva and tyr were not ( Figure 5 ) . We conclude that Sox10 expression can repress the Mitfa-mediated expression of most of the melanocyte differentiation genes tested , but that tyrp1b expression is resistant to this effect , and that the timing of tyrp1b expression is limited by mitfa expression . Our simple melanocyte GRN predicts that loss of Sox10 expression results , directly or indirectly , from expression of Mitfa . There are no published reports of Mitf ( positively or negatively ) regulating sox10 expression , but we observed strong transcriptional activation of sox10 when Mitfa was overexpressed in early zebrafish embryos ( Figure 4 ) . This result was surprising since it is in direct contradiction to the predictions of our model , although it does suggest the possibility of Mitfa-mediated sox10 regulation in vivo . To begin to assess whether this might be direct regulation of the sox10 promoter by Mitfa , we asked whether GFP was activated in the Tg ( -7 . 2sox10:GFP ) reporter line , in which a 7 . 2 kb fragment of the promoter proximal region of sox10 genomic DNA drives GFP expression [37] . In the presence of mitfa overexpression , we noted clear GFP expression in transgenic fish at both an early time point ( 6 hpf ) , as well as a later ( 10 . 5 hpf ) one ( Figure 6; Table 2 ) , consistent with possible direct regulation . In contrast , in the same experiment , very few ( 6% ) embryos injected with sox10 RNA showed GFP expression at 6 hpf , whereas essentially all transgenic embryos showed GFP by 10 . 5 hpf , consistent with the idea that Sox10 does not directly regulate this reporter construct , but that Mitfa expression induced by Sox10 can do so . To test this suggestion that sox10 mRNA only results in expression of the Tg ( -7 . 2sox10:GFP ) transgene via production of Mitfa , we asked whether expression of the transgene fails in mitfa mutant embryos . Thus , we repeated the experiment from Figure 6 in mitfa mutant , Tg ( -7 . 2sox10:GFP ) embryos . As a positive control , we injected Tg ( -7 . 2sox10:GFP ) ;mitfaw2/w2 embryos with mitfa RNA; this frequently resulted in GFP expression at both 6 hpf ( 31/108 ( 29% ) injected embryos ) and 10 . 5 hpf ( 27/111 ( 24% ) ) . In contrast , injection of sox10 mRNA in mitfa mutants did not result in GFP expression at either 6 hpf ( 1/112 ( 1% ) ) or 10 . 5 hpf ( 0/105 ( 0% ) ) . Interestingly , these data suggest that , in contrast to dct and other differentiation genes , Mitfa-dependent expression of sox10 is not repressed by the presence of Sox10 . The 7 . 2 kb of sox10 regulatory sequences in the Tg ( -7 . 2sox10:GFP ) transgene contains 6 concensus M boxes , making it plausible that Mitfa binds directly to this promoter . To begin to narrow the region of the sox10 promoter likely to mediate this response to Mitfa , we repeated these experiments in the Tg ( -4 . 9sox10:GFP ) line [28] in which the 5′ three M boxes are absent . Interestingly , this transgene shows no response to injected Mitfa at 6 hpf ( Figure S4 ) . Our data suggest that Mitfa can regulate sox10 expression , but these experiments , examining the reporter in the context of zebrafish blastomeres , do not necessarily reflect the promoter's response in melanoblasts . To address more directly how sox10 expression might be regulated by Mitfa in the endogenous situation i . e . in the melanocyte lineage , we examined sox10 expression in mitfa mutants; if Mitfa is necessary for repression of sox10 we predicted that mitfa mutants should show persistent sox10 expression . We examined mitfa mutants at 72 hpf , a stage when wild-type embryos show no detectable sox10 expression in melanocytes , but show strong expression in the peripheral nervous system and ear ( Figure 7A , 7B ) . In mitfa mutants , in addition to the peripheral nervous system expression , we see readily detectable sox10 expression in the position of the dorsal stripe ( Figure 7D , 7E ) . Furthermore , in mitfa mutants we also see a similar pattern of mitfa expression in this same region ( Figure 7F ) , strongly suggesting that these cells are neural crest-derived melanocyte precursors that are unable to differentiate due to the lack of functional Mitfa protein . We tentatively conclude that Mitfa can regulate the sox10 promoter , and that this interaction is likely to have a repressive function in vivo in differentiating melanocytes . In considering whether any known factors might contribute to this loss of sox10 expression in melanocytes , we noted the persistence of sox10 expression described in colgate/hdac1 mutants [38] . Histone deacetylase1 is a component of multiple complexes that modify chromatin , resulting in selective repression of gene expression . Consistent with the predictions of our model , hdac1 mutants show both persistent sox10 expression in neural crest cells and poor melanocyte differentiation , although the connection between these phenotypes was not addressed . To assess whether persistent sox10 expression in melanocytes was associated with the delay in differentiation , we used chemical inhibition of histone deacetylase function [39] at the time of early melanocyte differentiation , asking whether this resulted in poor melanocyte differentiation and if this correlated with persistence of sox10 expression . Trichostatin A was applied in each of four time windows: 12–48 hpf , 24–48 hpf , 30–48 hpf and 36–48 hpf . Embryos treated in the 12–48 hpf window showed severe morphological defects , lacking anterior head , but also showed a dramatic reduction in melanocyte pigmentation ( data not shown ) . Those treated from 24–48 hpf again showed severe reductions in melanocyte differentiation ( Figure 8D–8F ) . Although these embryos were of normal morphology , they did appear to show slight retardation , having a morphology similar to approximately 36 hpf embryos . However , comparison of the degree of melanisation of an untreated 36 hpf embryo with the nominally 48 hpf Trichostatin A-treated embryos indicated a clear reduction beyond that expected from delayed development . Later treatment windows showed only weak effects on melanocyte differentiation ( data not shown ) . Using in situ hybridisation we were further able to show that treated embryos showed substantially elevated levels of persistent sox10 expression in melanocytes ( Figure 8N , 8Q ) . Furthermore , our model requires Hdac-mediated repression of sox10 expression to be Mitfa-dependent; hence it predicts that Trichostatin A treatment of mitfa mutants would not result in further elevation of sox10 levels above those of untreated mitfa mutant controls . An experimental test of this prediction showed that , indeed , sox10 expression in mitfa mutant embryos is not further elevated by Trichostatin A treatment ( Figure S5 ) . Taken together , our data lead us to conclude that repression of sox10 expression in the melanocyte lineage is both Mitfa-dependent and Hdac-dependent , ( most likely mediated by Hdac1 [38] ) , and that these mechanisms contribute to the differentiation of zebrafish melanocytes in vivo . Our experimental data was consistent with the major predictions of the simple melanocyte GRN that we had proposed . To assess the GRN more rigorously , and to develop a more quantitative understanding of the model , we turned to mathematical modelling . We constructed a simple dynamical model of the GRN based upon ordinary differential equations , where the transcript concentrations were considered as dynamic variables . Our model aimed to describe the mutual regulation of the genes involved in the GRN by simple activatory and repressive dynamics , and the response of the GRN to external activatory signals , designated Factor A . Studies of both mouse and zebrafish have identified multiple enhancers that drive sox10 gene expression in neural crest and its derivatives [37] , [40] , [41] , [42] , [43] . The factors binding those enhancers are only poorly characterised in both species , but may include Lef/Tcf ( downstream of Wnt signalling ) , Sox9 , FoxD3 , Pax and AP2 . Since this regulation is poorly understood , for the purposes of our modelling we combine these factors into one composite Factor A . It is currently unclear whether in a neural crest cell context these signals are merely transient , or are constantly available . However , given the highly dispersive nature of neural crest cells , we might assume external signals , like Wnt , may be rather transient . Similarly , zebrafish sox9 , sox10 , foxd3 , pax and tfap2 are all downregulated in neural crest cells as they differentiate into melanocytes [34] , [44]–[48]; this work ) . Nevertheless , the data from mitfa mutants in Figure 7 indicate that , at least in the vicinity of the dorsal neural tube , one or more components of Factor A remain present at 72 hpf at least . Consequently , for the purposes of our modelling studies , we assumed that Factor A was constant throughout embryonic development . We explored the rigorous predictions of this initial melanocyte GRN ( Model A , Figure 9A ) by direct simulation with a widespread exploration of parameter space . Given the lack of quantitative knowledge of most parameters , we restricted ourselves to assessing under which conditions ( i . e . parameter value sets ) the model predicted i ) the long-term maintenance of mitfa expression , ii ) an initial increase of sox10 , leading to its maximal expression at intermediate times , and iii ) long-term loss ( or downregulation , i . e . below a detection threshold ) of sox10 expression , as we have observed in differentiating melanocytes . Direct numerical integration of the ODE system of Model A revealed that the model predicts that both mitfa and sox10 expression are maintained ( Figure 9B ) . However , we found that no parameter settings allowed us to obtain an appreciable difference between sox10 maximal expression and its steady-state value ( see Figure S6 ) , as implied by requirements ii ) and iii ) above . Maintenance of both mitfa and sox10 arises because the sox10-inducing signals ( Factor A ) are maintained , and these in turn maintain Mitfa expression . Our experimental data above indicates that sox10-inducing signals do seem to persist , at least in the vicinity of the neural tube . However , we note that experimentally , maintenance of Mitfa can be uncoupled from production of Sox10 . Our previous study showed that in sox10 mutant neural crest , transient expression of mitfa is sufficient to generate stable ( to 5 dpf at least ) melanocyte differentiation ( Elworthy et al , 2003 [16] ) . Since this demonstrates that stable melanocyte differentiation can occur in the absence of Sox10 activity if Mitfa is provided even transiently , we rejected Model A as too simplistic . In addition , we noted that it did not incorporate the complexities of Mitfa-mediated regulation of Sox10 as revealed by our experimental studies . Consequently , we explored the features of a revised model ( Model B , Figure 9A ) incorporating modifications expected to correct these deficiencies . Firstly , we introduce a Sox10-independent positive feedback loop on Mitfa ( Factor Y ) . Secondly , we add our demonstration that Mitfa-dependent activation of Hdac contributes to the repression of Sox10 . Model B predicts that in mitfa mutant embryos , mitfa transcription should be substantially decreased , due to the absence of the positive feedback through Factor Y . In situ hybridisation shows that mitfa expression in mitfaw2 mutants is distinctly decreased at 30 and 36 hpf ( [13] , and data not shown ) , but given that this mutant results in a premature stop codon , nonsense-mediated decay might also explain the lowered mRNA levels . We thus supplemented these observations with analysis of embryos homozygous for the single amino acid substitution ( I121S ) allele , mitfab692 [49] . In these mutants , we again observed an unambiguous substantial reduction in the levels of mitfa transcripts in the mutant embryos ( Figure 10A , 10B ) , thus providing support for the biological validity of Factor Y . Mitfa itself is a clear candidate for Factor Y , and indeed in mouse Mitf functions in conjunction with Lef1 and b-catenin to regulate the Mitf promoter [50] . As an initial test whether Mitfa might regulate its own promoter , we asked whether injection of mitfa mRNA would induce transcription of the endogenous mitfa gene . We used an in situ hybridisation probe for the 3′-UTR of mitfa , since the injected mRNA lacks these sequences , as well as examining dct induction as a positive control for Mitfa activity . We saw induction of both dct and mitfa expression upon injection of RNA encoding WT mitfa ( Figure 10C , 10D , 10G , 10H ) , whereas neither were seen after injection of RNA encoding either of the Mitfa mutants , Mitfa ( b692 ) or Mitfa ( w2 ) ( Figure 10E , 10F; data not shown ) . We conclude that a Sox10-independent , Mitfa-dependent Factor Y , predicted from mathematical modelling ( and perhaps Mitfa itself ) , is likely to play a major role in maintaining melanocyte differentiation . Contrary to our intuition , mathematical simulation of Model B showed that this revised model still failed to generate the required downregulation of sox10 under conditions where mitfa was maintained ( see Figure S7 ) . Furthermore , it failed to predict two aspects of the phenotype in sox10 mutant embryos . We found that three further refinements to produce a third model ( Model C , Figure 9A ) were required for the model to reproduce the experimentally-demonstrated behaviour , as we discuss in the next section . The first modification required is a change to the way that Hdac1-mediated repression functions on sox10 expression . In Model B , we postulated that Hdac1 represses Mitfa-dependent sox10 transcription . However , we found that this was inadequate to allow repression of sox10 expression in the wild-type ( Figure 9C ) , since constant Factor A persists ( Figure S7 ) . In this context , the identification of Hdac as a repressive factor becomes rather striking , since the effects of deacetylation might be expected to affect multiple enhancer elements . As we have noted experimentally , sox10 expression is repressed in differentiating melanocytes , so in Model C we show Hdac as repressing Factor A-dependent sox10 expression , as well as Mitfa-dependent activation of sox10 transcription ( Figure 9A ) . This model now reproduces the wild-type observations ( Figure 9D and Figure S8 ) . Secondly , we found that it is crucial to incorporate a threshold response within the Factor Y-mediated feedback in Model B . In the absence of such a threshold , the positive feedback of Factor Y ensures that in sox10 mutants the absence of melanocyte differentiation is only an unstable state associated with [mitfa] = 0 , since even the lowest level expression of mitfa would be expected to trigger positive feedback leading to high level mitfa expression and subsequent melanocyte differentiation ( Figure 9C , sox10+Mitfa ) . The biological observations are unambiguous – even vaguely normal looking melanocytes are exceptionally rare in sox10 mutants ( RNK , pers . obs . ) – suggesting that the positive feedback loop with Factor Y must exhibit threshold behaviour , so that the [mitfa] = 0 state is stabilised at low levels of Mitfa or of Y . In both sox10 and mitfa mutants expressing Mitfa under the sox10 promoter , melanocyte rescue is relatively unlikely ( 70% of embryos show no melanocytes , and most embryos showing rescue show <10 melanocytes per embryo [16] ) , but when it does occur melanocyte morphology and differentiation appear normal , consistent with the GRN being bistable . To account for this behaviour , we have incorporated a threshold response to the Factor Y feedback loop . Thirdly , Model B failed to predict the low level derepression of melanocyte differentiation genes in sox10 or sox10;mitfa double mutants ( data not shown ) . One solution to this problem , a Sox10-independent Factor Z driving ( low level ) expression of melanocyte differentiation genes , is incorporated into Model C ( Figure 9A ) . Our efforts to model Factor Z initially assumed that it was driven by Factor A , and thus remained constant . However , under these assumptions , we were unable to reproduce the very weak and transient expression of differentiation genes observed experimentally . Instead , we made the assumption that Factor Z is activated by an unknown Factor B , and Factor B is only transiently expressed in the melanocyte lineage . Mathematical exploration of this model shows that , whilst the non-zero wild-type steady state seen before in Model B is conserved , Model C also reproduces the gene expression patterns seen in sox10 , mitfa and sox10;mitfa mutants ( Figure 9D ) . In particular , expression of dct ( representing the melanocyte differentiation genes repressed by Sox10 ) is seen at low levels in mitfa , sox10 and sox10;mitfa mutants , but this is weakest and most transient in mitfa mutants . This modelling is only useful in so far as it allows us to correctly predict novel features of the biology . We chose to explore candidates for Factor Z . Such genes would have no prominent role in wild-type melanocytes ( i . e . loss of gene function would not have a melanisation defect ) , but they would need to be expressed in neural crest cells and to drive low level melanisation in sox10 mutants; in addition they would be only transiently expressed in melanocyte progenitors . In adult human melanocytes SOX9 is likely to regulate DCT [30] . There are two zebrafish orthologues of SOX9 , but neither sox9a nor sox9b nor sox9a;sox9b mutants show a loss of melanisation [34] . Unlike sox9a , sox9b is expressed in early neural crest cells , but then is downregulated ahead of sox10 in progenitors for all except craniofacial cartilage ( data not shown; [34] ) . We used previously published sox9b morpholinos [51] to address whether morpholino-mediated knockdown of Sox9b would result in loss of residual melanin in sox10 mutants ( Figure 11 ) . The numbers of residual melanised cells in sox10 mutants at 2 days post fertilisation ( dpf ) was significantly reduced in embryos injected with 0 . 5 ng of each sox9b morpholino compared with embryos injected with sox9b mismatch morpholinos ( Figure 11A–11C ) . We deduce that Sox9b can drive Sox10 and Mitfa-independent melanisation displayed by sox10 mutants . We conclude that Sox9b shows the characteristics predicted for Factor Z and that it at least contributes to this role in zebrafish . Furthermore , our data provides biological validation of Factor Z , a second feature of the melanocyte GRN predicted as a result of the mathematical modelling . We also note the transient expression of sox9b in NCCs , broadly consistent with our deductions from the modelling above .
In this study we have used a combination of genetic experimentation and mathematical modelling to build upon our initial description of melanocyte specification under the control of Sox10 [16] . We have considerably expanded and refined the GRN associated with melanocyte specification and differentiation in embryonic zebrafish ( Figure 12 ) . We have shown multiple new features , including 1 ) Sox10-mediated repression of many Mitfa target genes; 2 ) the transient nature of Sox10 expression in differentiating melanocytes , resulting from 3 ) Mitfa-dependent repression of Sox10 , likely via 4 ) a mechanism involving Hdac1 complex; and 5 ) Sox10-independent weak activation of melanogenesis genes . An early comparison of the core GRN of melanocytes in mouse and zebrafish had concluded that they were evolutionarily divergent [24] . That comparison focused on a basic description of the role of Sox10 in melanocyte differentiation , noting that in zebrafish there was no requirement beyond melanocyte specification ( i . e . activation of mitfa ) , whereas it was required positively both for melanocyte specification ( Mitf expression ) and differentiation ( Tyr expression ) in mouse . The more extensive examination of the zebrafish GRN presented here both supports the suggestion of some evolutionary divergence in the role of Sox10 , but also identifies a series of new features that will need to be examined in the mouse system . The data in the Hou et al study show that Mitf is not sufficient to rescue melanisation in Sox10 mutant neural crest cells , at least in primary cultures of neural crest cells , since Sox10 function is also required to drive Tyr expression [24] . Our data validate our previous conclusion that ongoing Sox10 function is not necessary for melanocyte differentiation in zebrafish in vivo , since mitfa expression in early neural crest cells was sufficient to fully rescue melanocyte differentiation , even up to 5 dpf [16] . However , we now show that Sox10 does have a role beyond melanocyte specification ( i . e . transcriptional activation of mitfa ) , although it appears to be purely repressive . Certainly , the effects of Sox10 on Tyr expression in mouse ( synergistic activation with Mitf ) and zebrafish ( antagonistic repression ) are in stark contrast . These data now make untenable the conclusion reached by Hou et al that the differences in the role of Sox10 might explain the differences in timing of melanisation in mammals ( late ) and fish ( early ) [24] . Further work to define in much greater detail the melanocyte GRN in each species will allow identification of the key differences between them actually controlling the distinctive timing of melanisation . Our observations in zebrafish beg the question of whether there is Sox10-dependent repression of melanocyte genes in vivo in mouse . Such studies are hindered by issues of sensitivity of whole mount in situ hybridization and the difficulties of directly comparing gene expression levels in melanocytes of wild-type and mutant strains , but one recent paper attempts to standardise the analysis of gene expression for multiple melanocyte markers in E11 . 5 mouse embryos . Using their semi-quantitative scoring system , Gpnmb ( but not Dct , Si , or Tyr ) expression is detectable in Sox10LacZ/LacZ mutants but not in MitfMi/Mi mutants [52] , providing a hint that Sox10-dependent repression of melanocyte differentiation genes may occur in mouse . It certainly seems surprising that two homologous cell-types , with striking conserved phenotypic characteristics , might show such a substantial change in their GRN . Comparison of GRNs in an evolutionary context is still in its infancy , but already examples of substantial differences between the circuitry of homologous cell-types are known . For example , in echinoderm development , conserved gene expression in homologous domains of sea urchins and sea stars often results from divergent regulatory inputs i . e . the output is conserved , but the regulatory mechanism has diverged [53] . Conceptually , it is trivial to imagine how mutations in regions near the binding site of an activatory transcription factor might allow binding of a co-repressor at that promoter . It will be exciting to identify the molecular basis for the change in Sox10 function . But what might be the biological function of the Feed-Forward Repression by Sox10 ? In the mouse sympathetic neuron , Kim et al suggest that this circuitry delays differentiation and maintains multipotency [35] . Delay of melanocyte differentiation and maintenance of progenitor multipotency is an attractive hypothesis in the zebrafish too . Recent study of an mitfa:GFP transgenic line indicates that not all neural crest cells that turn on mitfa will become melanocytes , since some will form iridophores instead ( Curran et al . , 2010 ) . Thus , in zebrafish expression of mitfa does not represent commitment to the melanocyte lineage; the Feed-Forward Repression loop we have defined might contribute to that maintenance of multipotency in the early melanocyte precursor . Loss of Sox10 expression would then be necessary for commitment to a differentiated state . In this context , it is intriguing that mouse melanocytes , which retain Sox10 expression , appear to have also retained multipotency , which can be exhibited when isolated and cultured [54] . We have proposed that Sox10 functions to delay melanocyte differentiation in embryonic zebrafish . Likewise , a similar conclusion was reached for the role of Pax3 in adult mouse melanocyte stem cell differentiation . Thus Lang and colleagues demonstrated that Pax3 acted with Sox10 to drive transcription of Mitf , whilst feed-forward repression by Pax3 delayed expression of dct [55] . Pax3 morphants are not described as having a dramatic melanocyte differentiation phenotype , but the detailed timing of melanocyte differentiation was not examined [44] . Our initial investigations using Pax3 morpholinos ( MN and RNK , data not shown ) have failed to detect an effect on either wild-type or sox10 mutant melanogenesis , so it remains unclear whether the role for Pax3 is conserved in fish . One key feature of the zebrafish melanocyte GRN that we have uncovered is the rapid down-regulation of sox10 during early differentiation . A major task will be to elucidate the molecular basis for this . Our study only begins to address this issue , indicating that sox10 repression in melanocytes is Mitfa-dependent , but leaves open whether sox10 is a direct target of Mitfa . Development of further tools for the zebrafish , especially good antibodies for Mitfa to allow ChIP-chip or ChIP-seq studies , will allow this important question to be addressed definitively . Our initial data provide a strong hint that the effect of Mitf , whether direct or indirect , on sox10 is highly context dependent; Mitfa activates the sox10 promoter in the context of embryonic blastomeres , whereas it represses the same promoter in the context of melanoblasts . We note that the 7 . 2 kb genomic DNA fragment in the Tg ( -7 . 2sox10:GFP ) reporter that responds to Mitfa contains 6 consensus M boxes , whereas 3 of these are missing in the Tg ( -4 . 9sox10:GFP ) that does not [56] . Testing whether Mitfa directly regulates sox10 in vivo via one or more of the 5′ M boxes is a priority for future work . We hypothesize that the presence of a repressive cofactor in melanoblasts alters the effect of Mitfa on the sox10 promoter . Little is known of repressive cofactors in zebrafish melanocyte development . Zebrafish histone deacetylase1/colgate ( hdac1/col ) mutants showed delayed melanocyte differentiation; whilst sox10 expression in early neural crest was indistinguishable from wild-type , sox10 expression was prolonged to at least 52 hpf , although it was unclear if these phenotypes were causally linked [38] . We have shown here that chemical inhibition of Hdac function during the phase of early melanocyte differentiation results in prolonged sox10 expression in differentiating neural crest cells , and in impaired melanogenesis . This is strikingly consistent with the core GRN we have identified here , and supports the hypothesis that Mitfa-dependent repression of sox10 requires Hdac1 . However hdac1 expression is both maternal and zygotic [57] , so transcriptional regulation of hdac1 itself by Mitfa is unlikely to explain the repression of sox10 in differentiating melanocytes . We speculate that Mitfa may regulate recruitment to the sox10 promoter of an Hdac1 complex [58] , resulting in deacetylation of this chromatin and repression of sox10 transcription . The identification of Mitfa-dependent activation of the Hdac complex proved crucial to explain the repression of sox10 transcription . In our modelling we initially assumed that Mitfa-dependent repression affected only the regulation by Mitfa itself , switching it from an activator to a repressor . However , modelling the GRN in this way proved ineffective , because it failed to shut-down sox10 transcription , apparently due to the fact that whilst the Mitfa influence was repressed , input from Factor A persisted , and hence Factor A-dependent expression became dominant . The realization that Hdac complex mediated the Mitfa-dependent repression immediately provided a resolution to this problem , since deacetylation would be expected to repress activity of many/all enhancers of the sox10 gene , making it likely that Factor A-dependent sox10 expression , as well as Mitfa-dependent expression , would be inactivated in the wild-type situation . In contrast , in the mitfa mutant situation , Factor A remains , so that we see persistent sox10 and mitfa expression , just as observed in vivo . Satisfyingly , this was exactly the behaviour we saw when we modeled the GRN in the light of this insight . Hence , whilst the presence of Factor A seems to persist , as revealed by the mitfa mutant phenotype , our intuition that the influence of Factor A would be transient in the wild-type situation appears to be well-founded , resulting from the global shut-down of sox10 transcription mediated by Hdac complex . We have demonstrated for the first time that in the presence of Sox10 , many Mitfa-mediated transcriptional responses are repressed . At first glance , it is surprising therefore that when we over-express Mitfa in zebrafish blastomeres , melanocyte differentiation genes are expressed robustly , despite the observation that sox10 is also expressed . We propose that the explanation lies in the timing of expression of Sox10 protein . When sox10 mRNA is injected alone , Sox10 protein forms before mitfa can be transcribed . Thus , Mitfa protein is functioning in the context of substantial amounts of Sox10; in contrast , when mitfa is expressed alone , Mitfa protein is functioning before sox10 transcription and hence is working in the absence of Sox10 protein . The test of this is the coinjection of both sox10 and mitfa mRNAs; in this context both Sox10 and Mitfa proteins would be formed together and hence again Mitfa would be functioning in the context of Sox10 protein . The prediction is that melanocyte differentiation genes would be repressed; this prediction is directly borne out by our experimental test ( Figure 5 ) . We conclude that our data is , in fact , consistent in suggesting that Sox10 represses Mitfa-mediated melanocyte differentiation . Nonetheless melanocyte differentiation in vivo occurs whilst sox10 transcripts remain detectable ( Figure 1 ) . We propose that , in part , the explanation lies in Sox10-mediated repression depending more on the ratio of Sox10:Mitfa proteins: our preliminary data exploring the effects of changed ratios of sox10:mitfa supports this [56] . In mouse sympathetic neuron differentiation , Sox10 heterozygotes show derepression of Phox2A , but normal expression of MASH1 and Phox2B , indicating that here higher levels of Sox10 are required for repression of differentiation than for specification [35] . In addition , the explanation likely lies in the complex integration of multiple factors as inputs on melanocyte differentiation gene expression . Thus , here we have identified Sox9b as an unexpected factor driving melanocyte differentiation . Given that , as we show here , sox9b expression is not detectable in differentiating melanocytes , this role must be transient , and restricted to the early phase of melanocyte development . Whilst melanisation is consistently repressed in sox10 mutants injected with sox9b morpholinos , effects on residual dct expression were more variable; whereas sox10 mutant embryos injected with the mismatch morpholino showed low level dct expression , this expression was sometimes reduced in sox9b morphant;sox10 mutant embryos , although not statistically significant overall ( MN and RNK , data not shown ) . We suggest that at these early stages of melanocyte differentiation , dct expression reflects the integration of multiple activatory ( Mitfa , Sox9b , others ? ) and inhibitory ( Sox10 , others ? ) inputs . Our mathematical modelling here ( Figure 9D ) shows that this scenario can generate a convincing reproduction of our semi-quantitative in situ observations . The challenge for the future will be in vivo quantitation of the various key parameters of the model in order to examine how precisely the model and the in vivo situation match each other . Our mathematical modelling approach , used iteratively with experimental data , has made specific predictions about the properties of currently unidentified factors in melanocyte differentiation . Importantly , we illustrate the power of our systems biology approach by experimentally identifying Sox9b as a factor fulfilling the properties of Factor Z . Our data here on melanocytes extends the evidence for partial redundancy of Sox10 and Sox9b in neural crest development initially shown for sensory neurons [28] . Indeed , we noticed that sox9b morphants also show significantly reduced numbers of ‘escaper’ iridophores too ( MN and RNK , data not shown ) , suggesting this partial redundancy between these closely-related transcription factors may be a general feature . Our modelling also implied the activity of a Sox10-independent , Mitf-dependent transcriptional activator of Mitfa , Factor Y , providing a positive feedback loop to allow stable melanocyte differentiation . We demonstrate that in mitfa mutant zebrafish embryos , mitfa expression is reduced compared with wild-type siblings consistent with our suggestion of a role for Mitfa in maintaining mitfa expression . Consistent with this , we also show that overexpression of Mitfa results in rapid , precocious expression of the endogenous mitfa gene . Likewise , while Mitf expression in mouse E11 . 5 embryos is prominent throughout the body , in MitfMi/Mi mutants it is weakened and only detectable in the tail , the developmentally youngest region [52] . These data strongly support the suggestion from our modelling that maintenance of mitfa expression is ( directly or indirectly ) dependent upon Mitfa function , and that this feedback is conserved in mouse melanocytes too . Apart from Sox10 , several other transcription factors have been shown to regulate Mitf [59] . One candidate for Factor Y is CREB , acting downstream of elevated cAMP induced by Melanocyte Stimulating Hormone ( MSH ) /Melanocortin Receptor 1 ( Mc1R ) signalling [60] . MSH has a clear role in background adaptation , and Mc1R expression is maintained throughout embryonic development [61] , [62] . However , current evidence for the role of Mc1R in melanisation in zebrafish based on morpholino knockdown is conflicting [63] , [64] . In our attempts to reproduce these morpholino studies we saw a transient decrease in melanisation , consistent with [63] , but this seemed to be in large part due to embryonic retardation , indicating that , in agreement with [64] , Mc1R signalling in zebrafish is unlikely to play a major role in melanocyte melanisation ( LV and RNK , data not shown ) . We conclude that Mc1R signalling is not likely to contribute to Factor Y , at least in the embryonic melanocytes . Understanding the mechanisms stabilizing the differentiated melanocyte fate is likely to have particular relevance for our understanding of melanoma . Levels of the steady state activity of Mitf appear to be crucial to the melanoma phenotype , with high Mitf activity associated with differentiation and lowered levels with proliferation and melanoma [65] . Several factors identified as regulating Mitf in development , also play major roles in melanoma; for example , WNT/b-catenin dependent regulation of MITF transcription has been demonstrated by chromatin immunoprecipitation and plays a major role in the transformed phenotype by promoting both proliferation and survival of melanoma cells [66] . A mouse melanoma model generated by combining melanocyte-specific expression of both constitutively active β-catenin and activated N-ras generates frequent melanomas [67] . In the context of our work , it is interesting that melanocytes from this strain frequently become immortalised , and do not fully pigment [67] . In conclusion , our systems biology approach has identified several new and unexpected features to the core GRN underlying melanocyte specification and differentiation in vivo . We have demonstrated a role for Sox10 in antagonising Mitfa-dependent differentiation; have firstly predicted , then identified Sox9b as part of , a factor with a transient role in Mitfa-independent melanisation observed in sox10 and sox10;mitfa mutants; have predicted and then shown that mitfa expression is , directly or indirectly , Mitfa-dependent; and have provided the first indication that Mitfa might negatively regulate sox10 expression in differentiating melanocytes . Both the latter mechanisms are likely to be major factors stabilising differentiation of melanocytes in zebrafish . The stage is now set for a comprehensive analysis of the zebrafish melanocyte GRN , by incorporation into the model of other known and unknown regulatory functions combined with a network analysis of the motifs identified therein , in order to truly understand the basis for stable differentiation of this medically-important cell-type . We suggest that application of our approach to other medically-important cell-types is likely to be valuable .
This study was performed with the approval of the University of Bath ethics committee and in full accordance with the Animals ( Scientific Procedures ) Act 1986 . Embryos were obtained from natural crosses and staged according to Kimmel et al . [68] . We used the sox10t3 allele [29] , the mitfaw2 [13] allele except where stated otherwise , when we used mitfab692 [49] , and the Tg ( -4725sox10:GFP ) ba3 and Tg ( -4 . 9sox10:EGFP ) ba2 lines [28] , [37] . RNA in situ hybridization was performed according to Thisse et al . [69] , except probes were not hydrolysed and embryos were incubated at 68°C in hybridization steps . Probes used were sox10 [15] , dct [36] , mitfa [13] , silva ( ZIRC cb397; [70] , tyrosinase [71] , tyrp1b ( clone number 6894514 from Geneservice , GenBank reference CB353867 , subcloned as an EcoRI/XhoI fragment into Bluescript ) , gch [72] , xdh [72] and paics ( Plasmid and probe generated by T . Chipperfield and C . Nelson ) . Antibody staining with anti-Sox10 ( 1∶10000 , [73] ) and Alexa Fluor 488 ( 1∶2000 , Invitrogen , A21206 ) was performed largely as Ungos et al . [74] . Embryos were viewed using an Eclipse E800 ( Nikon ) using either DIC or fluorescence microscopy as appropriate . Embryos were scored for Sox10 and sox10 expression by scoring 20 pigmented melanocytes in each of 5 embryos at each time point . One cell stage embryos were injected with RNA using standard methods as in Dutton et al . [15] . RNA was produced and recovered using the mMESSAGE mMACHINE and MEGAclear kits ( Ambion ) from hs>sox10 and hs>sox10m618 templates linearized with Asp718 [15] or CS2+mitfaWT and CS2+mitfaw2 linearised with Not1 [13] . sox10 , sox10m618 and mitfaw2 RNA were diluted to a concentration of 25 ng/µl , mitfa RNA was diluted to 6 . 25 ng/µl including 0 . 0005% Phenol Red . Embryos were injected with 4 . 6 nl RNA and grown for 6 or 10 . 5 hours at 28 . 5°C . Embryos were then processed for in situ hybridisation or scored for GFP fluorescence using an MZ12 dissecting microscope ( Leica ) . DNA sequence was submitted to TRANSFAC public version 6 . 0 using the Pattern Search for Transcription Factor Binding Sites ( PATCH 1 . 0 ) interface . Parameters were set to look for vertebrate transcription factor binding sites of 6 bp or more with the maximum number of mismatches being set at zero [75] . Trichostatin A ( TSA , [R- ( E , E ) ]-7-[4- ( Dimethylamino ) phenyl]-N-hydroxy-4 , 6-dimethyl-7-oxo-2 , 4-heptadienamide ) ( Sigma-Aldrich ) was kept as a 5 mM in DMSO stock solution ( 0 . 2 µm-filtered ) at −20C . Batches of embryos were treated with 1 µM Trichostatin A in Petri dishes , during each of four time windows ( from 12 hpf to 48 hpf , from 24 hpf to 48 hpf , from 30 hpf to 48 hpf and from 36 hpf to 48 hpf ) at 28 . 5°C . Control embryos received equivalent doses of DMSO alone . Melanocyte phenotypes of live embryos were documented at 48 hpf under a Nikon E800 microscope; embryos were anesthetized with Tricaine ( Sigma-Aldrich ) and mounted on slides under coverslips in 30% methylcellulose . RNA was extracted from samples of 40 embryos of each genotype ( decapitated after anaesthesis with Tricaine ) using TRIREAGENT ( Sigma-Aldrich , T9424 ) . First strand cDNA was synthesized using the Invitrogen First strand cDNA synthesis kit with Superscript III and random hexamers . Real time quantitative PCR was performed in duplicate using SYBR Green I PCR Master Mix ( Roche ) and a Lightcycler II machine according to the manufacturer's instructions . Primers were designed spanning an intron using Primer3 Plus software ( http://www . bioinformatics . nl/cgi-bin/primer3plus/primer3plus . cgi . ) . The following primers were used: gapdh: forward ACCAACTGCCTGGCTCCT , reverse TACTTTGCCTACAGCCTTGG; mitfa: forward CTGGACCATGTGGCAAGTTT , reverse GAGGTTGTGGTTGTCCTTCT; dct: forward TCTTCCCACCTGTGACCAAT , reverse CTGATGTGTCCAGCTCTCCA; trp1b: forward CGACAACCTGGGATACACCT , reverse AACCAGCACCACTGCAACTA . Gene expression was normalized against zebrafish gapdh expression in wild-type embryos . Quantitative RT-PCR data were analysed using the ( ΔΔCt ) method [76] . Student's t-test with Bonferroni correction for multiple comparison were performed using GraphPadPrism 5 . 0 to test the null hypothesis that there was no significant difference in gene expression levels between mitfa and sox10 mutants . In all tests , difference was considered significant if p<0 . 017 . We constructed a mathematical model for gene regulation as a one stage process: binding and unbinding of transcription factors ( TFs ) to DNA was assumed to regulate protein production in a single step of synthesis , without explicit modelling of intermediate mRNA levels . The model was expressed in terms of a system of ordinary differential equations ( ODEs ) . Binding and unbinding of TFs were described as faster processes than protein synthesis and degradation . This allowed us to solve the transcript dynamics in conditions of quasi-equilibrium for the TFs . The result was a description of both activatory and repressive regulation in terms of Hill-like functions . By using appropriate combinations of Hill functions , Models A , B and C ( Figure 9A ) were then described mathematically . The derivation is presented in the accompanying Text S1 . Models were investigated by direct numerical integration and , in the case of Model C for the sox10 mutant , by steady-state analysis . The steady-state analysis was obtained by setting time derivatives to zero , and by solving analytically the corresponding set of algebraic equations . This gave information about the long time behaviour of this GRN , and allowed us to draw conclusions independent of the particular set of chosen parameters . The time-dependent solution was computed numerically by using a standard finite differences algorithm ( Euler ) . Parameter values were chosen so as to reproduce the sought behavior , constrained by available experimental evidence whenever possible . For instance , knowledge about typical time scales of the relevant concentrations fixed gene expression and decay rates . Furthermore , the robustness of our conclusions with respect to the chosen parameter values was assessed by plotting the steady state value of Mitfa , and the steady state and the maximal values of Sox10 , as functions of the different activatory and repressive regulations between mitfa and sox10 in all studied models ( see Figures S6 , S7 , S8 ) . Here our aim was not to identify a unique parameter set that reproduced the experimental data , but rather to assess to what extent our conclusions might be broadly independent of the specifically chosen parameters . In this sense our results should be taken as qualitative , given the lack of knowledge of most parameter values , but still representative of typical dynamical behavior .
|
In a multicellular organism , one genome is used to make numerous different cell-types . This must require the activity of all these genes to be configured into multiple distinct and stable active states , each corresponding to one of the different cell-types characteristic of a tissue . The stable active states of differentiated cell-types contrast with the different , and transient , states characteristic of multipotent stem cells . We know little of the key features of these states that regulate the switch of a stem cell to stable differentiation . Here we examine this issue in the melanocyte , a genetically well-characterised cell-type , using a combination of dynamic mathematical modelling and experimental manipulation . In humans , disruption of the melanocyte state results in congenital and degenerative pigmentary diseases , whereas their destabilisation is likely to be an important factor in initiating melanoma . Our work predicts , validates , and identifies several novel features to the gene regulatory network of the zebrafish melanocyte , including one stabilising the differentiated state . Our study demonstrates the utility of this systems biology approach to understanding the genetic basis for differentiated cell states .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"animal",
"genetics",
"cell",
"differentiation",
"gene",
"function",
"animal",
"models",
"developmental",
"biology",
"model",
"organisms",
"gene",
"expression",
"biology",
"systems",
"biology",
"zebrafish",
"genetics",
"gene",
"networks",
"genetics",
"and",
"genomics",
"cell",
"fate",
"determination"
] |
2011
|
An Iterative Genetic and Dynamical Modelling Approach Identifies Novel Features of the Gene Regulatory Network Underlying Melanocyte Development
|
To date there are no approved antiviral drugs for the treatment of Ebola virus disease ( EVD ) . While a number of candidate drugs have shown limited efficacy in vitro and/or in non-human primate studies , differences in experimental methodologies make it difficult to compare their therapeutic effectiveness . Using an in vitro model of Ebola Zaire replication with transcription-competent virus like particles ( trVLPs ) , requiring only level 2 biosafety containment , we compared the activities of the type I interferons ( IFNs ) IFN-α and IFN-ß , a panel of viral polymerase inhibitors ( lamivudine ( 3TC ) , zidovudine ( AZT ) tenofovir ( TFV ) , favipiravir ( FPV ) , the active metabolite of brincidofovir , cidofovir ( CDF ) ) , and the estrogen receptor modulator , toremifene ( TOR ) , in inhibiting viral replication in dose-response and time course studies . We also tested 28 two- and 56 three-drug combinations against Ebola replication . IFN-α and IFN-ß inhibited viral replication 24 hours post-infection ( IC50 0 . 038μM and 0 . 016μM , respectively ) . 3TC , AZT and TFV inhibited Ebola replication when used alone ( 50–62% ) or in combination ( 87% ) . They exhibited lower IC50 ( 0 . 98–6 . 2μM ) compared with FPV ( 36 . 8μM ) , when administered 24 hours post-infection . Unexpectedly , CDF had a narrow therapeutic window ( 6 . 25–25μM ) . When dosed >50μM , CDF treatment enhanced viral infection . IFN-ß exhibited strong synergy with 3TC ( 97 . 3% inhibition ) or in triple combination with 3TC and AZT ( 95 . 8% inhibition ) . This study demonstrates that IFNs and viral polymerase inhibitors may have utility in EVD . We identified several 2 and 3 drug combinations with strong anti-Ebola activity , confirmed in studies using fully infectious ZEBOV , providing a rationale for testing combination therapies in animal models of lethal Ebola challenge . These studies open up new possibilities for novel therapeutic options , in particular combination therapies , which could prevent and treat Ebola infection and potentially reduce drug resistance .
As of December 13 , 2015 , the current outbreak of Ebola virus disease ( EVD ) in West Africa has resulted in 28 , 633 cumulative cases and 11 , 314 deaths [1] . Two potential vaccine candidates , rVSVΔG-ZEBOV and ChAd3-EBO Z , have shown durable protection from lethal Ebola challenge in mice [2] and macaques [3] respectively , and are part of the phase II/III PREVAIL trial in Liberia and Guinea ( https://clinicaltrials . gov/ct2/show/NCT02344407 ) . Other potential therapeutics , such as convalescent plasma and the antibody cocktail ZMapp [4] have been approved for an emergency phase II/III trial in Guinea ( https://clinicaltrials . gov/ct2/show/NCT02342171 ) and a phase I trial in Liberia ( https://clinicaltrials . gov/ct2/show/NCT02363322 ) , respectively . However , to date there is no licensed vaccine or treatment for EVD , although improvements in supportive care are increasing survival rates [5] . Repurposing antivirals used for other viral infections , based on knowledge of mechanisms of action , has prompted accumulating interest in the application of different nucleoside/nucleotide analogs and type I interferons ( IFNs ) for the treatment of Ebola virus disease ( EVD ) . Experimental nucleoside analogs may have therapeutic efficacy for EVD , given the evidence of protection in primate and rodent disease models , 2–6 days after lethal Ebola or the related hemorrhagic Marburg virus challenges [6 , 7] . Favipiravir , a viral polymerase inhibitor , provides 100% protection when administered 6 days after challenge with a lethal dose of Ebola virus [6] and has been evaluated in the phase II/III JIKI trial in Guinea ( https://clinicaltrials . gov/ct2/show/NCT02329054 ) . TKM-100802 , a cocktail of siRNAs targeting VP35 and L polymerase and brincidofovir ( BCV ) , a viral polymerase inhibitor that has activity against dsDNA viruses such as adenovirus and cytomegalovirus [8] , were also considered for treatment against EVD . The brincidofovir trial was halted , ostensibly because of projections of low recruitment . Despite infecting different target cells , Ebola and HIV-1 share many similar features early in their replication cycle . Both are RNA viruses that package a viral polymerase ( L for Ebola , RT for HIV-1 ) required for early replication in the cytosol of the host cell [9] . Homology-based structural prediction of the RNA-dependant RNA polymerase of Ebola indicates the polymerase contains conserved structural motifs in the catalytic palm subdomain similar to viral DNA polymerases [10] , supportive of nucleoside analogs potentially inhibiting Ebola replication . Inhibiting HIV-1 reverse transcription with nucleoside analogs such as lamivudine ( 3TC , cytidine analog ) , zidovudine ( AZT , thymidine analog ) or tenofovir ( TFV , adenosine monophosphate analog ) is the basis for highly active antiretroviral treatment ( HAART ) [11 , 12] . Nucleoside analogs are on the WHO list of essential medicines and can be deployed in limited resource settings [13] . Moreover , AZT binds RNA through G-C and A-U bases [14] , prompting us to evaluate whether these nucleoside analogs might also inhibit Ebola replication . Type I IFNs mediate diverse biological effects , including cell type-independent antiviral responses and cell type-restricted responses of immunological relevance . IFNs inhibit viral infection by preventing viral entry into target cells and by blocking different stages of the viral replication cycle for different viruses . Moreover , type I IFNs have a critical role in linking the innate and adaptive immune responses to viral challenge . IFN-α/β expression occurs as the earliest non-specific response to viral infection . Indeed , viruses have evolved immune evasion strategies specifically targeted against an IFN response , confirming the importance of IFNs as antivirals . This immune evasion strategy is relevant when one considers the IFN response to Ebola infection [15] . Ebola proteins VP24 and VP35 inhibit host cell systems that lead to IFN production and also inhibit events associated with an IFN response [16–18] . VP24 blocks the binding of importins to phosphorylated STAT1 , preventing STAT1 nuclear translocation required for transcription of interferon simulated genes [16] . VP35 binds viral dsRNA , preventing dsRNA degradation [17] and inhibits the phosphorylation of IRF-3 and the SUMOylation of IRF-3 and IRF-7 , thereby limiting IFN production [18] . Despite these virally-encoded mechanisms to limit an IFN response to infection , different rodent and non-human primate studies provide evidence for IFN-induced partial protection: the effects of IFN-α/β treatment in lethal Ebola virus infection reduced viremia and prolonged survival [19–21] . Thus , a potential therapeutic effect for IFNs as monotherapy in EVD , or in combination with other anti-Ebola therapies , has not been resolved .
We employed an established mini-genome system to rapidly evaluate candidate drugs that could inhibit Ebola Zaire replication under BSL 2 conditions [22–24] . At the outset we established the experimental conditions for infection with replication and transcription-competent virus like particles ( trVLPs ) , by examining luciferase activity under various transfection and drug treatment conditions , which included transfection with viral support protein plasmids ( S1 Fig ) . We included treatment with maraviroc , a CCR5 inhibitor , that would have no effect on trVLP entry and infection , thereby serving as a negative control for subsequent treatment regimens . In a first series of experiments , we examined the inhibitory effects of IFN-α ( 0 . 5μM/10 , 000 U/mL ) , IFN-ß ( 0 . 2μM/1 , 000 U/ml ) , TOR ( 5μM ) , CDF ( 100μM ) , FPV ( 100μM ) , and a combination of 3TC , AZT and TFV ( 5μM each ) on trVLP infection of 293T cells ( Fig 1 ) . Specifically , the 293 T cells were treated with the different drugs at four different times relative to infection with trVLP , as indicated . We provide evidence that for each of the individual drugs and for the triple drug combination , at the doses indicated , trVLP infection of 293 T cells is inhibited when treatment is initiated at +2 , +6 or +24 hours post-infection . Interestingly , TOR , an estrogen receptor modulator discovered in a high throughput screen as a potent inhibitor of Ebola [25] , significantly reduced viral luciferase activity at all time-points tested . For IFN-α , IFN-ß , TOR and FPV treatments , maximal inhibition of trVLP infection was achieved when the cells were treated prior to challenge with trVLP . By contrast , pre-treatment with CDF at 100μM , 24 hours prior to infection with trVLP , resulted in enhanced infection . In subsequent dose-response studies , we compared the inhibitory effects of IFN-α , IFN-ß , TOR , CDF , FPV , 3TC , AZT or TFV when administered 24 hours post trVLP infection ( Fig 2 ) . The data in Fig 2I summarize the IC50 dose for each drug . The IFNs exhibited the lowest IC50 values at 0 . 016μM for IFN-ß and 0 . 038μM for IFN-α . The data show a log-fold difference in IC50 values for IFN-α and IFN-ß when compared in terms of U/ml , the norm for antiviral activity measurements ( Fig 2A and 2B ) . TOR had the next lowest IC50 ( 0 . 36μM ) and completely inhibited infection at doses > 5μM ( Fig 2C ) . TFV had an IC50 at 0 . 98μM . CDF , 3TC and AZT all exhibited similar IC50 values in the dose range 4 . 2–7 . 8μM , while FPV had the highest IC50 of the nucleoside analogs at 36 . 8μM . At their IC50 concentration , none of these drugs directly inhibited luciferase reporter activity ( S2 Fig ) . We observed a relatively small antiviral dose range for CDF ( 1 . 5–25μM ) ( Fig 2D ) , beyond which the drug appeared to enhance viral infection ( S3 Fig ) . In cell viability assays we observe that at doses >10μM CDF affect cell viability , confounding the interpretation of the effects of CDF on viral replication . In an orthogonal assay to confirm these findings , we next measured viral replication and transcription by qRT-PCR , following trVLP infection . trVLP-infected cells were either left untreated , or treated with the different drugs 24 hours post-infection , then viral replication and transcription evaluated 24 hours later ( Fig 3 ) . All treatments , with the exception of TOR , significantly reduced the amount of genomic vRNA detected within cells ( Fig 3A ) and all treatments significantly reduced the synthesis of cRNA and mRNA isolated from infected cells ( Fig 3B ) . Notably , IFN-ß treatment of trVLP-infected cells resulted in the greatest reduction in viral replication and transcription . Next we examined the effectiveness of two and three drug combinations on trVLP infection . We first examined 28 two-drug combinations , using each drug at its IC50 value , and used the median-effect equation and combination index theorem [26] to determine drug synergy , additive or sub-additive effects ( Fig 4A ) . Synergy is defined as greater than additive effect when drugs were combined ( CI<1 ) , additive as the effect expected when combining each drug ( CI = 1 ) and sub-additive as a smaller than expected additive effect ( CI>1 ) . When administered 24 hours post-infection , many of the two-drug combinations showed strong synergism in inhibiting trVLP replication ( Fig 4J ) , with IFN-β + 3TC demonstrating the greatest synergism ( 97 . 3% inhibition , CI = 0 . 028 ) . 3TC was synergistic with all seven other drugs tested . Notably , when CDF was used in combination with FPV , AZT , TFV or IFN-α , it produced a sub-additive effect . Next we tested all possible 56 three-drug combinations , using each drug at its IC50 value , to assess whether adding a third drug enhanced efficacy compared with two-drug combinations ( Fig 4B–4I ) . This series of experiments served to validate our two-drug findings , as synergistic two-drug combinations such as IFN-β + 3TC and IFN-β + AZT , predicted strong synergy for the triple drug combination of IFN-ß + 3TC + AZT . As anticipated from the two-drug polygonogram , CDF was sub-additive when combined in three-drug combinations ( Fig 4E ) . This was most evident even when CDF was administered in conjunction with two-drug combinations that had shown strong synergy , such as IFN-β + 3TC or FPV + TFV , further indicating that CDF diminishes the antiviral effects of other drugs . IFN-ß , 3TC , AZT and TFV all promoted strong synergism when included in triple drug combinations , with IFN-β + AZT specifically providing strong synergism when combined in three unique triple therapies . From these two-drug and three-drug screens , we calculated the combination index ( CI ) and fractional inhibition ( Fi ) ( Fig 4J and 4K ) . Many of the synergistic drug combinations ( i . e . low CI ) included one nucleoside analog and an IFN , while those drug combinations that were sub-additive all included CDF . IFN-β was predominant in the most efficacious two- and three-drug combinations . In particular , IFN-β + 3TC and IFN-β + 3TC + AZT consistently exhibited the strongest synergism and highest Fi when administered 24 hours post-infection . Refer also to S1 and S2 Tables . In a final series of experiments , in order to validate our findings from the trVLP infection studies , we examined the antiviral effectiveness of IFN-ß , IFN-α , TOR , FVP , AZT , 3TC and TFV in 293T cells infected with ZEBOV ( ZEBOV contained an eGFP reporter ) . CDF was excluded from these experiments . Initial dose-response studies were conducted at doses reflective of those used in the trVLP experiments in Fig 2 . A higher dose of each drug was required to inhibit ZEBOV infection compared with trVLP infection ( S4 Fig ) . Using the IC25 of each drug , we next evaluated 2 and 3 drug combinations for additive or synergistic effects against ZEBOV infection . All seven 2 drug combinations were synergistic ( low CI ) ( Fig 5A ) , similar to the most synergistic combinations against trVLP in Fig 4J . IFN-β + 3TC proved to be the most synergistic 2 drug combination , analogous to trVLP infection . Of the most synergistic 3 drug combinations identified in the trVLP infection system , all seven exhibited synergy against ZEBOV infection , with IFN-β + 3TC + AZT and IFN-β + TOR + AZT exhibiting the strongest synergy ( Fig 5B ) . The CIs determined from trVLP infection correlated well with those determined using ZEBOV infection; specifically , the correlation coefficients ( R2 values ) confirm this ( Fig 5C and 5D ) .
In September 2014 , the WHO hosted a conference to facilitate development of a global action plan to deal with the Ebola outbreak in West Africa . Delegates from affected West African countries , ethicists , scientists , health care providers , logisticians and representatives from different funding agencies were in attendance . A committee had been struck to evaluate the different vaccine candidates and therapeutic interventions being developed , which subsequently received an overwhelming number of submissions for consideration , and was hampered by an inability to compare antiviral effectiveness , since in vitro and pre-clinical in vivo model systems vary , treatment regimens vary from prophylaxis to post-exposure administration , and direct readouts of antiviral efficacy differ . Moreover , given the virulence and high mortality associated with EVD , all of these studies have been conducted under BSL 4 conditions , limiting the number of laboratories that can engage in these antiviral studies . Cognizant of these limitations , we employed the trVLP model system to compare the antiviral effectiveness of eight antiviral candidates from three drug classes . We evaluated their antiviral activities in the context of inhibition of Ebola replication , using this mini-genome model that allows for rapid comparisons among compounds under BSL 2 conditions . The tetracistronic minigenome represents the most sophisticated in vitro replication model of Ebola virus to date . trVLPs proceed through every replication step as wild-type Ebola virus , and have been tested in multiple cell lines . Using TOR , there has been some validation of the trVLP assay . Specifically , TOR has been evaluated in limiting Ebola virus infection of VeroE6 and HepG2 cells , and exhibited IC50 values of 0 . 2 μM and 0 . 03 μM , respectively [27] , in line with the IC50 dose for TOR ( 0 . 36μM ) observed with trVLP infection . Likewise , the IC50 identified in the trVLP system for FPV ( 36 . 8μM ) , is consistent with that of 67μM recorded using Ebola virus infection [6] , suggesting that this Ebola mini-genome system has relevance for screening potential antiviral compounds . Indeed , our validation studies using ZEBOV ( ZEBOV-eGFP ) suggest that the trVLP infection model has utility as an in vitro screening assay when comparing different drugs as monotherapies or in 2 and 3 drug combinations . As mentioned , the Ebola virus encodes in its genome factors that limit a type I IFN response to infection [16–18] . Yet , both rodent and non-human primate studies suggest that IFN-α and IFN-ß treatment can confer partial protection from infection , reducing viremia and prolonging survival [19–21] , suggesting that it may be possible to override the inhibitory effects of the virus by treatment with IFN . At the outset , we conducted a series of experiments to compare the antiviral activities of IFN-α and IFN-ß in the trVLP infection system , and our findings suggest that whether treatment is administered prior to or post-infection , both IFN-α and IFN-ß exhibit antiviral activity . These findings only have relevance for the direct antiviral activities of these IFNs , since the effects of IFN-α or IFN-ß on immune modulation for viral clearance cannot be determined using this system . Nevertheless , these data contributed to the decision to conduct a clinical trial of IFN-ß treatment for EVD in Guinea . We provide evidence that the nucleoside/nucleotide analogs 3TC , AZT , TFV , FPV and CDF inhibit Ebola trVLP replication in vitro . The results with 3TC are in contrast to published data that show no evidence for 3TC inhibiting Ebola virus infection in vitro [27] . These studies examined the antiviral effectiveness of 3TC when administered one hour prior to infection , in contrast to our studies that have focused on post-exposure protection . In cells , the kinetics of 3TC phosphorylation are such that a minimum of four hours are required for optimal activity , perhaps distinguishing why our 24 hour pre-treatment , specifically a combination treatment , offered protection . Post-exposure treatment with 3TC and the other nucleoside/nucleotide analogs we examined , would more likely reveal activity against viral RNA synthesis than pre-treatment . When comparing the IC50 values of each of the nucleoside analogs that we tested , TFV exhibited the lowest IC50 at ~1μM . Whether this reflects the fact that this adenosine monophosphate analog only requires two phosphorylation events to become an active drug versus three for the other nucleoside analogs , remains undetermined . Extensive published data reveal both the safety profiles [11 , 28 , 29] and the biodistribution of 3TC , AZT and TFV in the circulation and liver [30 , 31] , the same compartments where Ebola infects monocytes , macrophages , dendritic cells , endothelial cells and hepatocytes . Moreover , drug interactions with other nucleoside analogs have been well studied: e . g . tenofovir disoproxil fumarate , when used alone or in combination with emtricitabine effectively prevents HIV-1 infection in antiretroviral pre-exposure prophylaxis ( PrEP ) [29] . Our studies also revealed that the active metabolite of brincidofovir , CDF , has a narrow therapeutic window of efficacy ( 6 . 25–25μM ) when assessed in the trVLP assay , enhancing viral replication at higher doses when added either prior to or post-infection . In cell viability assays , CDF exhibits cytotoxicity at doses >10μM . These findings suggest that caution is required if CDF is to be considered further for the treatment of EVD , specifically that phase I/II trials define the safety profile of this drug for EVD . Another advantage of this in vitro system is that it allowed us to evaluate various 2 and 3 drug combinations and demonstrates that combination treatments limit viral replication up to 97 . 3% . A benefit of combination treatment is the potential to limit/avoid the emergence of drug resistance . Interestingly , IFN-ß was predominant among all the 8 antivirals considered in terms of contributing very strong synergism in combination treatments: e . g . IFN-ß + 3TC; IFN-ß + 3TC + AZT . Using this system , we observe that FPV , when administered 24 hours post-infection , has an IC50 of ~ 37μM . To date , the phase II/III JIKI trial examining the efficacy of FPV against EVD has reported only modestly encouraging results . In our 2 drug combination treatment studies we show that , with the exception of CDF , whenever FPV is included , synergy occurs , effectively reducing the CI . It may transpire that for treating EVD , FPV is most effective in a drug combination regimen . Viewed altogether , we present an in vitro Ebola trVLP screening system , that requires only level 2 biocontainment , which allowed us to compare the antiviral activities of 8 compounds , either alone or in combination . We provide evidence that IFNs are effective inhibitors of Ebola replication , with IFN-ß exhibiting greater efficacy over IFN-α , or when used in combination with nucleoside analogs . We infer from our data that whether IFN-ß treatment is administered 24 hours prior to , or up to 24 hours post-infection , reduced Ebola replication is achieved . As additional antiviral therapeutic candidates become available , we now have the capability to measure and compare their direct antiviral activities with the existing panel . This allows for rapid in vitro evaluation and the opportunity to prioritize antiviral candidates for further pre-clinical and clinical trial studies .
We employed an established mini-genome system to rapidly evaluate candidate drugs that could inhibit Ebola Zaire replication under BSL 2 conditions [22] . The mini-genome encodes 3 of the 7 Ebola proteins ( VP24 , VP40 and GP1 , 2 ) and a luciferase reporter gene . Expression plasmids for the remaining four Ebola nucleocapsid proteins ( L , NP , VP30 and VP35 ) were also included during transfection . Cell culture conditions and virus infections were performed as previously described [22] . Briefly , 80 , 000 producer 293 T cells ( American Type Culture Collection; ATCC , Rockville , USA ) were seeded in individual wells of 24-well plates in 400μL Dulbecco’s Modified Eagle Medium ( DMEM ) containing 10% FBS , 1% penicillin and 1% streptomycin , and grown in 5% C02 atmosphere at 37°C . Cells were transfected with the viral replication protein plasmids ( L , NP , VP30 , VP35 ) , a tetracistronic Ebola mini-genome and the T7 polymerase , using the CalPhos Mammalian Transfection Kit ( Clontech Laboratories ) . 24 hours later , medium was replaced with 800μL DMEM with 5% FBS . The replication and transcription-competent virus like particles ( trVLPs ) were harvested 3 days later . Virus stock was frozen at -80°C . For infection , 293 T target cells were seeded at 80 , 000 cells in 400μL of DMEM supplemented with 10% FBS . Target cells were then transfected with the four viral replication protein plasmids , as well as Tim-1 , to allow efficient virus binding and entry . 24hr post-transfection , 25μL of trVLP stock was diluted in 600μL of DMEM with 5% FBS , warmed to 37°C for 30 min , then added to target cells . Medium was removed the following day and replaced with 800μL DMEM with 5% FBS . Four days post-infection , the medium was aspirated and cells were re-suspended in 200μL of 1x Renila Luciferase Assay Lysis Buffer ( Renilla Luciferase Assay System , Promega ) . Lysates were assayed for luciferase activity . We generated recombinant ZEBOV expressing enhanced green fluorescent protein ( eGFP ) from cDNA clones of full-length infectious ZEBOV , as previously described [32] . The eGFP reporter protein was expressed as an eighth gene , and the virus exhibited an in vitro phenotype similar to wild-type ZEBOV . Notably , in vivo , incorporation of GFP into wild-type ZEBOV results in some attenuation of disease [32] . All work with infectious ZEBOV was performed in biosafety level 4 ( BSL4 ) , at the National Microbiology Laboratory of the Public Health Agency of Canada in Winnipeg , Manitoba . 30 , 000 293 T cells were seeded in 96-well plates in 100μL DMEM with 10% FBS . 24 hours thereafter , the medium was replaced with 100μL DMEM with 10% FBS containing ZEBOV-GFP at an MOI of 0 . 1 . 24 hours post-infection , the medium was removed and replaced with 200μL of DMEM with 5% FBS , or 190μL DMEM with 5% FBS and 10μL of single or combinations of drugs . eGFP fluorescence was measured 3 days post-infection using a Synergy HTX Multi-Mode Microplate Reader ( BioTek ) . For these experiments , we used toremifiene citrate ( TOR; Sigma ) , cidofovir hydrate ( CDF; Sigma ) favipiravir ( FPV , T-705; Cellagen Technology ) , lamivudine ( 3TC; Sigma ) zidovudine ( AZT ) , tenofovir ( TFV ) maraviroc ( MVC; NIH AIDS Reagent Program ) , Infergen ( IFN alfacon-1 , Pharmunion Bsv Development Ltd . ) or human interferon beta-1a ( IFN-β , Avonex; Biogen ) . Forty-eight hours after trVLP infection , medium was aspirated from 293 T cells that had either been left untreated or treated with the various drugs and total RNA extracted from cell lysates with 500μL of TRIzol ( Thermo Fisher Scientific ) . cDNA synthesis was performed on 5 μg of total RNA , using the First-Strand cDNA Synthesis Kit ( GE Healthcare Life Sciences ) , according to the manufacturer’s instructions . A 20 μl reaction also contained bulk first-strand cDNA reaction mix , DTT solution and 40 pmol of one of two trVLP specific primers [33]: vRNA forward ( 5’-GGC CTC TTC TTA TTT ATG GCG A -3’ ) , or cRNA/mRNA reverse ( 5’-AGA ACC ATT ACC AGA TTT GCC TGA-3’ ) . Both primers were synthesized by the Center for Applied Genomics ( The Hospital for Sick Children , Toronto , Canada ) . Real-time qPCR reactions ( 25 μl ) were conducted in duplicate , using the Rotor-Gene RG-3000 thermocycler ( Corbett Research , Montreal , Canada ) . Each reaction contained 100 ng template cDNA , 12 . 5 μL 2 x SYBR Green PCR Master Mix ( Applied Biosystems , Warrington , UK ) , 300 nM of both the forward ( vRNA ) and reverse ( cRNA/mRNA ) primers , and PCR grade H2O ( Roche Diagnostics , Indianapolis , USA ) . Samples lacking reverse transcriptase ( No RT ) during first-strand cDNA synthesis served as negative controls . Cycling parameters were as follows: initial denaturation at 95°C for 10 min , followed by 40 cycles of amplification with 95°C for 15 seconds , 56°C for 30 seconds , and 60°C for 30 seconds . Biological triplicates in the drug-treated groups were normalized to the average Ct of infected cells given DMSO solvent alone , by the 2-ΔCT comparative CT method . Dose-response cytotoxicity/viability assays were conducted in 293 T cells 4 days post-infection for each of the drugs examined , either alone or in the various combinations indicated , using the MTT assay as previously described [34] . Means were compared using a two-tailed , unpaired Student’s t test and corrected for multiple comparisons . For all figures , ( * ) denotes a p value <0 . 05 , ( ** ) denotes a p value <0 . 01 and ( *** ) denotes a p value <0 . 001 . Error bars shown are the standard error around the mean ( SEM ) . Synergy between two and three-drug combinations , combination index ( CI ) and dose-reduction index ( DRI ) were calculated with CompuSyn Version 1 . 0 [26] . The coefficient of determination ( R2 ) was determined for simple linear regressions .
|
Studies to evaluate the effectiveness of candidate antiviral drugs to inhibit Ebola virus infection have been hampered by the availability and access to level 4 containment facilities . Using a mini-genome model system that generates Ebola virus-like particles that infect cells , we have been able to screen a panel of candidate drugs for antiviral activity , under normal level 2 containment . We compared the activities of 8 different antivirals from 3 drug classes , including drugs repurposed for the treatment of Ebola: type I interferons and nucleoside analogs . Our data indicate that IFN-ß is a potent inhibitor of Ebola virus , contributing to the decision to conduct a clinical trial of IFN-ß treatment for Ebola virus disease in Guinea . Moreover , we identified that 2 and 3 drug combinations inhibit Ebola replication when administered 24 hours post-infection . Drug combinations have important implications for clinical use , since lower doses of each drug are administered , potentially decreasing side-effects and , based on different mechanisms of action , there is less likelihood for the emergence of drug resistance . These studies set the stage for both preclinical and clinical evaluation .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2016
|
A Rapid Screening Assay Identifies Monotherapy with Interferon-ß and Combination Therapies with Nucleoside Analogs as Effective Inhibitors of Ebola Virus
|
The molecular composition of the cannabinoid type 1 ( CB1 ) receptor complex beyond the classical G-protein signaling components is not known . Using proteomics on mouse cortex in vivo , we pulled down proteins interacting with CB1 in neurons and show that the CB1 receptor assembles with multiple members of the WAVE1 complex and the RhoGTPase Rac1 and modulates their activity . Activation levels of CB1 receptor directly impacted on actin polymerization and stability via WAVE1 in growth cones of developing neurons , leading to their collapse , as well as in synaptic spines of mature neurons , leading to their retraction . In adult mice , CB1 receptor agonists attenuated activity-dependent remodeling of dendritic spines in spinal cord neurons in vivo and suppressed inflammatory pain by regulating the WAVE1 complex . This study reports novel signaling mechanisms for cannabinoidergic modulation of the nervous system and demonstrates a previously unreported role for the WAVE1 complex in therapeutic applications of cannabinoids .
Psychoactive cannabinoids derived from marijuana as well as diverse endocannabinoids , such as anandamide and 2-arachidonylglycerol ( 2-AG ) bind to and activate signaling via cannabinoid type 1 and type 2 ( CB1 and CB2 ) receptors , which belong to the G-protein coupled receptor ( GPCR ) family . Gene deletion studies have uncovered the tremendous significance of the CB1 receptor in mediating several key biological functions of ( endo ) cannabinoids in the adult central nervous system ( CNS ) , including learning and memory , pain and analgesia , neuronal excitability and seizures , fear acquisition and extinction , and appetite control , amongst several others [1–3] . CB1 receptor is abundantly expressed in inhibitory interneurons at presynaptic terminals , where it is activated by retrogradely acting endocannabinoids synthetized at the postsynaptic site [4 , 5] . CB1 receptor is also found to be expressed at more moderate levels in excitatory neurons ( principle cells ) in the CNS . Cell type-specific CB1 receptor knockout mice have particularly revealed key roles for CB1 receptor expressed in excitatory neurons in appetite , neuroprotection , epilepsy , fear , anxiety , pain , and analgesia , amongst others [6–10] . Recent studies have shown that far from enfolding their functions in the adult nervous system alone , endocannabinoids are important regulators of neuronal development; particularly , the CB1 receptor has been implicated in collapse and navigation of axonal growth cones in inhibitory as well as excitatory neurons [11–13] . Interestingly , the vast knowledge on the biological functions of endocannabinoids as well as exogenously-applied cannabinoids at the cellular and systemic levels [1 , 14] is contrasted sharply by the relative scarcity of studies addressing molecular components of cannabinoid receptor assembly and signaling partners . CB1 receptor primarily couples to Gi but can also couple with Gq/11 or G12/13 in a context-dependent manner , which have primarily represented the focus of previous studies [15–17] . To date , however , very little is known about which other proteins assemble with CB1 receptor in a multiprotein signaling complex to unfold its biological actions . The few interactions that were tested on a candidate protein basis pertain largely to trafficking and chaperone proteins modulating the endocytosis and intracellular trafficking of CB1 [18–22] . There are , to date , no reports on unbiased proteomic screens for CB1 receptor in neural tissue . Indeed , proteomic screens on GPCRs are notoriously challenging because of technical difficulties in solubilizing GPCR protein complexes . This study now reports novel members of the CB1 receptor complex found via a proteomic screen in mouse brain in vivo . We particularly report on the functional significance of CB1 receptor interactions with the Wiskott-Aldrich syndrome protein-family verprolin-homologous protein 1 ( WAVE1 ) /SCAR1 complex . WAVE proteins act as nucleation promoting factor linking upstream signals to the activation of actin related protein 2/3 ( ARP2/3 ) , which in turn activates actin nucleation [23 , 24] . Using a diverse set of models and in vitro as well as in vivo analyses , we demonstrate that CB1-WAVE1 interactions play a key role in dynamically regulating the actin cytoskeleton in developing and adult neurons , which contribute to prominent functions of endocannabinoids in the brain and spinal cord . Importantly , we demonstrate that cannabinoids structurally remodel synaptic spines by regulating the activity levels of WAVE1 and report a novel function for WAVE1 in mediating inflammatory pain via structural and functional plasticity of spinal neurons .
A proteomic strategy combining immunoprecipitation and mass spectrometry ( MS ) was used to identify new potential interacting partners of CB1 receptor . Owing to limited success of currently available antibodies against CB1 receptor in immunoprecipitation experiments , we generated recombinant adeno-associated virions ( rAAV ) expressing enhanced green fluorescent protein ( EGFP ) -tagged CB1 ( rAAV-EGFP-CB1 ) , in which the N-terminus is fused to EGFP , which we have previously shown to bind and respond to cannabinoids in an identical manner as wild-type CB1 receptor [22] . rAAV-EGFP-CB1 virions were stereotactically injected into the frontal and parietal cortex of adult mice ( Fig 1A ) and yielded a strong expression of EGFP-CB1 in cortical neurons , which was mostly localized to the neurophil , consistent with the targeting of CB1 receptor to axonal and dendritic segments [25] . We then proceeded to optimize membrane protein-enriched preparations and test a large number of detergents in varying concentrations and combinations to obtain a good solubilization of EGFP-CB1 receptor from cortical membranes , which is important for successful and specific pull-down of interacting protein assemblies without destroying protein–protein interactions in a functional complex . Western blot analysis on the immunoprecipitate of the solubilized fractions using a GFP-nanotrap antibody [26] revealed a single band around 250 KDa in CB1-EGFP-transfected HEK293 cell lysates following solubilization with Na-cholat ( S2A Fig ) , which represents the multimeric state of the CB1 receptor [27] . These optimized conditions , when applied to rAAV-EGFP-tagged CB1-transduced brain tissue , yielded high molecular weight forms of CB1 receptor when blotted against GFP and CB1 receptor ( Fig 1B , left blot , black arrow ) . Additionally , a band at approximately 95 KDa was also observed in pulldowns from brain tissue ( Fig 1B , left blot , yellow arrow ) , corresponding to the size of monomeric CB1-EGFP ( CB1: 64 KDa and EGFP: 28 KDa ) . Immunoblotting with an anti-CB1 receptor antibody additionally yielded a band just below 70 KDa , indicating immunoprecipitation of the endogenous CB1 receptor monomer ( Fig 1B , middle blot , blue arrowhead ) . We observed GFP protein in the lysates from rAAV-GFP transduced ( control ) brain tissue , but not in the immunoprecipitated material ( Fig 1B right blot ) . GFP-nanotrap-immunoprecipitated materials from solubilized mouse cortex preparations were then analyzed by high-resolution nanoflow liquid chromatography tandem MS ( nano-LC MS/MS ) in 5 independent immunoprecipitation experiments ( a comprehensive list of all proteins identified in the MS analyses as potential CB1 interactors is given in S1 Table ) . Apart from CB1-EGFP and endogenous CB1 receptor , several known interacting partners of CB1 receptor , such as adenylate cyclase [28] , cannabinoid receptor-interacting protein 1 [20] , clathrin coat assembly proteins [29] , and guanine nucleotide-binding protein Gq [17] were coimmunoprecipitated by GFP-nanotrap with high relative peptide query ( rPQ ) in rAAV-CB1-EGFP-mice , but not in rAAV-GFP-expressing control mice ( S1B Fig; loading control S1A Fig; S1 Table ) . Importantly , several members of the Wiskott-Aldrich syndrome protein family verprolin homologous protein 1 ( WAVE1 ) complex and its upstream signaling components were consistently coimmunoprecipitated with CB1 receptor with a high specificity in rAAV-CB1-EGFP-expressing mice , but not in rAAV-GFP-expressing mice . The WAVE1 complex consists of WAVE1 ( also known as SCAR1 ) , Abelson-interacting protein 1/2 ( ABI1/2 ) , NCK-associated protein 1 ( NCKAP1 , also known as NAP1 ) , cytoplasmic FMR1-interacting protein 2 ( CYFIP2 , also known as PIR121 or SRA1 ) and HSPC300 [30] . Under basal conditions , proteins of the WAVE1 complex are positioned to block the VCA domain of WAVE1 , thereby inhibiting WAVE1 activity towards ARP2/3 ( see scheme in Fig 1D , left image ) . This autoinhibition is then released upon binding with the active form of the RhoGTPases , Rac1 , leading to the activation of ARP2/3 by WAVE1 and thereby to actin nucleation [31] ( Fig 1D right image ) . Our analyses revealed that four constituents of this protein complex , namely WAVE1 , ABI1/2 , NCKAP1 , and CYFIP2 , as well as the upstream activator , Rac1 , were coimmunoprecipitated with solubilized CB1 ( Fig 1C ) . We confirmed results from MS experiments by blotting immunoprecipitates with antibodies recognizing diverse components of the WAVE1 complex . Specific bands corresponding to WAVE1 , CYFIP2 , NCKAP1 , and Rac1 were observed in samples coimmunoprecipitated by the GFP nanotrap in rAAV-EGFP-tagged CB1-expressing mice but not in rAAV-GFP-expressing mice ( Fig 1E ) . Using HEK293 cells heterologously transfected with CB1-EGFP or GFP alone for validation of the above results , we observed that the anti-EGFP antibody pulled down endogenous WAVE1 from cells expressing CB1-EGFP , but not from cells expressing GFP alone ( S2A Fig ) . Finally , we additionally validated the interaction in native brain tissue by immunoprecipitating endogenously expressed CB1 from wild-type mouse cortex . We observed that WAVE1 protein was coimmunoprecipitated with endogenous CB1 when lysates of wild-type mouse cortex were employed , but not in lysates derived from mice lacking CB1 in a global manner ( CB1-/- mice ) [32] , thereby establishing in vivo relevance as well as specificity of the interaction ( Fig 1E ) . In primary cortical neurons cultured from E15 mouse embryos infected with rAAV-CB1-EGFP virions and analyzed at 3 d in vitro , a stage at which they are still developing , marked overlap was observed between CB1-EGFP and anti-WAVE1 immunoreactivity in the somata and growth cones of processes ( Fig 1F , note the common domain occupied by CB1-EGFP and WAVE1 at the leading edge within the broad growth cone structure labelled via Phalloidin staining of actin in the inset in Fig 1F ) . Importantly , in nontransfected primary embryonic cortical neurons , WAVE1 also colocalized with native anti-CB1 immunoreactivity , the latter being in axonal growth cones , in which CB1 was markedly found in the developing axons ( S2B Fig ) as reported elsewhere [11] . To further test colocalization of WAVE1 and CB1 , we exploited the large size and elaborate morphology of COS ( CV-1 in Origin with SV40 genes ) cells . In COS7 cells cotransfected with WAVE1 and GFP ( control ) , heterologously-transfected WAVE1 showed a highly specific distribution , which was predominantly nuclear and perinuclear , with a much lesser degree of localization at the cell membrane than the cell interior ( Fig 2A , upper panels ) . In cells cotransfected with CB1-EGFP and WAVE1 , CB1-EGFP colocalized with WAVE1 in the cell membrane as well as in the cytoplasm , the latter being the predominant locus of overlap ( Fig 2A , lower panels ) . However , in CB1-EGFP-transfected COS cells treated for 45 min with the CB1 agonist , arachidonyl-2’-chloroethylamide ( ACEA ) , there was a marked increase in the proportion of WAVE1 localized at the cell membrane ( Fig 2A lower panels; quantitative analyses from several experiments are shown in Fig 2B ) , and there was a significant increase in colocalization of WAVE1 and CB1 ( inset in lower panels of Fig 2A; quantitative summary in Fig 2C ) . In cells cotransfected with WAVE1 and GFP ( control ) , no significant change was observed in the localization of WAVE1 at the cell membrane in ACEA-treated and vehicle-treated groups ( Fig 2A upper panels; Fig 2B ) . This suggests that activated CB1 and WAVE1 colocalize in the cell membrane and that activation of CB1 redistributes WAVE1 towards the cell membrane . It is well-recognized that Rho-family GTPases , to which Rac1 belongs , are subject to tight spatiotemporal regulation , and that multiple subcellular pools of any given Rho GTPase can operate simultaneously yet independently of each other in both time and space [33] . Therefore , instead of globally pulling down Rac1 from cell lysates , we spatiotemporally imaged Rac1 activity in growth cones of cortical neurons nucleofected with a biosensor ( Raichu-Rac1 [34] ) , a construct consisting of YFP , a ligand domain ( PAK CRIB ) , a flexible linker peptide , a sensor domain ( Rac1 ) and CFP ( Fig 3A , typical examples of fluorescence resonance energy transfer [FRET] expression in neurons shown in S3A Fig ) . Binding of Rac1 to GTP leads to FRET of excitation energy from CFP to YFP , i . e . , the active state of Rac1 corresponds to an increase in YFP to CFP fluorescence ratio ( FRET ratio; Fig 3A , right image ) . Although bleaching leads to a small decrease in total fluorescence intensity over time , the FRET ratio stays constant [34] , as observed upon treatment of neurons with vehicle over 45 min ( typical examples in Fig 3B and quantitative summary from ten neurons from five independent cultures is shown in Fig 3C ) . Nerve growth factor ( NGF ) , which activates Rac1 in growth cones [35] , served as a positive control , and was observed to evoke a significant increase in FRET ratio over 1 h treatment , as compared to vehicle ( DMSO , end-dilution 1:30 , 000 in neurobasal medium; Fig 3C and S3B Fig ) . Using this model system , we observed that the CB1 receptor agonist ( ACEA ) or inverse agonist ( AM251 ) bidirectionally altered the activity of Rac1 in growth cones of cortical neurons ( typical examples in Fig 3B and quantitative summary from 10–16 neurons from five independent cultures at 0 , 15 , 30 , and 45 min post-treatment is shown in Fig 3C; detailed time course over 45 min is shown in S3C Fig ) . Upon application of the agonist ACEA , growth cones showed a gradual decrease in FRET signals , reaching significant at 30 min and further progressively decreasing in magnitude until 45 min ( Fig 3C and S3C Fig ) . In contrast , the CB1 receptor inverse agonist , AM251 , induced a rapid and marked increase in the FRET ratio within a few minutes of treatment ( see time course in S3C Fig ) , which stayed constant over 45 min of analysis ( Fig 3C ) . Importantly , neither ACEA nor AM251 induced any significant changes in FRET ratio in growth cones of cortical neurons cultured from CB1-/- mice ( Fig 3D ) , although NGF-induced increase in FRET ratio was preserved in CB1-/- neurons ( Fig 3D and S3D Fig ) , thereby validating that the changes in Rac1 activity observed in wild-type neurons are specific and receptor-dependent . No changes were observed in FRET ratios over somata upon treatment with ACEA , AM-251 , or NGF as compared to vehicle ( S3D Fig ) , consistent with lack of morphological changes in the somata in response to these agents . Given that Rac1 activity is critically linked to disinhibition of WAVE1 signaling onto the Arp2/3 complex , these results predicted that agonism and inverse agonism at CB1 receptor bidirectionally modulate WAVE1 activity . We then studied the extent of serine phosphorylation in WAVE1 as a surrogate parameter for the activation status of WAVE1 . WAVE1 can be phosphorylated on three serine sites , namely S310 , S397 , and S441 , corresponding to an inhibited state and the dephosphorylated state has been linked to WAVE1 activation [36] . In developing primary cortical neurons , WAVE1 showed basal levels of phosphorylation on Ser397 ( pS397-WAVE1 ) and treatment with a CB1 receptor-specific agonist ACEA ( 100 nM ) significantly enhanced the proportion of pS397-WAVE1 over basal state as well as in comparison to vehicle . Representative example of Western blot analysis of pS397-WAVE1 and β-tubulin ( loading control ) are shown in Fig 3E , and the corresponding western blot analysis of total WAVE1 expression is shown in S3E Fig . Quantitative summary from at least four independent experiments of pS397-WAVE1 as a function of β-tubulin levels or total WAVE1 levels is shown in Fig 3G and S3E Fig , respectively . Conversely , treatment with CB1 receptor inverse agonist AM251 ( 600 nM ) significantly decreased the proportion of pS397-WAVE1 over basal as compared to vehicle treatment ( Fig 3G and S3E Fig ) . Neither ACEA nor AM251 affected pS397-WAVE1 levels in cortical neurons cultured from CB1-/- mice ( Fig 3F and Fig 3G ) or in wild-type neurons pretreated with pertussis toxin ( PTX ) ( 100 ng/ml ) , a specific inhibitor of Gi signaling ( Fig 3F and Fig 3G ) . These results thus show that a bidirectional change in CB1 receptor activity leads to a corresponding , Gi-dependent bidirectional alteration in the serine phosphorylation status of WAVE1 . We employed LifeAct-GFP , which constitutes the first 17 amino acids of Abp140 protein tagged with GFP and preferentially binds to filamentous actin ( F-actin ) rather than actin monomers ( G-actin ) and thereby permits a direct visualization of F-actin [37] . In LifeAct-GFP-transfected developing cortical neurons ( examples shown in Fig 4A ) , we observed that intensity of labeled F-actin decreased gradually upon exposure to ACEA over basal ( preapplication ) values , reaching significance over basal values at 60 min following ACEA application ( Fig 4B ) . In contrast , vehicle-treated neurons demonstrated comparable intensity over 60 min , indicating that ACEA-induced decrease in LifeAct-GFP intensity stems from F-actin destabilization rather than from bleaching . Moreover , the area covered by F-actin shrunk significantly upon ACEA treatment , in contrast to vehicle ( Fig 4C ) . Conversely , AM251 treatment led to an expansion of the area covered by F-actin within 30 min postapplication and enhanced the intensity of F-actin ( Fig 4B and Fig 4C ) , thus indicating enhanced stability of F-actin and actin polymerization . In contrast to wild-type neurons ( above ) , ACEA and AM251 did not affect actin dynamics in developing cortical neurons cultured from CB1-/- mice ( Fig 4D and Fig 4E ) . Furthermore , we observed that both modes of actin modulation by cannabinoids are lost upon down-regulation of WAVE1 expression . In cultured cortical neurons nucleofected with WAVE1-specific siRNA ( Fig 4J and Fig 4K; 59 . 6 ± 12 . 5% knockdown as compared to scrambled siRNA ) , we observed that neither ACEA nor AM251 affected actin dynamics in contrast to neurons transfected with scrambled siRNA ( Fig 4F to Fig 4I ) . Because siRNAs can have significant off-target effects , we implemented an independent siRNA#2; targeting WAVE1 at an independent locus as compared to the siRNA#1 employed above would be important for ruling out off-target effects of the formerly-used WAVE1 siRNA ( Materials and Methods ) . We verified the degree of WAVE1 knockdown with siRNA via western blot analysis ( example in S4A Fig and quantitative summary in S4B Fig ) and , in Life-Act-GFP assays , observed exactly the same phenotypic effects as siRNA#1 ( S4C Fig ) , establishing that these are indeed caused by WAVE1 down-regulation . Thus , WAVE1 links CB1 receptor with modulation of actin dynamics in developing neurons . Activation of CB1 receptor has been reported to induce growth cone collapse in developing GABAergic neurons [12] as well as in excitatory neurons of the cortex [11] . We treated developing cultured cortical neurons with cannabinoids or vehicle for 1 hr and identified axonal growth cones via staining for Tau 1 ( axonal marker ) and Phalloidin ( labels actin ) , as well as phosphorylated growth-associated protein 43 ( pGAP43 ) , which is expressed in stable and expanding growth cones , but not collapsing or retracting growth cone [38] . As described previously , axonal growth cones could be categorized into three different morphologies ( illustrated by examples shown in insets in Fig 5A ) : 1 ) normal or stable growth cone , characterized by an expanding cone larger than its proximal area and existence of pGAP43 staining ( seen in about 66 . 9 ± 3 . 4% of vehicle-treated neurons ) , 2 ) collapsed growth cone , characterized by a thin-shaped cone and lack of pGAP43 staining ( seen in about 7 . 1 ± 1 . 7% of vehicle-treated neurons ) , and 3 ) a large flat growth cone , characterized by the enlarged filopodia-rich rounded shape of the cone and strong pGAP43 staining ( seen in about 26 ± 2 . 7% of vehicle-treated neurons ) ( Fig 5B ) . We observed that a significantly higher number of ACEA-treated neurons showed collapsed morphology as compared to vehicle-treated neurons ( Fig 5B ) . In contrast , inverse agonism at CB1 receptor led to a further significant decrease in the proportion of neurons showing collapsed growth cones ( Fig 5B ) . Conversely , the incidence of large flat growth cones was significantly higher in AM251-treated neurons and significantly lower in ACEA-treated neurons as compared to vehicle ( Fig 5B ) . Modulation of growth cone morphology by ACEA and AM251 was not observed in cortical neurons derived from CB1-/- mice ( Fig 5B ) and was also inhibited in wild-type neurons pretreated with a Gi inhibitor , PTX ( 100 ng/ml ) or a Rac1 inhibitor , CAS 1090893-12-1 ( 50 μM ) as compared to the respective vehicles ( Fig 5C ) . Importantly , neurons with markedly reduced expression of WAVE1 via siRNA-mediated knockdown did not show significant effects on growth cone morphology in contrast to neurons expressing scrambled ( control ) siRNA ( Fig 5D ) . Thus , taken together , these results reveal causal relationships between CB1 receptor signaling , the WAVE1 complex , actin rearrangement and structural modulation in developing neurons . To further study the impact of CB1-WAVE1 interactions on neuronal morphology , we employed cortical neurons matured over 4 wk in vitro . Although CB1 receptors are believed to be mainly expressed presynaptically , in axonal varicosities , several studies have previously reported that CB1 can be additionally localized in dendrites of neurons [39–41] . To ascertain the expression pattern of CB1 in mature primary neurons , we studied the distribution of endogenously expressed CB1 ( Fig 6A ) using a previously well-characterized antibody ( [10]; lack of staining in CB1-/- mice is shown as a negative control in Fig 6A , lower panel ) . In addition to the widespread distribution of CB1 in axonal compartments ( negative for the dendritic marker protein , MAP2 ) , these neurons showed a marked distribution of CB1 in MAP2-positive dendrites . Both dendrites and axons showed hot spots of CB1 expression , reminiscent of varicosities and dendritic spines ( see overlay in Fig 6A , right column and Fig 6B ) . Indeed , when we infected mature cortical neurons with rAAV- Ca2+/calmodulin-dependent protein kinase II ( CamKII ) -GFP virions , which express GFP in dendritic compartments , including the shafts and spines , we observed a marked colocalization with endogenous CB1 in dendritic shafts and in several , but not all , dendritic spines ( arrowheads in examples shown in Fig 6C ) . To morphologically pinpoint the localization of endogenous CB1 in synaptic spines , we then costained for the classical postsynaptic density protein , PSD-95 , and observed a colocalization in a large number , but not all , PSD-95-positive spines ( arrowheads in Fig 6C ) . Similar results were obtained with exogenously-transfected CB1-EGFP ( Fig 6D ) , which showed a marked colocalization with MAP2-positive dendrites in addition to being targeted to thin , MAP2-negative axons ( Fig 6D ) . Finally , we tested primary spinal cord neurons that were matured over 4 wk in culture and observed identical results—endogenous CB1 was markedly localized to MAP2-positive dendrites ( arrows in Fig 6E ) in addition to being expressed in MAP2-negative axons ( arrowheads in Fig 6E ) . To address whether cannabinoids also regulate actin dynamics and thereby bring about structural remodeling in adult neurons via the WAVE1 complex , we then studied dendritic spines in wild-type cortical neurons matured over 3 wk in vitro and nucleofected with LifeAct-mCherry , thereby labelling F-actin in dendritic spines . To address dynamic changes in actin polymerization , we performed photobleaching of F-actin-bound LifeAct-mCherry in identified spines by focusing a light beam over individual spines and thereafter recorded the rate of fluorescence recovery after photobleaching ( FRAP ) [42] ( image examples are shown in Fig 7A ) . Since LifeAct is known for its relatively low binding affinity for F-actin , this method may introduce indirect measures of F-actin turnover . We therefore restricted our quantitative analyses to the total recovery of the fluorescence mobile fraction of LifeAct-mCherry , which is indicative of the newly formed F-actin itself , at steady-state 150 s postphotobleaching . Quantitative analyses from five independent experiments revealed that the total recovery of the fluorescence mobile fraction of LifeAct-mCherry postphotobleaching was significantly lower in ACEA-treated neurons as compared to vehicle-treated neurons ( Fig 7B and Fig 7E ) . To test the relevance of WAVE1 for cannabinoid-induced modulation of F-actin assembly , we had to establish a method for knocking down WAVE1 in cortical cultures that had been maintained for several weeks in culture to develop synaptic spines ( see Materials and Methods , example and quantitative summary of knockdown as compared to scrambled control shown in Fig 7D ) . FRET measurements on spines were performed 3 d after siRNA transfection , and WAVE1 knockdown but not scrambled siRNA blocked ACEA-induced remodelling of the actin cytoskeleton in dendritic spines ( Fig 7C , Fig 7E and S5A Fig ) . Similarly ACEA failed to affect F-actin assembly in individual spines of primary cortical neurons derived from CB1-/- mice ( Fig 7E and S5B Fig ) . Thus , ACEA-modulated actin assembly in single postsynaptic spines of mature cortical neurons in a CB1 and WAVE1-dependent manner . Because dendritic spines undergo considerable turnover and actin polymerization in spines is required for spine stability , formation , and development [43] , the above results would predict that over time , activation of CB1 receptor results in a net reduction in spine density . To test this prediction , we transduced cultured cortical neurons with rAAV-expressing GFP under promoter elements of the mouse CaMKII gene ( rAAV-CaMKII-GFP ) , thereby enabling visualization of dendritic spines on excitatory neurons 3–4 wk following culture and rAAV delivery ( examples in Fig 7E ) . Neurons were treated with ACEA ( 100 nM ) or vehicle ( DMSO , 1:30 , 000 in phosphate-buffered saline [PBS] ) for 24 h , and GFP-labelled dendritic spines on secondary and tertiary dendrites were imaged via confocal microscopy and morphologically categorized into “thin”- , “stubby”- , and “mushroom”-shaped spines as described previously [44] . We observed a net , significant reduction in dendritic spine density in neurons treated with ACEA as compared to vehicle ( Fig 7G ) . Interestingly , morphological classification indicated that this change came about only with mushroom-shaped spines ( Fig 7G ) , which are considered to represent stable and mature spines [44] . These results show that activation of CB1 dynamically modulates F-actin assembly in dendritic spines in mature cortical neurons and suggests that the net content of mature spines in excitatory neurons decreases over prolonged exposure to cannabinoidergic agonists . With a view towards testing the significance of these processes at the level of the whole organism , we sought a model system that would enable linking structural changes in spines in vivo with a functional change ( e . g . , in behavior ) and also enable addressing the contribution of WAVE1 thereof using molecular interventions . We chose neurons located in the spinal dorsal horn in adult mice , owing not only to their involvement in the analgesic effects of cannabinoids but also because they undergo significant synaptic structural remodeling in inflammatory pain [45] . We employed a model of inflammatory pain based upon unilateral injection of complete Freund’s adjuvant ( CFA ) to the plantar surface of mouse hind paw , in which hypersensitivity to mechanical stimulation of the hind paw with von Frey filaments develops within 24 h of CFA injection ( see leftward shift from basal values in the stimulus intensity-response frequency curves shown in black color Fig 8A ) . Consistent with previous reports of analgesic actions of spinally-applied cannabinoids [6] , we observed that intrathecal delivery of ACEA to the lumbar spinal segments over the 24 h period post-CFA injection ( three doses of 2 pmol ACEA given every 12 h ) significantly attenuated inflammatory hyperalgesia ( red-colored stimulus-response curves are shown in Fig 8A ) . In ensuing morphological analyses of the spinal cords of mice analyzed behaviorally in the above experiment , we observed that the above-described functional changes were well-aligned with corresponding alterations in synaptic spine density on spinal neurons . Indeed , the development of inflammatory mechanical hypersensitivity in the CFA model is known to be associated with enhanced density of dendritic spines on neurons of the spinal laminae II and V [45] , which play a key role in spinal nociceptive processing [46] . Likewise , in Golgi-staining experiments , we observed that CFA-injected mice treated intrathecally with vehicle showed increased spine density on the secondary or tertiary dendrites of spinal neurons in laminae II and V; however , spine remodeling did not occur in spinal neurons of mice treated over 24 h with ACEA ( typical examples in Fig 8B and quantitative summary from 12–16 neurons from at least four mice shown in Fig 8C; these mice represent the same cohort of mice which were analyzed in behavioral studies shown in Fig 8A ) . Thus , spinal CB1 receptor activation suppressed nociceptive activity-induced dendritic spine remodeling in the spinal cord and inflammatory hypersensitivity in the same set of mice in vivo . This gives rise to the question whether structural modulation by cannabinoids involves the WAVE1 complex . WAVE1 is ubiquitously expressed from E9 , and the expression is restricted to CNS ( including spinal cord ) , starting E15 until adulthood [47] . There are no reports on the potential contributions of WAVE1 to nociceptive modulation . Similar to our analyses on cortical neurons ( Fig 3E to Fig 3G ) , we observed that WAVE1 shows serine phosphorylation in the basal state in the lumbar spinal cord segments involved in nociceptive processing in adult mice . Representative example of western blot analysis of pS397-WAVE1 and β-tubulin ( loading control ) are shown in Fig 8D and the corresponding western blot analysis of total WAVE1 expression is shown in S6A Fig . Quantitative summary for pS397-WAVE1 levels expressed as a function of either βIII-tubulin levels or total WAVE1 levels is shown in Fig 8E and S6B Fig , respectively . Our results indicate that activity of the WAVE1 complex in the spinal cord is limited under basal conditions . Interestingly , at 24 h following intraplantar hind paw injection of CFA in mice , the relative proportion of pSer397 WAVE1 decreased significantly , indicating an increase in WAVE1 activity over the period of initiation and expression of inflammatory hypersensitivity ( Fig 8D , Fig 8E and S6B Fig ) . In mice receiving intrathecal ACEA over 24 h , CFA-induced decrease in pSer397-WAVE1 still occurred , but to a significantly lower magnitude than CFA-injected mice receiving intrathecal vehicle ( Fig 8D , Fig 8E and S6B Fig ) . Thus , in parallel to decreasing nociceptive activity-induced spine remodeling in the spinal dorsal horn , ACEA-attenuated nociceptive activity-induced spinal activation of WAVE1 . To directly test a potential causal relationship between spine remodeling and WAVE1 , we performed similar experiments employing the CFA model in mice in which the spinal expression of WAVE1 was down-regulated by RNA interference in vivo . Inducing a partial knockdown of WAVE1 to about 70% of basal value ( Fig 9A and Fig 9B ) was preferable in the present experiments because it is known from previous studies that total knockout or a more substantial knockdown of WAVE1 decreases spine density in CNS neurons in conditions of basal activity [48] . Indeed , we observed that under basal conditions , the density of dendritic spines on laminae II/V spinal neurons was comparable in mice injected intrathecally with WAVE1-siRNA and scrambled-siRNA ( Fig 9C ) . Consistent with the above , WAVE1-siRNA and scrambled-siRNA-injected mice showed comparable basal sensitivity to mechanical hind paw stimulation prior to induction of inflammation ( stimulus intensity-response frequency curves represented by dashed lines in Fig 9D ) . Importantly , unlike scrambled-siRNA-injected mice , WAVE1-siRNA-injected mice did not show enhancement of spine density at 24 h after intraplantar hind paw CFA injection ( Fig 9C ) . In the same group of mice , the development of inflammatory hypersensitivity ( represented by solid lines in Fig 9D ) was partially , but significantly , attenuated in WAVE1-siRNA-injected mice as compared to scrambled-siRNA-injected mice ( Fig 9D ) . Thus , a partial knockdown of WAVE1 expression in the lumbar spinal cord attenuated inflammatory hypersensitivity and abrogated activity-dependent spine remodeling in spinal dorsal horn neurons . Finally , we asked whether modulation of WAVE1 mechanistically contributes to cannabinoidergic analgesia in the inflammatory pain setting . In scrambled siRNA-expressing mice , intrathecal administration of ACEA over 24 h following injection of CFA in the hind paw decreased the development of inflammatory hypersensitivity as compared to vehicle-treated mice ( see solid lines in Fig 9E , upper graph ) in a manner similar to our results from naïve mice described in Fig 8A above . However , in WAVE1 siRNA-expressing mice , ACEA did not significantly attenuate inflammatory hypersensitivity as compared to vehicle administration ( see solid lines Fig 9E , lower graph ) . Dashed lines in both graphs in Fig 9E represent basal mechanical sensitivity in each of the respective groups prior to the induction of hind paw inflammation via CFA injection . In line with these behavioral observations , the above-described scrambled siRNA-injected mice demonstrated a decrease in CFA-induced spine remodeling in spinal neurons upon ACEA administration in comparison with vehicle ( Fig 9F ) , similar to results obtained in naïve mice ( Fig 8C ) . In contrast , loss of analgesic action of ACEA in mice with spinal WAVE1 knockdown was accompanied by failure of ACEA to attenuate CFA-induced spine remodeling in WAVE-siRNA-injected mice ( Fig 9F ) . It is to be noted , however , that a loss of CFA-induced spine remodeling , which is already inherent to the WAVE1-siRNA-injected group of mice , may have led to a ceiling effect . Taken together , these in vivo experiments indicate a requirement for spinal WAVE1 for the analgesic actions of spinally-delivered cannabinoids in inflammatory pain and reveal a close link between cannabinoidergic modulation of structural plasticity in spines and cannabinoid-induced analgesia .
This study uncovers new signalling mechanisms downstream of the CB1 receptor activation by identifying four components of the WAVE1 complex and Rac1 as physical interactors of the CB1 receptor . We elucidated the functional significance of this protein assembly in two key model systems , which have been critically associated with biological functions of the WAVE1 complex , namely growth cones of developing neurons and synaptic spine turnover in mature neurons . Our results indicate that a physical association between Rac1-WAVE1 and CB1 receptor provides a direct molecular link between the CB1 receptor and actin rearrangement . This newly identified interaction gives mechanistic insights not only into known functions of cannabinoid agonists in the structural modulation of developing neurons but also provides evidence for a role for cannabinoids in modulation of actin assembly in dendritic spines on mature neurons of the CNS . This function is evident in diverse CB1 receptor-expressing neuron populations as demonstrated in developing as well as mature cortical neurons ex vivo and in spinal dorsal horn neurons in vivo . The importance at the whole organismal level is indicated by our finding that cannabinoid agonists suppress nociceptive activity-induced structural plasticity of synaptic spines in conjunction with their analgesic actions in inflammatory pain in a WAVE1-dependent manner . The CB1 receptor complex has remained elusive , although a few interacting proteins have been described based upon candidate approach [18 , 49] . This study reports an open , unbiased proteomic screen for CB1 receptor-interacting proteins in the mouse brain . Owing to their multiple transmembrane domains and structural peculiarities , GPCRs are notoriously problematic in immunoprecipitation analyses [50] . Our strategy of stereotactically expressing EGFP-tagged CB1 receptor via rAAV vectors coupled with GFP-Nanotrap on cortical membranes and LC-MS analyses enabled overcoming technical hurdles; given the propensity of GPCRs , including CB1 receptor , to heteromerize , our strategy pulled down CB1 receptor heteromers as well as several known protein interactors of CB1 receptor , such as the CB1 receptor-interacting protein CRIP1a [20] , thereby validating the strategy and finding novel members of the CB1 receptor signaling complex . Given our observation that the CB1 receptor also interacts with Rac1 and modulates its activity , cannabinoidergic modulation of WAVE1 activity may occur downstream of Rac1 since the WAVE1 complex is one of the most important effectors of Rac1 in bringing about actin remodelling [30] . We also additionally observed a bidirectional change in serine phosphorylation status of WAVE1 upon agonism and inverse agonism at the CB1 receptor in neurons , which , though modest in magnitude , was consistent and specifically mediated by CB1-Gi signalling . The WAVE1 complex has been suggested to act as a convergence point for integrating signals from diverse pathways that link extracellular signals as well as cell-autonomous programs to the modulation of the actin cytoskeleton [48] . cAMP signalling can affect WAVE1 activity via cyclin-dependent kinase 5 ( CDK5 ) -mediated serine phosphorylation [36] . However , whether the cAMP-CDK5 pathway is interlinked with Rac1-mediated WAVE1 regulation or whether these run in parallel and represent independent modes of regulation of WAVE1 activity is not understood . Moreover , a recent study suggests that WAVE1 carries a highly conserved binding peptide motif , the WAVE regulatory complex ( WRC ) -interacting receptor sequence , which can facilitate interactions with diverse types of receptors , including some GPCRs [51] . However , we found that this motif is not conserved in the CB1 receptor . Given our observation that apart from WAVE1 , three of the four remaining components of the WAVE1 signalling complex were coimmunoprecipitated with the CB1 receptor , it is plausible that any of them interact directly with the CB1 receptor and thereby mediate cannabinoidergic modulation of WAVE1 indirectly . The CB1 receptor may thus regulate WAVE1 serine phosphorylation as well as Rac1-mediated WAVE1 activation via steric modulation . Alternatively , CB1 receptor interactions with members of the WAVE1 complex may position WAVE1 optimally in its vicinity to facilitate regulation of WAVE1 activity by signalling mediators downstream of the activated G-protein . In this regard , it is interesting to note that cannabinoidergic modulation of WAVE1 activity , as well as downstream cellular effects thereof , were sensitive to Gi inhibition . Importantly , our results on the close colocalization between WAVE1 and CB1 in COS7 cells and in neurons , and importantly showing redistribution of WAVE1 to the cell membrane in cells treated with a CB1 agonist point to a key physical and functional link between CB1 and WAVE1 . This was further strengthened by our observations that all of the effects of cannabinoidergic modulation on the actin cytoskeleton as well as morphology of neurons in vitro and in vivo as well as inflammatory pain behaviour were disrupted upon manipulating WAVE1 expression . Another interesting observation was that agonists and inverse agonists acting via CB1 receptor bidirectionally altered the basal activation status of Rac1 and WAVE1 and , consequently , bidirectionally modulated actin nucleation , thereby implicating a considerable basal tone of activity of the CB1 receptor . It cannot be ruled out that this occurs in a ligand-dependent manner , because despite the absence of serum and supplements during assays , traces of ligands may persist in the medium . On the other hand , there is rising evidence for ligand-independent constitutive activity of a number of GPCRs , including CB1 receptor [40 , 52] . The constitutively active "on" state of CB1 receptor can be shifted reversibly into constitutively inactive "off" state through conformational changes of the receptor , and thus bidirectionally modulate Rac1 and components of the WAVE1 complex , leading to changes in WAVE1 activity . In this two-state model , the existence of ligands , either agonist or inverse-agonist , would shift the proportion of the receptors to the active or inactive conformation [52 , 53] . Irrespective of which of the above scenarios comes into play , the observation that tuning CB1 receptor activity up or down profoundly influences the Rac1-WAVE1-actin nucleation axis places the CB1 receptor in a key position to sense diverse endocannabinoids and lipid modulators of CB1 receptor and bring about localized regulation of actin dynamics . How actin dynamics are affected by cannabinoids in developing neurons is not mechanistically well understood . Here , direct visualization of F-actin in growth cones revealed that cannabinoid agonists destabilize F-actin in growth cones , whereas deactivation of CB1 receptor activity by inverse agonists leads to enhanced actin nucleation in growth cones . This cannabinoidergic modulation of actin dynamics was entirely in line with the changes in morphology of axonal growth cones brought about by CB1 receptor agonists and inverse agonists in developing cortical neurons . Interestingly , a very recent study has reported a role for endocannabinoid in stabilizing microtubules via degradation of SCG10/stathmin-2 protein in developing neurons [54] , which is complementary to our results on actin remodeling . Taken together , these findings indicate a prominent role of the CB1 receptor in cytoskeletal rearrangement . Most analyses of RhoGTPase modulation are carried out via pull downs on cell lysates thereby representing net RhoGTPase activity over diverse cells as well as diverse compartments of the cell . Imaging studies , however , have revealed that within a cell , RhoGTPase activity varies substantially in a highly compartmentalized manner , e . g . , between the leading edge and the cell interior . Here , direct visualization of localized Rac1 activity within the growth cone revealed a decrease in Rac1 activity within minutes of CB1 receptor activation , whereas total Rac1 in somata did not change considerably . This , taken together with the observation that inverse agonism at the CB1 receptor conversely increased Rac1 activity , suggests that a basal tone of CB1 receptor signaling controls basal Rac1 via protein–protein interactions in growth cones of cortical neurons , which can be dynamically tuned to ( endo ) cannabinoid content in the environment . Given that RhoA and Rac1 exert opposing effects on actin polymerization , by activating RhoA [12] and deactivating Rac1 ( present study ) , agonists at CB1 receptor would be optimally positioned to bring about acute , localized disassembly of F-actin , leading to collapse of growth cones . Moreover , the two RhoGTPases likely act in differential subcellular domains , with RhoA highly functional in the periphery of lamellipodial segments of the growth cone , and Rac1-WAVE1 being expressed throughout the growth cone bodies , especially in the tips and shafts of filopodia [55 , 56] , which is consistent with our observations of RaichuRac activity in FRET imaging experiments , thereby leading to concerted filopodial and lamellipodial retraction upon CB1 receptor activation . Conversely , our results showed that CB1 receptor inverse agonists stabilize the F-actin assembly in growth cones concurrently to Rac1 activation . This suggests that an acute decrease in the local tone of cannabinoidergic signaling would enable growth cones to temporarily stabilize and enlarge , a step which is important not only in axonal navigation , but also facilitates branching in axons [57] . Importantly , the data indicate that cannabinoidergic modulation of actin remodelling via WAVE1 occurs in developing as well as mature neurons and spans distinct systems , such as the cortex and the spinal cord . Furthermore , we found strong evidence for localization of CB1 in axonal as well as dendritic compartments in mature neurons , including in postsynaptic spines . The highly restricted nature of our manipulation of actin assembly in individual spines in FRAP experiments , which was rapidly altered upon cannabinoidergic modulation , coupled with the evidence for CB1 localization in dendritic spines makes it likely that the functional effects evoked by cannabinoids in this study indeed result from CB1 activation in dendritic spines . However , we cannot rule out that previously described presynaptic functions of CB1 , such as modulation of neurotransmitter release from a presynaptic locus , play a role in the effects observed here . Our observation on cannabinoidergic modulation of dendritic spines is particularly interesting given that these sites of synaptic transfer are highly motile structures that govern key functions and act as important sculptors of plasticity in the CNS . Here , direct visualization of actin assembly via FRAP experiments in adult cortical neurons revealed that CB1 receptor activation significantly limits the conversion of G-actin to F-actin , leading over time to a depletion of mature spines with an elaborate mushroom morphology , which are believed to mediate synaptic stability and increased synaptic efficacy such as long-term potentiation ( LTP ) [44] . Moreover , these results also expand our understanding of the therapeutic effects of cannabinoids and clarify mechanistic contributions . Here , we utilized spinally-mediated analgesic functions of CB1 receptor agonists as another model system to test the impact of CB1-WAVE1 interactions on structural and functional plasticity at the level of the whole organism . We observed that intrathecal delivery of the CB1 receptor-specific agonist ACEA countered inflammatory mechanical hypersensitivity when given prior to the inflammatory insult , i . e . , before the onset of pathological nociceptive activity . That this functional modulation is associated with structural remodeling was revealed by the observation that parallel to attenuation of inflammatory hypersensitivity behavior , ACEA also decreased nociceptive activity-induced spine remodeling in spinal neurons , a structural correlate for spinal potentiation and central sensitization . Interestingly , the results of this study also describe a novel function for WAVE1 in the modulation of pain , particularly in mechanisms of nociceptive sensitization . Since spinal dendritic spine remodeling has been suggested to underlie diverse pathological pain states , including neuropathic pain associated with nerve lesions , diabetes , and HIV neuropathy [58] , it will be interesting to address in future studies whether WAVE1 mechanistically contributes to underlying processes and whether spinal cannabinoids may constitute an effective therapy . Despite its breadth spanning diverse ex vivo and in vivo analyses and model systems , this study has several limitations . First , given that the WAVE1 complex is large and has numerous protein components that coassemble with the CB1 receptor , the precise nature , order , and molecular structure of the interactions has not been worked out and will constitute a large study in their own merit . Furthermore , it is important to note that the expression of the CB1 receptor is not restricted to the excitatory neurons , which were mainly addressed in this study , but is highly pronounced in axons of GABAergic neurons [59] . It remains to be addressed whether CB1-WAVE1 interactions play a role in cannabinoidergic modulation of interneurons . Finally , WAVE1 may have additional downstream targets apart from actin nucleation , such as kinesin-1 , CRMP-2 and profiling [30 , 60] , and it remains to be determined whether and how these are related to cannabinoidergic mechanisms . In summary , our findings identify cannabinoids as key upstream regulators of the WAVE1 complex via physical interactions within the CB1 receptor assembly , which play a fundamental role in actin dynamics and structural modulation in growth cones during development , as well as in activity-induced plasticity of dendritic spines in adult life . This study uncovers novel mechanisms for developmental functions as well as therapeutic applications of cannabinoids and implicates the WAVE1 complex as a mediator of plasticity processes leading up to inflammatory pain .
All animal experiments were performed in accordance with the EU guidelines 2010/63/EU and the German TierSchG and TierSchVersV . All animal experimental protocols were approved by the local governing body Regierungspräsidium Karlsruhe ( license numbers T43/13 , T49/14 , G86/11 and G192/14 ) and were performed in accordance with their ethical guidelines . All in vivo experiments were done in age-matched ( 8 wk ) male C57Bl6 mice . Conventional CB1 global knockout mice ( CB1-/- ) have been described previously [32] . A rAAV vector with an AAV1/AAV2 chimeric backbone was used to express GFP or CB1 tagged with EGFP on its N-terminal end [22] . Viral particles were produced using AAV-293 HEK cell line ( Stratagene ) , a derivation of HEK-293 that produces higher viral titres . The cells were transfected with equimolar amounts of pAM-EGFP-tagged CB1 or pAM-GFP plasmid , pDP1rs helper plasmid and pDP2rs helper plasmid ( PlasmidFactory ) that were previously mixed in a transfection buffer ( 140 nM NaCl , 25 mM HEPES , 750 μM Na2HPO4 , and 125 mM CaCl2 , pH 7 . 05 ) . Three days following transfection , the cells were lysed with a lysis buffer ( 150 mM NaCl , 50mM Tris-HCL pH 8 . 5 ) to harvest the viral particles . To purify the viral particles , the attained lysate were loaded into a pre-equilibrated heparin agarose column , incubated for 1–2 h and washed at least four times with equilibration buffer ( 1x PBS , 1 mM MgCl2 , 2 . 5 mM KCl , pH 7 . 2 ) . The viral particles were then eluted out of the column with elution buffer ( equilibration buffer with 0 . 5 NaCl and pH 7 . 2 ) and then washed and centrifuged in Amicon-ULTRA filter to get rid of the salt and concentrate the particles . Following harvest , the viral titre was determined , and brain injection of virions ( 2–5 * 109 particles/ml , 500 nl ) in C57Bl6 adult mice were performed stereotactically as previously described [22] . Briefly , C57Bl6 adult mouse at 8–10 wk of age were anesthetized with fentanyl/medetomidine/midazolam ( 4:6:16; 0 . 7 μl/g , i . p . ) , and after the surgical anesthesia is reached ( lack of response to noxious stimuli ) , the fur on the head was shaved and cleaned with 70% ethanol . The head of the animal was then fixed on the Stereotaxic Alignment System ( Kopf instrument , model 1900 ) , leveled , and incised carefully along the midline with adequate local application of lidocaine during incision . Ten coordinate points were determined as such that these were distributed evenly across the brain with five points at each hemisphere by using the bregma as point [0 , 0] . Afterwards , craniotomy was performed carefully with handheld drill on determined points . Following this , injection of rAAV ( injection depth 500 μm; injection volume 500 nl of the rAAV stock , diluted in PBS; 2–5 * 109 particles/ml ) into the exposed cortex was performed very slowly using a glass micropipette that was fixed into the holder of the stereotaxic arm ( injection depth 500 μm ) . After injection , we waited another 2–3 min before withdrawing the glass micropipette carefully to avoid backflow of the virus into the surface . After all injections were done , the skull was cleaned with sterile PBS and sprayed once again with lidocaine solution , and the skin was sutured . Afterwards , the animal was let to recover , and we waited at least 4 wk to make sure that the recombinant in vivo expression from the AAV vector was sufficient before the mouse was used for further experiments . Cortex tissue or cultured cells were homogenized with bench homogenisator in ice-cold hypotonic buffer ( 20 mM Hepes pH 7 . 5 , 5 mM EDTA , 10 mM DTT , and 1x protease inhibitor cocktail ( Roche ) ) . After a brief centrifugation at 800 G 4°C , the supernatant of the homogenate was further subjected to ultracentrifugation at 125 . 000 G 4°C for 45 min to concentrate membrane-bound proteins in the pellet . The pellet from this ultracentrifugation was then solubilized for at least 2 h , 4°C in solubilization buffer ( 1% ( w/v ) Na-cholate , 5 mM EDTA , 100 mM NaCl , 20 mM Hepes pH 7 . 5 , 10 mM DTT and 1x protease inhibitor cocktail ( Roche ) ) . Subsequently , this solubilized lysate was ultracentrifuge at 100 . 000 G , 4°C for 45 min to separate the solubilized membrane protein ( supernatant ) from the unsolubilized fragments ( pellet ) [61] . Following solubilization , the lysates were further diluted with hypotonic buffer in 1:3 ratio . GFP-nanotrap beads [26] ( Chromotek ) or protein A/G PLUS agarose beads ( Santa Cruz ) coupled with rabbit-α-CB1 ( Frontier Science ) were then added into the solution and the mixture was incubated for at least 2 h or overnight ( α-CB1-coupled beads ) at 4°C in an overhead tumbler . For GFP-nanotrap immunoprecipitation , the beads were collected and washed three times with ice-cold hypotonic buffer in Pierce Microcentrifuge Spin Column ( Thermo Fischer ) . The beads were then resuspended in hypotonic buffer , mixed with 5 x Laemmli buffer in 4 to 1 ratio , and cooked at 95°C with vigorous shaking in a thermoblock for 5 min and was used for further analyses . For conventional immunoprecipitation using protein A/G beads coupled to α-CB1 , following overnight incubation with solubilized lysate , the beads were collected and washed at least five times with ice-cold washing buffer ( 20 mM Hepes pH 7 . 5 , 5 mM EDTA , 10 mM DTT , 1x protease inhibitor cocktail , and 400nM NaCl ) . Subsequently , the beads were briefly washed with Elution Buffer ( Thermo Scientific ) once . The beads were then incubated with Elution Buffer for at least 1 min , following which the eluate was collected . 1M TRIS buffer pH 9 . 5 was added in the dilution of 1:20 to the eluate to neutralize the acidic Elution Buffer , mixed with 5 x Laemlli buffer , cooked at 95°C with vigorous shaking in a thermoblock for 5 min , and subjected to further analyses . Immunoprecipitates were subjected to LC MS/MS analysis using a nanoHPLC system ( Eksigent , Axel Semrau ) coupled with an ESI LTQ Orbitrap mass spectrometer ( Thermo Fisher ) . Attained MS/MS spectra results were searched against the swiss prot protein database , selected for Mus musculus ( 16 , 338 entries ) using the Mascot software ( Matrix Science ) . Five independent immunoprecipitates were subjected to MS analysis . Subsequently , the uninterpreted MS/MS spectra were searched against the swiss prot protein database , selected for M . musculus ( 16 , 338 entries ) using the Mascot software ( Matrix Science ) . The algorithm was set to use trypsin as proteolytic specificity , assuming carbamidomethyl as a fixed modification of cysteine , and oxidized methionine and deamidation of asparagines and glutamine as variable modifications . Mass tolerance was set to 100 ppm and 0 . 5 Da for MS and MS/MS , respectively . Only protein hits with a probability of p < 0 . 05 for a random match were listed [62] . MS data was evaluated using rPQ score [63] . rPQ score is the ratio of peptide queries obtained for any given protein from GFP-nanotrap immunoprecipitate of the rAAV-EGFP-tagged CB1-transduced brain and control ( eluates of the rAAV-GFP transduced brain lysate ) . For proteins that were not detected in the controls , a detection limit of 0 . 125 ( queries ) was used as a denominator in the rPQ score . Proteins with rPQ values above four were regarded as specific . Cultured cortical neurons were prepared from C57Bl6 mouse embryos from embryonic day 16 ( E16 ) as previously described [64] and transfected using nucleofection device ( Lonza ) and P3 Primary Cell 4D-Nucleofector Kit ( Lonza ) either with plasmids and/or siRNA according to the respective experiments . For experiments on developing neurons , approximately 20 , 000 cells/cm2 were seeded on coverslips inside a 12-well plate ( growth cone modulation assay ) or 10 cm petri dish ( WAVE1 phosphorylation assay ) . For experiments requiring mature cultured neurons , approximately 120 , 000 cells/cm2 were seeded on 14 mm or 24 mm diameter coverslips and cultured for 4 wk . The medium was changed 1 d after preparation with fresh B27-supplemented neurobasal medium ( 500 ml Neurobasal medium , 10 ml B27 , 150 mM GlutaMax , penicillin/streptomycin 50 U/ml , 0 . 88 μl β-mercaptoethanol ) . For older culture , half of the medium was replaced every 6 d with a mixture of B27-medium and N2-supplemented ( Life Technologies ) neurobasal medium ( 500 ml Neurobasal medium , 5 ml N2 , 150 mM GlutaMax , penicillin/streptomycin 50 U/ml , 0 . 88 μl β-mercaptoethanol ) in ratio of 1:2 . To asses modulations of growth cones and WAVE1 phosphorylation , three days in vitro ( DIV ) cultured neurons were treated with different compounds: ( 1 ) DMSO ( Calbiochem ) ( used as vehicle for all compounds , unless stated otherwise , diluted to 1:30 , 000 in neurobasal medium ) ( 2 ) ACEA ( Sigma ) , working end-concentration 100 nM or ( 3 ) AM251 ( Sigma ) , working end-concentration 600 nM . Treatment lasted for 1 h ( for growth cone assay ) or 45 min ( for WAVE1 phosphorylation assay ) . Subsequently , the cultured neurons washed with prewarmed PBS and swiftly fixed with ice-cold 4% paraformaldehyde ( PFA , Sigma ) for growth cone modulation assay . For WAVE1 phosphorylation assay , after washing with prewarmed PBS , the cultured neurons were then lysed with ice-cold 1x RIPA buffer and shaken in an overhead tumbler for 20 min at 4°C . Lysed neurons were then pelleted with centrifugation at 13 . 000 RPM at 4°C and the supernatant was collected , mixed with 5 x Laemmli buffer , boiled with vigorous shaking at 95°C for 5 min , and subjected to immunoblotting . For Gi-inhibition , cultured neurons were pretreated with 100 ng/ml PTX , diluted in water ( Sigma ) overnight ( approximately 20 h ) prior to the treatments with ligands . For inhibition of Rac1 , neurons were pretreated with 50 μM Rac inhibitor II or Z62954982 ( Calbiochem ) for 4 h prior to treatment . WAVE1 inhibition was done by nuclecofecting cells WAVE1 siRNA before plating them . To analyse and quantify dendritic spines , the neurons were transduced with rAAV-CaMKII-GFP at 7 DIV and cultured further for at least another 2 wk . The mature cultured neurons were then treated with either vehicle ( DMSO 1:30 , 000 ) or ACEA ( 100 nM ) for 24 h . Following fixation , neurons were imaged with laser-scanning confocal microscope ( Leica SP2 AOBS ) using a 100X objective ( Leica ) . Only spines from 2nd order dendrites or higher were taken into account . COS7 cells were cultured in complete DMEM ( Gibco ) medium on poly-D-lysine ( Sigma ) coated glass coverslips . After reaching sufficient confluency ( approximately 60%–70% ) , cells were transfected with either GFP or CB1-EGFP together with mWAVE1 plasmid ( OriGene ) using Lipofectamine 2000 ( Invitrogen ) . At least 24 h post transfection , the cells were treated either with DMSO ( 1:30 , 000 ) or with 100 nM ACEA for 45 min , washed with prewarmed PBS , fixed in ice-cold 4% PFA for 20 min , and washed twice with RT PBS before storing or used in immunostaining against WAVE1 and actin filaments . Following immunostaining , cells that are either positive for GFP or CB1-EGFP were individually imaged using SP8 confocal microscopy ( Leica ) with the exact same conditions throughout . Captured images were then analyzed using Fiji software . A ROI of the cell membrane ( membrane ROI ) of each individual cell was created by creating a binary mask of the Phalloidin staining at the cell membrane . The ratio of WAVE1 intensity against the Phalloidin intensity at the corresponding membrane ROI was used to quantify the WAVE1-immunoreactivity at cell membrane . A second ROI was created for WAVE1 and CB1-EGFP colocalization ( colocalization ROI ) . To achieve this , the image of WAVE1 immunostaining of individual cell was overlapped to the corresponding image of CB1-EGFP using the “AND” command in the Fiji software . The ratio of CB1-EGFP total intensity at the colocalization ROI to the CB1-EGFP total intensity of whole cell was used as the quantification of the intensity of CB1-EGFP-WAVE1 overlap . Cultured cortical neurons were nucleofected with pRaichu-Rac1 cDNA plasmid ( Prof . Michiyuki Matsuda ) and plated with the density of approximately 20 , 000 cells/cm2 on 24 mm cover slips . The cultures were then starved stepwise by replacing half of the culture media with fresh nonsupplemented neurobasal medium twice on 2nd DIV and a couple hours prior to measurement which is done on 3rd DIV . For FRET measurement , the cover slips with starved cells were mounted on a cover slips holder filled with nonsupplemented F12 medium ( Gibco ) . Cells were imaged with an inverted microscope ( Nikon Ti-E , Nikon ) with a Nikon 100x Plan Apo objective with NA 1 . 4 . The microscope was equipped with a hardware autofocus ( Nikon PFS ) and a stage top incubation system ( INU series WSKM , TOKAI HIT , Japan ) . The FRET sensor was excited with a metal halide light source ( Intensilight , Nikon ) gated by a 425/26 nm band path filter . The emission was separated by a two-camera adapter ( TuCam , Andor Technology ) with a beam splitter at 509 nm onto two scientific complementary metal oxide silicon ( sCMOS ) cameras ( sNeo , Andor Technology ) ) . The first camera was gated by a CFP filter ( 465/30 nm ) and the second by a YFP filter ( 550/49 nm ) . All filters were of the hard-coated kind and purchased from AHF , Tübingen . Acquisition was driven by NIS-Elements Software 4 . 1 ( Nikon ) with simultaneous acquisition of CFP and YFP channel by hardware triggering between the cameras . For measurement of basal activity prior to any stimulation , time-lapse image series were acquired with 100 ms exposure time every 45 s for 15 min . After the basal measurement , the cells were treated with following pharmacological agents: DMSO ( diluted to 1:30 , 000 in F12 ) , ACEA ( 100 nM ) , AM251 ( 600 nM ) and NGF ( Sigma ) ( diluted in PBS , working end-concentration 100 ng/ml ) , and another image series were taken with the same conditions for 45 min . Image processing was done using NIS-Elements Software 4 . 1 . Channels were aligned by a linear transformation ( rotation and shift ) . In both channels , the background was subtracted frame by frame by defining a background region of interest . For each frame , a ratio image YFP/CFP was calculated and represented as pseudocoloured intensity map , which relatively corresponds to the state of activated Rac1 . Regions in the nuclear and growth cone areas were defined , and average ratios in these areas plotted against time . The raw ratios from individual experiments were divided by the average basal value to get the normalized values time traces . Cultured cortical neurons were nucleofected either with LifeAct-GFP cDNA plasmid ( Ibidi ) only , or together with WAVE1 siRNA or scrambled siRNA ( 100 nM , SCBT ) before plating them on 24 mm cover slips with the density of 20 , 000 cells/cm2 . The following siRNAs were used for transfection: On the 3rd DIV , time-lapse images and treatment were done in a similar fashion as the FRET imaging described above , but without beam splitter and only with one sCMOS camera . Moreover , LifeAct-GFP was excited with a metal halide light source gated by a 482/18 nm band path filter . Image processing was done using Fiji . Briefly , images were converted into binary image , and a mask of the growth cone area was created . The masking was superimposed back into the original 16 bit image , and the area and intensity of the growth cone were measured . For each cell , measurements were taken from images at 0 , 30th , and 60th minutes ( for intensity additionally at 15th and 45th minutes ) . The raw values of each time point were normalized against the basal value at 0 minute and presented as percentile ratio . Cultured cortical neurons were prepared as previously described and nucleofected with LifeAct-mCherry cDNA ( Ibidi ) . After approximately 25 DIV , the matured cultured neurons were transfected either with WAVE1 siRNA #1 or scrambled siRNA ( both at the concentration of 25 nM ) using HiPerFect kit ( Qiagen ) . Three days after siRNA transfection , the matured cultured neurons were imaged at 37°C using an inverted microscope ( Nikon Ti-E , Nikon ) coupled to A1R confocal system ( Nikon ) with an oil immersion objective ( Nikon Plan Apo λ 60x NA 1 . 40 ) . 561 nm Argon-laser was used to excite LifeAct-mCherry . Photobleaching of LifeAct-mCherry was acquired by constant laser exposure for 25 s and concentrating the exposure to a predefined circular ( 3 μm diameter ) ROI that is targeted to individual spines . The laser power used to capture images and bleaching was set to 5% of the maximum laser capacity . To examine the recovery of LifeAct-mCherry after photobleaching on these spines , series of images were taken at 5 s interval for the first 2 . 5 min and further at 30 s interval for another 2 . 5 min . Following background subtraction , the fluorescence intensity of target spines was normalized to the fluorescence intensity of a nearby unbleached “reference” spine . These normalized values were then compared to the average intensity of 5 images on basal state ( prior to bleaching ) to get the FRAP values . Using Prism software ( GraphPad ) , these values were fitted to one-phase exponential function: y = y0 + a ( 1 − exp ( −Kx ) ) , where y0 , a , and K are offset , maximum value and time constant , respectively . Mobile fraction was calculated from the average of the last five images and set as the maximum recovery value ( y0 + a ) . Immunohistochemistry and immunoblotting were done as described previously [45] , with a minor change in blocking solutions . For immunostaining , blocking was done using PBS containing 3% IgG-free BSA ( Jackson Immuno Research ) and 4% normal horse serum . For immunoblotting immunoprecipitation lysate , blocking was done using PBST containing 3% BSA ( Roth ) and 3% nonfat Milk ( Roth ) . Immunostaining and–blotting were done with following primary antibodies: polyclonal goat anti-CB1 ( Prof . Ken Mackie ) ; polyclonal rabbit anti-CYFIP2 ( Sigma ) ; polyclonal goat anti-pGAP43 ( SCBT ) ; monoclonal mouse anti-GFP ( Neuromab ) ; polyclonal rabbit anti-NCKAP1 ( Sigma ) ; monoclonal mouse anti-Rac1 ( BD Bioscience ) ; monoclonal mouse anti-Tau1 ( Merck ) ; polyclonal rabbit anti-βIII-Tubulin ( Sigma ) ; polyclonal goat anti-WAVE1 ( R&D ) ; polyclonal rabbit anti-WAVE1 ( Sigma ) polyclonal rabbit anti-pospho-WAVE1 ( Sigma ) . Actin fibres were visualized with TRITC-conjugated Phalloidin ( Invitrogen ) . Golgi–Cox staining was performed using an FD Rapid GolgiStain Kit ( FD Neurotechnologies ) according to the manual and as previously described [45] . Following impregnation steps , the tissues were shock-frozen in iso-pentane solution precooled in dry ice and afterwards cut with cryostat in sections with 90 μm thickness and mounted in gelatin-coated glass slides . The cut tissues were allowed to dry for at least overnight , then stained by immersing them in a mixture of staining solution from the kit , dehydrated in ethanol , cleared in xylene , and mounted using Eukitt ( O . Kindle ) mounting media . For analysis purposes , only impregnated neurons with diameter of at least 20 μm that were located within the laminae II to V of the spinal dorsal horn were chosen . Moreover , only secondary or tertiary dendrites that were projected in the direction of the dorsal horn were further analyzed . Images were captured using an upright microscope ( Nikon NiE , Nikon ) equipped with Nikon Plan Apo objective set and high resolution CCD camera ( Nikon DS-Ri1 , Nikon ) . Acquisition process was driven by NIS-Elements software 4 . 1 ( Nikon ) . siRNAs were prepared as previously described [65] . Briefly , siRNA ( for each mouse ) was diluted with water up to a quarter of the desired end volume needed for intrathecal delivery and diluted further with 10% glucose solution up to half of the desired end volume . The PEI solution ( Polyplus ) was prepared in similar fashion , first dilution with water up to quarter of the end volume and then again with 10% glucose solution up to half of the end volume . The diluted siRNA solution was then mixed with the diluted PEI solution and incubated for at least 15 min at RT before injection . For down-regulation of WAVE1 , 1 . 2 μg siRNA ( scrambled siRNA or WAVE1 siRNA ) ( SCBT ) was prepared in 15 μl-PEI mix solution for each mouse . ACEA and DMSO were diluted in artificial cerebrospinal fluid ( ACSF; 119 mM NaCl , 26 . 2 mM NaHCO3 , 2 . 5 mM KCl , 1 mM NaH2PO4 , 1 . 3 mM MgCl2 , 10 mM Glucose , 2 . 5 mM CaCl2 , pH 7 . 3 ) . ACEA was diluted to reach the concentration of 200 nM , and DMSO was accordingly diluted 1:500 , 000 in ACSF . 10 μl of this diluted solution was used for every round of treatment . CFA Sigma was injected unilaterally in the intraplantar surface of the hind paw in mice ( 30 μl ) , whereas control mice were injected with 0 . 9% saline [66] . Mechanical sensitivity was measured by applying punctuate pressure using Von Frey filaments ( Ugo Basile ) of different strength [66] . The filaments were applied perpendicularly to the plantar surface in the middle of the mouse’s hind paw with an upward force just sufficient to bend the microfilament . A positive response is a paw withdrawal before the filament bending , and we measured frequency of response out of five applications . Force required to elicit 40% response is considered mechanical thresholds . All behavioural measurements were done in awake , unrestrained , age-matched male mice by individuals who were blinded to the treatment of the mice being analyzed . All image analyses were done by experimentor that was blinded to the identity of the samples being analyzed . All data are expressed as mean ± SEM . Individual statistical analyses are stated in the respective figure legends . Changes with p < 0 . 05 were considered to be significant . No statistical methods were used to determine sample sizes .
|
One of the most interesting features of the endocannabinoid system ( a group of neuromodulatory lipids and their receptors , which promotes homeostasis in a variety of physiological processes ) is its ability to counteract nociception or pain . This function is largely mediated by the receptor component of the endocannabinoid system . One of the most-studied types of cannabinoid receptors , the cannabinoid receptor 1 ( CB1R ) , exerts its antinociceptive function at all levels of the central nervous system , from the periphery up to the brain . Despite numerous studies on the role of CB1R and its antinociceptive effect , our knowledge of the molecular mechanisms underlying this particular feature is still lacking . In this study , we identify the WAVE1-complex—known to be involved in actin nucleation—as novel interacting partners of CB1R . We observe a functional relationship between the WAVE1-complex and CB1R in the regulation of actin filaments in developing as well as mature cultured neurons . Furthermore , we show that inflammation-induced structural plasticity in spinal neurons that contributes to hyperalgesia is regulated by CB1R in a WAVE1-dependent fashion . These findings expand our understanding of CB1R signaling and of the physiological as well as pathological context of pain .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2015
|
The Cannabinoid Receptor CB1 Interacts with the WAVE1 Complex and Plays a Role in Actin Dynamics and Structural Plasticity in Neurons
|
Chromosome movements are a general feature of mid-prophase of meiosis . In budding yeast , meiotic chromosomes exhibit dynamic movements , led by nuclear envelope ( NE ) -associated telomeres , throughout the zygotene and pachytene stages . Zygotene motion underlies the global tendency for colocalization of NE-associated chromosome ends in a “bouquet . ” In this study , we identify Csm4 as a new molecular participant in these processes and show that , unlike the two previously identified components , Ndj1 and Mps3 , Csm4 is not required for meiosis-specific telomere/NE association . Instead , it acts to couple telomere/NE ensembles to a force generation mechanism . Mutants lacking Csm4 and/or Ndj1 display the following closely related phenotypes: ( i ) elevated crossover ( CO ) frequencies and decreased CO interference without abrogation of normal pathways; ( ii ) delayed progression of recombination , and recombination-coupled chromosome morphogenesis , with resulting delays in the MI division; and ( iii ) nondisjunction of homologs at the MI division for some reason other than absence of ( the obligatory ) CO ( s ) . The recombination effects are discussed in the context of a model where the underlying defect is chromosome movement , the absence of which results in persistence of inappropriate chromosome relationships that , in turn , results in the observed mutant phenotypes .
Classical cytological studies have shown that during the zygotene stage of meiosis , chromosome ends are tightly and specifically associated with the nuclear envelope ( NE ) and move coordinately into a “bouquet” configuration such that they are localized within a sub-area of the nuclear periphery . Upon exit from this stage , during early pachytene , telomeres again redistribute throughout the nuclear periphery ( reviewed in [1]; [2]–[4] ) . In budding yeast , these global effects are achieved by means of highly dynamic , actin-dependent , telomere-led movements , which , after initiating at the onset of zygotene , continue into pachytene [2] , [5] , [6] . Recent work from one of our laboratories shows that telomeres and associated nuclear envelope ( NE ) segments move via passive association with nucleus-hugging segments of dynamic cytoskeletal actin cables that tend to form in the vicinity of the spindle pole body ( SPB; [6] ) . A different mechanism has been elucidated for fission yeast; telomeres are tightly and specifically associated with the SPB and the entire complex moves dynamically along microtubules via interaction with the dynein motor complex [7] . Studies in fission yeast , budding yeast ( references below ) , rat , and mouse [8] have shown that , in accord with their special functions , telomeres of meiotic chromosomes become robustly associated with the NE in complexes comprised of both meiosis-specific proteins and proteins recruited from the mitotic program . In S . pombe meiosis , Bqt1 and Bqt2 connect the telomere binding protein Rap1 , which associates with telomeres through interactions with Taz1 , to the spindle pole body protein Sad1 . Sad1 is a member of the SUN domain family of proteins that localize to the NE [3] , [9] . Sad1 is also known to interact with the spindle pole body binding ( SPB ) protein Kms1 [10] . The final telomere/SPB cluster is thought to form through interactions between Bqt1 , Bqt2 , Rap1 , Taz1 , Sad1 , and Kms1 [3] , [7] . In budding yeast , two components of meiotic telomere-NE ensembles have been identified thus far: Ndj1 , also called Tam1 [11] , [12] , and Mps3 [13] . Ndj1 is a meiosis-specific protein that mediates association of telomeres to the NE; as a result , in the absence of Ndj1 , global and dynamic chromosome movements are severely reduced ( [2] , [11] , [12] , [14] , [15]; this work ) . Mps3 , which is present in mitotic as well as meiotic cells , most likely has two roles [13] . First , it interacts directly with Ndj1 such that the two proteins display a mutually dependent requirement for telomere localization to the NE . Second , Mps3 is a SUN domain protein , which suggests that it may mediate interactions between telomeres and cytoskeletal determinants . Recent studies have shown that rapid movement of yeast pachytene chromosomes involves passive association of telomere/NE ensembles to dynamically moving actin cables . Within this mechanically integrated complex , force is exerted on the NE component and transduced via telomere/NE complexes through the NE to the associated chromosome end [6] . The functional role ( s ) of global and dynamic chromosome movements for meiosis , though widely discussed , are not established . In both fission yeast and budding yeast , situations in which telomere localization is aberrant or the motion mechanism is directly abrogated reveal diverse defects . In S . cerevisiae , ndj1Δ strains show levels of crossing over similar to wild-type ( WT ) , partially disrupted crossover ( CO ) interference , modestly increased ectopic recombination , delayed formation of tight juxtaposition of homologs including delayed formation of the synaptonemal complex ( SC ) , defective progression of recombination intermediates into mature recombinants , increased MI nondisjunction , and decreased spore viability [11] , [12] , [16]–[18] . In S . pombe , similar phenotypes are observed for mutants defective in telomere localization; however , in contrast to budding yeast , CO-levels are significantly decreased ( e . g . [3] , [19] ) . Furthermore , in this organism , meiosis does not involve CO interference or the SC , and it has been difficult to perform a detailed analysis of recombination intermediates and their timing . These findings have led to suggestions that motion might play a direct role in recombination and/or homolog juxtaposition . However , the pleiotropic nature of these effects have made it difficult to distinguish defects that are direct consequences of the absence of motion rather than indirect effects and/or those that result from aberrant telomere biology irrespective of motion . We and others [1] , [6] , [20] , [21] have argued that the primary role of movement is to eliminate aberrant topological relationships among chromosomes , e . g . entanglements or “interlocks” and/or other types of unprogrammed connections among nonhomologous chromosomes , a possibility that has not yet been directly assessed in any organism . The present study began with a search for mutations that affect recombination through telomere-dependent effects . The hallmark phenotype conferred by the ndj1Δ mutation is increased nondisjunction of homologs at the MI division , and a screen for mutants defective in chromosome segregation during meiosis [22] identified three additional genes with weak chromosome missegregation phenotypes . We began by further characterizing these missegregation phenotypes . We show that one of these genes , CSM4 , is required specifically for regular segregation of homologs , analogously to NDJ1 , and by several additional criteria , encodes a third participant in the meiotic telomere/NE interactions involved in motion . We further show that the role of Csm4 is distinct from that of either Ndj1 or Mps3 . Finally , we analyzed diverse csm4Δ and ndj1Δ phenotypes for motion at zygotene and pachytene , recombination ( by genetic and physical approaches ) , SC morphogenesis , and meiotic progression . The observed phenotypes suggest a role for chromosome motion that can explain all observed effects and also supports the idea that the primary role of motion is regularization of topological relationships among chromosomes . Related and complementary findings are presented in the accompanying paper by Shinohara and colleagues ( [23] , see also [24] ) .
Ndj1 ( also called Tam1 ) was the first identified component of yeast telomere/NE ensembles [11] , 12 . The hallmark phenotype of ndj1Δ is nondisjunction of homologs at the first meiotic division , with an accompanying modest decrease in spore viability , to 62–82% , as compared to 92–98% in WT [11] , [12] . To identify new mutations that affect chromosomal events during meiosis , and in particular recombination , we focused on three genes , CSM2 , CSM3 , and CSM4 ( chromosome segregation in meiosis; [22] ) , whose corresponding mutations confer phenotypes similar to those of ndj1Δ: decreased spore viability and aberrant meiotic chromosome segregation . We began by further characterizing the nature of the chromosome mis-segregation defect in csm mutants . Spore viability patterns revealed that csm4Δ confers a pattern that is diagnostic of homolog nondisjunction: an excess of tetrads containing 0 , 2 , or 4 viable spores as compared to 1 or 3 viable spores [25] . Of the three csm mutants , only csm4Δ displays this pattern ( Figure 1A , data not shown ) . Csm4 was identified by bioinformatic analysis as a 156 amino acid tail-anchored membrane protein . Consistent with this designation , Csm4 was observed in both the endoplasmic reticulum and the perinuclear membrane when overproduced in mitotic cells [26] . The homolog nondisjunction phenotype of csm4Δ was confirmed and extended as follows: Taken together , these analyses show that the primary segregation defect in csm4Δ mutants is homolog nondisjunction . Similar results are reported by Kosaka et al . [23] . Comparison of isogenic strains reveals that the phenotype of csm4Δ is significantly stronger than that of ndj1Δ ( Figure 1A ) . As previously mentioned , csm4Δ mutants display a spore viability pattern indicative of nondisjunction ( 4 , 2 , 0>3 , 1 viable spores ) . This pattern is more severe in csm4Δ vs . ndj1Δ ( Figure 1B ) . Further , a larger percentage of two-spore viable tetrads are sisters in csm4Δ ( 88% ) compared to ndj1Δ ( 69% ) . As judged by the first two of these phenotypes , the double mutant defect is very similar to ndj1Δ , but slightly weaker ( Figure 1 ) . These same patterns are also apparent in the percentage of four viable spore asci and overall spore viability ( Figure 1A and B ) . These mutant phenotypes imply functional interaction between Csm4 and Ndj1 with respect to MI homolog disjunction . The observed epistatic relationship is intriguing . First , it is the weaker phenotype that dominates ( is epistatic to ) the stronger phenotype . Second , the occurrence of slight synergy implies that not only is Ndj1 strongly required for the csm4Δ phenotype but conversely , Csm4 is weakly required for the ndj1Δ phenotype . Importantly , the above conclusions do not reflect differences in sporulation efficiencies . For WT , 82% of cells yielded asci with three or four spores and 89% yielded asci with at least one spore ( n = 234 cells examined ) ; csm4Δ , ndj1Δ , and csm4Δ ndj1Δ all exhibited similar reductions in both categories: 50 , 51 , and 64% respectively , and 73 , 59 , and 78% respectively ( n = 229 , 221 , and 238 cells examined , respectively ) . Association of telomeres with the NE , occurrence of the bouquet configuration , and dynamic telomere movements were assessed , in SK1 isogenic strains ( Table S1 ) by analyzing the disposition of Rap1-GFP foci . Rap1 localizes prominently and focally at telomeres and less markedly throughout chromatin ( Figure 2A , panel i; [5] ) . Our approach can detect foci that correspond to single bivalent telomeres at pachytene ( R . Koszul , unpublished data ) , and thus , for earlier stages , should be sensitive enough to detect clusters of two ( or more ) unpaired homolog telomeres or four ( or more ) individual chromatids . Since bouquet formation involves colocalization of telomeres near , but not at , the SPB ( Introduction ) , we used strains in which the SPB was also labeled , with Spc42-RFP ( Figure 2B , panel i; Figure S1; Table S1; Materials and Methods ) . Cells were taken through synchronous meiosis under standard conditions ( Material and Methods ) . In such cultures , at any given time point , the majority of nuclei are in one particular stage . Specifically , at t = 2 , 3 , 4 and 5 h , the majority of nuclei are in G2 , leptotene , zygotene , and pachytene respectively , as defined by fluorescence activated cell sorter ( FACS ) analysis and SC status ( e . g . [6] and below ) . Zygotene and pachytene nuclei can thus be defined operationally by the population average behavior of nuclei at t = 4 and t = 5 h , respectively , albeit with some “contamination” from other stages at each time point . Organization of Rap1-GFP foci was analyzed in nuclei of living cells by 3D acquisition , in which a series of 400 nm optical z-sections are taken over time ( Figure 2A; n = 50 cells at every time point; 10 planes total , exposure time of 900 ms; Material and Methods ) . In WT mitotic cells ( t = 0 h in SPM ) , nuclei could be sorted by visual inspection into two categories ( Figure 2A , panel i , panel ii , t = 0 ) : ∼60% showed a ring of Rap1-GFP foci located in the periphery of the main chromosomal mass , with no clearly discernable internal foci , implying that telomeres are located “peripherally” . The remaining ∼40% clearly showed internal foci , implying a “dispersed” disposition pattern . Similar categories have been seen in other studies ( e . g . [15] ) . In contrast , by 2 h after initiation of meiosis , most nuclei were in the peripheral configuration ( Figure 2A , panel ii ) . This progression presumably reflects complete migration of telomeres to the nuclear periphery via formation of meiosis-specific telomere/NE complexes that have assembled in early prophase . Since meiotic telomere/NE association at G2/leptotene is a regular feature of meiosis in many organisms ( e . g . [30] , [31] ) , we infer that yeast exhibits this same progression but with a prior “background” from mitotic telomere/NE association . Living cells were also analyzed for the movement of Rap1-GFP foci . For this purpose , the focal plane of the microscope was set at the top of each examined nucleus so that movements around the nuclear periphery could be observed in apparent two dimensions . Frames were taken at one-second intervals over a period of one minute . The positions of the spots present in such focal planes were recorded and analyzed using SpotTracker2D ImageJ plug-in [32] , when the amplitude of the displacement was limited , or manually at t = 4 h ( below ) . Such analysis was performed for 5–12 Rap1 foci taken from a minimum of 5 different nuclei ( yielding a total of 340 one-second step-sizes for both time points ) . At t = 2 h , when telomeres have reached their peripheral localization , the average velocity of movement ( v ) through two-dimensional space was 0 . 07±0 . 05 ( S . D . ) µm/sec . Further , these step sizes exhibit nearly ( but not perfectly ) a Gaussian distribution , suggesting that all foci are behaving similarly ( Figure 2C , red curve; for details see figure legend and Materials and Methods ) . The same features are also seen previously at t = 0 h ( average velocity 0 . 06±0 . 05 µm/sec with a near-Gaussian step-size distribution ) , in accord with the fact that active motion has not yet begun by t = 2 h [6] . In contrast , at t = 4 h , foci exhibit an increased average velocity of 0 . 23±0 . 23 µm/sec , in accord with earlier studies [5] , [6] . Moreover step-sizes no longer fit a Gaussian distribution ( Figure 2C , red curve ) ; instead , there appear to be two types of movement , with a majority of steps being smaller and corresponding to a near-Gaussian distribution ( Figure 2C , blue curve; ∼77% of total ) plus a minority of much larger steps . The two apparent subpopulations exhibit velocities of ∼0 . 1 and ∼0 . 45 µm/sec , respectively , both of which are greater than the velocity observed at t = 2 h ( ∼0 . 07 µm/sec ) . The existence of two such populations is in good agreement with the fact that , during the period of active actin-mediated movement , only a subset of telomeres are directly coupled to the motion-generating mechanism while others are either unaffected or dragged along passively [2] , [6] , [24] . To determine the overall disposition of telomeres at various stages , formaldehyde-fixed nuclei were analyzed in 3D by collection of an appropriate set of “z-sections” ( Materials and Methods; n = 100 nuclei for each time point ) . Nuclei in which bright Rap1-GFP foci ( i . e . the telomeres ) were in a peripheral configuration ( above ) were scored with respect to whether most of the signals were or were not detectably clustered and , if so , whether those clusters occurred in the vicinity of the SPB ( representative examples in Figure 2Bi; details in Figure S1 ) . SPB-associated colocalization was defined as “bouquet” . Such configurations include both “loose bouquet” and “tight bouquet” ( Figure 2B , panel i ) , a distinction previously documented for yeast and several other organisms ( e . g . Sordaria , D . Zickler , personal communication; [5] , [33] , Kosaka et al . , accompanying paper [23] ) . Bouquet nuclei gradually increased in frequency from 2 h after meiosis induction , peaked at t = 4–5 h ( i . e . zygotene/pachytene ) , and then diminished dramatically when cells entered pachytene in accord with expected loss of the bouquet configuration at this stage ( Figure 2B , panel ii; analogous results obtained in a second independent experiment , not shown ) . As also noted in early studies ( e . g . Ref . [5] , [13] , [15] , [24] ) , the proportion of bouquet nuclei is low ( ∼20% ) even at the peak time points . This likely reflects the fact that zygotene nuclei have the potential for telomeres to be in the bouquet configuration but are undergoing such complex dynamic telomere movements that all telomeres are only present in a common area some fraction of the time [5] . ndj1Δ/csm4Δ mutants and WT were analyzed for telomere-related events in parallel . All three mutants exhibit a WT mitotic-like configuration at t = 0 . However , during meiosis , ndj1Δ telomeres fail to progress to a fully peripheral localization pattern ( as shown previously; [15] ) while , in contrast , csm4Δ telomeres behave indistinguishably from WT ( Figure 2A , panel ii ) . Thus , while Ndj1 is required for meiosis-specific telomere/NE association , Csm4 is not , as also shown by Kosaka et al [23] . Further , the ndj1Δ csm4Δ double mutant exhibits the ndj1Δ phenotype ( Figure 2A , panel ii ) , in accord with previous indications that Ndj1 localizes to telomeres and directly mediates their meiotic NE targeting [15] . ndj1Δ/csm4Δ mutants were analyzed for telomere movement at t = 0 and at zygotene , the time of which was defined for all three mutants by analysis of SC formation ( below ) . Differences among different situations were evaluated for significance by comparison of step-size distributions by parametric tests ( Materials and Methods ) . By this criterion , the following patterns emerge: ( i ) At t = 2 h , all three mutants exhibit velocities of movement similar to that seen in WT . ( ii ) At zygotene , all three mutants exhibit significantly less movement than WT , implying that Csm4 , like Ndj1 [2] is required for active motion ( Figure 2C; see also [6] , [24] . Since telomeres are still NE-associated in the absence of Csm4 , these findings suggest that this molecule is involved in the motion-producing force-generating process per se . ( iii ) Interestingly , from t = 0 to zygotene , there is a small but significant increase in motion in the absence of Ndj1 but no significant change in the absence of Csm4 ( Figure 2C ) . There is also no significant increase when both proteins are absent . This suggests that csm4Δ is partially epistatic to ndj1Δ with respect to zygotene motion ( see Discussion ) . Other studies further show that Ndj1 is not required for the NE deformations that signal actin-mediated motion while absence of Csm4 completely abrogates such motions [6] . Thus , the residual Csm4-dependent movement observed in ndj1Δ appears to reflect residual movement that is independent of meiosis-specific telomere/NE association , e . g . via mitotic-like or non-specific associations . In the absence of Csm4 , in contrast , telomeres may simply be “not moving” or may actually be “locked in place . ” In accord with abrogation of telomere/NE association and/or chromosome movement , there is no detectable bouquet formation in ndjΔ , csm4Δ , or the ndj1Δ csm4Δ double mutant ( Figure 2B , panel ii ) . This is also consistent with data reported for ndj1Δ [15] and for csm4Δ by Kosaka et al . [23] . We also explored the cytological localization of Csm4 during meiosis in relation to the localization of telomeres using a strain ( EAY1797 ) carrying an integrated Csm4-GFP fusion driven from the native CSM4 promoter and the Rap1-RFP fusion ( Materials and Methods ) . Intrinsic Csm4-GFP fluorescence is sufficiently weak so that localization can only be assessed with anti-GFP antibody in fixed cells , and even then , with substantial background staining . Nonetheless , at mid-prophase , Csm4 can be seen in foci around the periphery of the nucleus ( Figure 2D , left ) . These foci often overlap with strong foci of Rap1-RFP ( Figure 2D , right ) . These images provide evidence suggestive of NE localization of Csm4 and a tendency for association with telomeres . A strain expressing only Rap1-RFP does not show such patterns ( data not shown; strain NKY4005 ) . Defects in MI homolog segregation often reflect defects in the formation of crossovers ( COs ) . Further , it would be interesting to know whether/how telomere dynamics affect recombination . We therefore examined recombination in csm4Δ by both genetic ( this section ) and physical analyses ( below ) . We examined crossing over in 12 different intervals by tetrad analysis in WT and csm4Δ ( Figure 3 , Tables 1 , S1 , and S2 ) . The csm4Δ mutation conferred a 30–40% increase in the level of COs for all four intervals in the SK1 congenic strains . In the analysis of complete tetrads , the URA3-LEU2 and ADE2-HIS3 intervals were significantly different from WT ( G-test , p<0 . 007 , 95% confidence level , Dunn-Sidak correction , [34] , [35] ) but the LEU2-LYS2 ( p = 0 . 07 ) and the LYS2-ADE2 ( p = 0 . 013 ) were not ( Figure 3A ) . However , in the spore analysis , only the LEU2-LYS2 interval ( p = 0 . 014 ) was not significantly different from WT ( p<0 . 007 , Figure 3A ) . Similarly , in isogenic SK1 strains , CO frequencies were increased in csm4Δ mutants at four out of eight analyzed intervals in complete tetrads and at six out of eight intervals in the spore analysis ( G-test , p<0 . 05 , 95% confidence level ) . At the HIS4-LEU2 interval on chromosome III , CO levels were indistinguishable between WT and csm4Δ in both data sets ( Figure 3B ) . Oh et al . [36] showed that the sgs1ΔC795 mutation conferred an ∼20% increase of map distance in SK1 strains that was primarily due to an increase in the frequency of NPD tetrads . Based on this observation the authors suggested “… that a fraction of the events that would normally form single crossovers in WT cells gives rise to closely spaced double crossovers in sgs1ΔC795 cells . ” We did not see a similar pattern in csm4Δ mutants . Three of four genetic intervals ( all but LEU2-LYS2 ) in the SK1 congenic strain displayed significantly different PD:NPD:TT distributions in csm4Δ compared to WT , even when the NPD class was ignored ( G-test , p<0 . 05 ) . This observation suggested that the csm4Δ mutation did not increase map distances by specifically increasing the frequency of closely spaced double crossovers . This conclusion is reinforced by physical analysis of DNA events: species representing large joint molecules are overrepresented relative to other types of joint molecules in sgs1Δ [36] but not in csm4Δ ( compare Figure 7C with Figure S2C ) . In WT meiosis , occurrence of a CO in one region of a chromosome is accompanied by a reduced probability that one will also occur in a nearby region , a phenomenon known as "CO interference” . We assayed interference in csm4Δ by three different methods . One approach utilizes the method of Malkova et al . [37] , which evaluates the occurrence of interference in adjacent intervals by utilizing all of the information contained in complete tetrads ( Figure 4 , Table S3 ) . In WT , interference was observed for all three interval pairs . In contrast , csm4Δ strains showed reduced interference in two intervals and no significant interference in the third . One interpretation of these data is that interference does not extend as far from the initial crossover site in csm4Δ strains as it does in WT . A second approach evaluated the “coefficient of coincidence” ( COC ) . For a given pair of intervals , the COC is the ratio of the observed frequency of double CO events to that expected if COs in the two intervals occurred independently . In accord with results obtained using the Malkova et al . [37] method , the csm4Δ mutant exhibited a modest reduction in interference in all four intervals analyzed ( Table 2 ) . A third approach to interference analysis is the calculation of the ratio of observed non-parental ditypes ( NPD ) which reflects the occurrence of a four-strand double crossover , to that predicted by the number of single crossovers detected ( NPD ratio , [38] , [39] ) . By this criterion , we were unable to determine a difference in interference between WT and csm4Δ in all three intervals measured ( Table 2 ) . It is not clear why interference as assessed by NPD ratios is less affected by csm4Δ than when assessed by other methods . One possible explanation for the disparity between the COC and NPD ratio measurements is that NPD measurements may be affected by “chromatid interference” . Chromatid interference is a restriction on the independence of chromatid selection during CO recombination and has not been previously observed in yeast [37] , [40] . Correspondingly , WT and the csm4Δ mutant both exhibited a 1∶2∶1 ratio of exchanges involving two , three , or four chromatids in the URA3-LYS2-HIS3 interval , implying an absence of chromatid interference in both cases ( data not shown ) . During meiosis , the formation of COs , as opposed to noncrossovers ( NCOs ) , is promoted by a large number of proteins that are specifically dedicated to this process . Among these , the Msh4-Msh5 complex appears to act around the time of CO/NCO differentiation [41] , while Mlh1-Mlh3 appears to act later , likely during double Holliday junction ( dHJ ) resolution ( [42]–[44]; N . Hunter , personal communication ) . In both the msh5Δ csm4Δ and mlh1Δ csm4Δ double mutants , recombination levels are significantly lower at all examined intervals than levels seen with the csm4Δ alone ( G-test , p<0 . 007 , Dunn-Sidak correction , Figure 3 , Table 1 ) . This suggests that , while absence of Csm4 affects the level of COs , those COs are still occurring via the normal Msh5/Mlh1-dependent pathway . Conversely , in msh5Δ and mlh1Δ mutant backgrounds , absence of Csm4 increases CO levels about two-fold above the single msh5Δ and mlh1Δ mutant levels , suggesting that the effect of csm4Δ on CO levels is upstream and/or independent of the msh5Δ and mlh1Δ effects . Although the msh5Δ is only significantly different from its corresponding double mutant at the URA3-LEU2 interval in the spore dataset , the mlh1Δ recombination levels differ significantly from mlh1Δ csm4Δ at two out of four intervals in the tetrad dataset and three out of four intervals in the spore dataset ( G-test , p<0 . 025 , Dunn-Sidak correction ) . Furthermore , spore viability in the double mutants ( Figure 1; csm4Δ msh5Δ = 22%; mlh1Δ csm4Δ = 42% ) was much lower than any of the single mutants alone ( Figure 1; csm4Δ = 64%; msh5Δ = 36%; mlh1Δ = 68% ) . Taken together , these genetic interactions suggest that Csm4 acts independently of Msh4-Msh5 and Mlh1-Mlh3 . Physical analysis of csm4 msh4 mutants by Kosaka et al . [23] is consistent with this observation . ndj1Δ conferred a 30–40% increase in CO frequencies at all intervals , indistinguishable from the increase seen in csm4Δ ( G-test , p<0 . 007 , Dunn-Sidak correction ) . Crossover interference is also similarly affected in ndj1Δ and csm4Δ . These unusual phenotypes in both mutants provide strong support for Ndj1 and Csm4 playing similar roles with respect to recombination . In direct confirmation of this conclusion , the csm4Δ ndj1Δ double mutant is indistinguishable from either single mutant with respect to increases in CO levels in all four genetic intervals analyzed ( G-test , p<0 . 017 , Dunn-Sidak correction , no intervals are significantly different between csm4Δ and csm4Δ ndj1Δ and only one interval is significantly different between ndj1Δ and the double mutant , Figure 3 , Table 1 ) and interference phenotypes ( Figure 4; Table 2 ) . We note that a previous study also detected reduced interference in ndj1Δ but did not detect increased CO levels [11] . Strain background effects are likely responsible for this difference . Non-Mendelian ( non-2:2 ) segregation of an allele , often referred to as “gene conversion” , implies that a recombination interaction has occurred between homologs rather than sisters . In budding yeast , gene conversion events are usually manifested as 1∶3 or 3∶1 segregation patterns of individual alleles [45] . In the congenic SK1 background , gene conversion levels at TRP1 , URA3 , LEU2 , LYS2 , ADE1 , and HIS3 loci occurred at levels ranging from 0 to 0 . 8% of tetrads in WT and at indistinguishable levels in csm4Δ . The relatively low levels of gene conversion observed for these markers may reflect the fact that they mostly involve 1–3 kb heterologies . We also examined gene conversion at 11 loci marked by a variety of mutation types ( point mutations and insertions/deletions ) . In these SK1 strains [46] , non-Mendelian segregation frequencies ranged from 0 . 2% to 5 . 3% of tetrads in WT and 0 . 2% to 5 . 2% in csm4Δ derivatives . The total frequencies of gene conversion at all loci were 14 . 7% in WT and 17 . 4% in csm4Δ , with no chromosome- or locus-specific differences detectable . Gene conversion frequencies reflect the combined effects of a couple of variables: the frequency of recombination initiation at/near the locus and the probability that an event initiated on one homolog will chose a partner duplex on the other homolog rather than on the sister chromatid . The simplest possibility is that csm4Δ has little effect on either of these features , although a balanced effect on both parameters cannot be excluded . Homolog disjunction requires the presence of at least one interhomolog connection , created by the combined effects of a CO and the cohesion between sister chromatids centromere-distal to that CO . Homolog disjunction also requires appropriate reductional functioning of homolog centromere/kinetochore complexes and the efficient release of chiasma-maintaining sister connections . Because csm4Δ exhibits higher than WT levels of COs , it seems unlikely that homolog nondisjunction in csm4Δ results from the absence of a CO . On the other hand , in WT meiosis , special mechanisms ensure that each homolog pair experiences at least one CO ( the so-called “obligatory” CO ) even when overall CO levels are reduced ( for recent discussion see [34] ) . Thus , it remained possible that Csm4 is required for the occurrence of the obligatory CO . We examined the presence or absence of COs on chromosomes that had undergone nondisjunction using a system developed by Rockmill et al . ( [29]; Figure 5 , Table S4 ) . This system allows for the selection , purification , and genetic analysis of spores disomic for chromosome III in the BR strain background ( Table S1 , Materials and Methods ) . As a baseline for this analysis , map distances for six intervals spanning 167 kb of the 317 kb chromosome III were determined from four-spore viable tetrads in WT ( 390 tetrads dissected , 317 four-spore-viable , 93% spore viability ) and csm4Δ ( 697 tetrads dissected , 203 four-spore viable , 53% spore viability ) in the BR strain background . This spore viability pattern observed in csm4Δ was similar to that seen for both the congenic and isogenic csm4Δ SK1 strains . However , unlike what we observed in the SK1 strain background , the recombination frequencies were similar in WT and csm4Δ ( only two out of six intervals were significantly different , G-test , p<0 . 025 , Dunn-Sidak correction , see comment on strain background effects below ) . Tetrad analysis of the csm4Δ derivatives of these strains showed that 9 . 3% ( 17/182 ) of the two-spore-viable tetrads dissected displayed nondisjunction of chromosome III . This value is similar to what was seen in the congenic SK1 strain background ( 7 . 8% ) . Tetrad analysis also revealed that 85% ( 154/182 ) of csm4Δ two-spore-viable tetrads were sisters , consistent with meiosis I nondisjunction , and again , similar to that seen in the congenic SK1 strain background ( 88% ) . These data , along with the spore viability profile ( data not shown ) , show that again , aberrant segregation in csm4Δ strains resulted primarily from homolog nondisjunction . From sporulated csm4Δ cultures , we selected and analyzed 185 random spores disomic for chromosome III . In this analysis , CO levels were examined in a manner that accounted for the inability to detect homozygosity of dominant markers in the csm4Δ disomic spores and all tetrad information was converted to single spore data to allow direct comparison between the disome and tetrad data ( Materials and Methods ) . Interestingly , the distribution of COs among the examined intervals was significantly different from that observed among chromosomes that experienced regular segregation: when compared to csm4Δ tetrads , csm4Δ disomes displayed significantly increased levels in the iNAT-iLEU2 and iLEU2-MAT intervals on the right arm of the chromosome and a significantly decreased level of crossing over in the HIS4-iTHR1 region ( G-test , p<0 . 025 , Dunn-Sidak correction , Figure 5 , Table S4 ) . In addition , the total map distance for six intervals on chromosome III was higher for the disomes ( 85 cM ) compared to the WT ( 78 cM ) and csm4Δ ( 71 cM ) complete tetrads ( Figure 5 , Table S4 ) . Thus , homolog nondisjunction in csm4Δ does not appear to result from absence of the obligatory CO . More generally , chromosomes are not mis-segregating because of lack of recombination events , too many recombination events , or because they were only receiving crossovers in inappropriate locations ( e . g . , telomeres , centromeres ) . The altered distribution of crossovers seen in csm4Δ disomes also differed from what was previously seen in disomes isolated from WT and sgs1Δ strains . In these backgrounds , elevated levels of crossing over were seen at all loci with the highest levels found at those closest to the centromere , consistent with PSSC causing the majority of the mis-segregation events detected [29] . Rockmill et al . [29] hypothesized that the increase in centromere-proximal crossing over in WT and sgs1Δ strains caused PSSC events through the loss of sister chromatid cohesion . However , our data are not consistent with this scenario . The crossovers seen in csm4Δ disomes were not consistently higher or lower across the chromosome length , were not localized to a specific chromosomal position ( e . g . centromeres ) , and were clearly not aiding in proper chromosome segregation . Thus there is no clear pattern or trend from these data that can explain how such a changed distribution can cause chromosome mis-segregation . A different type of explanation for homolog nondisjunction is presented below ( Discussion ) . To address the nature of recombination in csm4Δ/ndj1Δ meiosis in more detail , we assayed , in synchronously initiated meiotic cultures , physical events at the HIS4LEU2 locus of chromosome III ( Figure 6A ) , where virtually all events emanate from a single DSB hot spot . Since Csm4 and Ndj1 are implicated in telomere status and dynamics ( below ) , we also asked whether mutant recombination phenotypes depend upon the presence of chromosomal telomeres in cis to the assayed locus . For many phenotypes we analyzed recombination between HIS4LEU2 loci present on circular versions of chromosome III as well as on normal linear chromosomes III . Presence of the circular chromosome was confirmed for all analyzed strains ( Figure S3A ) . All strains examined are isogenic SK1 derivatives ( Table S1 ) . The levels of CO and noncrossover ( NCO ) products were determined at the end of meiosis using a two-dimensional gel approach ( Figure 6B; [34] ) . In all three mutants ( ndj1Δ , csm4Δ , and ndj1Δ csm4Δ ) , both types of products were present at high levels ( Figure 6C ) . Correspondingly , DSBs form at the very similar levels in all four strains , as assessed in a rad50S background [47] where their turnover to later intermediates is blocked ( Figure 6D and S3B ) . Genetic analysis ( above ) detected modest increases in COs and non-Mendelian segregations , which presumptively represent total events and thus NCOs as well as COs . Such increases are not obvious in the present study at HIS4LEU2 ( Figure 6C; see also Figure 7C “COs” ) ; however , slightly increased levels of DSBs are reported from analogous analysis of a slightly different version of HIS4LEU2 by Kosaka et al . [23] . rad50S data also show that DSBs occur in a timely fashion in all three mutants , with small differences in timing among different strains that are well within standard culture-to-culture variation ( Figure 6D ) . Since the timing of DSB formation reflects the timing of DNA replication [48] this suggests that DNA replication also occurs with normal timing , which we have confirmed directly in all three mutants by FACS analysis ( data not shown ) . After DSB formation , however , progression through ensuing steps of recombination is severely delayed , for both linear and circular chromosomes . These steps were analyzed by one-dimensional gels that display DSBs and CO products plus two-dimensional gels that display single-end invasions ( SEIs ) and double Holliday junctions ( dHJs ) , two branched species on the pathway to formation of CO products ( [49] , [50] , [51]; Figure 7A and B; Figure S3C ) . All three intermediates ( DSBs , SEIs , and dHJs ) occur at higher than normal levels and peak at later than normal times in all three mutants , with very similar results for linear and circular chromosomes ( Figure 7C ) . This pattern is indicative of delayed progression , as discussed in detail below . Correspondingly , while CO products form at high levels in all cases , they appear with a substantial delay in all three mutants , for both linear and circular chromosomes ( Figure 7C ) . An additional type of one-dimensional gel analysis of the linear chromosome strains reveals that the same is also true for NCO products , which are delayed to the same extent as CO products in all three mutants ( Figure S2A and B ) . The detailed effects of ndj1Δ/csm4Δ mutations on recombination progression are elucidated by further analysis of the primary data , by two approaches [51] . First , the lifespan of an intermediate , given by the area under the corresponding primary data curve , defines the time spent by a given intermediate at that stage; thus , an increase in the lifespan of a species implies a delay in progression out of the corresponding step . The lifespans of DSBs , dHJs and SEIs all increased in each of the three mutants , relative to WT , for both linear and circular chromosomes , with the biggest increase for DSBs ( Figure 7D ) . Thus , all three mutants confer defects in all three corresponding steps , with the biggest delay in progression out of the DSB stage and lesser delays in progression from SEIs to dHJs and progression from dHJs to COs . Second , cumulative curve analysis defines the percentage of cells that have “entered” a particular stage as a function of time after initiation of meiosis , with “time of entry” defined as the time at which 50% of cells have carried out the corresponding step . Once again , a delay in progression is seen as an increase in the time interval between the entry time for one step and the entry time for the successive step . Among the three transitions examined , the biggest effect of the mutations is on the difference between the time of DSB formation and the time of SEI formation ( i . e . the DSB-to-SEI transition ) , as expected from lifespan analysis , with smaller ( or no ) differences seen for the other two other transitions ( SEI formation to dHJ formation , dHJ formation to CO formation; Figure 7E ) . We note that similar effects have been observed not only in the two complete experiments presented in Figure 7C but in a third set of experiments involving a different set of linear chromosome strains ( Figure 7D and E , “Lin2” ) . We also note that while previous work suggested no delay in the DSB-to-SEI transition in ndj1Δ [16] , reanalysis of that data suggests that the same delay was observed in that study as is reported here . We conclude that: ( i ) absence of Ndj1 and/or Csm4 confers delayed progression at every individual assayable step of recombination but most prominently at the DSB to SEI transition; ( ii ) that linear and circular chromosomes behave quite similarly with respect to these effects , though minor differences are not excluded; and ( iii ) that delays are not accompanied by any significant reduction ( or obvious increase ) in the level of final CO and NCO products . There is also a strong tendency for csm4Δ to confer the strongest effects among the three analyzed mutations ( Figure 7D and E ) . Effects of ndj1/csm4 mutations on three other aspects of recombination were also examined for both linear and circular chromosomes ( Figure S2C and D ) . First , “large joint molecules” ( LJMs ) , indicative of multi-chromatid interactions [36] , occur in ndj1/csm4 mutants as in WT meiosis ( Figure S2C ) . Moreover , direct comparison of LJM and dHJ levels reveals that the two species are affected coordinately , with no indication that the mutants have increased LJM levels as observed in certain other mutants ( Figure S2D; [36] ) . Second , ectopic recombination , which occurs between the molecularly-inserted LEU2 locus at HIS4LEU2 and the endogenous leu2 locus [52] , is slightly elevated in all three mutants , as compared to WT , as seen at very late time points ( Figure S2C ) , and as previously observed for ndj1Δ [18] . Third , for the linear chromosome , all three mutants exhibit a significant , but somewhat reduced , ratio of inter-homolog versus inter-sister dHJs ( ∼2 . 7∶1 versus ∼5∶1 for WT; Figure S2C ) . This difference could reflect: ( i ) defective homolog partner choice at the time that choice is made ( concomitant with DSB formation; [53] , K . K . and N . K . unpublished ) ; ( ii ) deterioration of homolog bias thereafter; and/or ( iii ) a differential role of Ndj1/Csm4 in progression of inter-homolog CO interactions versus inter-sister CO interactions . No such difference is observed for the circular chromosome; perhaps this is related to the fact that it does not exhibit such strong inter-homolog bias in WT ( Figure S2C ) . We note that related analysis of linear chromosome recombination in csm4Δ by Kosaka et al . [23] also reveals delays at all assayable steps , very similar to the delays reported here , and , coordinately , delays in formation of COs and NCOs . The two studies differ somewhat with respect to reported effects on the levels of COs and NCOs , perhaps because slightly different assays and HIS4LEU2 alleles were used . However , in both cases , high levels of both products do occur . All three ndj1/csm4 mutations confer delays in the occurrence of MI . The extent of the delay is greatest for csm4Δ , smallest for ndj1Δ , and intermediate for the double mutant . This is a highly reproducible effect . It has been observed in both linear and circular chromosome experiments ( Figure 7C ) and in all of the many other experiments performed with these mutants in the current and previous studies using the SK1 background ( [6] , [16]; data not shown ) . In a number of mutants , delayed and/or inefficient occurrence of MI results from delayed recombinational progression . This is also true for ndj1Δ/csm4Δ mutants: elimination of recombination initiation completely eliminates the MI delay in all three mutant strains ( see below ) . Appearance of COs and NCOs marks the end of recombination . Since MI delays are due to delays in recombination , it might be expected that , once these products appear , the mutants should exhibit no further delay in progression . Specifically: occurrence of MI should be delayed to the same extent as occurrence of COs . However , there are hints that this is not the case: occurrence of MI is even further delayed than is occurrence of COs , dramatically for two csm4Δ experiments and less dramatically for other mutants and/or other experiments ( Figure 7D ) . Moreover , since all MI delays are completely dependent upon recombination initiation ( below ) , this discrepancy seems to imply that , even after the majority of recombinational interactions are completed ( as seen by appearance of the high levels of COs and NCOs as detected by DNA analysis at HIS4LEU2 ) , a minority of interactions ( which do not make a significant contribution to total DNA-detected events ) remain unresolved and are either completed much later or not at all ( Discussion ) . We further find that the delays in occurrence of MI in all three mutants ( Figure 7 , also shown in Figure 8B , left panel ) are completely eliminated if initiation of recombination is eliminated by the spo11 ( Y153F ) mutation ( Figure 8B , right panel ) , as seen previously for ndj1Δ [16] and for csm4Δ by Kosaka et al . [23] . This effect is in accord with the fact that recombination defects trigger MI defects in several other situations ( e . g . [41] , [54] ) . For csm4Δ we further determined that the MI delay was eliminated by a rad17Δ mutation ( Figure S4 ) , which is known to alleviate MI delays resulting from recombinational blocks in other situations ( e . g . ndj1Δ; [55] ) . Spore viability in rad17Δ csm4Δ was dramatically reduced as compared to either single mutant , as would be expected from the compromise of a checkpoint that monitors aberrant recombinational progression [55] , [56] . Morphogenesis of the SC is readily monitored in whole cells using Zip1-GFP as described previously [41] , [57]: cells containing focal Zip1-GFP are in leptotene; those with an incomplete complement of Zip1 linearities are in zygotene , corresponding to formation of SC; and those containing a maximum complement of Zip1 linearities are in pachytene , a morphology corresponding to full length SC ( Figure 8A , panel i; e . g . [57] ) . csm4Δ , ndj1Δ , and ndj1Δ csm4Δ mutants all exhibit abnormal kinetics of progression into and out of the zygotene and pachytene stages ( Figure 8A , panels ii ) . Lifespan analysis ( described above ) further shows that all three mutants remain in both stages longer than WT ( Figure 8A , panel iii ) . Cumulative curve analysis ( described above ) further shows that all three mutants exhibit delayed onset of zygotene and delayed onset of pachytene ( Figure 8A , panel iv ) . These defects can be attributed to defects in progression of recombination ( above ) : onset of zygotene is triggered by CO-designation [58] , progression from zygotene to pachytene mirrors the progression of CO-designation and/or SC formation , and exit of pachytene is dependent upon completion of recombination [41] , [54] . In accord with these defects , some nuclei ( ∼20% ) exhibit large aggregates of Zip1-GFP , i . e . polycomplexes ( data not shown ) . In our analysis , pachytene appears more prolonged than zygotene; Kosaka et al [23] suggest that zygotene is more severely affected than pachytene . This may represent slight differences in progression in the two experimental protocols or between the particular strains examined . Noted above , however , was the peculiar fact that , in all three mutants , onset of MI is delayed more than completion of recombination product formation . We favor the idea that there are a small minority of recombinational interactions which persist , undetected by DNA analysis , after most interactions are fully completed ( Discussion ) . If this were true , and given that exit from pachytene is dependent on completion of recombination , and in turn , licenses onset of MI , it could be expected that all mutation-dependent effects would be complete by the end of pachytene , with no further mutant-dependent delay between pachytene exit and MI . This appears to be the case: in all three mutants , exit from pachytene is followed by MI by an interval of time that is the same as , or less than , that observed in WT ( Figure 8A , panel v ) .
Our work defines Csm4 as a direct participant in meiotic telomere/NE dynamics , in functional linkage with Ndj1: ( i ) Csm4 is required for telomere dynamics , similarly to and dependent upon Ndj1-mediated telomere/NE association . ( ii ) Csm4 partially colocalizes with telomeres along the NE and , correspondingly , deletion of its putative membrane-spanning domain confers a nearly-null phenotype ( S . Z . and E . A . , unpublished observations ) . ( iii ) Similar phenotypes and strong genetic interactions are observed for csm4Δ and ndj1Δ mutations with respect to recombination , recombination-coupled SC formation , and occurrence of the MI division . The absence of Csm4 does not discernibly alter meiosis-specific association of telomeres with the NE but strongly abrogates rapid zygotene telomere movements ( as well as dynamic telomere-led movements of pachytene chromosomes; [6] ) , and the tendency for telomeres to colocalize in the vicinity of the SPB at zygotene ( the “bouquet” ) . This latter tendency , seen on a population average basis , likely reflects spatial biasing of rapid telomere movements due to the preferential colocalization of actin cables near the SPB [6] . Moreover , the absence of Csm4 places telomeres in an immobile state that can be partially reinvigorated if meiosis-specific telomere/NE association is absent ( with ndj1Δ ) . Telomere-led chromosome movement is dependent upon actin [2] , [5] , [6] . This movement occurs because of association of telomeres to nucleus-hugging cytoplasmic actin cables which are , themselves , dynamic [6] . Thus , an obvious specific basis for the csm4Δ motion defect would be a failure of NE-associated telomeres to become physically and/or functionally coupled to these actin cables . Our work confirms and extends results from analyses of ndj1Δ showing that a mutation ( s ) which affects telomere/NE dynamics also affects meiotic recombination [2] , [11] , [12] , [16] . While it is difficult to be certain that alterations of recombination are a direct consequence of reduced chromosome movement , rather than being a secondary or unrelated effects of altered telomere biology , the current study provides evidence supportive of a direct connection and of a synthetic model for exactly how abrogation of motion might confer such effects . The csm4/ndj1 recombination phenotypes are different from those conferred by most recombination mutants because they involve delays in progression , at multiple steps , through what appears otherwise to be a normal and efficiently executed process . The existence of this phenotype supports the idea that particular factors are required specifically for timing of events rather than execution . A very similar timing phenotype has recently been described for the budding yeast pch2Δ mutant [57] , although this mutation confers delays primarily in pachytene events rather than at immediate post-DSB steps . The effects of pch2Δ are proposed to be mediated via the regulatory signal transduction kinase Mec1/ATR . The same could be true in the present case , with the addition that earlier events might involve both Mec1/ATR and its relative , Tel1/ATM , which is implicated in events immediately following DSB formation [61] . MI delays of csm4Δ ( above ) and ndj1Δ [55] are fully alleviated by elimination of Rad17 , implying alleviation of effects triggered by delays at any and all stages of recombination . In accord with action at multiple stages in the current situation , absence of Rad17 , or one of its collaborators , is known to alleviate MI arrest conferred by defects at diverse stages of recombination: including DSB exit ( dmc1Δ; [56] ) , progression of CO-designated DSBs to later stages ( zip1Δ; [55] , [56] ) and timely progression through pachytene ( pch2Δ; [55] ) . Cytological studies suggest that impediments to completion of presynaptic coalignment can also trigger a local response that includes destabilization of chromosome axes around the affected position ( s ) , e . g . at the site of an interlock in Bombyx [20] or a structural heterozygosity in mouse [67] . Thus , Rad17-dependent progression delays in csm4/ndj1 mutants may be part of a standard “checkpoint damage response” . We note , however , that Mec1/ATR and Tel1/ATM are involved in promoting progression of unperturbed WT meiosis , as well as “checkpoint damage sensing” ( for discussion see [57] ) . The same might well be true of Rad17 and its collaborators , in both WT ( as shown by Grushcow et al . [52] ) and , at least to some extent , in csm4/ndj1 meiosis . Perhaps these components function to “gate” the signal transduction response such that the rate of progression is appropriately sensitive to the status of the entire population of recombinational interactions in a given nucleus rather than proceeding on a more autonomous clock . The ultimate raison d'être of meiotic prophase is the proper segregation of homologs at the MI division . This process , in turn , requires the presence of one or more COs between homologous non-sister chromatids . Correspondingly , MI mis-segregation events are often associated with decreased reciprocal recombination levels [25] , [68]–[73] ) . However , the current work provides three lines of evidence that , surprisingly and contrary to earlier presumptions , homolog nondisjunction in csm4Δ is not attributable to an absence of COs or , more specifically , to absence of the first “obligatory” CO . First , homologs that have nondisjoined in csm4Δ do not exhibit a deficit of COs ( Figure 4 ) . A caveat in our analysis is that we were unable to measure telomere distal crossovers in the strains that displayed nondisjunction . Second , the ndj1Δ and csm4Δ mutations have very similar effects on CO formation while ndj1Δ has a much less severe effect on homolog nondisjunction than csm4Δ; in the double mutant , it further reduces nondisjunction below the csm4Δ level . Third , the primary defect of recombination and downstream events is a temporal delay of a process that eventually proceeded to completion . Nondisjunction events would more likely result from the inefficient execution of a particular process . One interesting possibility is that some of the COs in csm4/ndj1 mutants “fail to ensure disjunction” because of a defect in the relationships between sisters . Indeed , there are hints of abnormal sister relationships from detectable increases in PSSC events in these mutants , as shown previously by Conrad et al . [13] in tetrad analysis . Sister relationships are important in three respects: First , sister chromatid cohesion distal to the site of exchange is vital for the stabilization of the physical manifestations of crossing over , chiasmata , which hold the homologous pair together [74]–[76] . Second , at the sites of crossovers , cohesion must be relaxed in order to allow for exchange of the chromosome arms [77]–[79] . Third , sister cohesion along arms distal to the sites of COs must be released during anaphase I . Thus , the csm4Δ/ndj1Δ defect could be either a deficit of cohesion or , more intriguingly , a failure of cohesion to be properly released either at the site of the CO , along arms distal to the CO site , or specifically at telomeres . The scenario presented above , in which a deficit of motion results in defective immediately post-DSB steps of recombination , could also explain a defect in sister relationships . It has recently been shown that CO-designation at leptotene/zygotene is accompanied by local destabilization of chromosome axes; presumably as the first step in differentiation and separation of sister chromatids specifically at these sites [80] . At a site where an initiating DSB fails to establish a normal recombinosome/axis relationship , CO-designation might still occur with respect to DNA events but without accompanying effects on sister relationships . Alternatively , impeded completion of DSB/partner interactions could trigger a local loss of sister connectedness which extends down the chromosome arm ( s ) . Linkage of all csm4/ndj1 phenotypes to a single common cause is supported by the fact that , for homolog nondisjunction as for other effects , csm4Δ confers a stronger defect than ndj1Δ . On the other hand , Csm4/Ndj1 could be involved directly in sister chromatid cohesion , along arms or in centric regions , as an independent aspect of their molecular functions . Indeed , the third component of yeast telomere/NE dynamics , Mps3 has been implicated as a direct general participant in sister chromatid cohesion , in both mitotic and meiotic cells [13] , [81] . The ndj1Δ csm4Δ double mutant nondisjunction defect is slightly weaker than that of ndj1Δ , rather than being the same as or slightly greater than in ndj1Δ . Thus , for this phenotype , the defect in each single mutant is subtly dependent upon the presence of the WT gene product corresponding to the other mutation ( “partial reciprocal epistasis” ) . In the context of effects on sister cohesion , a possible explanation is that both mutations have two effects , conferring both a reduction in the number of sister chromatid connections and defective release of those connections that do occur . In this case , each mutation would reduce the number of connections and thus , synergistically , the number of aberrant connections remaining to interfere with MI homolog segregation . Explanations for homolog nondisjunction that do not involve sister cohesion can also be envisioned . For example , nondisjunction could simply be an additional consequence of the presence of entanglements , which might affect one or more homolog pairs . Alternatively , homolog nondisjunction may result from an excess of COs ( e . g . [82] ) . Our genetic data argue against such a model for csm4Δ mutants: we observed that the csm4Δ mutation decreased the meiotic viability of msh5Δ and mlh1Δ mutants ( Figure 2 ) and this would not be expected if it resulted in more COs . A third possibility would attribute homolog nondisjunction to an excess of multi-chromatid events resulting from failure to resolve precursor large joint molecules ( e . g . as in sgs1 [36] ) . However , there is no evidence that csm4Δ/ndj1Δ mutants exhibit a sgs1-like defect ( above ) . We construct a coherent model where abrogation of motion confers a defect in completion of early DSB/partner interactions ( partner identification or ensuing creation of bridges between homolog axes ) which , in turn , explains all other observed mutant defects as described above and in earlier studies . Phenotypes of motion-defective mutants in S . pombe have been explained similarly , though in less detail , as a partial defect in recombination and “pairing” [3] , [9] , [19] . However , diverse alternative explanations for some or all of the observed effects are not critically excluded . Future studies must now critically address predictions of this model , for yeast and for other organisms , e . g . assessment of local DSB/partner interactions and occurrence of aberrant topological relationships , on a per-cell basis .
Yeast strains are listed in Table S1 . Strains were grown in either yeast extract-peptone-dextrose ( YPD ) or minimal selective media [83] . Sporulation plates were prepared as described previously [84] . All incubations were performed at 30°c for the experiments presented in Figures 1 , 3 , 4 , 5 , and S4 , Tables 1–3 , and Tables S1 , S2 , S3 and S4 . When required , geneticin ( Invitrogen ) , nourseothricin ( Hans-Knoll Institute fur Naturstoff-Forschung ) , and hygromycin B ( Calbiochem ) were included in YPD media as described [85] , [86] . Plasmids and integrating vectors were introduced into yeast strains using standard methods [87] . The EAY1108 and EAY1112 SK1-congenic strains were described in Argueso et al . [27] and the NH942 and NH943 SK1-isogenic strains were described in de los Santos et al . [46] . The BR4635-8Bα and BR4256-5Ba strains are derivatives of those described in Rockmill et al . [29] . This BR strain set was used because it was specifically designed to measure crossing over on a chromosome that had experienced a nondisjunction event [29] . All diploids homozygous for coding region deletion mutations in CSM4 , NDJ1 , MSH5 , MLH1 , and RAD17 were created by sequential transformation of the parental strains and the mutations were marked with the KANMX4 , NATMX4 , or HPHMX4 as shown in Table S1 [85] , [86] . Details on how the mutations were introduced into these strains are available upon request . CSM4 was mutagenized by overlap PCR [88] to create the single-step integrating plasmid bearing the N-terminal GFP-Csm4 integrating vector pEAI242 . Details on how this plasmid was made are available upon request . pEAI242 was linearized with SacI and SphI prior to transformation . We tested the functionality of the N-terminal GFP-Csm4 construct by integrating it into the EAY1108/EAY1112 background where WT displays 97% spore viability ( n = 1199 tetrads ) and csm4Δ displays 64% spore viability ( n = 1164 ) . The integration strains displayed 92% spore viability ( n = 40 ) , indicating that the GFP-Csm4 fusion is functional . All images were analyzed using ImageJ [97] and/or Metamorph functions . Deconvolution of 2D and 3D acquisitions was performed using AutoDeblur .
|
In meiosis , cells specified to become gametes ( eggs or sperm ) undergo a single round of DNA replication followed by two consecutive chromosomal divisions . In most organisms , the proper segregation of chromosomes at the first meiotic division is mechanically dependent upon genetic exchange , or crossing over , at homologous sites along chromosomes . This process is highly regulated so that every pair of matched chromosomes , regardless of size , receives at least one crossover . In humans , defects in this recombination process can lead to a variety of chromosome aneuploidy syndromes . During early stages in meiosis , the ends of chromosomes , called telomeres , associate with the envelope of the nucleus and undergo highly dynamic movements . We identified a new component of the movement-generating system , Csm4 , in budding yeast . In the absence of Csm4 , the telomeres associate with the nuclear envelope but are locked in an immobile state . In addition , strains lacking Csm4 show delayed recombination progression and high levels of chromosome mis-segregation at the first meiotic division . These findings suggest that , during meiosis , Csm4 is involved in coupling telomere complexes to the movement-generating system and that chromosome motion is important for the completion of early steps in recombination .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"developmental",
"biology/germ",
"cells",
"molecular",
"biology/recombination",
"cell",
"biology/cell",
"growth",
"and",
"division",
"cell",
"biology",
"genetics",
"and",
"genomics/nuclear",
"structure",
"and",
"function",
"molecular",
"biology/chromosome",
"structure",
"genetics",
"and",
"genomics/chromosome",
"biology",
"genetics",
"and",
"genomics/gene",
"function"
] |
2008
|
Csm4, in Collaboration with Ndj1, Mediates Telomere-Led Chromosome Dynamics and Recombination during Yeast Meiosis
|
The primate visual system consists of a ventral stream , specialized for object recognition , and a dorsal visual stream , which is crucial for spatial vision and actions . However , little is known about the interactions and information flow between these two streams . We investigated these interactions within the network processing three-dimensional ( 3D ) object information , comprising both the dorsal and ventral stream . Reversible inactivation of the macaque caudal intraparietal area ( CIP ) during functional magnetic resonance imaging ( fMRI ) reduced fMRI activations in posterior parietal cortex in the dorsal stream and , surprisingly , also in the inferotemporal cortex ( ITC ) in the ventral visual stream . Moreover , CIP inactivation caused a perceptual deficit in a depth-structure categorization task . CIP-microstimulation during fMRI further suggests that CIP projects via posterior parietal areas to the ITC in the ventral stream . To our knowledge , these results provide the first causal evidence for the flow of visual 3D information from the dorsal stream to the ventral stream , and identify CIP as a key area for depth-structure processing . Thus , combining reversible inactivation and electrical microstimulation during fMRI provides a detailed view of the functional interactions between the two visual processing streams .
The primate visual cortex consists of a large number of cortical areas that collaborate to compute neural representations of the external world . In both the dorsal and the ventral visual stream of the macaque monkey , these cortical networks are organized into distinct hierarchies that provide progressively more elaborate representations of specific stimulus features such as motion [1] , color [2] , and disparity [3 , 4]; stimulus categories such as faces [5] and bodies [6]; and actions such as reaching , grasping , and saccadic eye movements [7–11] . To understand the computations at each level within such a hierarchy , it is crucial to know the components of the network , the anatomical connectivity between these regions , and the properties of individual neurons . However , how visual information flows between the different components of the network is frequently unknown . Understanding how visual information flows is particularly challenging when investigating the functional interactions between the dorsal and the ventral visual streams . Lesion studies have demonstrated that each visual processing stream is specialized [12 , 13] and can function—at least to some extent—independently of the other stream [14 , 15] . Both anatomical [16] and functional [17–19] evidence strongly suggest that the dorsal and ventral stream interact during object vision , but no study has been able to show conclusively which information is transferred between the two streams . To compute the depth structure of objects ( e . g . , concave or convex ) , the visual system recruits cortical regions in both the ventral and the dorsal visual stream [20–23] . Here , we first charted the network of cortical areas implicated in three-dimensional ( 3D ) object vision using fMRI in awake monkeys . We then reversibly inactivated the caudal intraparietal area ( CIP ) during fMRI and observed widespread but highly selective changes in fMRI activations in the depth structure network comprising both posterior parietal and inferotemporal cortex ( ITC ) . Moreover , CIP inactivation caused a significant reduction in performance in a depth structure categorization task . Electrical microstimulation of CIP during fMRI mainly activated areas outside the 3D network , indicating that the effects of CIP inactivation may originate as an indirect effect through posterior parietal cortex . To our knowledge , these results provide the first evidence for a causal contribution of a dorsal stream area to fMRI activations in the ventral visual stream .
To obtain an overview of the cortical network that is sensitive to depth structure , we scanned four monkeys in a block design ( 423 runs in total; Monkey M , 113 runs; Monkey R , 127 runs; Monkey S , 94 runs; and Monkey K , 89 runs ) in 21 scanning sessions ( Monkey M , 5; Monkey R , 6; Monkey S , 4; and Monkey K , 6 ) . We calculated the interaction between the factors “curvature” and “disparity” on the group data ( using 89 runs for each monkey ) , by computing the contrast [CS–CC]–[FS–FC] , where CS = curved stereo , CC = curved control , FS = flat stereo and FC = flat control . As in previous studies [20 , 22 , 25] , this contrast aims to identify regions that are activated more strongly by curved stimuli ( compared to their control stimuli ) than by flat stimuli ( compared to their control stimuli ) presented at different disparities . The network of cortical areas sensitive to depth structure was remarkably extensive ( Fig 1A ) . In line with previous studies [17 , 20 , 25] , the anterior lateral bank of the intraparietal sulcus ( IPS ) was more strongly activated by curved surfaces than by flat surfaces at different disparities . However , in contrast to previous investigations [20 , 22] , we also observed additional activations in the caudal IPS region—comprising both the lateral and the medial bank of the caudal IPS . Based on the anatomical locations of depth-structure-related activations ( see inset in Fig 1A ) , this caudal IPS region consisted of area CIP on the lateral bank and the posterior intraparietal area ( PIP ) on the medial bank of the IPS . For illustrative purposes , we calculated the percent signal change ( PSC ) of the curvature x disparity interaction effect along a path drawn through the IPS of the left hemisphere from caudomedially ( area PIP ) to rostrolaterally ( anterior intraparietal area or AIP , Fig 1B ) , for each of the four monkeys separately . Note that highly similar results were obtained for both hemispheres . Fig 1B illustrates that in every animal , both the caudal and the anterior lateral IPS showed significant depth-structure-related activations . In occipital cortex , we measured significant activations related to depth structure in and around the lunate sulcus and the inferior occipital sulcus , corresponding to dorsal and ventral areas V3 , V4 , and parts of V1 and V2 [26] . Surprisingly , a prominent activation pattern related to depth structure sensitivity was also observed in ITC . Both hemispheres showed strong activations in posterior ITC ( which partially corresponds to TEO [26] ) , on the shoulder and in the lower bank of the superior temporal sulcus ( STS ) , and more anteriorly in anterior IT ( AIT ) ; posteriorly near the posterior middle temporal sulcus ( PMTS ) on the temporal convexity , and anteriorly on the shoulder and lower bank of the rostral STS , a region most likely corresponding to the recording area in previous single-cell studies [21 , 28] . We calculated the percent signal change along a path through temporal cortex running from posterior inferotemporal cortex ( PIT ) to AIT ( Fig 1A ) . Each of the four subjects showed significant ( although to varying degrees; compare monkey M to monkey S: one sample t tests , p < 0 . 05; Fig 1C ) depth-structure-related activations in the ITC: a posterior patch corresponding to PIT , and more anterior activations in AIT ( largely corresponding to area TE , Fig 1C ) . Additional , but smaller , activations related to depth structure sensitivity were located in the anterior subsector of ventral premotor cortex ( area F5a , significant in three out of four monkeys [22 , 29] ) and in the prefrontal cortex ( area 46 and possibly area 45 in the left hemisphere ) . To assess the role of area CIP in the network of cortical areas sensitive to depth structure , we reversibly inactivated CIP in one hemisphere while the animals were passively fixating curved and flat surfaces during fMRI . To verify that we had successfully inactivated CIP , we first investigated the effect of muscimol injections on the fMRI activations elicited by curved surfaces in the caudal IPS . Because it was difficult to dissociate the fMRI activations in the medial bank from those in the lateral bank of the caudal IPS , we evaluated the effect of muscimol in a single region of interest ( ROI ) that encompassed both PIP and CIP ( but see S2A Fig for the effects on the medial and lateral bank of the caudal IPS ) . Muscimol injections significantly reduced ( unpaired t test , T ( 304 ) = 2 . 434; p = 0 . 015 ) the fMRI activations evoked by curved surfaces in the caudal IPS of the inactivated hemisphere ( Fig 2A ) . Analysis of Variance ( ANOVA ) on the PSC in the caudal IPS ROI with factors condition and inactivation ( muscimol versus saline ) showed that the main effect of condition was highly significant ( F ( 3 , 1216 ) = 17 . 15; p < 0 . 001 ) , but because muscimol caused fewer deactivations in the control conditions ( curved control and flat control ) , the main effect of inactivation was not significant ( F ( 1 , 1216 ) = 1 . 52; p = 0 . 22 ) , while the interaction between the factors condition and inactivation almost reached significance ( F ( 3 , 1216 ) = 2 . 36; p = 0 . 06 ) . Saline injections did not produce any significant reduction in PSC , since the size of the curvature x disparity effect was comparable between saline sessions ( average PSC over runs = 0 . 4 ) and the first fMRI experiment ( PSC = 0 . 41; T = 0 . 3471 , p = 0 . 73 ) . We plotted the depth-structure interaction effect ( curvature x disparity ) under saline and muscimol conditions ( at p < 0 . 05 FWE corrected for multiple comparisons ) on coronal images of the MRI template ( Fig 2B , group data , green voxels indicate depth structure activations during saline but not during muscimol sessions ) . Muscimol injections in the lateral bank of the caudal IPS reduced the depth-structure-related activations both in the medial and in the lateral bank of the caudal IPS ( ROI defined on the basis of the saline injections; significant reduction of the curvature x disparity interaction effect in the PSC in the caudal IPS ROI , T = 1 . 98; p = 0 . 049 ) . We also calculated the effect of muscimol injection on the depth-structure-related PSC within the caudal IPS region that was inactivated , which was estimated based on the anatomical MRI after the injection of the contrast agent Dotarem ( S1B Fig ) . Restricting the analysis of the effect of muscimol injection to a ROI defined by the sum of all Dotarem ROIs of the three monkeys resulted in a significant effect of inactivation on the PSC of the main effect of disparity ( T[304] = 2 . 8194; p = 0 . 005127 ) . Finally , the path drawn along the IPS illustrated in S2A Fig also shows that CIP inactivation caused a reduction in PSC both in the lateral ( CIP ) and in the medial ( PIP ) bank of the caudal IPS . Note that Fig 2 also shows a very small number of red voxels in the inactivated hemisphere , indicating higher activations following CIP inactivation . These results are , however , most likely due to imperfect warping to the anatomical template . Next , we analyzed the effect of reversible inactivation of CIP on the group data of the anterior IPS ROI , which consisted of the posterior and anterior subsectors of area AIP [17] . Reversible inactivation of area CIP markedly reduced the anterior IPS activation elicited by curved surfaces ( T[304] = 2 . 6288; p = 0 . 009 ) but did not significantly affect the other conditions ( Fig 3A ) . A two-way ANOVA with factors condition and inactivation ( muscimol versus saline ) revealed highly significant main effects of both condition ( F[3 , 1216] = 13 . 91; p < 0 . 001 ) and inactivation ( F[1 , 1216] = 13 . 98; p = 0 . 0002 ) but no significant interaction ( F[3 , 1216] = 0 . 42; p = 0 . 7 ) . Surprisingly , the effect of CIP inactivation was not uniform across the anterior IPS region: reversible CIP inactivation mainly reduced the curvature x disparity interaction effect in the most caudal part of the anterior IPS ROI ( largely corresponding to the posterior subsector of AIP ) , but exerted little influence on the most rostral part of the anterior IPS ROI ( compare left and right panels in Fig 3B; see also S2A Fig , left panel ) . For illustrative purposes , we calculated the PSC in muscimol and saline sessions on a path through the IPS . The decrease in the PSC of the interaction effect reached approximately 0 . 36%—which was comparable to that measured in CIP ( 0 . 4% ) —in the caudal part of the anterior IPS ROI ( see S2A and S3A Figs for individual monkeys ) . Overall , reversible inactivation of area CIP resulted in a significant ( T = 2 . 43; p = 0 . 015 ) reduction in the curvature x disparity interaction effect in the ipsilateral anterior IPS ( Fig 3B; for contralateral effects see also S4 Fig ) . We verified that the effect of CIP inactivation on the anterior IPS was also present in each subject . S3A Fig shows the curvature x disparity interaction effect on the echo planar images ( EPIs ) of the three monkeys during saline ( left panels ) and during muscimol ( right panels ) . In each animal , we measured a highly significant ( monkey R: T[122] = 3 . 4252; p = 0 . 000837; monkey M: T[104] = 3 . 3155; p = 0 . 00126018; monkey S: T[213] = 4 . 4215; p = 1 . 5617e-5 ) reduction in the PSC of the interaction effect in the anterior IPS region of the inactivated ( right ) hemisphere ( PSC values in S4 Fig ) . Thus , reversible inactivation of area CIP caused a robust and reproducible decrease in depth-structure-related fMRI activations in the anterior IPS , indicating a causal contribution from area CIP to the visual activations in the anterior lateral bank of the IPS that were elicited by 3D curved surfaces . Our second objective was to investigate the effect of CIP inactivation on the depth-structure-related activations in the ITC . In saline sessions , both PIT and AIT were significantly activated in all stereo conditions , but more so by curved surfaces ( significant curvature x disparity interaction effect , group data ) . CIP inactivation did not affect the fMRI activations in PIT in any of the conditions ( Fig 4A , left panel ) . However , the ROI of AIT showed a significant reduction in activation in all conditions ( Fig 4A , right panel ) , while the most significant effect of CIP inactivation on AIT ( T[304] = 4 . 171196; p = 0 . 0004 ) was observed in the curved stereo condition . The main effects of inactivation ( F[1 , 1216] = 13 . 61; p = 0 . 0004 ) and stereo condition ( F[3 , 1216] = 12 . 44; p < 0 . 0001 ) were highly significant ( ANOVA ) in the ROI of AIT , whereas the interaction between the factors inactivation and condition almost reached significance ( F[3 , 1216] = 0 . 4; p = 0 . 07 ) . As in the anterior IPS region , CIP inactivation caused a significant ( T = 2 . 714; p = 0 . 007 ) decrease in the curvature x disparity interaction effect in AIT . On the group data of the curvature x disparity interaction effect illustrated in Fig 4B , it is clear that CIP inactivation reduces the depth-structure-related activation in the rostral lower bank of the STS , a part of AIT ( note that Fig 4B only illustrates the depth-structure-related activations in AIT; those of PIT are shown in Fig 3B ) . The magnitude of the effect of CIP inactivation on AIT was surprisingly large compared to that in the IPS areas: CIP inactivation reduced the PSC of the curvature x disparity interaction effect in AIT to 52% of the level measured during saline sessions , compared to 43% for the caudal IPS region and 64% for the anterior IPS region . A two-way ANOVA of the PSC evoked by curved surfaces calculated across runs with factors area ( anterior IPS versus AIT ) and inactivation ( muscimol versus saline ) , and monkey as a nested factor , showed a highly significant main effect of area ( F[1 , 608] = 66 . 87; p < 0 . 0001 ) and inactivation ( F[1 , 608] = 20 . 77; p < 0 . 0001 ) , but no significant interaction ( F[1 , 608] = 0 . 12; p = 0 . 73 ) . As with the IPS region , we calculated the decrease in the PSC along a path through the ITC ( S2A Fig , right panel ) from posterior ( PIT ) to anterior ( AIT ) , and found a reduction in the interaction effect induced by CIP inactivation in AIT , but not in PIT . To demonstrate the consistency of the effect of CIP inactivation on AIT , we plotted the curvature x disparity interaction effect during saline and muscimol sessions on the EPIs of the individual monkeys ( S3B Fig ) . In every animal , we observed a decrease in depth-structure-related fMRI activations in the ipsilateral AIT , which was significant in monkeys M ( T[104] = 2 . 9587; p = 0 . 003825 ) and R ( T[122] = 2 . 1583; p = 0 . 03 ) but not in monkey S ( T[213] = 1 . 84778; p = 0 . 0660; PSC in S4 Fig ) . Note , however , that monkey S showed a significant effect of CIP inactivation in the more anterior part of AIT ( T[213] = 2 . 0265; p = 0 . 043 ) . Thus , although slightly more variable compared to the anterior IPS region , reversible CIP inactivation caused a pronounced decrease in the fMRI activations related to depth structure sensitivity in AIT , implying that CIP contributes to the depth structure sensitivity of the anterior IPS and to that of ventral stream area AIT . Finally , in order to visualize the effects of CIP inactivation on the depth structure network more directly , we subtracted the group average results for the contrast [ ( curved stereo–curved control ) – ( flat stereo–flat control ) ] during muscimol sessions from the same contrast during saline sessions ( S3C Fig ) . We observed significant effects of CIP inactivation in the caudal IPS , in the anterior IPS , and in AIT , consistent with the previous analyses . In the previous group analyses , we combined the data from equal numbers of runs from all three monkeys collected in different sessions . Although the results were consistent across subjects , we observed considerable variability between sessions . In S5 Fig , we plotted the PSC for every individual muscimol and saline session in the three main regions of interest: caudal IPS , anterior IPS , and AIT . We could reproduce the overall effect of CIP inactivation in different sessions in every animal , but remarkably , two out of ten muscimol sessions ( monkey R , session 2 and monkey S , session 3; compare PSC in the current muscimol session with averaged PSC over saline sessions ) did not show any effect of CIP inactivation , not even in the caudal IPS ROI . Technical factors ( e . g . , suboptimal injection volume , lower muscimol activity ) or varying compensatory mechanisms ( see discussion ) may account for this intersession variability . For comparison , we also show the PSC in the same ROIs in the no-injection experiments ( which were scanned months before the muscimol and saline sessions ) . In saline sessions , two of our subjects ( M and S ) showed a significant curvature x disparity interaction effect in area F5a of the ipsilateral hemisphere . Reversible inactivation of CIP significantly reduced the depth-structure-related activations in F5a in these two monkeys ( monkey M: T[104] = 3 . 0105; p = 0 . 003273; monkey S: T[213] = 3 . 2245; p = 0 . 0015 ) . The decrease in the interaction effect in F5a of the third monkey ( R ) was not significant . In view of the anatomical connections between F5a and area AIP [16 , 17] , the effect of CIP inactivation on F5a was in all likelihood a result of its effect on the anterior IPS . However , CIP inactivation did not affect every node of the depth structure network , since we observed no effect of CIP inactivation on curvature x disparity interaction in the occipital ROI comprising ventral and dorsal V2 , V3 , and V4 ( p = 0 . 4847 ) , nor in PIT ( p = 0 . 1145 ) ( see also S3C Fig ) . We also explored whether other regions ( defined on the basis of anatomical ROIs ) outside the depth structure network were significantly affected by reversible inactivation of CIP . Since the main effect of disparity ( contrast all stereo stimuli versus all control stimuli ) produced widespread activations throughout occipital , temporal , parietal , and frontal cortex ( S6 Fig ) , we calculated the effect of CIP inactivation on the main effect of disparity in 21 predefined anatomical ROIs in the right ( inactivated ) hemisphere ( the effect of CIP inactivation on the main effect of disparity along a path in the IPS and in ITC is illustrated in S2B Fig ) . The ROI of area FST , in the fundus of the STS , was the only region outside the higher-order activations showing a significant effect of CIP inactivation on the main effect of stereo ( p < 0 . 01 , Bonferroni corrected for multiple comparisons ) . Moreover , dorsal and ventral area V4 were the only ROIs showing a significant ( p < 0 . 05 , Bonferroni corrected for multiple comparisons ) reduction in visually-driven activations ( contrast all conditions–fixation ) caused by CIP inactivation . ( For an overview of the effects of CIP inactivation on the contralateral hemisphere , see S4 Fig ) . Because electrical microstimulation of the rostral lower bank of the STS , part of AIT , predictably alters the performance of monkeys in a depth structure categorization task [30] , and since CIP inactivation reduced the fMRI activation in the rostral lower bank of the STS , we wondered whether CIP inactivation would also affect psychophysical performance in this task . To that end , we trained two of the monkeys ( S and R ) to categorize depth structure ( convex versus concave ) in random-dot stereograms presented at three positions in depth and at various disparity coherences ( between 10% and 50% coherence ) . After stable psychophysical performance had been achieved , we injected muscimol in area CIP of the left hemisphere of monkey R and the right hemisphere of monkey S during depth structure categorization , in alternating sessions with saline injections ( monkey R ) or no injection ( monkey S ) . At 20% disparity coherence , both monkeys showed significant decreases in percent correct during reversible inactivation of area CIP ( Fig 5 ) . The decrease in depth structure categorization performance elicited by CIP inactivation was small ( monkey S: -4%; monkey R: -2% ) but highly significant in both animals ( monkey S: p = 0 . 001 , monkey R: p = 0 . 007 ) . In addition , monkey S was also significantly impaired at the higher disparity coherences ( 30%: p = 0 . 0001; 50%: p = 0 . 001 ) . Note that the smaller effect of CIP inactivation in monkey R may have been related to the saline controls in this monkey ( instead of no injection , as in monkey S ) . We also plotted the percent concave responses as a function of percent coherence to investigate the presence of a general bias in the animals' responses ( S7A Fig ) . The behavioral data were fitted using logistic regression . In both monkeys , the slope of the fitted psychometric function was significantly less steep in muscimol sessions ( difference in slope for monkey S = 0 . 156 , p = 0 . 005 , and for monkey R = 0 . 108 , p = 0 . 004 ) compared to control sessions , but no response bias was found . Consistent with a previous study [24] , the mean horizontal eye position traces , which indicate commitment of the monkey to a specific choice , began to differ for convex and concave choices at 220 ms ( monkey S ) and 300 ms ( monkey R ) after stimulus onset in the absence of muscimol ( saline or no injection , ROC analysis , S7B Fig ) . The difference in eye positions appeared 20 ms ( monkey S ) and 10 ms ( monkey R ) later when CIP was inactivated ( S7C Fig ) . Similarly , CIP inactivation slightly delayed the response times ( i . e . , the time between stimulus offset and saccade onset ) by 18 ms in monkey S ( two-way nested ANOVA with factors session and saline/muscimol , F = 7 . 99; p = 0 . 02 ) and by 10 ms ( ANOVA , F = 436 . 15; p < 0 . 001 ) in monkey R . Thus , a single unilateral injection of muscimol in area CIP resulted in a significant degradation of performance in a depth structure categorization task . To determine whether CIP is effectively connected to AIT , we electrically stimulated CIP during fMRI in two monkeys ( four sessions/30 runs in monkey M , who was also used in the inactivation experiments , and four sessions/36 runs in monkey D ) . In monkey M , we stimulated at the center of the stereo-activation in the lateral bank of the caudal IPS , whereas in monkey D , we stimulated in a site in the lateral bank of the caudal IPS where we recorded selective responses to planar disparity-defined surfaces ( S8A Fig ) . In both animals , CIP-EM elicited strong activations ( Fig 6A , p < 0 . 05 FWE corrected ) in the IPS areas PIP ( in the medial bank of the caudal IPS ) , VIP , and in the most posterior ( dorsal ) part of LIP , but also in the medial parietal areas V6 and V6A ( Fig 6B , one-tailed t tests on PSC: p <0 . 05 , Bonferroni corrected for multiple comparisons ) . Results for the two monkeys were quite similar ( see S8C Fig for individual results ) . However , no significant fMRI activations were observed in any of the temporal ROIs , nor in parietal area AIP ( Fig 6A ) . We verified the absence of any increase in PSC in the functionally-defined ROIs of PIT and AIT ( Fig 6B , rightmost bars ) . Because CIP-EM strongly activated area PIP in the medial bank of the caudal IPS , we also stimulated area PIP in three monkeys ( R and M from the CIP inactivation experiment: five sessions/40 runs and three sessions/34 runs respectively , and D from the CIP-EM experiment: three sessions/31 runs ) . The location of the stimulation site was determined by the depth-structure activations in the fMRI experiment ( in monkeys M and R ) , and by single-cell selectivity for disparity-defined planar surfaces ( in monkey D and R , example neuron in S8B Fig ) . In parietal cortex , the pattern of activations elicited by PIP-EM was highly similar to that evoked by CIP-EM ( Fig 6C and 6D ) : the anatomical ROIs of areas CIP , VIP , V6 , and V6A were significantly activated by PIP-EM ( one-tailed t tests on PSC: p < 0 . 05 , Bonferroni corrected for multiple comparisons ) . However , in contrast to CIP-EM , we also observed strong activations in area V3A and ( weaker ) activations in areas V3 and V4 . Nonetheless , PIP-EM did not activate the more anteriorly located AIT , nor the anatomical ROI of AIP . We verified the lack of increased PSC in the functionally-defined ROIs of PIT and AIT ( Fig 6D , rightmost bars ) . Thus , the reduction in depth-structure-related activations in AIT and the anterior IPS consequent to CIP inactivation were not paralleled by EM-elicited activations in these regions .
To elucidate how the dorsal and ventral stream interact during 3D object vision , we capitalized on the recruitment of posterior parietal and inferotemporal regions during fMRI when viewing 3D stimuli . We report here that area CIP contributes causally to the 3D object-related fMRI activations in the anterior IPS and , surprisingly , also to those in AIT , one of the end-stages of the ventral visual stream . Moreover , CIP inactivation significantly affected performance in a depth-structure categorization task . In contrast , electrical microstimulation of CIP during fMRI activated neither AIT nor the anterior IPS . To our knowledge , these results represent the first evidence that 3D object information in the dorsal stream affects visual processing in the ventral visual stream . To our knowledge , this is the first study in which fMRI , reversible inactivation , electrical microstimulation , and behavioral measurements were combined to investigate the impact of a cortical area ( CIP ) on other nodes of the network in which it is embedded . EM-fMRI can reveal the effective connectivity of patches of neurons with similar properties within a cortical area [17 , 31 , 32] , but cannot easily determine which information is being transferred between the different nodes of the network . Here , we measured the effect of CIP inactivation on the fMRI activations elicited by different visual stimuli ( curved and flat surfaces ) in posterior parietal and inferotemporal cortex , so that we could infer which aspect of visual processing was affected in these remote areas . Thus , inactivation-fMRI in combination with different visual stimulus conditions can clarify in a much more detailed way how cortical networks operate , above and beyond effective connectivity . In contrast to previous studies [20 , 22] , we observed robust activations related to depth structure sensitivity in the caudal IPS , in area PIT ( partially corresponding to TEO [26] ) and in AIT . The most anterior fMRI activation in AIT was located in the rostral lower bank and shoulder of the STS , in line with previous single-cell studies [21 , 28] . Undoubtedly , technological improvements ( 3T scanner , 8-channel coil , and a 4-fold higher spatial resolution ) have contributed to this result . Moreover , in contrast to the previous studies , all monkeys in the present study were first trained to discriminate depth structure in curved surfaces [24] . Hence , the combination of higher signal-to-noise , higher spatial resolution , and stringent behavioral control most likely explains the discrepancy between our fMRI results and previous studies . It should be noted that our fMRI results identified the activation in PIT as a very likely second—and possibly even more important—input stage for AIT , consistent with previous anatomical studies [33 , 34] . Thus , the depth structure-sensitive part of AIT that subserves depth structure categorization [30] may receive multiple inputs from both dorsal and ventral visual streams . Our results are also relevant for the functional organization of the human visual cortex and human–monkey homologies . Several studies [35–37] have identified a region in the caudal IPS in humans that is activated strongly by stimuli that are known to activate CIP neurons in monkeys ( planar surfaces in depth ) , and [4] showed stronger fMRI activations elicited by near/far disparity compared to zero disparity in the monkey CIP and in the human caudal IPS . Note that the fMRI activations in the human caudal IPS are frequently located more medially in the caudal IPS , whereas the monkey CIP is located in the caudal lateral bank of the IPS . However , in our study , both the lateral ( CIP ) and the medial ( PIP ) bank of the IPS were more activated by curved surfaces than by flat surfaces . Similarly , numerous studies have investigated the putative human homologue of area AIP using fMRI , e . g . , [4 , 18 , 38–41]; for a discussion on human and monkey AIP , see [10] . In addition , it has been proposed [10 , 36 , 42] that the visual object responses in the anterior IPS depend on 3D information represented in CIP , and , to our knowledge , our study provides the first causal evidence that this is indeed true . Finally , the strong depth-structure-related activation in PIT ( partially corresponding to TEO ) on the lip of the STS is a new finding ( previous research showed activations for near–far disparities in a different part of PIT , in the fundus of the STS [3] ) that may be highly relevant to the organization of the human occipitotemporal cortex . Using the same stimuli as in the current study , [40] observed a single ventral stream fMRI activation in the human lateral occipital complex ( LOC ) , which was interpreted as potentially homologous to the monkey AIT , because [22] also observed a single fMRI activation in AIT of the monkey . However , because of the susceptibility artefact ( loss of fMRI signal ) in the human temporal lobe , it was difficult to determine whether more anterior regions in occipitotemporal cortex were also sensitive to depth structure . In the current study , we measured a much more extensive network in the monkey ITC , with significant depth-structure-related fMRI activations in PIT and in AIT . Hence , the LOC activation in humans may be homologous to the monkey PIT . We targeted CIP because of the significant depth-structure-related activation in the caudal IPS in our first fMRI experiment , and because [4] has also reported strong fMRI activations evoked by near–far disparities in CIP . Moreover , electrical microstimulation of the most posterior subsector of AIP ( pAIP ) during fMRI activates both CIP and PIP [17] . Earlier single-cell studies had demonstrated that CIP neurons respond selectively to the 3D orientation of planar surfaces defined by binocular disparity [43 , 44] and other depth cues [45 , 46] and , to some extent , even to curved surfaces [47] . Hence , CIP may be an important—and possibly even the first—processing stage in the dorsal stream for the computation of changes in disparity along a surface , as opposed to absolute or relative disparity , which are represented—at least at the single-cell level ( see [48] ) —at earlier stages in the visual hierarchy such as areas V2 [49] , V3 , and V3A [50] . Several observations argue against an attentional interpretation of our inactivation results . First , if CIP inactivation were to indirectly ( through LIP ) reduce attentional gain in different parts of cortex , this gain change would not be specific to the 3D-shape interaction contrast . Indeed , one would expect lowered activity for all ipsilateral visual activations . However , although we observed widespread activations to disparity ( contrast all stereo versus all control stimuli ) throughout visual cortex , CIP inactivation resulted in limited decreases in disparity-induced activity: the ROI of area FST was the only region ( out of 21 predefined anatomical ROIs in the right/inactivated hemisphere ) outside the areas with higher-order activations showing a significant reduction in disparity activation following CIP inactivation . Furthermore , despite widespread visual activations to stimuli in the entire stimulus set ( contrast all conditions-fixation ) , CIP inactivation only caused reductions in visually-driven activity in area V4 . Thus , CIP inactivation did not result in global ( ipsilateral ) gain reductions in visual cortex . Instead , CIP inactivation caused highly specific , 3D-structure-related , activity reductions . Given that 3D-structure-related areas contribute to 3D-structure discrimination performance ( e . g . , [30] ) , it seems likely that these targeted 3D-structure-related activity decreases had an adverse effect on the monkeys' performance during 3D-structure categorization , as supported by our data . Finally , even though we cannot entirely exclude an attentional interpretation , our main conclusion remains that reversible inactivation of a visual area in the dorsal stream ( CIP ) affects visual activations in the ventral stream . The effect of reversible inactivation of CIP on remote areas could be either direct ( through direct cortico-cortical connections ) or indirect ( through other areas ) . The clearest example of an indirect effect was the decrease in fMRI activations in F5a after CIP inactivation , which must have been a result of its effect on AIP , with which F5a is connected and shares many properties [17 , 51 , 52] . Previous studies have shown that EM-fMRI furnishes an almost exact replica of the anatomical connectivity of an area [17 , 31 , 32] . Here , CIP-EM did not activate AIT nor any other ventral stream area , indicating that the effect of CIP inactivation on AIT was most likely indirect . This assertion is supported by anatomical connectivity studies , which have revealed that CIP connects primarily with other areas in the IPS such as LIP and ( to a lesser extent ) AIP [53] , and only very weakly with AIT [54] . Since depth-structure-selective patches in pAIP are effectively connected with AIT [17] , the reduction in fMRI activations in AIT caused by CIP inactivation was most likely established by means of its effect on AIP . These results clearly illustrate that EM-fMRI complements inactivation-fMRI when interpreting direct and indirect effects of reversible inactivation . Although caution is warranted ( since the absence of an EM-induced fMRI activation does not rule out the presence of a direct anatomical connection ) , the absence of anterior IPS activations during CIP-EM suggests that the CIP-inactivation effects on the anterior IPS were also indirect . To determine how CIP may have affected pAIP , we mapped the CIP-EM-induced activations onto the depth-structure-related activations of monkey M , which was the only animal involved in both the CIP-EM experiments and the stereo fMRI experiments . The overlap between these two types of activations in both dorsal LIP and VIP ( Fig 7A ) suggests that the CIP-inactivation effect on the anterior IPS was mediated through ( dorsal ) LIP and/or VIP . Moreover , the depth-structure activations at these anterior–posterior levels in the IPS were significantly reduced after CIP inactivation ( S9 Fig ) . These results are in agreement with a previous anatomical tracer study [55] showing that CIP is connected to area AIP through a series of projections along the lateral bank of the IPS . We did not activate pAIP when stimulating CIP , although weak direct CIP–AIP connections have been described [16 , 55] . Hence , EM-fMRI may primarily reveal the strongest anatomical connections of a given cortical area . Alternatively , our EM-fMRI experiments were more specific than anatomical tracer studies , since we selectively targeted patches within CIP selective for higher-order disparity . Since microstimulation in depth-structure-selective patches in pAIP activated both CIP and PIP [17] , the possibility exists that , even at this level of the visual hierarchy , cortico-cortical connections may be unidirectional , in line with recent anatomical studies [54] . The EM-fMRI experiments clearly demonstrated that CIP and PIP are strongly interconnected , but the pattern of PIP connectivity outside parietal cortex differed markedly from that of CIP . Only PIP-EM activated the early visual areas V2 , V3 , and V3A . Moreover , the CIP and PIP EM-fMRI experiments have revealed , to our knowledge , for the first time how the depth structure network , which is confined to the dorsolateral and ventral visual stream , is connected to areas belonging to the dorsomedial stream [56–58]: both CIP and PIP-EM elicited strong activations in the anatomical ROIs of areas V6 and V6A . Recent studies suggest that V6A neurons respond selectively to different grip types and even to different objects [59 , 60] , similar to neurons in AIP [61] . Hence , our data also reveal the pivotal role played by CIP , where the dorsolateral and dorsomedial pathways intersect when processing objects for grasping . We demonstrated the causal effect of CIP inactivation upon the anterior IPS region and AIT by means of fMRI , which represents an indirect measure of neural activity based on changes in blood oxygenation [62 , 63] . Hence , our results cannot determine what the effect of CIP inactivation may be on the neural responses in these higher-order areas . In theory , CIP inactivation may merely evoke subthreshold modulations in its target areas ( possibly measurable at the level of the local field potential ) without affecting the neuronal firing rate or selectivity . However , the perceptual effect of CIP inactivation on depth structure categorization must presumably have altered the firing rates of neurons at some distance from the injection site . Since electrical microstimulation of AIT neurons has a profound and predictable effect on depth structure categorization [30] , the most likely route through which CIP inactivation influenced depth structure categorization is through AIT . It is important to emphasize that a number of single-cell studies have already characterized the depth structure selectivity of individual neurons in almost all nodes of this network in great detail: in AIT [21 , 24 , 28 , 30 , 64] , AIP [23 , 25 , 65] , F5a [29 , 52] , and CIP [47] . Despite the inherent assumptions associated with fMRI [25] , these single-cell studies provide strong support for the notion that the curvature x disparity interaction contrast we used is effectively an fMRI indicator of higher-order disparity sensitivity at the neuronal level . The small but highly significant effect of CIP inactivation on accuracy in a depth-structure categorization task was similar in magnitude to that reported in a recent inactivation study in the ITC [66] . Several factors may explain why we did not observe larger effects in this task . We used a fixed-duration paradigm with a relatively long stimulus duration ( 800 ms ) , whereas previous studies [30 , 67] have reported reaction times of less than 300 ms in a reaction-time version of the same task . Hence , our long stimulus duration may have allowed time for additional processing of the stimulus . In addition , we observed residual depth-structure-related activations in the caudal IPS after muscimol injection , indicating that we may have only partially inactivated CIP . Note also that a previous CIP inactivation study observed a significant deficit in the discrimination of surface orientation in only three out of six injections [45] . More importantly , CIP inactivation did not significantly alter the activations in PIT , nor in earlier visual areas such as V3 and V4 . Thus , our muscimol injection in CIP may have affected only a small portion of the extensive depth structure network , such that other areas ( possibly in the contralateral hemisphere as well ) remained functional . A previous human fMRI study [68] showed that parietal activations correlate negatively with disparity coherence and psychophysical performance during 3D shape judgments . Hence , especially at lower disparity coherences , compensatory mechanisms may emerge that can obscure the effect of reversible inactivation . To our knowledge , our study is the first systematic attempt to dissect the neural circuitry subserving 3D object vision by means of an integrated approach combining fMRI , reversible inactivation , electrical microstimulation , and single-cell recordings . Fig 7B and 7C summarizes the main findings of this and a previous EM-fMRI study [17] . Despite the extreme interconnectedness of the cortical circuitry , it becomes possible to discern how 3D object information may be transmitted throughout the dorsolateral stream , from the relatively early stages in areas CIP and PIP , through LIP , pAIP , and aAIP to the motor system , and where these areas interact with the dorsomedial and the ventral stream . Thus , progress in our understanding of the neural basis of 3D object vision provides a glimpse into the intricate organization of , and functional interactions within , large-scale cortical networks throughout visual cortex , including their target areas in the motor system .
In total , five male rhesus monkeys ( monkey K: 4 kg , 4 y old; monkey M , 6 kg , 5 y old; Monkey R: 5 kg , 6 y old; Monkey S: 4 kg , 5 y old; Monkey D: 7 kg , 11 y old ) participated in the present experiments . Animal housing and handling were in accordance with the recommendations of the Weatherall report , allowing locomotor behavior , social interactions , and foraging . All animals were pair- or group-housed ( two to four animals per group; cage size at least 16–32 m3 ) with cage enrichment ( toys , swings , foraging devices ) at the primate facility of the KU Leuven Medical School . The natural light/dark cycle was followed , and all experiments were performed during daytime ( between 8 AM and 8 PM ) . Animals were fed daily with standard primate chow supplemented with bread , nuts , raisins , prunes , and fruits . The animals received their daily water supply during training and experiments , or ad libitum in the cages in between training or experimental periods . Monkeys K , M , R , and S were previously not used in other studies; monkey D was enrolled in other electrophysiological studies prior to this one . Animal care and experimental procedures complied with the national and European guidelines ( Directive 2010/63/EU ) and were approved by the Ethical Committee of the KU Leuven ( project number P063/2010 ) . No animals were sacrificed for this study , and except for monkey M , all animals are currently enrolled in other experiments . The implant of monkey M has been removed to prepare the animal for retirement . Each monkey was implanted with an MRI-compatible head fixation post on the skull using ceramic screws and dental acrylic using isoflurane anesthesia and under sterile conditions . Six w after surgery , the monkeys were trained to maintain fixation upon a spot inside a 1 . 5° , electronically-defined window while 3D stimuli were presented foveally . After adequate performance in passive fixation had been reached , all animals were trained in a depth structure categorization task [16] . The stimulus set consisted of curved and flat random-dot stereograms in which depth was defined by horizontal disparity ( dot size 0 . 08° , dot density 50% , vertical size 5 . 5° ) , presented on a grey background . All stimuli were generated using Matlab ( R2010a , MathWorks ) and were gamma-corrected . Similar to [20] and [22] , we used a 2-by-2 design ( S1A Fig ) with factors curvature ( curved versus flat ) and disparity ( stereo versus control ) . In the inactivation-psychophysics experiments and during training , dichoptic presentation of the stimuli was achieved by means of a double pair of ferroelectric liquid crystal shutters ( DisplayTech ) operating at 60 Hz each . The shutters opened and closed in synchrony with the vertical retrace of the display monitor ( VRG , P46 phosphor , vertical refresh rate is 120Hz ) . There was no measurable cross-talk between the two eyes [23] . The position of the right eye was recorded by means of an infrared-based camera system sampling at 500 Hz ( EyeLink 1000 , SR Research ) . Simultaneously with the stimulus presentation , a bright square at the right bottom of the display ( invisible to the monkey ) was presented , detected , and registered as photocell pulses by a photodiode attached to the display . The photocell pulses and the recorded eye movements were digitized and processed at 20 kHz on a single digital processor ( DSP , C6000 series , Texas Instruments ) . In the fMRI experiments , red/green versions of the stimuli were used , and red/green stereoglasses were placed before the eyes to provide dichoptic presentation of the stimuli . The stereo-curved condition consisted of three types of smoothly curved depth profiles ( 1 , ½ , or ¼ vertical sinusoidal cycle ) together with their antiphase counterparts obtained by interchanging the monocular images of the two eyes ( disparity amplitude within the surface: 0 . 5° ) . Four different circumference shapes ( example in S1 Fig ) were combined with each of these six depth profiles , and each combination could appear at two different positions in depth ( mean disparity +/– 0 . 5° ) , creating a set of 48 curved surfaces . In the stereo-flat condition , flat surfaces ( using the same four circumference shapes ) were presented at 12 different positions in depth , such that the disparity content was identical to that in the stereo curved condition . Finally , in the control conditions , we presented one of the monocular images ( either belonging to one of the stereo curved stimuli of to one of the stereo-flat stimuli ) to both eyes simultaneously ( curved-control and flat-control ) . These control conditions consisted of the same binocular input as the stereo conditions , since they contained exactly the same monocular images as the corresponding stereo condition . In the inactivation–psychophysics experiments , the stimuli were double-curved ( along the horizontal and vertical axes ) convex and concave surfaces with a circular circumference shape , and were presented with different disparity coherences ( 10%–50% ) , as in [24] . Monkeys had to indicate the perceived depth structure of a random-dot stereogram ( presented at the fixation point for a fixed duration of 800 ms ) by means of an eye movement ( to the left for concave or to the right for convex ) in order to obtain a liquid reward . The depth structure of the random-dot stereogram was solely defined by horizontal disparity as a two-dimensional radially-based Gaussian surface that could be either convex or concave . In order to encourage discrimination of the 3D-profile rather than absolute disparity , the 3D stimuli were presented at three different positions in depth ( near , at the fixation plane , and far ) . Since the disparity gradient was present only along the surface of the stimulus and not at the boundary [65] , monocular cues were absent . Task difficulty was manipulated by varying the percentage of dots defining the surface , i . e . , the disparity coherence . A random disparity ( drawn from a uniform distribution [-0 . 5 degrees , 0 . 5 degrees] ) was applied to dots that were not enclosed in the surface . All stimuli contained the same number of dots , regardless of the disparity coherence or position in depth of the 3D-shape . The contours of all 3D stimuli were circular . Monkeys K and M were trained until they reached a performance of 80% correct for a 100% disparity coherence stimulus ( typically after 3 to 4 w of training ) , whereas monkeys R and S were trained with disparity coherences between 100% and 10% until they performed above chance level at the 10% disparity coherence stimuli . Four monkeys ( M , K , S , and R . ) were trained to sit in a sphinx position in a plastic MRI-compatible monkey chair positioned in the horizontal bore of the magnet . The stimuli were rear-projected from a Barco 6300 LCD projector onto a translucent screen in front of the monkey at a 57 cm distance . Red/green stereoglasses were placed before the eyes to provide dichoptic presentations of the stimuli . A pupil corneal reflection tracking system ( Iscan ) was used to monitor the position of the eye at 120 Hz during scanning . Maintenance of fixation within the 1 . 5° electronically-defined window around the fixation point was required in order to be rewarded . The inter-reward-interval was systematically decreased ( from 2 , 000 to 900 ms ) when the monkey maintained fixation within the window , to encourage long , uninterrupted sequences of fixation . Prior to the scanning sessions , a contrast agent ( monocrystalline iron oxide nanoparticle [MION]; Feraheme , AMAG pharmaceuticals ) was injected to enhance the signal-to-noise ratio [69 , 70] and spatial selectivity of the MR signal [71] . BOLD signals depend on a combination of cerebral blood volume ( CBV ) , blood flow , and oxygen extraction , whereas MION measurements depend solely on CBV . Since an increase in brain activation produces a decrease in MR signal in MION CBV maps , the polarity of all signal-change values was inverted to account for differences between MION CBV and BOLD activation maps . A radial transmit-only surface coil and custom-built eight-channel phased-array receive coils were positioned closely around the monkey’s head . Functional images were acquired with a 3 . 0 Tesla full body scanner ( TIM Trio , Siemens ) , using a gradient-echo single-shot T2*-weighted echo-planar imaging sequence ( 40 horizontal slices , TR = 2 s , TE = 17 ms , 1 . 25 mm isotropic ) . Data were acquired using a block design whereby each block ( or condition , each 24 s long ) consisted of 12 functional volumes embedded in a time series of 222 volumes ( 444 s ) . Stimulus frequency was 1 Hz , and each condition ( block ) was repeated twice within each time series . The presentation order of the conditions was pseudo-randomized . Voxels showing a significant interaction between the factors curvature and disparity , i . e . , where ( curved-stereo versus curved-control ) was greater than ( flat-stereo versus flat-control ) , were deemed to be sensitive to the depth structure of surfaces [20] . After the first fMRI-experiments , a recording chamber was implanted vertically above the right CIP , under isoflurane anesthesia and aseptic conditions . The position of the recording chamber was centered on the local maximum of the fMRI activation in the posterior part of the IPS at Horsley-Clarke coordinates between 4 and 8P , and between 8 and 12L for the three monkeys . To target the activated voxels in area CIP in the lateral bank of the posterior IPS showing a significant interaction between curvature and disparity , we first obtained an anatomical MRI ( resolution 0 . 6 mm isotropic ) using glass capillaries filled with a 2% copper sulfate solution inserted into several positions of a standard recording grid ( Crist Instruments , spacing 1 mm ) placed inside the recording chamber . The fMRI activation maps were then warped onto this MRI template . After determining the anterior–posterior and medial–lateral position in the grid , we calculated the estimated depth of the injection sites and verified this by means of an anatomical MRI that was obtained after an injection of the contrast-agent Dotarem ( 4 μl of a 2% solution ) using a 10 μl Hamilton syringe , as shown in S1B Fig . To minimize damage to the cortex of the lateral bank of the IPS , we inserted the tip of the needle until reaching the transition between white and gray matter , so that the solution could spread to the neighboring cortex . At least one week later , we started the inactivation-fMRI experiments . In each muscimol session , 4 μl of the GABA-A agonist muscimol ( Sigma , 10 mg/ml ) was injected with a 10 μl Hamilton syringe connected to a 33-gauge stainless needle . The injection of muscimol always took place immediately preceding ( less than 30 min ) the fMRI measurements and outside the scanner . The syringes were inserted into a stainless-steel guiding tube that was placed inside the grid . The volume of the micro-infusions at each site was slowly delivered in small steps of 1 μl every minute to avoid pressure damage . Each muscimol scanning session alternated with a control saline scanning session , in which an equal amount of saline was injected at the same site prior to the scanning sessions . Saline sessions were separated from muscimol sessions by at least 24 h . After the injection of either muscimol or saline , we ran the standard fMRI experiment as described earlier . We collected a total of 219 muscimol runs and 227 saline runs , in 10 sessions each . The numbers of runs were equalized for saline and muscimol sessions within each monkey ( monkey M , 51 runs; monkey S , 107 runs; monkey R , 61 runs ) , and over monkeys for the group analysis by excluding runs randomly , yielding 51 runs per monkey for saline and 51 runs per monkey for muscimol sessions . We followed the same procedures when testing the effect of reversible inactivation of CIP on depth structure categorization . However , in these behavioral experiments , four disparity coherences were used ( 10% , 20% , 30% , and 50% coherence , convex and concave surfaces ) , and muscimol sessions were interleaved with two control sessions without ( monkey S ) or with ( monkey R ) saline injections . All procedures have been described elsewhere [17] . Briefly , in every EM-fMRI session ( monkeys M , D , and R ) , a Platinum/Iridium electrode ( impedance 50–200 kΩ in situ , FHC , Bowdoinham , ME ) was inserted in the grid through glass capillaries serving as guide tubes ( FHC , Bowdoinham , ME ) . A platinum wire served as ground . To verify the stimulation positions , structural MR images ( 0 . 6 mm resolution ) were acquired in every scan session ( prior to the start of the fMRI experiment ) while the electrode was located at the exact stimulation site inside a standard recording grid ( Crist Instruments , Hagerstown , MD; S8 Fig ) . During EM-fMRI sessions , all animals were sedated using a 0 . 25/0 . 5 cc mixture of ketamine ( Nimatek ) and medetomidine ( Domitor ) ( administered every 45 min ) . The animals were video-monitored during sedation , and body temperature was maintained using a heating pad . Note that we recently demonstrated highly comparable EM-fMRI effects in sedated and awake states in area AIP [17] . Using sedated animals , we can exclude any possible influence of attentional state . In every EM-fMRI session , the EM signal was produced using an eight-channel digital stimulator ( DS8000 , World Precision Instruments ) in combination with a current isolator ( DLS100 , World Precision Instruments ) . We switched between stimulation and no-stimulation blocks ( each lasting 40 s ) , for a total duration of 480 s . During stimulation blocks , a single EM train was applied ( on average ) every 3 s . Stimulation trains lasted 250 ms and were composed of biphasic square-wave pulses ( repetition rate 200 Hz; amplitude 1 mA ) . Each EM pulse consisted of 190 μs of positive and 190 μs of negative voltage , with 0 . 1 ms between the two phases ( total duration: 0 . 48 ms ) . The timing of the EM pulses during the fMRI experiment was computer-controlled . Correction for body-motion artifacts was performed with an off-line SENSE ( sensitivity encoding ) reconstruction of the images [72] . Data were analyzed using statistical parametric mapping ( SPM5 ) and BrainMatch software , using a fixed-effects GLM . Spatial preprocessing consisted of realignment and rigid coregistration with a template anatomy ( M12 ) [73] . To compensate for echo-planar distortions in the images as well as inter-individual anatomical differences , the functional images were warped to the template anatomy ( M12 ) using non-rigid matching BrainMatch software . The algorithm computes a dense deformation field by the composition of small displacements , minimizing a local correlation criterion . Regularization of the deformation field is obtained by low-pass filtering . The functional volumes were then resliced to 1 mm3 isotropic and smoothed with an isotropic Gaussian kernel ( full width at half maximum: 1 . 5 mm ) . To avoid higher-order distortion , a nonrigid slice-by-slice distortion correction was applied to fit a fixed-effect general linear model ( GLM ) . We used a fixed-effects analysis ( as is common in monkey fMRI experiments [74 , 75] ) because our sample size was too small for effective power in a random effects analysis . SPM5 was used to perform a voxel-based analysis as previously described [31] . To improve the GLM , six motion-realignment parameters and two eye-movement parameters were added as covariates of no interest . Eye traces were thresholded within the 1 . 5° ( horizontally ) x 2° ( vertically ) window , convolved with the MION response function and subsampled to the TR of 2 s . The functional volumes were resliced to 1 mm3 isotropic and smoothed with an isotropic Gaussian kernel ( FWHM: 1 . 5 mm ) . For illustrative purposes ( Fig 1A ) , the SPM5 activation maps of the group analysis were plotted on flattened or coronal representations of the M12 anatomical template , using Caret software ( version 5 . 64; http://brainvis . wustl . edu/wiki/index . php/Caret:About ) . Individual activation maps were plotted on the coronal slices of an averaged EPI image of each monkey , to minimize warping errors . In a manner analogous to previous studies [69 , 70 , 76 , 77] , depth structure sensitivity was defined by a significant interaction ( p < 0 . 05 FWE rate corrected , FWE for multiple comparisons on the entire brain ) between the factors curvature ( curved or flat ) and disparity ( present or absent ) in the block design . Group data were analyzed as fixed effects with an equal number of volumes per monkey , supplemented with single-subject analysis . To illustrate depth-structure sensitivity along the IPS and along the ITC in individual animals , we defined two paths along the IPS and along ITC , and then plotted the percent signal change of the curvature x disparity interaction effect in each point along the path [20 , 22] . The path was determined manually on the group data to include , insofar as possible , all activations related to depth structure . Therefore , one point was marked in coronal sections every millimeter , in the center of the activations , along the IPS or STS . For the analysis of the effects of CIP inactivation on the curvature x disparity interaction in each individual animal , we determined the borders of the ROIs based on the activated clusters in the interaction-contrast T-map of the control saline scanning sessions for each animal ( on the average EPI ) , at uncorrected level for monkey M and R ( p < 0 . 001 ) , and at corrected level for monkey S ( FWE , p < 0 . 05 ) . As in the group analysis , we identified two regions in the IPS ( anterior IPS and CIP/PIP ) and two regions in the superior temporal sulcus ( PIT and AIT ) . Unpaired t tests ( p = 0 . 05 Bonferroni corrected ) were then applied to each ROI using data from individual runs of muscimol and saline sessions . Regions of interest were functionally defined based on the clusters of contiguous voxels that were significantly ( FWE , p < 0 . 05 ) activated in the contrast [CS–CC]–[FS–FC] ( in which CS = curved stereo , CC = curved control , FS = flat stereo , FC = flat control ) in the group data ( four monkeys ) of the standard fMRI experiment ( without saline or muscimol ) . Based on these data , we defined four ROIs for which we wanted to assess the effect of reversible inactivation of area CIP: two regions in the intraparietal sulcus ( anterior IPS and CIP/PIP ) and two regions in the inferior temporal cortex ( PIT and AIT ) . The functionally-defined ROI of PIT was located on the shoulder of the lower bank of the STS , largely corresponding to OTd ( average within a sphere of 2 mm diameter [20] ) . The functionally-defined ROI of AIT was located more anteriorly in area TE , the most anterior part of the ITC . Our AIT ROI overlapped somewhat with PITv of [26] in its most caudal portion , but because the activations were mostly located in area TE , on the temporal convexity and more anteriorly in the lower bank of the STS , we considered all activated voxels in this ROI as belonging to AIT . We calculated the interaction effect between the factors curvature ( curved versus flat ) and stereo ( disparity versus control ) on the percent signal changes within all activated voxels of the saline sessions in these ROIs with the MarsBaR ROI toolbox for SPM , for muscimol sessions and saline control sessions independently . We then performed two-tailed t tests ( p = 0 . 05 Bonferroni corrected for multiple comparisons ) on the data of individual runs to evaluate the effect of CIP inactivation on our four ROIs . In these same four ROIs , we also calculated the main effect of disparity using the contrast ( curved stereo + flat stereo ) – ( curved control + flat control ) , and used two-tailed t tests across runs to assess the effect of CIP inactivation . Anatomical ROIs ( V1v , V1d , V2v , V2d , V3v , V3d , V3a , V4v , V4d , 45a , 45b , F5a , F5p , FST , LST , MSTd , MSTv , MT , UB1 , UB2 , STP ) were defined on a Caret Atlas based on previous studies [26] .
|
Many different areas of the brain are utilized to analyze the three-dimensional shape of the objects we see and grasp ( e . g . , a sphere compared to a disk ) . In this study , we temporarily inactivated one of these areas in the monkey parietal cortex and measured the effect on the network of brain areas involved in three-dimensional shape processing using functional brain imaging . Surprisingly , reversible inactivation of a parietal area not only reduced the functional activations in other parietal areas , but also in distantly separate brain areas such as the inferotemporal cortex . Moreover , parietal inactivation also caused a significant perceptual deficit in a depth-structure discrimination task . Electrical microstimulation of the inactivation site revealed that almost all of the observed effects were indirect , i . e . , arising through a different brain area . Thus , a combination of functional brain imaging , causal perturbation methods , and behavioral measurements in monkeys can elucidate the flow of three-dimensional object information in the primate brain .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[
"medicine",
"and",
"health",
"sciences",
"diagnostic",
"radiology",
"functional",
"magnetic",
"resonance",
"imaging",
"brain",
"vertebrates",
"social",
"sciences",
"neuroscience",
"animals",
"mammals",
"cerebral",
"hemispheres",
"magnetic",
"resonance",
"imaging",
"primates",
"right",
"hemisphere",
"brain",
"mapping",
"vision",
"animal",
"management",
"eyes",
"animal",
"performance",
"neuroimaging",
"research",
"and",
"analysis",
"methods",
"imaging",
"techniques",
"monkeys",
"animal",
"cells",
"head",
"agriculture",
"psychology",
"radiology",
"and",
"imaging",
"cellular",
"neuroscience",
"diagnostic",
"medicine",
"anatomy",
"cell",
"biology",
"neurons",
"biology",
"and",
"life",
"sciences",
"ocular",
"system",
"sensory",
"perception",
"cellular",
"types",
"amniotes",
"organisms"
] |
2016
|
Posterior Parietal Cortex Drives Inferotemporal Activations During Three-Dimensional Object Vision
|
Cascading activity is commonly found in complex systems with directed interactions such as metabolic networks , neuronal networks , or disease spreading in social networks . Substantial insight into a system's organization can be obtained by reconstructing the underlying functional network architecture from the observed activity cascades . Here we focus on Bayesian approaches and reduce their computational demands by introducing the Iterative Bayesian ( IB ) and Posterior Weighted Averaging ( PWA ) methods . We introduce a special case of PWA , cast in nonparametric form , which we call the normalized count ( NC ) algorithm . NC efficiently reconstructs random and small-world functional network topologies and architectures from subcritical , critical , and supercritical cascading dynamics and yields significant improvements over commonly used correlation methods . With experimental data , NC identified a functional and structural small-world topology and its corresponding traffic in cortical networks with neuronal avalanche dynamics .
Cascade-like dynamics is characterized by the succession of events , or processes , that are causally related , and is frequently encountered in many complex systems ( networks ) across disciplines . For example , single cells in living organisms maintain metabolic , protein and gene-interaction networks with mostly unidirectional signaling cascades in which nodes represent metabolites , proteins and genes respectively [1]–[3] . At the next higher level of cell to cell interactions such as the brain , pyramidal neurons in the cortex connect with thousands of other neurons [4] thereby supporting cascades of neuronal activity in the form of waves [5] , neuronal avalanches [6] and synfire chains [7] , [8] . Cascade-like dynamics also occurs in many social networks such as the spread of epidemics [9] and gossip [10] in human networks as well as human travel itself [11] . This cascading dynamics carries the signature of the underlying statistical interdependencies between the interacting nodes , which are summarized by the functional network topology , represented by adjacency matrix indicating whether two nodes interact or not , and architecture [12] , represented by a weighted graph which additionally indicates the magnitude of each interaction . The relationship between the cascading dynamics and the functional network is often poorly understood , even though reconstructing the network from the observed dynamics can provide crucial insights into the causal interactions between the nodes as well as the overall functioning of a complex system [13] . Of similar challenge remains the problem of how the functional architecture relates back to the structural organization of a network , that is to its physical nodes and physical connections between nodes [14] . While very similar dynamics can arise from fundamentally different network structures , e . g . for small neuronal networks with diverse elements [15] , for large networks such as the human cortex the global brain dynamics has been shown to reflect fairly accurately the underlying structural connectivity , i . e . cortex anatomy [16] , [17] . It is therefore critical to identify new approaches that provide insight into the functional and structural organization of a network based on the observed dynamics . Correlations in the dynamics between nodes have been successfully used to identify functional links in relatively large networks such as obtained from MEG or fMRI recordings of brain activity ( e . g . [18]–[20] ) . A pure correlation approach , however , is prone to induce false connectivities . For example , it will introduce a link between two un-connected nodes , if their activities are driven by common inputs [21] , [22] . More elaborate approaches such as Granger Causality [23] , partial Granger Causality [24] , partial directed coherence ( for a review see [25] ) , and transfer entropy [26] partially cope with the problem of common input , however , these methods require extensive data manipulations and data transformations and have been mainly employed for small networks [27] , [28] . Here , we propose a new method that efficiently reconstructs the functional architecture of a network from the dynamics . In the theoretical part of the manuscript , we first introduce two different Bayesian approaches to reconstruct the network topology from the observed cascades: ( 1 ) the Iterative Bayesian ( IB ) , and ( 2 ) the Posterior Weighted Averaging ( PWA ) with equal link priors . We then use PWA to derive the Normalized Count ( NC ) approach , a simple and efficient nonparametric algorithm that requires very little knowledge about the dynamical rules underlying activity cascades . We show that the NC , which is a hybrid between a Bayesian approach and a correlation method , performs almost as well as the IB when the exact probabilistic rules of the dynamics known . Using simulations , we demonstrate the utility of these algorithms for reconstructing random , small-world and scale-free network architectures from activity cascades modeled by subcritical , critical , and supercritical branching processes . We apply our approach to neuronal avalanches , which are the activity cascades in the brain . It has been shown [6] , [29] , [30] that they spontaneously emerge in superficial layers of cortex , both in vitro ( acute slices and slice cultures ) [29]–[32] and in vivo [33] . They have also been demonstrated recently in the spike activity of dissociated cortex cultures [34] , [35] . The network architecture that gives rise to neuronal avalanches is currently not known , although neuronal avalanches have been simulated in networks with scale-free [36] , [37] , fully connected [38] , random [39] , and nearest-neighbors [37] , [40] topologies . Here we demonstrate a small-world functional topology of neuronal group formation in neuronal avalanches .
For each pair of nodes , we can determine some scalar measure of connectivity . For example , these can be node to node correlations , or the FC approach ( Equation 19 ) , or the NC approach using Ultimately , we are trying to use these estimates as a measure of directed influence or causal traffic for each link in the underlying network . However , these measures will also include a contribution from non-causal correlations arising when pairs of nodes are active close in time but had a common ancestor at some prior time during the cascade , or share common inputs directly . We thus have to determine the statistical significance for each of the scalar connectivity estimates . The null-model is obtained by randomizing the recorded activity cascades using constrained pairwise shuffling . In this randomization procedure , the times of two randomly selected , active nodes will be switched , such that the node active at time , will be assigned time and vice versa . This shuffling method is straightforward to implement for continuous time events , in which case the time interval distribution will be preserved . For binned data , one will encounter situations where the time bin already has node active , and vice versa , in which case the shuffle is aborted and a new pair of nodes is sought . Shuffling in this way preserves the average activity at each node as well as the occupation of time bins with active nodes and thus the dynamical regime of the underlying branching process ( see Results ) . To obtain the resampled dataset , the pairwise switching is repeated times , being comparable to the total number of active nodes in the dataset . By repeating this procedure , resampled datasets are obtained , each with its corresponding estimate . We use the distribution of the to determine the threshold value for the given significance level . The number of shuffled replicates used to obtain the connectivity estimate at a significance level , where is the “over-shuffling” factor , usually 5 or 10 . We obtain the topology , i . e . , the adjacency matrix of the estimated network at the significance level as ( 20 ) and the architecture as ( 21 ) Hence the reconstructed network is a weighted , directed graph , , which depends on the prescribed level of confidence , and is supposed to be a measure of causal traffic in the network . Note that by using shuffling , we can determine a separate threshold for each link , thus reducing the bias towards more active nodes and reducing the contribution from correlations in the absence of interactions . When comparing reconstruction results using shuffling and individually derived thresholds with results based on a single common threshold in order to determine the significance of links , we always used the best possible ( oracular ) single threshold , since in our simulations the original network was known . We also investigated in our simulations if the threshold in the IB approach is indeed optimal and it turns out that choosing anywhere in the range between 0 . 1 and 0 . 9 yields very similar estimates . We simulated the branching process dynamics on 4 different network topologies ranging from a random connectivity with low clustering to a small-world connectivity with high clustering [44] . For the Erdös-Rényi ( ER ) network , nodes were connected randomly with fixed probability resulting in an average node degree and randomly assigned link directionality . In the Watts-Newman ( WN ) network [45] , each node had outgoing links to its nearest neighbors , after which new links were added randomly with probability to introduce long-range connections . This algorithm produces a small-world topology with a high clustering coefficient and an average degree similar to the topology described by Watts and Strogatz [44] . In our simulations we used . Neither the ER nor the WN topology take into account that many networks self-organize and expand through growth , e . g . cortical neuronal networks . We therefore also tested two growth models that achieve a small-world topology with high clustering coefficients . The Barabasi-Alberts ( BA ) [46] model uses a preferential attachment rule in which the probability of attachment from a new node is proportional to the node degree of the existing nodes . Each new node establishes new outgoing links starting initially with disconnected or fully connected nodes . The resulting topology is scale-free in which the degree distribution decays according to a power law with a slope of −3 . Here we use and an all-to-all connectivity for the initial network seed . The BA model requires a new node to attain some knowledge about the degree distribution in the network , which might pose a problem for large networks . In contrast , spatial growth networks [47] do not require global information about the existing network during development . We used the Ozik-Hunt-Ott ( OHO ) network [48] , which is initialized with nodes on a circle and all-to-all connectivity . In this network , a new node , whose location is chosen randomly on the circle , attaches preferentially to its nearest neighbors with outgoing links , hence its growth rule is named geographical preferential attachment . The OHO network is not scale-free , but has a clear small-world property with a high clustering coefficient that is independent of the number of nodes . Its average node degree is simply given by for large networks . In our simulations , we used . The initial seed for the OHO network is the network with an all-to-all connectivity . We note that for both growth models the number of outgoing links was for each node and that both models incorporate a subnetwork ( the initial seed ) with maximal clustering that is particularly difficult to reconstruct in the supercritical dynamical regime . For each topology , we created specific network architectures by using constant individual link activation probabilities , or alternatively , by drawing from a uniform distribution , or truncated Normal distributions ( e . g . truncated within the range [0 , 1] and then scaled to ) . Different dynamical regimes for each topology were explored on networks with nodes and an average node degree of . The quality of network reconstruction as a function of reconstruction algorithm , network topology , and network architecture was studied using nodes and , which approximates the number of electrodes from planar integrated micro-electrode array recordings for neuronal avalanches and the corresponding node degree . For the BA and OHO network , the average degree is discretized since it directly depends on the integer parameter , ( for undirected case ) . Here we used . The branching process dynamics was simulated as follows . A source node was selected randomly according to some initiation probability distribution ( see below ) and activated . In the next time step , all outgoing links emanating from will have a chance to activate its neighbors ( targets ) with the corresponding link activation probability . Each activated target now becomes a source for the next generation of active nodes , and this is repeated for successive time steps until no active nodes are found . Heterogeneity in node initiation was simulated by assigning the node initiation probability from a truncated Gaussian profile , , where is the normalized set of ordered node indices so that all nodes span the profile from is the heterogeneity parameter . Thus , the probability of choosing the center node ( the most active one ) was a factor of times larger than the probability of choosing the two edge nodes ( the least active ones ) . We used , hence the ratios were ≈1 . 65 , 7 . 4 , 2 . 7×105 respectively . We evaluated three different dynamical regimes of the branching process . In the critical regime , one active node at time on average will lead to exactly one active node in the next time step and the distribution of avalanche sizes obeys a power law with a slope of −1 . 5 [49] . In the ER network , the critical regime is reached if the average link probability , for and for WN networks , . Conversely , sub- and supercritical regimes of the branching process were simulated at , respectively . For the BA and OHO networks , a power law spanning a large range of avalanche sizes was difficult to identify , although their sub- and supercritical regimes were similar to those in ER and WN networks . We therefore used for those simulations a value for that yielded the closest fit to a power law size distribution between the sub- and supercritical regimes ( see also Figure 2 ) . A refractory period ensured that an avalanche ended once , or before , all nodes in the network were activated , a constraint that assured termination of the process particularly when simulating supercritical dynamics . Random node activation independent from the ongoing dynamics , i . e . due to noise or external inputs , was implemented such that any node on the network could be activated with probability per time step , expressed as . We used a level of 20% for all simulations with noise , which translated on average into the random activation of one node every five time steps , independent from the ongoing dynamics . Note that randomly activated nodes did not initiate new cascades , otherwise they would increase reconstruction efficiency since the patterns of activity in the ‘noise-induced’ cascades would also be influenced in the same manner by the underlying network that we are trying to reconstruct . While noise was used universally , in some instances we also tested the robustness of the algorithms to time jitter , implemented such that every active node at time was displaced into time bin with 20% chance . We applied the NC , FC , IB , and SS algorithms to different instances of the simulated cascade dynamics on all four network topologies and different architectures . Because the algorithms were described in detail in the Theory section , here , we focus on additional , practical issues . When reconstructing a network using IB , we used a cut-off value for the number of active nodes considered , , above which the IB iteration is skipped . Those iterations would take a significant portion of the evaluation time and yield only a slight gain in the posterior probability . While this diminished somewhat the performance of the IB particularly in the supercritical regimes , larger values of would have resulted in impractically long reconstruction times . In order to establish significance for various network parameters , we used two randomization techniques , the Erdös-Rényi randomization ( ER ) and the degree sequence preserving randomization ( DSPR ) [50] , [51] . In ER randomization , links were completely randomized in order to obtain an ER network with an equivalent number of nodes , links , and weight distribution as in the original network . This randomization destroys any correlations and changes the node degree distribution . In the DSPR , two directed links were chosen randomly between four different nodes , and then the target nodes of the two links were switched preserving the degree distribution . This is repeated many times , and in our implementation the number of such switches is equal twice the number of the links in the network ( number of links that have not been switched even once is less than 2% ) . Finally , for each of the network reconstructions , the total error , , was expressed as the number of links that differed between the reconstructed network and the original network relative to the total number of links in , ( 22 ) This error counts both false positives , i . e . an estimated link does not exist , as well as false negatives , i . e . an existing link was not identified , and because is usually sparse , the error can far exceed 100% of the true number of links . The error was averaged over 10 different realizations for each topology and expressed as mean±standard deviation , if not stated otherwise . When comparing two networks , neither of which represents the “gold standard” , we use the following two measures for comparison . One is , , the percent difference in topology , similar to , but now expressed as the total number of the differences relative to the number of the links that exist in either of the two networks . This is a less stringent measure than the , and the maximal error is limited to 100% . The second is the Pearson correlation coefficient between the link weights among the common links in the two networks , , or alternatively , among the links that are in either of the two , . In order to reduce a potential bias in reconstruction efficiency from arbitrarily selecting a particular significance level , we chose the best reconstruction obtained from the significance levels . Using an over-shuffling factor of 10 , best reconstructions for NC and FC were generally obtained at . In our simulations , we can also measure the traffic of causal activations through any given link by summing all the activations that actually occurred between its source and target nodes . The resulting traffic for each link was compared with the reconstructed link weights ( see Equation 21 ) to study traffic estimates using FC and NC . Coronal slices from rat dorsolateral cortex ( postnatal day 0–2; thick ) were attached to a poly-D-lysine coated 8×8 multi-electrode-array ( MEA; Multichannelsystems , Germany ) and grown at in normal atmosphere in standard culture medium without antibiotics for 4–6 weeks before recording ( for details see [29]–[32] ) . In short , spontaneous avalanche activity was recorded outside the incubator in normal artificial cerebrospinal fluid ( aCSF ) under stationary conditions ( laminar flow of 1–2 ml/min ) for up to 10 hrs . For long-term , pharmacological experiments a second set of cultures was recorded inside the incubator ( for details on long-term recording conditions see [29] ) . In short , MEAs with cultures were placed onto storage trays inside the incubator , which were gently rocked ( ≈200 s cycle time ) . For recording , single cultures grown on the MEAs for 5–6 weeks were placed into a head stage ( MultiChannelSystems , Inc . ) , which was affixed to a second tray within the incubator and which had the exact same motion as the primary storage tray . This allowed recording from cultures inside the incubator in culture medium under conditions identical to growth conditions . Bath application of the AMPA glutamate-receptor antagonist 6 , 7-dinitro-quinoxaline-2 , 3 ( 1H , 4H ) -dione ( DNQX , Sigma ) was used to reduce synaptic excitability in the cortical network . DNQX was directly added to the culture chamber . For wash , the medium was replaced with normal pre-conditioned culture medium . Analysis was based on the following time periods of spontaneous activity: 2–5 hr before , 15–20 hr during DNQX and 2–5 hr after 19 hr of washing of the drug . Spontaneous local field potentials ( LFP ) were low-pass filtered at 50 Hz and sampled continuously at 1 kHz at each electrode . Negative deflections in the LFP ( nLFP ) were detected by crossing a noise threshold of −3 SD followed by negative peak detection within 20 ms and nLFP peak times and nLFP amplitudes were extracted . Neuronal avalanches were defined as spatiotemporal clusters of nLFPs on the MEA . In short , a neuronal avalanche consisted of a consecutive series of time bins with width that contained at least one nLFP on any of the electrodes . Each avalanche was preceded and ended by at least one time bin with no activity . Without loss of generality , the present analysis was done with bin width , estimated individually [30] . ranged between for different sets of cultures . Avalanche size was defined as ( 1 ) the number of active electrodes that constitute an avalanche , i . e . the number of nLFPs , and ( 2 ) as the sum of absolute nLFP amplitudes on active electrodes . In the former case , size ranged from 1 to 60 ( corner electrodes were missing on the array ) , whereas in the latter case size ranged from ( lowest detection level of an nLFP ) up to several thousands of .
During activity cascades , an active node on average can activate less than 1 , exactly 1 , or more than 1 node in the next time step in correspondence to the subcritical , critical , and supercritical dynamical regime of a branching process . We therefore identified these three dynamical regimes for each of the 4 topologies by calculating the corresponding cascade size distributions on networks with N = 5000 nodes , and a constant activation probability for all links . For both the WN and ER networks , the critical probability , , was characterized by a cascade size distribution that followed a power law with a slope of −1 . 5 as predicted by theory [49] ( Figure 2; for ER; for WN ) . Conversely , an exponential distribution characterized the subcritical regime in which most cascades engaged only few nodes , whereas in the supercritical regime , a bimodal size distribution revealed that cascades stayed either relatively small or engaged most of the network . For the BA network , the distribution of cascades sizes in the subcritical regime followed a power law with a slope of ≈−3 for sizes <10 , suggesting that cascades in that regime were dominated by the degree distribution ( slope −3 ) . In contrast , the supercritical regime was identified by a bimodal size distribution . At the transition to the supercritical regime , the BA network revealed a power law slope close to −1 . 5 for a small range of avalanche sizes ( 10 to 100 at ) , which we used to identify the critical dynamics . For the OHO network , a critical regime was indicated at ( mean field prediction was 0 . 085 ) at which the cascade size distribution revealed a corresponding power law with slope of −1 . 5 ( Figure 2 ) , from which it deviates for large cascade sizes . Thus , given the constraints of a constant , the critical regime in the current simulations represented an approximation of a true critical dynamics for both the BA and OHO network ( Figure 2 ) . The characteristic size distributions for each dynamical regime suggest a varying efficiency in reconstructing networks based on the observed activity cascades . For the subcritical regime , we expect fewer ambiguous situations with multiple source nodes ( Figure 1C ) and thus better accuracy in network reconstruction . These smaller cascades , however , contain fewer links that can be estimated per unit time , which should slow the reconstruction progress . The opposite holds for the supercritical regime where large cascades allow for a larger percentage of links to be estimated per unit time , while the reconstruction accuracy might decrease due to an increase in ambiguous situations . Consequently , we expect the critical dynamical regime to achieve a balance between these opposing tendencies in network reconstruction . Additionally , in subcritical regime much greater number of initial events will not propagate at all , in which case a reconstruction step cannot be performed . Thus , it takes much longer time to collect the same number of STES in the subcritical regime than it does in critical or supercritical regimes . We quantified the relationship between the dynamical regime and the reconstruction efficacy by plotting the total reconstruction error as a function of number of propagation steps , , which is the total number of successive time bins that both contain at least one active node . This was done for all three regimes and all four algorithms ( Figure 3; ER topology , , uniform link activation probability for avalanche initiation; see also Figure 4B ) . For both FC and NC , the significance of a link was based on 1000 shuffles . For the IB algorithm , the correct value of was used in the dynamic term ( Equation 4 ) . In our initial evaluation without noise , the IB algorithm was superior in reconstructing the network in all three dynamical regimes . As predicted from the cascade size distributions , its reconstruction efficiency was higher in the critical regime compared to the subcritical regime ( Figure 3A , left , open arrows ) . Importantly , the IB algorithm further improved in the supercritical regime demonstrating its robust handling of situations with common inputs , where it achieved a high efficiency that is possible links were estimated in approximately the same number of propagation steps in order to reach a reconstruction accuracy of 1% . Similarly , the correlation algorithm FC , while being less efficient than the IB algorithm , faired better in the critical regime when compared to the subcritical regime . However , it failed in the supercritical regime to achieve 1% accuracy even for up to 106 propagation steps demonstrating its sensitivity to correlations due to common inputs ( Figure 3A , left , red arrow ) . Importantly , our newly developed NC algorithm clearly overcame the weakness of the FC algorithm and demonstrated its efficiency in all three regimes ( Figure 3A , left , black filled arrow ) . We note that the error reported is calculated with respect to the number of existing links in the network , i . e . ≈600 links for out of 3 , 600 possible links . Hence a reported error of 1% is equivalent to about 1/6 = 0 . 167% overall error in deciding whether a link existed or not . The simple SS algorithm , by avoiding ambiguous situations , performed surprisingly well for all regimes and was comparable to the performances of the IB and NC algorithm . However , the SS algorithm was highly sensitive to noise and relied on the assumption that the observed activations completely arose from the intrinsic dynamics . In fact , when we repeated our simulations in the presence of 20% noise ( Figure 3A , right ) , SS failed entirely in all regimes resulting in errors significantly larger than 100% . Equally important , the IB algorithm now required 4–5 times more propagation steps to reach an accuracy of 1% in the supercritical regime; a sensitivity to noise that originated from the iterative development of the priors over time ( Figure 3A , right , open arrow ) . In the presence of noise , only the NC algorithm robustly reconstructed networks with similar efficiency in the critical and supercritical regime thereby performing even better than the IB in the supercritical regime ( Figure 3A , right ) . In comparison to the standard correlation approach , the NC algorithm provided about 50% improvement in the critical regime and more than a 10-fold improvement to achieve 3% accuracy in the supercritical regime . These results demonstrate that NC performed best given ( 1 ) its simplicity , requiring no assumptions about the network connectivity or network dynamics , ( 2 ) its high accuracy for all three regimes , and ( 3 ) good reconstruction efficiency of about 2 . 7 propagation steps per potential link ( total links ) for the critical and supercritical regime at 1% reconstruction error . Correlation methods in network reconstruction commonly utilize a single , global threshold to identify links , e . g . links are assumed to exist for all pairwise node correlations that are above a minimal correlation value ( e . g . [18] , [20] , [52]–[54] ) . However , heterogeneous node activation frequencies , as well as other conditions , might require different significance thresholds for each link . For the networks in Figure 3 , we compared the efficiency in network reconstruction when establishing link significance using either shuffling or , alternatively , a fixed , best possible threshold for both the FC and NC algorithm in the presence of 20% noise . While shuffling performed slightly worse in the subcritical regime , it significantly improved reconstruction accuracy in the critical and supercritical regime ( Figure 3B ) . For the FC algorithm , shuffling was necessary for an accurate estimation in the critical regime , but it was insufficient in the supercritical regime where the error remained high above 1% , even for large numbers of propagation steps ( Figure 3B , red arrow ) . For the NC algorithm , shuffling was required to accurately reconstruct a network with supercritical dynamics ( Figure 3B , black arrow ) . The results , here plotted for , were similar for ( data not shown ) . This analysis clearly demonstrates that correlation based methods benefit from using shuffling estimates for thresholds in the critical regime . On the other hand , the NC algorithm in combination with shuffling is required for network reconstructions in the supercritical regime . The reconstruction results were obtained on a relatively small network with , and a question arises on how well it performs for larger networks . Since the network model we are trying to reconstruct has binary parameters , it is natural to expect that the number of needed samples , i . e . propagation steps , for the same reconstruction error should at least increase proportionally to . Using NC to reconstruct an ER topology from the cascades in the critical dynamical regime , we demonstrate ( Figure 4A ) that the number of propagation steps required for 1% reconstruction accuracy scales approximately linearly with the total number of potential links in the network , i . e . it scaled as , making it a potentially useful algorithm for reconstructing larger networks . Of particular concern for network reconstruction are situations in which nodes rarely participate in cascade initiations . For example , initiation sites of neuronal avalanches differ up to an order of magnitude in avalanche initiation rate [29] , [32] . Such heterogeneity should make it more difficult to reconstruct the topological neighborhood of less active nodes . Nevertheless , as shown in Figure 3B , the NC algorithm accurately reconstructed networks with heterogeneities in node initiation frequency up to a factor of 268 , 000∶1 for all three dynamical regimes and with only a slight increase in computation for critical and supercritical regimes . Finally , we tested the robustness of the IB , FC and NC algorithms in reconstructing networks with heterogeneous activation probabilities even though the reconstruction algorithms assume a fixed In addition , we introduced a temporal jitter of 20% when binning activity cascades as to account for temporal imprecision in cascade measurements . As before , the noise level was 20% and the node initiation heterogeneity was set to . Under these conditions , the IB failed ( Figure 4C ) to reconstruct the networks to 1% accuracy for all dynamical regimes . Similarly , FC was robust in subcritical and critical regimes , but it failed to reach below a 10% error in the supercritical regime . In contrast , NC always reached below 1% reconstruction accuracy , and performed the best in all regimes . The performance of NC can be further improved in supercritical regimes when the knowledge of the branching parameter , is taken into account , as in ( Figure 4C ) . The NC algorithm also allowed for a robust and accurate reconstruction of network topologies that differed from random connectivity . We tested its performance for 4 different topologies and all three dynamical regimes in comparison to the FC algorithm ( Figure 5; and reconstructed with and 1000 shuffles ) . While the FC algorithm failed for the OHO topology in the critical regime , the NC algorithm reconstructed all topologies in the subcritical as well as critical regime ( Figure 5B ) . Significantly , the FC algorithm failed to reconstruct any of the small-world topologies in the supercritical regime , while the NC algorithm reconstructed the WN as well as the BA network , demonstrated here up to an accuracy of 0 . 1% . Only the OHO network provided a limit above 1% in the efficacy in network reconstruction ( Figure 5B ) . This limit most likely arises because a supercritical dynamics will engage all nodes most of the time in a highly clustered manner at which pairwise shuffling becomes too constrained ( i . e . shuffling two active nodes between two different time points ) . The errors due to reconstruction will most likely be false positives and random in nature . Hence the overall network parameters ( average clustering coefficient , mean path length , average degree ) might or might not be affected significantly by the errors of this order of magnitude . Accordingly , we plotted the reconstructed network parameters as a function of propagation steps for the OHO network in the supercritical regime . As can be seen from Figure 5B , even seemingly high error rates of 10% did not significantly affect the clustering coefficient , while the average degrees are biased to larger values , indicating that most of the errors are false positives . The traffic on a network , i . e . the network flow , is one of the most important aspects that characterizes network functionality [55] . It was reliably estimated by NC for all three dynamical regimes and most topologies . We studied the correlation between the known link activation probabilities and the estimated link weights on an ER network for which link activation probabilities were drawn either from a uniform distribution or a truncated normal distribution between [0 , 1] with ( , and 20% noise ) . In Figure 6A it is shown that for both uniform and normal distributed activation probabilities , NC did significantly better than FC in relating the reconstructed weights to the original weights prescribed as , particularly in the supercritical dynamics . Furthermore , when correlating the estimated with the actual traffic in the network , calculated during the simulation , we found that NC provided a very good measure of the traffic between two nodes ( slope close to 1; Figure 6B and 6C ) . In contrast , FC significantly underestimated the traffic for increasingly higher traffic values ( slope ≪1 ) . These results , obtained on an ER network topology , were also confirmed for small-world topologies , where NC reliably estimated the traffic on the WN and BA network for all three dynamical regimes . Only for the supercritical regime on the OHO network did the NC algorithm estimate the traffic poorly ( Figure 6D , black dots ) . However using further improved the reconstruction in traffic similar to that of an equivalent ER network ( R = 0 . 72; data not shown ) . Given that the avalanche dynamics can be realized on different topologies ( see Figure 2 ) , we used the robust performance of the NC algorithm for different dynamical regimes and widely varying network topologies in order to reconstruct the functional topology and architecture of real neuronal networks that display neuronal avalanches recorded with integrated planar micro-electrode arrays ( MEA ) from neuronal cortex cultures . Spontaneous activity in these cultures is characterized by negative deflections in the local field potential ( nLFP ) indicative of a local synchronization within a subgroup of neurons near the electrode ( Figure 7A–D; [30] ) . The organization of nLFPs in the neuronal network takes on the form of complex spatiotemporal patterns that evolve over successive time bins ( Figure 7E and 7F ) . These patterns , when interpreted as successive node activations ( see Figure 1B ) , were used to reconstruct the functional network topology and network architecture . Under normal conditions , the dynamics that emerges in this system [29] is characterized by neuronal avalanches whose sizes obey a power law with a slope of −1 . 5 for avalanche sizes measured in terms of integrated nLFP amplitude or number of nLFPs indicative of a critical state ( Figure 7G , [6] , [56] , [57] ) . Importantly , the power law in avalanche sizes correlates with a sequential activation of local neuronal groups that is analog to a critical branching process [29]–[32] . In the absence of any knowledge of the real underlying network organization , we reasoned that the reconstructed network architecture might be reliable if its features converged with increasing number of propagation steps in the reconstruction process , e . g . as shown for the simulated OHO network in Figure 5B . Indeed , the network parameters such as the clustering coefficient , , and average node degree , , remained largely constant beyond 30 , 000 propagation steps . This was in agreement with our simulation results , where NC achieved a smaller than 1% error estimate for all topologies in the critical regime within a similar range of propagation steps ( Figure 5 ) . Importantly , despite the relatively small network size of and an average degree of , the clustering coefficient of was significantly higher than what would be expected for corresponding randomized versions of the network . Similarly , we also plot the excess clustering , a network parameter ( not a reconstruction error in ) that measures the clustering coefficient in the network that is beyond the one of an equivalent randomized version of the network . Results for indicate that the high clustering coefficient was not simply due to saturation by adding more and more links into a small network ( Figure 8A and 8B ) . These networks have nearly a linear relationship between the node degree and its strength , i . e . the summed weights of all links at a node , , with ( Figure 8D ) while Figure 8E shows the node in- and out-degree distributions . The weight distribution of the links revealed an exponentially decaying tail demonstrating the presence of a few links with large traffic ( Figure 8F ) . Given that the relatively high clustering was achieved with a small network diameter of ( Figure 8A and 8B ) , which was similar to those of the equivalent randomized networks , our findings demonstrate that the neuronal cultures with neuronal avalanche dynamics establish a small-world topology as previously reported in abstract form [58] , [59] . The functional network topology of the cortex in vitro cultures ( and acute slices [31] ) derived from neuronal avalanches is compared to the results reported for various neural systems in Table 1 . The networks range from full brain and cortical networks among different anatomical and functional areas of the brain [16] , [44] , [60]–[63] to cortical slices and cultures , as well as the neural network of the nematode C-elegans [44] . The table also shows the results for 21 cortical networks binned at ( 14 were acquired in the course of the previous studies , and combined with the current set of 7 , also re-binned to the same ) . The networks and the sources of this data are listed in the caption . One should note that these networks , with exception of the C-elegans are not very sparse , in which case the clustering coefficient will depend on the size of the network , as the table roughly indicates . A better comparison between these different systems can be achieved by using the excess clustering , found in the range between 0 . 13 and 0 . 32 , and which shows no obvious dependence on network size or sparsity . Functional connectivities are dynamically modulated even on a millisecond time scale [21] , [22] . For example , the functional connection of a single synapse , i . e . its efficacy to elicit a spike in a post-synaptic neuron , depends on the depolarization of the post-synaptic neuron , which itself is linked to the neuron's inputs from within the network , i . e . level of network activity . This suggests that the functional small-world topology reconstructed from the dynamical cascades , which captures the spatiotemporal organization of spiking activity [33] , might change with a change in network activity . On the other hand , local synaptic plasticity mechanisms such as spike-timing dependent plasticity [64] are expected to translate successive neuronal activations as reflected in the spontaneous dynamical cascades into a corresponding increase in synaptic strength thereby establishing a structural correlate of the observed dynamics . In that case , the network organization might be expected to be relatively robust to a decrease in overall activity levels . By taking advantage of the NC algorithm to reconstruct network architectures in subcritical and critical regimes , we tested the robustness of the functional small-world topology to acute changes in network activity . We acutely reduced the efficacy of excitatory glutamatergic fast synaptic transmission in the cultured networks by bath application of the AMPA receptor antagonist DNQX ( n = 3 networks ) . As expected , of DNQX significantly reduced the rate of spontaneous cascades by . Thus , in order to compensate for the reduced number of propagation steps per time , networks were reconstructed from ≈20 hr of activity in the presence of DNQX compared to 2–5 hrs of the control and wash condition . DNQX also reduced the formation of large avalanches leading to size distributions more similar to that of a subcritical state , which clearly deviated from the power law with a slope of −1 . 5 for the pre and wash condition ( Figure 9A ) . DNQX significantly reduced the traffic on the network , which under normal conditions revealed an exponential distribution ( Figures 8 and 9B ) . Despite these significant reductions in cascade rate and size as well as link traffic , the small-world topology of the critical network obtained before and after DNQX , nevertheless , was reliably reconstructed during DNQX as indicated by the similarity in the clustering coefficient with increasing number of propagation steps ( Figure 9C ) . On average , , as well as was not different between controls and DNQX . A detailed link-by-link comparison using , between the “pre”↔“wash” showed an error of and correlations , . Similarly , a comparison between “pre”↔“DNQX” , and “DNQX”↔“wash” yielded , respectively . When the comparison were made between the randomized versions of each network ( ER randomization ) , the results were virtually the same for all three cases , . These results show that while these networks are far from identical , their overlap is significantly larger than expected by chance .
The Bayesian approaches described here differ from the so-called Bayesian networks , or belief networks [68]–[70] , which specialize in the reconstruction of directed , acyclic graphs with a smaller number of configurations to be explored . In order to reconstruct cyclic graphs , “loopy” Bayesian network approaches [71] can be used , however , they are , even in their approximate form , NP-hard [72] . Bayesian networks are particularly useful in small networks when precise Bayesian inference is required for each link . In contrast , the IB or PWA approaches in the present study are meant for the reconstruction of large networks from large datasets . For that purpose we derived and tested new methods for reconstructing the functional network topology and traffic from dynamical network cascades . We made the Bayesian methodology feasible by dividing the observations and the network into individual target activations with the corresponding active subnetworks ( STES ) . The essential computational reduction was achieved by using the assumptions of ( a ) only the events in the near past ( the source nodes ) are a potential cause for an activation event in the cascade and ( b ) the activation events of two different target nodes that have common source nodes are independent . Both assumptions make sense in neuronal networks such as the cortex , in which events in the near past predominantly influence the present state of a neuron and where the synaptic transmission of a neuron at different postsynaptic sites is independent . All these methods rely on the assumption that the underlying dynamics is stochastic . A fully deterministic dynamics would not allow to discriminate direct from indirect influences . To combine individual STES and to obtain the reconstructed network , , we used the IB and PWA approach . They enable one to improve the reconstruction reliability whenever additional knowledge about the dynamics ( or priors in the case of PWA ) becomes available . They are computationally feasible , since their computational complexity is simply the number of STES , , times the complexity of the individual STES . We will assume that the needed in an observation for a given reconstruction accuracy is ( as was found for NC , see Figure 4A ) . Hence , the complexity of the IB is , where is the average number of over all STES . It will be likely that is a function of in the critical and supercritical regimes , but less so in the subcritical regime . When , the exponential complexity of IB can be managed to some degree by introducing a cut-off value , , thus reducing the complexity to , but keeping a large pre-factor . The computational complexity of individual STES in PWA will in most cases be equal or less than . For NC , the individual STES have complexity , hence , the NC has the same low complexity as FC and other correlation methods , , but it produces much better estimates of causal traffic and connectivity , making it a candidate algorithm for the reconstruction of large networks . Note , that most of the computational demand in NC comes from shuffling , whose complexity also is . Technical considerations of this algorithm are discussed in the next paragraph ( see also Text S1 for the implementation summary ) . The PWA approach can also be extended to include situations when the cascade propagation speed is highly heterogeneous , i . e . the continuous time approach is necessary , and/or when the amplitudes of the events need to be considered . This will require some knowledge , or experimental estimate , on how temporal differences and event amplitudes will affect the activation probabilities ( see Equation 3 ) . In these cases , the equivalent of the expression in Equation 16 becomes ( 23 ) where is the link activation probability for the link connecting the active source node and the target node . This expression is obtained in the limit of . A simple inclusion of the weights can also be obtained by treating , in which case is not the number of active nodes but the total strength of the sources . This more general framework , requiring the simulation of continuous time dynamics and varying amplitudes was beyond the scope of this manuscript . Although PWA was derived from Bayesian considerations , strictly speaking it is not a Bayesian method , particularly not the NC algorithm . When PWA uses uniform priors , one can argue that it is essentially a maximum likelihood method . The difference , however , with the maximum likelihood approach is that we use uniform priors on the links , but not the configurations themselves , which are the elements of our sample space . Thus , different configurations will get assigned different prior probabilities . When the prior probabilities for the existence of any link , are small , or are assigned based on the sparsity of a network , the existence of a link can be established using a nonparametric measure similar to correlation . Historically , arguments have been made that , in situations where prior knowledge is not available , a precise choice of the prior probability is not crucial [73] as long as the choice is smooth in the region of high likelihood . Thus , a uniform and sufficiently small probability will lead to essentially the same final estimate [74] . The general methodology of PWA and IB was derived in our Theory section . We then tested a particular nonparametric instance of PWA , the NC algorithm , with the goal of reconstructing large networks from large records of a point process dynamics . The NC is essentially a weighted correlation measure , with the weight inversely proportional to the number of potential source nodes . This weighting is not arbitrary , and if one uses a different weighting factor , e . g . , it does not perform as well as NC ( data not shown ) . If one assumes small prior probabilities for each link , this result becomes intuitive , since the posterior probability for the existence of simultaneous links is negligible , hence each link's probability is inversely proportional to the number of possibilities , i . e . active source nodes . Importantly , we did not assume that is small , but only that it is equal to the sparsity of the network and that the dynamics is near the critical point . This indicates that the validity of the NC algorithm does not rely on the precise choice of . The more elaborate IB approach with fully known dynamics established a benchmark that was closely met by the NC algorithm . The NC algorithm returns the link weights that are an approximate measure of the causal traffic across each link . In this paper we tested , using the simulations of a branching point process on a network , the case when the activation probabilities do not depend on the magnitude of the events and the event times are discrete . More general cases can be addressed using an appropriate activation function in equation 3 , and using a different weighting factor for PWA ( see Equation 23 ) . Shuffling of the original time series is commonly used to establish a priori statistical distributions for the null-hypothesis . Our results clearly demonstrate that pairwise shuffling significantly improves the reconstruction accuracy in the critical and supercritical regime . On the other hand , this method imposes strong limitations resulting in a conservative model that not only maintains the average activity rate of each node , which prevents the introduction of correlations due to rate modulation [22] , but also the exact lifetime and size distribution of cascades , thus ensuring that the shuffled raster remains in the same dynamical regime . This shuffling method reaches its limits in the supercritical regime with highly synchronized cascades , e . g . when almost all nodes become active within 1 time step for most cascades , in which the constraints of the pairwise shuffling limit its statistical power . Similarly , pairwise shuffling becomes constrained in the subcritical regime because of the limited number of nodes participating in cascades . Alternative methods combined with pairwise shuffling , such as temporal jittering , using a smaller portion of the raster to determine thresholds , or limiting total number of shuffles , might improve reconstruction efforts further in these cases . The ad hoc use of a global threshold in order to extract a functional connectivity from correlation matrices is often justified by providing a range of thresholds for which the obtained results are robust [18] , [20] , [52]–[54] . In the present study , we obtained thresholds for each potential link , which significantly outperformed the global threshold approach in the critical and supercritical regimes . The calculation of a probability value using a conservative model , i . e . maintained firing rate and cascade sizes and durations also naturally allows these thresholds to be interpreted in terms of significance for individual link existence . As shown in Figure 8C , topological features were shown to be robust for different significance thresholds . Our simulation of the branching process incorporated a refractory period during which a node remained inactive before being able to participate in a cascade again . Thus , the simulated dynamics represents a branching process only in the limit of large number of nodes . Notably , refractory periods for nodes are common in many real systems , where they arise from energy limitations such as transport capacities and where they serve several major purposes , such as limiting the rate with which each node engages in the network dynamics and terminating cascades in the supercritical regime . In the temporal domain , refractory periods support the formation of non-recurrent dynamics in an otherwise recurrent network . For example , in neuronal networks , each neuron after its action potential is not responsive to the near future neuronal feedback [76] , or in epidemics [9] typically studied in Susceptible-Infected-Removed models [77] , in which infected individuals acquire immunity against re-infection supporting the view of epidemic spread as an essential forward cascade with little recurrence . While we have addressed the existence of different dynamical regimes on different topologies , we have not studied comprehensively all possible issues that might affect the dynamics of the network , e . g . network modularity [78] . Despite the dynamic feed-forward aspects of most cascades , the resulting functional architecture is not limited to acyclic graphs because potentially recurrent links between nodes that do not engage in one cascade can be active during other times . In the present study , we derived the directed , weighted functional architecture of superficial cortical layers [29] , [31] grown on planar integrated micro-electrode arrays . We demonstrated that a small-world functional topology of neuronal avalanches is robust to an acute reduction in network traffic , suggesting that it potentially arises from a corresponding structural small-world topology of cortical micro-circuits . The neuronal avalanche dynamics that arises in these layers in vitro parallels layer formation in the intact animal [33] . The reconstruction of the architecture was based on neuronal avalanches , dynamical cascades that form in analogy to a critical branching process [29] , [30] for which our simulations show robust and accurate network reconstruction using the NC algorithm . The estimated clustering coefficient stabilized as predicted from our network simulations . Importantly , a similar topology was recovered from acute , subcritical network dynamics in the presence of DNQX . This suggests that the subgraph described by a cascade does not depend on the overall state of the network , but might underlie structural components of the network as formed by the number and strengths of neuronal connections . A small-world topology combines short distances between network sites with high clustering that allows for diverse functionality of subgraphs , as shown recently for sensory activities in the visual cortex of the cat [79] . Previous studies in dissociated neuronal cultures have quantified dynamical cascades during spontaneous neuronal activity using a variety of measures such as conditional probability [80] , pairwise delayed-correlation indices [81] , and sequential ordering [82] . Additionally , functional topologies were derived using correlation methods with global correlation thresholds [83]–[85] . As shown in the present study , the correlation approach might not adequately address functional connectivity , particular for dissociated cultures which have been shown to display supercritical dynamical cascades [82] . Despite these potential limitations , correlation and mutual information based methods derived non-directed functional small-world topologies from spontaneous activity in dissociated cortical cultures [52] , [86] , in line with our topological findings for the neuronal avalanche dynamics in layered cultures . Our study further quantified the network traffic , which was characterized by an exponential tail distribution similar to what has been found for the weight distribution in dissociated neuronal cultures [52] and airport traffic networks [55] . These characteristics of the small-world architecture formed by neuronal avalanches provide important constraints for future simulations of this type of cortical dynamics .
|
In many complex systems found across disciplines , such as biological cells and organisms , social networks , economic systems , and the Internet , individual elements interact with each other , thereby forming large networks whose structure is often not known . In these complex networks , local events can easily propagate , resulting in diverse spatio-temporal activity cascades , or avalanches . Examples of such cascading activity are the propagation of diseases in social networks , cascades of chemical reactions inside a cell , the propagation of neuronal activity in the brain , and e-mail forwarding on the Internet . Although the observation of a single cascade provides limited insight into the organization of a complex network , the observation of many cascades allows for the reconstruction of very robust features of network organization , providing valuable insight into network function as well as network failure . The current work develops new algorithms for an efficient reconstruction of relatively large networks in the context of cascading activity . When applied to the brain , these algorithms uncover the structural and functional features of gray matter networks that display activity cascades in the form of neuronal avalanches .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"mathematics/statistics",
"computational",
"biology/computational",
"neuroscience",
"computational",
"biology/metabolic",
"networks",
"computational",
"biology",
"computational",
"biology/signaling",
"networks",
"neuroscience",
"neuroscience/theoretical",
"neuroscience"
] |
2009
|
Efficient Network Reconstruction from Dynamical Cascades Identifies
Small-World Topology of Neuronal Avalanches
|
The On-Off direction-selective ganglion cell ( DSGC ) in mammalian retinas responds most strongly to a stimulus moving in a specific direction . The DSGC initiates spikes in its dendritic tree , which are thought to propagate to the soma with high probability . Both dendritic and somatic spikes in the DSGC display strong directional tuning , whereas somatic PSPs ( postsynaptic potentials ) are only weakly directional , indicating that spike generation includes marked enhancement of the directional signal . We used a realistic computational model based on anatomical and physiological measurements to determine the source of the enhancement . Our results indicate that the DSGC dendritic tree is partitioned into separate electrotonic regions , each summing its local excitatory and inhibitory synaptic inputs to initiate spikes . Within each local region the local spike threshold nonlinearly amplifies the preferred response over the null response on the basis of PSP amplitude . Using inhibitory conductances previously measured in DSGCs , the simulation results showed that inhibition is only sufficient to prevent spike initiation and cannot affect spike propagation . Therefore , inhibition will only act locally within the dendritic arbor . We identified the role of three mechanisms that generate directional selectivity ( DS ) in the local dendritic regions . First , a mechanism for DS intrinsic to the dendritic structure of the DSGC enhances DS on the null side of the cell's dendritic tree and weakens it on the preferred side . Second , spatially offset postsynaptic inhibition generates robust DS in the isolated dendritic tips but weak DS near the soma . Third , presynaptic DS is apparently necessary because it is more robust across the dendritic tree . The pre- and postsynaptic mechanisms together can overcome the local intrinsic DS . These local dendritic mechanisms can perform independent nonlinear computations to make a decision , and there could be analogous mechanisms within cortical circuitry .
The On-Off direction-selective ganglion cell ( DSGC ) of the mammalian retina spikes vigorously to moving stimuli , but only weakly to stationary light spots . It responds most strongly over a limited range of stimulus directions , and the direction producing the maximal response is called the “preferred” direction , while a stimulus moving in the opposite direction , called the “null” direction , produces little or no response [1] . We refer to such directionally-tuned spike responses as “direction-selective” . On-Off DSGCs are sharply bistratified neurons that respond with a transient depolarization and burst of spikes at both the onset ( “On” response ) and termination ( “Off” response ) of a bright stimulus within the receptive field . Similarly the leading edge of a bright bar crossing the receptive field will produce a transient On-response , and , if the bar is wide relative to the dendritic extent and the speed low enough , the trailing-edge will produce a distinct , temporally separate Off-response . In their original description of the DSGC , Barlow and Levick [2] noted that direction-selective spike output was produced for stimuli that covered only a small fraction of the dendritic arbor . They proposed that the synaptic mechanism comprised “subunits” that were repeated in an array across the receptive field . In contrast to most ganglion cells , which initiate spikes in the axon initial segment , the DSGC initiates spikes in the dendritic tree [3] . The dendritic spikes are thought to propagate to the soma and initiate a somatic spike , similar to neurons in other regions of the brain where dendritic spiking is important for signal processing [4] . These observations suggest that some type of local dendritic processing could provide the basis for the proposed subunits . Evidence for dendritic spiking in the DSGC was observed in low amplitude “spikelets” , which appear when somatic spiking is suppressed by local application of tetrodotoxin ( TTX ) to the soma , or by hyperpolarizing the soma [3] . Dendritic spikes are hypothesized to initiate somatic spikes with high probability because they are rarely seen under normal conditions . Further , both somatic and dendritic spiking responses are strongly tuned to preferred-direction stimuli , whereas the somatic graded potential shows relatively weak directional tuning ( Figure 1a ) [1]–[3] . This implies that the DSGC does not employ the mechanism used by most other ganglion cells for synaptic integration , where spikes initiated at the soma reflect the summation of synaptic inputs over the dendritic tree [5] . Instead it suggests that DSGC dendrites sum synaptic inputs and generate local spikes which then propagate to the soma , in the process amplifying the responses' directional selectivity . In addition to dendritic spiking in the DSGC , other mechanisms are also important for generating its direction-selective response . GABAergic inhibition is essential , and presynaptic mechanisms render both excitatory and inhibitory synaptic inputs to the DSGC directionally-tuned [6] , [7] . Both excitatory and inhibitory inputs vary in amplitude and relative timing as a function of direction . Further , postsynaptic integration of excitatory and inhibitory inputs has been hypothesized to contribute to DS signals [8]–[12] . Postsynaptic inhibition resulting from null direction movement could produce DS signals in two ways: it could block the propagation of dendritic spikes or it could block their initiation [2]–[4] , [13] ( Figure 1b ) . However , the relative contributions of presynaptic and postsynaptic mechanisms to the DS spiking of the DSGC remains unclear . Initial theoretical studies suggested that postsynaptic mechanisms might suffice [14] and this received some experimental support [15] . However , more recently , presynaptic mechanisms have appeared to be the most significant [6] , [7] , [16] , [17] . We wanted to revisit this issue to delineate the relative contributions of presynaptic and postsynaptic mechanisms in a calibrated model . To investigate how dendritic processing of synaptic PSPs ( postsynaptic potentials ) could amplify DS , we constructed multi-compartment biophysical models of DSGCs , digitized from tracer-injected morphologies calibrated to physiological data obtained prior to tracer injection . We stimulated the models with moving light bars that activated synaptic inputs . The goal was to explore how morphology , voltage-gated channels , and synaptic inhibition affect the initiation and propagation of dendritic spikes , and to compare these with the known physiological properties . Our simulations show that sub-threshold PSPs from the distal dendritic regions of the On-Off DSGC are heavily attenuated by propagation to the soma , but that spikes initiated within local dendritic regions can propagate with high probability to the soma and back-propagate to the remainder of the dendritic tree . Therefore active amplification of DS appears to take place during spike initiation in the dendrites .
To explore whether the morphology of the DSGCs would provide for local dendritic processing , we measured the electrotonic properties . DSGC dendrites branch extensively , with higher-order dendrites tending to loop back towards the soma and many dendritic tips throughout the dendritic field [19] , [20] . Dendritic diameter decreases with branching , ranging from 2–3µm for proximal dendrites to less than 0 . 5µm at terminal branches [21] . These morphological properties are typical of many neurons , especially retinal ganglion cells , and result in higher local input resistance and shorter electronic lengths as one moves away from the soma [22]–[25] . We mapped the input resistance for DSGCs in models that included all the voltage-gated channel types , and found , as in classical studies [22] , that dendritic Rin increased with distance from soma , ranging in proximal dendrites from 150–200MΩ , to greater than 1GΩ for distal dendrites ( Figure 2d ) . This implied that for a given excitatory synaptic conductance the distal dendrites generated larger PSPs than proximal dendrites . Next we explored how dendritic morphology influenced the passive spread of PSPs within DSGC dendrites . The PSP from a single excitatory synapse ( red spot , Figure 2a–c ) produced strong local depolarization ( magenta branches ) that declined steeply with distance from the current-injection point [22]–[26] . We measured the degree of attenuation as a function of synaptic location by comparing the PSP amplitude in the dendrite with that at the soma ( see Methods ) . Measurements of PSPs at single synapses at many points across the dendritic tree showed that dendritic PSP amplitude increased sharply with distance from soma , in line with the local input resistance values ( Figure 2d ) . The corresponding somatic PSP amplitude was weakly dependent on synapse location , and declined gradually as the input was moved away from the soma ( Figure 2f ) . This weak spatial dependence arises because the soma is centrally located , and , as evident from the relatively slow time-to-peak and decay time of the somatic PSPs , tends to reflect the overall depolarization reached after the charge injected into the dendrite has spread through the cell . Thus , under physiological conditions , at each point within the dendritic arbor , the synaptic depolarization will comprise a slower , spatially averaged component , generated by the total input to the cell , and a faster , higher-amplitude component generated by local inputs in the electrotonically isolated regions [23] , [24] . We further quantified the electrotonic isolation by estimating the local space constant across the dendritic tree ( see Methods ) , and found that the space constant of most dendritic loci was less than the distance to soma ( Figure 2e ) , consistent with spatially localized PSPs . Overall the simulations suggest that the dendritic tree of the DSGC is composed of high-gain electrical subunits that can independently integrate synaptic input . This is true for both passive models and our calibrated active models , even when channel densities are perturbed from their “standard” values ( see Methods ) . These high-gain subunits are proposed to generate the directional signals that drive the direction-selective dendritic spikes , which in turn enhance the directional tuning , as reported previously [3] . To measure the excitability of dendritic regions , we simulated dendritic spiking in models with uniformly high ( gNa1 . 6 = 40 mS/cm2 ) dendritic Na+ channel densities . We activated a single synapse and measured Gthresh , the “conductance threshold” for spiking , at locations sampled evenly and independently across the dendritic tree ( see Methods , Figure 3 ) . The locus of spike initiation was not always at the point of input but typically nearby , usually over an entire subregion ( 50–100 µm dia ) within 1ms . Spikes did not initiate at the soma except for very proximal synaptic locations . Our first expectation was that Rin would be the predominant determinant of Gthresh , i . e . Gthresh would be inversely proportional to Rin , however the scatter of the points in Figs . 2d–f show that Rin is not the overriding factor . A small number of locations at intermediate distances from the soma had higher thresholds than would be predicted from Rin alone ( asterisks in Figure 3a–c ) . These regions had few nearby distal branches with high Rin that could be charged up to produce a spike , and experienced significant axial current flow through a high-conductance pathway to the soma [24] . This reduced the current available to charge the local capacitance , causing a rate-of-rise insufficient to produce a spike , but a PSP amplitude high enough to inactivate Na+ channels ( more depolarized than −50mV ) . We also found that the threshold was higher at some of the extreme dendritic tips , due to their higher axial resistance , which reduced their ability to excite more proximal dendrites . However , the majority of locations had bi-directional current paths with proximity to highly excitable terminal dendrites , and therefore had a low spike threshold . These effects implied that the spike threshold of a single dendritic location was dependent on the properties of the local dendritic region . Overall , the distal dendrites of the DSGC , which cover most of the dendritic field [19] , [21] , comprise electrotonically isolated local regions with high gain and low spike threshold , and these regions are capable of independently integrating synaptic input to generate a dendritic spike . In models with dendritic initiation of spiking , we observed that when a dendritic spike reached the soma it invariably spread throughout the entire cell and obliterated any simultaneous dendritic spikes . In these models , when a dendritic region received excitatory input , the dendrites within the local region of 50–100 µm in extent depolarized toward spike threshold . When several such regions received simultaneous excitatory input , the one that reached spike threshold first generated a full-blown spike that propagated to the soma within 1–2 ms , and then back-propagated into the other dendrites within 1–2 ms rendering them refractory ( see Video S1 ) . In other systems , impedance mismatches due to morphology can cause spike propagation to fail when dendritic Na-channel density is low [24]–[26] . Live recordings have shown that most ganglion cells initiate spikes in the axon/soma and actively propagate spikes into the dendrites , which do not initiate spikes [27] , [28] . Thus the dendritic Na-channel densities of most retinal ganglion cells must be high enough to actively propagate spikes but not high enough to initiate them [28]–[30] . However , the DSGC initiates dendritic spikes , so starting from a Na-channel density considered normal for most ganglion cells , 25 ms/cm2 , we set the Na-channel density high enough so that each dendritic spike successfully propagated to the soma and initiated a somatic spike ( see Methods - Calibration , Figure 4a; [3] ) . To explore the requirements for successful dendritic spike propagation , we examined models with dendritic Na-channel densities lower than our calibrated models . In these low-dendritic-Na-channel models , synaptically-evoked spike propagation efficiency was low ( Figure 4b , see Methods ) , because most spikes failed at a thick proximal dendrite branch-point , where they were attenuated by shunting from the large capacitance and low axial resistance . A linear density gradient with a higher proximal density of Na+ channels improved propagation ( Figure 4c ) . Another consequence of this gradient was a smaller difference in spike threshold between proximal and distal dendritic regions ( not illustrated ) . The F/I curve ( see Methods ) for somatic current injection was primarily affected by proximal dendritic Na+ density , and had a slightly lower slope for the gradient model , but still fit within the observed variability of physiological data . We considered the conditions under which a sub-threshold depolarization could facilitate spike initiation . In the real DSGC , light stimulation by moving bars often produced a 5–10mV somatic depolarization 50–100ms prior to spiking . We found that propagation in models with low Na- channel density was also facilitated when somatic and proximal regions were depolarized either by injecting current at the soma or stimulating proximal regions with synaptic input . Transiently depolarizing proximal areas compensated for loss of current due to proximal high membrane conductance by bringing Na-channels closer to activation threshold . This suggested that a combination of proximal depolarization and high proximal Na-channel density could promote the successful propagation of dendritic spikes in the real DSGC . A previous study [3] indicated that dendritic spiking is responsible for the small spikelets seen in somatic recordings , but did not determine whether the spikelets represented full-blown dendritic spikes , or what parameters affected the distribution of spikelet amplitude . To explore these issues we ran a series of simulations in which a subregion of the dendritic tree was stimulated with a spot of light , and recordings made under normal conditions or with the Na-channels in the soma removed , thus simulating TTX application to the soma , which blocked somatic spiking as in the previous study ( Figure 5 ) . We found that each region had a characteristic excitability and ability to transmit spikelets of a certain amplitude to the soma , and that these properties varied across the dendritic tree ( Figure 5a–f ) . Some regions were more sensitive than others and would spike more readily with a weak stimulus , and some regions were relatively insensitive to spiking . To test the effect of these differing excitabilities on a typical response , we ran a simulation of a bar passing over the DSGC , and recorded the somatic spiking and dendritic spiking in 2 locations ( Figure 5g , h ) . The somatic spike train showed vigorous spiking separated by ∼20 ms where the bar passed between 2 regions of high excitability separated by a non-spiking region . As the stimulus passed across each region of high excitability , it initiated full-blown dendritic spikes that propagated to the soma and back-propagated throughout the dendritic tree ( see Video S1 ) . The previous study had shown that somatic PSPs during null direction stimulation , which were devoid of superimposed spikes , were often as large or larger than PSPs during preferred direction stimulation that produced vigorous spiking ( see also Figure 1 ) . We hypothesized that this directional difference was due to local inhibition that suppressed dendritic spike initiation in the null direction without reducing somatic PSP amplitude . We next tested the question whether inhibition functions in the DSGC dendritic tree mainly to prevent propagation of spikes , or whether it serves to prevent spike initiation . To explore the ability of inhibition to modulate the spiking properties of the DSGC , we ran simulations with different spatial arrangements of inhibitory synapses . Initially we wanted to evaluate how much “on-the-path” inhibition was required to suppress dendritic spike propagation . For these simulations we applied a 75µm “spot” of shunting inhibition ( ∼30 synapses , ∼300–3000 pS/synapse , reversal potential ∼Vrest ) over an area that covered the soma and proximal dendrites , while stimulating a distal region with a spot of excitatory input ( ∼30 synapses , ∼100 pS/synapse , reversal potential = 0mV , ) ( Figure 6a ) . Previous work has shown that the total peak inhibitory input to DSGCs is around ∼10nS [7] , [31] , however , given the limited visibility of synaptic currents for a somatic recording electrode , the actual inhibitory input to the dendrites will be somewhat larger ( see Methods; [32] ) . We performed simulations in which we adjusted the magnitude of inhibition and excitation in the dendrites so that the conductance measured at the soma matched that recorded previously [7] . These results indicated that the actual synaptic conductance was likely to be about a factor of two larger than recorded at the soma ( see Methods ) . Nonetheless we found that applying on-path inhibition of up to 5 times the observed values ( 50nS ) , even within a relatively small dendritic region , as suggested by prior theoretical work on non-spiking input [15] , was insufficient to prevent dendritic spike propagation and produced only a modest attenuation in the spike amplitude ( Figure 6b , black trace ) . Increasing the inhibition to 85nS did attenuate the dendritic spikes and prevent activation of a somatic spike . In this case , the dendritic spikes appeared at the soma as low amplitude “spikelets” ( Figure 6c , black trace ) . We performed these simulations for excitation and on-path inhibition in several regions in the dendritic tree on several different cell morphologies , and all gave similar results showing that to be effective , the inhibition would have to be unrealistically strong . The reason , we found , was that to attenuate an actively propagating spike , the inhibitory conductance locally within the region of propagation must be larger than the peak activated Na-channel conductance . Further , we found that the precise timing of the spikes relative to inhibitory input over 50 ms was not important for blocking propagation , as long as there was substantial overlap [15] , because the key factor was amplitude of the inhibitory conductance relative to the Na-channel conductance . The on-path inhibition also attenuated the dendritic PSP produced by excitatory input ( Figure 6b , c gray traces ) . Given that dendritic spikelets are rarely observed at the soma of the DSGC during light stimulation [3] , and that an unrealistically-high inhibitory conductance was needed to shunt propagating dendritic spikes , our conclusion from this set of experiments was that in the real cell , null-direction inhibition is much more likely to block spike initiation rather than spike propagation . We next wanted to determine how much inhibition was required to suppress dendritic spike initiation under the same conditions . The receptive field of the DSGC has both spatial and temporal components [2] , [8] , which are widely believed to result from spatially offset inhibition that trails excitation in the preferred direction . Because the DSGC's distal dendrites are electrotonically isolated , we hypothesized that a response observed in a local region could not represent electrotonic spread from synaptic inputs outside that region . Therefore , responses evoked in a local dendritic region would reflect the spatial localization of the stimulus , and not a time-delayed signal spreading from adjacent regions . To separate spatial from temporal effects within the local dendritic region , we first isolated temporal effects with a stationary , spatially distributed excitatory “spot” of synaptic input ( dia = 50µm ) , preceded or followed by ( Δt = −50 to +50 ms ) a superimposed spot of inhibitory input ( Figure 7a ) . Kinetic and conductance parameters of the synapses were selected so that excitatory and inhibitory conductances matched physiologically-observed values [7] . The excitatory synapses ( 200pS/synapse ) and inhibitory synapses ( 275pS/synapse ) both incorporated a transient temporal filter ( τ = 50ms , high pass ) . In order to assess the required level of inhibition , we adjusted the amount of leading inhibition ( i . e . arriving prior to excitation ) in time ( Δt = −50ms ) to just prevent dendritic spike initiation ( Figure 7b , d ) . When the temporal order was reversed , and inhibition was delayed , excitation depolarized the dendrites enough to generate spikes prior to the inhibition's onset ( Figure 7c ) . Thus physiologically realistic levels of inhibition ( 4–10nS ) can interact locally with excitation to produce a local directional difference in the PSP amplitude . A nonlinear spike threshold dramatically amplified this difference to produce a strongly direction-selective spike output ( Figure 7a , b ) . We called this type of temporal processing the “postsynaptic DS” mechanism because it relied exclusively on interactions within the dendritic tree to generate DS . We next wanted to determine how well the model performed for a spatio-temporal stimulus , essentially identical to one that is regularly used for studying these cells . Previous work has shown that the excitatory and inhibitory inputs to DSGCs are already directional [6] , [7] , [16] , [17] , with inhibition being larger in the null than the preferred direction , and excitation being larger in the preferred than null direction . We explored how the model could reproduce the responses of the DSGCs under conditions where synaptic inputs were activated throughout the dendritic arbor according to the motion of the stimulus . For the “presynaptic” DS mechanism , the excitatory and inhibitory conductances at a dendritic locus varied with direction but were activated at the same time , and for the “postsynaptic” DS mechanism , the conductances remained constant with direction but were activated with an asymmetrical spatio-temporal offset . For the postsynaptic mechanism , we set inhibition with a spatial offset , to generate a temporal offset that was dependent on bar direction ( See Methods ) . Stimulation in the pref direction activated excitatory synapses in advance of inhibitory synapses . As bar direction approached null , inhibition was set to overlap more with excitation , and completely overlapped excitation temporally and spatially in the null direction . Our baseline spatial offset produced inhibition that trailed excitation by ∼50ms in the pref direction . We also tested temporal offsets of 75ms , 150ms , and 200ms . When calibrating the model , we adjusted the spatial offset of the inhibition , and the magnitudes of the inhibitory and excitatory conductances so that the waveshape of the somatic currents measured with voltage clamp matched those recorded from a typical cell ( Figure 8a , b; see Methods ) . To fit the currents in the preferred direction the total excitatory conductance was 6 . 5nS and inhibitory conductance was ∼2 . 5nS , while in the null direction excitation was 3 . 5nS and inhibition was 6nS . These preferred/null ratios of excitation and inhibition are within the ranges reported previously for DSGCs [7] . Once calibrated , we measured the directional-difference in the PSP amplitude for a model without Na-channels as a function of the distance from the soma ( Figure 9 ) . The simulations included either the presynaptic mechanism from Figure 8 , where both amplitude and waveshape of the PSPs depended on direction , the postsynaptic mechanism , where only the temporal offset between excitation and inhibition depended on direction and the amplitudes did not vary , or both mechanisms . The results showed that the directional-difference was largest in the peripheral dendrites , which also corresponded to the areas of highest excitability ( Figure 3 ) . The model reproduced the relatively small directional-difference in the somatic PSP amplitude that is seen in real recordings ( compare Figure 3d and Figure 1 ) . The magnitude of the postsynaptic mechanism was largest for the peripheral dendrites but dropped to almost zero near the soma ( Figure 9 ) . The reason was that the magnitude of the postsynaptic mechanism was directly related to the input resistance and the PSP amplitude ( Figure 2 ) . To determine the contribution of the DSGC's morphology to its direction-selective response , we ran simulations with a moving bar in a simplified model without Na-channels that included only excitatory synaptic input to the DSGC that did not vary according to bar direction , while recording responses at the soma and throughout the dendritic tree . We measured the DS index ( 0 = non-directional , 1 = fully directional; see Methods ) and vector angle of the dendritic PSPs evoked by a bar stimulus that moved alternately in eight directions , and found that the distal dendrites had a weak “intrinsic DS” , with preferred directions that pointed radially outward from the approximate geometric center of the dendritic arbor ( Figure 10a , c ) . This intrinsic DS resulted from spatial summation in dendrites similar to that described in models of starburst amacrine cells [33]–[36] . The directional asymmetry results from partial isolation between a dendritic compartment and the soma , which delays summation of the somatic PSP with the dendritic PSP during centripetal motion [34] . Because the somatic response represents the summation of PSPs from all the dendrites , the effects of the intrinsic DS tend to cancel out resulting in little intrinsic DS measured at the soma . However , the responses in most of the distal dendrites were clearly direction-selective , tuned to the centrifugal direction . To explore the interaction between presynaptic DS and intrinsic DS , we configured the bar stimulus with the above-described “presynaptic DS” mechanism , where excitation was strong in one direction ( 0° ) and weaker in the opposite direction ( 180° ) . Inhibition for this input was set to be the opposite , weakest when the bar moved at 0° and strongest at 180° . We then ran a series of simulations as before , one for each dendritic location , measuring the DS index and vector angle . On the null side of the dendritic tree ( closest to an advancing null stimulus ) where the intrinsic DS of the distal dendrite agreed with the presynaptic DS , the directional difference of the PSPs was 2-fold or more that observed without presynaptic DS ( Figure 10a , c , c , g ) , On the pref side of the dendritic tree ( the side from which a preferred stimulus originates ) , the results showed that the presynaptic mechanism can override the intrinsic DS , producing a directional difference in the evoked PSPs opposite to the local intrinsic DS signal ( Figure 10b , f , d , h ) . This analysis demonstrates that the intrinsic DS at each dendritic location can be large enough to enhance or reduce the local directional difference in the PSP amplitude produced by addition of the postsynaptic and presynaptic DS mechanisms ( Figures 9 , 10 ) . The intrinsic DS mechanism enhanced the DS responses on the null side of the dendritic arbor , and conversely , weakened DS signals on the preferred side of the arbor ( Figure 10 ) . It is interesting to note that there is a well documented “non-DS” zone located on the preferred side of the DSGC [2] , [37] , within which directional responses are much weaker or even absent . These results suggested that the effects of intrinsic dendritic DS may account for the non-DS zone . When Na-channels were included , the model reproduced the DS spiking response of the cell . The Na-channels amplified the preferred PSPs more than null PSPs within a local region because the preferred PSPs were more depolarized ( Figure 11 ) . We tested the effect of different Na-channel densities , and found that this selective amplification effect occurred in both sub-threshold mode and when spikes were initiated ( Figure 11 ) . The spike threshold within local dendritic regions effectively amplified the directional difference of the PSPs to produce strongly direction-selective somatic spikes . To determine the role of dendritic spiking relative to the other DS mechanisms identified above ( presynaptic , postsynaptic , and intrinsic ) , we simulated a bar passing over the DSGC in different directions , and measured the magnitude of the spike and PSP responses and their DS index ( Figure 12a ) . We adjusted the excitatory and inhibitory inputs so that the DS index of the PSPs was ∼0 . 2 ( Figure 12b , similar to that recorded from real cells ) , and found that the DS index of the resulting spikes was ∼0 . 8 , about 4-fold higher than for the PSPs ( Figure 12c ) . We measured the DS index with different Na-channel densities and in addition compared them to a uniform density with a gradient . Higher Na-channel densities , although they tended to generate more spikes , did not increase the DS index . Instead , the lower densities and the gradient gave a higher DS index , because they gave a greater difference in spiking between preferred and null directions . We simulated local TTX application to the soma , as was done experimentally [3] , by turning off somatic Na-channels . The DS index of the resulting spikelets was 0 . 5 , which was higher than PSPs alone but lower than for full-blown somatic spikes ( Figure 12d ) . This implied that , besides carrying dendritic signals to the soma , the role of spikes is to amplify the directional difference of the PSPs received by the DSGC , and that direction-selective spiking is generated at least in part by postsynaptic non-linearities . We next considered the interactions between the presynaptic and postsynaptic mechanisms . We simulated the presynaptic and postsynaptic mechanisms independently and then combined them to explore how each one contributes to produce directional-differences in dendritic PSPs and to direction-selective spiking ( Figure 13 ) . The strength of the synapses was set to produce peak excitatory and inhibitory conductances within physiologically-observed ranges [7] . As above , we simulated the presynaptic DS mechanism by modulating the time-course of the synaptic conductances , and the postsynaptic mechanism with spatially offset inhibition . For simulations with active Na-channels and either presynaptic or postsynaptic mechanisms alone , spiking was strong in the preferred and weak in the null direction ( Figure 13a , b ) , but the presynaptic mechanism produced a stronger DS index than the postsynaptic mechanism ( Figure 13a , b ) . When both mechanisms were combined , the DSGC again spiked in the preferred direction but not the null , and the DS index was the greatest . Thus , pre- and postsynaptic mechanisms cooperated to produce directional differences in the dendrites ( Figure 12 ) , which were then non-linearly amplified with a spike threshold to produce the DSGC's spiking response ( Figure 13 ) . Finally , our morphological models inevitably contain uncertainties as to the dendritic diameter and the surface membrane resistivity that could affect the dendritic space constants , which in turn can influence the degree of dendritic isolation . Because the findings presented here predict that dendritic isolation within the DSGC is an important biophysical factor for generating its directional selectivity , we explored how the DS response was affected by changes in the space constant of the dendritic tree . We ran simulations in 8 different directions with different values of the dendritic axial resistance ( Ri ) . A high value of axial resistance diminished the spread of axial current through the dendrites , which decreased the space constant ( Figure 2 ) and amplified the presynaptic DS mechanism without changing the relative responses in different directions ( Figure 14a ) . A high value of axial resistance also diminished the effect of shunting by the leading inhibition of the postsynaptic mechanism , increasing the number of spikes in both preferred and null directions ( Figure 14b ) . A reduced value of Ri had opposite effects . When both mechanisms were combined , the resulting directional selectivity was intermediate between that for the presynaptic or postsynaptic mechanisms alone ( Figure 14c ) . Once we had developed intuition about how the dendritic tree attenuates PSPs but not spikes , the apparent paradox of Figure 1 was straightforward to understand . A simulation of a somatic recording duplicated the lack of correlation between the PSP amplitude and spiking ( Figure 15 ) . From the previous simulations , we learned that spikes propagate from the dendrites and depolarize the somatic voltage quickly enough to initiate somatic spikes , even from a membrane potential hyperpolarized below spike threshold by 5–10 mV ( Figures 4–7 ) . The dendritic spike is not visible because the somatic spike overlays it precisely [3] ( Figure 4 ) . The recordings shown in Figures 1 and 15 show the result of somatic spiking summed with a compound PSP generated by synaptic conductances across the dendritic tree . From inspection of the spikes in the preferred direction ( gray trace ) , the after-hyperpolarization ( bottom envelope , Figure 1 , 15a ) of the spikes progressively depolarizes by a few mV through each spike burst . The explanation is that the origin of the PSPs and thus their relative amplitude changes depending on the position of the moving bar . The first spikes start when the bar passes over the distal tips of the dendrites . The corresponding somatic PSPs are attenuated by a few mV ( Figure 2f ) . Later spikes in the burst initiate from more proximal dendritic regions , and the corresponding PSPs are less attenuated at the soma . Note , however , that this somatic recording does not reflect the amplitude of the distal PSPs – they are unattenuated by electrotonic decay and thus have a large directional difference to initiate robust spiking . The recordings from the null direction of Figures 1 and 15 ( black trace ) show a compound PSP with greater amplitude but without initial spiking . These recordings reflect PSPs from a more proximal dendritic location that are less attenuated than from a more peripheral dendritic location . The PSPs from this more proximal region are insufficient to cause local spiking because they are shunted by the proximity to the soma . Although the null direction PSPs initiate hardly any spiking , they propagate without much attenuation to the soma and so appear larger than the preferred direction PSPs . Further , because the soma is hyperpolarized 5–10 mV below spike threshold , any dendritic PSP that propagates toward the soma also tends to be attenuated and hyperpolarized , reducing the probability that it will reach spike threshold after back- propagating distally .
Our simulations indicate that local dendritic processing follows from the dendritic structure , and that a purely postsynaptic model can produce strong directional signals ( Figure 13 ) . One might then ask why presynaptic mechanisms have also evolved . Without presynaptic computation of DS , the directional selectivity of the DSGC would suffer because the postsynaptic mechanism decays to almost zero near the soma ( Figure 9 ) and is reduced on the preferred side of the dendritic tree by the intrinsic DS within the dendrites ( Figure 10 ) . Thus presynaptic mechanisms can overcome limitations inherent in postsynaptic processing and produce a more robust system . However , the presence of a non-DS zone in many cells suggests that in many cases presynaptic mechanisms are not strong enough to overcome the intrinsic dendritic bias . This is consistent with a previous report showing that the strength of the presynaptic DS signal is very variable across the population of cells [7] . Clearly a relatively strong presynaptic mechanism would produce a strong and consistent DS signal at the soma ( Figure 9 ) . Our results predict that the variability in the strength of the presynaptic DS signal will be correlated with the variability in the strength of DS in the somatic PSP , with cells having a relatively weak presynaptic DS component also displaying weak DS in somatic PSPs , as illustrated in Figures 1 and 15 . Further work will be required to fully explore the interactions of presynaptic and postsynaptic mechanisms in the DSGC . The circuitry that generates the presynaptic DS is currently under intense scrutiny and is beyond the scope of this study . One of the consequences of dendritic initiation of spiking , revealed by the simulations , is that when a dendritic spike reaches the soma it will spread throughout the entire cell ( see Video S1 ) and obliterate any other simultaneous dendritic spikes [40] . The result is that the dendritic region with the lowest spike threshold will dominate the responses of the cell , because that region will reach threshold first , and therefore will also recover from the ensuing refractory period first , giving a role of “winner-take-all” to the most excitable regions ( Figure 3 ) . The occurrence of dendritic “hot-spots” was predicted by models in which identical synaptic inputs are distributed across the dendritic arbor ( Figure 3d–f ) . Such results raise the question whether the responses of DSGCs are dominated by inputs from only a few dendritic regions , or whether cellular mechanisms exist that even out sensitivity across the dendritic arbor so that dendritic spike initiation is equally likely from any point . Although the answer to this question is unknown , the results of live recordings suggest that typical DSGCs initiate spikes in only a few local regions [3] . Our tests of density gradients in Na+ channels suggest that the excitability could be regulated by a nonuniform density of Na- and K-channels ( Figures 3 , 4 ) . One criticism of voltage-clamp recordings of neurons having synaptic inputs on an extended dendritic tree , especially the DSGC in which dendritic tips are isolated from the soma , is that estimates of conductance are inaccurate because the cell is not adequately space-clamped . To determine how accurate measurements of conductance are in cells of this type , we simulated voltage-clamping the soma and measured synaptic conductances according to the established protocol [6] , [7] . These simulations indicated that estimated conductances differed from the actual ones by 50–100% ( see Methods ) . The accuracy of the estimate of excitatory conductance was greater than that of the inhibitory estimate because voltage clamp errors were greater at depolarized clamp potentials due to axial resistance and the relatively hyperpolarized dendritic membrane , leading to a reversal potential more positive than expected . These simulation results emphasize that a major advantage of computational models is the ability to look closely at mechanisms that would be difficult to study in the real neural system . The model allows the experimenter to estimate a range of errors , taking into account the accuracy of the data provided , and to identify what mechanisms in the neural system are responsible for the errors . Thus , with the dendritic morphology and a few simple assumptions and measurements , the actual conductances can be determined with a greater accuracy . Because our results depend on a theoretical model , it is reasonable to ask how relevant they are to the real neural circuit . The simulations were sequentially calibrated , starting first with spike shape and amplitude ( see Methods ) , then excitability with injected current ( F/I plot ) , and finally proceeding to higher level tests of the spikelet amplitude and behavior . Although the original morphology was derived from careful measurements , in most cases from confocal stacks , some imprecision in the diameters of the reconstructed dendrites is inevitable . We took this into account by bracketing the diameters using an additional multiplicative factor in the models , then verifying that the overall dendritic surface area and time constant were correct by matching the charging curve with injected current . We verified that the results did not depend on a unique combination of parameters , for example , the particular morphology of the dendritic tree , or some unique combination of channel types or their densities - all of our conclusions are based on phenomena that emerged from the simulations . For example , the intrinsic weak DS found in the dendritic system , although derived from the morphology and biophysical membrane parameters , was robust and did not depend strongly upon a particular choice of model parameters ( see Methods ) . The local initiation of dendritic spikes described here that propagate with high probability to the soma represents a general mechanism for performing independent nonlinear computations leading to a decision [41] . For example , a complex cortical cell sums signals nonlinearly from its presynaptic neurons [42] . The synaptic signals originate from a large number of presynaptic neurons , and the amplification performed in any local subregion by nonlinear summation of PSPs in subthreshold mode can independently amplify the signal , potentially leading to a spike [39] . The spike generated by this process can override the processing of other local regions along the propagation route . When a dendritic spike propagates to the soma and axon it provides the neuron with an all-or-none decision based on the nonlinear processing performed by any of the independent local computational subunits [43] .
Experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee at Oregon Health and Science University and NIH guidelines . Dark-adapted , pigmented rabbits were surgically anesthetized with sodium pentobarbital and the eyes removed under dim-red illumination . The animals were then killed by anesthetic overdose . All subsequent manipulations were performed under infrared illumination ( >900nm ) or under dim red light ( >620nm ) . The anterior portion of the eye was removed and the eyecup was transected immediately above the visual streak . The ventral piece was used exclusively in all experiments . The retina was dissected from the eye , and a 5 by 5 mm section of central retina was adhered photoreceptor side down , to a circular glass cover-slip coated with poly-L-lysine ( Sigma ) or Cell-Tak ( BD Bioscience , USA ) and placed in the recording chamber ( ∼0 . 5 ml volume ) . The preparation was continuously perfused ( ∼4 ml/min ) with oxygenated bicarbonate-buffered Ames medium [44] , pH 7 . 4 maintained at 34–36°C . The major electrolytes in Ames medium are: 120 mM NaCl , 23 mM NaHCO3 , 3 . 1 mM KCl , 1 . 15 mM CaCl2 , and 1 . 24 mM MgCl2 . Patch electrodes were pulled from borosilicate glass to have a final resistance of 4–8 MΩ . For extracellular loose-patch recording , the electrodes were filled with Ames medium . For intracellular recording the electrodes were filled with the following electrolytes: 110 mM K-methylsulfonate , 10 mM NaCl , 5 mM Na-HEPES , 1 mM K-EGTA , 1 mM Na-ATP , and 0 . 1 mM Na-GTP . For multi-photon imaging experiments 50–100 µM of Alexa Fluor 488 hydrazide ( Invitrogen Corporation , USA ) was included in the pipette solution . The liquid junction potential of 10 mV was subtracted from all voltages during analysis . The retina was visualized with infrared differential contrast optics , and ganglion cells with a medium soma diameter and a crescent-shaped nucleus were targeted as potential DSGCs [20] . An extracellular electrode was applied to the soma under visual control through a hole in the inner-limiting-membrane above the cell of interest , and a loose patch recording was formed . After establishing that the ganglion cell was a DSGC and determining its preferred direction ( see below ) , the extracellular recording electrode was removed and an intracellular patch-electrode applied to the same cell for whole-cell recording . During whole cell recordings voltage signals were filtered at 2–4 kHz through the 4-pole Bessel filter of the EPC10 Double patch clamp amplifier ( HEKA Electronics Incorporated ) , and digitized at 20–50kHz . To minimize series resistance errors during whole-cell current-clamp recordings , 10ms hyperpolarizing current pulses were applied and the bridge was balanced to eliminate any instantaneous voltage offsets . Bridge balance was monitored periodically during the recordings . Light stimuli , generated on a CRT computer monitor ( refresh rate , 60 Hz ) , were focused onto the photoreceptor outer segments through a 40× ( NA 0 . 8 ) Zeiss water-immersion objective . The percent stimulus contrast was defined as C = 100 * ( L-Lmean ) /Lmean , where L is the stimulus intensity and Lmean is the background intensity . C was set between 30 and 100% . The standard moving stimulus comprised a light or dark bar , moving along its long axis at 800–1200 µm/s on the retina . All light stimuli were centered with respect to the tip of the recording electrode , and thus also with the soma of the ganglion cell . The bar's width was 250 µm , and its length was set to achieve a 1–2 second separation of the leading- and trailing-edge responses . The stimulus area was limited by the aperture of the microscope objective , and covered a circular region 0 . 5 mm in diameter , which reduced the antagonistic effect evoked by stimulating the surround . The leading edge of the stimulus bar commenced at one edge of the stimulus area and moved until the trailing edge reached the opposite edge . Thus , both leading and trailing edges of the stimulus traversed the whole receptive field of the recorded cell , which evoked both the On- and Off-responses of the DSGC . A Zeiss Axioskop 2 FS mot equipped with a LSM 510 meta NLO scanhead and a mode-locked Ti/Sapphire laser ( Chameleon; Coherent , USA ) was used to capture images of DSGC morphology . After break-in , Alexa Fluor 488 hydrazide ( Invitrogen Corporation , USA ) included in the recording pipette diffused rapidly into the dendritic tree . In some cases the recording electrode was removed from the cell body after the cell had filled with dye before imaging took place . The dye was excited using mode-locked laser light from the chameleon laser tuned to 800–920 nm , and emitted light was collected through the objective , filtered through a BG 39 filter , and detected and digitized with the Zeiss LSM 510 system . To aid with digitizing stacks of images of tracer-injected cell morphologies , we wrote additional software routines called from the “Image-J” image processing software package . Using Image-J the operator manually traced the cell's dendritic segments and branching pattern , measuring diameters with the caliper tool . Our software saved the morphologies as a collection of nodes and cables . The morphologies were then imported into the Neuron-C simulator [45] , [46] , and endowed with voltage-gated channels ( see “Channel densities” below ) . Semi-random arrays of presynaptic cells ( see below ) were constructed automatically by the simulator with a regularity ( mean/SD ) of 6–10 , and synapses were connected between the presynaptic cells and the closest dendrite of the DSGC if it was within a threshold distance ( typically 10 um ) . We set the compartment size small enough ( 0 . 03 lambda or less ) so that each synapse from a presynaptic array of cells was connected to a different compartment , preserving spatial accuracy . Five morphologies were digitized from confocal stacks and studied along with another more detailed morphology ( “ds1e” ) , which had been traced with a Neurolucida system ( Microbrightfield , Inc ) . Two morphologies explored in detail here , “DS060825” and “ds1e” had ∼750 and ∼3000 compartments respectively . The simulations were run on an array of 15 computers each with 2 or 4 AMD Opteron cores for a total of 48 CPUs , allowing simulations with 50 parameter sets to be run in 24–48 hours . We performed several types of simulations: calibration , receptive field mapping , single flashed spot , and moving bar . In calibration simulations , we injected various levels of current into the soma and measured the spiking response . Each simulation took ∼1 hour of computing time , and 20–50 simulations were typically run in parallel . In the mapping simulations , we chose a set of points ( nodes ) in the dendritic tree , and for each point , a protocol measured the conductance threshold ( see below ) . These simulations took roughly 30 minutes per dendritic node , and a sample of several hundred nodes was required for an accurate map of dendritic properties . In single spot simulations , a small spot of synaptic input was turned on over a portion of the dendritic tree and the postsynaptic and soma voltages were recorded . The length of these simulations depended on the spot duration but typically took less than 30 minutes . In the moving bar experiments ( see below ) , we ran 8 simulations in parallel for each of the 8 directions of motion ( 360°/8 = 45° increments ) , each of which took ∼45 minutes of computing time . We tested variations in many parameters , including morphology , synaptic input parameters , and channel density parameters , which multiplied the number of necessary simulations , for a total of ∼200 , 000 simulations to produce the results in this paper . We measured the attenuation from a dendritic point to the soma by stimulating the point with a low-conductance synapse ( 200 pS ) , and computing the voltage attenuation as the ratio of the dendritic and somatic PSP amplitudes [25] . An attenuation less than 1 indicated a dendritic PSP smaller than the somatic PSP . We also computed “synaptic transfer” , a measure of attenuation less sensitive to dendritic Rin , as the ratio of the PSP amplitudes independently evoked by a dendritic synapse and by a somatic synapse . We performed this measurement over the extent of the dendritic tree by testing many points in independent simulations , producing maps of the dendritic attenuation and input resistance properties ( not illustrated ) . We computed an approximation to the steady-state space constant ( λest ) for various points in the dendritic tree to estimate a dendritic region's capability of independently integrating synaptic input . λest was computed between two points i and j in the dendritic tree by re-arranging the formula for steady-state voltage decay in a passive infinite cable to give:where distij is the distance between points i and j . A single synapse was turned on for 100ms to stimulate point i . The simulations were performed using an active model that included Na , Kdr , KA , Ca , KCa , and Ih channels , and with a synaptic conductance ( 50 pS ) which always produced a sub-threshold PSP . The quantity was computed from the steady-state voltages of points within 20–60 um of the site of stimulation and then averaged to give . This method thus estimated the space constant based on the local dendritic structure under realistic conditions . While exploring the dendritic Na+ channel density necessary to generate dendritic spikes in response to synaptic input , we found that some regions were more excitable than others , i . e . they produced more spikes . In order to quantify a region's “excitability” , we measured the efficacy of a single synapse to elicit a dendritic spike . The synapse had an exponential release function with a time constant of decay that was longer than the extent of the experiment , and remained “on” for 100ms unless a spike occurred . For a given point in the dendritic tree , we determined the “conductance threshold” ( Gthresh ) as the minimum synaptic conductance necessary to elicit a dendritic spike , using an automatic binary search algorithm . This algorithm was run independently on a set of points selected uniformly from the dendritic tree . For each point the algorithm started after the model had equilibrated at a steady-state resting potential , and the model's equilibrated state ( voltage of each compartment , synapse states , and channel states ) was saved for later use . The initial conductance of the synapse was set halfway between the range of 100pS and 5nS ( ∼2 . 5nS ) . If a spike occurred within a short interval ( 25–100 ms ) , the conductance was set to the midpoint of the lower conductance range ( 100pS to 2 . 5nS ) , but if no spike occurred , the conductance was set to the midpoint of the higher conductance range ( 2 . 5nS to 5nS ) . The model was then reset to its original equilibrated state from the saved file and the process was repeated with the new conductance value in the reduced conductance range . The algorithm determined Gthresh with an accuracy of 100pS in 7–8 iterations , sufficient to discern the large relative differences in Gthresh between distal and proximal regions . Because the model was noiseless , there was no uncertainty in the measurement of Gthresh . Although almost all points tested initiated dendritic spikes , the points differed in their ability to successfully propagate spikes to the soma and initiate a somatic spike . To quantify the success of dendritic spike propagation , we injected a synaptic input with the threshold conductance at each dendritic location in independent simulations , recorded dendritic and somatic voltages , and divided the number of somatic spikes by the number dendritic spikes , calling this “propagation efficiency” . The value of propagation efficiency ranged from 0 to 1 , with 1 indicating that each dendritic spike successfully propagated to the soma and initiated a somatic spike . For the purpose of defining biophysical properties , the morphology of each model was partitioned into 5 regions: dendrites , soma , hillock , thin segment , and axon [28]–[30] . The On-Off DSGC has a bistratified dendritic tree separated into On and Off layers , and each morphology had 3–4 dendritic systems which arose from primary dendrites at the soma . We found each dendritic system was spatially separate , and some arborized in both the On and the Off layers [19] . Dendrites in the Off layer were on average more distant radially from the soma than those in the On layer . The exact diameter of each dendritic segment that results from digitizing a tracer-injected image is difficult to establish , although the relative diameters between segments can be established with more certainty [21] , [47] . Because the diameter of a dendritic segment determines its surface area , capacitance , and axial resistance , we explored the effects of deviations from the digitized morphology . The diameter of each dendritic segment was bracketed by scaling by 0 . 5 and 1 . 5 , which linearly scaled the dendrites' capacitance and quadratically scaled their axial resistance , affecting the spatial spread of current from the soma [28] , [29] , [48] . This in turn affected the electrotonic separation between the soma and terminal dendrites , as well as the charging curve , i . e . the voltage trajectory up to the first spike , and time-to-first spike during current injection . Our results were qualitatively similar for models with scaled dendritic diameters . It should also be noted that control simulations where the synaptic inputs were switched to obtain the opposite preferred direction , showed qualitatively similar results , i . e . the synaptic mechanisms could be configured to produce DS in any arbitrary direction for a given morphology . We set the channel density for each morphological region with fast inactivating Na , Kdr , transient KA , high-threshold CaL , Ih , and KCa channels similar to previous models [28] , [30] , [48] , [49] ( Table 1 ) . We set reversal potentials for Na+ at +65mV [50] and for K+ at −100mV , which approximated Goldman-Hodgkin-Katz ( GHK ) potentials calculated for the channel permeabilities assumed in the simulation from the internal recording electrode solution and the external Ames medium [44] , [51] . The somatic and dendritic Na+ channel densities were encapsulated by two separate parameters . We calibrated these parameters against phase plots from physiological data from rabbit . In order to allow dendritic spikes to initiate and propagate to the soma , the Na-channel density on the dendrites was increased beyond that necessary to produce somatic spike back-propagation into the dendrites [26]–[30] . To generate realistic peak values of dV/dt during the rising phase of a somatic spike we reduced the Na+ channel density in the soma and hillock while preserving relatively high dendritic Na+ channel densities . To match the physiological data , we slightly altered channel parameters such as the activation and inactivation offsets , and rate multipliers ( see below ) . The channel kinetics were normalized in the simulator software to 22°C , and we took a Q10 value of 2 . 3 for Na+ channel activation as an overall temperature coefficient to match channel kinetics at 35°C [48] . Recent immunocytochemical evidence shows that Nav1 . 2 channels are initially expressed at the thin segment during early development but later replaced by Nav1 . 6 channels [52] . Retinal ganglion cells in Nav1 . 6-null mice exhibit impaired ( lower ) firing rates , and apparently compensate for the missing channel type by increasing the density of Nav1 . 2 channels [53] . These developmental findings suggested that NaV1 . 6 channels play the dominant role in spike generation . NaV1 . 6 channels are known to generate a higher persistent current following a spike , leading to a faster recovery from after-hyperpolarization ( AHP ) , which might be responsible for the shorter inter-spike interval observed in wild type Nav1 . 6 mice [54] . Nav1 . 6 channels activate at more hyperpolarized potentials than Nav1 . 2 , which could affect spike shape and rate [54] , [55] . We explored the differences between Nav1 . 2 and Nav1 . 6 spike trains using preliminary single compartment models containing either NaV1 . 2 and NaV1 . 6 sodium channels . We found that spikes recorded in the DSGC in response to somatic current injection exhibited a similar fast recovery from AHP that we could only match in the model with the inclusion of Nav1 . 6 channels . To determine the best match using existing models of Na+ channel types for the spike shapes measured in the DSGC , we approximated the kinetics of Nav1 . 2 and Nav1 . 6 channels with Markov models [30] , [56] , [57] . We explored the differences between Nav1 . 2 and Nav1 . 6 spike trains using preliminary single compartment models containing either NaV1 . 2 or NaV1 . 6 sodium channels . Both models started with identical K channel densities and kinetics , but one contained Nav1 . 2 channels , and the other Nav1 . 6 channels . To simplify initial calibration of the model , we started with an existing Markov model of Nav1 . 2 Na-channel type and developed it for an approximate match with the real cell's spiking properties , then set the NaV1 . 6 model with similar parameters . We then applied a constant current input to the 2 models and adjusted the densities and kinetics of the NaV1 . 6 channels to produce the best match by eye for spike amplitude , after-hyperpolarization , frequency , and phase plot . In this process we found that the NaV1 . 6 type at any particular voltage was more activated and therefore exhibited a larger open probability . To produce comparable spike amplitude and frequency , we gave the NaV1 . 6 channel rate function an offset of 10mV depolarized from the original Markov activation rate function [57] , and to produce comparable spike shapes , we set the NaV1 . 6 density 2–3 times lower than the NaV1 . 2 density . We then took this set of parameters as the initial basis for the spiking properties of our multi-compartment model of the DSGC , and further modified them during the process of Calibration . Because we found the distal regions of the DSGC to be more excitable , we tested the effects of a higher proximal Na+ channel density on dendritic spike propagation and the spatial distribution of dendritic excitability . Recent evidence suggests that some retinal ganglion cell dendrites have a high proximal Na+ channel density , although it is not known whether these cells are DSGCs [58] . Previous modeling studies suggest that dendritic Na-channels are necessary for normal spiking [28]–[30] , so we set Na-channel density as a gradient where Na+ channel density was high in the proximal regions ( gNa1 . 6 = 45mS/cm2 ) and declined linearly as a function of integrated cable distance from the soma to a baseline value ( gNa1 . 6 = 20mS/cm2 ) for the most distal dendrites , and explored the effect of this gradient on dendritic spiking . We ensured that the minimum density of the most distal dendrites was still high enough to allow initiation and propagation of dendritic spikes ( propagation efficiency ∼1 ) , as well as backpropagation of somatic spikes [26] . In a series of initial simulations , we explored the electrotonic properties of the dendritic tree . We found that as the distance from the soma to a dendritic locus increased , the input resistance increased , and the amplitude of a somatic PSP evoked from a constant-strength synapse decreased [22] , [33] , [59] . This raised the question of whether a compensatory mechanism could modulate the PSP amplitude in the dendritic tips . Because there was evidence for Ih currents in the recordings from the real DSGC , we tested its effect in the dendrites . In other neural systems , an Ih channel gradient with increasing density and decreasing activation offset with distance from the soma can reduce such a tendency by dampening excitability in distal dendrites [60] . To study the effects of a non-uniform distribution of Ih on dendritic excitability , we ran some simulations with dendritic Ih channel densities that started at a baseline value close to the soma ( gIh = 0 . 001mS/cm2 ) and increased linearly with distance from soma to roughly 10 times the baseline value ( gIh = 0 . 01mS/cm2 ) . In those simulations , in order to prevent over-excitability from increased Ih in the distal regions of the dendrites , we ramped the activation voltage of Ih channels down with distance to 10mV more hyperpolarized in the distal regions than in proximal regions [60] . In most simulations , we included synaptic inputs from bipolar and small-field inhibitory amacrine cells . The presynaptic cells were modeled as passive single compartments controlled by a voltage clamp directly set by the stimulus . Each presynaptic cell compartment provided one synapse onto a dendritic compartment of the DSGC . The stimulus for the presynaptic inhibitory amacrine cells was typically spatially offset to simulate the amacrine cells' spatially offset inhibition . The presynaptic voltage passed through threshold and exponential release functions , and the resulting neurotransmitter release was low-pass filtered ( tau = 2ms , [30] , [45] , [46] ) . To implement noisy vesicle release , the level of released transmitter controlled a nonstationary Poisson ( random ) release function . The filtered transmitter then passed to a postsynaptic ligand-gated channel , modeled as a Markov 7-state AMPA receptor [61] , or a Markov 5-state GABAA receptor [62] . Binding of transmitter to these receptors produced a postsynaptic conductance , whose maximum value was set for each simulation , and ranged from 50pS to 5nS . The reversal potentials for excitatory and inhibitory channels were 0 mV and −68 mV , respectively . Although bipolar and amacrine cells presynaptic to ganglion cells typically make several synaptic contacts [63]–[65] , we included only 1 synapse per presynaptic cell for simplicity . This was equivalent to several synapses each with a proportionately smaller conductance within the local dendritic region . To simulate light stimulation over a receptive field , while avoiding the complications of photo-transduction , light responses of each bipolar and amacrine cell were generated via a “transduction element” which transformed a light intensity in space I ( x , y ) to a voltage-clamped potential . For example , for DS060825 , ∼220 excitatory cells and 180 inhibitory cells were randomly distributed across the On or Off layer and synapically connected to the DSGC's dendritic field . Each transduction element that connected to a cell was assigned a location in space that corresponded to the soma of that cell . Excitation and inhibition were controlled by independently-modulated light stimuli mapped to the same dendritic field . Standard conductance values used except where noted otherwise were excitatory , 80 pS/synapse , inhibitory , 95 pS/synapse . Spatially leading or trailing inhibition was simulated by delaying the onset of the excitatory or inhibitory light stimulus , respectively . For a stimulus moving at velocity vbar , delaying the onset of excitatory or inhibitory stimuli by a time Δt produced a spatial offset of Δx = vbar/Δt . The biophysical parameters provided for each morphology were calibrated to match the F/I curve , spike shape ( via phase plot ) , and ISI curve produced by current injection recordings in the cell from which it was digitized ( Figure 16 ) . This produced channel kinetics and densities mentioned in the “Channels” subsection ( see above ) similar to previously published models of retinal ganglion cells [28]–[30] , [48] , [49] . In total , five morphologies were modeled . The channel densities thus obtained were closely constrained , because the dendritic Na- and K-channel density is inversely related to the slope of the firing rate vs . input current function [28] , [48] . The reason is that if the dendritic channel densities are low , there can be no local dendritic spike initiation , which causes the charge from one spike to surge into the dendrites and return quickly to the soma to initiate another spike too soon . With active back-propagation of spikes into the dendritic tree , the membrane gets charged by the spike and then discharged by K-channels , so any extra charge is prevented from propagating to the soma [28] . Ion channel densities and kinetics were calibrated to electrophysiological and pharmacological data . When Ih channels were blocked by application of ZD7288 , the DSGC hyperpolarized to 10–20mV below resting potential ( data not shown ) . To simulate this effect , we set the reversal potential of the leak conductance to −100mV , distributed Ih channels across the soma and dendrites , and configured them and the other channel types to produce a steady-state resting potential ranging from −70 to −80 mV . Dendritic leak reversal potential was set to assist in calibrating the spike rate , which is particularly sensitive to dendritic channel activation in ganglion cells because their Na and K channels are relatively inactive during the inter-spike interval [28] , [48] , [66] . We calibrated voltage offsets and densities for Nav1 . 6 and Kdr channels by matching phase plots of spikes from physiological recordings ( Figure 16e ) . The Nav1 . 6 and Kdr channel activation curves were offset depolarized by 4 . 5mV and 17mV , respectively . Offsets that significantly varied from these produced mismatched phase plots and a voltage threshold for spiking that differed from the real data . Calcium channels of both high-voltage-activated ( HVA ) L-type , and transient low-voltage-activated ( LVA ) T-type have been found in the soma and dendrites of retinal ganglion cells [67]–[70] . We included L-type Ca2+ channels , modeled as a discrete Markov channel with m3-like kinetics [71] , and we set the Ca2+ channel density uniform across the soma and dendrites . We modeled intracellular [Ca2+] dynamics in the soma with 10 diffusion shells , each 0 . 1µm thick , with a Ca2+ pump set to give a decay constant of ∼100ms [30] . In the simulations [Ca2+]i increased linearly with spike rate , as has been directly observed in DSGC dendrites [3] . In many types of ganglion cells , Ca2+-activated K channels ( KCa ) reduce the firing rate during a prolonged current injection [72] . We included two types of KCa channels , a small conductance , high [Ca2+] affinity , voltage-independent sKCa channel with an activation time constant near 100ms [73] , and another sKCa channel with a higher [Ca2+] affinity and activation time constant near 300ms [74] . KCa channel densities were distributed uniformly across the soma and dendrites , set to match the cell's frequency-current and ISI curves produced by spike trains at various levels of current injection ( Figure 16d–f ) . The calcium system ( Ca2+ channels , pump , and KCa channels ) was configured such that [Ca2+]i never exceeded 1 µM during repetitive spiking [30] . Both KCa channel types were active during the inter-spike interval but did not significantly affect spike shape . In some simulations , we added excitatory and inhibitory synaptic input from a moving bar stimulus ( vbar = 1000µm/s ) , calibrated to evoke a response similar to physiological data ( see above , and Figures 6 , 11a , b ) . We adjusted the spatial separation of the excitatory and inhibitory stimulus components ( see “Synaptic Input” ) and their corresponding synaptic conductances to approximate the wave shapes of currents ( Vhold = −75 and 0 mV ) recorded in a typical DSGC . We set the rise time for postsynaptic potentials ( PSPs ) to ∼1 ms and the time constant of decay for EPSCs and IPSCs to 50ms . For simulations of bar sweeps in 8 directions , we modeled the presynaptic mechanism with overlapping excitatory and inhibitory synapses . The synaptic strength per synapse for excitation ( ge ) and inhibition ( gi ) in each direction θ was computed as:The equations allowed for an arbitrary pref direction to be assigned , typically . For postsynaptic inhibition , we used a similar equation involving the onset of a temporally delayed inhibition instead of conductance strength:where is maximum separation in seconds between the onset of excitation and the onset of inhibition . Given a velocity , a temporal offset of produced a spatial offset of . At a given locus we quantified the direction selectivity of the response by stimulating at evenly-spaced angles distributed over 360 degrees . At each angle the response comprised a vector with length equal to the response amplitude and direction equal to the stimulus direction:where is the spike or peak PSP response vector for a bar swept at angle , and is the magnitude of the response . The vector sum represented the directional response , and its length , normalized by the sum of response amplitudes , represented the “DS index” or DSI , and ranged from 0 to 1 [3] , [7] . For comparing PSP and spiking responses , the peak PSP was computed by first digitally removing spikes ( “spike-blanking” ) [3] . For some tests , we calculated the directional difference between PSPs as the peak amplitude of the preferred direction PSP minus the peak amplitude of the null direction PSP . In order to determine how postsynaptic inhibition suppresses spikes in the DSGC during null-direction stimulation , we first attempted to determine the magnitude of the inhibitory synaptic conductances as measured from the soma . Excitatory and inhibitory synaptic conductance components are often estimated from the currents recorded at the excitatory and inhibitory reversal potentials ( e . g . [17] ) , or by measuring currents over a range of holding potentials and calculating the excitatory and inhibitory synaptic conductances from the synaptic current-voltage relation [6] , . With either approach , incomplete space-clamp inevitably leads to errors in the magnitudes of the conductance estimates [32] . To investigate how estimates of the synaptic conductance derived from recordings at the soma deviate from actual conductances , we simulated synaptic input to the DSGC model and estimated the conductances during somatic voltage clamp ( Figure 17 ) . During a small voltage step , accurately fitting the capacitive transient in the DSGC requires a sum of exponentials , implying that the cell is not isopotential [7] . We stimulated a distal area with a spot of co-localized excitatory and inhibitory input , and verified that the dendrites were not isopotential with the soma ( Figure 2a ) . When the soma was voltage clamped at holding potentials above or below resting potential , current leaked out through the dendrites and distal current flow was impeded by axial resistance , causing a voltage difference in the distal dendrites . At more depolarized holding potentials the model was less isopotential , and the postsynaptic current produced by the spot , computed by summing all individual synaptic currents , was larger than the synaptic currents recorded at the soma . This produced a lower I–V slope at each time point during the synaptic response , which led to a more depolarized synaptic reversal potential estimate ( Figure 2c ) and an underestimate of the total conductance ( Figure 17d , black ) . For this spot stimulus , the excitatory and inhibitory synaptic conductances were underestimated by factors of ∼40% and ∼50% , respectively ( Figure 17d , red , blue ) . Inhibitory conductances were underestimated more than excitatory conductances because space clamp errors were greater at depolarized clamp potentials , leading to a lower slope on the I/V plot and a reversal potential more positive than expected . A more positive synaptic reversal potential is interpreted as a relatively larger excitatory component or smaller inhibitory component . When the spot of synaptic input was localized over the soma and proximal dendrites , errors in the synaptic conductance estimates were minimal . In a similar model using a moving bar , the synaptic conductances were under-estimated by a similar amount in the distal dendrites and the soma .
|
The On-Off direction-selective ganglion cell ( DSGC ) found in mammalian retinas generates a directional signal , responding most strongly to a stimulus moving in a specific direction . The DSGC initiates spikes in its dendritic tree which are thought to propagate to the soma and brain with high probability . Both dendritic and somatic spikes in the DSGC display strong directional tuning , whereas postsynaptic potentials ( PSPs ) recorded in the soma are only weakly directional , indicating that postsynaptic spike generation markedly enhances the directional signal . We constructed a realistic computational model to determine the source of the enhancement . Our results indicate that the DSGC dendritic tree is partitioned into separate computational regions . Within each region , the local spike threshold produces nonlinear amplification of the preferred response over the null response on the basis of PSP amplitude . The simulation results showed that inhibition acts locally within the dendritic arbor and will not stop dendritic spikes from propagating . We identified the role of three mechanisms that generate direction selectivity in the local dendritic regions , which suggests the origin of the previously described “non-direction-selective region , ” and also suggests that the known DS in the synaptic inputs is apparently necessary for robust DS across the dendritic tree .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[
"physiology/sensory",
"systems",
"neuroscience/sensory",
"systems",
"physiology/neuronal",
"signaling",
"mechanisms",
"biophysics/theory",
"and",
"simulation",
"computational",
"biology/computational",
"neuroscience",
"neuroscience/neuronal",
"signaling",
"mechanisms",
"neuroscience/theoretical",
"neuroscience"
] |
2010
|
Dendritic Spikes Amplify the Synaptic Signal to Enhance Detection of Motion in a Simulation of the Direction-Selective Ganglion Cell
|
Recently , enterovirus 71 ( EV71 ) has caused life-threatening outbreaks involving neurological and cardiopulmonary complications in Asian children with unknown mechanism . EV71 has one single serotype but can be phylogenetically classified into 3 main genogroups ( A , B and C ) and 11 genotypes ( A , B1∼B5 and C1∼C5 ) . In Taiwan , nationwide EV71 epidemics with different predominant genotypes occurred in 1998 ( C2 ) , 2000–2001 ( B4 ) , 2004–2005 ( C4 ) , and 2008 ( B5 ) . In this study , sera were collected to measure cross-reactive neutralizing antibody titers against different genotypes . We collected historical sera from children who developed an EV71 infection in 1998 , 2000 , 2005 , 2008 , or 2010 and measured cross-reactive neutralizing antibody titers against all 11 EV71 genotypes . In addition , we aligned and compared the amino acid sequences of P1 proteins of the tested viruses . Serology data showed that children infected with genogroups B and C consistently have lower neutralizing antibody titers against genogroup A ( >4-fold difference ) . The sequence comparisons revealed that five amino acid signatures ( N143D in VP2; K18R , H116Y , D167E , and S275A in VP1 ) are specific for genogroup A and may be related to the observed antigenic variations . This study documented antigenic variations among different EV71 genogroups and identified potential immunodominant amino acid positions . Enterovirus surveillance and vaccine development should monitor these positions .
Human enteroviruses include over 100 serotypes and usually cause self-limited infections , except polioviruses and enterovirus 71 ( EV71 ) which frequently involve neurological complications [1] , [2] . Although EV71 was first described in 1969 , a retrospective analysis shows that this virus circulated in the Netherlands as early as 1963 [3] . Recent molecular evolution studies predicted that EV71 could have emerged in the human population around 1941 [4] . Globally , two patterns of EV71 outbreaks have been reported: small-scale outbreaks with low mortality and large-scale outbreaks with high mortality . The latter pattern occurred in Bulgaria with 44 deaths in 1975 , in Hungary with 45 deaths in 1978 , in Malaysia with 29 deaths in 1997 , in Taiwan with 78 deaths in 1998 , in Singapore with 5 deaths in 2000 , and recently in China with more than 100 deaths every year after 2007 . Due to its tremendous impact on healthcare systems , development of EV71 vaccines is a national priority in some Asian countries [2] . EV71 has one single serotype as measured by hyperimmune animal antiserum but can be phylogenetically classified into 3 genogroups ( A , B and C ) and 11 main genotypes ( A , B1∼B5 and C1∼C5 ) by analyzing the most variable capsid protein sequences ( VP1 ) [1] . Recently , one new genogroup was only detected in India [5] . Genogroup A viruses were isolated in 1970 in the United States and were not detected globally again until 2008 . In an investigation of a HFMD outbreak in central China in 2008 , Yu et al identified five EV71 isolates which were closely related to genotype A based on analysis of the VP1 gene [6] . In contrast , genogroups B and C are widely circulating in the world with different evolution patterns [7] , [8] . Interestingly , genogroup replacements of EV71 have been well documented in Taiwan and Malaysia [1] , [2] . In Taiwan , nationwide EV71 epidemics with different predominant genotypes occurred in 1998 ( C2 ) , 2000–2001 ( B4 ) , 2004–2005 ( C4 ) , and 2008 ( B5 ) [9]–[11] . In this study , sera from EV71-infected children were collected to measure cross-reactive neutralizing antibody titers against different genotypes , which are critical to understand the drivers of genogroup replacement and viral diversity , and for selection of vaccine strains .
Institutional review board approvals were obtained from Chang Gung Memorial Hospital , and National Taiwan University following the Helsinki Declaration . Written informed consents were obtained from parents/guardians on behalf of all child participants . To avoid confounding effects of EV71 re-infections , historical sera were collected from young children who were under 5 years of age and infected with different EV71 genotypes in 1998 ( genotype C2 , 10 sera ) , 2000 ( genotype B4 , 5 sera ) , 2005 ( genotype C4 , 2 sera ) , 2008 ( genotype B5 , 5 sera ) , or 2010 ( genotype C4 , 3 sera ) [10] , [12]–[14] . These sera were used to measure cross-reactive neutralizing antibody titers against all 11 EV71 genotypes . Twelve strains of the 11 EV71 genotypes were used in the study , including two genotype C4 viruses which were isolated in 2005 and 2008 , respectively . Eight of these twelve viruses were isolated in Taiwan and the other four viruses ( genotype A , B2 , B3 and C3 ) had not circulated in Taiwan ( Table 1 ) . All viruses were amplified in rhabdomyosarcoma ( RD ) cells using Dulbecco's Minimum Essential Medium ( DMEM ) containing fetal bovine serum 2% v/v and penicillin/streptomycin . Virus titers ( 50% tissue culture infectious doses , TCID50 ) were determined in RD cells using the Reed-Muench method . The P1 region of the EV71 genome encodes four capsid proteins including VP1 , VP2 , VP3 and VP4 proteins , which are involved in the induction of immune response and the infection of cells [15]–[17] . Therefore , the P1 regions of 11 EV71 genotypes were sequenced to identify correlations between genetic and antigenic variations . Viral genomic RNA was extracted from 140 µL of virus culture isolates using a QIAmp Viral RNA kit ( Qiagen , USA ) according to the manufacturer's instructions . cDNA of EV71 was synthesis by SuperScript II Reverse Transcriptase ( Invitrogen , USA ) . PCR reactions were performed by specific primers and KAPA HiFi DNA Polymerase ( Kapa Biosystems , USA ) . Primers used in this study are listed in Supporting Table S1 . Nucleotide sequences of P1 regions ( 2586 bp ) were aligned and analyzed by the Mega 4 software ( Molecular Evolutionary Genetics Analysis software version 4 . 0 ) [18] . Phylogenetic trees were constructed by the neighbor-joining method using the Maximum Composite Likelihood method and the prototype CA16 strain ( CA16/G-10 ) as the outgroup virus . The reliability of the tree was estimated using 1 , 000 bootstrap replications . Nucleotide sequences analyzed in this study have been submitted to GenBank . Laboratory methods for measuring EV71 serum neutralizing antibody titers followed standard protocols [19] , [20] . Briefly , 50 µL of two-fold serially diluted sera and virus working solution containing 100 TCID50 of EV71 were mixed on 96-well microplates and incubated with RD cells . A cytopathic effect was observed in an inverted microscope after an incubation period of 4–5 days . Each serum dilution includes three replicates and the neutralization titers were read as the highest dilution that could result in a reduction of the cytopathic effect in at least two of three replicate wells . Each test sample was run simultaneously with cell control , positive serum control , and virus back titration . If the ratios of neutralizing antibody titers between different genotypes were greater than 4 , we measured neutralizing antibodies titers at least three times to confirm the accuracy of tests . Large tabular serological data are hard to summarize and are recently analyzed using antigenic cartography ( i . e . , antigenic map ) [11] , [21] , [22] . Briefly , antigenic cartography is a way to visualize and increase the resolution of serological data , such as neutralization data . In an antigenic map , the distance between a serum point S and antigen point A corresponds to the difference between the log2 of the maximum titer observed for serum S against any antigen and the log2 of the titer for serum S and antigen A . Thus , each titer in a neutralization assay table can be thought of as specifying a target distance for the points in an antigenic map . Modified multidimensional scaling methods are used to arrange the antigen and serum points in an antigenic map to best satisfy the target distances specified by the neutralization data . The result is a map in which the distance between points represents antigenic distance as measured by the binding assay [11] . In this study , an antigenic map was generated using a web-based analytic tool [22] . Neutralizing antibody titers were log transformed to calculate the geometric mean titers ( GMTs ) , and their 95% confidence intervals ( 95% CI ) . The GMTs of cross-reactive neutralizing antibody titers were further used to generate an antigenic map using a web-based analytical tool [22] . The relative positions of strains and antisera were adjusted such that the distances between strains and antisera in the map represent the corresponding ratios between homologous and heterologous neutralizing antibody titers . Differences between homologous and heterologous neutralizing antibody titers were tested for statistical significance by the nonparametric tests ( NPAR1WAY Procedure ) using SAS software ( SAS Institutes , Cary , NC ) . Nucleotide sequences analyzed in this study have been submitted to GenBank ( accession numbers JN874547–JN874558 ) .
Twenty-five sera were collected from 25 young children who were infected with EV71 genotype C2 , B4 , C4 , B5 , and C4 in 1998 , 2000 , 2005 , 2008 and 2010 , respectively . Cross-reactive neutralizing antibody titers against 11 EV71 genotypes are shown in Table 2 . Overall , all EV71-infected children had detectable neutralizing antibody titers against 11 EV71 genotypes . Interestingly , homologous neutralizing antibodies titers were not always higher than heterologous neutralizing antibody titers . As shown in Table 2 , serum neutralizing antibody titers against the homologous genotype ( C2 ) in children infected in 1998 varied over 100-fold and they were grouped into two groups ( low and high titers ) for further analysis . In addition , children infected in 2005 and 2010 were merged for further analysis because they were all infected with genotype C4 . GMTs of neutralizing antibody titers against 11 genotypes are shown in Figure 1 . Overall , children infected with genotype C2 , C4 , B4 and B5 had lower GMTs ( >4-fold difference ) against genotype A than other genotypes . In contrast , antigenic variations between genogroup B and C did not have a clear pattern . We further merged neutralizing antibody titers against different genotypes within the same genogroup to calculate GMT for further comparisons . As shown in Figure 2 , children infected with genotype C2 and C4 had similar GMT against genogroup B and C but children infected with B4 and B5 had higher GMTs against genogroup B than against genogroup C . We further constructed the antigenic map using GMT of cross-reactive neutralizing titers presented in Figure 1 . Overall , genotypes in genogroup B and C clustered together and genotype A was found to be outside of the cluster ( Figure 3 ) . To investigate the correlation between genetic and antigenic variations of EV71 genotypes , nucleotide and deduced amino acid sequences of P1 regions of 11 EV71 genotypes ( 12 viruses ) were analyzed . Pairwise comparisons of P1 regions have shown that the nucleotide ( amino acid ) differences within EV71 genogroup were 0 . 049∼0 . 151 ( 0 . 005∼0 . 015 ) for Genogroup B and 0 . 042∼0 . 135 ( 0 . 005∼0 . 013 ) for Genogroup C and the nucleotide ( amino acid ) differences were 0 . 209∼0 . 224 ( 0 . 02∼0 . 026 ) between Genogroup A and B , 0 . 21∼0 . 235 ( 0 . 018∼0 . 024 ) between Genogroup A and C , and 0 . 188∼0 . 228 ( 0 . 021∼0 . 032 ) between Genogroup B and C ( Table 3 ) . Overall , the nucleotide differences in the P1 region within genogroup were much lower than that between genogroups ( 0 . 042∼0 . 151 vs . 0 . 188∼0 . 235 ) but the differences in amino acid sequences were not as abundant as found in nucleotide sequences ( 0 . 005∼0 . 015 vs . 0 . 018∼0 . 032 ) ( Supporting Table S2 ) . Genetic variations in VP1 , VP2 , VP3 and VP4 genes were also analyzed ( Supporting Table S2 ) . Interestingly , nucleotide differences in VP1∼VP4 were similar but no amino acid differences were observed in VP4 gene , which may exclude influence of VP4 on antigenic evolution of EV71 . Phylogenetic analyses based on nucleotide sequences of the P1 , VP1 and VP1+VP3 regions are shown in Figure 4 . Overall , the phylogenetic trees generated using the P1 and VP1+VP3 regions indicated that genogroups B and C were distinct from the genotype A; however , the phylogenetic tree based on the VP1 region suggested that genogroup A is clustered with genogroup C . Overall , the phylogenetic relationship among the EV71 genotypes did not match with the antigenic relationship observed in this study . To further determine the amino acid differences related to the observed antigenic variations shown in Fig 1 and 3 , the deduced amino acid sequences of P1 regions were aligned to reveal that five amino acid signatures ( N143D in VP2; K18R , H116Y , D167E , and S275A in VP1 ) are specific for genogroup A and may be related to the observed antigenic variations ( Figure 5 ) .
EV71 has one single serotype as measured by hyperimmune animal antiserum but antigenic variations have been reported recently in human studies [9]–[11] . Using sera collected from young children with primary infection of genotype B5 , two studies detected partial antigenic differences between genogroup B and C but not between viruses in the same genogroup ( B5 and B4 viruses ) [9] , [10] . Kung et al . did not detect significant antigenic differences between genotypes B4 and C4 viruses using acute-phase sera from EV71 inpatients [23] . A serological survey in healthy Japanese children and adults detected partial antigenic differences between genotype B5 and A viruses but not among different genotypes in genogroup B and C that had previously circulated in Japan [24] . By constructing an antigenic map using 14 children sera , however , Huang et al . detected antigenic differences between genogroup B and C , and also between B5 and B4 viruses [11] . In a monkey study , Arita et al . [25] found that monkeys immunized with live-attenuated EV71 vaccine ( genotype A ) induced similar ( <4-fold difference ) antibody responses against genotype B1 but lower ( ≧4-fold difference ) antibody responses against genotype B4 , C2 and C4 . In our study , we found that children infected with genotype C2 , C4 , B4 and B5 had lower GMTs ( ≧4-fold difference ) against genotype A than other genotypes but antigenic variations between genogroup B and C did not have a clear pattern , which is different from the Huang study [11] . It is hard to compare different studies which had small sample size and employed different human sera and laboratory procedures , in particular the cell lines ( RD cells vs . Vero cells ) and virus strains used in the neutralization assay . A network to harmonize laboratory procedures including standard sera and viruses is required to make the comparison possible . Moreover , the clinical and epidemiological significance of the antigenic variation requires longitudinal serological studies to clarify . Most clinical studies , including our study faced the limitation of small sample size due to the difficulty of collecting large amounts of serum samples from young children . Ideally , suitable animal models should be developed to generate a panel of antisera for monitoring EV71 antigenic variations , as ferrets served for influenza surveillance [26] . Representative EV71 clinical isolates could be selected for monitoring antigenic variations using the animal antisera . The clinical isolates with significant antigenic variations detected using animal antisera would be further evaluated using children post-infection sera . Currently , five EV71 vaccine candidates are under evaluation in clinical trials , including three genogroup C viruses and two genogroup B viruses [27] . Based on the cross-reactive neutralizing antibody presented in the current study , genogroup B and C viruses are expected to induce protective neutralizing antibodies against genogroup B and C viruses but not genogroup A viruses . Interestingly , genogroup A viruses have disappeared for over 35 years but re-emerged in China in 2008 . In an investigation of a HFMD outbreak in central China in 2008 , Yu et al identified five EV71 isolates which were closely related to genotype A based on analysis of VP1 genes but these genogroup A viruses did not spread widely [6] . Reasons for the reemergence of genotype A in central China are not clear , and the full genomic sequences of the isolates should be performed to clarify the issue . Recently , novel genotype C2-like viruses were detected in Taiwan in 2008 and children infected with genotype C4 , C5 , B4 and B5 viruses had much lower ( >100-fold ) serum cross-reactive neutralizing antibody titers against the novel C2-like virus than against the homologous viruses . Interestingly , these novel C2-like viruses were recombinants of genotype C2 and B3 viruses but they did not spread widely [9] . Based on historical poliovirus studies , immunodominant neutralizing epitopes mainly locate on VP1 and VP2 proteins . Recently , binding sites of two EV71 mice neutralizing monoclonal antibodies were identified using synthetic peptide technology to locate at amino acid position 211–225 of VP1 protein and amino acid position 136–150 of VP2 protein , respectively [16] . The importance of these linear epitopes in the human immune response is not clear . In the current study , we combined human serological data and viral genetic sequence data to identify five amino acid positions ( 4 on VP1 protein and 1 on VP2 protein ) related to antigenic variations . Only one of these five positions ( VP2-143 ) was also identified in the mice monoclonal antibody studies . The clinical significance of these five positions needs to be verified using reverse genetics to generate mutant viruses . Recently , the 3-dimensional structures of EV71 capsid proteins have been published [28] , [29] . Structural studies elucidating interaction between EV71 capsid proteins and neutralizing antibodies will help understand the mechanism of vaccine-induced immunity and design better vaccines . Traditionally , the phylogenetic relationship of EV71 genotypes has been widely analyzed using VP1 nucleotide sequences [1] . Interestingly , a recent study found that the VP1-based phylogenetic tree is not similar to the complete genome-based phylogenetic tree [7] . Our study also found that the phylogenetic trees based on VP1 and P1 nucleotide sequences differ slightly . Specifically , genogroup A is close to genogroup C in the VP1-based phylogenetic tree but this relationship was not found in the P1-based phylogenetic trees . It is well known that enteroviruses including EV71 frequently recombine at the junction of structural ( P1 ) and non-structural ( P2 or P3 ) genes [8] , [30] . Therefore , the P1 gene is suitable for phylogenetic analysis but the complete genome is required for detection of gene recombination . However , the P1 gene ( about 3000 nucleotides ) is much larger than the VP1 gene ( about 890 nucleotides ) and the P1 gene may not be readily available . The combined VP1+VP3 gene ( about 1600 nucleotides ) is much shorter than the P1 gene but could generate a similar phylogenetic tree to that based on the P1 gene . Overall , the VP1 gene is good enough for defining genotypes of genogroup B and C viruses , but it would be better to analyze the phylogenetic relationship between genogroup A viruses and other genogroup viruses based on the VP1+VP3 or P1 genes . From an evolutionary perspective , a recent analysis of 628 EV71 VP1 sequences estimated that EV71 emerged in the human population around 1941 and evolved more quickly in the past 20 years [4] . It is unclear why EV71 has evolved more quickly in the past 20 years . Recombination , being a common occurrence among enteroviruses , might be the likely explanation for the emergence of EV71 , but it would require full genome analysis to better understand the mechanism of EV71 evolution , which is critical to long-term success of EV71 vaccination programs .
|
Recently , enterovirus 71 ( EV71 ) has caused life-threatening outbreaks in tropical Asia . EV71 has one single serotype but can be phylogenetically classified into 3 main genogroups and 11 genotypes ( A , B1∼B5 and C1∼C5 ) . In Taiwan , nationwide EV71 epidemics with different predominant genotypes occurred in 1998 ( C2 ) , 2000–2001 ( B4 ) , 2004–2005 ( C4 ) , and 2008 ( B5 ) . In this study , historical sera from children infected with these 4 genotypes were collected to measure cross-reactive neutralizing antibody titers against 11 genotypes . In addition , amino acid sequences of P1 proteins of the tested viruses were compared . Serology data showed that children infected with genogroup B and C consistently have lower neutralizing antibody titers against genogroup A ( >4-fold difference ) . Antigenic variations between genogroup B and C could be detected but did not have a clear pattern . Five amino acid signatures are specific for genogroup A and may be related to the observed antigenic variations . Vaccine development should monitor the antigenic and genetic variations to select vaccine strains .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"infectious",
"diseases",
"pediatrics",
"and",
"child",
"health",
"pediatrics"
] |
2013
|
Cross-reactive Neutralizing Antibody Responses to Enterovirus 71 Infections in Young Children: Implications for Vaccine Development
|
Rabies is a viral zoonosis affecting mammal species and causes large economic losses . Included among the neglected diseases , it is still insufficiently addressed by governments and the international community , despite formal surveillance and control programs . This study used a dataset of 10 , 112 rabies diagnoses in animals provided by the Brazilian passive surveillance system from 2001 to 2012 . The positivity rate of the tested samples was 26 . 4% , and a reduction in the total samples sent during the last six years was observed . The kernel density map indicated case concentration in the south region and a decrease in density of rabies cases in the second period studied ( 2007 to 2012 ) . The directional trend of positive rabies diagnoses remained in the south region , as shown by the standard deviational ellipse . The spatial scan statistic identified three large clusters of positive diagnoses , one in the first period ( 2001-2006 ) and two in the second period ( 2007-2012 ) , indicating an expansion of risk areas . The decrease in rabies cases from 2006 to 2012 does not necessarily reflect lower viral circulation or improvement in actions by epidemiological surveillance; this decrease could indicate a deficiency in epidemiological surveillance during the observation period due to the increase in the silent areas . Surveillance should maintain an increasing or constant number of tests during the years in addition to a reduction in the number of outbreaks of rabies , which would indicate a lower positivity rate . The findings in this study indicate deterioration in the effectiveness of the passive surveillance for rabies . The number of rabies cases , total number of tests performed and positivity rate are good indicators for evaluating passive surveillance . This paper can function as a guide for the assessment and improvement of the actions in passive surveillance of rabies .
Rabies is a viral zoonosis due to a Lyssavirus infection associated with neurological expression due to encephalitis or meningoencephalitis . It is one of the oldest known infectious diseases in the world but remains a neglected zoonotic disease , insufficiently addressed by governments and the international community [1] , [2] . Most countries in the Americas have been declared free of human cases of dog-transmitted rabies , there is now only notification of human rabies transmitted by dogs in Bolivia , Peru , Honduras , Haiti , Dominican Republic , Guatemala and some states in northern and northeastern Brazil [3] . The urban human rabies , transmitted by dogs and cats , has decreased from 73 cases in 1990 to 17 cases in 2003 in Brazil [4] . Currently , vampire bat-transmitted rabies is a major public health problem in the subtropical and tropical areas in the Americas , from Mexico to Argentina [5] . However , in terms of rabies cases transmitted by all species in the period from 2001–2012 , 129 human rabies cases were notified [6] . Particularly , cases in which humans were bitten by bats have increased in Brazil [7] . The urban cycle , particularly including domestic dogs , has been controlled [2] . However , sylvatic cycles are expanding with an increasing number of diagnoses in species such as a fox Cerdocyon thous and a common marmoset Callythrix jacchus [8] in Brazil , but vampire bats also play a main role in rabies transmission [9] . Consequently , the occurrence of rabies virus in vampire bats is reflected by the cattle rabies incidence [10] [11] . There are three vampire bat species , Desmodus rotundus , Diphylla ecaudata and Diaemus youngii . Two species feed only on blood of wild birds , and one species , D . rotundus , feeds on livestock and could be a transmitter of Lyssavirus causing bovine paralytic rabies , a source of large economic losses [12] , [13] . Outbreaks in livestock transmitted by vampire bats were first observed between 1906 and 1908 in the State of Santa Catarina in Brazil , when approximately 4000 cattle and 1000 horses and mules died due to paralytic rabies [12] . Since this episode , Brazil has applied measures to control the bat-transmitted rabies . A total of 9 , 277 rabies cases were reported in Brazil from 2002 to 2009 ( 88 . 0% in cattle , 10 . 0% in horses and 2 . 0% in other species ) [14] . Currently , the use of warfarin to reduce the vampire bat population and the livestock vaccination against rabies are regularly conducted [12] . Rabies in Brazil cause large economic losses in the productive sector due to animal deaths and in the public sector through costs in surveillance and control programs [15] , [16] . Brazil has 211 . 28 million bovines , the second largest herd in the world , and the country also contains other herbivore species . Minas Gerais State has the largest equine population , the second largest cattle population , and the highest milk production [17]; therefore , it can be used as a model to evaluate rabies surveillance in Brazil . The Brazilian Program for Rabies Control in herbivores aims to prevent the disease in cattle by focusing on the control of vampire bats ( Desmodus rotundus ) , strategic vaccination and active and passive epidemiological surveillance [18] , [12] . However , cases due to bat-transmitted rabies are largely underreported . The aim of this paper was to understand the spatial and temporal distribution of the rabies cases and to analyze this information to confront the robustness of these results with the effectiveness of passive surveillance to identify trends in disease behavior and the dynamics of the surveillance .
Minas Gerais is located in the southeastern region of Brazil , has an area of 586 , 528 Km2 , and includes 853 municipalities , politically grouped in 66 small aggregate regions ( SARs ) and 12 large aggregate regions ( LARs ) [19] . SARs and LARs were adopted as basic areas for this study ( Fig . 1 ) to facilitate the analysis interpretation . This study was developed using data from the government agency responsible for animal health in the State ( Instituto Mineiro de Agropecuária—IMA ) , which covers diagnoses from animals suspected of rabies between 2001 and 2012 . The samples originated from all regions of the State and were sent voluntarily by farmers or by both private and public veterinarians . The variables in the database were animal species , year , month , location ( municipality ) and the rabies test results . The results were geo-referenced using ArcGIS 9 . 3 [20] , digital map files from the Brazilian Institute of Geography and Statistics ( IBGE ) with the political administrative borders of the large aggregate regions ( LARs ) , small aggregate regions ( SARs ) and municipalities [19] . The rabies surveillance database in the GIS platform was used to map the distribution of the disease and also applied other spatial and temporal analysis . To identify trends in the disease behavior and the dynamics of the surveillance , tables and graphs were used to describe these patterns . The study was divided into two sub periods to highlight patterns; the first study period was from 2001 to 2006 , and the second was from 2007 to 2012 . The samples sent to the government animal health laboratory were subjected to the direct immunofluorescence technique and to the biological proof ( inoculation in mice or cells ) . Differential diagnoses were performed by histopathology and immunohistochemistry based on the guidelines of a specific manual with techniques for herbivore rabies control [18] . A standard deviational ellipse , using 1 standard deviation , was calculated to find any directional trend among positive results in the municipalities considering all species . The Kernel density estimation ( search radius = 100 Km ) used to assess the intensity of positive results on the surface ( km2 ) , which allowed us to identify the areas of higher concentration of cases along the smoothed surface that was generated , using ArcGIS version 9 . 3 . 1 [20] . The space-time scan statistical analysis ( Poisson model ) to identify spatial clusters was applied only on herbivorous species in which it was possible to know its animal population ( bovine , equine , ovine and caprine ) , the analysis was performed using SatScan version 9 . 2 [21] , and the parameters were high rates , months as the time variable and the maximum spatial cluster size with upper limit 50% of the population at risk . The studies on epidemiological surveillance of rabies involve three epidemiologic indicators: number of rabies outbreaks , the total number of tests performed and the relationship between positive tests and total tests ( the positivity rate ) using data from all species [22] . These indicators were selected due to the passive nature of the rabies data used to analyze surveillance in Minas Gerais State . The relationship between positive results and the total tests performed was used to measure the reporting level of the rabies surveillance . Values equal or close to 1 express less surveillance actions , and values equal or close to 0 indicate an increase in surveillance activities [22] . This ratio was calculated for each municipality and for each month in the respective years , and the result was consolidated on a graph and represented on maps . A projected epidemiological scenario was estimated based on the analysis of the best level of surveillance in the period studied; from 2001 to 2005 , rabies surveillance achieved the recommended positivity rate without compromising the total number of tests performed . Therefore , the average positivity rate from 2001 to 2005 was used in the subsequent years in conjunction with the true total positive samples sent each year , resulting in the expected total of tests that should have been performed to ensure adequate surveillance in the projected scenario .
The passive surveillance system for rabies in Minas Gerais performed 10 , 112 rabies diagnoses between 2001 and 2012 ( Table 1 ) . Among the total results , 2 , 670 were positive , corresponding to 26 . 4% ( 25 . 5% to 27 . 2% , confidence interval of 95% ) , and 7 , 442 were negative . In addition to herbivore species , which are important in Brazilian livestock production , additional species such as swine , canine , feline , bats and other wild animals were also included in the analysis ( excluding only the analysis using the Poisson Model ) . The percentages of positive animals for each species were the following: bats ( 2 . 42% ) , bovine ( 40 . 6% ) , equine ( 28 . 6% ) , caprine ( 23 . 8 ) , ovine ( 12 . 0% ) , swine ( 7 . 3% ) , canine ( 3 . 4% ) , feline ( 0 . 8% ) , and others ( 7 . 9% ) . Among the 853 municipalities in Minas Gerais , 361 ( 42 . 3% ) did not have any rabies diagnoses in the 12 years of study , 212 ( 24 . 8% ) of which have not sent samples for diagnoses; these municipalities were classified as silent areas for rabies ( Table 2 ) . Between 2001 and 2006 , 384 municipalities ( 45 . 0% ) diagnosed positive cases; 297 ( 34 . 8% ) did not send samples , and of the 7 , 203 samples sent in the period , 1 , 759 ( 24 . 4% ) were diagnosed as positive . From 2007 to 2012 , 317 municipalities ( 37 . 2% ) were positive and 359 ( 42 . 0% ) did not send samples . Of the 2 , 909 samples in the second period , 911 ( 31 . 3% ) were diagnosed as positive . Among the municipalities diagnosed as positive in the first period , 80 ( 20 . 8% ) did not send samples in the second period of the study and were classified as silent areas . The standard deviational ellipse indicated a tendency in the south of the State , in the direction west-east ( Fig . 2 ) . As observed in the period from 2001 to 2006 , the distribution was more concentrated in western Minas Gerais , mostly in the Triângulo Mineiro/Alto Paranaíba region ( 11 ) . However , between 2007 and 2012 , the distribution was more concentrated in the east . All regions reported cases of rabies , and fewer positive results were detected in northern and eastern Minas Gerais State ( Fig . 3A and 3B ) . Four LARs contained 66% of the municipalities ( 140 of 212 ) that had not sent any samples in the studied period; 48 were located in LAR-2 , 30 in LAR-1 , 9 in LAR-3 and 53 inside the LAR-7 . From 2001 to 2006 , 41 SARs included 93 municipalities that showed at least six cases ( Fig . 3A ) . From 2007 to 2012 , only 22 SARs included 32 municipalities that reported at least six cases ( Fig . 3B ) . The mean of cases reported in the first period was 1 . 24 cases/municipality/month , and 61 municipalities reported over 3 cases/month; in the second period , the mean was 1 . 13 cases/municipality/month , and 11 municipalities reported over 3 cases/month . The trend of diagnoses over time is presented in Fig . 4 . In the first period of the study , the positive results remained above 200 per year , with a mean of 293 cases per year . The negative results increased in the first three years and stabilized until 2005 . In 2006 and afterwards , the negative results decreased considerably , and the mean number of negative results was 907 per year . In the second period of the study , a different pattern was found in the results; the positive results were below 200 per year , and the mean was 151 cases per year . The negative result presented an accentuated decreased tendency , and the mean number of negative diagnostics was 333 tests/year . The kernel density map ( Fig . 5 ) showed a high concentration of cases in the southern region of the State during both testing periods . However , the case concentration reduced in the second period . The spatial-temporal analysis with Poisson model ( Fig . 5 ) identified a cluster in the southern region in the first period of the study . In the second period , two clusters were identified , including southern and eastern Minas Gerais . The kernel density map presented a reduction in the case concentration of rabies during the second period . However , the spatial scan statistic recognized an increase in the rabies risk area . The LARs with high concentration of cases and cluster areas in the second period ( using the two techniques ) were the following: Jequitinhonha ( 1 ) , Vale do Mucuri ( 3 ) , Vale do Rio Doce ( 7 ) , Metropolitana de Belo Horizonte ( 10 ) , Zona da Mata ( 9 ) , Oeste de Minas ( 8 ) , Campo das Vertentes ( 6 ) and Sul/Sudoeste de Minas ( 12 ) . The positivity rate of rabies diagnoses is spatially represented for each municipality in Fig . 6 . In the second period of the study ( Fig . 6B ) , areas with high positivity rates ( above 0 . 66 ) and areas with no reports of the disease increased compared to the first period ( Fig . 6A ) . The positivity rate of rabies diagnoses from 2001 to 2012 is also represented in Fig . 7 , synthesizing the behavior of the passive surveillance over time . Until 2005 , the indicators were favorable for rabies surveillance . In contrast , from 2006 to the end of the study , there was an increase in the positivity rate and a reduction in total samples sent . In Fig . 8 , the projected scenario was presented to ensure adequate rabies surveillance , considering the epidemiologic indicators from 2001 to 2005 , number of rabies outbreaks , the total number of tests performed and the positivity rate .
Rabies in Brazil is a complex situation and needs more research , and various socioeconomic and environmental factors need to be considered . Nevertheless , the method applied in this study allowed us to establish priorities for epidemiologic surveillance . The combined use of the total number of tests performed , number of rabies cases , and the positivity rate are good indicators to evaluate passive rabies surveillance . The approach in this paper using spatial analysis techniques combined with other sources of information serves as a guide for improving the actions in passive rabies surveillance . The method can also be used to evaluate the effectiveness of the actions by passive rabies surveillance and to improve the strategies already adopted by government programs to control and prevent the disease in Brazil and in other countries .
|
The approach proposed in this study can provide a valuable contribution to strengthen the efforts to reduce rabies incidence , a relevant public health problem worldwide . In Brazil during the period from 2001–2012 , 129 human cases of rabies were notified , with 128 deaths . It is imperative to decrease rabies virus circulation both in domestic and wildlife species , which has been reinforced by many studies demonstrating that rabies incidence is often much higher than the official reports . This study used a dataset of 10 , 112 rabies diagnoses in animals provided by the Brazilian passive surveillance and applied spatial analysis in addition to many indicators to highlight high positivity rate areas and accurately identify silent areas and regions where underreporting occurs . This type of approach is vital to establish priorities for epidemiologic surveillance strategically and to precisely focus actions in the required regions . These spatial analysis techniques combined with other sources of information can also be used to evaluate the effectiveness of passive surveillance for rabies and improve the strategies already adopted by government programs to control and prevent the disease . We consider our results an important achievement in envisioning effectiveness in the process to eliminate rabies worldwide .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[] |
2015
|
Trends in Animal Rabies Surveillance in the Endemic State of Minas Gerais, Brazil
|
The Eph receptor tyrosine kinases ( RTKs ) are regulators of cell migration and axon guidance . However , our understanding of the molecular mechanisms by which Eph RTKs regulate these processes is still incomplete . To understand how Eph receptors regulate axon guidance in Caenorhabditis elegans , we screened for suppressors of axon guidance defects caused by a hyperactive VAB-1/Eph RTK . We identified NCK-1 and WSP-1/N-WASP as downstream effectors of VAB-1 . Furthermore , VAB-1 , NCK-1 , and WSP-1 can form a complex in vitro . We also report that NCK-1 can physically bind UNC-34/Enabled ( Ena ) , and suggest that VAB-1 inhibits the NCK-1/UNC-34 complex and negatively regulates UNC-34 . Our results provide a model of the molecular events that allow the VAB-1 RTK to regulate actin dynamics for axon guidance . We suggest that VAB-1/Eph RTK can stop axonal outgrowth by inhibiting filopodia formation at the growth cone by activating Arp2/3 through a VAB-1/NCK-1/WSP-1 complex and by inhibiting UNC-34/Ena activity .
During development , axons navigate to their final destination by interpreting extracellular guidance cues through their growth cone . The Eph receptor tyrosine kinases ( RTKs ) and their ephrin ligands are involved in directing axons to their proper location [1] , [2] . Studies in vertebrate systems have identified a number of effectors in the Eph RTKs signaling pathway in axon guidance [2] . However , the molecular mechanism of how Eph RTKs regulate axon guidance is still incomplete . This is partly due to the large number of Ephrins and Eph RTKs that can engage in crosstalk [2] , [3] . The presence of a single Eph RTK , VAB-1 , in Caenorhabditis elegans can simplify the analysis of the signal transduction events from the receptor . The C . elegans VAB-1 Eph RTK is required for various aspects of neuronal development , including neuroblast movements , and axon guidance [4] , [5] , [6] , [7] . The molecules involved in VAB-1 signaling in axon guidance are still unknown . To resolve this issue , we used a genetic suppressor approach as well as a physical protein interaction approach and identified NCK-1 , WSP-1/N-WASP , UNC-34/Ena , and the Arp2/3 complex as molecules regulated by VAB-1 signaling in axon guidance . The Nck adaptor proteins are known actin cytoskeleton regulators , and have been shown to function downstream of several axon guidance receptors including Robo , Dcc and the Eph RTKs [8] , [9] , [10] , [11] . Although the function of Nck has been studied in various organisms , the biological function of NCK-1 in C . elegans has only been recently explored [12] . Furthermore , what molecules interact with the C . elegans NCK-1 is still unknown . The WASP protein family ( WASP and N-WASP ) are scaffolds that integrate multiple signaling pathways , leading to the formation of short branched actin filaments through the activation of the Arp2/3 complex [13] . The C . elegans N-WASP homolog , WSP-1 , functions in neuronal cell migration and axon guidance [14] , [15] . However , a connection between WSP-1 and a guidance receptor has not yet been established . The Ena/VASP proteins are involved in actin-dependent movements including neuronal migration and axon guidance , and are known for their role in promoting filopodia formation [16] . In C . elegans , the Ena/VASP homolog UNC-34 is required for proper neuronal cell migration , axon guidance and filopodia formation [14] , [17] , [18] , [19] , [20] . Previous work has shown that Ena/VASP proteins are versatile in their developmental roles and function in both repulsive and attractive cues . For example Ena/VASP are effectors of receptors for repulsive cues such as SAX-3/Robo , UNC-5/Netrin receptor and EphB4 , but they can as also act as effectors for attractive cues downstream of receptors such as UNC-40/DCC [21] , [22] , [23] , [24] , [25] . The Arp2/3 complex is a conserved family of actin nucleators and when activated results in the formation of an elaborate network of branched actin filaments similar to those found in lamellipodia [26] , [27] . In C . elegans , the Arp2/3 complex is required for axon guidance , and the initiation of growth cone filopodia downstream of an unidentified axon guidance signal [15] , [20] . In this paper , we describe some of the molecular events that allow the VAB-1 Eph RTK to regulate actin dynamics for axon guidance . We provide genetic and biochemical evidence to show that VAB-1 signals through NCK-1 and WSP-1/N-WASP , and negatively regulates UNC-34/Ena . We propose a model for PLM ( Posterior lateral microtubule ) axon termination whereby the VAB-1 Eph RTK is able to prevent axon extension by inhibiting growth cone filopodia formation . This is accomplished by negatively regulating the activity of the filopodia elongator UNC-34/Ena , and simultaneously activating Arp2/3 through a VAB-1/NCK-1/WSP-1 complex .
To identify VAB-1 Eph RTK effectors , we utilized transgenic animals carrying mec-4::myr-vab-1 ( quIs5 ) which encodes a constitutively active VAB-1 tyrosine kinase ( myristoylated-VAB-1 termed MYR-VAB-1 ) in the mechanosensory neurons [6] . In wild-type young adults , PLM neuron cell bodies are located in the tail region and have axons that stop at the centre of the animal ( Figure 1A ) . We previously showed that myr-vab-1 caused neuronal defects in the mechanosensory neurons , in particular the premature termination of PLM axons ( Figure 1A , 1B ) [6] . Since the MYR-VAB-1 behaves as a constitutively active VAB-1 RTK , we reasoned that mutations in effectors of the VAB-1 signal may suppress the neuronal defects . We used a candidate gene approach to examine genes with known roles in axon guidance and tested whether loss-of-function mutations could suppress the myr-vab-1 PLM premature termination phenotype . We identified nck-1 as a candidate effector of VAB-1 Eph RTK signaling . The nck-1 ( ok694 ) mutation partially suppressed the PLM axon premature termination ( Figure 1B ) , indicating that other effectors are involved in the MYR-VAB-1 signaling . The C . elegans genome encodes for only one nck-1 adaptor protein , and is most similar to the human Nck2 and Drosophila DOCK [12] . NCK-1 has all the domain features of the NCK adaptor proteins , including three SH3 domains followed by a single SH2 domain . We previously reported that the deletion allele nck-1 ( ok694 ) is predicted to be a null allele , thus all of our genetic work was carried out using the ok694 allele [12] . If NCK-1 is an effector of VAB-1 signaling then we would expect the nck-1 loss-of-function mutation to have a phenotype similar to that of the vab-1 loss-of-function . Indeed , previous work showed that both vab-1 and nck-1 mutants have similar neuronal defects , including an overextension in PLM axons ( Figure 1C ) [6] , [7] , [12] . To further confirm that nck-1 and vab-1 are in the same pathway in the PLM neurons , we analyzed the effect of the double mutation on the PLM axons . The vab-1; nck-1 double mutation did not enhance the PLM over extension phenotype ( Figure 1C ) , indicating that NCK-1 and the VAB-1 Eph receptor function in the same pathway to guide the PLM axons . To determine if the PLM defects observed in vab-1 and nck-1 animals were present at an earlier stage , we examined the PLMs of the first larval stage ( L1 ) ( see Methods ) . Wild-type L1s had PLM axons that were 103–114 µm long , and terminated at a region anterior to the tip of the ALM cell body ( 93% ) and is consistent with previous reports for L1 PLM lengths [28] ( Figure 2A ) . Both vab-1 and nck-1 animals had PLM axons that significantly overgrew beyond the wild-type termination region ( Figure 2A , 2B ) . This indicates that VAB-1 and NCK-1 are required at an early stage to prevent PLM axons from overgrowing beyond their normal termination region . We also showed that 96% of L1 myr-vab-1 transgenic animals had PLM axons that were undergrown when compared to wild-type ( Figure 2A , 2C ) . The PLM undergrowth defects caused by MYR-VAB-1 were significantly reduced by nck-1 ( ok694 ) ( 57% ) ( Figure 2C ) . These results are consistent with our analysis carried out in early adults , and further confirm that NCK-1 is an effector of VAB-1 signaling in PLM axon guidance . We previously showed that NCK-1 is expressed in various tissues including the nervous system [12] . In addition , like VAB-1 , NCK-1 can function cell autonomously in the mechanosensory neurons for PLM axon guidance [6] , [12] . If NCK-1 and VAB-1 function in the same pathway during neuronal development , then they should be localized in the same cells . Indeed , NCK-1 and VAB-1 were co-localized in some of the neurons , consistent with the role of NCK-1 as an effector of VAB-1 ( Figure 3A , 3B ) . However , the expression pattern of VAB-1 and NCK-1 did not overlap exactly , suggesting that both NCK-1 and VAB-1 have independent roles during development ( Figure 3A ) . Expression of NCK-1-GFP and activated VAB-1 ( MYR-VAB-1 ) in the mechanosensory neurons showed that NCK-1 did co-localize with activated VAB-1 in the PLM axon and cell body ( Figure 3B ) . In a parallel approach we used yeast two-hybrid screens to identify effectors of VAB-1/Eph RTK signaling and identified the full length NCK-1 as a binding partner of the VAB-1 intracellular kinase region . Yeast two-hybrid analysis showed that the NCK-1 SH2 domain is sufficient to bind VAB-1 and that VAB-1 tyrosine Y673 is crucial for the interaction with the NCK-1 SH2 domain ( Figure 4A ) . To further confirm the NCK-1/VAB-1 interaction we used GST-pull down assays . Deletion analyses confirmed that the SH2 domain is necessary and sufficient to bind VAB-1 ( Figure 4B ) . Furthermore , the NCK-1 interaction required an active tyrosine VAB-1 kinase since the NCK-1 SH2 domain did not bind a kinase inactive version of VAB-1 ( G912E ) ( Figure 4C , 4D ) . Since SH2 domains are known to bind phosphotyrosines we wanted to test how specific the NCK-1 SH2 domain is for VAB-1 . We found that four other SH2 domains ( MIG-10 , SEM-5 , ABL-1 , VAV-1 ) were unable to bind VAB-1 ( Figure 4E ) . In summary , NCK-1 interacts with VAB-1 in a kinase dependent manner , the interaction is mediated via the NCK-1 SH2 domain and the VAB-1 Y673 juxtamembrane tyrosine , and VAB-1 has high specificity for the NCK-1 SH2 domain . How does VAB-1 cause the PLM to stop once the VAB-1 Eph RTK is activated and adaptor proteins such as NCK-1 bind the receptor ? A previous report indicated that Ena/VASP was required for repulsion caused by EphB4 signaling in fibroblasts , but it was unclear how the signal was conveyed [24] . The Ena/VASP family are composed of an N-terminal EVH1 domain , a central PRO region and a C-terminal Ena/VASP homology II domain ( EVH2 ) [16] . We asked if NCK-1 could be the link between the Eph RTK and Ena/VASP . We first tested if NCK-1 and UNC-34 can directly interact . In vitro binding assays with bacterially expressed NCK-1 and UNC-34 confirmed that both proteins do indeed physically interact ( Figure 4F , 4G ) . Furthermore , we found that the PRO-EVH2 domains are required together to bind NCK-1 ( data not shown ) . We also showed that all three NCK-1 SH3 domains were able to bind UNC-34 ( Figure 4G ) . While nck-1 and vab-1 animals have overextended PLM axons , unc-34 animals have the opposite phenotype and have PLM axons that terminate prematurely ( Figure 1D , Figure 2C ) . This suggests that UNC-34 is involved in PLM axon extension , and reflects a known role of Ena/VASP in actin filament formation and elongation [16] , [29] . To understand the genetic nature of the interaction between nck-1 and unc-34 , we analyzed the nck-1 ( ok694 ) ; unc-34 ( e566 ) double and found that nck-1 partially suppressed the unc-34 PLM termination defect , while unc-34 did not suppress the nck-1 overgrowth ( Figure 1D , and data not shown ) . This suggests that , in PLM axon outgrowth , unc-34 may negatively regulate nck-1 . To provide further evidence for this genetic interaction we over expressed NCK-1 ( mec-4::nck-1 ) in the PLM neurons of unc-34 ( e566 ) animals and this resulted in a synergistic enhancement of the unc-34 PLM termination phenotype ( Figure 1D ) . Although we cannot conclusively rule out that nck-1 inhibits unc-34 , overall , our results suggest that UNC-34 can inhibit the function of NCK-1 and may do so by physically binding to it . Since UNC-34 and NCK-1 physically interact , we wanted to examine whether VAB-1 , NCK-1 and UNC-34 could form a complex in vitro . Surprisingly , although UNC-34 can bind strongly to NCK-1 , the introduction of VAB-1 abolished the binding between UNC-34 and NCK-1 ( Figure 5A Lane 4 , 5 ) . This result suggests that VAB-1 might be inducing its effect at the growth cone membrane by relieving the inhibition of NCK-1 that is caused by UNC-34 . To provide in vivo support of this we over expressed UNC-34 in the mechanosensory neurons ( mec-4::unc-34 ) and it significantly reduced the MYR-VAB-1 PLM premature termination phenotype ( Figure 1B ) . To gain more insight into the interaction between VAB-1 and UNC-34 , we sought to analyze the effect of the vab-1;unc-34 double on PLM axons . We found that the vab-1;unc-34 double mutant is synthetic lethal ( data not shown ) , so we used a mechanosensory specific unc-34 RNAi approach ( see experimental procedures ) . The unc-34 ( RNAi ) strain had PLM termination defects that were similar to unc-34 ( e566 ) ( Figure 1D ) . Analysis of the vab-1;unc-34 ( RNAi ) double showed that reducing the levels of unc-34 can rescue the PLM overextension defects seen in vab-1 ( dx31 ) ( Figure 1C ) , which is consistent with vab-1 inhibiting unc-34 function . Since the genetic data suggested that vab-1 negatively regulates unc-34 , we questioned if the activation of VAB-1 could affect the expression and/or localization of UNC-34 . Induction of MYR-VAB-1 via heat shock promoter did not change the localization of UNC-34 , but instead resulted in the reduction of UNC-34::GFP levels compared to wild-type animals ( Figure 5B ) . To test whether VAB-1's negative regulation can function cell autonomously in the PLMs , we expressed UNC-34::GFP only in the mechanosensory neurons ( via mec-4 promoter ) and it is expressed at high levels . When we introduce constitutively active VAB-1 only in the touch neurons ( mec-4::myr-vab-1 ) it reduced the UNC-34::GFP levels significantly ( Figure 5C ) . In summary , our binding assays and genetic analyses show that VAB-1 activation results in binding NCK-1 which in turn blocks the UNC-34 binding to NCK-1 , freeing NCK-1 from the negative influence of UNC-34 and in addition VAB-1 negatively regulates UNC-34 protein levels . Since mammalian Nck is known to physically bind and activate N-WASP to regulate actin filaments through the Arp2/3 complex [8] , [30] , [31] , we questioned whether VAB-1 is linked to the cytoskeleton through WSP-1/N-WASP . If WSP-1 acts downstream of VAB-1 , then the wsp-1 mutants should suppress the PLM termination defect caused by MYR-VAB-1 . Two wsp-1 alleles are predicted to affect the WSP-1 protein . The wsp-1 ( tm2299 ) is not well characterized , but is homozygous lethal and is predicted to be a null allele . The embryonic lethality is due to wsp-1 pleiotropy as WSP-1 is also required for cytokinesis during embryogenesis [14] . The wsp-1 ( gm324 ) allele is a well characterized deletion that removes exons 2 and 3 , furthermore , no WSP-1 protein nor mRNA can be detected , therefore wsp-1 ( gm324 ) is a strong loss-of-function allele [14] . wsp-1 ( gm324 ) displays some embryonic and larval lethality but can be maintained as a homozygote [14] , [15] , [32] . We chose to use the wsp-1 ( gm324 ) allele as it allowed us to bypass the embryonic lethality associated with the wsp-1 null allele . We found that wsp-1 ( gm324 ) could significantly suppress the MYR-VAB-1 PLM termination defect in young adults and L1s ( Figure 1B , Figure 2C ) . If WSP-1 is an effector of VAB-1 signaling then we would expect to see neuronal defects similar to vab-1 animals . It was previously reported that the wsp-1 ( gm324 ) had weak axon guidance defects , such as in the PDE and VD/DD neurons [15] . We report here that approximately 50% of wsp-1 ( gm324 ) animals have overextended PLM defects in young adults , and 42% PLM axon overgrowth in L1s ( Figure 1C , Figure 2B ) . Since the wsp-1 PLM overextension frequency is much greater than vab-1 ( Figure 1C ) , it implies that WSP-1 also functions independent of VAB-1 for PLM axon guidance . We also found that the vab-1 ( dx31 ) ;wsp-1 ( gm324 ) double mutants are synthetic lethal ( data not shown ) , which is consistent with WSP-1 functioning in parallel pathways with VAB-1 . The presence of WSP-1 in the VAB-1 signaling pathway suggests the possibility that the PLM termination phenotype caused by MYR-VAB-1 could be due to the activation of the Arp2/3 complex . WSP-1 , like its mammalian counterpart , is composed of an N-terminal Ena/VASP homology I domain ( EVH1; also known as WASP-homology-1 domain ( WH1 ) ) , a central section containing a basic region ( BR ) , a GTPase binding domain ( GBD ) and a proline-rich region ( PRO ) , and a C-terminal with two verprolin homology domains ( V; also known as WH2 ) , a cofilin homology domain ( C ) and an acidic domain ( A ) [13] , [14] , [32] collectively known as the VCA region . The C-terminal VCA regions of both WSP-1 and N-WASP have been shown to be sufficient for activating the Arp2/3 complex in vitro [32] , [33] . We utilized the C-terminal VCA region of WSP-1 to selectively activate the Arp2/3 complex in the mechanosensory neurons ( mec-4::wsp-1vca ) . The WSP-1VCA caused PLM premature termination defects that were very similar to MYR-VAB-1 ( Figure 1D ) . The activation of high levels of the Arp2/3 complex produces extensive short branched actin networks that prevent the formation of filopodia , and hence can inhibit axon extension [34] , [35] . Ena/VASP , on the other hand , promotes axon extension through filopodia formation and elongation [16] , [17] , [20] , [29] . Thus , activation of Arp2/3 complex and UNC-34/Ena have opposite roles in the axon growth cone , and perhaps Arp2/3 complex activation can antagonize the function of UNC-34/Ena . Since WSP-1/N-WASP is an activator of the Arp2/3 complex , we wanted to test genetically if wsp-1 can antagonize unc-34 function . Due to the synthetic lethality of wsp-1; unc-34 double mutants [14] , [36] , we analyzed the PLM axons in wsp-1; unc-34 ( RNAi ) animals . Tissue specific unc-34 RNAi resulted in the partial suppression of PLM overextension defects caused by wsp-1 ( gm324 ) ( Figure 1C ) , consistent with WSP-1/Arp2/3 activity antagonizing UNC-34 function . In summary , we show that WSP-1 functions in PLM axon termination , through various signaling pathways , including the VAB-1 Eph RTK . Our results suggest that MYR-VAB-1 is exerting its effect by activating the Arp2/3 complex through WSP-1 . We also suggest that WSP-1 can antagonize UNC-34 function by activating the Arp2/3 complex . We used in vitro binding assays to ask whether VAB-1 , NCK-1 and WSP-1 could form a complex . WSP-1 was able to bind NCK-1 ( Figure 6A , Lane 6 ) , but not VAB-1 ( Figure 6A , Lane 5 ) . However , WSP-1 was able to pull down VAB-1 in the presence of NCK-1 , indicating that a VAB-1/NCK-1/WSP-1 complex can occur ( Figure 6A , Lane 7 ) . Since NCK-1 is able to bind both UNC-34 and WSP-1 , we wanted to determine whether all three molecules can form a complex , or do UNC-34 and WSP-1 compete for NCK-1 binding . We first confirmed that WSP-1 was unable to bind UNC-34 ( Figure 6B , Lane 6 ) . We found that although WSP-1 binds NCK-1 , the presence of UNC-34 resulted in a 70% reduction of the NCK-1/WSP-1 complex ( Figure 6B , Lane 8 ) . This shows that UNC-34 can effectively compete with WSP-1 for NCK-1 binding . Furthermore we could not detect NCK-1/UNC-34/WSP-1 in a complex ( Figure 6B , Lane 8 ) . Interestingly , adding VAB-1 to the binding interaction increased the level of NCK-1 binding to WSP-1 , indicating that VAB-1 eliminated UNC-34's ability to compete for NCK-1 binding ( Figure 6B , Lane 9 ) . In summary , our binding assays show that VAB-1 , NCK-1 and WSP-1 form a complex , that UNC-34 competes with WSP-1 for NCK-1 binding , and that VAB-1 enables WSP-1 to outcompete UNC-34 for binding to NCK-1 . The VAB-1 RTK effectors NCK-1 and WSP-1 are known actin regulators and therefore implicate VAB-1 signaling in regulating actin cytoskeleton for axon guidance . To confirm this , we monitored the PLM growth cone of wild-type and myr-vab-1 transgenic animals at the time of hatching . In wild-type animals , most of the PLM growth cones exhibited dynamic changes and had many filopodia protrusions ( 70%; N = 20 movies ) ( Figure 7A , Video S1 ) . Transgenic myr-vab-1 animals , on the other hand , had growth cones that were less dynamic and were usually void of filopodia like structures with only 25% ( N = 16 movies ) showing some filopodia structures ( Figure 7B , Video S2 ) . Since our molecular and genetic data suggest that VAB-1 inhibits UNC-34/Ena function we also observed the growth cones of unc-34 ( e566 ) animals . We found that unc-34 ( e566 ) mutants , like myr-vab-1 animals , had growth cones void of filopodia structures with only 25% displaying filopodia structures ( N = 12 movies; not shown ) . Our results show that activated VAB-1 can affect the PLM growth cone morphology by inhibiting filopodia formation .
Our results show that the C . elegans NCK-1 adaptor protein is an effector of the VAB-1 RTK signal in vivo . Several lines of evidence indicate that VAB-1 and NCK-1 act together to regulate axon guidance . First , nck-1 and vab-1 animals have similar neuronal defects . Second , NCK-1 and VAB-1 physically interact and co-localize in similar neuronal cells and axons . Finally , the nck-1 loss-of-function suppresses the defects caused by the constitutively active VAB-1 . We found that NCK-1 binds the VAB-1 juxtamembrane tyrosine Y673 ( YEDP ) via its SH2 domain in a VAB-1 kinase dependent manner . This is consistent with the published binding specificity of the Nck SH2 domain , as well as reports of Nck1 binding to the second juxtamembrane tyrosine residue ( YEDP ) in EphA3 ( Y602 ) and EphA2 ( Y594 ) [37] , [38] , [39] . Interestingly , Nck adaptors have been reported to function downstream of Eph RTKs but it appears that the activated EphA RTKs are direct targets of Nck adaptors [38] , [39] , [40] , [41] , whereas Nck may indirectly interact with EphBs [11] , [42] , [43] . Considering that the intracellular region of VAB-1 is more similar to EphA receptors [4] , our results in C . elegans provides relevant insight into how mammalian EphA receptors could regulate the actin cytoskeleton for axon guidance . The Ena/VASP protein family is required in processes that involve dynamic actin remodeling such as platelet shape change , axon guidance and Jurkat T cell polarization [44] . The ability of Ena/VASP proteins to remodel actin stems from their ability to polymerize actin , which is required for filopodia formation and elongation [16] , [45] , [46] . In C . elegans , UNC-34/Ena functions in neuronal cell migration , axon guidance and epithelial filopodia formation [14] , [18] , [19] , [20] . Our results further confirm the role of UNC-34 in axon extension , because we show that the unc-34 ( e566 ) PLM axons terminated prematurely . The cause of early termination is likely due to a reduction of filopodia elongation in the growth cone , resulting in the persistence of more densely branched filaments that can slow axon migration . This is supported by the finding that unc-34 mutants have fewer filopodia structures on growth cones , and a reduced rate of growth cone migration [20] ( this work and our unpublished observations ) . In addition , mammalian studies show that depletion of Ena/VASP generates shorter and more densely branched filaments [47] . We propose that VAB-1 negatively regulates UNC-34 for PLM termination . This is supported by our observations that: 1 ) the loss-of-function unc-34 resulted in PLM axon defects similar to the hyperactive MYR-VAB-1; 2 ) over expressing UNC-34 in the PLM partially suppressed the MYR-VAB-1 phenotype; 3 ) tissue specific unc-34 RNAi suppressed the vab-1 PLM overextension defects; and 4 ) over expressing VAB-1 reduced the UNC-34 protein levels . Although we do not know the mechanism of the reduction of the UNC-34 protein levels displayed in the hyperactive VAB-1 , it is possible that UNC-34 , when removed from its adaptor NCK-1 , is more prone to degradation . In this case NCK-1 may play a dual role and may also promote UNC-34 function as well . It is also likely VAB-1 signaling could affect the unc-34 transcriptional level . Future experiments should resolve how VAB-1 regulates UNC-34 protein levels . Our finding that VAB-1 negatively regulates UNC-34/Ena is different from a previous report that shows mammalian EphB4 as an activator of Ena/VASP [24] . In fibroblast cells , the EphB4 receptor is thought to activate Ena/VASP to destabilize lamellipodia during cell repulsion and likely does so by promoting elongated actin filaments rather than a branched actin filament network . Although the Eph receptor signal transduction to Ena or UNC-34 is opposite ( activates vs . inhibits ) the role for UNC-34/Ena is conserved , because in both cases UNC-34 or Ena/VASP promotes actin filament elongation . Our results provide evidence that VAB-1/Eph RTK can regulate the actin cytoskeleton through its interaction with NCK-1 and WSP-1 . This is based on our observation that vab-1 , nck-1 and wsp-1 mutants share the same phenotype of PLM axon overextension , that both nck-1 and wsp-1 were able to partially suppress the MYR-VAB-1 PLM termination defect , that VAB-1 , NCK-1 and WSP-1 are able to form a complex in vitro , and that the activation of the Arp2/3 complex via the WSP-1 VCA domain resulted in PLM termination defects similar to MYR-VAB-1 . The role of N-WASP as a negative regulator of axon elongation has been shown by two separate reports , where the reduction of N-WASP resulted in the enhancement of axon elongation [48] , [49] . This phenotype is reminiscent of the PLM overextension defects we observed in wsp-1 animals . There have been conflicting reports on the role of the Arp2/3 complex in axon elongation , where some reports suggest that the Arp2/3 complex acts as a negative regulator of axon elongation [34] , [49] , while other reports show that the Arp2/3 complex is required for axon elongation [20] , [50] . A paper by Ideses et al . ( 2008 ) provided a potential resolution to this paradox by looking at the characteristics of actin assembly in the presence of variable amounts of Arp2/3 complex in vitro [35] . It is proposed that high levels of the Arp2/3 complex prevent the formation of filopodia bundles by promoting the extensive branching networks of actin with short tips . On the other hand , at low concentrations of Arp2/3 the actin filaments have longer tips and are further apart making it easier to form filopodia bundles [35] . Therefore , it would be expected that the complete elimination of Arp2/3 would prevent any neurite elongation . Similarly , the excessive activation of Arp2/3 would also prevent neurite elongation due to the increased levels of short , branched networks of actin filaments . In the C . elegans epithelial cells unc-34 and wsp-1 function redundantly for epithelial cell migrations [14] . However our results in PLM neurons suggest that WSP-1 and UNC-34 have opposite roles . Why the apparent paradox ? This is reminiscent of what has been observed for Ena/VASP proteins where some reports suggest Ena/VASP promotes actin dependent processes while others suggest Ena/VASP may inhibit actin dependent processes [51] . While the growth cones on axons and the leading edge of epithelial cells both require actin for movement , they might not be identical in the way the cell moves forward . Proteins such as Ena/VASP , N-WASP , and Arp2/3 are thought to promote actin polymerization , however these proteins also change the geometry of the actin filament network in addition to promoting actin assembly . Therefore the overall effects of such changes in the actin network may not be easy to predict with respect to cell movement since various concentrations of these actin regulators could lead to activation or inhibition of filopodia . Since WSP-1/N-WASP is an activator of the Arp2/3 complex and different levels of Arp2/3 can elicit different behaviors , WSP-1 may also have opposite effects depending on its level of activity . In addition , while most of our results are based on the PLM neurons it is very likely the roles of UNC-34 and WSP-1 and how they are regulated will be different in other neurons . N-WASP has been shown to interact in a complex with the mammalian EphB2 , through the adaptor molecule intersectin [52] . Furthermore , this complex of EphB2 , intersectin and N-WASP is required for dendritic spine formation , which consists mainly of a meshwork of branched filaments caused by the activation of the Arp2/3 complex [52] . C . elegans intersectin ( ITSN-1 ) is expressed in the nervous system , and it is enriched in presynaptic regions and has roles in neurotransmission [53] . Future work will determine whether the VAB-1/Eph interacts with ITSN-1 to connect WSP-1 . Our current work shows that the VAB-1 Eph RTK can signal through WSP-1/N-WASP through a different adaptor molecule , NCK-1 , and we propose , like the mammalian intersectin adaptor , this complex activates Arp2/3 to promote branched actin . We propose a model of how the proteins VAB-1 , NCK-1 , UNC-34 , WSP-1 and Arp2/3 function in axon growth cones for extension and termination ( Figure 7C ) . During PLM axon outgrowth , the growth cone is stimulated by an attractive cue that results in the accumulation of UNC-34/Ena at the growth cone . The result is a net forward movement due to the role of UNC-34/Ena in inhibiting actin capping proteins , and allowing filopodia elongation by polymerizing F-actin at the leading edge . In addition , UNC-34/Ena binds to the NCK-1 SH3 domains to prevent it from interacting with WSP-1 and participating in a signaling pathway ( s ) that would otherwise inhibit axon extension . It is also possible that the UNC-34/NCK-1 heterodimer could function together for actin polymerization or that NCK-1 binding could stabilize the UNC-34 protein . In this case NCK-1 acts positively with UNC-34 . However , since unc-34 and nck-1 mutants have opposite PLM axon phenotypes , it suggests that nck-1's role in axon outgrowth is more dispensable or redundant than its role in axon termination . Once the VAB-1/Eph RTK receives the signal to inhibit axon extension , VAB-1 is autophosphorylated and provides a docking site ( Y673 ) for NCK-1 . The NCK-1-SH2 domain binds the activated VAB-1 receptor and this disrupts the interaction between NCK-1 and UNC-34 to release the inhibitory effect of UNC-34 on NCK-1 . Through an unknown mechanism , we also show that VAB-1 negatively regulates the UNC-34/Ena protein levels . VAB-1/NCK-1 can now recruit and activate WSP-1 and all three proteins form a complex , which results in high levels of Arp2/3 activation , ultimately leading to a branched meshwork of actin filaments . The combined actions of VAB-1/Eph blocking UNC-34/Ena activity , while activating Arp2/3 through NCK-1/WSP-1 contributes to the molecular events required to stop the growth cone forward movement .
All C . elegans strains were manipulated as described by Brenner [54] . All alleles were isolated in the standard wild type Bristol strain N2 . All experiments were performed at 20°C unless otherwise indicated . The following strains were used in this study: N2 ( var . Bristol ) [54]; LGI: zdIs5[mec-4::gfp]; LGII: vab-1 ( dx31 ) , quIs5[mec-4::myr-vab-1]; LG IV: wsp-1 ( gm324 ) , LG V: unc-34 ( e566 ) ; LGX: , quIs6[unc-34::unc-34::gfp]; Unmapped: quIs16[hs::myr-vab-1] [55]; Extrachromosomal arrays ( this study ) : quEx131[mec-4::nck-1A] , quEx190[nck-1::nck-1A-gfp] [12] , quEx215[mec-4::unc-34::gfp] , quEx281[mec-4::unc-34] , quEx283 [mec-4::nck-1A::gfp] [12] , quEx321[mec-4::vca] , quEx338[mec-4::unc-34 RNAi] ( see tissue specific RNAi ) . Unless noted otherwise , all C . elegans strains were obtained from the C . elegans Genetics Stock Center , ( U . of Minnesota ) . To produce double stranded RNA ( dsRNA ) only in the mechanosensory neurons , we constructed a cloning vector ( pIC659 ) with head to head Pmec-4 promoters on each side of a Multiple Cloning Site ( MCS ) such that the sense and antisense strands of an inserted cDNA would be transcribed . The mec-4::unc-34 RNAi construct ( pIC727 ) was created by cloning an unc-34 cDNA fragment ( ATG start to the first SalI site , 388 bp ) into the pIC659 dual Pmec-4 RNAi cloning vector . The mec-4::nck-1A construct ( pIC313 ) was previously described in Mohamed and Chin-Sang ( 2011 ) . The mec-4::unc-34 construct ( pIC624 ) was generated by amplifying unc-34 cDNA and sub-cloning behind the mec-4 promoter . The same procedure was used to make the mec-4::unc-34::gfp ( pIC540 ) construct , but unc-34 was fused to gfp amplified from pPD95 . 75 . To create the mec-4::vca construct ( pIC673 ) , the VCA region of WSP-1 ( 9108–9606 of the wsp-1 gene; C07G1 . 4a in Wormbase ) was amplified from genomic DNA and cloned behind the mec-4 promoter . The unc-34::unc-34::gfp translation reporter was generated by a PCR fusion approach [56] using the following pieces: 1 . A ∼5 kb genomic region that includes 2 kb of 5′UTR and the first two exons of unc-34 , 2 . Exons 2–7 were amplified from RB2 cDNA library , and 3 . a 868 bp GFP fragment amplified from pPD95 . 75 ( gift from Dr . Andrew Fire ) . The expression of the UNC-34::GFP rescued the unc-34 ( e566 ) uncoordinated phenotype . Details of plasmid/PCR constructs and primer sequences are available upon request . Transgenic animals were generated by germ-line transformation as previously described [57] . The unc-34::unc-34::gfp translational reporter was injected at a concentration of 20 ng/µL , and one of the unc-34 rescuing lines ( quEx61 ) was integrated to create quIs6 . The mec-4::unc-34 construct was injected at a concentration of 30 ng/µL into mec-4::gfp ( zdIs5 ) ; mec-4::myr-vab-1 ( quIs5 ) . The mec-4::unc-34::gfp construct was injected at a concentration of 30 ng/µL into N2 . mec-4::unc-34RNAi , mec-4::vca and mec-4::nck-1 were injected into mec-4::gfp ( zdIs5 ) at 30 ng/µL . mec-4::nck-1 ( quEx131 ) was later crossed into unc-34 ( e566 ) , and mec-4::unc-34RNAi ( quEx338 ) was crossed into vab-1 ( dx31 ) and wsp-1 ( gm324 ) . Transgenic animals were identified by the co-injection marker pRF4/rol-6 ( 30 ng/µl ) , or odr-1::rfp ( 30 ng/µl ) [57] . At least two independent lines were isolated and analyzed . The data shown are from one representative line . Mixed stage animals were fixed and stained as described in Chin-Sang et al . ( 1999 ) [58] . Rabbit anti-VAB-1 antibodies ( antigen VAB-1-HIS6 ) and chicken or mouse polyclonal antibodies against GFP ( Chemicon ) were used at 1∶100 dilutions . Texas Red-conjugated goat anti-rabbit and FITC conjugated goat anti-chicken or anti-mouse secondary anti bodies ( Jackson's lab ) were used at a 1∶100 dilution . For Western blot analysis , antibodies were used at the following dilutions: anti-NCK-1 at 1∶500 , anti-VAB-1 at 1∶2500 , anti-MBP-HRP at 1∶8000 , anti-GST-HRP at 1∶4000 and 4G10 ( Upstate Inc . ) at 1∶2500 . Goat-anti-rabbit-HRP and goat-anti-mouse-HRP were used as at 1∶10000 dilutions on western blots . Relative band intensities in Figure 6B were quantified using at least two independent blots and analyzed using the National Institutes of Health Image J program . The mechanosensory neurons were visualized using the mec-4::gfp ( zdIs5 ) reporter . Young adult animals were scored as having PLM axon overextension or premature termination as described previously [6] . Outgrowth of the PLM axon happens during embryogenesis and continues to grow after hatching and most of its growth happens at the L1 stage . From L2 onwards to adulthood PLM growth is maintained relative to its termination point along the body [28] . To measure the L1 PLM axons , newly hatched L1s were synchronized in the absence of food for up to 12 hours . We found that although the worms were born in the absence of food that the PLM was still able to grow and the PLM axon lengths were equivalent to the length of animals developing for 2–3 hours post hatching . This corresponds to the Phase 1 or fast growth PLM growth phase [28] . Our wild-type reference strain ( zdIs5 ) had L1 PLMs with an average PLM length of 108 . 5 ( ±5 . 5 ) microns with a PLM length/total body length ( from head to tail ) ratio of 0 . 48 ( ±0 . 04 ) . L1 PLM axons were scored as overgrown if they were longer than 114 µm and had a PLM/total body length ratio of greater than 0 . 52 . L1 PLM axons were scored as under grown if the PLMs were shorter than 103 µm and had a PLM length/total body length ratio of length less than 0 . 44 . The L1 PLM axons were traced from photograph and measured in NIH Image J software . The wild-type neuron morphology was defined by analysis of neuronal GFP reporters and is consistent with the electron microscopic reconstruction of the C . elegans nervous system [59] . Animals were anesthetized using 0 . 2% tricaine and 0 . 02% tetramisole in M9 , and mounted on 3% agarose pads . Unless stated otherwise , fluorescent animals and images were analyzed using a Zeiss Axioplan microscope , Axiocam and Axiovision software . PLM growth cones were visualized using a mec-4::gfp ( zdIs5 ) reporter . Eggs were allowed to hatch for 5 minutes , and the newly hatched L1 animals were examined immediately on 3% agarose pads with a drop of 0 . 2% tricaine and 0 . 02% tetramisole in M9 . PLM growth cones were imaged with a Zeiss LSM710 confocal microscope at intervals of 20–30 s . Axons were scored positive for filopodia if time-lapse movies revealed at least 2 protrusions , and there were dynamic movements ( eg . growth and collapse ) of the these structures within the 10–15 minutes of filming . See Videos S1 and S2 for examples . Yeast cells were grown on standard and selective media as required [60] . The desired plasmids were transformed into yeast cells using the lithium acetate method [61] . For binding and deletion analysis , the pGBKT7 vector was used as bait and the pGADT7 vector ( Clontech ) as prey , and β-galactosidase activity was measured qualitatively by X-GAL overlay assays [62] . To identify interactions with VAB-1 , the Kinase Region ( 669 aa-985 aa ) of vab-1 was cloned into pGBKT7 ( pIC187 ) and used in a screen against the RB2 cDNA library ( gift from Dr . R . Barstead ) , and about 600 , 000 colonies were screened and two independent nck-1 cDNA clones were isolated . Site directed mutagenesis ( QuickChange , Stratagene ) of pIC187 was used to change the juxtamembrane tyrosine 673 changed to glutamic acid ( Y673E ) . The SH2 domains of NCK-1 , MIG-10 , SEM-5 , ABL-1 and VAV-1 were cloned into the activation domain of the pGADT7 vector . Primer sequences and details of plasmid constructs are available upon request . The following constructs were created by cloning the desired cDNA fragment into Glutathione-S-Transferase ( pGEX4T-2 , Amersham ) : pIC282 – NCK-1 SH2 domain ( 298 aa–397 aa ) , pIC297 – all three NCK-1 SH3 domains ( 1 aa–308 aa ) , pIC308 – 1st NCK-1 SH3 domain ( 1 aa–72 aa ) , pIC593 – 2nd NCK-1 SH3 domain ( 112 aa–186 aa ) , pIC309 – 3rd NCK-1 SH3 domain ( 198 aa–308 aa ) , pIC324 – full length ( F . L . ) NCK-1 ( 1 aa–397 aa ) , and pIC606 – F . L . UNC-34 ( 1 aa–454 ) . The following constructs were created by cloning the desired cDNA fragment into Maltose Binding Protein ( pMALtm-p2X , New England Biolabs ) : pIC225 – F . L . intracellular region of wild-type VAB-1 ( 581 aa–1117 aa ) , pIC119 – F . L . intracellular kinase deficient VAB-1 ( G912E ) , pIC603 – UNC-34 RPO-EVH2 domain ( 128 aa–454 aa ) , pIC605 – F . L . UNC-34 ( 1 aa–454 aa ) , pIC671 – UNC-34 PRO domain ( 128 aa–274 ) , pIC674 – UNC-34 EVH2 domain ( 246 aa–454 aa ) , pIC670 – WSP-1 VCA domain ( 334 aa–607 aa ) . pIC582 – His-6::VAB-1 ( 581 aa–1117 aa ) was described in Brisbin et al ( 2009 ) . All fusion constructs were expressed in E . coli Tuner ( DE3 ) . For Figure 4B , 4C , 4F , 4G and Figure 5A , a GST ‘pull-down’ assay was used to confirm the VAB-1 , NCK-1 and UNC-34 interactions . Soluble/purified ( Load ) MBP-VAB-1 , MBP-VAB-1 ( G912E ) , MBP-UNC-34 F . L . , MBP-UNC-34-PRO-EVH2 , MBP-UNC-34-PRO or MBP-UNC-34-EVH2 were incubated for 2–3 hrs at 4°C with soluble extracts containing either GST , GST-NCK-1 F . L . , GST-NCK-1-all SH3 domains , GST-NCK-1 ( 1stSH3 ) , GST-NCK-1 ( 2ndSH3 ) , GST-NCK-1 ( 3rdSH3 ) , GST-NCK-1 ( SH2 ) , His-6::VAB-1 ( 581 aa–1117 aa ) ( pIC582 ) or GST-NCK-1 F . L . coexpressed with pIC582 bound to 50 µl glutathione sepharose beads ( GE healthcare ) . Unbound fractions were collected , protein bound to GST beads were washed four times ( 25 mM Hepes , 10% Glycerol , 0 . 1% Triton-X , 285 mM NaCl ) , and a proportional loading of each sample was analyzed by standard SDS polyacrlyamide gel , followed by western blotting . All loads fused to MBP were detected using anti-MBP conjugated to HRP ( New England Biolabs ) . His6-VAB-1 was detected using Rabbit anti-VAB-1 antibodies ( antigen VAB-1-His6 ) ( Figure 5A ) . GST and GST-NCK-1 F . L . , and GST-NCK-1 deletion domains were detected either by Ponceau S or anti-GST conjugated to HRP . For Figure 6 , MBP ‘pull-down’ was used to confirm VAB-1 , NCK-1 , WSP-1 and UNC-34 interactions . Soluble extracts ( Load ) of GST-NCK-1 F . L . , His-VAB-1 ( pIC582 ) , GST-NCK-1 F . L . coexpressed with pIC582 , or GST-UNC-34 F . L . were incubated for 2–3 hours at 4°C with soluble extracts containing either MBP or MBP-WSP-1 ( 334 aa–608 aa ) bound to 100 µl amylose resin beads ( New England Biolabs ) . Unbound fractions were collected , protein bound to amylose beads were washed four times ( 20 mM Tris-Cl [pH 7 . 5] , 200 mM NaCl , 1 mM EDTA , 1 mM DTT ) , and a proportional loading of each sample was analyzed by standard SDS polyacrylamide gel , followed by western blotting . VAB-1 was detected by Rabbit anti-VAB-1 , GST fused proteins were detected by anti-GST conjugated to HRP , MBP and MBP-WSP were detected by anti-MBP conjugated to HRP .
|
The correct wiring of the nervous system depends on the ability of axons to properly interpret extracellular cues that guide them to their targets . The Eph receptor tyrosine kinases ( RTKs ) have roles in guiding axons , but their signaling pathways are not completely understood . In this study , we used the nematode Caenorhabditis elegans to study how the VAB-1 Eph RTK regulates the growth cone structure for axon guidance . Genetic and molecular data show that VAB-1 regulates the conserved molecules NCK-1 , WSP-1/N-WASP , and UNC-34/Ena . Our study provides a model of how the VAB-1 Eph RTK modulates the growth cone structure to inhibit axonal outgrowth . We show that activated VAB-1 can inhibit an NCK-1/UNC-34 interaction by binding to the NCK-1 SH2 domain . We also show that NCK-1 and WSP-1 can physically interact and that VAB-1/NCK-1 and WSP-1 form a complex in vitro . We suggest that the VAB-1 Eph RTK can contribute to the termination of axon outgrowth by two methods: 1 ) The VAB-1/NCK-1/WSP-1 complex activates ARP-2/3 to change the actin growth cone dynamics to that of a branched structure thus reducing the number of filopodia , and 2 ) VAB-1 inhibits axon extension by inhibiting UNC-34/Ena's function in actin polymerization .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"developmental",
"biology",
"molecular",
"cell",
"biology",
"molecular",
"development",
"genetics",
"molecular",
"genetics",
"biology",
"morphogenesis",
"neuroscience",
"genetics",
"and",
"genomics"
] |
2012
|
The Caenorhabditis elegans Eph Receptor Activates NCK and N-WASP, and Inhibits Ena/VASP to Regulate Growth Cone Dynamics during Axon Guidance
|
Organisms have increased in complexity through a series of major evolutionary transitions , in which formerly autonomous entities become parts of a novel higher-level entity . One intriguing feature of the higher-level entity after some major transitions is a division of reproductive labor among its lower-level units in which reproduction is the sole responsibility of a subset of units . Although it can have clear benefits once established , it is unknown how such reproductive division of labor originates . We consider a recent evolution experiment on the yeast Saccharomyces cerevisiae as a unique platform to address the issue of reproductive differentiation during an evolutionary transition in individuality . In the experiment , independent yeast lineages evolved a multicellular “snowflake-like” cluster formed in response to gravity selection . Shortly after the evolution of clusters , the yeast evolved higher rates of cell death . While cell death enables clusters to split apart and form new groups , it also reduces their performance in the face of gravity selection . To understand the selective value of increased cell death , we create a mathematical model of the cellular arrangement within snowflake yeast clusters . The model reveals that the mechanism of cell death and the geometry of the snowflake interact in complex , evolutionarily important ways . We find that the organization of snowflake yeast imposes powerful limitations on the available space for new cell growth . By dying more frequently , cells in clusters avoid encountering space limitations , and , paradoxically , reach higher numbers . In addition , selection for particular group sizes can explain the increased rate of apoptosis both in terms of total cell number and total numbers of collectives . Thus , by considering the geometry of a primitive multicellular organism we can gain insight into the initial emergence of reproductive division of labor during an evolutionary transition in individuality .
Organisms have increased in complexity through a series of major evolutionary transitions , in which formerly autonomous entities become parts of a novel higher-level entity [1]–[5] . Examples of this transition include the evolution of multicellular organisms from unicellular ancestors and eusocial “superorganisms” from multicellular ancestors . One of the primary benefits ascribed to major evolutionary transitions is the potential for the higher-level entity to evolve division of labor among its lower-level units [1] , [2] , a subject which has received a good deal of theoretical attention [6]–[11] . A salient form of this is reproductive division of labor in which some lower-level units forgo their contribution to reproduction of the higher-level entity . This is found in the germ-soma differentiation in multicellular organisms and worker/queen roles in eusocial insects [12]–[17] . While reproductive specialization is not strictly required for division of labor to provide a fitness benefit to the higher-level entity , it has evolved repeatedly in independent lineages [18]–[22] . Upon a superficial glance , the existence of such reproductive self-sacrifice seems to present an evolutionary paradox . How would such a self-destructive tendency be favored by a process ( natural selection ) that places a high premium on survival and reproduction ? The resolution of the paradox generally involves a situation in which the self-sacrifice improves the fitness of the higher-level unit [6] , [7] , [9] , [11]–[13] , [17] , [23]–[27] . For ease of discussion , let us call the lower-level entities “particles” and the higher-level entities “collectives” . Suppose the altruistic action of some particles allows other particles in their collective to found new collectives at a higher rate . If these founding particles possess a tendency for self-sacrifice ( which can occur if particles have high relatedness within collectives ) , then reproductive division of labor within collectives can evolve . We emphasize that such altruism must occur in a strict subset of particles within the collective and requires the plastic or stochastic expression of phenotypic traits at the particle level . Thus , while the logic for the evolution of reproductive self-sacrifice is sound , the mechanistic underpinnings could be complex . The precise way in which such differentiation evolves and its presence in the early stages of major transitions are largely unknown . A recent evolution experiment on the yeast Saccharomyces cerevisiae has provided a unique platform to address the issue of reproductive differentiation during an evolutionary transition in individuality [28] . In this experiment , populations of unicellular yeast were periodically exposed to a selective regime that rewarded cells that sank quickly in test tubes . During this setting , cells in clusters sink more quickly than independent cells , incentivizing group formation . Cluster-forming phenotypes evolved repeatedly via the retention of cell-cell connections after mitotic reproduction . These group-forming types outcompeted their unicellular ancestors , driving them to extinction in all 10 replicate populations within 60 days [28] . Clusters grew in size until the resulting physical strain caused them to fragment , yielding a form of group reproduction . As a result , the yeast evolved group formation and reproduction de novo . Interestingly , these yeast clusters soon evolved a secondary trait: a higher rate of cellular programmed death ( hereafter referred to as apoptosis ) . Why would a higher rate of cellular suicide , an ostensibly costly trait , be favored by natural selection ? A higher rate of apoptosis might have evolved because it increases collective-level reproduction . Since each cell in the group is connected solely to its parent and offspring cells [28] , it only takes a single break in any connection to produce two distinct collectives . Both physical strain and cell death can create such breaks and , consequently , increase the number of groups . Selection for a greater number of clusters could promote division of labor [6]–[11] , providing an explanation for the evolution of higher rates of apoptosis . Yet , the problem is that the selective regime seemingly rewards large clusters ( group size ) , not large numbers of clusters ( group fecundity ) . Moreover , the apoptotic mechanism of group reproduction acts in direct opposition to group viability . While there may be a benefit for groups to reproduce ( to reduce the risk of not being transferred due to random sampling error ) , as groups divide they become smaller and sink less quickly , making them less competitive against larger groups . It would appear that an optimal strategy would be for groups to grow as large as possible and divide infrequently . In contrast , when selection for large groups is stronger ( requiring faster settling ) , groups evolve higher rates of apoptosis and produce proportionally smaller propagules [28] . To address this conundrum , we build a series of mathematical and computational models . The first model explores the optimal way for clusters to split under a selective regime similar to the Ratcliff et al . experiment [28] . We see that it is possible for high rates of cluster division to be adaptive , but the opposite is also a possibility . This first model ignores various details about the yeast system for tractability , including the geometry of the clusters and cellular apoptosis . The second model explicitly considers the cellular arrangement within yeast clusters and the consequences of apoptosis on cluster reproduction . This model reveals how the geometric structure of the cluster interacts with apoptosis to affect the number and size of cluster offspring . We find that the organization of snowflake yeast imposes powerful limitations on the available space for new cell growth . By dying more frequently , cells in clusters circumvent space limitations , and , paradoxically , reach higher numbers . Finally , we demonstrate that selection for particular cluster sizes can explain the increased rate of apoptosis both in terms of total cell number and total numbers of collectives . Thus , considering the specific geometry of the clusters reveals the adaptive benefit of the evolution of reproductive self-sacrifice in the Ratcliff et al . experiment [28] , and a possible mechanism for the emergence of reproductive division of labor during an evolutionary transition in individuality .
In this section , we build an abstract model to get a rough understanding of how the experimental regime might select for different rates and forms of cluster division . When a cluster splits , it yields both new and smaller clusters . Thus , division simultaneously affects cluster reproduction and the prospects for viability under settling selection . How should a cluster balance fecundity against survival ? Similar to the experimental regime of Ratcliff et al . [28] , we use a framework with a growth phase followed by a selection event , enabling cluster division strategies to depend on time within the growth phase . Since cluster growth and division change the size of clusters , we permit splitting strategies to be size-dependent . Here we use a dynamic programming approach [29] , [30] to explore optimal cluster division strategies . We denote the probability a cluster of cells survives settling selection as . Since larger clusters settle faster than smaller ones , we assume is a non-decreasing function . In addition , we assume that division and growth of clusters occur for time steps prior to settling selection . We define to be the maximal reproductive output for a cluster of size ( ) at time ( ) . As a consequence , if fitness is measured in terms of number of clusters , and if fitness is measured in terms of the number of cells . Over each time step ( from to , where ) , we assume that clusters divide and then grow . Specifically , a cluster of cells at time point splits into two clusters of sizes and ( where ) . We note that this includes the case where the cluster does not divide ( i . e . , ) ; therefore , this framework allows us to track optimal cluster division rate as well as optimal size for cluster propagules . After division , the new clusters grow according to the function ; that is , a cluster that starts with cells ends the time step with cells . For instance , if every cell in a cluster doubles over a time step , then . We have the following backwards recursion for maximal reproductive output: ( 1 ) Suppose fitness is measured in terms of the number of clusters that survive selection ( i . e . , ) . In the Supplement , we prove that if , where is some positive integer greater than 1 and is concave ( e . g . , ) , then the optimal strategy is always to divide into halves ( or as close to halves as possible ) . If cell death is the means of cluster division , these conditions would predict the evolution of cell death mechanisms that produce equal sized cluster offspring . Generally in this case , higher splitting rates could be favored and cell death may be one way to accomplish this . However , there are several important caveats regarding this result . If is convex over some range of as might be found in a Hill function , then it can be optimal not to divide at all ( at least for some sizes; see Supplement ) . If fitness is measured in terms of the number of cells ( i . e . , ) rather than the number of clusters , it can be optimal not to divide even when is strictly concave ( see Supplement ) . In such cases , cell death rate would be predicted to decrease . Furthermore , the model lacks a mechanistic basis for cluster division . Such a basis follows from recognizing the geometry of the cluster . Yeast clusters form when mother cells remain attached to their budding daughter cells . Because a given mother cell can have multiple attached daughters , the cluster is a branched acyclic network ( i . e . , a multi-branched tree ) . Suppose cell death severs a single cell-cell connection . In such a case , a yeast cluster will produce two daughter clusters . However , the sizes of the daughter clusters are constrained by the network topology of the mother cluster . In terms of the above model , some values will not be possible and breaking a random link will make some values much more likely than others . Consequently , the model's implicit assumption that any value of is equally available to a dividing cluster is misplaced . Moreover , if the rate of cell death is constant , then larger clusters should expect more broken links ( and therefore more offspring clusters per unit of time ) , which is not currently captured by the above model . To address these issues directly , we explicitly incorporate cluster geometry into our second model . We describe the structure of a cluster by a tree graph in which nodes represent cells and edges represent physical attachments between cells ( Figure 1 ) . When a cell reproduces , its corresponding node in the tree gains an edge to a newly created node . This growth mechanism ensures that cells are only attached to their parent and offspring . For simplicity we begin the tree with only one node which represents the first mutant yeast cell to have the capacity to form clusters , call it Node 0 . Each time Node 0 reproduces it generates a branch which will continue to grow independently . Initially , we assume that all cells reproduce and do so at the same time . So with each successive generation the tree doubles its nodes , i . e . the cluster doubles in the number of cells . After generations the branches from Node 0 will be composed of cells depending on when the branch was initiated . The total number of cells in the tree is ( we note this is equivalent to in the model from the previous section ) . Due to the tree geometry , if a link/edge between two cells/nodes is severed then it will result in two distinct clusters , i . e . the cluster reproduces . Since both physical strain and cell death lead to cluster reproduction , we can view these as mechanisms for severing an edge between two nodes . It is also possible to divide clusters by removing a node rather than severing an edge . Removing a node with more than two connections , however , could result in “multiple births” , which is not typically observed experimentally . Thus , we assume clusters reproduce via link severance . Furthermore , in order to explain the experimental observations of Ratcliff et al . [28] , we consider cell death as the primary mechanism of link severance–although other mechanisms may also exist . Whatever the mechanism , the location of the severed edge plays a significant role in determining the sizes of the resulting cluster offspring . If an edge in the periphery is severed then one of the resulting clusters will be composed of only a single cell . In contrast , severing more central edges will result in more symmetry between offspring clusters . The particular manner by which cells die determines whether a severed edge is more likely to be in the periphery or the center . If cell death is completely random such that the centermost cells are just as likely to die as newly created cells and the tree is doubling in size every generation , then the severed edge is more likely to be peripheral . This is because at any time of the tree is newly created . As a result , there is a chance that death of a random cell will yield a “group” that is one ( dead ) cell in size by breaking its one and only link to the tree . The expected sizes of the offspring clusters after rounds of cell reproduction are and , and the ratio of the smaller offspring to the parent is less than after 10 generations ( the ratio , , after generations is ) . Such a small cluster may not be able to grow large enough to survive the selective regime and could be excluded from future growth and reproduction . If cell death is not completely random but rather related to age then central edges would be more likely to be severed . In the case that the oldest cell ( Node 0 ) dies , the sizes of the resulting offspring clusters will depend on which link is severed . Each link of Node 0 corresponds to one of its branches with cells . Without a bias as to which link is severed the smaller offspring would be expected to have cells . The ratio of this offspring to the parent after rounds of cell reproduction , , is . After generations is approximately which is 20 times larger than when cell death is completely random . Thus , weighting death towards older , more central cells increases the size of the smaller offspring . Experimental observations of early cluster offspring in the yeast system suggest that the smaller offspring may be closer to of the size of the parent [28] . To see how link severance via cell death can achieve such values , we consider again the death of Node 0 which yielded less offspring asymmetry than random cell death . The oldest branch of Node 0 , created in the first round of cell reproduction , is the only branch greater than 40% of the tree size– it is half of the size of the whole tree . The next oldest branch , created the second time Node 0 reproduces , is a fourth of the whole tree size . Each successive branch is half the size of the previous . If there is no bias in which branch becomes the offspring then the odds favor the branches that are much less than 40% . Instead of unbiased link severing , it could be that links are severed according to the size of the branch they are supporting . Bigger branches may produce more strain on their links compared to smaller branches and , therefore , may break more easily . Although there are many potential ways to bias severance in favor of bigger branches , we assume a simple biasing such that the probability a link is severed is directly proportional to the size of the branch . In this case , the ratio of the smaller offspring to the parent after generations , , is which approaches as increases . This matches experimental observations more closely and suggests that cluster division via cell death may be biased both in which cells die and which links are severed . Until now , we have operated under the unrealistic assumption that all cells in a cluster have the same constant rate of reproduction . Although each time a cell reproduces , the cluster increases in size and span; it also fills the limited volume at the center . As this space gets crowded , cells lose both access to nutrients and room for further reproduction . To determine how the tree geometry experiences volume constraints , we use a 3-dimensional model of growth in which cells occupy concentric shells surrounding the central node , Node 0 ( Figure 2 ) . By stretching the cluster along its longest diameter , this model maximizes the available space and sets an upper bound to the size capacity of the cluster . We assume that each cell is an identical sphere with radius . Cells occupy shells depending on how many links separate them from Node 0 . For example , the third shell is filled with cells that are 3 links from the center . The offspring of a cell occupies the next shell and , conversely , its parent is in the previous shell . Each shell encloses a volume equivalent to a sphere with radius . This volume ( ) can hold at most cells– ignoring issues concerning the maximum packing of spheres . In the growing cluster , the actual number of cells within the volume of a shell is simply the total number of cells in each interior shell . For a given shell after rounds of cell reproduction the total number of cells is ( see Figure 2 ) . Thus , the volume enclosed by shell is exceeded when the number of generations satisfies: ( 2 ) We calculate the lowest for which the volume bounded by each shell is exceeded and find shells 4–6 are the first to overflow at the twelfth round of cell reproduction ( ) . Even if a cell in shell 4 could relocate to shell 3 there is no room available because the volume defined by shell 4 has been exceeded . While there is still space in the volume contained by shells 7–12 , cells from the overcrowded volume cannot move here because they must remain connected to their parents in more interior shells . In addition to the volume constraint , there may be constraints regarding how many attachments ( edges ) a single cell can have . Experimental observations of cluster structure find that most cells are attached to only a few cells ( ) . If there is a limit to the number of attachments per node then this will alter the organization of a cluster ( Figure 3 ) . For example , a tree with maximum node degree of 3 will have just 3 branches emanating from Node 0 . Instead of doubling with each round of cell reproduction , the number of nodes in a branch follows a recursion: , where is the number of nodes episodes after the creation of the branch . Geometries with higher maximum node degrees ( hereafter called “degree capped” ) also feature recursive relationships such that in general , for a tree with degree cap of , with the first values following . This stems from an important distinction in trees with a degree cap: their size only increases with those cells created within the last generations . These recursive relationships relate cluster sizes with Fibonacci numbers such that trees without degree caps are simply Fibonacci sequences of infinite order . In all cases , the total number of nodes in the tree is simply twice the number in the largest branch . As the distribution of cells in branches is altered by limiting the number of node attachments so , too , is the expected size of offspring clusters . If Node 0 dies and there is no bias to which link is severed then the expected offspring size as a proportion of the parent is , where is the degree limit and is the number in the largest branch . This quickly approaches which is much greater than the value found in trees without limits to the number of node attachments . Consequently , cluster offspring are more equal in size . In trees without degree caps , biasing which link is severed according to branch size increased the symmetry of cluster offspring such that the expected size of the smaller offspring is of the parent's size . By comparison , biasing link severance in favor of bigger branches ( as done before ) has less of an effect in trees with degree caps . The expected size of the smaller offspring is of the parent's size for a degree cap of 3 and for a degree cap of 4 . While the most symmetric cluster reproduction is in trees with a degree cap of 3 and biased link severance , all trees with biased link severance produce an offspring that is between of the parent's size . It should be noted that a degree cap of 2 can do better but it can only form filaments rather than snowflake-shaped clusters . Not only do trees with degree caps produce more symmetric offspring but they can experience less severe volume constraints . Since limiting the number of attachments per cell reduces the size of a cluster , it effectively delays when clusters begin to run out of space . A tree without degree caps can only go through 11 rounds of cell reproduction before exceeding the available volume contained by a shell . A tree with degree cap of 4 , however , goes through 14 generations before encountering a limit at shells 8–11 . A tree with a degree cap of 3 can undergo even more generations , exceeding shells 14 and 15 on the 20th round of cell division . It reaches cells before encountering volume limitations which is twice that of trees with a degree cap of 4 ( cells ) and 5–10 times as many cells as trees without degree caps . Interestingly , trees with degree caps of 2 produce filaments , a common biological shape , that are free of any volume constraints . Thus far , we have examined the consequences on cluster reproduction of link severance due to death of a single cell . In practice , however , as clusters grow and reproduce , mechanisms of cell death interact with geometric constraints to create a population of clusters with a distribution of sizes . In addition , the death of a cell has downstream consequences by preventing future growth of a branch . To determine how such mechanisms interact , we simulate the population expansion from the first mutant capable of forming clusters , Node 0 ( see Methods ) . The population simulations show that the total number of living cells increases with the probability of cell death ( Figure 4A & B ) . This paradoxical result is a consequence of the constraints on cell reproduction due to degree caps and limited volume . For a degree cap of 3 , cells that reach the maximum degree ( 3 in this case ) stop reproducing . After 21 generations , many cells have reached the maximum degree and no longer contribute to the growth of the population . By dying , a link connecting two non-reproducing cells is broken . This allows one cell to reproduce again and start a new branch that increases the population by more cells than the cost of the dead cell . Since a cluster with degree cap of 3 does not encounter volume limitations until the 20th generation , near the end of the simulation , the volume constraint does not play a significant role in the increased cell population . In fact , it can be removed and the total number of cells still increases with higher rates of cell death . This is not true with clusters that have a degree cap of 4 ( Figure 4C ) . The higher degree cap reduces the extent to which fixing a maximum number of attachments constrains the population while at the same time increases the strength of the volume constraints– cells experience volume limitations by the 14th round of cell reproduction . So , both degree cap and volume constraints allow clusters to increase the number of living cells by increasing the frequency of cell death . In biological systems cell death may not be completely random but rather biased by age . Analytically , we showed that the age of the dead cell affects the expected sizes of cluster offspring . Here , we include a bias in the age at which cells die in the simulation by protecting reproducing cells from death; cells cannot die until a set amount of time has passed since their last reproduction . We expect this to act similarly to decreasing the death rate because fewer cells are susceptible to death . As such , we predict that the longer death is delayed the lower the final population . Instead , we find that delaying death has a variety of effects depending on the degree cap and the frequency of cell death ( Figure 4D ) . The results match our expectations when the probability of death is low ( ) or clusters are not degree capped . In contrast , when the probability of death is high ( ) , the number of cells in degree capped clusters increases if death is delayed . This effect is strongest when death is delayed only one round of cell reproduction , i . e . cells are susceptible when it has been at least one generation since their last reproductive event . As the delay gets longer the total number of cells decreases . For a degree cap of 4 , delaying death for 5 rounds of cell reproduction still produces more cells than when there is no delay , but this is not true for a degree cap of 3 . Thus , delaying death has different effects depending on the probability of death , the length of the delay , and the maximum node degree . Due to volume constraints and degree caps , apoptosis can increase both the number of cells and the number of clusters . Yet , the experimental regime rewarded cells that were in clusters above a certain size– this success might be measured as either the number of clusters or the number of cells in clusters . The frequency of cell death affects both the number and size distribution of clusters . To find which apoptosis rate yields the most clusters over different size thresholds , we compute the average number of clusters above threshold for different probabilities of death ( Figure 5A for degree cap of 3 and 5B for degree cap of 4 ) . For small cluster thresholds ( 25 cells ) , the highest probability of death produces the most cluster offspring . As the cluster threshold increases to cells , the probability of death that leaves the most cluster offspring decreases to . Larger size thresholds ( ) effectively reward clusters that never divide , and so the best strategy is to have the lowest probability of death ( here , ) . These trends also hold if the degree cap is 4 , but the higher probabilities of death ( and ) dominate for greater ranges of size thresholds . Moreover , these trends are the same if fitness is determined not by the number of clusters above threshold but rather the number of cells in those clusters . Once again , the higher probabilities of death are successful for size thresholds from 1 to . One notable difference is that for size thresholds between and , the highest probability of death , , produces the most clusters but not the most cells– the probability of death produces more cells in clusters above threshold . In determining which apoptosis rate produces the most clusters , we assumed that the probability of death is an evolvable trait . The same may be true of other features related to cluster organization or cell death such as degree cap and the age bias of cell death ( the death delay ) . To find which combination of these traits , “strategies” , yields the most clusters above threshold , we compare combinations of degree cap , probability of death , and death delay ( Figure 6A , B ) . For each combination of traits we grow a population of clusters from a single cell , Node 0 , and compute the distribution of cluster sizes . This is repeated 100 times and we compare the strategies across different cluster size thresholds . For weak thresholds that permit small clusters of less than 25 cells , the most clusters are left by those without degree caps who have a probability of death of and no death delay . This strategy also produced the most living cells without considering cluster size thresholds ( Figure 4D ) . For intermediate cluster thresholds between 25 and 1000 cells , a degree cap of 4 with a probability of death of is best . As the size threshold increases within this range so does the optimal death delay . For cluster size selection between and the best strategy shifts back to clusters without degree caps who have a probability of death of and death delays above 0 . The largest cluster size selection ( ) finds the lowest probabilities of death with all degree caps doing well . In general , these results hold if the number of generations in the simulations is reduced from 21 to 19 . We next consider how the best combinations of traits for different size thresholds fare in cluster offspring symmetry . We compute the average size of offspring cluster for the best strategies ( Figure 6C ) and find that they produce much smaller offspring than the observed experimentally: of the parent's size for small cluster selection , for intermediate clusters , and for large clusters . Although they fall short , only a degree cap of 3 with the highest probability of death left more symmetrical cluster offspring ( ) . The symmetry of offspring did not compensate for the limits such a stringent degree cap places on population size . To see if higher rates of cell death could be selected for in the context of evolving populations , we expand our simulations to include mutations and repeated cycles of growth and selection mimicking the transfers of the experiment [28] . We start the simulation with a single cell , Node 0 with no degree cap and a low rate of cell death ( ) . Along with each round of cell reproduction and death , there is a round of mutation . Cells mutate with a probability of ( assuming there are many mutations that affect the probability of cell death ) and are assigned a new probability of death randomly sampled from a uniform distribution between and . After the population grows cell divisions , we randomly pick clusters based on their diameter as a proxy for settling speed until we have 10% of the population . This process of growth and selection is repeated for 100 transfers ( Figure 7 ) . The result is a rapid growth in the average probability of death for the population . The probability of death reaches within a similar number of transfers as was found experimentally ( 60 transfers [28] ) .
Population simulations compute the growth and reproduction of clusters from a single cell , the first mutant to form clusters , Node 0 . The simulation approach involves agent-based tracking of cell information , including the cluster it belongs to , the shell it is in , its parent cell , its number of offspring , the size of each of its branches ( for biased link severance ) , and the time it last divided . Simulations have discrete time steps representing cell generations . At the start of a time step , all cells reproduce so long as they satisfy three conditions: 1 . they are alive , 2 . they have not reached the maximum degree ( degree cap ) , and 3 . there is room in the next shell where their offspring will reside ( volume constraints ) . Following reproduction , we implement cell death . Initially all cells are susceptible and have a constant probability of death . This assumption is relaxed at times to delay death until a cell has been unable to divide a certain number of time steps , effectively protecting younger cells from death . If a cell dies then one of its links is randomly severed , biased by the number of cells along that branch ( weighting ) . Dead cells remain attached to a cluster and no longer reproduce . Each simulation goes through 21 rounds of reproduction and death ( cluster reproduction ) which would allow planktonic cells to reach populations of cells . At the end of the simulation , the distribution of cluster sizes is computed as well as the total number of living cells in those clusters . Simulations are done using the numerical software MATLAB ( version 7 . 12 . 0 . 635 Natick , Massachusetts: The MathWorks Inc . , 2011 ) .
An experiment exploring the emergence of multicellularity observed the rapid evolution of groups from unicellular precursors in the yeast Saccharomyces cerevisiae when cultures were placed under selection for rapid settling through liquid medium [28] . Soon after the establishment of groups , cells also evolved a higher rate of apoptosis . Elevated cell death clearly lowers cell viability , but it would also seem to lower group viability . This is because settling selection favors large clusters and cell death facilitates group division , and thus size reduction . Why would natural selection favor elevated– as opposed to reduced– levels of apoptosis ? Here we show that the organization of the group and the constraints imposed by its geometry are instrumental in understanding the functional consequences of apoptosis . By increasing the frequency of cell death , both the number of cells and groups can increase . Thus , a trait which is harmful to the cells that express it ( they die ) acts as a form of suicidal altruism and is beneficial to both the long-term number of cells and group entities once the group structure is considered . Furthermore , this trait may play a key role in the evolutionary transition to multicellularity . With the transition from unicellularity to multicellularity there is an important shift in the level of organization and individuality [1]–[3] . A key requirement for multicellularity is formation of a cohesive group of cells . Group formation offers distinct advantages over a strictly solitary lifestyle such as protection from predation [31] , [32] , access to new niches [4] , and survival in harsh environments [33] . However , for groups to qualify as units of selection , they must also possess the capacity to beget group offspring [27] , [34] , [35] . In this experimental yeast system , clusters grew in size and as a result of cell death or physical strain they fragmented and thereby reproduced . As a result , the yeast simultaneously evolved group formation and a mode of reproduction de novo . The later evolution of increased cell death led to more frequent cluster reproduction , thereby , linking reproductive self-sacrifice at the lower-level to fecundity at the higher-level . From a certain perspective , the fitness of the apoptotic lower-level units is subjugated to elevate the fitness of higher-level units , which is taken to be a hallmark of an evolutionary transition in individuality [1] , [2] , [27] , [36] , [37] . Interestingly , evolution of increased cell death also acts to stabilize the transition to multicellularity . If a cell with a higher rate of death were to leave the context of its collective , it would not fare well in competition with other cells who never formed groups ( and never evolved greater apoptosis ) . In this way , the trait ratchets cells into a multicellular lifestyle by making them less competitive with their unicellular ancestors . This prevents abandonment of the collective and reversion to unicellularity . By tying the fate of cells to the fate of their groups , such context dependent traits stabilize primitive multicellular forms . The amount of stabilization provided by a context dependent trait would likely depend on the fitness tradeoff between unicellular and multicellular life . More stabilization is expected from traits that severely hamper the fitness of cells outside the group context . It is unknown whether such stabilizing traits are common but with the yeast system analyzed in this paper there is robust selection for increased apoptosis rates . Rather than finding a narrow range of conditions that selected for moderately higher rates of cell death , we found strong selection for high rates of cells death ( ) across a wide spectrum of cluster size thresholds . In fact , the only regime where increased cell death does not succeed is when groups need to be close to the maximum possible size . This regime selects for the lowest cell death rates and results in a single group encompassing the entire population . Otherwise , when size selection required minimum group sizes from 0 to , high rates of cell death allowed cells to circumvent limitations imposed by geometry . Interestingly , these limitations were of two different classes: limits to the number of connections and limits to space . The relative importance of each limitation depends on the geometry . Limits to connections are stronger for trees with a maximum node degree of 3 and under while limits to space are more restrictive when the maximum degree is 4 and higher . As a consequence , a gamut of different tree geometries encounter limitations to growth that robustly select for high rates of apoptosis . Although using selective regimes based on Ratcliff et . al . experiment [28] , favored high rates of cell death similar to those observed in the experiment , it did not match the same magnitude of offspring-parent ratio ( ) . One reason for the mismatch with experimental data could be the compounding effects of cell death on the reproduction of the clusters . Experimental populations undergo repeated generations of cell reproduction and death which alters the geometric arrangements of cells . In comparison , the population simulations of Figures 4-7 contain nascent clusters who have grown from a single cell over the course of only 21 generations . They may not have had enough time to accumulate dead cells which alters their structure and promotes the birth of larger cluster offspring . Another possible reason for the mismatch could be due to our implementation of cell death . If cell death occurs under conditions of low nutrient concentration or build-up of cellular waste products , then cells in the center of clusters may be substantially more likely to die , which would produce relatively larger offspring clusters . Also , our models do not explicitly incorporate physical forces within growing clusters , which could affect likely break points , and thus , relative size of offspring clusters . Our model implicitly assumes that the environment in which cells and groups grow is nutrient rich , and that the death of a cell provides the possibility for replacement by future cells . This allows apoptosis to overcome the cost of sacrificing a cell through the benefit of additional cellular reproduction . If , instead , the environment were nutrient poor and death of a cell did not guarantee replacement , then high rates of apoptosis would encounter an additional cost not reflected in our model , and would likely be less successful . It is possible that the model could be modified to consider cell survival as a function of crowdedness rather than cell fecundity . Cells in more crowded areas have less access to nutrients and by dying could create more access for neighbors , potentially improving their survival . These considerations lie outside the scope of this paper . In the experimental regime , as in the model , populations were grown in nutrient rich environments and so increased apoptosis led to both higher group and cell number . Still , it is important to recognize that the fitness consequence of traits depend on both the environment established by the group as well as its external environment . While there have been many theoretical studies on the evolution of division of labor within multicellular organisms , modeling that division in the context of the multicellular geometry represents an under-explored direction . Considering the fitness implications of group geometry reveals that the group represents a novel , dynamic environment , one constructed bottom-up by individual cells . As such , variations in cellular physiology affect the geometry of the cluster , which in turn affects cell growth and survival . For example , if a cell has a morphology that only permits three connections to other cells , then the maximum possible cluster size will be much smaller than a cellular morphology that permits four connections . Similarly , different group formations impose different selective pressures on the cells within groups . The difference between three and four connections determines when cells will run out of space within the group . Although we considered a simple model with identical cells defined by just a few properties ( maximum degree , apoptosis rate , and death delay ) , we found that these traits interact in surprising ways . For instance , increasing cell death increased the number of living cells but delaying death for cells– effectively reducing the apoptosis rate– had contrasting effects depending on the maximum degree and duration of the delay . We only investigated how altering cell death affects cluster size and the number of cells/clusters in the population , but it is possible that cells could evolve different shapes , sizes , or behaviors which modify whole group-level traits . In fact , recent work has shown that strong selection for faster settling results in the evolution of larger , more elongate cells , which increase both group size and settling speed [38] . The environment faced by cells in clusters is not uniform: cells in the interior should experience a lower concentration of resources ( as they must diffuse past other yeast that are consuming them ) and higher concentrations of waste products . These environmental gradients provide reliable cues that could allow a cell to determine its position within the cluster . As cells change their location within the geometry and experience different internal environments , it may be advantageous to adopt different strategies or forms . This raises the possibility that selection can favor location-specific morphological or behavioral differentiation . Indeed , this may provide an evolutionary origin to primitive multicellular developmental programs .
|
A major transition in evolution occurs when previously autonomous entities become co-dependent in the context of a higher-level entity . Such transitions include the evolution of multicellular organisms from unicellular ancestors and eusocial “superorganisms” from multicellular ancestors . The evolution of reproductive division of labor occurs after some of these transitions ( e . g . , germ-soma differentiation in multicellular organisms ) . Yet , how exactly this occurs is unknown . Here , we examine this issue in the context of an experimental model of primitive multicellularity that evolved a form of reproductive division of labor de novo . Cells within groups evolved higher rates of death . Through cellular death , groups of cells split apart and formed new groups in a form of collective reproduction . The evolution of this trait is puzzling since cells originally formed groups under selection for large size . Group splitting produces smaller groups which are less likely to survive in the experiment . We show that the organization of the group is key to understanding evolution of increased cell death . Due to the arrangement of cells , higher rates of cell death increase both the number of cells and groups that survive . Reproductive division of labor evolves because the group context changes the fitness value of traits .
|
[
"Abstract",
"Introduction",
"Results",
"Methods",
"Discussion"
] |
[
"evolutionary",
"biology",
"theoretical",
"biology",
"biology",
"and",
"life",
"sciences",
"microbiology"
] |
2014
|
Geometry Shapes Evolution of Early Multicellularity
|
To understand the complex relationship governing transcript abundance and the level of the encoded protein , we integrate genome-wide experimental data of ribosomal density on mRNAs with a novel stochastic model describing ribosome traffic dynamics during translation elongation . This analysis reveals that codon arrangement , rather than simply codon bias , has a key role in determining translational efficiency . It also reveals that translation output is governed both by initiation efficiency and elongation dynamics . By integrating genome-wide experimental data sets with simulation of ribosome traffic on all Saccharomyces cerevisiae ORFs , mRNA-specific translation initiation rates are for the first time estimated across the entire transcriptome . Our analysis identifies different classes of mRNAs characterised by their initiation rates , their ribosome traffic dynamics , and by their response to ribosome availability . Strikingly , this classification based on translational dynamics maps onto key gene ontological classifications , revealing evolutionary optimisation of translation responses to be strongly influenced by gene function .
The expression of genes can be considered as a two-stage process , beginning with transcription and the production of an mRNA , followed by translation of that mRNA into protein by the cell's ribosome population . Gene expression must be tightly regulated to control protein composition , enabling the cell to rapidly respond to a wide range of environmental conditions . For this reason , cells exert fine control over gene expression , both at the transcriptional [1] , [2] and post-transcriptional level [3]–[6] . One key mechanism of post-transcriptional control of gene expression is translational regulation . The process of translation can be divided in three main phases , namely initiation , elongation and termination . Whereas termination is generally believed to be a fast process and therefore not limiting for translation [7] , the respective contributions of initiation and elongation to translational regulation are still under debate [8] . On one hand , the translation initiation rate , or the rate at which ribosomes access the 5′ untranslated region ( 5′ UTR ) and start translating the ORF , is regulated in part by formation of secondary structures in the 5′ leader [9] , [10] . The presence of secondary structures inhibits the ability of an mRNA to sequester ribosomes , thereby lowering the effective translation initiation rate . The 5′ leader composition is characteristic of each mRNA , resulting in a heterogeneity of the ribosome recruitment process among the transcripts [11] , [12] . Despite the importance of this process in gene expression regulation , there are currently no estimates of in vivo , mRNA-specific translation initiation rates based on refined traffic models , and how they regulate genome-wide patterns of protein expression . On the other hand , there is increasing evidence that translation elongation itself controls gene expression , being regulated by the rate of supply of tRNAs , particularly in microorganisms with codon biased genomes . Within families of isoacceptor tRNAs , members are not all present at the same concentration in the cell , leading to variation in delivery times , and the introduction of stochastic pauses [13] . Such pauses control ribosome transit , regulating ribosome queue formation . There is evidence that a ramp of slow codons near the 5′ end of some open reading frames regulates the flow of ribosomes onto an mRNA [14] , [15] , and pausing during elongation on any mRNA will affect queue dynamics , and thus the flux ( or current ) of ribosomes along the mRNA . However , there is no knowledge of how , on a genome-wide scale , the dynamic flux of ribosomes along an mRNA might be crucial in regulating protein expression . Here , we address these two problems: first , we estimate mRNA-specific in vivo translation initiation rates on a genome-wide basis by integrating a computational model of mRNA translation with experimental datasets of ribosome occupancy . Crucially , we show that translation initiation rates are correlated with gene function . Second , we show that the translation dynamics response of each mRNA is characteristic of its gene ontology , by elucidating how ribosome traffic , moving with variable speed across the codon field , responds to a range of initiation rates . We also show that codon arrangement rather than codon usage , clearly separates mRNAs into distinct classes typified by their responses to variations of the translation initiation rate . This suggests that not only codon usage but also codon arrangement is a selectable determinant of gene expression .
Our model describes how ribosomes bind to the mRNA , move along it performing the translation , and dissociate from the mRNA at the stop codon , releasing the finished protein into the cytoplasm [16] . The mRNA is represented by a unidimensional lattice , with each site denoting a codon . Ribosomes are represented by particles occupying 9 codons [15] that attempt to bind the mRNA with a rate , provided that the binding region is not obstructed by another ribosome . The particle on-rate mimics the initiation of translation , in which several processes have been condensed into just one step . The factors influencing the initiation of translation , such as secondary structures in the 5′UTR , concentration of initiation factors and ribosome availability , are all included in this parameter and will be discussed below . Subsequently , ribosomes advance on the polynucleotide chain ( elongation ) following a two-state dynamics: ( 1 ) recognition of the cognate tRNA with rate depending on the codon , and ( 2 ) translocation towards the next codon with rate ( see Figure 1 ) . At the last codon , the ribosomes detach and release the protein with a rate ( termination ) . The cognate tRNA-capture rates can be estimated from data on tRNA abundances , which are assumed to be proportional to their gene copy numbers [17] , and by considering further corrections such as the wobble base pairing ( see Supplementary Information , Text S1 ) . Effects of competition for near-cognate and non-cognate tRNAS were found not to materially affect any of the conclusions of this study ( see Supplementary Information , Text S1 ) , and are therefore neglected . The translocation rate has been measured to be 35 [18] , and is codon independent . The termination rate is determined by the concentration of the release factors; the termination process is assumed to be fast , comparable to the translocation [7] . Moreover , the model takes into consideration steric interactions between ribosomes , so that even if a ribosome sitting on codon has already captured the cognate tRNA , it cannot translocate if the next codon is occupied by another ribosome . Hence , it is an exclusion process [19] exhibiting different regimes characterised by the flow of particles and by their density along the lattice . In particular , if the sequence contains slow sites , then queues of particles behind the slow sites or high density phases appear when the on-rate of particles is of the same magnitude as the bottleneck rate . In contrast to commonly used exclusion models [14] , [20] , our model accounts for the processes involved in the mechano-chemical ribosome cycle , condensing them in two main steps: capture of the tRNA and translocation . It includes the crucial fact that ribosomes can capture a cognate tRNA while they wait for the next lattice position to become vacant . In contrast , ribosomes from simpler exclusion models unrealistically “lose” immediately the captured tRNA if they cannot move to the next codon . This is a key difference , which leads to different dynamics of ribosome traffic and transitions between traffic regimes [16] . This effect is further enhanced by the fact that the time scales related to the capture of the tRNA and translocation are strongly separated , with the translocation being much faster . Furthermore , the two-state ribosome reproduces the dwell-times observed in single-molecule experiments [21] . In summary , our model predicts the current of ribosomes or translation rate , and the density of ribosomes on a particular mRNA ( number of ribosomes divided by the ORF length ) , taking as input the specific sequence of codons of the mRNA . Both the translation rate and the ribosome density are predicted as a function of the translation initiation rate , i . e . the rate at which ribosomes arrive at the start AUG codon; their functional dependence on the initiation rate thus varies from sequence to sequence as a consequence of different codon compositions and codon arrangements . The translation initiation rate , i . e . the rate at which ribosomes start translating the ORF , depends on many factors , such as the rate at which ribosomes attempt to bind the mRNA , the concentration of initiation factors and the presence of secondary structures in the 5′UTR region [9] , [11] , [12] . Despite the key role of this parameter , direct experimental evaluations are intractable , both in vivo and in vitro , with no direct measurements having been carried out to date . Previous works , such as [20] , could only estimate the translation initiation rate as the value that maximised the predicted correlation of the ribosome current with experimental data . Furthermore , has usually been considered as a unique , fixed value ( the same for each of the mRNAs ) , but it is well known that the translation initiation rate depends on several mRNA-specific factors , such as the structural properties of the mRNA leader region . Knowledge of mRNA-specific values of , therefore , would provide important insight into control of gene expression at the level of translation . Siwiak and Zielenkiewicz [22] present specific initiation rates , however with a simple model that neglects ribosome kinetics and traffic ( the comparison is discussed in the Supplementary Text S2 ) . Here we present a novel approach to identify the initiation rate of each individual mRNA for the whole genome . We first apply our translation model to all mRNA sequences of S . cerevisiae . The model predicts the translation rate and the ribosome density on each mRNA as a function of the translation initiation rate . Then , by utilising genome-wide experimental data of ribosome density from [23] for yeast grown under non-stressed conditions , we identify the physiological translation initiation rate as the one which , when used in our simulations , replicates the experimentally observed density: ( 1 ) This yields a value of the translation initiation rate for each mRNA as shown by the genome-wide distribution in Figure 2 . Using the genome-wide experimental data of ribosome density from Arava et al . [7] yields a very similar distribution of initiation rates ( see Section 4 of Supplementary Text S1 ) . The knowledge of this distribution reveals how translational regulation of gene expression works at the level of initiation by correlating the values of with the biological functions of the corresponding genes , encoded in their Gene Ontology ( GO ) annotations . In Figure 2 we split up this distribution in four parts , from small to high ( i ) – ( iv ) . Strikingly , significantly enriched GO annotations are identifiable in each of the regions . Messenger RNAs with an initiation rate below ( region ( i ) of Figure 2 ) contain a highly disproportionate number of regulatory proteins and proteins linked to transcription from Pol II promoters , mainly located in the nucleus , chromosome , membrane or protein complexes . In the range of from to ( region ( ii ) of Figure 2 ) we find other significantly over-represented terms such as cytoplasmic translation , ribosome biogenesis or oxoacid metabolic process , while genes with from to ( region ( iii ) of Figure 2 ) are primarily constituents of ribosomes . Very large initiation rates ( region ( iv ) of Figure 2 ) are characteristic of genes associated with the respiratory chain . However , most of genes falling in this region are not annotated ( a complete list of can be found in the Supplementary Table S1 and the details of the GO analysis , with the annotations found in each region and their enrichments , can be found in the Supplementary Table S2 ) . The assignment of initiation rates correlates with protein abundances typical of given GO categories: regulatory proteins are usually present at low levels . In contrast , proteins involved in translation , ribosome biogenesis and metabolic processes are abundant . This result is a signature of the divergent translational control that distinct genes exhibit at the level of initiation , suggesting that factors influencing , such as secondary structure in the 5′ leader region , have been shaped by evolution to contribute to the delicate balance of cellular protein composition . To show that the procedure introduced above can be applied under different conditions , we carry out a similar analysis under pheromone treatment by using the corresponding measurements of ribosome densities from [23] , and estimate the initiation rates under these conditions . The initiation rates do not substantially change , consistent with the finding by Mackay et al . [23] that only a small number of mRNAs exhibit altered densities after pheromone treatment . However , with our analysis we identify two mRNAs , SAG1 and HO , which exhibit a radical change in their initiation rate value under pheromone treatment . Importantly , these two mRNAs have been shown to present altered 5′UTR sequences that explain their significant ribosome density change [24] . Now we analyse the influence of the physical properties of the mRNA on the translation initiation rate by analysing the correlations of the identified rates with the presence of secondary structures in the 5′UTR and the length of the transcript . The physiological estimates of the initiation rate show a small but significant correlation with the free energy of the secondary structures in the 5′UTR ( Spearman's rank , p-value ) confirming that secondary structures may have an important regulatory role , as already suggested [9] , [25] , [26] , see Supplementary Information Text S1 . Remarkably , we find a strong negative correlation between the initiation rate and the length of the ORF ( Spearman's rank , p-value ) , see Figure 3 . Of relevance to this observation , Arava and coworkers [7] found that ribosome density counter-intuitively and systematically decreases with increasing the ORF length . In a subsequent work [27] , they reported that the explanation most consistent with their experimental investigation was that lower initiation rates predominate on longer mRNAs , exactly as we estimate in this work . In order to show that the genome-wide correlation between translational efficiency and biological function obtained above is not only the consequence of codon usage but is strongly influenced by the order in which codons are used in the mRNA , we simulate the translation of a library of randomised ORFs such that both amino acid sequence and codon composition remain identical . That means , two ORFs belonging to the library have exactly the same codon usage but the arrangement of these codons is different . Here we show that , even though all these randomised ORFs have exactly the same codon usage indices such as the CAI , codon adaptation index [32] , and tAI , tRNA adaptation index [33] , their predicted protein production rate can be very different . Figure 6 shows how the different values of predicted protein production rate ( ribosomal current for a fixed initiation rate ) are distributed for 2 , 000 synonymous randomised codon sequences of a typical Saccharomyces cerevisiae gene ( YPL106C ) . Figure 6 clearly shows that the relative positioning of codons has a crucial effect on the translation efficiency , suggesting that very different cellular production rates can be achieved through evolution of the codon arrangement . For instance , in the case shown in Figure 6 there is an increase of about from the lowest to the highest value of the translation rate . The variation of for different codon arrangements is a general result and does not depend on the gene or the chosen initiation rate ( for more information see Supplementary Information , Text S1 ) . We thus show that by randomising codon arrangement ( i . e . randomly exchanging the position of synonymous codons in a sequence ) , different protein production rates are obtained , even though codon usage remains fixed . This indicates that the codon arrangement has a highly significant role in determining the efficiency of translation . While several models of protein synthesis have been developed over the last decades [34] , the role of codon sequence and stochastic ribosomal movement has been investigated only recently . But even recent models typically treat the initiation rate as a fixed parameter , identical for all mRNAs , despite its key role in determining translational efficiency . In contrast , our model predicts the protein production rate as a function of the initiation rate . By then integrating genome-wide simulations with datasets of polysome sizes , we have identified the physiological value of the initiation rate for each mRNA . This set of values then leads to the prediction of the protein production rate for each transcript . This allows us to validate our model predictions with experimental data . Figure 7A is a scatter plot of the genome-wide simulations versus measured protein abundance from [30] . The model predictions for , where denotes mRNA abundance , correlate very well with the experimental protein abundances ( Spearman's rank = 0 . 64 , p-value ) , compared to other attempts such as the tAIc ( Spearman's rank = 0 . 38 , p-value ) . Our outcome is further improved when considering just transcripts loaded onto polysomes ( Spearman's rank = 0 . 66 , p-value ) , see ‘Materials and Methods’ and Supplementary Information , Table S1 . Moreover , as it can be appreciated from Figure 7A , the predictions from our model correlate very well with measured protein abundance for all ranges of gene expression , in contrast to other translation efficiency indices ( panels B , C , D ) , which exhibit a poor correlation for lowly expressed genes .
The phenotype exhibited by any cell is dictated by its proteomic composition . How much of each type of protein is expressed is governed by a range of factors , including the level of transcription and stability of the encoding mRNA , the half-life of the protein , and how efficiently its mRNA is translated . A number of strategies have been employed to predict translational efficiency , many of which utilise the observation that not all codons are used with equal frequency , and that codon usage frequency is proportional , at some level , to the abundance of the corresponding decoding tRNA species [35] , [36] . Initially , measures such as the codon adaptation ( CAI ) index were developed [32] , which correlate high protein abundance with over-use of the sub-set of codons found in a group of very-highly expressed genes , normally those encoding the ribosomal proteins . However , such approaches frequently struggle to predict the expression level of less abundant proteins . More recently , dynamic TASEP ( Totally Asymmetric Simple Exclusion Process ) models have been employed to simulate the flow of ribosomal traffic , including queuing interactions between adjacent ribosomes on the polysome [20] , [37]–[41] . Even though these models represent a big step towards a more complete description of the translation process , most of them miss one essential component , namely the mechano-chemical ribosome cycle . By including this mechanism into an exclusion process we showed that the mathematical description of translation becomes much more accurate [16] . Here we applied this model to simulate the translation of every mRNA in the transcriptome of S . cerevisiae leading to the estimates of the individual translation initiation rates unique to each of the 6 , 000 genes in yeast . We furthermore showed that mRNA sequences can be classified according to their ribosome traffic characteristics , and crucially , this classification maps to gene ontology assignments . Even though the role of the translation initiation rate has been shown to play a central role in translational control of gene expression [9] , to our knowledge no genome-wide estimations of these rates have been reported , considering ribosome traffic effects . The translation initiation rate , i . e . the rate at which ribosomes start translating the ORF , condenses many factors , such as cytoplasmic ribosome availability , initiation factors and secondary structures on the 5′UTRs , all of them strongly dependent on nutrients and stress conditions . Some approaches consider the translation initiation rate to be fixed for every transcript , thereby neglecting the key factors that make the initiation rate unique to each transcript . In contrast , by considering traffic dynamics , we determined the first genome-wide estimate of initiation rates for each and every mRNA ( Figure 2 ) by integrating our stochastic model of ribosome traffic with data of ribosome densities across all mRNAs [23] . Our analysis showed a wide range of values under these non-stress conditions . Importantly , the values are strongly correlated with gene function , explaining for example why translation of ribosomal protein mRNAs , which typically have a very high value , is very efficient . These values of are expected to be influenced by the degree of secondary structure of the 5′ leader sequence , and indeed we did find a significant correlation with the free energies of the secondary structures . The strongest connection involving was however a negative correlation with mRNA length , mirroring the findings from experimental research that described lower ribosome densities on longer mRNAs [7] , [15] . In contrast to the explanation that the effect could be caused by bottlenecks of slow codons [15] ( see Supplementary Information Text S1 ) this negative correlation supports the idea that due to the circular structure of mRNAs , the ends of shorter mRNAs can interact more easily than longer mRNAs , thereby promoting ribosome recycling [7] , [42] , [43] . Indeed following detailed experimental analysis using ribosome density mapping , Arava and colleagues concluded that lower densities on longer mRNAs are best explained by lower rates of translation initiation [27] , mirroring our findings in this work . To summarise , we interpret the correlation between the estimated initiation rates and the ORF lengths as a possible indication of a regulatory mechanism that allows circularised mRNAs to load ribosomes more efficiently onto their transcripts , leading to the observed ribosome-ORF length relationship . Our analysis furthermore identified two main distinct classes of mRNAs regarding their responsiveness to changes in the initiation rate : some sequences exhibited an abrupt change in the polysome size upon a change in , whereas smooth sequences showed a gradual increase . Calculations with artificial sequences revealed that sequences with rare codons in the main body of the ORF belong to the abrupt class , whereas sequences with either no rare codons or rare codons at the 5′ end belong to the smooth class [28] . Crucially , we note that the classification of mRNAs into smooth and abrupt responders maps onto particular gene ontological classifications . Smooth responder mRNAs as a class are highly over-populated with ribosomal protein mRNAs and translation factors . Conversely , the abrupt class contains disproportionate numbers of regulatory proteins , including nucleic acid-binding transcription factors , and cell cycle proteins . One reason why ribosomal protein mRNAs are predominantly of the smooth response type might relate to the massive manufacturing scale of ribosome biosynthesis; in yeast , ribosomal protein mRNAs account for nearly 30 percent of all mRNAs [44] , [45] . Smooth-type responses to must be of selective advantage for a cell , since if ribosome queues were established on such a large proportion of the cell mRNA population they would sequester a large numbers of ribosomes , with deleterious consequences for cell fitness . On the other hand , it has been recently found that cell-cycle regulated genes predominantly adopt non-optimal codon usage ( with no ramp of slow codons at the beginning , and therefore of the abrupt class ) to achieve elongation-limited mRNA translation; this can generate cell cycle-dependent oscillations in protein abundance induced by changes in the tRNA pool [46] . Therefore , it is apparent that the cell coordinates codon usage and codon arrangement to achieve translational gene expression control . Our results also showed important differences in the computationally deduced slope of the production rate curve in response to increasing . Some mRNAs are what we term highly geared , that is , small increases in produce relatively large increases in . This type of super-responsive mRNA was significantly enriched in regulatory proteins , which also have a relatively small initiation rate . We speculate that this might be a mechanism to facilitate rapid responses to changed environmental conditions , allowing , for example , rapid synthesis of transcriptional repressors that in physiological conditions are severely limited by the initiation ( low ) . Conversely , low geared mRNAs , where increases in produce proportionately lower responses in , were enriched in ribosomal proteins . Since ribosomal proteins are used to manufacture ribosomes , lower gearing of the responsiveness to may help prevent undesirable positive feedback effects . We furthermore classified mRNA sequences according to the maximal translation rate that they can achieve , i . e . their saturation value , and our analysis revealed that abrupt sequences have predominantly a small , whereas smooth sequences are characterised by a large . This correlates with the levels of the corresponding proteins: regulatory proteins are typically present in low abundance , whereas ribosomal proteins are highly abundant . Moreover , this might prevent possibly deleterious consequences of over-producing regulatory proteins , including cell cycle factors , during occasional bursts of ribosomal availability that would lead to a very large increase in the value of . In S . cerevisiae for example , this occurs upon sudden glucose depletion: translation initiation is rapidly inhibited [47] but some mRNAs ( including those involved in carbohydrate metabolism ) continue to be translated [48] , thereby being exposed to a spike in ribosome availability . Similar complex translational re-programming , coincident with a partial cell-wide shut down of translation initiation , occurs in response to oxidative stress [49] . Hence , by having a high responsiveness to and a low , abrupt sequences can have a very rapid gene expression upregulation , on one hand , but a controlled maximum translation rate , on the other hand . Figure 8 summarises our findings on the initiation rates and the consequences of different ( mRNA-specific ) dependencies of the protein production rate on the initiation step . Incidentally , this classification of proteins according to their translation dynamics , coincides with the classification according to protein stability . In [50] , the S . cerevisiae proteome was analysed using a clustering approach to classify proteins according to half-life , and the stable protein cluster was enriched with proteins involved in protein production , including ribosomal proteins and enzymes involved in amino acid metabolism . Moreover , the unstable protein cluster was enriched with cell cycle proteins and proteins involved in transcriptional regulation . Therefore , our analysis indicates that stable proteins tend to have a low responsiveness in their production rate to external changes which change the initiation rate , whereas unstable proteins production responses very effectively to external changes . Hence , our analysis strongly suggests that the cell coordinates dynamics of protein degradation with the dynamics of protein production . In summary , we have shown how our stochastic model representing the ribosome traffic flow on mRNAs is able to discern and describe the biological interplay between translation initiation and elongation , at a single-codon level . We have illustrated how the application of this model across the entire genome can be used to infer mRNA-specific translation initiation rates in vivo , and that selection of codon arrangement is likely to be an important mechanism to tune the translation system to meet the competing demands of ribosome biosynthesis and translation of all other mRNAs in the cell . With our approach , mRNA sequences can be classified according to their translation dynamics , mapping to key gene ontological classifications; codon arrangement plays a fundamental role in this classification , indicating that it is optimized through evolution to match the corresponding gene function . Moreover , gene-specific physiological values of initiation rate can be used to determine the translational efficiency for each mRNA; this allows the prediction of genome-wide protein abundances with a significant increase in correlation when compared with previous approaches ( Figure 7 ) . We foresee this type of analysis will be of great value to understand how the economics of translation are regulated on a cell-wide basis , and how codon arrangement is optimised to control gene expression in response to the translational remodelling that occurs in response to many environmental stresses .
For each mRNA sequence of S . cerevisiae we performed a stochastic simulation of translation , one mRNA at a time , following the rules explained above and summarised in Figure 1 . Our algorithm is a continuous time Monte-Carlo based on the Gillespie algorithm , and therefore it gives the real-time dynamics of the system . In each simulation of individual mRNAs we let the system reach the steady-state . Then we measured , at constant interval times , two quantities: the current of ribosomes along the mRNA , i . e . , how many ribosomes per unit time finish translation , and the density of ribosomes on the mRNA , i . e . , the total number of ribosomes divided by the length ( in codons ) of the mRNA . Therefore , the current gives the translation rate , and the density determines the polysome size . We then averaged these quantities over the entire time interval of the simulation . We ran the simulations for a broad range of initiation rates between 0 and 5 , making sure that the plausible physiological regime for variation was covered , and we fixed the other parameters as explained in the previous sections . The obtained curves and were then smoothed with a ten-points running average . The gradient of the translation rate at the physiological initiation rate is defined , for each mRNA , as the numerical derivative of the relation , computed at . It geometrically represents the slope of the curve at the physiological value . Since both the distributed physiological values of the initiation rates and different codon sequences cause a different dependence of on , the derivative differs from mRNA to mRNA . The maximal values and were defined as the mean of the last five simulation points of the current and the density , respectively , corresponding to the five largest values of considered . The translation efficiency of a transcript is defined as the protein production rate computed at the physiological value , . Denoting by the amount of a specific mRNA in the cell ( data from [30] ) , for any protein the quantity is an estimate of the protein abundance , see [20] . We also considered the effective amount of transcript involved in polysomes , , where can be found in [23] . The prediction slightly improves the correlation with measured protein abundance , as discussed in the ‘Results’ section .
|
Gene expression regulation is central to all living systems . Here we introduce a new framework and methodology to study the last stage of protein production in cells , where the genetic information encoded in the mRNAs is translated from the language of nucleotides into functional proteins . The process , on each mRNA , is carried out concurrently by several ribosomes; like cars on a small countryside road , they cannot overtake each other , and can form queues . By integrating experimental data with genome-wide simulations of our model , we analyse ribosome traffic across the entire Saccharomyces cerevisiae genome , and for the first time estimate mRNA-specific translation initiation rates for each transcript . Crucially , we identify different classes of mRNAs characterised by different ribosome traffic dynamics . Remarkably , this classification based on translational dynamics , and the evaluation of mRNA-specific initiation rates , map onto key gene ontological classifications , revealing evolutionary optimisation of translation responses to be strongly influenced by gene function .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"physics",
"systems",
"biology",
"molecular",
"cell",
"biology",
"statistical",
"mechanics",
"theoretical",
"biology",
"protein",
"translation",
"gene",
"expression",
"biology",
"computational",
"biology",
"molecular",
"biology"
] |
2013
|
Ribosome Traffic on mRNAs Maps to Gene Ontology: Genome-wide Quantification of Translation Initiation Rates and Polysome Size Regulation
|
Gene expression variation is extensive in nature , and is hypothesized to play a major role in shaping phenotypic diversity . However , connecting differences in gene expression across individuals to higher-order organismal traits is not trivial . In many cases , gene expression variation may be evolutionarily neutral , and in other cases expression variation may only affect phenotype under specific conditions . To understand connections between gene expression variation and stress defense phenotypes , we have been leveraging extensive natural variation in the gene expression response to acute ethanol in laboratory and wild Saccharomyces cerevisiae strains . Previous work found that the genetic architecture underlying these expression differences included dozens of “hotspot” loci that affected many transcripts in trans . In the present study , we provide new evidence that one of these expression QTL hotspot loci affects natural variation in one particular stress defense phenotype—ethanol-induced cross protection against severe doses of H2O2 . A major causative polymorphism is in the heme-activated transcription factor Hap1p , which we show directly impacts cross protection , but not the basal H2O2 resistance of unstressed cells . This provides further support that distinct cellular mechanisms underlie basal and acquired stress resistance . We also show that Hap1p-dependent cross protection relies on novel regulation of cytosolic catalase T ( Ctt1p ) during ethanol stress in a wild oak strain . Because ethanol accumulation precedes aerobic respiration and accompanying reactive oxygen species formation , wild strains with the ability to anticipate impending oxidative stress would likely be at an advantage . This study highlights how strategically chosen traits that better correlate with gene expression changes can improve our power to identify novel connections between gene expression variation and higher-order organismal phenotypes .
A fundamental question in genetics is how individuals with extremely similar genetic makeups can have dramatically different characteristics . One hypothesis is that a small number of regulatory polymorphisms can have large effects on gene expression , leading to the extensive phenotypic variation we see across individuals . In fact , gene expression variation is hypothesized to underlie the extensive phenotypic differences we see between humans and chimpanzees despite >98% DNA sequence identity [1 , 2] . This hypothesis is supported by numerous examples of gene expression variation affecting higher-order organismal traits . For example , human genome-wide association studies ( GWAS ) have found that a substantial fraction of disease-associated variants are concentrated in non-coding regulatory DNA regions [3–8] . Further examples include gene expression variation being linked to differences in metabolism [9–11] , physiology [12–16] , morphology [17–23] , and behavior [24–27] . While gene expression variation is pervasive , there is often a lack of obvious phenotypic change associated with differentially expressed genes . This can occur for a variety of reasons . First , a large fraction of expression variation has been postulated to be evolutionarily neutral with no effect on organismal fitness [28–30] . Second , co-regulation of genes that share the same upstream signaling network and transcription factors can lead to genes whose expression differences correlate with phenotype but are not truly causative . Finally , some gene expression differences may truly affect phenotype , but only under specific conditions . For example , the predictive power of expression quantitative trait loci ( eQTL ) mapping studies on higher-order phenotypes can be poor unless multiple environments are considered [31] . Similarly , tissue-restricted eQTLs are more likely to map to known disease-associated loci identified from GWAS than non-tissue-restricted eQTLs [32 , 33] . Thus , a major challenge for connecting gene expression variation to downstream effects on higher-order traits is the choice of which conditions and traits to examine . To this end , we have been leveraging natural variation in the model eukaryote Saccharomyces cerevisiae , and a phenotype called acquired stress resistance . Many studies have shown a poor correlation between genes that respond to stress and their importance for surviving stress [34–43] . Thus , we and others have argued that the role of stress-activated gene expression is not to survive the initial insult , but instead protects cells from impending severe stress through a phenomenon called acquired stress resistance [44 , 45] . Acquired stress resistance ( sometimes referred to as “induced tolerance” or the “adaptive response” ) occurs when cells pretreated with a mild dose of stress gain the ability to survive an otherwise lethal dose of severe stress . Notably , acquired stress resistance can occur when the mild and severe stresses are the same ( same-stress protection ) or across pairs of different stresses ( cross protection ) . This phenomenon has been observed in diverse organisms ranging from bacteria to higher eukaryotes including humans [44–50] . The specific mechanisms governing acquisition of higher stress resistance are poorly understood , but there are wide reaching implications . In humans , ischemic preconditioning ( transient ischemia followed by reperfusion—i . e . mild stress pretreatment followed by severe stress ) may improve outcomes of cardiovascular surgery [51–54] , while transient ischemic attacks ( “mini-strokes” ) may protect the brain during massive ischemic stroke [55–57] . Thus , understanding the genetic basis of acquired stress resistance in model organisms holds promise for mitigating the effects of stress in humans . A previous study found that a commonly used S288c lab strain is unable to acquire further ethanol resistance when pretreated with a mild dose of ethanol [44] . We found this phenotype to be surprising , considering the unique role ethanol plays in the life history of Saccharomyces yeast , where the evolution of aerobic fermentation gave yeast an advantage over ethanol-sensitive competitors [58] . Because ethanol is a self-imposed stress that induces a robust stress response [59–63] , we expected that ethanol should provoke acquired stress resistance in wild yeast strains . Indeed , this turned out to be the case , with the majority of tested wild strains acquiring resistance to severe ethanol following a mild ethanol treatment [45] . Furthermore , this phenotype correlated with extensive differences in the transcriptional response to acute ethanol stress in the lab strain when compared to a wild vineyard ( M22 ) and wild oak ( YPS163 ) strain ( >28% of S288c genes were differentially expressed at an FDR of 0 . 01 ) [45 , 64] . We performed linkage mapping of S288c crossed to a wild vineyard strain ( M22 ) and wild oak strain ( YPS163 ) , and observed numerous “hotspots” where the same eQTL loci affect the expression of a large number of transcripts ( anywhere from 10–500 transcripts per hotspot ) [64] . In the present study , we provide new evidence that one of these eQTL hotspot loci affects natural variation in acquired stress resistance , namely the ability of ethanol to cross protect against oxidative stress in the form of hydrogen peroxide . The causative polymorphism is in the heme-activated transcription factor Hap1p , which we show directly impacts cross protection , but not the basal resistance of unstressed cells . Finally , we show that the Hap1p effect is mediated through novel regulation of cytosolic catalase T ( Ctt1p ) during ethanol stress in wild strains . This study highlights how strategically chosen traits that are better correlated with gene expression changes can improve our power to identify novel connections between gene expression variation and higher-order organismal phenotypes .
We previously found that an S288c-derived lab strain was unable to acquire further ethanol resistance when pretreated with a mild dose of ethanol , in contrast to the vast majority of ~50 diverse yeast strains [45] . In addition to the S288c strain’s acquired ethanol resistance defect , ethanol also failed to cross protect against other subsequent stresses [44 , 65] . In nature , wild yeast cells ferment sugars to ethanol , and then shift to a respiratory metabolism that generates endogenous reactive oxygen species [66–68] . Thus , we hypothesized that ethanol might cross protect against oxidative stress in wild yeast strains . We tested this hypothesis by assessing whether mild ethanol treatment would protect a wild oak strain ( YPS163 ) from severe oxidative stress in the form of hydrogen peroxide ( H2O2 ) . Cross protection assays were performed by exposing cells to a mild , sublethal dose of ethanol ( 5% v/v ) for 60 min , followed by exposure to a panel of 11 increasingly severe doses of H2O2 ( see Materials and Methods ) . Confirming the observations of Berry and Gasch [44] , ethanol failed to cross protect against H2O2 in S288c , and in fact slightly exacerbated H2O2 toxicity ( Fig 1 ) . In contrast , ethanol strongly cross protected against H2O2 in YPS163 ( Fig 1 ) . The inability of ethanol to induce acquired stress resistance in S288c correlates with thousands of differences in ethanol-dependent gene expression in comparison to wild strains that can acquire ethanol resistance [45 , 64] . In light of this observation , and the known dependency of cross protection on stress-activated gene expression changes [44] , we hypothesized that differences in cross protection against H2O2 by ethanol may be linked to differential gene expression . To test this , we performed quantitative trait loci ( QTL ) mapping using the same mapping population as our original eQTL study that mapped the genetic architecture of ethanol-responsive gene expression [64] . Specifically , we conducted QTL mapping of both basal and acquired H2O2 resistance in 43 F2 progeny of S288c crossed with YPS163 ( see Materials and Methods ) . While we found no significant QTLs for basal H2O2 resistance , we did find a significant QTL peak on chromosome XII that explained 38% of the variation in cross protection ( Fig 2 ) . It is unlikely that our failure to detect a chromosome XII QTL for basal H2O2 resistance was due to a lack of statistical power , because two independent basal H2O2 resistance QTL studies using millions of S288c x YPS163 F2 segregants also found no significant associations at this locus [69 , 70] . Additionally , we estimated the heritability of phenotypic variation in basal resistance to be 0 . 79 , which is slightly above the median value estimated by Bloom and colleagues for 46 yeast traits [71] , and is only moderately lower than the heritability for cross protection ( 0 . 92 ) . Lastly , the shape of the distribution of phenotypes in the F2 were markedly different between basal and acquired H2O2 resistance , with basal resistance showing a transgressive segregation pattern and acquired resistance showing a continuous distribution ( S1 Fig ) . Altogether , these results suggest that the genetic basis of natural variation in acquired stress resistance is distinct from the basal resistance of unstressed cells ( see Discussion ) . The significant QTL for cross protection was located near a known polymorphism in HAP1 , a heme-dependent transcription factor that controls genes involved in aerobic respiration [72–74] , sterol biosynthesis [75–77] , and interestingly , oxidative stress [77 , 78] . S288c harbors a known defect in HAP1 , where a Ty1 transposon insertion in the 3’ end of the gene’s coding region has been shown to reduce its function [79] . In fact , we previously hypothesized that the defective HAP1 allele was responsible for the inability of S288c to acquire further resistance to ethanol . However , a YPS163 hap1Δ strain was still fully able to acquire ethanol resistance , despite notable differences in the gene expression response to ethanol in the mutant [45] . Likewise , despite previous studies implicating Hap1p as a regulator of oxidative stress defense genes [77 , 78] , HAP1 is apparently dispensable for same-stress acquired H2O2 resistance [47] . These observations suggest that the molecular mechanisms underlying various acquired stress resistance phenotypes can differ , even when the identity of the secondary stress is the same . Because we previously implicated HAP1 as a major ethanol-responsive eQTL hotspot affecting over 100 genes , we hypothesized that ethanol-induced cross protection against H2O2 may depend upon Hap1p-regulated genes . However , it was formally possible that HAP1 was merely linked to the truly causal polymorphism . To distinguish between these possibilities , we generated deletion mutations in the YPS163 background for every non-essential gene within the 1 . 5-LOD support interval of the QTL peak ( encompassing IFH1 –YCS4 ) . Of the 36 mutants tested , two showed significantly and highly diminished acquired H2O2 resistance ( Fig 3 and S2 Fig ) , hap1Δ and top3Δ ( encoding DNA topoisomerase III ) . To determine whether different alleles of HAP1 and/or TOP3 were responsible for natural variation in acquired H2O2 resistance , we applied an approach called reciprocal hemizygosity analysis [80] , where the TOP3 and HAP1 alleles were analyzed in an otherwise isogenic S288c-YPS163 hybrid background ( see Fig 4A for a schematic ) . In each of the two reciprocal strains , one allele of the candidate gene was deleted , producing a hybrid strain containing either the S288c or YPS163 allele in single copy ( i . e . hemizygous for TOP3 or HAP1 ) . While we found only mild allelic effects for TOP3 , the effects of different HAP1 alleles were striking ( Fig 4B and 4C ) . The hybrid strain containing the HAP1YPS163 allele showed full cross protection , while the strain containing the HAP1S288c allele showed none . Thus , we examined the effects of HAP1 on acquired H2O2 resistance further . Intriguingly , we found that the YPS163 hap1Δ mutant was unaffected for acquired H2O2 resistance when mild H2O2 or mild NaCl were used as mild stress pretreatments ( Fig 5 ) , suggesting that Hap1p plays a distinct role in ethanol-induced cross protection ( see Discussion ) . Finally , we performed allele swap experiments to examine the effects of the different HAP1 alleles in the original parental backgrounds . We introduced only the Ty element from HAP1S288c into the YPS163 HAP1 gene , and observed a loss of acquired H2O2 resistance similar to the YPS163 hap1Δ strain ( Fig 6 ) . We next tested whether repair of the defective hap1 allele in S288c could restore cross protection . Surprisingly , S288c repaired with HAP1 YPS163 was largely unable to acquire further H2O2 resistance ( Fig 6 ) . This additional layer of genetic complexity suggests that S288c harbors additional polymorphisms that affect cross protection . To determine whether this was due to allelic variation in TOP3 , the only other locus showing a difference in acquired H2O2 resistance , we genotyped each of the segregants at both the HAP1 and TOP3 loci . We identified two segregants with both the HAP1 YPS163 and TOP3YPS163 alleles that were nonetheless unable to acquire further resistance ( S3 Fig , S1 Table ) . These data , along with the continuous distribution of F2 phenotypes ( S1 Fig ) , is consistent with other loci outside of the chromosome XII QTL peak contributing to variation in acquired H2O2 resistance . Moreover , the causative alleles at these loci are apparently masked in YPS163-S288c hybrids that fully acquire H2O2 resistance , suggesting that they are recessive ( see Discussion ) . We also noted during the genotyping that a small number of segregants contained the HAP1 S288c ( or TOP3S288c ) allele but were still able to acquire further H2O2 resistance ( S3 Fig and S1 Table ) , suggesting that HAP1 function is conditionally necessary in certain genetic backgrounds . To determine whether this was due to a unique genetic background for YPS163 , we deleted HAP1 in three additional wild strains . A wild oak ( YPS1000 ) and wild vineyard ( M22 ) strain showed defects in acquired H2O2 resistance similar to that of the YPS163 hap1Δ strain , while a wild coconut ( Y10 ) strain showed a very slight defect ( S4 Fig ) . Altogether , these results are consistent with HAP1 being necessary for ethanol-induced cross protection against H2O2 in some genetic backgrounds , including those of several wild strains , but not others ( see Discussion ) . Because Hap1p is a transcription factor , we hypothesized that acquired H2O2 resistance relied on Hap1p-dependent expression of a stress protectant protein . We reasoned that the putative stress protectant protein should have the following properties: i ) a biological function consistent with H2O2 detoxification or damage repair , ii ) reduced ethanol-responsive expression in S288c versus YPS163 , iii ) be a target gene of the HAP1 eQTL hotspot , and iv ) possess evidence of regulation by Hap1p . We first looked for overlap between our previously identified HAP1 eQTL hotspot ( encompassing 376 genes ) and genes with significantly reduced ethanol-responsive induction in S288c versus YPS163 ( 309 genes ) [64] . Thirty-four genes overlapped for both criteria , including several that directly defend against reactive oxygen species ( TSA2 encoding thioredoxin peroxidase , SOD2 encoding mitochondrial manganese superoxide dismutase , CTT1 encoding cytosolic catalase T , and GSH1 encoding γ-glutamylcysteine synthetase ( Fig 7A and S1 Table ) ) . Of those 34 genes , 8 also had direct evidence of Hap1p binding to their promoters [81] ( Fig 7B and S1 Table ) , including CTT1 and GSH1 ( though both TSA2 and SOD2 have indirect evidence of regulation by Hap1p [82 , 83] ) . We first focused on CTT1 , since it is both necessary for NaCl-induced cross protection against H2O2 in S288c [84] , and sufficient to increase H2O2 resistance when exogenously overexpressed in S288c [85] . We deleted CTT1 in the YPS163 background , and found that ethanol-induced cross protection against H2O2 was completely eliminated ( Fig 8 ) . The complete lack of cross protection in the ctt1Δ mutant suggests that other peroxidases cannot compensate for the lack of catalase activity under this condition . Next , because CTT1 was part of the HAP1 eQTL hotspot ( Fig 7C , plotted using the data described in [64] ) , we tested whether the S288c HAP1 allele reduced CTT1 expression during ethanol stress . To do this , we performed qPCR to measure CTT1 mRNA induction following a 30-minute ethanol treatment ( i . e . the peak ethanol response [45] ) . Consistent with our previous microarray data [45 , 64] , we saw lower induction of CTT1 by ethanol in S288c relative to YPS163 ( Fig 9A ) . Moreover , we saw dramatically reduced induction of CTT1 in a YPS163 hap1Δ mutant compared to the wild-type YPS163 control ( Fig 9A ) . Further support that HAP1 is causative for reduced CTT1 expression was provided by performing qPCR in the HAP1 reciprocal hemizygotes , where we found that the HAP1S288c allele resulted in significantly reduced CTT1 induction compared to the HAP1YPS163 allele ( Fig 9A ) . To determine whether the differences in CTT1 induction across strain backgrounds also manifested as differences in each strain’s ability to detoxify H2O2 , we measured in vitro peroxidase activity in cell-free extracts . We compared in vitro peroxidase activity in extracts from unstressed cells and cells exposed to ethanol stress for 60 minutes ( i . e . the same pre-treatment time that induces acquired H2O2 resistance ( see Materials and Methods ) ) . For wild-type YPS163 , ethanol strongly induced peroxidase activity , and this induction was completely dependent upon CTT1 ( Fig 9B ) . Mirroring CTT1 gene expression patterns , the induction of peroxidase activity was reduced in a YPS163 hap1Δ mutant . Additionally , reciprocal hemizygosity analysis provided further support that lack of HAP1 function results in decreased peroxidase activity , as the hybrid containing the HAP1S288c allele showed significantly reduced peroxidase activity following ethanol stress compared to the hybrid containing the HAP1YPS163 allele ( Fig 9B ) . Notably , the hybrid containing the HAP1YPS163 allele had lower CTT1 induction and in vitro peroxidase activity following ethanol shock than wild-type YPS163 , despite equivalent levels of acquired H2O2 resistance in the strains . These results suggest that HAP1 may play additional roles in acquired H2O2 resistance beyond H2O2 detoxification , depending upon the genetic background ( see Discussion ) . Interestingly , S288c showed no induction of peroxidase activity upon ethanol treatment , despite modest induction of the CTT1 transcript . This result is reminiscent of Ctt1p regulation during heat shock in the S288c background , where mRNA levels increase without a concomitant increase in protein levels [84] . Thus , in addition to strain-specific differences in CTT1 regulation at the RNA level , there are likely differences in regulation at the level of translation and/or protein stability .
In this study , we leveraged extensive natural variation in the yeast ethanol response to understand potential connections between gene expression variation and higher-order organismal traits . Previous screens of gene deletion libraries have found surprisingly little overlap between the genes necessary for surviving stress and genes that are induced by stress . [34–43] . Instead , gene induction may be a better predictor of a gene’s requirement for acquired stress resistance [84] . Thus , we hypothesized that phenotypic variation in acquired stress resistance may be linked to natural variation in stress-activated gene expression . Our results provide a compelling case study in support of this notion—namely that a polymorphism in the HAP1 transcription factor affects natural variation in acquired H2O2 resistance , but not the basal H2O2 resistance of unstressed cells . Forward genetic screens have shown that the genes necessary for basal and acquired resistance are largely non-overlapping [34 , 36 , 84] , suggesting that mechanisms underlying basal and acquired stress resistance are distinct . We provide further genetic evidence to support this model . YPS163 hap1Δ mutants and the hybrid carrying the HAP1S288c allele had strong acquired H2O2 defects , but no differences in their basal H2O2 resistance ( Figs 4 and 6 ) . Moreover , the YPS163 hap1Δ mutant was affected only when ethanol was the mild pretreatment , and was able to fully acquire H2O2 resistance following mild H2O2 or mild NaCl ( Fig 5 ) . These results suggest that the mechanisms underlying acquired resistance differ depending upon the mild stress that provokes the response . Further dissection of the mechanisms underlying acquired stress resistance will provide a more integrated view of eukaryotic stress biology . Our results reveal a new role for Hap1p in cross protection against H2O2 that has been lost in the S288c lab strain . We propose that a major mechanism underlying ethanol-induced cross protection against H2O2 is the induction of cytosolic catalase T ( Ctt1p ) , and that in the YPS163 background , Hap1p is necessary for proper induction of CTT1 during ethanol stress . We based this mechanism on the following observations . First , over-expression of CTT1 in S288c is sufficient to induce high H2O2 resistance [85] . Second , a YPS163 ctt1Δ mutant cannot acquire any further H2O2 resistance following ethanol pre-treatment ( Fig 8 ) , suggesting that no other antioxidant defenses are able to compensate under this condition . Lastly , the defect in cross protection for the YPS163 hap1Δ mutant correlates with reduced CTT1 expression and peroxidase activity during ethanol stress ( compare Figs 6 and 9 ) . How Hap1p is involved in the regulation of CTT1 during ethanol stress remains an open question , but we offer some possibilities . Hap1p is activated by heme , thus promoting transcription of genes involved in respiration , ergosterol biosynthesis , and oxidative stress defense including CTT1 [75 , 76 , 78 , 82] . Because heme biosynthesis requires oxygen , Hap1p is an indirect oxygen sensor and regulator of aerobically expressed genes [74 , 75 , 86] . There is currently no evidence that heme levels are affected by ethanol stress , nor is there evidence that Hap1p is “super-activating” under certain conditions . Thus , we disfavor a mechanism of induction caused solely by Hap1p activation . Instead , we favor a mechanism where Hap1p interacts with other transcription factors at the CTT1 promoter during ethanol stress , leading to full CTT1 induction . One possibility that we favor is recruitment of the general stress transcription factor Msn2p , which plays a known role in acquired stress resistance [44 , 45] . We previously showed that a YPS163 msn2Δ mutant had no induction of CTT1 mRNA during ethanol stress [45] , suggesting that Msn2p was an essential activator for CTT1 under this condition . The CTT1 promoter region contains three Msn2p DNA-binding sites , two of which are ~100-bp away from the Hap1p binding site . Hap1p binding to the CTT1 promoter could help recruit Msn2p during ethanol stress , possibly through chromatin remodeling that increases accessibility of the Msn2p binding sites as proposed by Elfving and colleagues [87] . What is the physiological role of Hap1p-dependent induction of CTT1 during ethanol stress ? One possibility is that regulation tied to the heme- and oxygen-sensing role of Hap1p ensures that CTT1 induction only occurs under environmental conditions where reactive oxygen species ( ROS ) are most likely to be encountered—namely stressful conditions that are also aerobic . In the context of ethanol stress , aerobic fermentation would lead to subsequent respiration of the produced ethanol and simultaneous ROS production . Under these conditions , CTT1 induction leading to ethanol-mediated cross protection against ROS would likely confer a fitness advantage . On the other hand , during stressful yet anoxic conditions , Ctt1p and other ROS-scavenging proteins are likely unnecessary . Furthermore , because heme is not synthesized during anoxic conditions [74] , Hap1p would fail to induce CTT1 and other genes encoding non-essential heme-containing proteins . This may improve fitness by conserving energy used for biosynthesis and by redirecting limited heme to more essential heme-containing proteins . The S288c lab strain has long been known to possess a defective HAP1 allele [79] . Apparently , the defective allele arose relatively recently , as only S288c contains a HAP1 Ty1 insertion out of over 100 sequenced strains [88 , 89] . The lack of HAP1 function in S288c could be due to relaxation of selective constraint , though others have argued in favor of positive selection for reduced ergosterol biosynthetic gene expression [90 , 91] . Regardless , the loss of ethanol-induced acquired H2O2 resistance is likely a secondary effect of the loss of Hap1p function . Intriguingly , we did find that two ( non-S288c ) domesticated yeast strains also lack ethanol-induced cross protection against H2O2 ( S5 Fig ) , suggesting that phenotypic differences in acquired stress resistance may differentiate domesticated versus wild yeast . Because environmental stresses are likely encountered in combination or sequentially [92] , acquired stress resistance is likely an important phenotype in certain natural ecological settings . Future studies directed at understanding differences in acquired stress resistance phenotypes in diverse wild yeast strains may provide unique insights into the ecology of yeast . While our QTL mapping identified HAP1 as the major effector of cross protection , we note that additional complexity remains unexplained . Notably , despite the strong cross protection defect in the YPS163 hap1Δ mutant , some residual cross protection persists that is absent in S288c ( Fig 6 ) . Intriguingly , the residual cross protection is also absent in the hybrid carrying the HAP1S288c allele , suggesting the involvement of other genes depending upon the genetic background ( Fig 4B and 4C ) . It is known that yeast strains with respiratory defects have increased ROS sensitivity [93 , 94] , potentially due to increased programmed cell death [95] . It is possible that reduced respiratory activity and concomitant ROS sensitivity in strains lacking HAP1 is exacerbated by genetic interactions with other alleles . The lack of cross protection in S288c and the HAP1S288c hybrid correlates with the lack of inducible peroxidase activity following ethanol pretreatment in those strains . The lack of inducible peroxidase activity in S288c despite modest induction of CTT1 mRNA could be due to translational regulation , which is supported by the observation that while mild heat shock induces CTT1 mRNA , protein levels remain nearly undetectable [84] . Strikingly , the hybrid carrying the HAP1YPS163 allele still cross protects despite levels of CTT1 mRNA induction and peroxidase activity that are lower than in the YPS163 hap1Δ strain that is unable to acquire further resistance ( Fig 9 ) . These data suggest that HAP1 plays an additional role in ethanol-induced cross protection beyond H2O2 detoxification by Ctt1p . Moreover , the continuous distribution of the cross protection phenotype in the segregants ( S1 Fig ) and the results of allele swap experiments ( Fig 6 ) strongly implicate other genes and processes in this complex trait . Specifically , the lack of complementation by the HAP1YPS163 allele in the S288c background suggests that additional loci in S288c render HAP1 necessary but not sufficient for cross protection in this background . Moreover , our genotyping of the segregants at HAP1 revealed a small number that still possessed cross protection in the absence of functional HAP1 ( S3 Fig and S1 Table ) , suggesting that HAP1 is dispensable in certain genetic backgrounds . We examined the effects of hap1Δ mutations in other wild strain backgrounds and found two additional strains with a strong HAP1 requirement and a third strain with at most a mild HAP1 effect ( S4 Fig ) . This result , as well as those from other recent studies [96–98] , suggests that these types of genetic background effects are likely the rule rather than the exception . Future high resolution mapping experiments will be necessary to identify and characterize the source of these genetic background effects . Gene expression variation is extensive in nature and is hypothesized to be a major driver of higher-order phenotypic variation . However , there are inherent challenges to connecting gene expression variation to higher-order organismal traits . Hundreds to thousands of genes are often differentially expressed across individuals , so identifying which particular transcripts exert effects on fitness is difficult . By studying acquired stress resistance—a phenotype better correlated with stress-activated gene expression changes—we were able to uncover a novel connection between gene expression variation and an organismal trait .
Strains and primers used in this study are listed in S2 and S3 Tables , respectively . The parental strains for QTL mapping were YPS163 ( oak strain ) and the S288c-derived DBY8268 ( lab strain; referred to throughout the text as S288c ) . The construction of the S288c x YPS163 QTL mapping strain panel ( 44 F2 progeny ) is described in [99] ( kindly provided by Justin Fay ) . Genotypes for the strain panel are listed in S4 Table . During the course of analyzing HAP1 genotypes , we found one segregant ( YS . 15 . 2 ) to be a mixed population , so it was removed from subsequent analyses . Deletions in the BY4741 ( S288c ) background were obtained from Open Biosystems ( now GE Dharmacon ) , with the exception of hap1 ( whose construction is described in [45] ) . Deletions were moved into haploid MATa derivatives of DBY8268 , M22 , and YPS163 by homologous recombination with the deletion::KanMX cassette amplified from the appropriate yeast knockout strain [100] . Homozygous hap1Δ strains of YPS1000 and Y10 were generated by moving the hap1Δ::KanMX allele from the BY4741 background into the strains , followed by sporulation and tetrad dissection . All deletions were verified by diagnostic PCR . DBY8268 containing a wild-type HAP1 allele from YPS163 was constructed in two steps . First , the MX cassette from the hap1Δ::KanMX deletion was replaced with a URA3MX cassette , selecting for uracil prototrophy . Then , URA3 was replaced with wild-type HAP1 from YPS163 ( amplified using primers 498-bp upstream and 1572-bp downstream of the HAP1 ORF ) , while selecting for loss of URA3 on 5-fluoroorotic acid ( 5-FOA ) plates . Deletions and repair of HAP1 were confirmed by diagnostic PCR ( see S3 Table for primer sequences ) . YPS163 containing a HAP1S288c allele was constructed by first inserting a KanMX cassette into S288c 117-bp downstream of the Ty element to create JL1032 . We then amplified and transformed the Ty element into YPS163 using primers that annealed 103-bp upstream of the Ty element and 177-bp downstream of the KanMX cassette , generating JL1069 . Diploid strains for HAP1 and TOP3 reciprocal hemizygosity analysis were generated as follows . The hemizygote containing the wild-type S228c HAP1 allele ( JL580 ) was generated by mating JL140 ( YPS163 MATa hoΔ::HygMX hap1Δ::KanMX ) to JL506 ( DBY8268 MATα ho ura3 hap1 ) . The hemizygote containing the wild-type YPS163 allele ( JL581 ) was generated by mating JL112 ( YPS163 MATα hoΔ::HygMX HAP1 ) to JL533 ( DBY8268 MATa ho ura3 hap1Δ::KanMX ) . The hemizygote containing the wild-type S288c TOP3 allele ( JL1107 ) was created by mating JL1066 ( YPS163 MATa hoΔ::HygMX top3Δ::KanMX ) to BY4742 ( MATα TOP3 ) . The hemizygote containing the wild-type YPS163 allele ( JL1106 ) was created by mating JL1121 ( BY4741 MATa top3Δ::KanMX ) to JL112 ( YPS163 MATα hoΔ::HygMX TOP3 ) . All strains were grown in batch culture in YPD ( 1% yeast extract , 2% peptone , 2% dextrose ) at 30°C with orbital shaking ( 270 rpm ) . To identify possible promoter polymorphisms , the HAP1 promoters of the DBY8268 ( JL505 ) , YPS163 ( JL111 ) , and S288c HAP1YPS163 ( JL975 ) strains were amplified using primers that anneal 1091-bp upstream and 134-bp downstream of the HAP1 start codon . PCR products were purified with a PureLink PCR cleanup kit ( Invitrogen ) and sequenced by Sanger Sequencing ( Eurofins Genomics ) using a primer that anneals 498-bp upstream of the HAP1 start codon . Sequences were aligned to the S288c and YPS163 reference sequences using SnapGene v4 . 1 ( GSL Biotech ) . This verified the presence of a 1-bp indel within a poly-A stretch that differs between S288c and YPS163 . The S288c HAP1YPS163 ( JL975 ) strain contains the YPS163 HAP1 promoter sequence . Additionally , the YPS163 strain containing the HAP1S288c was constructed to only contain the Ty element and not the S288c promoter polymorphism . The HAP1 allele of each segregant for the QTL mapping panel was genotyped by differential PCR analysis where the same forward primer ( HAP1 int 3’ F ) was paired with two different reverse primers . One primer ( Ty R ) anneals specifically to the Ty element , yielding an 856-bp product when amplifying the S288c allele . The second primer ( HAP1 3’ end R ) anneals 3’ to the Ty element of HAP1S288c , yielding a 570-bp product for HAP1YPS163 and a 6 . 5-kb product for HAP1S288c . Each segregant was genotyped using both sets of primer pairs , and only one segregant ( YS . 15 . 2 ) appeared to contain both HAP1 alleles . Subsequent analysis of multiple colonies verified that YS . 15 . 2 was a mixed population , and thus it was removed it from all subsequent analyses . The TOP3 alleles of S288c and YPS163 contain two non-synonymous SNPs at nucleotide positions 1 , 398 and 1 , 422 . Segregant genotypes at TOP3 were determined by analyzing restriction fragment length polymorphisms . TOP3 was amplified using primers ( TOP3 up F and TOP3 down R ) that anneal ~500-bp upstream and downstream of the open reading frame , generating a 2 . 9-kb product . PCR products were digested with either 1 ) PstI , which cuts at position 1 , 248 only within the TOP3YPS163 ORF allele yielding 1 . 7- and 1 . 2-kb products , or ( 2 ) KflI , which cuts at position 1 , 155 only within the TOP3S288c yielding 1 . 6- and 1 . 3-kb products . Genotypes for HAP1 and TOP3 are listed in S1 Table . Cross-protection assays were performed as described in [44] with slight modifications . Briefly , 3–4 freshly streaked isolated colonies ( <1 week old ) were grown overnight to saturation , sub-cultured into 6 ml fresh media , and then grown for at least 8 generations ( >12 h ) to mid-exponential phase ( OD600 of 0 . 3–0 . 6 ) to reset any cellular memory of acquired stress resistance [85] . Each culture was split into two cultures and pretreated with YPD media containing either a single mild “primary” dose or the same concentration of water as a mock-pretreatment control . Primary doses consisted of 5% v/v ethanol , 0 . 4 M NaCl , or 0 . 4 mM H2O2 . Thereafter , mock and primary-treated cells were handled identically . Following 1-hour pretreatment at 30°C with orbital shaking ( 270 rpm ) , cells were collected by mild centrifugation at 1 , 500 x g for 3 min . Pelleted cells were resuspended in fresh medium to an OD600 of 0 . 6 , then diluted 3-fold into a microtiter plate containing a panel of severe “secondary” H2O2 doses ranging from 0 . 5–5 . 5 mM ( 0 . 5 mM increments; 150 μl total volume ) . Microtiter plates were sealed with air-permeable Rayon films ( VWR ) , and cells were exposed to secondary stress for 2 hours at 30°C with 800 rpm shaking in a VWR symphony Incubating Microplate Shaker . Four μl of a 50-fold dilution was spotted onto YPD agar plates and grown 48 h at 30°C . Viability at each dose was scored using a 4-point semi-quantitative scale to score survival compared to a no-secondary stress ( YPD only ) control: 100% = 3 pts , 50–90% = 2 pts , 10–50% = 1 pt , or 0% ( 3 or less colonies ) = 0 pts . An overall H2O2 tolerance score was calculated as the sum of scores over the 11 doses of secondary stress . Raw phenotypes for all acquired stress resistance assays can be found in S1 Table . A fully detailed acquired stress protocol has been deposited to protocols . io under doi dx . doi . org/10 . 17504/protocols . io . g7sbzne . Statistical analyses were performed using Prism 7 ( GraphPad Software ) . Phenotyping of the QTL mapping strain panel for basal and acquired H2O2 resistance was performed in biological duplicate . Because cross-protection assays on the entire strain panel could not all be performed at the same time , we sought to minimize day-to-day variability . We found that minor differences in temperature and shaking speed affected H2O2 resistance; as a result , we used a digital thermometer and tachometer to ensure standardization across experiments . Moreover , we found that differences in handling time were a critical determinant of experimental variability . To minimize this source of variability , all cell dilutions were performed quickly using multichannel pipettes , and no more than two microtiter plates were assayed during a single experiment . To ensure that replicates on a given day were reproducible , we always included the YPS163 wild-type parent as a reference . Single mapping scans were performed using Haley-Knott regression [101] implemented through the R/QTL software package [102] . Genotype probabilities were estimated at every cM across the genome using the calc . genoprob function . Significant LOD scores were determined by 100 , 000 permutations that randomly shuffled phenotype data ( i . e . strain labels ) relative to the genotype data . The maximum LOD scores for the permuted scans were sorted , and the 99th percentile was used to set the genome-wide FDR at 1% . This resulted in LOD cutoffs of 3 . 07 for QTL mapping of basal H2O2 resistance , and 4 . 24 for acquired H2O2 resistance . Broad-sense heritability ( H2 ) was estimated from the segregant data as described in [71] using a random-effects ANOVA model implemented through the lmer function in the lme4 R package [103] . H2 was estimated using the equation σG2 ( σG2+σE2 ) , where σG2 represents the genetic variance due to the effects of segregrant , and σE2 represents the residual ( error or environmental ) variance . The proportion of variance explained by a QTL was estimated using the equation 1−10 ( −2n*LOD ) , where n represents the number of segregants . Induction of CTT1 by ethanol was assessed by real-time quantitative PCR ( qPCR ) using the Maxima SYBR q-PCR Master Mix ( Thermo Fisher Scientific ) and a Bio-Rad CFX96 Touch Real-Time PCR Detection System , according to the manufacturers’ instructions . Cells were grown to mid-exponential phase ( OD600 of 0 . 3–0 . 6 ) as described for the cross-protection assays . Cells were collected by centrifugation at 1 , 500 x g for 3 minutes immediately prior to the addition of 5% v/v ethanol ( unstressed sample ) and 30 minutes post-ethanol treatment , which encompasses the peak of global expression changes to acute ethanol stress [45] . Cell pellets were flash frozen in liquid nitrogen and stored at -80°C until processed . Total RNA was recovered by hot phenol extraction as previously described [104] , and then purified with a Quick-RNA MiniPrep Plus Kit ( Zymo Research ) including on-column DNase I treatment . cDNA synthesis was performed as described [104] , using 10 μg total RNA , 3 μg anchored oligo-dT ( T20VN ) , and SuperScript III ( Thermo Fisher Scientific ) . One ng cDNA was used as template for qPCR with the following parameters: initial denaturation at 95°C for 3 minutes followed by 40 cycles of 95°C for 15 seconds and 55°C annealing and elongation for 1 minute . Cq was determined using regression analysis , with baseline subtraction via curve fit . The presence of a single amplicon for each reaction was validated by melt curve analysis . The average of two technical replicates were used to determine relative CTT1 mRNA abundance via the ΔΔCq method [105] , by normalizing to an internal control gene ( ERV25 ) whose expression is unaffected by ethanol stress and does not vary in expression between S288c and YPS163 [45] . Primers for CTT1 and ERV25 were designed to span ~200 bp in the 3’ region of each ORF ( to decrease the likelihood of artifacts due to premature termination during cDNA synthesis ) , and for gene regions free of polymorphisms between S288c and YPS163 ( see S3 Table for primer sequences ) . Three biological replicates were performed and statistical significance was assessed via a paired t-test using Prism 7 ( GraphPad Software ) . For peroxidase activity assays , mid-exponential phase cells were collected immediately prior to and 60 minutes post-ethanol treatment , to assess peroxidase activity levels during the induction of cross protection . Cells were collected by centrifugation at 1 , 500 x g for 3 minutes , washed twice in 50 mM potassium phosphate buffer , pH 7 . 0 ( KPi ) , flash frozen in liquid nitrogen , and then stored at -80°C until processed . For preparation of whole cell extracts , cells were thawed on ice , resuspended in 1 ml KPi buffer , and then transferred to 2-ml screw-cap tubes for bead beating . An equal volume ( 1 ml ) of acid-washed glass beads ( 425–600 micron , Sigma-Aldrich ) was added to each tube . Cells were lysed by four 30-second cycles of bead beating in a BioSpec Mini-Beadbeater-24 ( 3 , 500 oscillations/minute , 2 minutes on ice between cycles ) . Cellular debris was removed by centrifugation at 21 , 000 x g for 30 minutes at 4°C . The protein concentration of each lysate was measured by Bradford assay ( Bio-Rad ) using bovine serum albumin ( BSA ) as a standard [106] . Peroxidase activity in cellular lysates was monitored as described [107] , with slight modifications . Briefly , 50 μg of cell free extract was added to 1 ml of 15 mM H2O2 in KPi buffer . H2O2 decomposition was monitored continuously for 10 minutes in Quartz cuvettes ( Starna Cells , Inc . ) at 240 nm ( ε240 = 43 . 6 M-1 cm-1 ) using a SpectraMax Plus Spectrophotometer ( Molecular Devices ) . One unit of catalase activity catalyzed the decomposition of 1 μmol of H2O2 per minute . For each sample , results represent the average of technical duplicates . To assess statistical significance , four biological replicates were performed and significance was assessed via a paired t-test using Prism 7 ( GraphPad Software ) .
|
A major goal in genetics is to understand how individuals with different genetic makeups respond to their environment . Understanding these “gene-environment interactions” is important for the development of personalized medicine . For example , gene-environment interactions can explain why some people are more sensitive to certain drugs or are more likely to get certain cancers . While the underlying causes of gene-environment interactions are unclear , one possibility is that differences in gene expression across individuals are responsible . In this study , we examined that possibility using baker’s yeast as a model . We were interested in a phenomenon called acquired stress resistance , where cells exposed to a mild dose of one stress can become resistant to an otherwise lethal dose of severe stress . This response is observed in diverse organisms ranging from bacteria to humans , though the specific mechanisms governing acquisition of higher stress resistance are poorly understood . To understand the differences between yeast strains with and without the ability to acquire further stress resistance , we employed genetic mapping . We found that part of the variation in acquired stress resistance was due to sequence differences in a key regulatory protein , thus providing new insight into how different individuals respond to acute environmental change .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"peroxidases",
"chemical",
"compounds",
"quantitative",
"trait",
"loci",
"enzymes",
"gene",
"regulation",
"regulatory",
"proteins",
"enzymology",
"dna-binding",
"proteins",
"organic",
"compounds",
"fungi",
"transcription",
"factors",
"alcohols",
"proteins",
"gene",
"expression",
"chemistry",
"ethanol",
"genetic",
"loci",
"yeast",
"biochemistry",
"eukaryota",
"organic",
"chemistry",
"phenotypes",
"genetics",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"organisms"
] |
2018
|
Linkage mapping of yeast cross protection connects gene expression variation to a higher-order organismal trait
|
Lymphatic filariasis is caused by the parasitic worms Wuchereria bancrofti , Brugia malayi or B . timori , which are transmitted via the bites from infected mosquitoes . Once in the human body , the parasites develop into adult worms in the lymphatic vessels , causing severe damage and swelling of the affected tissues . According to the World Health Organization , over 1 . 2 billion people in 58 countries are at risk of contracting lymphatic filariasis . Very few drugs are available to treat patients infected with these parasites , and these have low efficacy against the adult stages of the worms , which can live for 7–15 years in the human body . The requirement for annual treatment increases the risk of drug-resistant worms emerging , making it imperative to develop new drugs against these devastating diseases . We have developed a yeast-based , high-throughput screening system whereby essential yeast genes are replaced with their filarial or human counterparts . These strains are labeled with different fluorescent proteins to allow the simultaneous monitoring of strains with parasite or human genes in competition , and hence the identification of compounds that inhibit the parasite target without affecting its human ortholog . We constructed yeast strains expressing eight different Brugia malayi drug targets ( as well as seven of their human counterparts ) , and performed medium-throughput drug screens for compounds that specifically inhibit the parasite enzymes . Using the Malaria Box collection ( 400 compounds ) , we identified nine filarial specific inhibitors and confirmed the antifilarial activity of five of these using in vitro assays against Brugia pahangi . We were able to functionally complement yeast deletions with eight different Brugia malayi enzymes that represent potential drug targets . We demonstrated that our yeast-based screening platform is efficient in identifying compounds that can discriminate between human and filarial enzymes . Hence , we are confident that we can extend our efforts to the construction of strains with further filarial targets ( in particular for those species that cannot be cultivated in the laboratory ) , and perform high-throughput drug screens to identify specific inhibitors of the parasite enzymes . By establishing synergistic collaborations with researchers working directly on different parasitic worms , we aim to aid antihelmintic drug development for both human and veterinary infections .
Lymphatic filariasis is a neglected tropical disease caused primarily by the parasitic nematodes Wuchereria bancrofti and Brugia malayi . The painful and disfiguring manifestations of this disease , also known as elephantiasis , can lead to permanent disability , causing an annual loss of approximately 5 . 5 million disability adjusted life years , affecting the poorest populations in Africa , Asia , and Latin America [1] . Current antifilarial therapies aim to eliminate filariasis through mass drug administration . However , in standard doses , the drugs used for this purpose ( diethylcarbamazine , ivermectin and albendazole ) are not effective against adult nematodes . As the adult worms can live in the human body for ca . 15 years [2] , patients need to undergo multiple rounds of treatment , increasing not only the cost of therapy , but also the risk of drug-resistant worms emerging [3–5] . Filarial worms are difficult to cultivate in vitro , so adult worms for laboratory studies have to be obtained from animal models . Marcellino et al . [6] successfully developed a whole-plate , motion-based screen for monitoring drug activity against macroscopic parasites ( WormAssay ) . This method was subsequently employed in screens against B . malayi [2] , leading to the identification of the antifilarial activity of the FDA-approved drug auranofin . Unfortunately , there is no small animal model for other filarial worms , such as W . bancrofti [7] or Onchocerca volvulus; hence , there is a requirement for novel assays in the search for better treatments targeting filariasis cell-based assays also require extensive optimization ) . An alternative to parasite-based assays is to use in vitro drug screens based on protein targets . However , in vitro target-based assays require careful ( and costly ) optimization of the screening platform for each individual target protein to be tested , and provide no information on whether the drug is likely to be taken up by cells or whether it has general cytotoxicity . To address these problems , we have developed and successfully validated a novel approach to high-throughput screens ( HTS ) for antiparasitic compounds using yeast [8 , 9] . Yeast cultures , which can be grown rapidly and at low cost , are ideal for use in automated screens [8–11] . Yeast cells are suitable hosts for the expression of nematode proteins [12–18] , including enzymes essential for different life-cycle stages of the parasites , many of which cannot be propagated in vitro [17] . We engineered Saccharomyces cerevisiae strains to express either different parasite drug targets [9] , or their equivalent human proteins , such that the growth of the yeast is dependent on the functioning of these heterologous proteins . We then transformed the engineered strains with plasmids expressing either CFP ( cyan fluorescent protein ) , Venus ( yellow fluorescent protein ) , Sapphire ( blue fluorescent protein ) or mCherry ( red fluorescent protein ) , to enable their in vivo labeling . Our engineered yeast strains are genetically identical , apart from expressing different heterologous drug targets and fluorescent labels that allow the growth of multiple strains to be followed in a single culture . These mixed cultures can be treated with chemical libraries to identify compounds capable of specifically inhibiting strains with the parasite targets but not their human counterparts . By these means , the drug sensitivity observed in a particular strain can be directly linked to the in vivo inhibition of the heterologous target protein . This approach has a number of significant advantages over conventional screens: it is very easy to set up for different drug targets; it is cheap , as the volumes used are very small and the yeast growth medium is inexpensive; it discriminates between compounds affecting parasite enzymes and human enzymes , and , by definition , active compounds must be able to enter living cells . In this work , we evaluated the potential of such yeast-based drug screens in the identification of novel antifilarial compounds . We constructed yeast strains expressing different B . malayi target proteins , and used them to screen for novel inhibitors of filarial enzymes . We utilized a publicly available small-chemical library ( 400 Malarial Box compounds; http://www . mmv . org/malariabox ) and identified compounds with significant inhibitory activity against the B . malayi enzymes , but little or no detectable activity against the equivalent human enzymes expressed in yeast . These first hit compounds were then validated in vitro against the closely related species , Brugia pahangi ( continuously cultivated in our laboratory ) with encouraging results , providing a proof of principle for this approach .
All animal protocols were carried out in accordance with the guidelines of the UK Home Office , under the Animal ( Scientific Procedures ) Act 1986 , following approval by the University of Glasgow Ethical Review Panel . Experiments were performed under the authority of the UK Home Office , project numbers 60/4448 and 60/3792 . The filarial enzymes selected for testing are listed in Table 1 . The coding regions of Bm1_22900 ( BmNMT ) , Bm1_01925 ( BmPGK ) , Bm1_29130 ( BmTPI ) , Bm1_ 49000 ( BmPIS ) , Bm1_48165 ( BmSAH ) , Bm1_38705 ( BmSEC53 ) , Bm1_11585 ( BmADE13 ) and Bm1_23075 ( BmCDC8 ) were PCR-amplified from an adult Brugia malayi cDNA library , kindly donated by the Filariasis Research Reagent Resource Center ( University of Georgia ) . These were cloned into the BamHI-PstI sites of pCM188 [19] to produce pCMBmNMT , pCMBmPGK , pCMBmTPI , pCMBmPIS , pCMBmSAH , pCMBmSEC53 , pCMBmADE13 and pCMBmCDC8 . These constructs placed the heterologous genes under the control of the TetO2 promoter . Synthetic genes encoding Brugia malayi Bm1_33465 ( BmCDC21 ) , Bm1_16500 ( BmKRS ) , Bm1_42945 ( BmMVD ) , Bm1_57600 ( BmRKI ) and Bm1_16300 ( BmDYS ) were synthesized by Geneart and sub-cloned into the BamHI-PstI sites of pCM188 [19] to produce pCMBmCDC21 , pCMBmKRS , pCMBmMVD , pCMBmRK1 , and pCMBmDYS . The coding regions of human TPI1 ( HsTPI ) , CDIPT ( HsPIS ) , AHCYL1 ( HsSAHa ) , AHCY/SAHH ( HsSAHb ) , PMM2 ( HsSEC53 ) , PUR8 ( HsADE13 ) , CDC8/DTYMK ( HsCDC8 ) and PPA1 ( HsIPP1a ) were PCR amplified from a cerebellum cDNA library , kindly donated by Dr . Nianshu Zhang ( University of Cambridge ) . These were cloned into the BamHI-PstI sites of pCM188 [19] to produce pCMHsTPI , pCMHsPIS , pCMHsSAHa , pCMHsSAHb , pCMHsSEC53 , pCMHsADE13 , pCMHsCDC8 and pCMHsIPP1a . Synthetic genes encoding Homo sapiens TYMS ( HsCDC21 ) , LysRS ( HsKRS ) , MVD ( HsMVD ) , and RPIA ( HsRKI ) were synthesized by Geneart and sub-cloned into the BamHI-PstI sites of pCM188 [19] to produce pCMHsCDC21 , pCMHsKRS , pCMHsMVD , and pCMHsRK1 . Plasmid maps for new constructs developed in this work are available in Figs A to AF in S1 Text . Plasmids encoding human NMT2 and PGK1 ( pCMHsNMT and pCMHsPGK ) are described in Bilsland et al [9] . PDR5 encodes the major drug export pump of Saccharomyces cerevisiae , hence we deleted PDR5 in all of our yeast strains to increase their susceptibility to the test compounds . Deletion of the yeast PDR5 coding sequence was performed as described previously [20] . pCMBmNMT constructs were transformed into nmt1Δ::KanMX/NMT1 pdr5Δ::HisMX/PDR5 strains ( BY4743 background [21] ) . pCMBmPGK constructs were transformed into pgk1Δ::KanMX/PGK1 pdr5Δ::HisMX/PDR5 strains ( BY4743 background ) . The same approach ( transformation of the heterologous construct into a yeast strain heterozygous for a deletion mutant of the orthologous yeast gene and heterozygous for pdr5 ) was employed for all subsequent constructs . Heterozygous strains harbouring the heterologous constructs were sporulated and derived haploids were selected for growth assays and drug screens . Strain genotypes are described in Table A in S1 Text . Saccharomyces cerevisiae , Brugia malayi and Homo sapiens orthologues were selected based on data available at: http://inparanoid . sbc . su . se/cgi-bin/index . cgi . Multiple protein sequence alignments were performed using https://www . ebi . ac . uk/Tools/msa/ , creating a pairwise identity matrix between each protein orthologue . Yeast strains were grown in 2 . 5 mL YPD ( 1% yeast extract , 2% peptone , 2% glucose ) cultures overnight at 30°C . Cultures were diluted 100 times in fresh YPD with 0 , 5 or 10 mg/L doxycycline to tune-down the expression of the target protein from the TetO2 promoter . Clear 384-well plates ( Corning ) were then filled with 70 μL of each diluted culture ( in triplicate ) . Plates were incubated in a BMG Optima plate reader and OD595 measures for each culture were acquired every 15 minutes . Growth scores were obtained by calculating the maximum exponential growth rate for each culture , and then multiplying this value by the yield of the culture ( yield = maximum OD595—minimum OD595 ) . Growth scores of the yeast strains dependent for growth on either the B . malayi or human coding sequence for the target enzyme were divided by the score for the corresponding wild-type strains ( BY4741 or BY4742 ) , to estimate their relative growth . Yeast strains were transformed with fluorescent plasmids expressing one of the following fluorescent proteins mCherry ( yEp_Cherry_LEU2 ) , Venus ( yEp_venus_LEU2 ) , Sapphire ( yEp_sapphire_LEU2 ) or CFP ( yEp_CFP_LEU2 ) [8] . This allowed us to monitor , in real time , the growth of 2–4 different yeast strains growing in competition . Fluorescently labelled strains were grown in 2 . 5 mL YNB ( 1 , 7 g/l yeast nitrogen base , 5 g/L ammonium sulphate , 2% glucose , and amino acid supplements ) overnight at 30°C . Cultures were pooled and diluted 50 times in fresh YNB . An aliquot ( 35 μL ) of the diluted pooled-culture was added to each well of black 384-well plates , together with 35 μL of YNB with 20 μM of each test compound ( final 70 μL of 1:100 cultures with 10 μM of the test compound ) . Each experiment was run in quadruplicate , with two replicates using one combination of fluorescent markers , and two replicates using the alternate combination , in order to minimize false-positive results . Plates were incubated in a BMG Optima plate reader and fluorescence measured every 25 minutes , for 48 hours , in 4 different channels: Venus ( excitation 500 nm/emission 540 nm ) , CFP ( 440 nm/490 nm ) , Sapphire ( 405 nm/510 nm ) and mCherry ( 580 nm/612 nm ) . We confirmed each of our hits by generating dose:response curves with 0 , 2 , 10 and 50 μM of each hit compound . Time-series fluorescence data from each channel was analyzed using a fitting procedure written in R , based on the model-free spline method of Kahm et al . [22] . The first 26 fluorescence measurements were used for analysis ( corresponding to 26 . 66 hours of incubation ) ; beyond this time , fluorescence decayed due to photo-bleaching . Specifically , a smoothed spline was fitted to the raw data for each channel and the first derivative through each point of the fit calculated . The highest derivative was taken as the maximum exponential growth rate ( μ ) . This gradient was extrapolated back towards the start of the growth curve by fitting a straight line of the form y = mx + c . The lag phase duration ( λ ) was estimated by finding the intercept of this line and the baseline fluorescence from the start of the experiment . The fluorescence yield was calculated by subtracting the baseline fluorescence from the maximum value of the fit . These parameters are summarized in Fig AG in S1 Text . Each compound was scored according to: Score = rate × yieldlag The average score for each compound was normalized to the average score for the DMSO control from the corresponding row in the plate . Compounds were scored for each pool ( marker swap ) independently . A compound was considered a hit when the relative growth score , in both marker combinations , of the recombinant strains expressing the parasite target protein was significantly lower than the growth score of the strain expressing the human enzyme . Female worms of B . pahangi were recovered from infected jirds ( the rodent , Meriones unguiculatus ) approximately four months post-infection , exactly as described previously [23] . Worms were washed once in Hank’s Balanced Salt Solution and then cultured individually in 24-well tissue culture plates containing 2 . 0 ml of RMPI-1640 ( Invitrogen , Cat No: 52400 ) supplemented with 5% heat-inactivated fetal bovine serum , 1% glucose and 100 units per mL of penicillin/streptomycin ( all Invitrogen ) for 24 hours to monitor viability and microfilarial output . Healthy worms were then selected for drug testing . Initially , each compound was tested in triplicate at concentrations of 50 μM and 10 μM . Control wells contained the appropriate concentration of the DMSO solvent . Parasites were also exposed to geldanamycin , a known Hsp90 inhibitor , at the same concentrations , as a positive control . All cultures were maintained at 37°C in 5% v/v CO2 in air and were monitored daily for viability . In a second experiment , adult worms were exposed to a wider concentration range of three selected drugs that had proved active in the first experiment . These were MMV396794 , MMV665941 and MMV666022 . Each drug was tested in triplicate at 25 , 10 , 5 and 2 . 5 μM .
We initially selected 15 enzymes ( Table 1 ) to test as anti-filarial targets according to the following criteria: highly expressed in the adult stages of the parasite; likely to be essential for the viability of the nematode in the human host; and have an essential yeast ortholog . As drugs such as DEC and ivermectin have potent microfilaricidal activity , we chose to focus on adult parasites to control the stages not affected by current therapies . Furthermore , adult worms are the cause of much of the pathology associated with lymphatic filariasis [24] . We obtained Wuchereria bancrofti and Brugia malayi cDNA libraries , prepared from mRNA extracted from adult stages of the nematodes , from the Filariasis Research Reagent Resource Center ( The University of Georgia ) . These cDNA libraries were used as templates for amplification by polymerase chain reaction ( PCR ) , and cloning into yeast plasmids . From the outset of our work , we noticed that the publicly available genome sequence of W . bancrofti contains multiple gaps , which frequently overlapped with our target genes . This made the design of primers for coding sequence ( cds ) amplification problematic . On the other hand , the genome sequence of B . malayi is very well annotated , facilitating the cloning procedure . Hence , we focused most of our cloning efforts on B . malayi targets . In spite of this , after multiple attempts and primer designs , we failed to amplify seven of the intended B . malayi targets ( Table 2 ) . This was probably due to either the absence of that particular cDNA in the libraries or to errors in the annotated sequence . The problem due to the low coverage of our cDNA library was overcome by synthesizing DNA encoding each of the selected Brugia targets to allow their expression in yeast . We identified two instances of errors in the published Brugia genome sequence by either sequencing our clones derived from cDNA ( BmSAH ) , or by phenotyping strains expressing a synthetic cds based on the predicted sequence of the BmMVD gene product . We observed that the synthetic human MVD could complement the yeast deletion very successfully ( 88% of wild-type growth ) , whereas the synthetic B . malayi MVD could not . Alignment of the publicly available B . malayi , Loa loa ( eye worm ) , human , and yeast sequences , demonstrated that the B . malayi sequence diverged from those of the other three species , strongly suggesting either a non-conserved insertion in the Brugia malayi protein ( which leads to the loss of function of the heterologous protein in yeast ) or a problem with genome assembly of the published B . malayi sequence ( Fig 1 ) . Similarly , sequencing of the cloned B . malayi adenosylhomocysteinase ( Bm1_48165 , BmSAH ) , demonstrated an insertion of 27 amino-acids ( aa ) between aa 88 and 89 of Bm1_48165 . Eight independent BmSAH plasmids were constructed and sequenced and the same insertion was always detected . As the same insertion is present in both yeast and human SAH ( Fig 2 ) , it suggests that there may be an error in the B . malayi genome sequence , or that B . malayi encodes more than one splice variant of the enzyme . Furthermore , the encoded protein from our clone is enzymatically functional in yeast , complementing the deletion of the orthologous yeast gene . We constructed seven additional strains where B . malayi gene products ( Bm1_22900/BmNMT , Bm1_01925/BmPGK , Bm1_29130/BmTPI , Bm1_49000/BmPIS , Bm1_33465/BmCDC21 , Bm1_57600/BmRKI , Bm1_16300/BmDYS ) were able to complement the essential functions of the yeast orthologous gene . The heterologous genes were cloned into the yeast expression vectors under the regulation of the TetO2 promoter [25] . This promoter is constitutively on; however , by adding doxycycline to the growth medium , it is possible to tune-down the expression of the construct . Decreasing the expression of the target enzyme facilitates the growth inhibition by the test compounds ( less target = more efficient inhibition ) . We have previously tested the inhibition of expression from the TetO2 promoter by addition of 1 , 2 , 5 , 10 , 20 , 50 or 100 mg/L of doxycycline [9] and found that 5 and 10 mg/L would be ideal for the current work . Hence , we performed growth assays with each of the heterologous strains in medium with 0 , 5 or 10 mg/L of antibiotic ( Fig 3 ) . However , we noticed that most of our strains require the full expression of the heterologous targets for optimum growth . Hence , drug screens employing the strains expressing B . malayi targets were performed in the absence of doxycycline . In addition to BmMVD , four other B . malayi clones constructed in our studies failed to complement the essential functions of their cognate yeast genes; these were: Bm1_38705 ( BmSEC53 ) , Bm1_16500 ( BmKRS1 ) , Bm1_11585 ( BmADE13 ) and Bm1_23075 ( BmCDC8 ) . With a view to establishing a sequence-similarity cut-off on which to base the selection of new targets , we calculated the percentage identity between yeast and filarial proteins that could successfully replace each other and those that did not . We found no clear correlation between sequence similarity and functional complementation; therefore , with the data collected so far , we cannot predict which filarial enzymes will be functional in yeast ( Fig 4 ) . The screening method developed in our laboratory allows the simultaneous screening of up to three parasite targets and of the human counterpart in a single assay [8 , 9] . Hence , to make best use of the screening efforts , we included yeast strains expressing Schistosoma mansoni PGK [9] , S . mansoni NMT [9] , Homo sapiens NMT [9] , H . sapiens PGK [9] , H . sapiens TPI , H . sapiens PIS and H . sapiens SAHa and SAHb ( S1 Text ) . These eight strains , as well as the five strains expressing the corresponding Brugia malayi targets ( Table 2 ) , were labeled by expression of fluorescent proteins and screened against each of the 400 compounds from the Malaria Box collection at a concentration of 10 μM . Comparing the growth of yeast strains expressing either parasite or human enzymes , we identified a number of compounds that specifically inhibited the growth of strains dependent on the parasite enzymes ( Table 3 ) . We then performed dose:response experiments with 0 , 2 , 10 or 50 μM of each hit compound to confirm our NMT , PGK and PIS hits , and confirmed the specificity of the compounds at the indicated concentrations ( Table 3 , last column ) . We performed drug-sensitivity assays of our nine Brugia malayi hits using adult Brugia pahangi worms . Two of the compounds were not soluble in the test conditions , but four of the soluble compounds killed the worms at concentrations of 10 or 50 μM . Furthermore , the three remaining compounds compromised the motility of B . pahangi ( Fig 5 ) . Of the four compounds that had pronounced effects on adult worms at 24h incubation , two had immediate effects on adult viability . For MMV396794 and MMV665941 , adult worms were dead within 3h of incubation at 50 μM concentration . These two compounds and MMV666022 were re-tested over a wider range of concentrations starting at 25 μM . The results were very similar , with worms incubated in MMV MMV396794 and MMV665941 at 25 μM dying within 3h of exposure . At 10 μM , all worms were dead in MMV665941 by 6h of exposure and at 5 . 0 and 2 . 5 μM were tightly coiled , although still motile at 6h , but dead by 24h . Compound MMV396794 killed all worms at 10 μM by 24h , with worms at lower concentrations being sluggish at 24h , although still alive . MMV666022 was the least active compound—although , after 48h , worms were dying at 25 and 10 μM , they were largely unaffected at 5 or 2 . 5 μM . Considering that the compounds used in our screens ( Malaria Box ) have low toxicity to human cells , this initial validation was extremely encouraging .
Bm1_22900 , Bm1_01925 , Bm1_29130 , Bm1_48165 , Bm1_16955 , Bm1_38705 , Bm1_33465 , Bm1_16500 , Bm1_11585 , Bm1_42945 , Bm1_57600 , Bm1_23075 , Bm1_16300 , Bm1_32340 , Bm1_49000 .
|
We have developed and validated a yeast-based high-throughput screening assay for the identification of specific inhibitors of filarial targets . We engineered yeast strains to functionally express parasite and human enzymes , labeling these with fluorescent proteins and growing them in competition in the presence of test compounds . These strains express different target proteins from Brugia malayi ( as well as their human orthologs ) and our results demonstrate that it is possible to identify compounds that can discriminate between filarial and human enzymes . Accordingly , we are confident that we can extend our assay to novel targets from Brugia malayi and other worms of medical and veterinary importance , and perform high-throughput screens to identify new drugs against different parasitic worms .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"invertebrates",
"medicine",
"and",
"health",
"sciences",
"cloning",
"animals",
"drug",
"screening",
"complementary",
"dna",
"cloning",
"age",
"groups",
"fungi",
"adults",
"model",
"organisms",
"brugia",
"forms",
"of",
"dna",
"pharmacology",
"dna",
"molecular",
"biology",
"techniques",
"dna",
"cloning",
"research",
"and",
"analysis",
"methods",
"saccharomyces",
"complementary",
"dna",
"molecular",
"biology",
"yeast",
"people",
"and",
"places",
"biochemistry",
"brugia",
"malayi",
"nucleic",
"acids",
"genetics",
"nematoda",
"biology",
"and",
"life",
"sciences",
"yeast",
"and",
"fungal",
"models",
"saccharomyces",
"cerevisiae",
"population",
"groupings",
"organisms"
] |
2016
|
Yeast-Based High-Throughput Screens to Identify Novel Compounds Active against Brugia malayi
|
Lymphatic filariasis ( LF ) infects approximately 120 million people worldwide . As many as 40 million have symptoms of LF disease , including lymphedema , elephantiasis , and hydrocele . India constitutes approximately 45% of the world's burden of LF . The Indian NGO Church's Auxiliary for Social Action ( CASA ) has been conducting a community-based lymphedema management program in Orissa State since 2007 that aims to reduce the morbidity associated with lymphedema and elephantiasis . The objective of this analysis is to evaluate the effects of this program on lymphedema patients' perceived disability . For this prospective cohort study , 370 patients ≥14 years of age , who reported lymphedema lasting more than three months in one or both legs , were recruited from villages in the Bolagarh sub-district , Khurda District , Orissa , India . The World Health Organization Disability Assessment Schedule II was administered to participants at baseline ( July , 2009 ) , and then at regular intervals through 24 months ( July , 2011 ) , to assess patients' perceived disability . Disability scores decreased significantly ( p<0 . 0001 ) from baseline to 24 months . Multivariable analysis using mixed effects modeling found that employment and time in the program were significantly associated with lower disability scores after two years of program involvement . Older age , female gender , the presence of other chronic health conditions , moderate ( Stage 3 ) or advanced ( Stage 4–7 ) lymphedema , reporting an adenolymphangitis ( ADL ) episode during the previous 30 days , and the presence of inter-digital lesions were associated with higher disability scores . Patients with moderate or advanced lymphedema experienced greater improvements in perceived disability over time . Patients participating in the program for at least 12 months also reported losing 2 . 5 fewer work days per month ( p<0 . 001 ) due to their lymphedema , compared to baseline . These results indicate that community-based lymphedema management programs can reduce disability and prevent days of work lost . These effects were sustained over a 24 month period .
Lymphatic filariasis ( LF ) is a parasitic infection that leads to damage of the lymphatic system , causing lymphedema of the legs , arms , breast , or genitals . These symptoms affect an estimated 40 million people , making LF the second-leading cause of disability globally [1] , [2] . Despite remarkable progress toward the interruption of LF transmission [3] , less attention has been paid to LF morbidity management and disability prevention , which remain critical problems in many endemic areas [4] . Though filarial infection causes initial lymphatic dysfunction , development and progression of lymphedema is thought to result from recurrent episodes of secondary bacterial infections , known as adenolymphangitis ( ADL ) . Patients with lymphatic damage are at increased risk for ADL episodes due to poor lymphatic drainage and predisposition to interdigital fungal infections , which can serve as a portal of entry for pathogenic bacteria [5] . ADL episodes are characterized by pain , swelling , and inflammation of the affected extremity , often accompanied by fever or chills . These episodes further damage lymphatic vessels and worsen lymphatic dysfunction , leading to an increased risk for additional ADL episodes [5] . While LF-associated lymphedema cannot be completely cured , low-cost , effective approaches to morbidity management are available for lymphedema patients [6] , [7] . Proper care of lymphedema , known as lymphedema management , has been shown to be effective in preventing disease progression , reducing limb swelling , and reducing the frequency of ADL episodes [6]–[8] . Lymphedema management includes regular limb washing , appropriate exercise , elevation of the affected limb , early treatment of bacterial and fungal infections , and use of proper footwear [5] . Morbidity control is of special concern in India , where an estimated 59 million people are infected with the parasites that cause lymphatic filariasis , approximately 19 . 6 million of whom exhibit symptoms of lymphedema , elephantiasis , or hydrocele [9] . LF predominately affects the poorest segments of India's population , and the associated morbidity and disability are compounded by stigmatization , strict caste and gender roles , and a lack of access to healthcare [10]–[12] . Since 2007 , the Indian non-governmental organization ( NGO ) Church's Auxiliary for Social Action ( CASA ) , has been providing community-based treatment of lymphedema in Orissa State , India ( Figure 1 ) [13] . The program currently serves more than 20 , 000 lymphedema patients and their families through a network of village volunteers , who are trained to provide home-based care and instruction in lymphedema management techniques . While previous studies have demonstrated improvements in patient quality of life and a reduction in ADL episodes after beginning lymphedema management , most have assessed patients over relatively short periods of time ( ≤1 year ) and on a smaller scale [14] , [15] . The objective of this study , therefore , was to evaluate the longer-term impact of a large-scale , community-based lymphedema management program on perceived disability and productivity among lymphedema patients using a validated disability-assessment tool [16] .
This project was submitted for human subjects review to the Center for Global Health at the Centers for Disease Control and Prevention ( CDC ) , Atlanta , Georgia , USA . The project was determined to be program evaluation under CDC policy prior to the implementation of the survey . Permission for the survey was obtained from the Orissa State Department of Health and Family Welfare . Participants were asked to give their written consent prior to participation . For those unable to write , consent was documented by recording the person's fingerprint or marking the signature line with an ‘X’ and by countersignature of survey personnel . For participants under 18 years of age , verbal consent of a parent or guardian was also obtained . Consent procedures were approved by CDC and the Orissa State Department of Health and Family Welfare . Khurda District , Orissa State , India , is located near India's east coast on the northern portion of the Bay of Bengal ( Figure 1 ) , and contains the state capitol of Bhubhaneswar . Khurda District has a population of approximately 1 . 9 million and is highly endemic for lymphatic filariasis caused by Wuchereria bancrofti , with surveys from 2001–2005 estimating between 22 , 500 and 235 , 000 microfilaria-infected persons [17]–[20] . CASA provides services to >20 , 000 lymphedema patients in the Orissa State . Study patients were enrolled from randomly selected villages in Bolagarh , one sub-district of Khurda district . The map shown in Figure 1 was generated by ArcMAP 10 . 1 software ( ESRI , Redlands , California , USA ) , using shapefiles downloaded from DIVA-GIS ( http://www . diva-gis . org/gdata ) . The study was conducted from July 2009–July 2011 in 30 villages in Bolagarh sub-district . Villages were eligible for inclusion in the study if they had not yet been enrolled in the lymphedema management program , and were not located in the immediate vicinity of a village that had already been enrolled in the program . Lymphedema patients were selected based on a house-to-house morbidity census conducted by CASA in 2003 and repeated prior to the start of the program . Patients were eligible for the study if they were ≥14 years of age and had reported lower leg swelling of at least three months duration . Patients with lymphedema of the breast , arm , or genitals ( in the absence of lower-limb lymphedema ) were not eligible for participation in the study . The study was powered to detect a 5% decrease in the frequency of ADL episodes , with a 15% drop-out rate , from baseline to 24 months post-enrollment in the lymphedema management program , with an alpha of 0 . 05 . In-person interviews with participating patients were conducted by trained local interviewers in Oriya , the local language . Interviews included questions regarding general demographic information , history of lymphedema , understanding of and compliance with lymphedema management , frequency of ADL episodes , and access to care . Lymphedema patients were evaluated prior to enrollment in the lymphedema management program , and again at 1 month , 2 months , 3 months , 6 months , 12 months , 18 months , and 24 months after enrollment in the program . Evaluation included a physical examination of the affected extremity and administration of a pre-tested questionnaire . Due to logistics issues , the 18 month data collection was not performed on time and therefore is not included in this analysis . The physical assessment of each patient was conducted by both a trained interviewer and a supervisor . Both the interviewer and the supervisor performed independent staging of the leg ( s ) and photographs were taken of the affected limb ( s ) . Staff used the 7-stage classification system developed by Dreyer and colleagues [5] to stage patients' degree of lymphedema . Where staging was inconsistent between the interviewer and supervisor , or with prior or subsequent staging , photographs were independently reviewed by two physicians with extensive LF experience ( P . Budge and L . Fox ) , and discrepancies were resolved . For this analysis , stages were combined into three categories: early lymphedema ( stages 1–2 ) , moderate lymphedema ( stage 3 ) , and advanced lymphedema ( stages 4–7 ) . An adenolymphangitis ( ADL ) episode was defined as a patient self-report of two of more of the following symptoms: redness , pain , or swelling of the leg or foot , with or without the presence of fever or chills , during any point in the previous 30 days . The World Health Organization Disability Assessment Schedule II ( WHO DAS II ) was administered to patients at each interview . The WHO-DAS II survey , created by the World Health Organization , is designed to assess daily function across six broad categories , or domains , including cognition , mobility , self-care , getting along with others , life activities , and participation in society [16] . The instrument measures an individual's perception of their disability through a series of questions scored on a 5-point scale ranging from 1 ( “No difficulty” ) to 5 ( “Extreme difficulty or cannot do” ) . The questions are based on the interviewee's perception of their experiences over the last 30 days . Taken together , these scores provide an overall assessment of total perceived disability , with higher scores corresponding to higher levels of perceived disability . This analysis used simple ( un-weighted ) scoring of the WHO-DAS II domains to calculate an overall disability score . Data were independently dual-entered into Epi Info 7 , ( Stone Mountain , 2008 ) and then checked for inconsistencies . Data cleaning and analysis were performed in SAS 9 . 3 ( Cary , North Carolina , USA ) . Paired T-tests were used to examine perceived disability changes over time and changes in mean days of work lost due to lymphedema . These paired analyses compared the disability or domain score at each time point to the same patients' scores at baseline—the baseline scores of patients not present at any given assessment were not included in that assessment's comparison . Mixed effects model linear regression was used to identify factors associated with changes in disability scores over time , taking into account correlations in the data over the entire 24 month study period . All variables that were statistically significant ( P≤0 . 05 ) on univariate analysis were included in the final predictive model , as were important demographic variables . Variables were checked for co-linearity before their inclusion in the final model . To examine the effect of loss to follow-up , sensitivity analyses using the methods listed above , but excluding those patients not present at study end , were performed .
A total of 457 patients were selected from 30 villages . Initially , 375 ( 82% ) met the inclusion criteria and agreed to participate in the study . Five patients were subsequently excluded from analysis , due to lack of lymphedema on examination ( n = 2 ) , failure to meet the age criteria ( n = 1 ) , or mislabeling of survey forms ( n = 2 ) . Fifty-four ( 14 . 6% ) patients were lost to follow-up during the 24 month study period . Over the course of the study , reasons for non-participation at any particular assessment were absence from the village at the time of the assessment ( 70% ) , refusal ( 7% ) , illness ( 6% ) , or death ( 17% ) . In total , the study encompassed over 658 person-years of observation time ( baseline to time of last follow-up ) . At enrollment participants averaged 57 . 2 years of age , and the majority were women ( 218 , 59% ) ( Table 1 ) . Most participants ( 298 , 81% ) were married , and only 75 ( 20% ) had more than a primary school education . Approximately half of the study population ( 49% ) identified “Homemaker/Housekeeper” as their primary occupation , while 57 ( 17% ) reported being unemployed or retired . More than 40% ( 162 ) of patients reported at least one chronic health condition other than lymphedema at one or more time points during the study . The most commonly reported chronic conditions were gastrointestinal problems ( 18% ) and high blood pressure ( 17% ) . The majority of patients had lymphedema classified as “early” ( Stage 1–2 ) ( 50% ) , or “moderate” ( Stage 3 ) ( 36% ) . Only 53 ( 14% ) of patients had lymphedema classified as “advanced” ( Stages 4–7 ) . Patients reported having experienced lymphedema symptoms for an average of 25 . 5 years ( range: 1 . 0–75 . 0 years ) . One hundred twenty-four ( 34% ) patients reported bilateral lymphedema . There were no statistically significant differences in demographic characteristics between baseline and the twenty-four month assessment except that significantly more patients were classified as having early lymphedema at 24 months as compared to baseline ( p = 0 . 0155 ) . A subset analysis of the 316 patients present at 24 months revealed that 55 of these patients ( 17% ) were in a lower stage category at study end compared to baseline , while 20 ( 6% ) were in a higher stage category ( data not shown ) . Among the 54 patients lost to follow-up by study end , 32 ( 59% ) had early lymphedema , 15 ( 28% ) had moderate lymphedema , and 7 ( 13% ) had advanced lymphedema ( data not shown ) . This did not vary significantly from the baseline characteristics of those who remained in the study . Composite disability scores from the WHO-DAS II questionnaire decreased from an average score of 66 . 2 at baseline to 60 . 4 at 24 months ( p<0 . 0001 ) , a decline of more than 9% . This reflects significant and sustained reduction in each of the six WHO-DAS II component domains , with the exception of mobility and self-care ( Figure 2 ) . Patients reported a 13% decrease in cognitive disability from baseline to twenty-four months post-enrollment ( p<0 . 0001 ) . Disability in the domain “Getting Along with Others” declined 12% ( P<0 . 005 ) , while disability in life activities decreased 7% ( p = 0 . 0046 ) . Difficulty participating in society decreased 11% ( p<0 . 0001 ) from baseline to twenty-four months . Disability in mobility also decreased slightly during the follow-up period ( 4% ) , as did scores for self-care ( 6% ) , though neither of these declines was statistically significant . After stratifying by lymphedema category , patients with the most advanced lymphedema ( Stages 4–7 ) saw the largest reductions in overall disability scores ( Figure 3 ) . Scores in this group fell approximately 13% between baseline and 24 months ( p = 0 . 0044 ) . Patients with moderate lymphedema ( Stage 3 ) reported a 10% drop in disability over 24 months ( p = 0 . 0011 ) , while patients with early stage lymphedema ( Stages 1–2 ) experienced a smaller percent reduction in scores ( 5% ) that did not reach statistical significance ( p = 0 . 0697 ) . A number of factors were significantly associated with total disability score on univariate analysis ( Table 2 ) . Factors associated with a decrease in perceived disability and lower WHO-DAS II composite scores included having at least a primary school education ( estimate: −6 . 9 , 95% CI: −10 . 4 , −3 . 4 ) , and being employed as a homemaker ( estimate: −5 . 6 , 95% CI: −7 . 9 , −3 . 4 ) or a worker or student ( estimate: −7 . 6 , 95% CI: −10 . 0 , −5 . 3 ) . Being currently married was also associated with a lower disability score ( estimate: −3 . 1 , 95% CI: −5 . 7 , −0 . 6 ) . The only individual component of lymphedema management that was significantly associated with reduced disability levels on univariate analysis was wearing shoes . Compared to those never wearing shoes , individuals reporting always wearing shoes while outside had disability scores 2 . 8 points lower ( 95% CI: −4 . 6 , −1 . 1 ) . When compared to baseline measures , time in the program was also associated with decreased disability scores for every time-point . Factors associated with an increased disability score in univariate analysis included belonging to the highest age quartile ( estimate: 10 . 9 , 95% CI: 7 . 0 , 14 . 8 ) , the presence of one or more additional chronic health problems ( estimate: 7 . 2; 95% CI: 4 . 3 , 10 . 1 ) , moderate ( estimate: 2 . 7; 95% CI: 0 . 9 , 4 . 6 ) or advanced lymphedema ( estimate: 11 . 7 , 95% CI: 8 . 3 , 15 . 0 ) , bilateral lymphedema ( estimate: 3 . 5 , 95% CI: 0 . 7 , 6 . 3 ) , the presence of interdigital lesions , which are fungal and bacterial infections in the interdigital web spaces , ( estimate: 5 . 1 , 95% CI: 3 . 3 , 6 . 8 ) , and having had an ADL episode in the previous 30 days ( estimate: 11 . 5 , 95% CI: 10 . 0 , 13 . 0 ) . The strongest predictor of perceived disability , however , was patients' self-reported health rating for the past 30 days . Patients reporting “Very bad” health had scores more than 38 points higher than those reporting “Very good” health ( estimate: 38 . 9 , 95% CI: 33 . 1 , 44 . 6 ) . In multivariate analysis several factors remained significantly associated with decreased disability scores on the WHO-DAS II ( Table 2 ) . After controlling for covariates , patients who reported being employed as a homemaker had WHO-DAS II scores 5 . 7 points lower than those who were unemployed ( 95% CI: −8 . 2 , −3 . 3 ) , while worker/student had scores that were 4 . 5 points lower than those who were unemployed ( 95% CI: −6 . 8 , −2 . 1 ) . The individual components of lymphedema management , including soap use , elevation of the affected limb , wearing shoes , and antifungal cream use were not significantly associated with disability scores after controlling for other covariates . However , patients who reported performing leg exercises more than once a week ( but less than once a day ) had scores 2 . 6 points lower than patients who never performed the exercises ( 95% CI: −5 . 2 , −0 . 1 ) . Time enrolled in the program was significantly associated with decreased disability scores through 12 months of program participation . Risk factors for increased overall disability that remained significant in multivariate analysis included belonging to the oldest age quartile , female gender , the presence of other chronic health problems , moderate or advanced lymphedema , the presence of interdigital lesions , and having had an ADL episode in the past 30 days . Patients belonging to the oldest age quartile had scores 7 . 9 points higher than patients in the youngest age quartile ( 95% CI: 4 . 0 , 11 . 7 ) , while women scored 5 . 7 points higher than their male counterparts ( 95% CI: 2 . 2 , 9 . 3 ) . Patients with lymphedema stage 4 or higher scored 8 . 4 points higher on the WHO-DAS II than patients with early stage lymphedema ( 95% CI: 5 . 0 , 11 . 8 ) . ADL episodes had the largest effect on disability scores in our model . Patients reporting an ADL episode in the previous 30 days had scores 10 . 6 points higher than those who had not reported an ADL episode ( 95% CI: 9 . 1 , 12 . 2 ) . In a multivariate model including the predictors above as well as patient self-reported overall health status , patient health rating during the previous 30 days remained the largest predictor of increased disability ( data not shown ) . After controlling for covariates , patients reporting “Very bad” health during the last 30 days scored approximately 33 points higher on total disability than patients who reported “Very good” health ( 95% CI: 27 . 2 , 39 . 4 ) . At each assessment time point , patients were asked about the number of days of work lost in the preceding 30 days due to lymphedema-associated disability . At baseline , patients reported an average of 6 . 4 ( 95% CI: 5 . 6 , 7 . 2 ) days of work lost due to disability in the previous 30 days ( Figure 4 ) . After enrollment into the lymphedema management program , the average number of days of work lost in the previous 30 days declined to 4 . 7 ( 95% CI: 4 . 0 , 5 . 4 ) at 2 months and 2 . 9 ( 95% CI: 2 . 4 , 3 . 4 ) at 6 months . At 24 months post-enrollment , the number of days of work lost remained significantly lower than baseline , at 3 . 9 ( 95% CI: 3 . 2 , 4 . 6 ) . Stratified by lymphedema stage , patients with advanced lymphedema reported missing more days of work due to their lymphedema in the previous 30 days at baseline than patients with early stage lymphedema ( 5 . 3 days vs . 10 . 4 days , p = 0 . 0265 ) ( Data not shown ) . However , patients with advanced lymphedema also saw the greatest reduction in days of work lost at 24 months , with a 44% decline from 10 . 4 to 5 . 9 days ( p = 0 . 0083 ) . Patients with moderate stage lymphedema also saw a significant decrease in days of work lost from 6 . 2 to 4 . 5 days . ( 28% , p = 0 . 0439 ) . The mixed effects model used to analyze factors associated with a change in disability score and the paired comparisons of perceived disability ( comparing each individual's score to their corresponding baseline score ) account for missing data , so exclusion of the 54 patients not present at study end should have little effect on the reported outcomes . To verify this , the analyses were repeated including only those 316 patients present at study end . This did not change the significance of any observed differences in perceived disability , and in all cases exaggerated the magnitude of the difference ( data not shown ) . Excluding the 54 patients not present at study end also made no difference in determining which variables were significant in the multivariate analysis , except to make the association between the 24 month assessment ( variable “Time” in Table 1 ) reach statistical significance ( data not shown ) .
This study had several limitations . First , survey results were based on patient recall and perceived disability during the previous 30 days and may be subject to recall bias . Patients enrolled in this study were included based on the presence of lymphedema in one or both legs . Because blood was not drawn to test for the presence of microfilaremia or filarial antigenemia the lymphedema management program may have enrolled patients with non-filarial lymphedema . Nevertheless , it is important to note that lymphedema management programs are recommended for lymphedema resulting from all causes . Additionally , the study is limited by the lack of a comparable control group not receiving the community-based lymphedema management program , as it is considered inappropriate to withhold knowledge of lymphedema management techniques from patients with lymphedema . In order to account for repeat measurements over time , we used a mixed effects model that incorporated time as a variable in both the univariate and multivariate analyses . Because compliance with foot care was dramatically increased at all assessments subsequent to baseline , our model likely underestimates the effect of compliance with foot care on overall disability score . Indeed , compliance with foot care becomes highly significant when time is taken out of the model ( data not shown ) . Finally , there was an increase in most WHO DAS II domain scores at 12 months compared to 6 months . It is not unexpected to see fluctuations in perceived disability from chronic diseases; more frequent or longer monitoring would provide a better sense of whether the benefits we have observed will be sustained . Future research will address the relationship between ADL episodes and lymphedema progression in this cohort . While the effects of lymphedema management on clinical disease , disability , and quality of life have been studied previously [6] , [7] , [14] , [22] , this is one of the first evaluations of a community-based lymphedema management program with 24 month longitudinal follow-up . Our findings indicate that community-based lymphedema management programs can reduce patient perceived disability and reduce the number of work days lost due to lymphedema symptoms . Significantly , these effects were maintained for two years following program enrollment . These data emphasize the need for national lymphatic filariasis elimination programs to prioritize morbidity management and disability prevention programs to improve the lives of those suffering from lymphedema associated with lymphatic filariasis .
|
Lymphatic filariasis ( LF ) is the world's second-leading cause of disability and causes limb lymphedema and elephantiasis in up to 15 million people and lymphedema or hydrocele in over 40 million people , worldwide . A massive global effort has been undertaken to eliminate LF as a public health problem . LF elimination is based on two pillars: ( 1 ) interruption of transmission and ( 2 ) treatment of clinical disease among those already affected . The Indian NGO , Church's Auxiliary for Social Action ( CASA ) , has been providing community-based treatment of lymphedema in Khurda District , Orissa State , India , since 2007 . We evaluated the impact of this treatment program on the participating patients' perceived disability using the WHO Disability Assessment Schedule II ( WHO-DAS II ) . After two years of enrollment in the program , patients had significantly lower levels of perceived disability . We found that being employed and time enrolled in the program were associated with significant reductions in disability scores . Compared to baseline , patients enrolled in the program for at least 12 months reported 2 . 5 fewer days of work lost in the previous 30 days due to their lymphedema . These findings indicate that participation in a community-based lymphedema management program can reduce patients' disability and prevent days of work lost due to lymphedema .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"public",
"health",
"and",
"epidemiology",
"clinical",
"research",
"design",
"behavioral",
"and",
"social",
"aspects",
"of",
"health",
"infectious",
"disease",
"epidemiology",
"parasitic",
"diseases",
"preventive",
"medicine",
"skin",
"infections",
"global",
"health",
"cohort",
"studies",
"neglected",
"tropical",
"diseases",
"lymphatic",
"filariasis",
"infectious",
"diseases",
"epidemiology",
"public",
"health",
"socioeconomic",
"aspects",
"of",
"health",
"prospective",
"studies"
] |
2013
|
Impact of Community-Based Lymphedema Management on Perceived Disability among Patients with Lymphatic Filariasis in Orissa State, India
|
The high rates of RNA virus evolution are generally attributed to replication with error-prone RNA-dependent RNA polymerases . However , these long-term nucleotide substitution rates span three orders of magnitude and do not correlate well with mutation rates or selection pressures . This substitution rate variation may be explained by differences in virus ecology or intrinsic genomic properties . We generated nucleotide substitution rate estimates for mammalian RNA viruses and compiled comparable published rates , yielding a dataset of 118 substitution rates of structural genes from 51 different species , as well as 40 rates of non-structural genes from 28 species . Through ANCOVA analyses , we evaluated the relationships between these rates and four ecological factors: target cell , transmission route , host range , infection duration; and three genomic properties: genome length , genome sense , genome segmentation . Of these seven factors , we found target cells to be the only significant predictors of viral substitution rates , with tropisms for epithelial cells or neurons ( P<0 . 0001 ) as the most significant predictors . Further , one-tailed t-tests showed that viruses primarily infecting epithelial cells evolve significantly faster than neurotropic viruses ( P<0 . 0001 and P<0 . 001 for the structural genes and non-structural genes , respectively ) . These results provide strong evidence that the fastest evolving mammalian RNA viruses infect cells with the highest turnover rates: the highly proliferative epithelial cells . Estimated viral generation times suggest that epithelial-infecting viruses replicate more quickly than viruses with different cell tropisms . Our results indicate that cell tropism is a key factor in viral evolvability .
RNA viruses are responsible for a disproportionate number of emerging human diseases , including influenza , ebola hemorrhagic fever , hantavirus pulmonary syndrome , and Middle East respiratory syndrome , which place tremendous health and economic burdens on both the developing and developed world [1] , [2] . In 2008 , rotavirus and measles virus caused the deaths of 570 , 000 children under the age of five , making them two of the leading killers of children worldwide [3] . In 2009 , it was estimated that rotavirus infections alone result in $325 million in medical treatment costs and $423 million in societal costs each year [4] . Further , the implementation of many intervention strategies has either failed or been delayed as a result of the evolutionary dynamics of these pathogens [1] , [5] , [6] , [7] , [8] , [9] . Differences in viral evolutionary dynamics , such as rates of evolution , can explain why certain viruses have the capacity to adapt to new host species , increase in virulence , or develop resistance to antivirals [7] , [8] , [9] , [10] , [11] . Therefore , understanding why some RNA viruses evolve more quickly can facilitate better prediction of their pathogenic and epidemiological potential [8] , [10] , [11] , [12] . Though extremely high nucleotide substitution rates are a defining feature of RNA virus evolution [1] , [13] , [14] , [15] , there have been few attempts to comprehensively examine the driving genomic and ecological factors behind these rates . Differences in the strength and direction of selection pressures on these viruses result in variation among their substitution rates [1] , [5] , [13] . However , while some general patterns have been observed in selection pressures , such as enhanced purifying selection on the structural proteins of arboviruses [16] , there have been no attempts to quantify the relationship between selection pressures and long-term viral substitution rates . The high rates of RNA virus evolution are most commonly attributed to their replication with error-prone RNA-dependent RNA polymerases ( RdRps ) [1] , [17] , but these nucleotide substitution rates are known to span at least three orders of magnitude [5] , [17] and do not correlate well with experimentally measured viral mutation rates [5] . Further , the substitution rates of some DNA viruses , which replicate with high-fidelity DNA polymerases , are comparable to the high substitution rates of RNA viruses [13] . Therefore , the polymerase error rate alone cannot explain the substitution rate variation in RNA viruses . Along with mutation rate , viral replication frequency directly impacts the rate at which mutations can be introduced , and ultimately fixed as substitutions [13] . Replication frequencies could be influenced by a variety of factors related to viral genomic architecture or ecology [13] . For example , weak negative correlations between viral genome lengths and substitution rates have been attributed to either enhanced replication frequencies or higher mutation rates in viruses with smaller genomes [15] , [17] , [18] , [19] . It has also been suggested that different transmission and infection modes result in differences in generation time , ultimately causing variation among per-year rates of synonymous substitution of RNA virus structural genes [5] . In this modern survey of mammalian RNA virus evolution rates , we generated and compiled published substitution rates of structural and non-structural genes produced by Bayesian coalescent analyses [20] . We analyzed these rates as a function of seven factors related to virus genomic architecture ( i . e . , genome length , genome sense , and whether or not the genome is segmented ) and virus ecology ( i . e . , target cell , transmission mode , host range , and whether the infection is acute or persistent ) . We also evaluated the relationships of viral substitution rates with dN/dS estimates , experimentally measured mutation rates , and estimated generation times . Though recombination undeniably plays a role in shaping viral evolutionary dynamics and could inflate substitution rate estimates [21] , [22] , we conservatively removed any potential recombinants from our datasets prior to analysis . Through this broad analysis , we were able to demonstrate that cell tropism , and its impact on viral generation time , has the greatest influence on rates of mammalian RNA virus evolution .
A review of the literature yielded 92 published Bayesian nucleotide substitution rate estimates for the structural genes of 35 different mammalian RNA viral species , and 21 published Bayesian rates for RdRps or a non-structural gene of 14 different viral species ( referred to collectively as “non-structural , ” Table S1 ) . These rates were supplemented with 26 novel Bayesian substitution rates of structural genes of 19 different viral species , and 19 novel Bayesian rates of non-structural genes of 16 different viral species ( Table S2 ) . Collectively , these rates span three orders of magnitude , ranging from 3 . 0×10−5 to 1 . 5×10−2 nucleotide substitutions per site per year ( ns/s/y ) and 2 . 0×10−5 to 1 . 3×10−2 ns/s/y for the structural genes and non-structural genes , respectively ( Table S1 ) . Plotting the levels of each variable by ascending mean substitution rate revealed similar patterns ( i . e . , the same ordering of levels ) for both the structural ( S ) and non-structural ( NS ) datasets in three of these variables , excepting transmission route . Viral substitution rates grouped according to target cell ( panels 1A and 1B ) , transmission route ( panels 1C and 1D ) , infection type ( panels 1E and 1F ) , and host range ( panels 1G and 1H ) are shown in Figure 1 . Substitution rates were also grouped by viral genomic architecture ( genome sense/strandedness , Figure 2A and 2B , and genome segmentation , Figure 2C and 2D ) and plotted against viral genome length ( Figure 2E and 2F ) . There were no apparent relationships between genomic properties and substitution rates ( Figure 2 ) , including no linear relationship between substitution rates and genome lengths in either dataset ( coefficient of determination , S: R2 = 0 . 06 , NS: R2 = 0 . 08 ) . dN/dS estimates calculated in this study were compiled with published estimates also calculated using the Single Likelihood Ancestor Counting ( SLAC ) method ( 56 structural gene dN/dS estimates , 33 non-structural gene dN/dS estimates total , Table S1 ) . ANCOVA analyses were performed separately on the structural and non-structural gene datasets to determine which , if any , of seven factors ( target cell , transmission route , infection mode , host range , genome length , genome sense , and genome segmentation ) significantly predict the nucleotide substitution rates of mammalian RNA viruses . To explore the many dummy-coded categorical variables , three analyses were run using different variable levels as the base levels ( see Methods for details , Tables 1 and 2 ) . For all of the ANCOVA analyses , the adjusted coefficient of determination ( ) was ≥0 . 73 , indicating that over 70% of the substitution rate variability can be explained by the predictor variables included in this study . Standardized residual plots identified only six potential outliers of the 118 structural gene rates and one potential outlier of the 40 non-structural gene rates ( Figure S1 ) , indicating that the data are normally distributed and therefore amenable to a general linear model . Regardless of the base levels , target cells were the only significant predictors of log-transformed substitution rates for both structural and non-structural genes ( Tables 1 and 2 ) , with cell tropism as the only significant predictor variable by type III sum of squares ( SS ) analyses ( P<0 . 0001 and P = 0 . 003 for the structural and non-structural gene datasets , respectively ) . Targeting epithelial cells or neurons was found to be the most significant predictor of structural gene rates in each analysis where these were not the base levels ( P<0 . 0001 , Table 1 , Figure 3 ) , while targeting neurons was found to be the sole significant predictor of substitution rates for the smaller non-structural gene dataset ( P = 0 . 009 , Table 2 , Figure 3 ) . Further , there was a high correlation between each viral species' estimated structural gene substitution rate and its corresponding non-structural gene rate ( 33 viruses , Pearson r = 0 . 87 , P<0 . 0001 ) . This suggests that if it were possible to calculate more non-structural rates , we would likely see results similar to those from the structural gene dataset . To minimize any potential bias introduced by using multiple published rates for a single viral strain or species , we conducted control analyses using datasets with only one rate per species . For species with multiple substitution rates in one of our datasets , we calculated the average log substitution rate and used that as the sole substitution rate for the species in the control analysis . These data were also normally distributed ( Figure S2 ) , but the for these analyses were slightly lower than for the full datasets ( S: = 0 . 65 , NS: = 0 . 70 , Tables S3 and S4 ) . These control results were consistent with those from the full dataset analyses: tropisms for epithelial cells or neurons were the most significant substitution rate predictors ( Tables S3 and S4 , Figure S3 ) . Because of the high correlation between the structural and non-structural gene rates , we combined the two datasets ( Figure 4 ) and performed a final set of three ANCOVA analyses using this combined dataset . The results from these analyses were nearly identical to those from the structural gene analyses ( Table S5 ) . The exception was that , in addition to cell tropism , Type III SS analysis also identified transmission route as a significant predictor variable ( P = 0 . 007 ) , though it was still less significant than cell tropism ( P<0 . 0001 ) . More specifically , in addition to different cell tropisms , transmission through arthropod vectors was also found to be a significant rate predictor in one of the three analyses ( P = 0 . 002 , Table S5 ) . To ensure that any substitution rate variability attributed to a given predictor variable was not significantly dependent on other predictor variables , we examined collinearity in all datasets . With the exception of the persistent infection variable , which was nested with the endothelial target cell variable and thus excluded , the ANCOVA analyses for the structural gene rate datasets and the combined rate dataset showed no significant collinearity ( no variance inflation factors ( VIF ) were greater than 10 ) . For the non-structural gene rate datasets , many different predictor variables had VIF>10 . However , subsequent analyses where each individual variable was removed did not significantly reduce collinearity in these datasets ( data not shown ) . Due to the consistent results between the structural and non-structural gene datasets , as well as those from the combined rate dataset , we concluded that correlations among independent variables did not significantly impact our results . Since target cells were found to be the only consistently significant predictors of substitution rates , a series of one-tailed t-tests was used to confirm which cell tropisms are associated with higher viral substitution rates than others . Viruses that target epithelial cells were found to have significantly higher structural gene substitution rates than viruses that target neurons , endothelial cells , or leukocytes ( Table 3 , P<0 . 0009 ) . Similarly , viruses that target epithelial cells were found to have significantly higher non-structural gene substitution rates than viruses that target neurons , hepatocytes , or leukocytes ( Table 4 , P<0 . 0007 ) . These results were recapitulated in the control datasets that only used one rate per viral species ( Tables S6 and S7 ) . It should be noted , however , that most of the viruses in this study that are classified as targeting leukocytes ultimately cause systemic infections and infect a wide variety of cell types . Consequently , viruses in the leukocyte target cell category had the most rate variation of all the target cell categories ( Figure 1 ) . Because transmission through arthropod vectors was also found to be a significant rate predictor in the ANCOVA analyses based on the combined datasets and because of the correlation between epithelial cell tropism and fecal-oral/respiratory transmission , we evaluated any significant variation among substitution rates of viruses with different transmission routes . Using a series of one-tailed t-tests , we found that viruses that are transmitted through the fecal-oral/respiratory route have significantly higher substitution rates than those transmitted by arthropod vectors ( P<0 . 0001 ) . However , we also compared different cell tropisms within each of these transmission routes . We found that fecal-oral/respiratory transmitted viruses that target epithelial cells have significantly higher substitution rates than those that target other cell types ( P<0 . 0001 , Figure 5 ) . Similarly , we found that neurotropic arboviruses have significantly lower substitution rates than arboviruses that target other cell types ( P<0 . 001 , Figure 5 ) . We also tested for linear relationships between viral substitution rates and other evolutionary parameters for which only smaller subsets of our datasets could be analyzed . Reliable experimentally measured mutation rates estimated as mutations per base per infectious cycle were only available for four different viruses included in this study ( poliovirus 1 [11] , [23] , [24] , hepatitis C virus [25] , influenza A virus [26] , [27] , [28] , influenza B virus [26] ) . Mutation rates measured as mutations per base per strand replication were only available for three viruses included in this study ( poliovirus 1 [29] , measles virus [30] , [31] , and influenza A virus [32] ) . These mutation rates were not significantly correlated with their corresponding substitution rate estimates ( r = 0 . 69 , P = 0 . 31 and r = −0 . 93 , P = 0 . 25 , for mutation rates measured as mutations per base per infection and mutation rates measured as mutations per base per replication , respectively ) . Similarly , there were no significant correlations between the estimated substitution rates and dN/dS estimates ( ρ = −0 . 02 , P = 0 . 88 and ρ = −0 . 07 , P = 0 . 68 , for the limited structural gene and non-structural gene datasets , respectively ) . ANCOVA and t-tests consistently revealed epithelial cell tropism and neurotropism as the most significant viral substitution rate predictors . Since these two cell types have some of the highest and lowest turnover rates , respectively , of all mammalian cells [33] , [34] , [35] , [36] , we sought to determine if there were any associations between host cell turnover rate and viral generation time . Using the model proposed by Sanjuán ( 2012 ) that relates the long-term substitution rate , K , to the mutation rate , μ , correcting for transient deleterious mutations , we were able to estimate generation times for the few viruses with reliable mutation rate estimates . This model , , with , ( G = genome length , g = generation time , sH = harmonic mean of the selection coefficient ) [15] , confirmed that influenza A virus , influenza B virus , and poliovirus , which target epithelial cells , have substantially shorter generation times ( <40 hours ) than hepatitis C virus , which targets hepatocytes ( >200 hours ) . These results , while based on a very limited dataset , provide quantitative evidence for a link between cell tropism and generation time . Shorter average generation times lead to more rounds of replication per year , which could neatly explain higher per-year substitution rates .
Variation in strength and/or direction of selection has frequently been invoked as a determinant of viral substitution rates [12] , [13] , [21] . While positive selection can certainly result in variation among very short-term substitution rates , purifying selection tends to dominate over longer timescales [21] , [45] , [46] , [47] . However , variation is observed in the strength of purifying selection due to differences in host ranges . For instance , as previously mentioned , viruses vectored by arthropods have unique evolutionary constraints placed on them by their host diversity [41] , [42] , [43] , [48] . While previous studies found that arboviruses are under stronger purifying selection than non-arboviruses [1] , [41] , [49] , we found that the dN/dS estimates based on structural genes of arboviruses were not significantly lower than those for non-arboviruses ( P = 0 . 19 ) . The dN/dS estimates based on non-structural genes of arboviruses were only moderately lower than those for non-arboviruses ( P = 0 . 04 ) . Further , we found no significant correlation between the estimated dN/dS and substitution rates , suggesting that detectable differences in selection pressures do not explain the variation in substitution rates of mammalian RNA viruses . To date , there are no data supporting a link between cell tropism and sustained differences in selection pressures . Compared to the slower evolution of DNA viruses , the evolution of RNA viruses is dominated by their high mutation rates [1] , [13] , [15] . Weak negative correlations between genome lengths and viral substitution rates have been attributed to a relationship between mutation rate and substitution rate , as smaller genomes could in theory withstand higher mutation rates than larger genomes [13] , [15] , [50] . However , while differences in spontaneous mutation rates appear to be significantly correlated to the long-term substitution rates of DNA viruses [15] , this linear relationship disappears past a certain mutation rate threshold: around 10−6 mutations per site per infectious cycle , the lower end of the mutation rate range of RNA viruses [13] , [15] . It is , therefore , not surprising that we found no significant correlation between substitution rates and the available , reliable mutation rate estimates . Additionally , a recent study of the retrovirus HIV-1 found that infection of different cell types did not lead to differences in mutation rate [51] , providing some evidence that mutation rate is not correlated with cell tropism . Together , these data suggest that mutation rate variation among different cell types is not driving higher substitution rates in epithelial-infecting mammalian RNA viruses . Ruling out selection , mutation rates , and recombination frequencies as drivers of RNA virus substitution rates implies that the rate variation is largely the result of variation in replication dynamics [5] , [13] . Enhanced replication frequencies ( shorter generation times ) have been used to explain a variety of the previously suggested links between virus ecology and substitution rate . For example , viruses in the acute phase of an infection generally replicate more frequently than those in a persistent infection , and viruses in a latent phase do not replicate at all [39] . Further , as an alternative to differential selection pressures , the argument that transmission mode drives viral substitution rates assumes that viruses that can be transmitted more rapidly will have shorter generation times ( e . g . , horizontal transmission vs . vertical transmission [5] , [52] , [53] ) . DNA viruses have shorter generation times in faster dividing cells [54] , [55] , but the associations between cell tropism and RNA virus generation time are less obvious , as RNA viruses do not depend on cellular replication machinery . However , there is evidence that for at least some RNA viruses , viral genome replication is highly dependent on host cell proliferation , with RNA synthesis occurring at much lower rates in poorly proliferating cells than in rapidly dividing cells [56] , [57] , [58] , [59] , [60] . For example , it has been repeatedly demonstrated that hepatitis C virus genome replication is enhanced in proliferating cells , perhaps due to higher levels of available nucleotides [59] , or because of higher levels of viral protein synthesis facilitated by nuclear translation initiation factors that only become available in the cytoplasm during cell division [58] . Similar dependence on cell proliferation for viral replication efficiency has been demonstrated in a number of picornaviruses [57] , [60] , [61] , [62] . Further , using the model proposed by Sanjuán ( 2012 ) , we found that viruses that infect epithelial cells have generation times that may be as much as 40-fold shorter than a virus that infects non-epithelial cells . This offers a possible mechanistic basis for our finding that viruses that target the fastest-dividing cells in the body ( intestinal and respiratory epithelial cells [34] , [35] , [36] , [63] ) have higher substitution rates than viruses that infect cells that turnover at very low rates , if at all ( neurons [33] , [35] , [64] ) . We are the first to provide statistical evidence that cell tropism predicts rates of mammalian RNA virus evolution , likely through its influence on virus generation time . These results offer a new perspective on why it has been difficult to create effective vaccines for viruses that infect epithelial tissue , such as rotavirus and enterovirus 71 [65] , [66] . Further , as it has been shown that higher rates of viral evolution can result in increased genetic diversity and higher epidemiological fitness [26] , [67] , [68] , the higher substitution rates of epithelial-infecting viruses predict increased evolvability and greater potential for emergence in novel host species [21] .
Long-term nucleotide substitution rates of mammalian RNA viruses were collected from the literature , with a focus on finding rates for the outer structural gene containing the major antigenic site ( s ) and non-structural ( preferably the RdRp ) genes . While the RdRp genes of the ( - ) ssRNA and dsRNA viruses are classified as structural , or virion-associated , genes [69] , they are generally thought to be more conserved and under very different selection pressures than the structural genes that interact with the host immune system [70] , [71] . We excluded retroviruses from analysis because they are known to have highly variable substitution rates due to time spent integrated into DNA genomes , where they evolve at the rate of their hosts' genome [13] , [72] . Viruses that predominately infect non-mammals , with mammals serving as incidental , dead-end hosts , were also excluded . Only rates estimated for individual viral species or strains were used , not those that aggregated multiple species into one analysis . Similarly , only rates from single gene analyses were included , not those based on full genomes or multiple gene alignments . In order to minimize any rate discrepancies that could result from variations among datasets ( e . g . , number of taxa , temporal range , portion of gene analyzed ) and/or subtle methodological variations [45] , [73] , [74] , [75] , [76] , [77] , only rates produced by Bayesian coalescent analyses of datasets composed of at least 30 taxa , isolated over a minimum range of 15 years and spanning at least 40% of the analyzed gene were included . Bayesian coalescent analyses provide estimates of viral evolution that are calculated over a longer range than simply the date range over which the taxa were isolated . This is because they determine the likely phylogenetic relationship among the isolates and infer substitution rates over the entire evolutionary history of the sampled taxa: over decades , hundreds , even thousands of years . These rates can therefore be considered “long-term” nucleotide substitution rates . Data regarding genomic architecture and ecology were obtained for all viruses with published substitution rates that met these criteria . We included multiple rates for a given virus when available , except when a single study examined multiple lineages and summarized the results in a single rate [78] , [79] , [80] , [81] . Corresponding dN/dS estimates were collected when available . These published substitution rates were supplemented with novel BEAST [20] rate analyses based on the sequence data available in GenBank ( accessed through Taxonomy Browser , http://www . ncbi . nlm . nih . gov/Taxonomy ) . Sequences for structural and non-structural genes with years of isolation available in GenBank or the literature were manually aligned using Se-Al v2 . 0a11 [82] . Sequences with GenBank or published information that indicated they were genetically manipulated or extensively passaged in the lab prior to sequencing were eliminated from further analysis . The final datasets also adhered to the conservative criteria described above for published datasets . As recombination events can lead to over-estimation of nucleotide substitution rates , each dataset was scanned for recombination using seven different algorithms ( RDP , GENECONV , Bootscan , MaxChi , Chimaera , SiScan , and 3seq ) implemented in RDP v3 . 44 [83] . Sequences implicated as recombinant by two or more algorithms were excluded from further analysis . These finalized alignments were deposited into Dryad ( doi:10 . 5061/dryad . 58ss8 ) . Modeltest v3 . 7 [84] was used to determine the best-fit model of nucleotide substitution for each dataset ( by AIC ) . Long-term nucleotide substitution rates were estimated using BEAST v1 . 5 . 4 [20] . Each dataset was run for at least 50 million generations and until all parameters had stabilized ( effective sampling size >200 ) . Each dataset was run with two different clock models ( strict and uncorrelated lognormal ) and three different demographic models ( constant , exponential , and Bayesian skyline ) . The best-fitting clock/demographic model combination for each dataset was determined using Bayes factors as implemented in Tracer v1 . 5 [85] . For each best set of priors , two independent runs were performed to ensure that the results were replicable , and a control analysis was run without the dataset to ensure that the priors were not controlling the outcome of the analysis . The Single Likelihood Ancestor Counting ( SLAC ) , codon-based maximum likelihood method available in the HYPHY package on the Datamonkey web server [86] was used to evaluate the strength of selection pressure on these datasets . In order to determine which factors most significantly predict substitution rates of mammalian RNA viruses , ANCOVA analyses were run using SPSS Statistics v21 ( IBM ) with log-transformed mean substitution rates as the dependent variable and seven overarching predictor variables ( target cell , transmission route , whether the infection is acute or persistent , host range , genome length , genome sense , and whether or not the genome is segmented ) . For each variable , different base levels were tested to ensure that the chosen base level did not significantly influence the results . Collinearity among the variables was also assessed , with variance inflation factors ( VIF ) greater than 10 indicating redundancy among variables . Separate ANCOVA analyses were run on the structural and non-structural gene datasets . As there were multiple published rates for some viral species and strains , additional analyses were run for both the S and NS datasets with only one substitution rate per virus species . When there were multiple rates for a given virus species , we calculated and used an average rate . One-tailed t-tests were subsequently run in R v2 . 14 . 1 [87] to provide an additional measure of significant directional variation among the log-transformed mean rates of different levels for any categorical variable that was found to be a significant rate predictor ( α = 0 . 01 , adjusted by Bonferroni correction for multiple comparisons ) in the ANCOVA analyses . Additional t-tests were also conducted using the control datasets with one rate per virus species . Additionally , though there were no dN/dS or mutation rate estimates available for all viruses used in this study , the available data for each variable were compared to corresponding log-transformed mean substitution rate estimates using Spearman rank correlation ( for dN/dS ) or Pearson correlation coefficient ( for mutation rates ) . Structural and non-structural gene rate estimates were also compared using Pearson correlation coefficient . All correlation analyses were performed in SPSS Statistics v21 .
|
RNA viruses are the fastest evolving human pathogens , making their treatment and control difficult . Compared to DNA viruses , RNA viruses replicate with much lower fidelity , which can explain why RNA viruses evolve significantly faster than most DNA viruses . However , there is tremendous variation among the evolutionary rates of different RNA viruses , which is not explained by variation in mutation rates . Here we present a survey of mammalian RNA virus rates of evolution , and a comprehensive comparison of these rates to different properties of virus genomic architecture and ecology . We found that cell tropism is the most significant predictor of long-term rates of mammalian RNA virus evolution . For instance , viruses targeting epithelial cells evolve significantly faster than viruses that target neurons . Our results provide mechanistic insight into why viruses that infect respiratory and gastrointestinal epithelia have been difficult to control .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"rna",
"viruses",
"viral",
"classification",
"virology",
"biology",
"microbiology",
"viral",
"evolution"
] |
2014
|
Cell Tropism Predicts Long-term Nucleotide Substitution Rates of Mammalian RNA Viruses
|
In Drosophila , the MSL ( Male Specific Lethal ) complex up regulates transcription of active genes on the single male X-chromosome to equalize gene expression between sexes . One model argues that the MSL complex acts upon the elongation step of transcription rather than initiation . In an unbiased forward genetic screen for new factors required for dosage compensation , we found that mutations in the universally conserved transcription elongation factor Spt5 lower MSL complex dependent expression from the miniwhite reporter gene in vivo . We show that SPT5 interacts directly with MSL1 in vitro and is required downstream of MSL complex recruitment , providing the first mechanistic data corroborating the elongation model of dosage compensation .
Drosophila dosage compensation is widely used as a model system to investigate how transcription is regulated by large scale chromatin modifications [1] . To equalize the expression of the X-linked genes between XY males and XX females , the single X-chromosome in males is hypertranscribed a modest , but essential ∼1 . 4–1 . 8 fold . This is accomplished by the MSL complex , which consists of at least five proteins and two noncoding roX ( RNA on X ) RNAs [2] . The complex contains the histone modifying enzymes MOF ( H4K16ac ) and MSL2 ( H2BK34ub ) [3] . MSL3 is a chromodomain protein implicated in MSL complex distribution to its target site [4] . MSL1 assembles the complex via discrete docking sites for MSL2 , MSL3 , and MOF . MLE is an ATPase/helicase with double stranded RNA binding motifs that associates with the complex in an RNA dependent manner . A long-standing puzzle is the biochemical mechanism by which the MSL complex up regulates X-linked genes , each of which is controlled by different transcription factors . An elegant model that solves this problem posits that MSL complex does not act with diverse gene-specific transcription factors to alter initiation , but rather at the elongation step of transcription common to all genes [5] . This proposal is supported by the higher resolution mapping of MSL complex binding and H4K16 acetylation within the bodies of actively transcribed X-linked genes with a bias towards the 3′ end [6]–[8] . Global nuclear run on analysis showed that compared to autosomes , the male X-chromosome has higher levels of transcriptionally engaged RNAPII ( RNA Polymerase II ) within the distal portions of the genes [9] . In contrast to this , a recent study detected increased RNAPII occupancy at the promoters of X-linked genes in males leading to the alternate idea that dosage compensation operates at the level of transcription initiation [10] . It is not clear whether decondensation of the chromatin fiber by H4K16 acetylation aids passage or recruitment of RNAPII enough to explain dosage compensation [11] , or if the MSL complex has additional interactions with the basal transcriptional machinery . To search for new factors involved in dosage compensation we performed an unbiased forward genetic screen that relies on a sensitive eye pigmentation reporter of MSL complex activity . This approach was designed to recover heterozygous mutations in genes that are essential for general transcription in both sexes but play an additional role in male dosage compensation . We recovered multiple alleles of Spt5 , a universally conserved transcription elongation factor . We found that SPT5 is required for dosage compensation in males and extensively colocalizes with the MSL complex on the X-chromosome . Moreover , we found that SPT5 and MSL1 directly interact with each other . We propose that SPT5 is required downstream of MSL complex recruitment to stimulate transcription elongation . The identification of SPT5 is strong mechanistic evidence supporting the elongation model of dosage compensation .
The eye color of roX1 transgenic males is a sensitive reporter of MSL activity [12] . When roX1 transgenes occasionally land in repressive chromatin , the miniwhite marker is epigenetically silenced so that females have solid white eyes . Males have spotted eyes because the MSL complex binds the autosomal roX1 transgene and locally modifies the chromatin allowing miniwhite expression in a fraction of cells . We have previously described a strategy for isolating mutations that increased local MSL activity [13] . Here we use a similar method to isolate mutations that reduce MSL activity . This approach has two important advantages . First , it allows identification of factors that are instrumental in achieving dosage compensation of the male X-chromosome in vivo , but may associate with the MSL complex too weakly or transiently to copurify with MSL proteins . Second , our genetic strategy retains a wild type allele of the relevant gene allowing us to capture factors that have additional essential functions . Homozygous mutations in such factors would be lethal to both sexes and thus would have been missed in earlier genetic screens based on the male specific lethal phenotype . We screened approximately 16 , 000 EMS mutagenized flies and identified 48 mutations that dramatically lowered the eye pigmentation in males ( Figure 1A and 1B , Figure S1 ) . It is difficult to estimate if the amount of pigmentation in any individual ommatidia changes . What the screen detects is a change in the fraction of ommatidia that do or do not derepress the miniwhite reporter linked to roX1 . Mutants that were also recessive lethal to both sexes were placed in complementation groups ( Figure S1 ) . We tested the ability of the modifier mutants to suppress eye pigmentation in multiple mosaic roX1 transgenes inserted in distinct repressive locations reasoning that those were more likely to affect dosage compensation rather than the particular silencing factors acting on flanking chromatin ( Figure S2 ) . One uninteresting mechanism that might produce this phenotype would be mutations that globally strengthened the repressive chromatin environment responsible for silencing of the miniwhite gene in our roX1 reporters . We tested the effect of the new mutants on In ( 1 ) wm4 which displays classic position effect variegation in both sexes . Most of the candidate modifiers of dosage compensation did not affect pigmentation in In ( 1 ) wm4 ( Figure 1C and 1D ) arguing against a global increase in repressive chromatin . Complementation group C was chosen for detailed analysis . Meiotic recombination placed the locus near the polytene bands 56C-F but none of the available chromosome deficiencies uncovered the mutation [14] . Closer inspection revealed a gap in the deficiencies where elongation factors Spt5 and Elongin-C are located ( Figure 1E ) . Available mutations in Elongin-C complemented all five group C alleles and we found no lesions in Elongin-C upon sequencing ( data not shown ) . However , when we tested the Spt5MGE-3 mutation [15] , it failed to complement all five group C alleles . Sequencing genomic DNA from these mutants identified one stop codon ( Q314X ) , two splice junction mutations ( E471Z and A680Z ) , and one missense mutation ( S14F ) ( Figure 1F ) . The modular structure of SPT5 is summarized in Figure 1G . Because SPT5 is such a critically important transcription elongation factor used by many genes , we were concerned that it appeared in our screen because reducing the level of any vital general transcription factor would lower expression of our eye color reporter . To address this concern , we screened the autosomal Bloomington Deficiency stock collection . We reasoned that if disrupting transcriptional efficiency in general affected our reporter , then many deficiencies would lower red pigmentation of the mosaic roX1 lines just like Spt5 mutations had . We crossed six different roX1 mosaic lines that carry roX1 transgenes in diverse chromatin environments to 190 deficiencies . The idea was that any deficiency that affected the eye coloration of multiple roX1 reporter lines was more likely to affect some aspect of dosage compensation rather than the particular repressive environment surrounding the different inserts . We found that only 10 intervals reduced MSL complex reporter activity ( Table S1 ) . Moreover , removing one copy of these 10 regions lowered MSL complex dependent red pigmentation across 4 or more of the mosaic roX1 lines supporting the notion that the relevant factors are somehow acting on dosage compensation . The deficiency screen shows that silencing the roX1 eye color reporter is an uncommon dominant haploinsufficient phenotype produced by only a few loci in the genome . Thus , the phenotype seen in Spt5 mutants is unlikely to be due to a general reduction of transcription . We were still concerned that the white eye color gene used in our dosage compensation reporter might be particularly sensitive to SPT5 levels . We turned to strong hypomorphic alleles of white to test this possibility . The wa and we alleles each carry different transposon insertions that greatly reduce their expression resulting in orange eyes [17] , [18] . On this background , small changes in white expression should be easily detectable by altered eye color . We crossed our Spt5 mutations into these two stocks and observed no difference in the eye pigmentation ( Figure S3 ) . We also crossed unrelated transgenes marked with miniwhite into our new Spt5 mutations and saw no change in eye pigmentation ( data not shown ) . This shows that a 50% reduction in SPT5 levels does not alter the phenotype from hypomorphic white alleles or miniwhite . We conclude that Spt5 mutations dramatically affected the probability that males overcome silencing not because of global reduction of transcription across the genome or the white promoter itself , but rather because SPT5 plays some role in dosage compensation to which our roX1 reporter is responsive . We used a sensitized genetic background to see if Spt5 affected dosage compensation of the X chromosome in addition to the roX1 eye color reporter transgene . Because SPT5 is essential for most transcription , homozygous null animals die early in development . The viability of Spt5/+ males demonstrates that dosage compensation must be adequate even with reduced SPT5 levels . The same is true for any of the msl/+ heterozygotes . However , males with limiting MSL complex might be more sensitive to reduced levels of SPT5 . Males missing either roX1 or roX2 are alive but males missing both roX RNAs have greatly reduced male viability [19] . In our genetic background such roX1 roX2 double mutant males are completely lethal but can be rescued by an autosomal roX transgene [20] . Restoring male viability under these conditions depends on abundant MSL subunits . Males heterozygous for msl1 or mle showed reduced viability when roX1 RNA is also limiting [13] . Similarly , reducing SPT5 selectively lowered male viability to ∼15% when roX1 RNA was limiting consistent with a role in dosage compensation ( Figure 2A ) . We assayed related factors to see how specific this phenotype was and found that lowering Elongin-C , another factor involved in elongation or Jil1 , the histone H3S10 kinase that associates with the MSL complex had no effect on male viability ( Figure 2A ) . However , Su ( Tpl ) S192 , a mutation in the elongation factor ELL did reduce male viability . Others have reported that ELL RNAi lines display male specific lethality consistent with a role in dosage compensation [21] . These results are consistent with SPT5 playing a central role in dosage compensation that becomes more obvious when MSL activity is limited by low roX1 RNA levels . Females normally lack dosage compensation because SXL blocks translation of msl2 mRNA . Ectopic dosage compensation can be induced in females by artificial expression of MSL2 by the [H83M2] transgene that escapes SXL regulation [22] . The inappropriate dosage compensation slows development resulting in delayed eclosion of adult females ( Figure 2B ) . The resulting females produce very few eggs and are sterile . If reducing Spt5 weakens dosage compensation then that might reduce the toxic effects of inappropriate dosage compensation in [H83M2] overexpression females . When [H83M2] females also carried a mutation in Spt5 , the female-specific developmental delay was modestly rescued ( Figure 2B ) . However , the more striking result was that the Spt5/+ [H83M2] females produced abundant eggs that successfully developed into larvae . This argues that SPT5 is needed for MSL2 to drive inappropriate dosage compensation in females . To further examine functional links between SPT5 and dosage compensation , we tested genetic interactions between the newly recovered Spt5 mutations and unusual gain of function msl1 alleles . We previously reported two missense alleles that partially disrupt the MSL1-MOF or MSL1-MSL3 interfaces [13] . Both mutations dominantly cause msl1*/+ males to produce solid red eyes ( more MSL activity , Figure 2E ) from the mosaic GMroX1-75C transgenic reporter whose basal pattern is mostly white with scattered small red sectors ( Figure 2C ) . Spt5 mutations alone reduce sectoring slightly ( Figure 2D ) . We constructed flies heterozygous for both the msl1P864L and Spt5S14F mutations that also carried the 75C dosage compensation eye color reporter . These males had white eyes ( Figure 2F ) . This shows that even MSL complex containing the overly active P864L subunit can only act on the roX1 reporter when full SPT5 levels are present . Taken together , these in vivo results indicate a role for SPT5 in male X- dosage compensation beyond its general role in transcription of the entire genome in both sexes . ChIP analysis found that SPT5 is enriched over the transcription start site ( TSS ) of most Drosophila genes , with additional binding across the transcribed regions [23] . At the level of polytene chromosomes , SPT5 binds many sites [16] , [24] and colocalizes imperfectly with the MSL complex on the male X-chromosome [6] . To examine this issue in more detail , we raised new SPT5 antibodies . The serum recognized a single band around 135 kDa on SDS-PAGE which is larger than the predicted 119 kDa ( Figure S4 ) . Anomalous migration for SPT5 was reported earlier [16] , [24] . The SPT5 serum recognized many bands on all polytene chromosomes in both sexes ( Figure 3B control panel and Figure S5A ) . Most of the X-linked bands overlapped with MSL1 staining in males , but a few bands stain for only SPT5 or MSL1 ( Figure 3B control panel and Figure S5A ) . The presence of SPT5 only bands is not surprising since several genes on the X escape dosage compensation [7] . A possible explanation for the MSL1 only bands will be presented below . In order to place SPT5 in the dosage compensation pathway , we focused on two of its most intensively studied roles . First , unphosphorylated SPT5 binds to and pauses RNA polymerase over the transcription start site ( TSS ) . Release from this 5′ pause requires phosphorylation by P-TEFb at multiple sites at the C-termini of both RNA polymerase large subunit and SPT5 [25]–[27] . One way SPT5 might aid dosage compensation is if MSL complex stimulated release of the paused RNAPII/SPT5 complex at male X-linked genes . We refer to this as the Pause Release Model . Extra X-linked transcripts would result from clearing the 5′ end of genes freeing them for additional rounds of initiation . After phosphorylation by P-TEFb , SPT5 switches to a positive elongation factor that accompanies RNA polymerase down the gene . The MSL complex might instead enhance the processive action of SPT5 preventing pausing and/or premature termination as RNAPII moved across X-linked genes ( Elongation Model ) . The Pause Release model is less appealing because it calls for MSL action at the TSS , when MSL complex is instead predominantly located farther downstream [6]–[8] . One way to explain this discrepancy would be if the MSL complex only fleetingly interacts with P-TEFb or SPT5 at the 5′ end but then travels across the gene with the elongating RNAPII . If true , this model predicts that loss of SPT5 would lower MSL complex occupancy of the male X-chromosome because MSL complex could not enter the body of genes . Unfortunately , we cannot generate Spt5 null tissue . However , we can approximate that condition by using elongation inhibitors DRB and flavopiridol that block P-TEFb phosphorylation of RNAPII CTD Ser2 and SPT5 that are necessary for pause-release and entry into elongation [23] , [28] . After exposure to these drugs , elongating RNAPII continues to the 3′ end of genes , but new RNAPII is trapped at the TSS , effectively stripping gene bodies of RNAPII and SPT5 . Treating salivary glands with either inhibitor removed actively elongating RNAPII ( Ser2P and Ser5P phosphorylated ) from all chromosomes , but paused RNAPII that is Ser5P phosphorylated was unchanged consistent with previous reports [28] ( Figure 3A and 3C ) . The banded SPT5 signal was strongly reduced on all chromosome arms after DRB or flavopiridol treatment ( Figure 3B and data not shown ) . More importantly , the MSL complex staining pattern remained unchanged following inhibitor treatment ( Figure 3 and data not shown ) . This shows that although MSL complex preferentially binds actively transcribed genes , binding persists for some time after the last polymerase has passed . This finding may explain the few loci bound by MSL complex but not SPT5 in untreated animals ( Figure S5D and S5E ) . These may be dosage compensated genes whose developmentally controlled transcription ceased prior to fixation . Modified histone H3K36me3 is found within active genes with a 3′ bias similar to the MSL complex . This modification may provide one component of MSL targeting specificity through the MSL3 chromodomain [29] , but the issue is contentious [4] . We saw no difference in H3K36me3 staining between flavopiridol treated and mock treated tissue ( Figure S6 ) consistent with earlier reports that these inhibitors only modestly lowered H3K36me3 [30] . To examine MSL binding at a higher resolution than is possible with polytenes , we turned to ChIP analysis of male S2 cells . We measured MSL1 binding to the 5′ and 3′ ends of two highly validated target genes after treatment with flavopiridol . If the scarcity of MSL complex at the 5′ ends of X-linked genes was caused by released RNAPII/SPT5 complex quickly carrying it into the body of genes , we might be able to trap MSL complex over the TSS by treatment with P-TEFb kinase inhibitors . The Pause Release Model predicts that flavopiridol treatment should cause the MSL signal to accumulate at the 5′ end of genes with a corresponding loss at the 3′ end . However , just as was seen with the polytene experiments , flavopiridol treatment did not alter MSL complex distribution as measured by ChIP ( Figure 3D ) . These results argue against the Pause Release Model and instead favor the idea that MSL complex acts upon SPT5 during active elongation . We tested whether the genetic interactions observed between SPT5 and dosage compensation might arise from direct physical contacts . Early attempts to purify intact MSL complex did not recover SPT5 as a partner suggesting that if such interactions occur , they are transient [31] , [32] . Dosage compensation in Drosophila is thought to have recruited an ancestral chromatin modifying complex found in most animals by evolving a new targeting strategy to the male X . If true , perhaps any SPT5-MSL interaction predates Drosophila dosage compensation and would be found in the most phylogenetically conserved regions of the complex . We tested the ancient PEHE domain of MSL1 that recruits MSL3 and MOF and forms a functionally critical core of the complex [13] . We asked whether purified subdomains of SPT5 ( Figure 4A ) could specifically pull down isolated MSL1 PEHE motif . We found specific binding between MSL1 PEHE and the N-terminal ( N ) and middle fragment ( M ) SPT5 fragments ( Figure 4B ) . The N fragment contains the NusG-like domain that interacts with the RNAPII clamp domain to encircle the template DNA and makes RNAPII processive [33] , [34] and one KOW motif ( Figure 4A ) . The M segment contains additional KOW domains . KOW domains found in other proteins bind either protein or RNA partners [35] . We failed to detect any interactions with the C-terminal region that is phosphorylated at multiple sites by P-TEFb . Although this analysis does not exclude additional contacts between other MSL subunits or roX RNAs with regions of SPT5 in vivo , the data show that SPT5 and MSL complex have the ability to interact via MSL1 PEHE . Although we focused our analysis on Spt5 , we wondered if the other modifier mutations found in the screen might identify a new class of factors needed for dosage compensation . We were able to map a few modifiers to previously characterized genes . In the case of deficiencies , we tested whether point mutations of candidate genes could recapitulate the effect of deficiency . That approach showed that Chromator was the relevant gene that dominantly suppresses the MSL complex dependent reporter expression in Df ( 3L ) BSC21 . CHRO is a chromodomain protein that localizes specifically to the interband regions and is implicated in maintaining chromosome structure [36] , [37] . Importantly , it copurifies with the MSL complex [31] underscoring the validity of our genetic approach to search for factors involved in dosage compensation . Additionally , we found that a complementation group from the EMS mutagenesis screen fell within the Df ( 2R ) vg-C ( Figure S1B ) . This interval contained three strong candidates , Spt4 , iswi , and Sin3A . Complementation tests eliminated iswi . In eukaryotes , SPT5 usually acts in a complex with SPT4 . No point mutations in Spt4 have been reported in flies , but the gene is not essential in yeast [38] . All the new EMS mutations instead failed to complement a known lesion in Sin3A [39] , the first indication that the SINA/RPD3 histone deacetylase complex may play a role in dosage compensation . No deficiency removes Spt5 so this region was not covered in our deficiency screen .
It is possible that dosage compensation in Drosophila is entirely a consequence of the known histone modifications carried out by its subunits , H4K16ac ( MOF ) and H2BK34ub ( MSL2 ) . However , if additional factors are required , new approaches may be needed to identify them . Biochemical purification is challenging due to the very large size of the MSL complex , the presence of the noncoding roX RNAs , and the fact that active MSL complex is tightly associated with transcribed chromatin . Extraction methods strong enough to release soluble MSL complex from chromatin may destroy critical contacts with key partners . Genetic approaches also face limitations . If an important partner performs additional functions beyond dosage compensation , mutations would likely be lethal to both sexes masking its interaction with the MSL complex . We developed an unbiased forward genetic screen able to detect subtle changes in MSL activity that are not large enough to prevent dosage compensation of the male X , but sufficient to alter a sensitive eye pigmentation reporter . This screen implicated Spt5 , a universally conserved transcription processivity factor for RNAPs , in the MSL pathway [27] , [33] , [40] , [41] . The validity of our approach is illustrated by the identification of mutations in known components involved in the process such as msl1 , mle and Chro [13] . The value of a genetic approach to detect protein interactions that may only be stable on actively transcribed chromatin is evident . The Drosophila protein interaction map ( DPiM ) identified dozens of proteins that are candidate interactors with SPT5 but surprisingly found no stable contacts with subunits of either RNA polymerase II or P-TEFb , the most highly validated partners known from other studies . This search also found no contacts with MSL subunit [42] , [43] . The technical difficulty most likely rests with the problem of isolating an enormous complex of many megadaltons tightly tethered to DNA . Multiple lines of evidence support a role for SPT5 in dosage compensation . The effect of Spt5 mutations on the white eye color reporter was entirely dependent upon the adjacent roX1 locus that can recruit soluble MSL complex to any location in the genome . Spt5 mutations had no effect on white or miniwhite gene expression when not linked to roX1 . Spt5 mutations acted on all mosaic roX1 reporter transgenes regardless of the chromatin environment surrounding the inserts . Interactions of Spt5 mutants and gain of function msl1 alleles suggest that SPT5 acts between MSL complex and RNA polymerase . Mutations in Spt5 selectively reduced male viability under limiting roX RNA conditions in a manner comparable to the effect of mle mutations . Additionally mutations in Spt5 partially suppressed the toxic effects of ectopic dosage compensation in females . An independent screen of the Drosophila deficiency collection showed that the Spt5 phenotype is rare . Removing one allele of almost any transcription related factor had no effect on the eye pigmentation levels of mosaic roX1 reporters arguing that dosage compensation is particularly sensitive to SPT5 protein levels . Finally , we found that the most ancient and conserved segment of the MSL1 protein physically binds to two different regions of the SPT5 protein consistent with the largely overlapping patterns of chromatin occupancy across the body of X-linked genes . In eukaryotes SPT5 along with SPT4 forms the DSIF ( DRB: 5 , 6-dichloro-1-β-D-ribofuranosylbenzimidazole Sensitivity Inducing Factor ) [27] . The highly conserved NusG like domain ( NGN ) docks to RNAP through its interaction with the RNAP clamp domain and closes the cleft where the tightly bent melted DNA template resides preventing RNAP from falling off the template [33] ( Figure 1G ) . The multiple KOW domains may contact either the emerging nascent transcript or other transcription factors . Although some studies indicated that SPT5 acts on a restricted set of genes [44] , genome wide ChIP analysis showed that SPT5 and RNAPII colocalize throughout the genome [23] , [28] . SPT5 arrests RNAPII near the transcription start site as the short nascent transcript emerges from the enzyme [25]–[27] . The highly regulated release from pause is controlled by the P-TEFb kinase phosphorylating multiple sites near the C-terminus of SPT5 and RNAPII CTD [25] , [27] , [45] . While regulated release from pause was originally described using the highly inducible hsp70 gene from Drosophila [46] , it is now recognized as a widespread step in transcriptional regulation [23] , [28] , [47] . Although other factors , such as cMyc , stimulate transcription through pause-release of SPT5 [28] , our results argue against a similar mechanism operating in Drosophila dosage compensation . MSL complex occupancy is lowest around the TSS and does not depend on continuous association with the elongating RNAPII/SPT5 to be enriched within the gene bodies . Instead we propose that the effect of SPT5 on dosage compensation is downstream of MSL complex recruitment ( Figure 5 ) . MSL complex mediated H4K16ac is enriched within the body of genes and drives decondensation of chromatin possibly facilitating easier passage of RNAPII [5] , [48] . It is plausible that within this chromatin domain , SPT5 impacts dosage compensation via its known interactions with SPT6 , which eases RNAPII passage by nucleosomal removal [16] , [24] , [40] , [49] , [50] and thereby improves the elongation rate of RNAPII . Alternatively , the interaction between MSL complex and SPT5 may increase elongation rates of dosage compensated genes on the X by enhancing RNAPII processivity [33] . We hypothesize that passage of a pioneer RNAPII generates certain transcription-specific epigenetic modifications such as H3K36me3 across a gene . These modifications recruit MSL complex from nearby X-linked sequence specific binding sites called Chromatin Entry Sites ( CES ) or High Affinity Sites ( HAS ) [1] . Once recruited , MSL complex is stably bound and stimulates elongation via its transient interaction with the transcribing SPT5/RNAPII ( Figure 5 ) . At least one point of this interaction is via the MSL1 PEHE domain and SPT5 NusG like and KOW domains . While we have focused on the analysis of Spt5 in this report , our genetic approach also yielded additional candidates . So far , we have mapped two of these to Chro and Sin3A . CHRO , a chromodomain protein copurifies with the MSL complex [31] . Interestingly , CHRO recruits and localizes with JIL-1 , a histone kinase that has also been implicated in dosage compensation [36] , [37] , [51] , [52] . The CHRO/JIL-1 kinase complex is thought to maintain chromosomal integrity [36] , [37] . It is conceivable that this complex plays a similar role in maintaining the specialized X-chromatin architecture in male flies . SIN3A , part of the SIN3A/RPD3 histone deacetylase complex is attracted by phosphorylated SPT5 and Ser2 phosphorylated CTD of RNAPII to deacetylate histones in the wake of transcribing RNAPII within the H3K36me3 chromatin domain [53] , [54] . Therefore , phosphorylated SPT5 , in addition to modulating processivity may also play a role in erasing transcription dependent acetylation via recruiting the SIN3A/RPD3 complex . This serves to suppress spurious transcription initiation from cryptic promoters within the coding region [53] , [54] . Alternatively , the SIN3A/RPD3 complex may play a role in MSL complex recruitment to the GAGA element rich MSL recognition elements ( MRE ) sequences via its interaction with GAGA factor [55] . Further enquiry into the specific role played by these newly identified factors will result in an improved understanding of the mechanism of dosage compensation . An independent RNAi screen using an MSL complex dependent luciferase expression as a reporter in S2 cells also identified a role for CHRO and SIN3A in dosage compensation [56] . Recovering overlapping cofactors from rather different genetic screens increases confidence that these strategies are identifying authentic components of the dosage compensation pathway . However , SPT5 was not found using the RNAi screen . This is not surprising since a general transcription factor such as SPT5 probably affects the expression of the normalizing control used in luciferase reporter assays . This again highlights the usefulness of an in vivo genetic strategy . Our results provide direct in vivo support for the elongation model of dosage compensation by linking the SPT5 elongation factor to the MSL complex [5] . The finding that the Drosophila males have ∼1 . 4 fold more transcriptionally engaged RNAPII at the distal ends of X-linked genes as compared to autosomes also supports the idea of increased elongation [9] . Conversely , a recent report that compared global RNAPII occupancy in males and females found an increase in RNAPII levels across the entire body of the gene including the promoters on male X-linked genes [10] . This observation raises the possibility that dosage compensation may operate at the level of transcription initiation . A caveat of this study is that only a subset of X-linked genes ( n = 242 ) had detectable RNAPII within the body of genes , possibly due to technical difficulties in the ability to detect elongating RNAPII . An alternate explanation for the results is that lowered RNAPII pausing [9] , [10] and increased elongation improves RNAPII recycling from the 3′ to 5′ end of genes possibly via gene looping interactions and may be reflected in ChIP seq studies as an increase in RNAPII levels at the promoter . Moreover , Conrad et al postulate that H4K16ac at promoters is the key to dosage compensation . However , H4K16ac at promoters occurs both on male autosomes and all chromosomes in females and is not specific to the male X [48] , [57] . On the other hand , H4K16ac within gene bodies is a unique feature of transcribed genes on the male X-chromosome and is therefore an attractive candidate to drive dosage compensation by improving RNAPII passage across the chromatin fibre during elongation . Mammals also contain a version of the MSL complex composed of MSL1 , 2 , 3 and MOF , but apparently lacking a large noncoding RNA component and RNA helicase [58] , [59] . Like flies , the human MSL complex , is bound within the bodies of genes with a distinct 3′ bias , acetylates histone H4K16 in the body of genes and increases transcription by approximately two fold [58]–[60] . Our results linking the most conserved domain of MSL1 with the conserved transcription elongation factor SPT5 in flies indicate that mammalian MSL complex is likely to also act upon transcription elongation .
Mutagenesis was performed as described [13] . Detailed mutagenesis scheme and protocol is included in Text S1 . For the deficiency screen the transgenic lines [w+ GMroX1-58D] , [w+ GMroX1-60F] , [w+ GMroX1-69C] , [w+ GMroX1-75C] , [w+ GMroX1-99F] and [w+ GMroX1-102C] were used . The full genotype of ΔroX1 , roX2 stock is y w roX1ex6 Df ( 1 ) roX252 [w+ cos4Δ4 . 3] [19] . Plasmids for bacterial expression of MBP fusion SPT5 protein fragments , SPT5-N ( aa 112–393 ) , M ( aa 389–733 ) and C ( aa 732–1054 ) were a kind gift from Dr . John Lis [16] . Antibodies were raised by Cocalico Biologicals , Pennsylvania . Polytene squashes were prepared as described in [61] . Primary antibodies were rabbit anti-MSL1 antibodies ( 1∶50 ) , guinea pig anti-SPT5 antibodies ( 1∶100 ) , mouse H5 monoclonal anti-Ser2P RNAP ( Covance , 1∶30 ) , mouse H14 monoclonal anti-Ser5P RNAP ( Covance , 1∶50 ) and rabbit anti-H3K36me3 ( Invitrogen , 1∶50 ) . Appropriate secondary antibodies were used in combinations that allowed for dual protein localization . Detailed protocol can be found in Text S1 . For inhibitor treatment 500 nM flavopiridol or 100 µM DRB was used . Briefly S2 cells were crosslinked with 1% formaldehyde and nuclei extracted in 15 mM HEPES , 5 mM MgCl2 , 0 . 2 mM EDTA , 0 . 5 mM EGTA , 10 mM KCl , 350 mM Sucrose , 0 . 1% Tween 20 , 0 . 5 mM PMSF , 1 mM DTT . Chromatin was sheared to 300–700 bp fragments and pulled down with anti MSL1 antibodies ( Gift from M . Kuroda ) . After several washes eluted chromatin was used . For real time PCR ( ABI 7900 qPCR model ) , SYBR green master mix ( ABI ) , 1 µM primers and 1 µl of input and ChIP DNA was used . Primer sequences are provided in Text S1 . MBP-SPT5N , MBP-SPT5M , MBP-SPT5C , the unrelated protein MBP-MCP ( MS2 phage coat protein ) , GST-MSL1 C-terminal domain fusion protein and GST were expressed and isolated from bacteria . Equivalent molar concentrations of MBP proteins bound to amylose beads were incubated with GST proteins . After three washes with 150 mM NaCl , 0 . 1% NP40 and 20 mM Tris for 10 mins at 4°C , proteins were eluted by boiling in SDS-loading buffer and separated on 8% SDS-polyacrylamide gels . Westerns were performed as described [13] . We used anti-GST antibodies ( Sigma ) to detect GST and affinity purified guinea pig anti-SPT5 sera to detect MBP-bound SPT5 . The proteins were visualized by using appropriate HRP conjugated secondary antibodies ( Jackson Immuno ) and lunimol reagent ( Santa Cruz ) .
|
Drosophila males hypertranscribe most of the genes along their single X chromosome to match the output of females with two X chromosomes . It had been difficult to imagine how the MSL dosage compensation complex could impose a modest , but essential , ∼two-fold increase by interacting with hundreds of different factors that control transcription initiation for such a diverse collection of genes . An alternative model proposed that dosage compensation instead acted at some step of transcription elongation common to all genes . We performed a genetic screen for mutations that subtly reduce dosage compensation and recovered mutations in the Spt5 gene that encodes a universally conserved elongation factor . SPT5 closes the RNA polymerase II clamp around the DNA template to prevent pausing or premature termination . We find that the dosage compensation complex genetically and physically interacts with SPT5 on actively transcribed genes providing direct molecular support for the elongation model of dosage compensation .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"molecular",
"cell",
"biology",
"genetic",
"screens",
"gene",
"expression",
"genetics",
"gene",
"regulation",
"epigenetics",
"biology",
"molecular",
"genetics",
"chromatin",
"genetics",
"and",
"genomics",
"dna",
"transcription",
"histone",
"modification"
] |
2012
|
Mutations in the Transcription Elongation Factor SPT5 Disrupt a Reporter for Dosage Compensation in Drosophila
|
Timings of human activities are marked by circadian clocks which in turn are entrained to different environmental signals . In an urban environment the presence of artificial lighting and various social cues tend to disrupt the natural entrainment with the sunlight . However , it is not completely understood to what extent this is the case . Here we exploit the large-scale data analysis techniques to study the mobile phone calling activity of people in large cities to infer the dynamics of urban daily rhythms . From the calling patterns of about 1 , 000 , 000 users spread over different cities but lying inside the same time-zone , we show that the onset and termination of the calling activity synchronizes with the east-west progression of the sun . We also find that the onset and termination of the calling activity of users follows a yearly dynamics , varying across seasons , and that its timings are entrained to solar midnight . Furthermore , we show that the average mid-sleep time of people living in urban areas depends on the age and gender of each cohort as a result of biological and social factors .
The daily activity of people varies across space and time from place to place , date to date , and hour to hour as a result of biological , societal , economic , and environmental factors , shaping the society where they live . Roughly speaking , each day humans do certain activities at specific times . There are many environmental factors ( cues or ‘zeitgebers’ ) involved in the entrainment of this clock , but as pointed out by Roenneberg et al . [1] , the most dominant is light and is associated with the light-darkness cycle determined by the daily rhythm of daylight . However , mainly in places not close to the equator , the timing and duration of daylight is subject to noticeable seasonal variation due to the yearly movement of the Earth around the Sun , and these changes have a direct influence on the kind and timing of different human activities . On the other hand , humans living in urban areas are also immersed in an environment full of cues that could influence the entrainment of the circadian clock . Artificial lighting , social practices and schedules ( work and school hours , workdays vs weekends ) , particularly for those living in big urban areas , could have a noticeable influence on the entrainment process . Social conventions impose characteristic schedules on individuals , and , at the population level we can expect people in urban areas to have periods of high activity between morning and evening , and periods of low activity ( resting ) during the night . The length and timings of human activity periods , specifically in urban areas , has important consequences for human health [2–5] , economy and power consumption [6] , and public transportation efficiency [7] . The human sleep wake cycle ( SWC ) , and its dynamics in particular , has been studied in recent years to understand the processes and cues that govern it [8] . Generally speaking , most research on human SWC has focused on experiments with small groups under controlled conditions [9 , 10] , or questionnaire studies [1 , 11–14] ( mainly using the Morning-Eveningness Questionnaire ( MEQ ) [15] and the Munich Chronotype Questionnaire ( MCTQ ) [16] ) . The use of these tools for studying SWC has proved to be very fruitful and effective , though having some limits on the domain of applicability [14] . In contrast , the ever-increasing availability of information communication technologies ( ICT ) combined with researchers’ ability to access large-scale ICT-generated datasets ( ‘Big data’ ) has made possible the study of human behaviour using a variety of reality ( data ) mining techniques . In particular , there are a number of examples where mobile phone datasets have been analyzed to study social networks [17–20] , sociobiology [21 , 22] , mental health [23] , mobility [24–27] , as well as social behaviour of cities [28 , 29] . Over the past decade or so , the existence and accessibility of these large population-level datasets , has allowed scientists to study intrinsic human behavioural and socio-evolutionary patterns in unprecedented and complementary ways , compared to other research approaches . Recently , datasets of mobile phone usage have also been used to study circadian rhythms , by analyzing individual’s mobile phone usage from the data captured by sensors [26 , 30–35] , or people’s communication patterns from their call detail records ( CDRs ) [31–33] . For example , one study used the mobile phone screen on-off sensor data to examine the sleep wake cycle of nine individuals , finding that most of the individuals varied their sleep time patterns between weekdays and weekends , as well as showing seasonal changes in their mid-sleep time [30] . In another study using mobile phones calls and text messages of a small number of individuals , it was shown that individuals can be classified as having morning type or evening type activity levels [31] . In our previous related work [33] , we quantified the resting periods of people from their mobile phone calling activity , showing that there is a counterbalancing effect between the afternoon and night time resting periods , due to an interplay between ambient temperature and sunlight . The use of CDRs as a tool for investigating the sleep/wake circadian rhythm , is in our view a promising new line of research as of that complements the other research approaches especially the large scale survey-based studies , pioneered by Roenneberg et al [11–13] . In this study , we apply reality mining techniques to users’ call records in a mobile phone communication network to study the dynamics of the users’ calling patterns by focusing on the periods of low activity , i . e . when almost no calls are made . Users of the mobile phone network typically have specific time periods during which their calling activity ceases , and we may assume that the SWC is bounded inside this period of inactivity . We observe that the daily calling activity time displays an interesting dynamics across the year through seasons and along different geographical zones . By studying these patterns we can gain insights into human activity patterns , and the SWC , in particular . Interestingly , the calling activity pattern changes with the day of the year and it is found to depend also on the geographical location ( latitude and longitude of the mobile phone user ) . From the circadian clocks involved in the daily rhythms of human societies , only those entrained to solar-based events depend also on the geographical location and on the day of the year . In this work , we use mobile phone calling activity at the population level to study how the onset and termination of the urban human activity in different cities is synchronized with the East-West progression of the Sun . Also , we analyzed the annual progression of the onset and termination of the calling activity , finding that they show a strong seasonal variation . We note that this behavior is similar to the annual dynamics of solar midnight , inferring that solar midnight is an important cue entraining the human circadian clock . Finally , we determine the mid-time of the period of low calling activity , which is bounded between the termination of calling activity each day and its onset on the next day . We interpret this mid-time to correspond to the mid-sleep time , and show that it is strongly dependent on the age and gender of the individuals in the population .
The mean time of the first call tF and of the last call tL of people in a city can be influenced by environmental , social , and economic factors , and their possible daily value could be distributed completely at random . However , we find that during the year and at different latitudes , despite the different factors influencing the shape of the distribution Pall , the onset and termination of calling activity follows a consistent pattern , and this characteristic behaviour allows us to compare the calling activity pattern of cities lying at different latitudes . If the onset or termination of the urban calling activity is socially driven , with fixed times for specific activities ( like office working hours from 9:00am to 6:00pm ) , one could expect that cities lying in the same time zone and at the same latitudes have similar calling activity timings ( onset and termination ) . However , we find that the onset and termination of calling activity synchronizes with the East-West sun progression , in such a way that cities lying in western locations start ( and terminate ) their calling activity after cities at eastern locations , with a delay difference corresponding to the time difference between their local meridians . In Fig 2A and 2B we show tL and tF for 5 different cities lying inside a latitudinal band centered at 42°N±40′ . The region including the 5 cities spans a longitudinal angle of 10 . 8° , and by taking one of the cities as a reference , other cities are located at −7 . 8° , −4 . 7° , −3 . 7° , and +3 . 0° from the reference city marked here with 0 . 0° . Then we compare the actual distributions PL and PF of the time of the last call and of the first call , respectively , for the 5 cities in the same latitudinal band , and find that PL and PF for western cities seem shifted to later times . However , when the distributions are shifted by an amount of time corresponding exactly with the time difference between the local meridian of the corresponding city and the reference city , the distributions visibly collapse onto each other , as can be seen in Fig 2C and 2D . In this case , the time shifts are +31 . 2 , +18 . 8 , +14 . 8 , and -12 minutes for the cities located at -7 . 7° , -4 . 7° , -3 . 7° , and +3° from the reference city at 0° , respectively . The distribution collapse shown in Fig 2 is obtained by introducing a time shift corresponding to the sun transit differences between cities . In order to quantify the exact delay between the distributions , we calculate the required time shift that should be introduced between the calling distributions to minimize the Kullback-Leibler divergence DKL between them ( see the Methods section ) . This measure is indicative of the similarity between the distributions , and is minimized when they are identical . We extend this analysis to include data from 30 cities , each one lying in one of the four latitudinal bands centered at 37°N ( 10 cities ) , 39 . 5°N ( 5 cities ) , 41 . 5°N ( 7 cities ) , and 42 . 5°N ( 8 cities ) . For each band , we choose one city lying near the mid point of the band as the reference , and calculate for all the cities in the band the average time shift between them and the reference city . This is done for each day of the week , averaging over 52 weeks of the year 2007 . The results are shown in Fig 3 , and it can be seen that the time shift that minimizes the divergence between the distributions corresponds to the delay between their local sun transit times . This synchronization appears stronger for the termination of the calling activity ( represented by the distributions PL ) . As this pattern is consistently present in all of the four analyzed latitudinal bands , we conclude that it is a general behaviour of the population living in the cities . This result is consistent with those reported by Roenneberg et al . [12] , obtained from MCTQ studies of people in Germany , distributed over a region that is 9° wide longitudinally . In their work , they take into account the population of the city by defining three population size categories , i . e . less than 300 , 000 inhabitants , between 300 , 000 and 500 , 000 inhabitants , and more than 500 , 000 inhabitants , while we classify each city of more than 100 , 000 inhabitants according to its latitudinal coordinate . Grouping the cities into latitudinal bands , we found a consistent entrainment to the East-West progression of the Sun , regardless of the population size of each city . This result implies that the termination ( last call of the day ) and onset ( first call of the next day ) of calling activities in cities at similar latitudes follow an external cue driven by solar events , and the time difference in these solar events between two different cities is reflected in the timings of their calling activity . We have shown that the cities located at the same latitude but at different longitudes have periods of low calling activity with different onset and termination times ( Figs 2 and 3 ) . This shift coincides with the difference between their local sun transit times , i . e . when the sun crosses the meridian of the city . This observation raises the question as to what external daily event induces such synchronization . As the delays correspond to the time period between the local sun transit times of the cities , it seems plausible to think that the sun functions as a cue for this entrainment . At the latitudes where the studied cities are located , the time difference between the sunset in the summer and in the winter is around 3 hours , if daylight saving is not taken into account , and the same holds for the time difference between sunrises . In contrast , the time difference between the mean time of the last calls between summer and winter is at most one hour [33] . However , there is a clear synchronization between the sun transit time and the timings of calling activity . This means that there should be an external clock functioning as a cue . On the other hand , from a biological perspective , the time when the secretion of melatonin reaches its maximum [37] lies close to midpoint between sunset and sunrise ( i . e . solar midnight ) , once the night is as dark as possible . It has been proposed that the mid-sleep time coincides with the time corresponding to maximum melatonin secretion [38 , 39] , and if the solar midnight shifts through the year , the time for the maximum melatonin secretion should follow a similar pattern , as well as the entrained mid-sleep time . In their study , Allenbradt et al . [40] , using the MCTQ approach , have reported that mid-sleep time ( on free-days ) changes from one season to another . In some of the studied populations , they found that there is a small but significant difference in the average mid-sleep time between the days when Daylight Saving Time is applied and other days . This lends support to our assumption that if the mid-sleep time shifts in response to seasons , the timings of the calling activity should be influenced by its variation . In such a case , when the human mid-sleep time occurs at later hours , the timings of the calling activity for the following days should also occur at later hours . In other seasons , when the mid-sleep time occurs earlier , the activity timings should also be shifted towards earlier hours . If this is the case , then solar midnight should be functioning as the cue to which the calling activity timings are entrained . The activity pattern is a consequence of the interplay between seasonal and geographical factors , as well as social and societal activities like work and/or school , transportation , eating and leisure activities . However , the latter require specific timings during the day , not necessarily controlled by the sleep/wake cycle . We have shown elsewhere [33] that the total period of low calling activity ( that is , the period between the termination and the onset of the calling activity ) is strongly correlated with the duration of daylight , showing seasonal changes similar to the mid-sleep time . In order to find any possible synchronization between the onset ( and termination ) of calling activity and solar midnight , we calculate the average of the mean times of the last call t ¯ L and that of the first call t ¯ F , for three sets of cities located at the latitudinal bands ϕ = 37°30′N ( seven cities ) , 40°20′N ( six cities ) , and 43°0′N ( eight cities ) . We compare t ¯ L , and t ¯ F with the yearly evolution of the solar midnight in a reference city within a given latitudinal band ( see Fig 4 ) . A detailed description of how t ¯ L and t ¯ F are calculated can be found in the Methods section . It can be seen that only t ¯ L resembles to some extent the dynamics of the solar midnight , with their two minima and at least one of their maxima occurring around the same days of those of solar midnight , although the relative amplitudes are not in correspondence . In addition , the discontinuities introduced by the daylight saving is visible in all the graphs , suggesting that the timings of the calling activity are not solely influenced by the socially-driven time , but instead are synchronized with an external ( astronomical ) clock . The period of low calling activity is bounded by the mean times of the last call during the night and of the first call in the morning . The duration of this period changes across seasons [33] and is strongly influenced by the length of the day ( or conversely by the length of the night ) . The mid-time of this low calling activity period should correspond to the average time of human low activity , i . e . when the majority of the urban population is sleeping . In chronobiology studies , the mid-sleep time , corresponding to the time when human sleep is in the middle of its cycle , has been found to vary with the age and gender of the individuals [11 , 41] . Despite the fact that each individual has a distinctive sleep-wake cycle , with a chronotype ranging from advanced sleep period ( morningness ) to delayed sleep period ( eveningness ) [42] , at the population level a characteristic mid-sleep time can be consistently calculated , taken simply as the average of individual mid-sleep times . From the mean times of the last call of the day , tL and of the first call tF of the next day , we define the period of low calling activity TLCA as the elapsed time between tL and tF , as a measure of the time when cities cease their activity . In Fig 5a , the width of the low activity period TLCA of the most populated city in the dataset is shown , for 4 different days of the week ( Tuesdays , Fridays , Saturdays and Sundays ) , as a function of the subscribers’ age and gender . There is a noticeable change of about 3 hours , moving from the age cohort of 20 to that of 40 year olds . After that rather abrupt increase , especially for Fridays and Saturdays , TLCA slightly decreases , reaching a local minimum value for the age cohort of 50 year olds , and then it increases again to reach the highest value at the age of 78 years . For the analyzed weekday ( Tuesday ) as well as for Sunday , TLCA increases almost monotonically with the cohort age , showing a small plateau for age cohorts between 45 and 58 . We have also tracked the midpoint of the inactivity period , defined as the mid-time between tL and tF . Due to its similarity with the average time in the middle of the sleeping period [41] , we interpret this minimum calling activity time as the mid-sleep time tmid , calculated simply as tmid = ( tL + tF − 24 ) /2 . Both quantities are found to depend on the age and gender of each cohort , as can be seen in Fig 5b . We find that , for certain age groups ( from 18 to 32 years old , and from 43 to 80 years old ) tmid occurs at a later time for women as compared to men , while in the age group of 33 to 42 years old , tmid for the men occur later . This finding differs somewhat from the reported mid-sleep times ( on free days ) in the chronotype questionnaire study based on the MCTQ [11 , 13] , where males show a later mid-sleep time for age cohorts younger than 38 years old . Also , there is a strong dependence on age , with younger age cohorts ( 20–30 year old ) having later tmid , i . e . around 30 minutes after that of the oldest age cohort ( 70-80 years old ) . This observation is in accordance with the observed chronotypes [41] , which are attributed to biological factors or internal clock being regulated by neuronal and hormonal mechanisms . We also found an unexpected rise of tmid for the age cohort of 45–65 year old individuals , which we suspect is entirely of social origin . Hence it seems that both biological and social factors play a role in changing tmid , i . e . shifting the period of low activity to later hours . In addition , we find that tmid varies across days of the week . On Fridays and Saturdays tmid occurs at a later hours compared with the other days . Similarly , the age cohort with the latest mid-sleep time tmid is different for different days of the week . On Saturdays , individuals in the age group 30 to 45 years old have the latest tmid , while for the other days of the week it is the 20–25 years old cohort which shows the latest mid-sleep time . The results of TLCA and tmid for the most populated city are also and consistently found in the next 5 most populated cities , as shown in the Supplementary Material ( S1 and S2 Figs , respectively ) .
In this study , we have found that the onset and termination of the period of low calling activity for people in cities at about the same latitude but at different longitudes are shifted according to their relative longitudinal separation . Cities westward from the easternmost analyzed city stop their activity later in line with the time delay of the sun transit time . This result suggests that a solar event acts as a cue for the circadian rhythm of the period of low calling activity with the SWC bounded inside . This result is consistent with those reported by Roenneberg et al . [12] , although strictly speaking the two studies cannot be compared directly as the focus of our study is on variation by latitude and theirs was on variation by population size of cities . In addition , we found that the seasonal variation of the termination of calling activity resembles the annual variation in solar midnight ( or solar noon ) . However , when the annual behaviour of activity termination is compared with other characteristic solar events like the sunrise and sunset , it appears to have a different functional form with different number of maxima and minima with different dates . Although , it seems likely that solar midnight ( or solar noon ) acts as a cue in the synchronization of the termination of the calling activity , further research is needed to confirm this . At the individual level , knowledge of the mid-sleep time and sleep duration allows the determination an individual’s chronotype [16] . However , at the population level , we could determine from the calling distributions the characteristic variation in the sleep duration and mid-sleep time as a function of the group age . The observed overall trends are in line with the earlier findings [41] and reveal an increase in the sleep duration and decline in the mid-sleep time with age . Several other intricacies are also evidenced at closer inspection . Firstly , the aspect of ‘social jetlag’ [43] , defined as the difference between the mid-sleep times on free days and that of work days , becomes apparent across all age groups . Interestingly , although social jet-lag is expected to give rise to extended sleep duration on free days as a compensatory effect , for young adults ( 20–25 ) we find that the sleeping periods are comparatively less on free days ( Friday and Saturday nights and the following mornings ) . Therefore , sleep deprivation is likely to be at a maximum for this age range . Second , previous observations suggest a monotonic decrease in the mid-sleep time from around 20 years of age , which can be attributed to endocrine factors [41] . In contrast , we observe a reversal in trend of the mid-sleep time such that at the age of 45 years it starts rising till 55 years of age , after which it decrease again .
Calling behavior varies seasonally , particularly the mean value and the width of the distributions of the first and last call vary across the year , being pushed towards the afternoon during winter and towards midnight during the summer . In spite of this seasonal variation , for a given day the calling distributions of different cities have similar shapes , and we exploit this similarity to calculate the delays between them to identify the temporal shifts of the distributions . The Kullback-Leibler divergence [45] is a measure of similarity between two distributions , commonly used in statistical analysis , for example when comparing one distribution obtained from data and another generated by a model . It reaches zero , its minimum possible value , when the distributions are identical , and it increases in value as the distributions become more and more dissimilar . In the case of the calling activity of different cities , the distributions are not identical but have a very similar shape . Applying Kullback-Leibler divergence to a pair of these distributions , it would reach a minimum value when these distributions overlap most , falling on top of each other and collapse to one . Thus , if we measure the amount of time one distribution should be shifted in order to minimize its divergence from the second distribution . The time shift would correspond the actual time delay between them . In order to quantify the actual time shift between the distributions PL of last calls for cities lying along different Longitudes , we proceed as follows . First , for all the cities within the band , we calculate all the distributions PL ( t , d ) between January 2nd and December 31st . For each day d , we fix PL ( t , d ) 0° of the city labeled ‘0°’ as the reference distribution , and for every other city c in the band , we compared the reference PL ( t , d ) 0° with time-shifted versions PL ( t + nΔ , d ) c of the distribution PL ( t , d ) c , with −5 ≤ n ≤ 8 and Δ = 5 min , to find the time shift n*Δ that minimizes the divergence DKL between them . Here , DKL is the Kullback-Leibler divergence measure , defined as DKL ( P , Q ) = ∑i Pi log ( Pi/Qi ) , with P , Q being the two discrete distributions . Once we find for each city the set {n*Δ} with all the time-shifts across the year , we calculate its average time-shift 〈n*Δ〉 , and plot it for all the cities in the band in the right column of Fig 3 . As the time for the mean time of the last call is different for different days of the week [33] , the average is calculated separately for each day of the week . We apply the same procedure for the time of the first call distributions PF , and the results are shown in the left column of Fig 3 . In order to find if there is any relation between tL and tF and the solar midnight , we have chosen 7 , 6 and 8 cities , lying in the latitudinal bands centered at ϕ = 37°30′ N , 40°20′ N , and 43°0′ N , respectively . For each city , we shift its corresponding distributions in accordance with its longitudinal difference to collapse all into one . Then we calculate the average mean time of the last call , t ¯ L ( d ) = 〈 t ¯ L ′ ( d , c ) 〉 , where , t ¯ L ′ ( d , c ) denotes the mean time of the last call for the shifted distribution for a city c belonging to the analyzed band during the day d , and 〈⋅〉 denotes the average over all cities lying within the band . Similarly , we calculate the average mean time of the first call t ¯ F ( d ) for the given latitudinal band . The quantities t ¯ L ( d ) and t ¯ F ( d ) are compared with the time at which the solar midnight occurs in the reference city of each band . It should be noted that in the original graphs there are days of national holidays and local festivities that introduce drastic pattern changes , which we filter out to construct the final graphs .
|
For humans living in urban areas , the modern daily life is very different from that of people who lived in ancient times , from which todays’ societies evolved . Mainly due to the availability of artificial lighting , modern humans have been able to modify their natural daily cycles . In addition , social rules , like those related to work and schooling , tend to require specific schedules for the daily activities . However , it is not fully understood to what extent the seasonal changes in sunrise and sunset times and the length of daylight could influence the timings of these activities . In this study , we use a new approach to describe the dynamics of human resting periods in terms of mobile phone calling activity , showing that the onset and termination of the resting pattern of urban humans follow the east-west sun progression inside the same timezone . Also we find that the onset of the low calling activity period as well as its mid-time , are subjected to seasonal changes , following the same dynamics as solar midnight . Moreover , with resting time measured as the low activity periods of people in cities , we discover significant behavioural differences between different age and gender cohorts . These findings suggest that the length and timings of the human daily rhythms , still have a sensitive dependence on the seasonal changes of the sunlight .
|
[
"Abstract",
"Introduction",
"Results",
"Discusion",
"Materials",
"and",
"methods"
] |
[
"ecology",
"and",
"environmental",
"sciences",
"medicine",
"and",
"health",
"sciences",
"engineering",
"and",
"technology",
"sleep",
"cell",
"phones",
"physiological",
"processes",
"probability",
"distribution",
"mathematics",
"longitude",
"chronobiology",
"geography",
"cartography",
"probability",
"theory",
"communication",
"equipment",
"circadian",
"rhythms",
"urban",
"areas",
"equipment",
"earth",
"sciences",
"geographic",
"areas",
"physiology",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"latitude",
"daylight"
] |
2017
|
Tracking urban human activity from mobile phone calling patterns
|
Pattern formation during development is a highly dynamic process . In spite of this , few experimental and modelling approaches take into account the explicit time-dependence of the rules governing regulatory systems . We address this problem by studying dynamic morphogen interpretation by the gap gene network in Drosophila melanogaster . Gap genes are involved in segment determination during early embryogenesis . They are activated by maternal morphogen gradients encoded by bicoid ( bcd ) and caudal ( cad ) . These gradients decay at the same time-scale as the establishment of the antero-posterior gap gene pattern . We use a reverse-engineering approach , based on data-driven regulatory models called gene circuits , to isolate and characterise the explicitly time-dependent effects of changing morphogen concentrations on gap gene regulation . To achieve this , we simulate the system in the presence and absence of dynamic gradient decay . Comparison between these simulations reveals that maternal morphogen decay controls the timing and limits the rate of gap gene expression . In the anterior of the embyro , it affects peak expression and leads to the establishment of smooth spatial boundaries between gap domains . In the posterior of the embryo , it causes a progressive slow-down in the rate of gap domain shifts , which is necessary to correctly position domain boundaries and to stabilise the spatial gap gene expression pattern . We use a newly developed method for the analysis of transient dynamics in non-autonomous ( time-variable ) systems to understand the regulatory causes of these effects . By providing a rigorous mechanistic explanation for the role of maternal gradient decay in gap gene regulation , our study demonstrates that such analyses are feasible and reveal important aspects of dynamic gene regulation which would have been missed by a traditional steady-state approach . More generally , it highlights the importance of transient dynamics for understanding complex regulatory processes in development .
Biological systems depend on time . Like everything else that persists for more than an instant , there is a temporal dimension to their existence . This much is obvious . What is less obvious , however , is the active role that time plays in altering the rules governing biological processes . For instance , fluctuating environmental conditions modify the selective pressures that drive adaptive evolutionary change [1 , 3–5] , time-dependent inductive signals or environmental cues trigger and remodel developmental pathways [6 , 7] , and dynamic morphogen gradients influence patterning , not only across space but also through time [8–16] . In spite of this , many current attempts at understanding biological processes neglect important aspects of this temporal dimension [17] . For practical reasons , experimental studies often glance over the detailed dynamics of a process , and focus on its end product or output pattern instead . Similarly , modelling studies frequently restrict themselves to a small-enough time window allowing them to ignore temporal changes in the rules governing the system . Accuracy is sacrificed and the scope of the investigation limited for the sake of simplicity and tractability . Although reasonable , and often even necessary , such simplifications can lead us to miss important aspects of biological regulatory dynamics . We set out to tackle explicitly time-dependent aspects of morphogen interpretation for pattern formation during animal development . As a case study , we use the gap gene network , which is involved in segment determination during the blastoderm stage of early development in the vinegar fly Drosophila melanogaster [18] . Activated by long-range gradients of maternal morphogens Bicoid ( Bcd ) and Caudal ( Cad ) , the trunk gap genes hunchback ( hb ) , Krüppel ( Kr ) , giant ( gt ) , and knirps ( kni ) become expressed in broad overlapping domains along the antero-posterior ( A–P ) axis of the embryo ( Fig 1 ) . The establishment of these domains is fast and dynamic . Subsequently , gap gene domain boundaries sharpen and domains in the posterior region of the embryo shift anteriorly over time ( Fig 1 ) . Towards the end of the blastoderm stage , gap gene production rates drop and domain shifts slow down . The blastoderm stage ends with the onset of gastrulation . The gap gene system is one of the most thoroughly studied developmental gene regulatory networks today . For our particular purposes , we take advantage of the fact that it has been extensively reverse-engineered using data-driven modelling . This approach is based on fitting dynamical models of gap gene regulation , called gene circuits , to quantitative spatio-temporal gene expression data [19–27 , 29 , 34 , 35] . Dynamical models capture how a given regulatory process unfolds over time . They are frequently formulated in terms of ordinary differential equations ( ODEs ) with parameter values that remain constant over time . Such equations represent an autonomous dynamical system . Central to the analysis of such dynamical systems is the concept of phase space and its associated features ( S1A Fig ) . Phase ( or state ) space is an abstract space that contains all possible states of a system . Its axes are defined by the state variables , which in our case represent the concentrations of transcription factors encoded by the gap genes . Trajectories through phase space describe how a system’s state changes as time progresses . The trajectories of a gap gene circuit describe how transcription factor concentrations change over time . All trajectories taken together constitute the flow of the system . This flow is shaped by the regulatory structure of the underlying network—the type ( activation/repression ) and strength of interactions between the constituent factors—which is given by the system’s parameters . Since these parameters are constant over time in an autonomous system , the trajectories are fully determined given a specific set of initial conditions . Once the system’s variables no longer change , it has reached a steady state . Steady states can be stable—such as attractors with converging trajectories from all directions defining a basin of attraction—or unstable—such as saddles; where trajectories converge only along certain directions and diverge along others . The type and arrangement of steady states , and their associated basins of attraction define the phase portrait of the system ( S1A Fig ) . There exist powerful analytical tools to analyse and understand the phase portrait and the range of dynamic behaviours determined by it . Geometrical analysis of the phase portrait enables us to build up a rich qualitative understanding of the dynamics of non-linear autonomous systems without solving the underlying equations analytically [36] . The application of dynamical systems concepts and phase space analysis to the study of cellular and developmental processes has a long history ( see [37–39] for recent reviews ) . In particular , it has been successfully applied to the study of the gap gene system . Manu and colleagues [22 , 23 , 40] examined the dynamics and robustness of gap gene regulation in D . melanogaster using diffusion-less gene circuits fit to quantitative expression data . These models have a four-dimensional phase space , where the axes represent the concentrations of transcription factors encoded by the trunk gap genes hb , Kr , gt , and kni . The analysis of these phase portraits yields a rigorous understanding of the patterning capabilities of the system . The analysis by Manu et al . [23] corroborated and expanded upon earlier genetic evidence [41] indicating that the regulatory dynamics responsible for domain boundary placement in the anterior versus the posterior of the embryo are very different . In the anterior , spatial boundaries of gap gene expression domains are positioned statically , meaning that they remain in place over time [42] . Stationary boundaries are regulated in two distinct ways [23] . ( 1 ) In the case of the posterior boundary of the anterior gt domain , different nuclei along the A–P axis have equivalent attractors positioned at different locations in phase space ( shift in attractor position ) ; ( 2 ) in the case of the posterior boundary of the anterior hb domain , system trajectories fall into different basins of attraction ( attractor selection ) ( Fig 2A ) . In both of these cases , patterning is largely governed by the position of attractors in a multi-stable phase space . In contrast , gap domain boundaries in the posterior of the embryo shift anteriorly over time [25 , 42] . In this region , the system always remains far from steady state , and the dynamics of gene expression are transient . Therefore , trajectories here are fairly independent of precise attractor positions . The model by Manu et al . [23] shows that posterior gap gene expression is governed by an unstable manifold ( Fig 2A ) . An unstable manifold is the trajectory connecting a saddle to an attractor ( S1A Fig ) . The authors demonstrate that this manifold has canalising properties since it compresses many incoming neighbouring trajectories into an increasingly smaller sub-volume of phase space over time [23] . This explains the observed robustness of posterior patterning . Moreover , the geometry of the unstable manifold provides an explanation for the ordered succession of gap genes that become expressed in each nucleus of the posterior region . Such an ordered temporal sequence of gene expression , if arranged appropriately along the A–P axis , creates the observed kinematic anterior shifts of gap domains over time ( Fig 2A ) . Despite its explanatory power , the analysis by Manu et al . [23] is limited in an important way . In order to simplify phase space analysis , the authors implement simplified dynamics of maternal morphogens Bcd and Cad in their model ( Fig 2A ) . They use a time-invariant exponential approximation to simulate the Bcd gradient and Cad is assumed to reach a steady-state profile about 20–30 minutes before gastrulation [22 , 23] . This steady-state profile is used for model analysis . ( Based on this , we will refer to this formulation as the static-Bcd gene circuit model in what follows ) . Although reasonable , these simplifications affect the accuracy of the model , since Bcd and Cad have their own expression dynamics on a similar time scale as gap proteins . The Bcd gradient decays and Cad clears from much of the posterior trunk region towards the end of the blastoderm stage ( Fig 2B ) [42] . This means that the autonomous analysis of the static-Bcd model is not well suited to investigate the dynamic interpretation of morphogen gradients . In particular , assuming autonomy makes it impossible to isolate and study the explicitly time-dependent effects of changing gradient concentrations on gap gene regulation and pattern formation . For this reason , we consider the dynamics of maternal morphogens explicitly in our model . We have obtained gap gene circuits that incorporate realistic time-variable maternal gradients of Bcd and Cad ( Fig 2B ) [26] . These gradients are implemented as external inputs to gap gene regulation ( see Models and Methods section ) . They are not influenced by any of the state variables and , thus , are parameters of the system . This means that our gap gene circuits become fully non-autonomous [54] , since certain parameter values now change over time . While non-autonomous equations are not significantly more difficult to formulate or simulate than autonomous ones , phase space analysis is far from trivial . As model parameters change , so does the geometry of the phase portrait , and consequently system trajectories are actively shaped by this time-dependence . Separatrices and attractors can change their position ( geometrical change ) , and steady states can be created and annihilated through bifurcation events ( topological change ) ( S1B Fig ) . In autonomous systems , bifurcations can only occur along the spatial axis of the model . In non-autonomous systems , they also occur in time , implying that trajectories can switch from one basin of attraction to another during a simulation run . We can think of time-variable phase portraits as embedded in parameter space . We call the combination of phase and parameter space the configuration space of the system . The configuration space on non-autonomous models hence encodes a much richer repertoire of dynamical mechanisms of pattern formation than autonomous phase space alone . This can complicate analysis and interpretation of the system considerably . Using a simple model of a genetic toggle switch , we have established a methodology for the characterisation of transient dynamics in non-autonomous systems ( S1B Fig ) , based on the analysis of instantaneous phase portraits [43 , 45] . Such portraits are generated by fixing the values of system parameters starting at a given point in time , and then determining the geometrical arrangement of saddles , attractors , and their basins under these “frozen” conditions . The overall non-autonomous trajectory of the system is given by a series of instantaneous phase portraits over time . With sufficiently high temporal resolution , this method yields an accurate picture of the non-autonomous mechanisms of pattern formation implemented by the system . These mechanisms can be classified into four broad categories [43]: ( 1 ) transitions of the system from one steady state to another , ( 2 ) pursuit of a moving attractor within a basin of attraction , ( 3 ) geometrical capture , where a trajectory crosses a separatrix , and ( 4 ) topological capture , where a trajectory suddenly falls into a new basin of attraction due to a preceding bifurcation event ( S1B Fig ) . This classification scheme can be used to characterise the dynamical repertoire of non-autonomous models in a way analogous to phase space analysis in autonomous dynamical systems . In this paper , we present a detailed analysis of a non-autonomous gap gene circuit . Specifically , we use the model to address the effect of non-autonomy , i . e . the effect of time-variable maternal gradient concentrations , on gap gene regulation ( Fig 2 ) . To isolate explicitly time-dependent regulatory aspects , we simulate gap gene expression in the presence and absence of maternal gradient decay . Using phase space analysis , we then identify and characterise the dynamic regulatory mechanisms responsible for the observed differences between the two simulations . Our analysis reveals that maternal gradient decay limits the levels of gap gene expression and controls the dynamical positioning of posterior domains by regulating the rate and timing of domain shifts in the posterior of the embryo .
Non-autonomous gene circuit models are based on the connectionist formalism introduced by Mjolsness et al . [21] , modified to include time-variable external regulatory inputs as previously described [26 , 34] . Gene circuits are hybrid models with discrete cell divisions and continuous gene regulatory dynamics . The basic objects of the model consist of nuclei arranged in a one-dimensional row along the A–P axis of the embryo , covering the trunk region between 35 and 92% A–P position ( where 0% is the anterior pole ) . Models include the last two cleavage cycles of the blastoderm stage ( C13 and C14A ) and end with the onset of gastrulation; C14A is further subdivided into eight time classes of equal duration ( T1–T8 ) . At the end of C13 , division occurs and the number of nuclei doubles . The state variables of the system consist of the concentration levels of proteins produced by the trunk gap genes hb , Kr , gt , and kni . We denote the concentration of gap protein a in nucleus i at time t by g i a ( t ) . Change in protein concentration over time is given by the following set of ODEs: d d t g i a ( t ) = R a ϕ ( u i a ( t ) ) + D a ( n ) g i - 1 a ( t ) + g i + 1 a ( t ) - 2 g i a ( t ) - λ a g i a ( t ) ( 1 ) where Ra , Da and λa are rates of protein production , diffusion , and decay , respectively . Diffusion depends on the distance between neighbouring nuclei , which halves at nuclear division; thus , Da depends on the number of preceding divisions n . ϕ is a sigmoid regulation-expression function representing coarse-grained kinetics of transcriptional regulation . It is defined as follows: ϕ ( u i a ( t ) ) = 1 2 u i a ( t ) ( u i a ( t ) ) 2 + 1 + 1 ( 2 ) where u i a ( t ) = ∑ b ∈ G W b a g i a ( t ) + ∑ m ∈ M E m a g i m ( t ) + h a ( 3 ) with the set of trunk gap genes G = {hb , Kr , gt , kni} , and the set of external regulatory inputs M = {Bcd , Cad , Tll , Hkb} . External regulator concentrations g i m are interpolated from quantified spatio-temporal protein expression profiles [26 , 42 , 46] . The dynamic nature of these profiles renders the parameter term representing external regulatory inputs ∑ m ∈ M E m a g i m ( t ) time-dependent; explicit time-dependence of parameters implies non-autonomy of the dynamical system ( see Introduction and [54] ) . Interconnectivity matrices W and E define interactions among gap genes , as well as regulatory inputs from external inputs , respectively . The elements of these matrices , wba and ema , are called regulatory weights . They encode the effect of regulator b or m on target gene a . These weights may be positive ( representing an activating regulatory input ) , negative ( representing repression ) , or near zero ( representing the absence of a regulatory interaction ) . ha is a threshold parameter that represents the activation state of target gene a in the absence of any spatially and temporally specific regulatory input . This term incorporates the regualtory influence of factors that are not expressed in a spatially specific manner ( for example , the pioneer factor Zelda [31] ) . Eq ( 1 ) determines regulatory dynamics during interphase . In order to accurately implement the non-instantaneous duration of the nuclear division between C13 and C14A , the production rate Ra is set to zero during a mitotic phase , which immediately precedes the instantaneous nuclear division . Mitotic schedule as in [26] . We determine the values for parameters Ra , λa , W , E , and ha using a reverse-engineering approach [19 , 25 , 26 , 34] . For this purpose , we numerically solve gene circuit Eq ( 1 ) across the region between 35 and 92% A–P position using a Runge-Kutta Cash-Karp adaptive step-size solver [26] . Models are fit to a previously published quantitative data set of spatio-temporal gap protein expression [26 , 42 , 46] ( see Fig 1 for gap gene expression patterns , and Fig 2B for dynamic Bcd and Cad profiles ) . Model fitting was performed using a global optimization algorithm called parallel Lam Simulated Annealing ( pLSA ) [47] . We use a weighted least squares cost function as previously described [26] . To enable comparison of our results to the static-Bcd gene circuit analysis by Manu et al . [23] , we keep model formalism and fitting procedure as similar as possible to this earlier study . Manu and colleagues fitted gene circuits including a diffusion term , but analysed the model with diffusion rates Da set to zero [23] . This diffusion-less approach reduces the phase space of the model from hundreds of dimensions to 4 by spatially uncoupling the equations and considering each nucleus independently from its neighbours . Dimensionality reduction is essential for geometrical analysis of phase space . Unfortunately , setting diffusion to zero in our best 3 ( of a total of 100 ) non-autonomous gene circuits fitted to data with non-zero diffusion terms leads to severe patterning defects ( see S2 Fig for common patterning defects ) . This is likely due to numerical , not biological issues , since we do find circuit solutions that correctly reproduce gap gene patterns both in the presence and absence of diffusion using an alternative fitting approach that fixes diffusion parameters Da to zero during optimization ( see below ) . To further facilitate comparison with the static-Bcd model , we constrained the signs of regulatory weights to those reported in Manu et al . [23] . In previoius work , we have verified this network structure extensively against experimental data [18 , 25 , 26 , 34] . Optimization was performed on the Mare Nostrum supercomputer at the Barcelona Supercomputing Centre ( http://www . bsc . es ) . One optimization run took approximately 35 min on 64 cores . The purpose of our reverse-engineering approach is not to sample parameter space systematically , but instead to discover whether there are specific model-fitting solutions that are consistent with the biological evidence and reproduce the dynamics of gap gene expression correctly . Global optimization algorithms are stochastic heuristics without guaranteed convergence , which means that for complex non-linear problems many optimization runs will fail or end up at sub-optimal solutions ( see also discussions in [24 , 26 , 33] ) . In order to find the best-fitting solution , we therefore select solutions from 200 initial fitting runs as follows: ( 1 ) we discard numerically unstable circuits; ( 2 ) we only consider solutions with a root-mean-square ( RMS ) score less than 20 . 0 as most circuits with scores above this threshold show gross patterning defects; ( 3 ) we use visual inspection to detect remaining gross patterning defects among selected circuits ( missing or bimodal domains , and disconnected boundaries . See S2 Fig ) as previously described [34] . Out of the resulting 7 highest scoring circuits , only 3 recover the shifting dynamics of posterior gap domains . In order to rule out diffusion as a pattern-generating mechanism in these circuits , we compared their performance in the presence and absence of diffusion ( see above ) . For this purpose , we used values of diffusion rates Da obtained by fitting our non-autonomous models with diffusion . All three circuits produce satisfactory gap gene patterns ( including anteriorly shifting posterior trunk domains ) whether diffusion is present or not . The best fit among these was selected for detailed analysis ( see S1 Table , for parameter values ) . The residual error of our best-fitting diffusion-less circuit ( RMS = 10 . 73 ) lies at the lower end of the range of residual errors for fully-non-autonomous circuits with diffusion , which range from RMS scores of 10 . 43 to 13 . 32 [26] . This lends further support to the notion that diffusion is not essential for gap gene patterning . Moreover , our previous work also shows that circuits which were fit without weighting the data show somewhat lower RMS scores of 8 . 71 to 10 . 11 despite exhibiting more patterning defects at late stages [26] . The RMS score of the static-Bcd model ( fit without weights ) is higher , at 10 . 76 [22] . Taken together , this implies a slightly better quality-of-fit of our fully non-autonomous diffusion-less model compared to the static-Bcd diffusion-less circuits of Manu et al . [22] . We characterise the time-variable geometry and topology of phase space in our fully non-autonomous gap gene circuit for every nucleus in a sub-range of the fitted model between 35 and 71% A–P position . This restricted spatial range allows us to simplify the analysis by excluding the influence of terminal gap genes tll and hkb on patterning ( similar to the approach in [22] ) . We aim to identify those features of configuration space that govern the placement of domain boundaries , and thus the patterning capability of the gap gene system . We achieve this by generating instantaneous phase portraits for the model [43 , 45] at 10 successive points in time ( C13 , C14A-T1–8 , and gastrulation time ) . To generate an instantaneous phase portrait , all time-dependent parameter values—i . e . those corresponding to the profiles of external regulators—are frozen at every given time point . This yields an autonomous system for each point in time , for which we can calculate the position of steady states in phase space using the Newton-Raphson method [48 , 49] as implemented by Manu et al . [23] . We classify steady states according to their stability , which is determined by the corresponding eigenvalues ( see S1A Fig ) . Nuclei express a maximum of three trunk gap genes over developmental time , and only two at any given time point . Therefore , we project four-dimensional phase portraits into lower-dimensional representations to visualise them more easily . This yields a graphical time-series of instantaneous phase portraits for each nucleus , which allow us to track the movement , creation , and annihilation of steady states ( typically attractors and saddles ) by bifurcations . The transient geometry of phase space governs the non-autonomous trajectories of the system . We classify the dynamic behaviours exhibited by these trajectories into transitions , pursuits , and captures according to our previously established methodology ( see Introduction and S1B Fig ) [43] .
Previously published non-autonomous gap gene circuits suggest a specific regulatory structure for the gap gene network in D . melanogaster ( Fig 3A ) [26] . This structure is consistent with the network predicted by the static-Bcd model of Manu et al . [23] , and with the extensive genetic and molecular evidence available in the published literature on gap gene regulation [18] . Unfortunately , it is difficult to derive insights about dynamic regulatory mechanisms from a static network diagram . Computer simulations help us understand which network interactions are involved in positioning specific expression domain boundaries across space and time [24–26 , 34] . Although powerful , this simulation-based approach has its limitations . It cannot tell us how expression dynamics are brought about: for instance , why some gap domain boundaries remain stationary while others shift position over time . To gain a deeper understanding of the underlying regulatory dynamics , we analyse the configuration space of a fully non-autonomous gene circuit through instantaneous phase portraits ( S1B Fig ) [43] , analogous to the autonomous phase-space analysis presented by Manu and colleagues [23] ( Fig 2 ) . This type of analysis requires diffusion-less gap gene circuits to keep the dimensionality of phase space at a manageable level . We obtained fully non-autonomous gap gene circuits that lack diffusion through model fitting with diffusion parameters Da fixed to zero and interaction signs constrained to those of previous works ( as described in “Models and Methods” ) . This resulted in a set of three selected , well-fitted circuits . The network topology of these gene circuit models correspond to that shown in Fig 3A . The following analysis is based on the best-fitting model with a root mean square ( RMS ) residual error of 10 . 73 , which constitutes a slight overall improvement in quality-of-fit compared to static-Bcd models ( see “Models and Methods” and [22 , 26] ) . Its regulatory parameter values are listed in S1 Table . This diffusion-less non-autonomous gene circuit accurately reproduces gap gene expression ( Fig 3B ) . In particular , it exhibits correct timing and relative positioning of domain boundaries . Together with the fact that it fits the data equally well as equivalent circuits with diffusion ( see “Models and Methods” , and [26] ) , this confirms earlier indications that gap gene product diffusion is not essential for pattern formation by the gap gene system [23 , 25] . Interestingly , previously published diffusion-less static-Bcd circuits show rugged patterns with abrupt “on/off” transitions in expression levels between neighbouring nuclei [23] . In contrast , diffusion-less fully non-autonomous circuits produce smooth spatial expression patterns with a graded increase or decrease in concentration levels across domain boundaries . This is because non-autonomy , with its associated movement of attractors and separatrices over time , provides increased flexibility for fine-tuning expression dynamics over time compared to models with constant phase-space geometry ( see below ) . In biological terms , it suggests that the expression of smooth domain boundaries does not strictly require diffusion . Although diffusion undoubtedly contributes to this process in the embryo , its role may be less prominent than previously thought [23 , 25] . We used our non-autonomous gap gene circuit to assess the effect of maternal gradient decay on gap gene regulation . One way to isolate this effect is to compare the output of the fully non-autonomous model—with decaying maternal gradients—to simulations using the same model parameters , but keeping maternal gradients fixed to their concentration levels early during the blastoderm stage ( time class C12 ) . As shown in Fig 4 , the relative order and positioning of gap domains remain unaffected when comparing models with fixed versus time-variable gradient concentrations . This indicates that maternal gradient decay is not strictly required for correct pattern formation by gap genes . We do observe , however , that maternal gradient dynamics significantly affect the levels of gap gene expression throughout the trunk region of the embryo ( Fig 4 , shaded areas ) . While early expression dynamics are very similar in both models ( time classes C12–T2 ) , they begin to diverge at later stages . The fully non-autonomous model reaches peak expression at T2/T4 , but the autonomous model without maternal gradient decay overshoots observed expression levels in the data between T4 and T8 . This indicates that maternal gradient decay leads to decreasing activation rates at the late blastoderm stage , thereby regulating the timing and level of peak gap gene expression . Such a limiting regulatory effect of maternal gradients has been proposed before [25 , 42] , but has never been tested explicitly . Interestingly , the overshoot occurs in different ways in the anterior and the posterior of the embryo . In the anterior , maximum concentrations of Hb and Kr across each domain remain unchanged , but levels of expression keep increasing around the Kr/Gt interface , rendering the domain boundaries steeper and less smooth in the simulation without maternal gradient decay ( Fig 4 , asterisk ) . In the posterior , we observe increased levels of Kni and Gt across large parts of their respective expression domains ( Fig 4 , arrows ) . These effects are asymmetric: both posterior Kni and Gt domains exhibit an anterior expansion , while the posterior boundary of the Kni domain is not affected . Considering that both of these domains shift towards the anterior over time ( Fig 1 ) [25 , 42] , we interpret this as follows: maternal gradient decay not only decreases the rate of expression at late stages in the posterior region , but also leads to a slow-down of gap domain shifts , thereby limiting the extent of the shift . In the autonomous simulation without maternal gradient decay , both Kni and Gt domains keep on moving , which explains the observed expansion and increase of expression levels towards the anterior part of the domain . We asked whether the differing effects of maternal gradient decay in the anterior and the posterior of the embryo depend on the presence of different regulatory mechanisms in these regions [23] . To validate this hypothesis , we need to understand and characterise the dynamic mechanisms underlying gene regulation in our non-autonomous model . We achieve this through analysis of the time-variable phase spaces of nuclei across the trunk region of the embryo using the methodological framework presented in the Introduction ( S1B Fig; see [43] for details ) . To briefly reiterate , this analysis is based on the characterization of the changing phase space geometry that shapes the trajectories of the system . The shape of a trajectory indicates typical dynamical behaviors , that can be classified into four distinct categories—transitions , pursuits , as well as geometrical and topological captures—each showing particular dynamic characteristics . These categories provide mechanistic explanations for the dynamic behavior of the system . For every nucleus , we then compare these non-autonomous mechanisms to the autonomous mechanisms of pattern formation found in the static-Bcd model [23] . This direct comparison allows us to identify the causes underlying the observed effects of maternal gradient decay on the temporal dynamics of gap gene expression . In agreement with Manu et al . [23] , we find different patterning modes anterior and posterior to 52% A–P position . Just like in static-Bcd models , anterior expression dynamics are governed by convergence of the system towards attractors in a multi-stable regime . In contrast , our model differs from that of Manu et al . [23] concerning posterior gap gene regulation . We find that a monostable spiral sink drives gap domain shifts in the posterior of the embryo; this differs markedly from the unstable manifold observed in static-Bcd gap gene circuits [23] . An in-depth analysis and biological discussion of spatial pattern formation driven by this mechanism goes beyond the scope of this study . It is provided elsewhere [44] . Here , we focus on temporal aspects of gene regulation and pattern formation , namely the regulation of the velocity of gap domain shifts by maternal gradient dynamics in the posterior of the embryo .
In this paper , we have examined the explicitly time-dependent aspects of morphogen gradient interpretation by a gene regulatory network; the gap gene system of the vinegar fly D . melanogaster . Using a fully non-autonomous gap gene circuit , we compared the dynamics of gene expression in the presence and absence of maternal gradient decay . We find that dynamic changes in the concentration of maternal morphogens Bcd and Cad affect the timing and rate of gap gene expression . The precise nature of these effects differs between the anterior and the posterior region of the embryo . In the anterior , gradient decay creates smooth domain borders by preventing the excessive accumulation of gene products across boundary interfaces between neighbouring gap domains . In the posterior , gradient decay limits the rate of gap gene expression , and therefore the extent of gap domain shifts , towards the end of the blastoderm stage . A temporal effect on gene expression rates is translated into slowing rates of domain shifts , which in turn alter the spatial positioning of expression boundaries . As a consequence , gradient decay stabilises spatial gap gene patterns before the onset of gastrulation . An effect of maternal gradient decay on gap gene expression rates has been suggested before—based on the analysis of quantitative expression data [25 , 42] . However , only mechanistic dynamical models—such as the non-autonomous gap gene circuits presented here—can provide specific mechanisms and quantitative causal evidence for this aspect of gap gene regulation . Our analysis suggests that maternal gradient decay—specifically , the disappearance of Cad from the abdominal region of the embryo—has an important role in regulating the timing of gap gene expression as well as limiting the rate and extent of gap domain shifts in the posterior of the embryo . This result is consistent with experimental data indicating that Cad affects gap domain shifts . Mutants lacking maternal cad , which show a reduced level of Cad protein throughout the blastoderm stage [28] , show a delay in the shift of the posterior domains of kni and gt [32 , 44] . However , Cad does not seem to act exclusively . An indirect role of Bcd in regulating gap domain shifts through altering gap-gap interactions was suggested by a modelling study [30] . It remains unclear whether Cad is also involved in mediating this effect . Finally , a recent study of Bcd-dependent regulation of hb postulated an additional mechanism for gap gene down-regulation that acts before maternal gradient decay occurs [2] . This could have an indirect effect on the timing of late ( Bcd-independent ) hb regulation , which may mediate the direct effect of Bcd decay on late hb expression we are observing in our models . To better understand the mechanistic basis for the observed differences in patterning between the anterior and the posterior , we analysed the time-variable phase portraits in our non-autonomous model [43] . In agreement with a previous study based on autonomous phase space analysis of static-Bcd gap gene circuits [23] , we find that two distinct dynamical regimes govern gap gene expression anterior and posterior to 52% A–P position ( Fig 8 ) . Stationary domain boundaries in the anterior are governed by regulatory mechanisms that are equivalent in static-Bcd and fully non-autonomous models ( our work and [23] ) : they take place in a multi-stable dynamical regime where the posterior boundary of the anterior Gt domain is set by the movement of an attractor in phase space , and the posterior boundary of the anterior Hb domain is set by attractor selection ( i . e . the capture of transient trajectories in the non-autonomous case ) ( Fig 8 , left ) . Attractor movement in fully non-autonomous models leads to smooth expression boundaries , which are absent in the static-Bcd case . In contrast , static-Bcd and non-autonomous models suggest different mechanisms for gap domain shifts in the posterior of the embryo . While these shifts are controlled by an unstable manifold in the static-Bcd gene circuit model [23] , we find a pursuit mechanism featuring a monostable spiral sink to govern their behaviour in our fully non-autonomous analysis ( Fig 8 ) . The spiralling geometry of transient trajectories imposes temporal order on the progression of gap genes being expressed . If arranged appropriately across nuclei in the posterior of the embryo , this temporal progression from Kr to kni to gt to hb leads to the emergence of the observed kinematic domain shifts [44] . It is important to note that similar regulatory principles can be found in all three solutions of our fully non-autonomous model that reproduce gap-gene patterning correctly both in the presence and absence of diffusion . We have chosen the most structurally stable solution for detailed analysis . The other two circuits show more variability of regulatory features both across space and time . Still , both of these models consistently exhibit multi-stability in the anterior , and spiral sinks as well as transiently appearing and disappearing limit cycles in the region posterior to 52% A–P position . This indicates that the two main dynamical regimes described here—stationary boundaries through attractor selection in the anterior vs . shifting gap domain boundaries through spiralling trajectories in the posterior—are reproducible across model solutions . It is important to note that non-autonomy of the model is not strictly required for the spiral sink mechanism to pattern the posterior of the embryo . Simulations with fixed maternal gradients demonstrate that domain shifts can occur in an autonomous version of our gap gene circuit ( see Figs 4 and 7 ) . The reason why earlier models [22 , 23] do not feature spiral sinks remains unknown although one possibility is that fitting in the absence of diffusion somehow benefits characterisations of posterior pattern formation in terms of oscillatory behaviours . In spite of this , there are two reasons to consider the mechanism proposed here an important advance over the unstable manifold proposed by Manu et al . [23] . The first reason is technical: non-autonomous gap gene circuits—implementing correct maternal gradient dynamics—are more accurate and stay closer to the data than the previous static-Bcd model . The fact that the quality of a reverse-engineered model usually depends on the quality of its fit to data implies that our model provides more accurate and rigorous predictions than previous efforts . The second reason is conceptual: although it is difficult to interpret an unstable manifold in an intuitive way , it is straightforward to understand the spiral sink as a damped oscillator patterning the posterior of the embryo . The presence of an oscillatory mechanism in a long-germband insect such as D . melanogaster has important functional and evolutionary implications , which are discussed elsewhere [44] . Analysis of an accurate , non-autonomous model is required to isolate and study the explicitly time-dependent aspects of morphogen interpretation by the gap gene system . Here , we have shown that such an analysis is feasible and leads to relevant and specific new insights into gene regulation . Other modelling-based studies have used non-autonomous models before ( see , for example , [16 , 26 , 34 , 50–53] ) . However , none of them have directly addressed the proposed role of non-autonomy in pattern formation [17] . Our analysis provides a first step towards a more general effort to transcend this limitation in our current understanding of the dynamic regulatory mechanisms underlying pattern formation during animal development .
|
Animal development is a highly dynamic process . Biochemical or environmental signals can cause the rules that shape it to change over time . We know little about the effects of such changes . For the sake of simplicity , we usually leave them out of our models and experimental assays . Here , we do exactly the opposite . We characterise precisely those aspects of pattern formation caused by changing signalling inputs to a gene regulatory network , the gap gene system of Drosophila melanogaster . Gap genes are involved in determining the body segments of flies and other insects during early development . Gradients of maternal morphogens activate the expression of the gap genes . These gradients are highly dynamic themselves , as they decay while being read out . We show that this decay controls the peak concentration of gap gene products , produces smooth boundaries of gene expression , and slows down the observed positional shifts of gap domains in the posterior of the embryo , thereby stabilising the spatial pattern . Our analysis demonstrates that the dynamics of gene regulation not only affect the timing , but also the positioning of gene expression . This suggests that we must pay closer attention to transient dynamic aspects of development than is currently the case .
|
[
"Abstract",
"Introduction",
"Models",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"invertebrates",
"gene",
"regulation",
"animals",
"blastulas",
"animal",
"models",
"developmental",
"biology",
"drosophila",
"melanogaster",
"model",
"organisms",
"systems",
"science",
"mathematics",
"experimental",
"organism",
"systems",
"molecular",
"development",
"embryos",
"morphogenesis",
"pattern",
"formation",
"drosophila",
"research",
"and",
"analysis",
"methods",
"blastoderm",
"embryology",
"computer",
"and",
"information",
"sciences",
"embryonic",
"pattern",
"formation",
"gene",
"expression",
"dynamical",
"systems",
"insects",
"arthropoda",
"morphogens",
"genetics",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"organisms"
] |
2017
|
Dynamic Maternal Gradients Control Timing and Shift-Rates for Drosophila Gap Gene Expression
|
The bacterial colicin-immunity proteins Im7 and Im9 fold by different mechanisms . Experimentally , at pH 7 . 0 and 10°C , Im7 folds in a three-state manner via an intermediate but Im9 folding is two-state-like . Accordingly , Im7 exhibits a chevron rollover , whereas the chevron arm for Im9 folding is linear . Here we address the biophysical basis of their different behaviors by using native-centric models with and without additional transferrable , sequence-dependent energies . The Im7 chevron rollover is not captured by either a pure native-centric model or a model augmented by nonnative hydrophobic interactions with a uniform strength irrespective of residue type . By contrast , a more realistic nonnative interaction scheme that accounts for the difference in hydrophobicity among residues leads simultaneously to a chevron rollover for Im7 and an essentially linear folding chevron arm for Im9 . Hydrophobic residues identified by published experiments to be involved in nonnative interactions during Im7 folding are found to participate in the strongest nonnative contacts in this model . Thus our observations support the experimental perspective that the Im7 folding intermediate is largely underpinned by nonnative interactions involving large hydrophobics . Our simulation suggests further that nonnative effects in Im7 are facilitated by a lower local native contact density relative to that of Im9 . In a one-dimensional diffusion picture of Im7 folding with a coordinate- and stability-dependent diffusion coefficient , a significant chevron rollover is consistent with a diffusion coefficient that depends strongly on native stability at the conformational position of the folding intermediate .
The study of proteins that fold in an apparent two-state-like manner [1] has led to tremendous advances in protein folding biophysics [2 , 3] . In line with the consistency [4] and minimal frustration [5] principles , the energy landscapes of these proteins may be pictured as smooth funnels with little ruggedness [6–8] . However , the consistency between local and nonlocal interactions is never perfect [4] . Frustration exists [5] in biomolecules and can sometimes serve important biological functions [9] . It is physically intuitive that energetically favorable nonnative interactions can occur [10] . Through improved experimental techniques , nonnative interactions are now known to be more prevalent than previously appreciated [11 , 12] . From a fundamental biophysical standpoint , a better understanding of the presence and absence of nonnative interactions is key to deciphering biomolecular recognition and to assessing our grasp of basic protein energetics [13] . As one of the earliest definitive examples of nonnative effects in single-domain proteins , the folding kinetics of bacterial immunity protein Im7 and its homolog Im9 are well characterized [14 , 15] . Despite their very similar native structures ( Fig 1A and 1B ) , a large body of experimental work demonstrates that Im7 folds via an intermediate stabilized by nonnative contacts , whereas Im9 folding is essentially two-state [16–22] . The relative simplicity of the Im7/Im9 system makes it well suited for an informative case study . Unlike some of the larger proteins ( number of residues n ≳ 100 ) such as cytochrome c [23] and ribonuclease A [24] that fold in a more complex manner [25] , Im7 and Im9 folding is not complicated by a heme or disulfide bonds . Indeed , in view of many single-domain proteins that can fold with no apparent nonnative effects , the nonnative interactions in Im7 are likely a consequence of functional constraints [26 , 27] . It is noteworthy in this connection that the biological functions of Im7 and Im9 are evolutionarily related by promiscuous interactions [28] that are probably underpinned by nonnative excited-state conformations [29] . Theory and computation have provided valuable insights into the Im7/Im9 system . Experimental Φ-values were used as constraints in conformational sampling to derive putative folding transition states of these proteins [27 , 30] . The results suggest a functional origin for the nonnative interactions in Im7 [27] . In a separate effort , an equilibrium intermediate state was predicted for Im7 using a Gō-like model that assumes no favorable nonnative interaction [31] . However , although topological frustration and heterogeneity in contact density can , in some cases , lead to kinetic and equilibrium folding intermediates in the absence of favorable nonnative interactions [32–34] , a subsequent computational study indicates that Im7 folding cannot be explained by native-centric interactions alone [26] . Instead , nonnative effects arising from “localized frustration” [35] was seen as necessary for rationalizing the peculiar behaviors of Im7 [26] . Consistent with this finding as well as with experiment , a sequential stabilization algorithm for predicting folding pathway was not able to reach the Im7 native structure because of kinetic trapping; but the same algorithm was successful in accessing the Im9 native structure [36] . A clear kinetic difference between Im7 and Im9 is manifested by their chevron plots of logarithmic folding and unfolding rates versus denaturant concentration [11] . The folding arm of the Im7 chevron at pH 7 . 0 and 10°C exhibits a significant rollover , whereas that of the Im9 does not [16 , 18–20] . The present study addresses this basic distinction between Im7 and Im9 by direct simulations of folding/unfolding rates . Because each chevron plot is a summary of kinetic and thermodynamic data from a large set of folding/unfolding trajectories [13] , it is not yet practical to employ all-atom molecular dynamics [37 , 38] for the extensive computation necessary to produce model chevron plots . Moreover , current molecular dynamics forcefields are probably insufficient to rationalize highly cooperative folding behaviors such as that of Im9 because the forcefields tend to over-predict nonnative effects [38 , 39] . Therefore , as an interim method that has been applied elsewhere [40–42] , we develop tractable explicit-chain coarse-grained models [43] to tackle the chevron behaviors of Im7 and Im9 , as these behaviors have not been addressed by direct simulations to date . We model nonnative effects using “hybrid” formulations that augment structure-based native-centric interactions with physics-based , sequence-dependent transferrable energy terms [44 , 45] . Limitations notwithstanding , this approach has been accounting for an increasing number of experiments [13 , 43 , 46–51] . By comparing nonnative interactions that do [52] and do not [47 , 49] reflect the variation of hydrophobicity among nonpolar residues [53] , we find that the difference between the Im7 and Im9 chevrons is well rationalized by nonative interactions involving large hydrophobic residues . The present study addresses also the relationship between conformational diffusion and folding intermediates . Diffusion is a useful concept [54–59] in describing physical effects of solvent and internal friction in folding [60–63] . Whereas mild internal friction likely arises from the particulate nature of the solvent [62] and correlated dihedral rotations along the polypeptide [63] , elevated internal friction in compact chains [60] can emerge more generally from topological frustration [32 , 33] and favorable nonnative interactions [10 , 54] . As discussed below , the Im7 chevron rollover in our model is associated with a coordinate- and stability-dependent coefficient of one-dimensional diffusion , with a strong anticorrelation between native stability and diffusion rate at the position of the transiently trapped intermediate . Notably , the smallest diffusion coefficients at these trapped positions can be orders of magnitude smaller than those encountered in two-state-like folding .
The equilibrium free energy profiles computed near the models’ transition midpoints ( Fig 1C and 1D ) show no dramatic difference between Im7 and Im9 . The free energy barrier is lower for Im7 than for Im9 in the db models ( dotted curves ) ; but this trend is reversed when the nonnative interactions in the db+hϕ and db+MJhϕ models are included ( dashed and solid curves ) . Nonnative interactions in these models slow down folding for Im7 but speed up folding for Im9 . Unlike previous Im7 models that exhibit a significantly populated equilibrium intermediate [26 , 31] ( which is apparently not quite in line with the success of two-state fitting of experimental equilibrium data for wildtype Im7 [22] ) , folding in our models is thermodynamically two-state as their folding/unfolding barriers under midpoint conditions are quite high ( ≳ 5kBT , where kB is Boltzmann constant and T is absolute temperature ) . The only hint of an Im7 folding intermediate is a small dip in the Im7 profiles ( Fig 1C ) at Q ≈ 0 . 85 that is absent in the Im9 profiles ( Fig 1D ) . This feature by itself is no definitive evidence for complex folding kinetics , however . Under much stronger folding conditions , folding in our models becomes downhill [64] . Now even less difference is seen in Fig 2A between the equilibrium free energy profiles of Im7 and Im9 under zero-denaturant conditions ( ΔG/kBT ≈ −10 . 5 and −12 . 0 , corresponding to the experimental folding free energy of approximately −24 . 9 kJ mol−1 for Im7 [19] and −28 . 2 kJ mol−1 for Im9 [15] at pH 7 . 0 and 10°C; see Fig 2B ) . The approximate linearity of native stability versus interaction strength ϵ/T ( Fig 2B ) allows ΔG/kBT to be used as a proxy for denaturant concentration [42] in model chevron plots . Fig 3 shows that the folding-arm rollover and lack thereof , respectively , in the experimental chevrons for Im7 and Im9 at pH 7 . 0 and 10°C [16 , 18–20] is captured by the db+MJhϕ but not the db and db+hϕ models , suggesting that the Im7 rollover arises from the strong nonnative interactions among the large hydrophobic residues as modeled by db+MJhϕ ( S1 Fig ) . The difference between Im7 and Im9 folding cannot be explained by native interactions alone ( as in db ) or the more generic nonnative hydrophobic effects in db+hϕ . The chevron rollover in the db+MJhϕ Im7 model is a consequence of transient yet long-lived trapped conformations at Q ≈ 0 . 85 ( Fig 4A ) , which do not appear in Im9 folding under similarly strong folding conditions ( Fig 4B ) . An overview of Im7 and Im9 folding kinetics is afforded by their kinetic profiles , which show a deep minimum at Q ≈ 0 . 85 for Im7 ( Fig 4C ) but not for Im9 ( Fig 4D ) . Determined from folding trajectories alone [59] , kinetic profiles are more useful than free energy profiles for identifying folding intermediates . The Im7/Im9 difference is not apparent from the free energy profiles because , on one hand , kinetic trapping is minimal when folding is only weakly favored ( Fig 1C ) . On the other hand , when folding is strongly favored ( Fig 2A ) , the contribution from folding trajectories to an equilibrium profile is overwhelmed by that from unfolding trajectories , viz . , the resident time in the folded state is much longer than that in the unfolded and intermediate states . Consequently , the deep well at Q ≈ 0 . 85 in Fig 4C translates to merely a small kink around the same Q value in Fig 2A . A physical account of the main difference between Im7 and Im9 folding kinetics is thus provided . Many mutational experiments are rationalized below as well . Because of their simplicity , however , db+MJhϕ models are limited in some respects . For instance , the midpoint folding rate of Im7 is ≈ 1/5 that of Im9 in this model ( Fig 3C ) ; but the experimental midpoint rate of Im7 ( ≈ 1 . 2–3 . 0 s−1 [19 , 65] ) is ≳ 40 times that of Im9 ( ≈ 0 . 03 s−1 [15 , 20] ) . Moreover , whereas the folding and unfolding arms of the simulated chevron plots are quite symmetric around the transition midpoint , experimental unfolding rate exhibits a much weaker denaturant dependence than folding rate [16 , 18–20] . These drawbacks are typical of topology-based models [42] , which are more apt for folding than for unfolding kinetics [43 , 66] . But this limitation has little bearing on our analysis of folding kinetics . Improved modeling likely requires special stability-enhancing energies that have minimal effects on folding kinetics [67 , 68]; but such efforts are outside the scope of the present work . Structural properties of our simulated Im7 intermediate ( Fig 5 ) are largely in agreement with mutagenesis experiments , which indicate that the intermediate is stabilized by nonnative interactions between Helix IV and the open end of the Helix I-Helix II hairpin involving residues L3 , I7 , F15 , V16 , L18 , L19 , L34 , L37 , L38 , F41 , V42 , I68 , and I72 [19] . Notably , 12 of these 13 residues are involved in the most populated 20 nonnative hydrophobic contacts ( with > 80% probability of occurrence ) in the Im7 intermediate simulated using db+MJhϕ ( Fig 5A , upper triangle ) . The only exception is V42 , for which the most probable nonnative contact V36–V42 has nonetheless a 73% occurrence probability in the simulated intermediate . Among the 20 most probable nonnative contacts in the simulated Im7 intermediate , three are between the N-terminal segment and Helix II [L3–V33 ( 94% ) , L3–L34 ( 85% ) , I7–V36 ( 92% ) ] , eight are between Helices I and II [F15–L34 ( 92% ) , F15–V36 ( 99% ) , F15–L37 ( 97% ) , F15–L38 ( 85% ) , V16–L37 ( 92% ) , V16–L38 ( 80% ) , L19–L38 ( 80% ) , L18–L34 ( 96% ) ] , four are between different residues in Helix II [V33–F41 ( 98% ) , L34–F41 ( 99% ) , V36–F41 ( 90% ) , V36–I44 ( 82% ) ] , and five are between Helices II and IV [L37–V69 ( 99% ) , L37–I72 ( 96% ) , L38–I68 ( 99% ) , L38–V69 ( 99% ) , L38–I72 ( 91% ) ] . Helix III hardly contributes to the intermediate-stabilizing nonnative contacts in the model . The most likely nonnative contact in the intermediate ensemble that involves Helix III , L38–L53 , has an occurrence probability of only 17% . Our computed probabilities of contacts are in line with experiments indicating that Helices I and IV are fully formed but Helix II is partly formed in the Im7 intermediate [14] . In Fig 5A , intrahelical contacts between residues i , i + 4 are present but less probable for Helix II ( residues 32 to 45 ) than for Helices I and IV ( residues 12 to 26 and 65 to 78 ) . Experiment indicates also that Helix III is absent [14] but it is present in our simulated Im7 intermediate . This limitation of the model is likely related to its simple treatment of native interactions . Nonetheless , in agreement with experiment , amino acid substitutions in Helix III result only in small changes in folding rate in the db+MJhϕ model ( see below ) . A snapshot of the simulated Im7 intermediate state is shown in Fig 5B wherein each of the highlighted nonnative contacts has ≥ 80% occurrence probability except M1–L18 ( 56% ) in the Im7 intermediate ensemble ( Fig 5C ) . All except one ( V42 ) of the 13 residues identified by mutagenesis experiments ( see above ) to be stabilizing the Im7 intermediate are represented in the highlighted nonnative contacts . We have verified that structures very similar to the Cα intermediate conformation in Fig 5B are physically realizable by constructing a corresponding atomic structure [69] with added sidechains [70] ( S2 Fig ) . Our simulated Im7 kinetic intermediate is stabilized by nonnative interactions ( S1 Fig ) . As such , it is diametrically different from the equilibrium intermediates simulated using purely native-centric models [31] with heterogeneous Gō energies [71] . Instead of being a product of nonnative effects , equilibrium intermediates in such Gō-like models arise from their reduced folding cooperativity [72] , which can lead to three-state-like free energy profiles for Im7 and Im9 ( S3 Fig ) ; but such features are at odds with experiment . Effects of select mutations in the db+MJhϕ model for Im7 are examined through their folding kinetic profiles [59] ( Fig 6 ) . Some mutations reduce the depth of the kinetic trap at Q ≈ 0 . 85 relative to that of the wildtype ( WT ) while others lead only to negligible changes . We compute also the rates of reaching the intermediate position at Q ≈ 0 . 85 and the folded state at Q = 0 . 98 from initially unfolded conformations . The former rate ( ≈ 3 . 9 × 10−7 for WT , in units of reciprocal number of time steps ) varies little , whereas the latter overall folding rate ( = 5 . 0 × 10−8 for WT ) is sensitive to mutation . The overall folding rate correlates , albeit imperfectly , with the depth of the Q ≈ 0 . 85 minimum . The general trend of variation of the simulated folding rates is largely in line with that of the experimental intermediate-to-native folding rates kin [19] or kIN [65] for the single mutants ( = 238 s−1 for WT ) in Fig 6A . For both simulation and experiment , folding rate remains essentially unchanged for three mutants ( simulated rate in units of 10−8 , experimental rate in s−1 [19 , 65] ) : I54A ( 4 . 4 , 200 ) , I72V ( 5 . 0 , 250 ) , A77G ( 4 . 9 , 235 ) [ ( 5 . 0 , 238 ) for WT]; and is speeded up for four mutants: F15A ( 30 . 7 , 550 ) , L34A ( 40 . 2 , 1850 ) , L37A ( 49 . 6 , 450 ) , L38A ( 31 . 1 , 1600 ) . Folding rate remains essentially unchanged experimentally but is speeded up in simulation for five mutants; nevertheless the simulated increase is less than that for mutants that fold faster in experiments: F15Y ( 19 . 3 , 220 ) , V16A ( 10 . 7 , 220 ) , V33A ( 19 . 7 , 238 ) , V36A ( 22 . 3 , 190 ) , F41Y ( 9 . 3 , 186 ) . However , the present model cannot account for the dramatic experimental increase in folding rate and the disappearance of folding-arm rollover for F41L ( kin = 5000 s−1 [19] , ≈ 21 times of that of WT ) because F and L have similar MJ energies [52] . For this mutant , the simulated rate 3 . 6 × 10−8 is smaller than that of WT . Even mutating F to a non-hydrophobic in the model cannot produce the experimental effect of F41L , viz . , the simulated rate for the F41G mutant is 2 . 68 times that of WT but is far from sufficient . To account for the dramatic impact of F41L , future theoretical studies will need to pursue subtle effects beyond our simple treatment of hydrophobicity , perhaps by considering energetics specific to aromatic residues [13 , 73] . Consistent with experiment [14] , L53A/I54A has a negligible kinetic effect on Im7 in our model ( Fig 6B ) , which is in line with the small experimental Φ-values of ≈ 0 . 03–0 . 16 and kin = 200 s−1 for L53 and I54 in Helix III [19] . In contrast , many double mutants with hydrophobicity-reducing substitutions in Helices I and II can dramatically destabilize the folding intermediate and thus speed up Im7 folding ( Fig 6B ) . These predictions should be testable by future experiments . However , because mutations in our models change only the nonnative but not the native interactions , as it stands our approach cannot address mutations such as L18A/L19A/L37A that prevent Im7 folding [22] . The three-state kinetics of Im7 is related to its hydrophobic composition . Im7 has 32 hydrophobic residues ( 17 with stronger and 15 with weaker hydrophobicities; Fig 1 ) whereas Im9 has 28 ( 15 and 13 in the two categories ) . In Helix II , Im7 has two more hydrophobics ( V33 , V42 ) and the stronger L38 instead of the weaker V37 in Im9 . In Helix IV , Im7 has I72 instead of Im9’s V71 . Notably , V33 , L38 , and V72 are involved in 10 of the 20 most probable nonnative contacts in the simulated Im7 intermediate listed above . Im7 and Im9 have nearly equal numbers of native contacts involving Helices I and IV ( 54 and 50 , respectively , for Im7 and 53 and 49 for Im9 ) . But the number of native contacts involving Helix II is 52 for Im7 ( residues 32 to 45 ) and 62 for Im9 ( residues 30 to 44 ) . The native contact density of Helix II is thus appreciably lower for Im7 ( 52/14 = 3 . 71 ) than for Im9 ( 62/15 = 4 . 13 ) . With lower local native-centricity and higher local hydrophobicity ( Fig 7 ) , Im7’s Helix II—which contains two more hydrophobic residues than Im9’s as shown by the sequences at the bottom of Fig 7 ( see also discussion above ) — is more prone to nonnative contacts than Im9’s Helix II . Indeed , Helix II is involved in all of the 20 most probable nonnative contacts in the simulated Im7 intermediate . We emphasize that the critical factor here is the local native contact density of Helix II but not necessarily the overall native contact density of the protein . Im7 has fewer native contacts than Im9 ( 154 versus 164 ) in our models; yet the simulated Im7 intermediate remains essentially unchanged even if the number of Im7 native contacts is increased to 161 by using Swiss-PdbViewer [69] to construct additional contacts in its less ordered N-terminal region . Moreover , the trend seen here is not limited to our specific definition of native contacts . To assess the robustness of our inference , we have also applied the CSU software , which employs detailed analysis of interatomic contacts and interface complementarity to determine native contacts [74 , 75] . Under the CSU criterion , the total number of native contacts is very similar for Im7 and Im9 ( 177 and 180 respectively; see upper-left map in Fig 7 ) . Nonetheless , similar to the observation above , the local density of CSU-defined native contacts of Helix II is also appreciably lower for Im7 ( 59/14 = 4 . 21 ) than for Im9 ( 67/15 = 4 . 47 ) . Experiments on Im9 have shown that V37L/V71I and V37L/E41V/V71I can lead to three-state folding [15 , 21] and folding-arm rollover at pH 7 . 0 and 10°C [21] . Computationally ( S4 Fig ) , these mutations deepen somewhat the shallow minimum at Q ≈ 0 . 85 in the Im9 kinetic profile ( A and C of S4 Fig ) . But the effect is insufficient to account for experimental data , indicating that further effort is needed to better model the balance between native and nonnative interactions in Im9 . For instance , if the native interaction strength of L33 ( which acts as a “gatekeeper” [76] ) in Helix II was reduced , much deeper Im9 kinetic traps would develop for V37L/V71I and V37L/E41V/V71I ( B and D of S4 Fig ) . Although our present model does not address mutational effects on native interactions , this result indicates nonetheless that L33 mutations that reduce the native interaction strengths ( e . g . , by substituting it with a less hydrophobic residue ) may lead to less cooperative folding of Im9 . This suggested behavior should be testable by future experiments . The above analysis of the interplay between local native contact density and hydrophobicity suggests that the different folding kinetics of wildtype Im7 and Im9 may also be seen in variants of the homogeneous db+hϕ model ( KHP = 1 as defined in Methods ) with stronger nonnative hydrophobic interaction strengths ( KHP > 1 ) . Consistent with this idea , S5 Fig shows that a signficant folding intermediate population starts to develop at KHP = 1 . 3 for Im7 but no corresponding folding intermediate is observed for Im9 at the same KHP . Two comments are in order here . On one hand , the result in S5 Fig from an alternate formulation of hydrophobicity reinforces our general notion that local native contact density and hydrophobicity are the main physical underpinnings for the Im7-Im9 kinetic difference . On the other hand , a strength of ≳ 1 . 3 for the homogeneous nonnative hydrophobic interaction is needed to achieve the desired Im7-Im9 difference , whereas the heterogeneous nonnative hydrophobic interaction strengths in the db+MJhϕ model that produce a similar effect average only to 1 . 0 ( see Methods; note that even at KHP = 1 . 3 , the minimum at Q ≈ 0 . 85 in ( A of S5 Fig ) is shallower than that in Fig 4C ) . Physically , KHP ≳ 1 . 3 is problematic because it implies that nonnative interaction strength is ≳ 30% higher than native interaction strength . For this reason and considering the obvious limitation of the homogeneous approach that it cannot address effects of mutations among hydrophobic residues , the more refined db+MJhϕ approach is adopted in the present study . The Im7/Im9 system is instructive in elucidating nonnative effects and kinetic trapping in the diffusion picture of protein folding [54–59] . Conformational diffusion models with a coordinate and stability-dependent diffusion coefficient on a one-dimensional free energy profile were constructed for two-state-like [57] and downhill [58] folding; but corresponding modeling for folding with a significant chevron rollover has not been much explored . In this regard , it is noteworthy that the rollover in our Im7 db+MJhϕ model appears across only ≈ 8% variation in interaction strength ( ϵ/kBT = 1 . 37 and 1 . 48 , respectively , for midpoint and zero denaturant ) . In contrast , rollover-like features for two-state-like and downhill folders emerge over much wider ranges of interaction strength [58] . The restraining-potential method [56 , 58] in Methods is used to compute Q- and ΔG-dependent autocorrelation function CQ ( t ) ( Fig 8 ) and diffusion coefficient D ( Q ) ( Fig 9 ) . The restraining-potential method directly addresses the escape probability from a given Q . Rather than seeking a good fit by Bayesian analysis [55] , we adopt this method to explore possible limits of the diffusion picture by testing the consistency between diffusive accounts of restrained and unrestrained chain kinetics . The most notable Im7/Im9 difference presents itself around the Im7 kinetic trap at Q ≈ 0 . 8–0 . 9 . Here a dramatic deepening of a dip in D ( Q ) with increasing native stability is seen for Im7 but not for Im9 , whereas D ( Q ) for other Q-values is not very sensitive to ΔG ( Fig 9 ) . Achieving numerical convergence of the computed D ( Q ) in the Q ≈ 0 . 85 region of Im7 is difficult because of kinetic trapping . To delimit theoretical possibilities , we obtain lower and upper bounds of D ( Q ) for Im7 in this region , respectively , by initializing restrained runs from kinetically trapped and random conformations ( Fig 9 ) . Im7 chevrons may now be computed in the diffusion model; but considerable variation ensues ( shaded area in Fig 10A ) because of numerical uncertainties . The rollover trend of the simulated Im7 chevron is among the predicted possibilities . However , when matched against explicit-chain kinetics , D ( Q ) is found to be underestimated by an overall factor of e2 . 7 ≈ 15 ( Fig 10 ) , indicating that the method for computing D ( Q ) [56 , 58] needs to be improved or that a one-dimensional diffusion perspective is of limited applicability here . Despite these uncertainties , it is clear that a D ( Q , ΔG ) that decreases exponentially with ΔG at the trap position Q ≈ 0 . 85 ( Fig 10B ) is necessary to reproduce the folding-arm rollover for Im7 ( Fig 10A , circles ) . The required variation of D ( Q , ΔG ) at this position , which spans two orders of magnitude , is reassuringly consistent with the lower bound estimated by initiating restrained runs from the kinetic trap . In the absence of such a strong dependence of D ( Q , ΔG ) on ΔG , the predicted folding arm would become essentially linear ( top dashed line in Fig 10A ) . In the same vein and consistent with our explicit-chain results ( Fig 3 ) , no folding-arm rollover is produced by the diffusion model for Im9 . To recapitulate , our explicit-chain models account physically for the strikingly different folding kinetics of Im7 and Im9 in terms of prevalent nonnative interactions among large hydrophobic residues in Im7 but not in Im9 . The proteins’ different experimental chevron behaviors are rationalized by our simulation . The same phenomenon may also be described by a one-dimensional diffusion process with a very small and strongly stability-dependent diffusion coefficient at the position of the Im7 kinetic trapped intermediate . Our model interaction schemes are tentative [13 , 46] . For instance , possible contributions to nonnative interactions from electrostatic [48 , 51 , 77] and aromatic [73] effects are not taken into account . Nonetheless , by comparing different modeling schemes as controls and contrasting Im7 and Im9 behaviors under the same scheme , we arrive at a physical picture that is largely in agreement with experiment . As observed experimentally [65 , 78] , Helices I and IV are essentially formed while Helix II is partially formed in our simulated Im7 intermediate . Kinetic effects of many mutations in our model are consistent with experiment , including those involving Helix III ( Figs 3–7 ) , demonstrating the versatility of the hydrid modeling approach to nonnative effects [13] . Several limitations of our model are noted . In particular , the short Helix III is present in our simulated Im7 intermediate but experimentally that is apparently not the case [65 , 78] . To address this issue , more sophisticated treatments of local conformational propensity [36 , 79] and sidechain effects [13 , 42] are probably needed . Indeed , the rich repertoire of experiments on the Im7/Im9 system , such as those on pH [18 , 21] and temperature [15] effects , offers ample data for testing extensions of our models . Perhaps the most useful insight from the present effort is that the peculiar folding kinetics of Im7 vis-à-vis that of Im9 is closely related to their difference in the balance between local native contact density and hydrophobicity . This principle embodies a competition between native topology and nonnative interactions [49] and is likely applicable to protein dynamics and biomolecular processes in general . As such , it should be examined in detail and extended to other forms of nonnative interactions in future investigations .
Three related Cα chain models for Im7 and Im9 are considered , namely the db , db+hϕ , and db+MJhϕ models . Common to these models is a set of native-centric interactions with desolvation barriers for each protein . Folding and unfolding kinetics is simulated by Langevin dynamics [80] . Desolvation barrier ( db ) is a robust feature in hydrophobic interactions [81] that tends to enhance folding cooperativity [40 , 82] . Indeed , for some proteins such as ribosomal protein S6 , Cα models with db lead to highly cooperative folding behaviors that are consistent with experiments [49] whereas models without db exhibit only weak folding cooperativity [76] . Here , following Ref . [80] , the pairwise db energy is defined by a contact minimum well depth of ϵ = 1 . 0 , a db height of ϵdb = 0 . 1ϵ , and a solvent-separated minimum well depth of ϵssm = 0 . 2ϵ . The db model is purely native-centric with the total interaction potential , denoted here as EN , equal to Vtotal in Ref . [80] . The same interaction strength is applied to all native-centric interactions . The native contact sets for Im7 and Im9 are constructed using the same criterion [80] . A pair of residues i , j belongs to the native contact set if at least one pair of their non-hydrogen atoms , one from each residue , are less than 4 . 5 Å apart in the Protein Data Bank ( PDB ) structure . The PDB Cα separation between i , j is denoted by r i j n . The total number of native contacts in the set , Q ˜ n , is equal to 154 and 164 , respectively , for Im7 and Im9 ( Fig 7 ) . We have explored using alternate “flavored” native-centric interaction strengths [72 , 83] in accordance with the residue-dependent contact energies in Ref . [71] but , interestingly , the resultant models for Im7 and Im9 fail to fold cooperatively . Favorable nonnative interactions are included in db+hϕ and db+MJhϕ . Using a hybrid formulation [13 , 43–51 , 84–92] , the total interaction potentials ET of these models are given by ET = EN + EHP , where E HP = ∑ i n ∑ j = i + 4 n K HP κ i j exp [ − ( r i j − σ h ϕ ) 2 ] is the sum of sequence-dependent nonnative contact energies over i , j that are both hydrophobic ( hϕ ) , defined to be the eight amino acids A , V , L , I , M , W , F , and Y [47] . rij is the Cα distance between i , j during simulation ( 1 ≤ i , j ≤ n , where the total number of residues n = 87 for Im7 and n = 86 for Im9 ) ; and σhϕ = 5 . 0 Å . The nonnative hϕ interactions in the db+hϕ model are homogeneous with κij = −1 . 0 irrespective of hydrophobic residue type and KHP = 1 . 0 as in Refs . [47 , 49] , whereas the nonnative hϕ interactions in the db+MJhϕ model are heterogeneous , with κij = Δϵij where Δϵij is the contact energy in Table V of Miyazawa and Jernigan [52] and KHP = 1 . 8 such that the average hϕ energy KHP⟨κij⟩ over all possible 8 × 7/2 + 8 = 36 hϕ pairs is equal to −1 . 0 ( the KHP κij values range from −0 . 216 for A-A to −1 . 584 for F-F ) . This average hϕ interaction energy of −1 . 0 is essentially maintained by the average hϕ energies over all possible nonnative hϕ contact pairs ( defined below ) for the Im7 and Im9 sequences in the db+MJhϕ models . Those average energies are equal to −0 . 994 for wildtype Im7 ( 412 possible nonnative hϕ pairs ) and −0 . 998 for wildtype Im9 ( 306 possible nonnative hϕ pairs ) . MJ-type potentials [52 , 71] are derived from the statistics of native contacts in the protein structure database . Because protein native structures do not contain many significantly unfavorable contacts , MJ potentials are not expected to describe repulsive interactions between amino acid residues with accuracy [93] . Nonetheless , they do provide a crude account of the relative strengths of favorable physical interactions between residues . In fact , it has long been known that MJ potentials for nonpolar pairs reflect the combined hydrophobicities of the two contacting residues [94 , 95] , as is illustrated by the good correlation ( Fig 3b of [96] ) between a set of MJ energies [71] and the experimental octanol-water transfer free energies of amino acids [53] . In this regard , although there are considerable variations among experimental hydrophobicity scales for all twenty types of amino acids [96 , 97] , a higher degree of consistency among different experimental scales is seen for the hydrophobic ( nonpolar and non-charged ) amino acids themselves [98] . Taking these considerations together , we view MJ energies between nonpolar residues as a reasonable coarse-grained model of the underlying physics of hydrophobicity . Thus , they should be applicable to favorable nonnative hydrophobic interactions and represent a more refined model than those with homogeneous hydrophobic interaction strengths . In our models , two hydrophobic residues i , j that are not in contact in the native PDB structure are considered to be in a nonnative contact if |i − j| > 3 and rij < 8 . 0 Å ( Fig 5 ) . The total number of nonnative contacts in a conformation is denoted by nHP ( S1 Fig ) . All non-bonded energies in our models are temperature independent and pairwise additive . For simplicity , temperature dependence and nonadditity of interactions [99–102] are not considered here . We consider a residue pair i , j in the native contact set to be in contact during the folding/unfolding simulation when r i j ≤ r i j n + 1 . 5 Å; i . e . , when rij is not larger than that of the db and therefore within the attractive well of the contact minimum . We use Q , the number of native contacts divided by Q ˜ n , as progress variable of folding [103 , 104] . A free energy profile in units of kBT corresponds to −ln P ( Q ) where P ( Q ) is the normalized conformational population distribution as a function of Q ( Figs 1 and 2 ) . As was introduced before [59] , the kinetic folding path ( FP ) profiles , −ln PFP|s ( Q ) , is the negative logarithm of average fractional resident time PFP as a function of Q along folding trajectories wherein the notation “|s” indicates that equal weight is assigned to every folding trajectory [59] ( Figs 4 and 6 ) . Chevron plots are simulated using change in native stability by varying the simulation temperature as a proxy for variation of denaturant concentration [105] ( Fig 3 ) . With a low Langevin viscosity , this approach is computationally efficient and is appropriate for our present purpose because the trend ( shape ) of model chevron rollover is apparently unaffected by variation over a wide range of Langevin viscosities [101] . Recent tests also indicate that the model chevron plots thus obtained are very similar to those simulated using more sophisticated coarse-grained sidechain models that account for denaturant dependence by experimental transfer free energies [13 , 41 , 42] . We use the restraining ( bias ) potential method [55 , 56 , 58 , 106] to estimate Q-dependent diffusion coefficients at different simulation temperatures ( hence different free energies of folding ΔG ) . Following Ref . [55] , a Q-dependent diffusion coefficient is given by D ( Q ) = var ( Q ) τ corr ( Q ) ( 1 ) for a given ΔG . Here the variance var ( Q ) ≡⟨Q ( t0 ) 2⟩t0−⟨Q ( t0 ) ⟩t02 , where ⟨…⟩t0 denotes time averaging over different t0 values; the correlation time τ corr ( Q ) = ∫ 0 ∞ C Q ( t ) d t where the autocorrelation function [54 , 107] C Q ( t ) = ⟨ Q ( t + t 0 ) Q ( t 0 ) ⟩ t 0 - ⟨ Q ( t 0 ) ⟩ t 0 2 var ( Q ) ( 2 ) is Q-dependent . The var ( Q ) and CQ ( t ) for determining D ( Q , ΔG ) ( Figs 8 and 9 ) are estimated using bias potentials V bias ( Q , Q 0 ) = K Q Q ˜ n 2 ( Q − Q 0 ) 2 , where the prescription in Ref . [108] is used to treat Q as a continuum variable . Unless specified otherwise , KQ = 0 . 1ϵ is used with 64 different Q0 values for Im7 or Im9 . This choice of KQ is similar to that in Ref . [56] and serves to localize conformational fluctuations to Gaussian-like distributions ( S6 Fig ) . D ( Q ) is quite insensitive to lowering KQ by at least a factor of two ( S7 Fig ) . This method for determining D ( Q ) is exact if the diffusion process is truly governed by the Smoluchowski equation and KQ is sufficiently large so that variation of free energy G ( Q ) within a constrained conformational ensemble is essentially linear in Q . The applicability of this approach to protein folding , however , hinges on whether the dynamics along Q is Markovian to a good approximation [55] . For protein folding , D ( Q ) estimated by the restraining-potential method does exhibit a weak dependence on KQ [58] . We have checked our restraining-potential methodology against that of Xu et al . [58] by using a KQ value that produces conformational distributions similar to theirs . Our D ( Q ) for chymotrypsin inhibitor 2 at transition midpoint matches well with theirs ( S8 Fig ) . D ( Q ) can also be estimated using Bayesian analysis [55] . For one dipeptide system , the Bayesian-estimated D ( Q ) was verified to be very similar to that from restraining potentials [55] . Here we use only the restraining-potential method . Once D ( Q ) is in place for a given native stability ( free energy of folding ) ΔG , the folding MFPT in our nonexplicit-chain models of one-dimensional conformational diffusion ( Fig 10 ) is computed using the discretized form [59] ( M F P T ) D = ∑ Q = Q D Q N P eq ( Q ) - 1 ∑ Q ′ = 0 Q P eq ( Q ′ ) / D ( Q ) ( 3 ) of the general formula [54 , 109] ( M F P T ) D = ∫ Q D Q N d Q ∫ 0 Q d Q ′ 1 D ( Q ) exp [ G ( Q ) - G ( Q ′ ) k B T ] , ( 4 ) where Peq ( Q ) is the normalized equilibrium conformational population at Q . The boundary values QN and QD for the native ( folded ) and denatured ( unfolded ) states are the same as that in our explicit-chain simulations ( Fig 2 ) . Alternatively , MFPT can be computed using Kawasaki Monte Carlo ( MC ) dynamics by generalizing the formulation in Ref . [59] to coordinate-dependent D ( Q ) , viz . , the transition probability from Q to Q + δQ is now given by A − 1 D ( Q ) D ( Q + δ Q ) exp [ − δ G ( Q ) / k B T ] where δG ≡ G ( Q + δQ ) − G ( Q ) and A is a constant . The above geometric mean D ( Q ) D ( Q + δ Q ) may also be replaced by the arithmetic mean [D ( Q ) + D ( Q + δQ ) ]/2; the two means are equal in the limit of D ( Q + δQ ) − D ( Q ) → 0 . In addition to MFPT , Kawasaki MC is useful also for providing distribution of folding times and other properties of individual trajectories .
|
In order to fold correctly , a globular protein must avoid being trapped in wrong , i . e . , nonnative conformations . Thus a biophysical account of how attractive nonnative interactions are bypassed by some amino acid sequences but not others is key to deciphering protein structure and function . We examine two closely related bacterial immunity proteins , Im7 and Im9 , that are experimentally known to fold very differently: Whereas Im9 folds directly , Im7 folds through a mispacked conformational intermediate . A simple model we developed accounts for their intriguingly different folding kinetics in terms of a balance between the density of native-promoting contacts and the hydrophobicity of local amino acid sequences . This emergent principle is extensible to other biomolecular recognition processes .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Methods"
] |
[] |
2015
|
Native Contact Density and Nonnative Hydrophobic Effects in the Folding of Bacterial Immunity Proteins
|
Fasciola hepatica is not only responsible for major economic losses in livestock farming , but is also a major food-borne zoonotic agent , with 180 million people being at risk of infection worldwide . This parasite is sophisticated in manipulating the hosts’ immune system to benefit its own survival . A better understanding of the mechanisms underpinning this immunomodulation is crucial for the development of control strategies such as vaccines . This in vivo study investigated the global gene expression changes of ovine peripheral blood mononuclear cells ( PBMC ) response to both acute & chronic infection of F . hepatica , and revealed 6490 and 2364 differential expressed genes ( DEGS ) , respectively . Several transcriptional regulators were predicted to be significantly inhibited ( e . g . IL12 and IL18 ) or activated ( e . g . miR155-5p ) in PBMC during infection . Ingenuity Pathway Analysis highlighted a series of immune-associated pathways involved in the response to infection , including ‘Transforming Growth Factor Beta ( TGFβ ) signaling’ , ‘Production of Nitric Oxide in Macrophages’ , ‘Toll-like Receptor ( TLRs ) Signaling’ , ‘Death Receptor Signaling’ and ‘IL17 Signaling’ . We hypothesize that activation of pathways relevant to fibrosis in ovine chronic infection , may differ from those seen in cattle . Potential mechanisms behind immunomodulation in F . hepatica infection are a discussed . In conclusion , the present study performed global transcriptomic analysis of ovine PBMC , the primary innate/adaptive immune cells , in response to infection with F . hepatica , using deep-sequencing ( RNAseq ) . This dataset provides novel information pertinent to understanding of the pathological processes in fasciolosis , as well as a base from which to further refine development of vaccines .
The trematode parasite Fasciola hepatica ( liver fluke ) is the causative agent of a global disease ( fasciolosis ) that is of major health , welfare and economic importance in domestic animals ( cattle , sheep ) [1] . This parasite can also infect and complete its lifecycle in a wide range of other mammals including humans [2] . Fasciolosis is currently recognized as a major food-borne zoonosis , since human infection cases are distributed broadly , including South America , the Middle East , and Asia [3–5] , where it is estimated that 2 . 4 million people are infected every year , and 180 million people are at risk of infection [2 , 6 , 7] . Fasciolosis is initiated when the definitive hosts ingest vegetation or water contaminated with an encysted larval parasite ( metacercariae ) . Following hatching from the cyst in the small intestine , the newly excysted juveniles ( NEJ ) penetrate through the intestinal wall and into the liver . Usually from 4–6 days post infection ( acute stage of fasciolosis ) , the immature flukes migrate through the hepatic parenchyma , and their burrowing and feeding behaviors results in traumatic tissue damage ( hemorrhage ) and later repair through fibrosis during the next 5–6 weeks . The chronic stage is typically established since the flukes reach the biliary ducts where they mature to adults by 8–10 weeks post infection [8] . F . hepatica has a wide definitive host range , widespread geographical distribution and ability to survive long term in the host , which depends on sophisticated methods of modulating host immune responses to benefit parasite survival [9] . Some currently known methods employed by the parasite include polarization towards a Th2 response , suppression of Th1/Th17 responses , alternative activation of macrophages ( characterized by high arginase activity and low iNOS levels ) , induction of eosinophil apoptosis , and inhibition of dendritic cell maturation ( summarized in [9–11] ) . However the cellular and molecular mechanisms involved in these phenomena are not fully understood . Some efforts have been made to explore gene expression changes in host liver followed by F . hepatica infection , and these studies yielded important information on the molecular mechanisms involved in host physiological and pathological changes [12 , 13] . The main transcriptomic changes identified by these studies were associated with host metabolism , tissue-repair , liver injury and hepatic toxicity; however gene expression changes related to immune response were very limited . In terms of control strategies against F . hepatica , triclabendazole ( TCBZ ) is currently the first choice as a fasciolicide [14] . However , the development of resistance to this compound across Europe , Australia , and in some countries of South America [15–18] has raised concerns about the sustainable control of fasciolosis into the future . Therefore , a ‘greener’ alternative option for control , such as vaccination , is urgently needed . In the last decades , great efforts have been made to identify and test several F . hepatica antigens as vaccine candidates . For instance , cathepsin L1 proteinase plays multiple key roles in the physiological activity and parasitism of F . hepatica including digestion of blood , penetration of host tissue , evasion of host humoral immunity , egg production and immunosuppression [19 , 20] . A recombinant mutant version of this protein , rmFhCL1 , has shown protection against F . hepatica infection in cattle [21] . However , the current vaccine candidates , although showing promise , have not always been consistent in the degree of protection elicited . Monitoring gene expression changes of host immune cells during a vaccine trial would be useful to understand the vaccine effect . In addition , a more complete understanding of the ways in which F . hepatica infection modulates the immune response will also benefit the rational design of vaccines which may be able to overcome these immunoregulatory effects . Peripheral blood mononuclear cells ( PBMC ) consist of several cell types such as T cells , B cells , Natural Killer Cells and monocytes , which play a crucial role in both innate and adaptive immune response to F . hepatica infection . We hypothesized that analysis of the transcriptomic changes in the PBMC in response to vaccination and infection would reveal the molecular mechanisms underlying the immune response induced by the vaccine and host-fluke interaction . In this study , we conducted a vaccine trail on sheep and monitored the transcriptomic changes of PBMC induced by vaccine and F . hepatica infection . Although the immunization did not induce any detectable changes at transcriptional level , we identified a large number of DEGs induced by acute and chronic stage of F . hepatica , respectively . The present study is one of the first global transcriptomic analysis of ovine PBMC , in response to vaccine and infection with F . hepatica using RNA sequencing ( RNAseq ) , and provides important new information that enhances our understanding of fasciolosis from an immunological perspective .
Six-month-old lambs ( n = 8 ) used in this study were obtained from the flock at UCD Lyons Research farm , and verified by serology and faecal egg counts as free from F . hepatica infection . Animals were housed under normal husbandry conditions in fluke-free sheep-pen at UCD Lyons Research Farm . Briefly , three doses of experimental F . hepatica vaccine , a mixture of 200 μg rmFhCL1 , a recombinant protein F . hepatica Cathepsin L1 ( provided as a gift from Professor John Dalton , McGill University ) , and the adjuvant Montanide™ ISA 70 VG ( gently provided by SEPPIC ) , was administered to four lambs ( Animal IDs are V1 , V2 , V3 , and V4 ) at two-weeks intervals ( Fig 1 ) . One week after the last vaccination , these animals plus four controls ( Animal IDs are C1 , C2 , C3 , and C4 ) were orally infected with 90 F . hepatica metacercariae each ( Baldwin Aquatics , Inc ) . The effect of the vaccine was evaluated based on the measurement of fecal egg count ( FEC ) at 10 and 16 weeks post infection ( wpi ) , and fluke burden ( FB ) at 16 wpi . Blood was collected into heparinized tubes for each animal at four time-points: pre-vaccination ( Time-point 1 , T0 ) , pre-infection ( T1 , just right before inoculation of F . hepatica metacercariae ) , 1 week post infection ( T2 ) , and 14 weeks post infection ( T3 ) . PBMC were isolated from heparinized blood immediately using Histopaque ( Sigma ) . Briefly , 9 ml heparinized blood was diluted in 9 ml of complete medium ( RPMI GLUTAMAX , 1% non-essential amino acids , 10% foetal calf serum ( FCS ) , 1% Penicillin Streptomycin ) . The diluted blood was added to a Leucosep tube with 15 ml Histopaque ( Sigma ) . After centrifuging at 1304 x g for 10 min without the brake , the white PBMC layer was collected from the filter and washed in medium without FCS . The cell pellet was then re-suspended in 1 ml red cell lysis buffer for 1 min . Following two washing steps , the pellet was finally re-suspended in 4 ml complete medium . Cells were counted and their viability checked by trypan blue staining . Total RNA was extracted from PBMC immediately by using E . Z . N . A Total RNA Kit ( Omega ) , according to the manufacturer’s’ instructions . Purified RNA was treated with RNA-free DNase I ( Qiagen ) . RNA quantity was evaluated using Nanodrop 1000 spectrophotometer ( Thermo Fisher Scientific ) , and RNA quality using an Agilent 2100 Bioanalyzer with an RNA 6000 Nano LabChip Kit ( Agilent Technologies ) . Only samples with A260/280 ratios > 2 . 0 and RNA Integrity Numbers ( RIN ) of ≥ 7 . 8 were used for sequencing . Thirty-two RNAseq libraries were prepared using TruSeq Stranded mRNA Sample Preparation Kit ( Illumina ) following the manufacturer’s instructions . The concentration of each amplified library was measured using Qubit assays ( Life Technologies ) and the size distribution was assessed on a Bioanalyzer using the DNA1000 kit ( Agilent Technologies ) . Library concentration was normalized to 10nM and pooled for multiplex sequencing . The thirty two RNA-seq libraries were prepared in 4 pools and delivered to the Research Technology Support Facility , Michigan State University , for sequencing . After quantitation and validation by Qubit ( Life Technologies ) and quantitative real-time PCR ( KAPA Biosystems ) each pool was loaded on a single lane of an Illumina HiSeq 2500 Rapid Run flow cell ( v1 ) . Sequencing was performed using TruSeq Rapid SBS reagents ( Illumina ) in a 2x100bp ( PE100 ) format . All RNAseq data has been made available via the NCBI GEO repository [22] , under the accession number GSE71431 . The first stage in sample processing was quality evaluation of the unprocessed FASTQ files using the software FastQC v0 . 10 . 0 . After filtering out adapter sequence reads and removing poor quality reads , the clean fastq files were aligned to the sheep genome ( Oar v3 . 1 , available in the ENSEMBL website [23] ) using the alignment software STAR [24] version 2 . 3 . 0e_r291 ( the parameter used is described in S1 Table ) . The gene annotation for the sheep genome ( Oar3 . 1 Ensembl release 79 ) was downloaded from ENSEMBL . After alignment , the gene counts were generated using the software featureCounts [25] from the Subreads software , package version 1 . 3 . 5-p4 . Transcripts per million ( TPM ) were calculated based on a simple library size normalization . In order to visualize the overall structure of the data and to identify potential outliers or mislabeled samples of the data , Principal Components Analysis ( PCA ) and Between Groups Analysis ( BGA ) was carried out . The BGA plots were generated using the MADE4 package [26] in an R/Bioconductor [27] . Differential gene expression was calculated using the Limma Voom package [28] . Normalization factors were calculated from the raw counts using EdgeR [29] . The design matrix was generated using Puma [30] fitting main effects of “group” ( vaccinated or control ) and time along with interactions of “group” with “time” . The “animal ID” was treated as a random effect and “time” was treated as a discrete variable . The empirical Bayes function in Limma [31] was used to determine the significantly expressed genes . A false discovery rate ( FDR ) of 5% was chosen as a cut-off . The lists of differentially regulated genes for different comparisons were annotated using the R/bioconductor BiomaRt package [32] . Functional pathway analysis was conducted through the use of IPA ( Ingenuity Pathway Analysis , QIAGEN ) . Since the sheep is not currently a supported species in IPA; human orthologues for the relevant sheep genes were used . The human orthologue information was downloaded from BiomaRt and merged with the significant genes . In cases where a single sheep gene was annotated to multiple human genes just one human orthologue was retained for the purposes of pathway analysis . The lists of DEG were used as the input for IPA analysis . ESEMBL ovine ID numbers for genes mentioned in the text were list in Table 1 . The curated canonical pathways from Ingenuity Knowledge Base that were enriched in the DEG dataset were determined by using a right-tailed Fisher’s exact test . The enrichment p-value is calculated by assessing the probability of a pathway being randomly selected from all of the curated pathways , and a cut-off of p-value ≥ 0 . 05 was used as the threshold for significant pathways . The IPA “upstream analysis” feature was used to analyze the upstream regulators . The Ingenuity Knowledge Base was used to predict the expected causal effects between upstream regulators and DEG targets . By calculating an overlap p-value and an activation z-score , the analysis gives a prediction of the status of the upstream regulator . Fifteen differentially expressed genes ( DEGS ) involved in immunological or signaling pathways were selected for validation using quantitative real-time PCR ( qPCR ) . cDNA was synthesized from 250ng total RNA from each sample used for RNAseq library preparation using the High-Capacity cDNA Reverse Transcription Kit ( Applied Biosystems ) , according to the manufacturer’s instructions . In a 20 . 0 μl reaction , the cDNA was diluted 1:20 with distilled water prior to use . Primers were designed based on the corresponding cDNA sequences ( obtained from ESEMBL ) , and primer efficiencies were determined using a 1:4 dilution series over 7 points , and efficiencies for all primers were between 95–110% ( S2 Table ) . RT-qPCR was performed using SYBR Green Master Mix ( Applied Biosystems ) on a 7300 Real-Time PCR System ( Applied Biosystems ) . 20μl reaction volume contained 5μl diluted cDNA samples ( or appropriate controls ) , 10μl of SYBR Master Mix and 300 nM final concentration of each primer . RT-qPCR reaction was performed as following cycling parameters: 10 min at 95°C ( heat-activation step ) ; 40 cycles of 15 sec at 95°C , 1 min at 60°C . The specificity was confirmed by melt curve analysis . Using the GeNorm algorithm within the qBase+ software package ( Biogazelle ) the stability of 8 potential reference genes ( including GUSB , ATP , PGK1 , GAPDH , B2M , ACTB , TBP and RPL19 ) was assessed ( see the corresponding gene IDs in Table 1 ) . RPL19 and TBP were shown to be the most stability expressed ( M<0 . 15 ) [33] . Calibrated normalized relative quantities ( CNQR ) were calculated using the qBase+ software [34] . Normal distribution of fold-change values was checked using the Shapiro-Wilk test in the SPSS statistical package ( IBM Corp ) . Two-tailed paired sample t-tests was used to assess statistically significant gene expression fold-changes .
There was no significant difference in faecal egg count ( FEC ) or fluke burden ( FB ) between vaccinated and control animals , indicating no protective effect elicited by vaccination in this trial ( S1 Fig ) . The 32 RNAseq libraries ( representing vaccinated ( n = 4 ) and control ( n = 4 ) groups from 8 animals at four time-points T0-T3 ) were sequenced on an Illumina HiSeq 2500 platform and generated mean values per library of 20 . 3 million paired-end reads , of which 16 . 0 million reads ( 81 . 8% ) uniquely mapped to ovine genome ( S3 Table ) . Post alignment quality control ( QC ) analysis shows no evidence of poor quality or outlier libraries ( S2 Fig ) , and aggressive normalization ( e . g . quantile normalization ) was not required . In total 21236 genes are annotated in the Ensembl sheep gene annotation . Genes having a raw count of fewer than 10 reads in fewer than 4 samples were filtered out and removed from the analysis . A total of 13309 genes were retained and used for subsequent differential expression analysis . Both PCA and BGA were performed in order to carry out a preliminary separation of samples . The main source of variation between the samples is infection as can be seen from the PCA plot in ( Fig 2A ) . The first principal component separates pre-infection samples ( T0 + T1 ) from post-infection samples ( T2 + T3 ) . There is a trend on the second axis with the vaccinated animals shifted slightly along the axis compared to the controls . There is however a lot of overlap between the groups . Using the supervised clustering of BGA ( between groups analysis ) the same general trends are evident i . e . the biggest effect is the infection effect with T0+T1 and T2+T3 clustering together ( Fig 2B ) . Further differential expression analysis identified very few DEGS between vaccinated and control groups for each time-point individually . Only 12 DEGS were detected at T2 , and none was found at any other time-points . When the three post-vaccination time-points ( T1-T3 ) were combined together for this analysis , 36 DEGS ( S4 Table ) were identified between vaccinated and control animals , due to the increased power given the bigger sample sizes . However manual inspection of these ‘vaccination effect’ genes showed that these 36 genes were systematically different in these animals and were also differentially expressed in the time-point 0 ( T0 ) samples . The samples designated vaccinated ( V ) were from animals infected with F . hepatica metacercariae at time-point 1 ( T1 ) therefore there is no experimental difference between control and vaccinated animals at T0 . It is likely that these differences instead represent inter-animal variation ( likely to be expression quantitative trait loci , eQTLs ) and are not relevant to the vaccination effect . Therefore the 36 DEGS representing inter-animal variation were excluded from further analysis . Since the biggest effect was infection of F . hepatica rather than vaccination , and in order to increase the statistical power and accuracy for the subsequent analysis [35] , we combined the vaccinated and control samples together from each time point , giving a sample size of 8 biological replicates per time-point . To explore the gene expression changes in the acute stage of infection , we compared the data from T2 with T1 , and obtained 3134 and 3356 significantly up-/down- regulated DEGS respectively ( FDR-adjusted p-value ≤ 0 . 05 ) ( S5 Table ) . In addition , 1248 significantly up-regulated and 1116 significantly down-regulated DEGS were detected in T3 compared to T2 ( S5 Table ) , which represent the transcriptomic diversity between chronic and acute stage of infection . Notably , the DEGS from both comparisons were all equally distributed between up- and down-regulated . This differs from other transcriptomic studies based on liver tissue from animals infected with F . hepatica in which the DEGS were biased to up-regulation [12 , 13] . DEGS from T2vsT1 and T3vsT2 were compared according to the direction of expression ( Fig 3 ) . 5398 DEGS ( 2740 up-regulated and 2928 down-regulated ) were observed in T2vsT1 only . 1533 DEGS ( 811 up-regulated and 722 down-regulated ) were detected uniquely in the comparison of T3vsT2 , demonstrating that gene expression level was significantly regulated by chronic rather than acute F . hepatica infection . 428 DEGS ( 215 up-regulated and 213 down-regulated ) were observed as occurring in both T2vsT1 and T3vsT2 , and displayed the same direction of expression . 182 genes were up-regulated at the acute stage but subsequently down-regulated at the chronic stage of infection , while 223 genes were oppositely regulated over the time course . S2 Table summarizes the results of qRT-PCR data for the 15 selected DEGS from several canonical pathways of interest . Although the fold change values for the expression of some genes measured by RNAseq or qRT-PCR were quite different , the gene expression patterns of all DEGS in terms of fold-change direction and statistical significance were reproducible by qPCR analysis . DEGS from T2vsT1 and T3vsT2 were fitted into known ‘canonical’ pathways of an IPA database to explore the potential PBMC cellular pathways modulated following F . hepatica infection . In total , 182 and 55 canonical pathways were significantly enriched at T2vsT1 and T3vsT2 , respectively ( FDR-adjusted p-value ≤ 0 . 05 ) ( S6 Table ) . The top 10 canonical pathways from two comparisons are shown separately in Table 2 . Notably , a number of these pathways in T2vsT1 are associated with innate host-defense mechanisms , including nitric oxide production in macrophages ( ranked 1 ) , IL-6 signaling ( ranked 3 ) , phagosome formation ( ranked 4 ) , and toll-like receptor signaling ( ranked 10 ) . Partial pathway graphs of ‘TGF-β signaling’ and ‘Nitric oxide production in macrophages’ are shown in Figs 4 and 5 , respectively . In contrast , ‘antigen presentation’ ( ranked 3 ) is the only adaptive immune-related pathway in the top-10 overrepresented pathways in T3vsT2 . The relative expression level of selected genes from pathways of interest are presented in Table 3 . DEGS and IPA upstream regulator analysis were also used to predict the upstream transcriptional regulators that may be responsible for gene expression changes observed during infection , based on prior knowledge stored in the IPA gene database . Upstream regulators are defined as any molecules that can affect the expression of another molecule , including transcription factors , cytokines and micro-RNAs . Several predicted regulators associated with the adaptive immune response are shown in Table 4 . Some Th1/Th17-associated and pro-inflammatory cytokines ( including IFN-γ , TNF , IL15 , IL12 ( complex ) , IL17F and IL18 ) as upstream regulators were predicted to be inhibited with z-score < -2 . Fig 6 shows the interaction of downstream target genes regulated by IL18 and IL12 ( complex ) , which highlight the overlapped downstream genes with the same direction of activation ( e . g . , IFNG , CCL3 , CCR5 , CD244 , CSF1 , CSF2 , CXCL10 , CXCL8 , HLX , IDO1 , IL18BP , IL18RAP , IL6 , NCR1 and NOS2 ) . In addition , an important regulator cytokine during F . hepatica infection , IL10 , showed a trend of activation with z-score of 0 . 423 ( overlap p-value = 3 . 84E-16 ) .
It is interesting in Table 4 that the upstream regulators IL18 and IL12 were all predicted to be dramatically inhibited with z-scores of -3 . 52 and -3 . 58 respectively . Previous studies in mice have shown that the combination of IL18 and IL12 can inhibit IL4 dependent IgG1 production , and enhance IgG2a production in B cells in vivo/vitro [39] . It is known that infection of F . hepatica induce a Th2-biased immune response characterized by high titers of specific IgG1 antibodies and virtually no specific IgG2 [40 , 41] . Notably , another important cytokine involved in IgG2a class switching , IL27 ( down-regulated in T2vsT1 ) , appears in the downstream target genes of IL12 complex ( Fig 6 ) . IL27 is produced by activated APC such as macrophages and DC [42 , 43] . Previous studies have indicated that IL27 plays a role in induction of IgG2a class switching in mouse spleen B cells activated with anti-CD40 or LPS , while IL27 inhibited IgG1 class switching induced by IL4 in activated B cells [44] . Since these supporting evidences are derived from mice , future studies should be performed to investigate the role of these cytokines in IgG1/IgG2 class-switching in sheep . We hypothesize that this isotype bias seen strongly in the immune response to F . hepatica in sheep is attributable to the attenuated regulation effect of IL18 and IL12 on B cells . Three mature miRNAs were predicted to be significantly activated ( miR-16-5p and miR-155-5p ) or inhibited ( miR-217-5p ) for downstream regulation at the acute stage of infection ( Table 4 ) . Particularly , miR-155-5p is predicted to be significantly activated ( z-score = 2 . 1; overlap p-value = 5 . 61E-08 ) with 55 predicted target genes in our dataset ( such as IFNG , IL6 , IL1A , TNF and chemokine ligands ) . The mature miR-155 ( miR-155-5p ) has been demonstrated to be one of the five major miRNAs that is specific for hematopoietic cells including B cells , T cells , monocytes and granulocytes [45] . MiR-155-5p has been shown to play a role in pathogen-induced immunity [46] by shaping the transcriptome of lymphoid cells that control diverse biological functions from inflammation to immunological memory [47 , 48] . Therefore this lends an impetus for exploring the role of miR-155-5p in F . hepatica infection . Previous studies have proposed that TGFB1 plays a central role in fibrosis during F . hepatica infection [37] . Indeed , it is well known generally that persistent TGFΒ1 signaling leads to excessive fibrosis and ultimately scarring of internal organs [49] . Through downstream SMAD signaling , TGFB1 induces the transcription of gene COL1A1 , encoding collagen type I , which is the major fibrous collagen and plays a central role in wound-healing [49] . As shown in Fig 4 , at the acute stage of infection , we found that TGFΒ1 and the related genes TGFBR1 , SMAD 3/4 were all up-regulated , and maintained thereafter ( because we didn’t detect them as DEGS in T3vsT2 ) . Importantly the COL1A1 gene was up-regulated at the acute stage of infection ( log2FC = 3 . 87 , T2vsT1 ) , and then rose up to an even higher level at the chronic stage ( log2FC = 3 . 33 , T3vsT2 ) , indicating that persistent TGFΒ1 function plays a key role in accumulation of ECM ( Extracellular Matrix ) and fibrosis during the whole course of F . hepatica infection . This is in agreement with previous finding in the fluke-infected sheep liver tissue that genes involved in fibrosis and ECM formation ( e . g . TGFΒ1 and COLIA1 ) were up-regulated at 8 wpi [13] . In addition , a powerful fibrosis-promoting molecule [50] plasminogen activator inhibitor-1 ( PAI-1 ) was also up-regulated in T2vsT1 but not in T3vsT2 , indicating its potential role in pro-fibrogenesis during the infection . On the other hand , we noticed that an inhibitory-Smad protein—SMAD7—was also up-regulated in T2vsT1 . SMAD7 is involved in a negative feedback loop which can suppress the fibrogenic progress mediated by SMAD2/3/4 signaling [51–53] , indicating that SMAD7 may play a role in controlling fibrosis during the early stage of ovine fasciolosis . Usually during chronic liver injury , the transcription of SMAD7 is blocked by the pSmad3L pathway [54] , and the lack of SMAD7 induction might lead to constitutive fibrogenic TGFΒ1 production [53] . In this study , SMAD7 expression was maintained at a similar level at the chronic stage of infection , since SMAD7 does not appear in the list of DEGS of T3vsT2 . This indicates that SMAD7 may play a negative regulatory role in fibrosis formation during the chronic stage of ovine fasciolosis . Considering that fibrosis is often limited in ovine as compared with bovine fasciolosis , it would be interesting to investigate the transcription of SMAD7 during F . hepatica infection in cattle . We hypothesize that the expression level of SMAD7 in PBMC of cattle would be significantly lower at the chronic stage than that the acute stage of F . hepatica infection , while the expression of COL1A1 gene would significantly increase from acute to chronic infection . NOS2 , encoding inducible nitric oxide synthase ( iNOS ) was extremely down-regulated ( log2 FC value was -27 . 4 , p < 0 . 05 ) at the acute stage of infection , and ranked at the top of all down-regulated genes ( S5 Table ) . This downregulation was seen also at the timepoint representing the chronic stage of infection ( since NOS2 was not detected in DEGS data T3vsT2 ) . iNOS is able to convert arginine into citrulline and nitric oxide ( NO ) , which is considered a defense mechanism in response to pathogen invasion [55] . This result is consistent with previous observation that ovine macrophages failed to generate nitric oxide when incubated with NEJ of F . hepatica in vitro [56] . Regulation of iNOS at the transcriptional level is complex and cell- / species- specific [57] . While there are numerous transcription factors involved in iNOS expression , activation of the IFNγ-regulated transcription factor STAT-1α and thereby activation of the iNOS promoter have been demonstrated to be the essential step for iNOS induction in murine , rat and human cells , and all mammalian iNOS promoters contain several homologies with STAT-1α binding sites ( GAS , Interferon-Gamma-Activated Sequence ) [57 , 58] . It was observed that iNOS induction in murine macrophages was blocked with a disrupted STAT1 gene [59] . In man , the inhibition of iNOS expression by various compounds has been attributed to the inhibition of the JAK-STAT-1 pathway [57] . The data described here shows that the gene expression of IFNγ , IFNγ-receptor , STAT1 and JAK were all significantly down-regulated at T2 compared to T1 ( Fig 5 ) . Although the mechanism of iNOS induction in ovine cells has not yet been described , our results indicate that the attenuated IFNγ-JAK/STAT pathway is likely account for the extreme down-regulation of iNOS during F . hepatica infection . Previous studies have shown that F . hepatica is capable of generating alternatively-activated macrophages ( AAM ) characterized by having high arginase activity and low levels of iNOS [11] . We hypothesize that this dramatic down-regulation of iNOS is mainly attributed to alternative activation of macrophages during infection . Since iNOS can be expressed not only in macrophages but also in other immune cells such as CD4+ T cells [60] , an interpretation of the suppression of iNOS in this study needs to consider the role of other cells in PBMC population . Previous vaccine studies have suggested that iNOS expression and subsequent NO production is important for an effective host response against the early migrating liver fluke [61] . It will be worthwhile , therefore , to investigate the relationship between the IFNγ-JAK/STAT pathway and iNOS induction in future vaccine/challenge trials with F . hepatica . Toll-like receptors ( TLR ) , belonging to the family of pathogen-associated pattern recognition receptors , are usually expressed in sentinel cells and play a key role in the innate immune system . Previous studies suggest that some parasites , for instance , Leishmania , Entamoeba and Trypanosoma can down-regulate TLR expression [62] . Our results showed that F . hepatica infection significantly downregulated the mRNA expression of TLR1 , TLR5 , TLR6 , TLR7 and TLR10 in PBMC at the acute stage of infection ( T2 ) . The transcription of TLR1 , TLR5 and TLR7 then rose again to a certain extent at the chronic stage of infection ( T3 ) ( Table 3 ) . As the potential transcriptional regulators , the activation states of TLR1 , TLR3 , TLR4 , TLR5 and TLR7 were predicted to be inhibited ( Table 4 ) , indicating an attenuated role of them in downstream regulation . Previous studies have suggested that helminth infection can modulate the expression level of TLR and/or their function . For example , stimulation with live microfilariae of Brugia malayi can significantly downregulate the mRNA expression of TLR3 , TLR4 , TLR5 and TLR7 in human monocyte-derived DC [63] . In cattle , the excretory/secretory products of F . hepatica restricted TLR2/TLR4-mediated activation , which may control excessive inflammatory-induced pathology during F . hepatica infection [64] . Surprisingly , in our study , TLR4 mRNA expression increased to a peak at the acute stage , then decreased to original levels at the chronic stage . In general , the present results suggest that F . hepatica may suppress TLR pathways in the host by down-regulating the expression of TLR1 , 5 , 6 , 7 and 10 , in particular during acute stage , which may be an important evasion strategy . Apoptosis of immune cells has been considered as an immunosuppressive strategy used by F . hepatica during infection . [65] demonstrated in vitro that F . hepatica-derived ES was able to induce apoptosis in eosinophils ( Eo ) by a caspase-dependent mechanism . Further studies suggested that Fh-derived ES induce Eo apoptosis via ROS-mediated mitochondrial-membrane potential loss . In addition , F . hepatica-derived ES has also been observed to induce apoptosis in peritoneal macrophages in vitro , but the mechanism remains unclear [66] . As shown in S6 Table , the z-scores of the pathways ‘Death Receptor Signaling’ and ‘Apoptosis Signaling’ were 1 . 5 and 1 . 3 respectively in T2vsT1 ( with p-value < 0 . 05 ) , indicating an overall increase in the activity of both pathways in T2 compared to T1 . Therefore , F . hepatica may be capable of inducing apoptosis in PBMC during infection . In general , apoptosis is mainly induced through two distinct pathways , the death receptor pathway ( extrinsic ) and the mitochondrial pathway ( intrinsic ) [67] . For the initial step of the extrinsic apoptosis pathway , most death receptors and their corresponding ligands were up-regulated in T2vsT1 including TNF-alpha/TNFR1 , TNF-alpha/TNFR2 , APO3L/DR3 and death receptor 4 , 5 , 6 , while the FAS receptor was down-regulated ( Table 3; S3 Fig ) . TNF-α is expressed in a wide range of cells but mainly produced by macrophages [68] , and is the major extrinsic mediator of apoptosis . Previous in vitro studies suggest that fluke-derived molecules such as fatty acid binding protein ( Fh12 ) may suppress the expression of TNF-α in macrophages [69 , 70] . However , this is not inconsistent with our result showing up-regulation of TNF-α expression at one wpi due to the previous observation of the mixed Th1/Th2 response at the early stage of infection . The up-regulation of both TNF-α and TNFR1/TNFR2 suggest a possibility that F . hepatica could induce the extrinsic apoptosis pathway in PBMC via TNF-TNFR , rather than by the Fas/Fasl model . In the intrinsic pathway of apoptosis , mitochondria release pro-apoptotic proteins [71] . Previous studies suggest that an activated intrinsic pathway is involved in Fh-induced Eo apoptosis , as indicated by the increase level of pro-apoptotic protein cytochrome c [10] . This is consistent with our data , in that the transcription level of the CYCS gene , encoding cytochrome c , was up-regulated at one wpi , and increased to an even higher level at 14 wpi . However , other pro-apoptotic proteins released by mitochondria were down-regulated , including DIABLO , AIFM1 and DFFB ( S5 Table ) . Both the extrinsic and intrinsic apoptosis pathways are caspase-dependent . Previously , it has been shown that in FhESP-induced apoptotic Eo , the activation of caspase -3 , -8 and -9 is increased [10] . Caspase-8 is the major initiator in the extrinsic apoptotic pathway , and may also mediate changes in mitochondrial function [72] . Surprisingly , in our study , the transcription of Caspase-8 was down-regulated at one wpi . On the other hand , the most important executioner , caspase-3 , was up-regulated at the chronic stage of infection ( Table 3 ) . There was no obvious trend in genes governing the regulation of the pro-apoptotic targets of caspases which are associated with the various morphological changes occurring during apoptosis . For example , the downstream effector genes ACTB ( ACTIN ) and SPTAN1 were up-regulated in T2vsT1 , while LMNA , and DFFB were down-regulated , indicating a mixed-effect on apoptosis in PBMC ( S5 Table; S3 Fig ) . In summary , our data suggest that F . hepatica infection may have an overall role in the activation of PBMC apoptosis . Both extrinsic and intrinsic pathways may be involved and may work synergistically in infected animals . TNFα/TNFR seem to be involved in the initiation of extrinsic apoptosis . Cytochrome c may play a persistently pro-apoptotic role via the intrinsic pathway during infection . However , our data was obtained using a complex PBMC population , it is not surprising that a complex pattern of relevant gene changes was observed . Apoptosis is clearly occurring during F . hepatica infection , but it was not possible to pinpoint the exact mechanism involved . Taking into account that different mechanisms may be involved in apoptosis of diverse cell species , further study directed at the apoptosis of individual cell types is an obvious next step . Th17 is a relatively newly recognized subset of T helper cells distinct from Th1 and Th2 lineages , characterized by the expression of the RAR-related orphan receptor ( ROR ) family transcription factor RORγt , IL23 receptor ( IL23R ) , and producing effector cytokines including IL17A , IL17F , IL21 and IL22 [73] . The role of IL17 cells in helminth-driven immune responses is still not fully understood . Studies with Schistosoma species , for example , have indicated that the Th17 response contributes significantly to severe immunopathology [74 , 75] , and likely in an IL23 dependent manner [76 , 77] . Additionally , several studies on nematode parasites also support the robust association of Th17 and IL17 with pathology [78–80] . To date , there is no consistent evidence that Th17/IL17 have a protective effect against helminth infection [81] . In case of Teladorsagia circumcincta infection in sheep , the Th17 cytokine expression correlates with susceptibility to infection [82] . During Schistosoma haematobium [83] and Echinostoma caproni [84] infection , the Th17-related cytokines ( IL17 , IL21 and IL23 ) were associated with protection against parasite re-infection/infection respectively . Also , in a vaccination trial in mice using F . hepatica-derived synthetic peptides , the most protective peptides stimulated the production of high IFN-γ , IL4 and IL17 levels , indicating that Th17 cells may be involved in protective immune responses [85] . Here , we observed an attenuated Th17 response to F . hepatica infection with significantly down-regulated expression of the orphan retinoic acid nuclear receptor ( ROR ) family transcription factor RORγt , IL23/IL23R and IL17F/IL17RC ( Table 3 ) . The transcription factor RORγt plays an essential role in differentiation of Th17 cells [86] . Although IL23 is not the differentiation factor of Th17 [87] , IL23 plays a fundamental role in stabilizing the Th17 lineage and expanding Th17 responses [86] . Notably , we observed a dramatic down-regulation of IL23R with a fold change of 211 . Previous studies have shown that pathogenic Th17 cells exhibit a unique transcriptional signature , including high IL23R expression , distinguished from non-pathogenic Th17 cells [88] . This downregulation of IL23R in our study may indicate an attenuation of pathogenicity of Th17 cells during infection . Th17 cells are the major source of IL17/IL17A and IL17F in many types of adaptive immunity . IL17F shares the strongest sequence homology with IL17A in the IL17 family , and both cytokines promote the generation of pro-inflammatory cytokines and chemokines [89] . Importantly , dysregulated IL17A and IL17F production can result in excessive pro-inflammatory cytokine expression and chronic inflammation , leading to severe tissue damage [89] . Our data showed that one of the IL17F receptor genes , IL17RC , and its downstream chemokine gene CXCL5 , were also down-regulated during F . hepatica infection ( Table 3 ) . Compared to IL17A , IL17F was found to be a more potent inducer of CXCL5 [90] , a known neutrophil recruiter , indicating that F . hepatica may avoid excessive neutrophil recruitment during infection by suppressing the IL17F/CXCL5 axis . In summary , our data suggest F . hepatica may inhibit the differentiation and stability of Th17 cells and further attenuate their role in promoting immunopathology , which in turn may benefit the parasite’s survival within the host . In addition , as Th17 responses have been proposed to be protective against F . hepatica [85] , it would be interesting to investigate the correlation between the expression levels of IL17-associated genes ( RORγt , IL23/IL23R and IL17F ) and protection levels in future vaccination trials . Epidemiological and experimental studies suggesting an inverse correlation between helminth prevalence and incidence of autoimmune diseases ( summarized from [91] ) , indicate a protective effect of helminth infection against the development of autoimmune disease . Recent studies have demonstrated that this protective ability is associated with the suppression of host Th17 response during helminth infection [92–97] . A study in mice showed that F . hepatica can attenuate the induction of experimental autoimmune encephalomyelitis ( EAE ) , a rodent model of human multiple sclerosis ( MS ) , by inhibiting Th1 and Th17 responses through a TGFB-dependent mechanism [94] . The authors hypothesized that fluke-induced TGFB may suppress Th17 cells by inhibiting IL23 that would normally promote their development or expansion . Consistent with this hypothesis was the observation that two major secretory antigens of F . hepatica in their recombinant forms ( rFhCL1 and rFhGST-si ) suppressed the differentiation of Th17 cells by altering the function of dendritic cells ( DC ) to secrete reduced levels of IL-23 [98] . Our finding supports this observation , as we observed down-regulation of IL23/IL23R after infection . Previous studies have shown that IL23-deficient ( IL23p19-/- ) mice were resistant to EAE [99] , and IL23-activated Th17 cells exhibited a higher capacity to transfer EAE than IL12-activated Th1 cells [100] . A study based on the IL-12/23 subunit ( p35 , p19 , or p40 ) deficient mouse model demonstrated that the IL23/IL17 axis ( Th17 response ) rather than the IL12/IFNγ axis ( Th1 response ) is essential for the establishment of EAE [99] . The importance of the IL-23/IL17 axis is also supported in human MS [101] . In addition , a chemokine-chemokine receptor system CCL2 ( or MCP-1 ) -CCR2 pathway has been proposed to be essential for development of EAE/MS as this signaling pathway might play an important role in migration of Th17 cells to MS lesions [101] . This is supported by the recent observation that Th17-expressed CCR2 drives Th17 recruitment to the inflamed central nervous system ( CNS ) during EAE [102] . Interestingly , our data showed a significant down-regulation of CCR2 and its ligand CCL2 induced by F . hepatica infection ( Table 3 ) , suggesting a potential role of F . hepatica in inhibition of Th17 recruitment into EAE/MS lesions . We hypothesize that F . hepatica infection may attenuate EAE/MS by suppressing these key genes involved in the IL23/IL17 axis and CCL2/CCR2 signaling pathways . Further identification of specific , or combinations of F . hepatica-derived molecules involved in suppressing these genes would be for important for further investigation of their immunotherapeutic potential against human MS and other autoimmune diseases . In conclusion , this is one of the first study which describes gene expression changes of host PBMC in response to F . hepatica infection . The study was carried out in sheep , which , along with cattle , represent one of the two major livestock species that are the targets for development of vaccines as a control measure for fasciolosis . Overall , our study revealed a Th2-biased immune response to infection , and suppression of Th1/Th17 responses , as expected . The potential regulatory ability of IL12 , IL18 and miR155-5p in PBMCs during F . hepatica infection were proposed and further investigation is warranted . Notably , based on the analysis of the data , we hypothesize that up-regulated TGFβ1 , involved with induction of fibrosis during infection , may work through SMAD2/3/4 signaling . The inhibitory-Smad protein ( SMAD7 ) may play a key role in limitation of fibrosis formation in ovine fasciolosis . The comprehensive down-regulation of Toll-like receptors ( e . g . TLR1 , 5 , 6 , 7 and 10 ) and up-regulation of death receptors ( e . g . TNFR1 , TNFR2 , DR3 , 4 , 5 and 6 ) was observed in PBMC obtained from sheep during the early stage of infection , and this indicates that F . hepatica may attenuate the inflammatory response through altering the function of sentinel cells and inducing apoptosis in PBMC . Finally , F . hepatica likely inhibits the differentiation and stability of Th17 cells through down-regulation of the transcriptional factor RORγt and IL23/IL23R . The dataset provided here , from one of the major target livestock species provides information pertinent to understanding of the immune response to fasciolosis , as well as a base from which to further refine development of vaccines . This study also shed a light on helminth-mediated immunoregulation as it may impact on control of allergic and immune-mediated diseases of man and animals .
|
Fasciola hepatica ( liver fluke ) is not only of major health , welfare and economic importance in ruminants , but also an emerging zoonosis . This parasite is sophisticated in manipulating the host’s immune system to benefit its own survival . In this study we investigated global gene expression changes of the primary innate/adaptive immunity-related cells ( peripheral blood mononuclear cells ) from sheep pre- and post- infected with F . hepatica , which revealed the underpinning mechanisms behind various aspects of fluke-induced immunomodulation , including fibrosis , nitric oxide production , regulation of Toll-like receptors , apoptosis of immune cells , and Th17 differentiation . We hypothesis that activation of pathways relevant to fibrosis in ovine chronic infection , may differ from those seen in cattle . The dataset provided here provides information pertinent to understanding of the immune response to sheep fasciolosis . Due to the lack of studies on human fasciolosis , this information is also valuable for exploring human immune response to F . hepatica infection , as well as for further refine development of vaccines .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"cell",
"death",
"medicine",
"and",
"health",
"sciences",
"pathology",
"and",
"laboratory",
"medicine",
"gene",
"regulation",
"immunology",
"cell",
"processes",
"tropical",
"diseases",
"fascioliasis",
"parasitic",
"diseases",
"regulator",
"genes",
"gene",
"types",
"neglected",
"tropical",
"diseases",
"immune",
"system",
"proteins",
"proteins",
"gene",
"expression",
"pathogenesis",
"immune",
"response",
"toll-like",
"receptors",
"biochemistry",
"signal",
"transduction",
"helminth",
"infections",
"cell",
"biology",
"host-pathogen",
"interactions",
"apoptosis",
"genetics",
"biology",
"and",
"life",
"sciences",
"immune",
"receptors"
] |
2016
|
Transcriptomic Study on Ovine Immune Responses to Fasciola hepatica Infection
|
Overlaying differential changes in gene expression on protein interaction networks has proven to be a useful approach to interpreting the cell's dynamic response to a changing environment . Despite successes in finding active subnetworks in the context of a single species , the idea of overlaying lists of differentially expressed genes on networks has not yet been extended to support the analysis of multiple species' interaction networks . To address this problem , we designed a scalable , cross-species network search algorithm , neXus ( Network - cross ( X ) -species - Search ) , that discovers conserved , active subnetworks based on parallel differential expression studies in multiple species . Our approach leverages functional linkage networks , which provide more comprehensive coverage of functional relationships than physical interaction networks by combining heterogeneous types of genomic data . We applied our cross-species approach to identify conserved modules that are differentially active in stem cells relative to differentiated cells based on parallel gene expression studies and functional linkage networks from mouse and human . We find hundreds of conserved active subnetworks enriched for stem cell-associated functions such as cell cycle , DNA repair , and chromatin modification processes . Using a variation of this approach , we also find a number of species-specific networks , which likely reflect mechanisms of stem cell function that have diverged between mouse and human . We assess the statistical significance of the subnetworks by comparing them with subnetworks discovered on random permutations of the differential expression data . We also describe several case examples that illustrate the utility of comparative analysis of active subnetworks .
Developments in genomic and proteomic technologies in recent years have given us numerous methods for capturing high resolution snapshots of cellular processes . The end result of a genome-scale experiment is typically a long list of candidate genes that provide a basis for further , more detailed , follow up experiments . For example , gene expression microarrays are a popular approach for identifying differentially expressed genes between two cell types or experimental conditions , and this technology typically yields several hundred to a few thousand differentially expressed genes in a typical comparison [1] , [2] . While there are sometimes obvious biological processes represented within these lists , developing precise hypotheses from such a long list of candidates can be challenging . Although to varying degrees , this is also true of other genome-scale experiments or screens ( e . g . Genome wide association studies [3] or genetic interaction screens [4] ) . In short , the bottleneck in genomic research has quickly moved from the production of high-quality data to interpretation and hypothesis generation . One powerful approach that has been used to aid in the interpretation of candidate genes lists is integrative analysis with complementary genome-scale data . For example , in a landmark study , Ideker et al . addressed the challenge of interpreting lists of significantly differentially expressed genes by overlaying them on a protein-protein interaction network [5] . They found that certain groups of differentially expressed genes tend to cluster together on the interaction network , building confidence that the signature was indeed biologically relevant and suggesting that entire physical modules were differentially expressed together . This approach has since been extended to several other scenarios , all demonstrating the utility of this idea . For example , Rajagopalan et al . extended Ideker's method to larger , literature-curated biological networks [6] . Others incorporated co-expression scores to favor selected edges of the protein interaction network [7] , [8] , [9] , [10] . Dittrich et al . later formulated the problem as an integer linear programming optimization problem [10] . Recent work has also extended this idea to show that sample classification based on expression profiles can also take advantage of complementary structural information in protein-protein interaction networks [11] . In separate studies , groups have compared and aligned the structure of protein-protein interaction networks across species [12] , [13] . The basic approach adopted by these methods is to identify subgraphs with conservation at the protein sequence level ( nodes ) as well as at the physical or functional interaction level ( edges ) . This approach has been used to suggest core pathways that are conserved across species and to build confidence in individual protein-protein interactions based on the co-occurrence in multiple species [12] , [13] . However , to our knowledge , no one has yet applied this idea to study network-based patterns of expression across species . We propose that just as protein-protein interaction networks can be mined for conserved patterns , differential expression patterns overlaid on biological networks can be aligned to identify conserved patterns of expression , which we call conserved active subnetworks . In this study , we describe a novel approach for identifying conserved active subnetworks in interaction networks across multiple species . Given differential expression measures representing analogous phenotypes in two different species and corresponding interaction networks ( for example , protein-protein interaction networks ) , our approach identifies tightly connected network modules that show a high degree of differential expression , i . e . dense subnetworks , and are conserved in both networks . This is in contrast to previous approaches , which focused on using differential expression or other activity scores to identify dense subnetworks in protein-protein interaction networks for a single species [5] , [6] , [7] , [8] , [9] , [10] , [11] . In addition to addressing the new question of conservation of network patterns across species , our approach presents a scalable solution to active subnetwork identification , which has typically been restricted to relatively sparse protein-protein interaction networks . Sparse coverage of current protein-protein interaction studies limits the ability to match patterns across species . Recent work in area of genomic data integration helps to address this issue . Several approaches now exist which integrate interaction and other information to infer functional associations between genes , to form functional linkage networks [14] , [15] , [16] . Such approaches can incorporate protein-protein and genetic interactions , gene expression , protein localization , phenotype , and sequence data; and have been applied now in many species including yeast , bacteria , worm , fly , plants ( Arabidopsis ) , mouse , and human [14] , [15] , [16] , [17] , [18] , [19] , [20] , [21] . These networks are often significantly denser than protein-protein interaction networks and include hundreds of thousands or even millions of weighted edges that reflect confidence in gene-gene functional relationships . The power ( and challenge ) in using functional linkage networks is that they capture a broad range of functional relationships that have relevance for defining network modules: for example , physical interactions between proteins , co-expression , regulatory relationships , or shared mutant phenotypes . This is in contrast to protein-protein interaction networks which focus on physical interactions between proteins , our knowledge of which is relatively limited in many species , particularly higher eukaryotes . A more detailed comparison of functional linkage and protein-protein interaction networks and the implications for their use for active subnetwork discovery is provided as Supplementary Material ( see a detailed discussion in Text S1 , Note 1 , “Implications of using functional linkage vs . physical interaction networks for active subnetwork discovery” ) . Given their more comprehensive coverage of a broad variety of gene relationships , functional linkage networks should allow for more sensitive discovery of networks that are differentially expressed under various conditions of interest . However , with their broader coverage also come several computational issues . Given the fact that functional linkage networks are orders of magnitude more dense than protein-protein interaction networks , existing algorithms for the discovery of dense subnetworks do not easily scale to this problem . Using functional linkage networks from human and mouse as a basis , we applied our scalable cross-species network discovery approach to identify conserved subnetworks that are differentially active in stem cells relative to differentiated cells based on parallel gene expression studies in mouse and human . We show that these conserved patterns are not likely to have occurred by chance , and that they are enriched for known as well as novel stem cell and differentiation-related processes . Another useful application of our approach is to find functional modules which have diverged or which have been rewired across the two species , which has been previously approached using expression data alone [22] . We designed a variation of our cross-species network search approach to find a number of species-specific networks , which likely reflect differences in the active cellular program between mouse and human pluripotent stem cells . Finally , we demonstrate the usefulness of our algorithm by discussing specific examples of subnetworks discovered , some of which highlight the potentially novel candidate genes involved in the maintenance of stem cell pluripotency .
We developed an algorithm to find conserved active subnetworks across species ( Figure 1 ) . Our approach requires lists of differentially expressed genes and corresponding fold change values in two different species , assumed to represent analogous conditions . The aim of our approach is to overlay gene activity scores on the respective functional linkage or interaction networks to discover dense subnetworks with a large number of differentially active genes with similar expression patterns in both species . Our approach assumes a set of orthologous clusters for the two species of interest and weighted linkage networks in both species , although it can be also applied to binary interaction networks ( e . g . protein-protein interaction networks [23] ) . Briefly , subnetworks are simultaneously grown in both species from seed genes by adding nearby genes in the interaction networks that maximize the average activity score of the subnetwork while at the same time maintaining a minimum desired clustering coefficient of the genes in the subnetwork ( see Materials and Methods for details ) . Subnetwork growth is stopped when the average activity score reaches a minimum threshold . This process is then repeated with each differentially active gene in either species serving as the seed . The result is a set of highly clustered subnetworks with a high density of matched differential expression in both species ( see Materials and Methods for details ) . To test our subnetwork discovery method , we compiled a compendium of gene expression data for mouse and human pluripotent stem cells . Briefly , 249 mouse and 132 human expression profiles were obtained from several independent datasets from the Gene Expression Omnibus ( GEO ) database [24] ( Table S3 , S4 ) . Our goal was to identify subnetworks whose activity was associated with the maintenance of stem cell pluripotency in both human and mouse . It has been shown that human embryonic stem ( ES ) lines across the world are identical in expression of key pluripotency markers like Nanog and Pou5f1 , but they can show remarkable differences in expression of other lineage specific markers such as AFP , possibly due to different culture conditions and varying levels of spontaneous differentiation in cultures [25] . Thus , we reasoned that a large compendium of data in both species could support a more robust differential expression analysis , free of any biases from individual studies or cell lines . To group expression profiles at similar stages of differentiation , we used non-negative matrix factorization ( NMF ) [26] , which is an unsupervised clustering method ( see Materials and Methods for details ) . Clusters resulting from NMF clearly separated the expression profiles of undifferentiated , pluripotent cells from those that were in early stages of differentiation or late stages of reprogramming . Differential expression analysis ( SAM ) was then performed between these two classes of samples to identify a set of genes that change in expression as the pluripotent cells start to exit the self-renewal program during differentiation ( see Materials and Methods for details ) . This clustering and differential expression analysis process was performed independently on the mouse and human expression data . The genes deemed significant by this analysis were labeled with activity scores reflecting normalized fold change values ( see Materials and Methods for details ) and used as input for our subnetwork discovery approach . It is important to note that the method for differential expression analysis ( or other means of generating activity scores ) is completely independent of the subnetwork discovery algorithm . Our large compendium of stem cell expression data for mouse and human provided an interesting setting for subnetwork discovery , but our approach could also be applied to activity scores derived from more standard , single-dataset differential expression studies , assuming comparable datasets are available for two different species ( see Text S1 , Note 2 , “neXus applied to single dataset differential expression study” and Figure S1 for an example ) . We applied our subnetwork discovery approach to the results of the stem cell differential expression analysis and functional linkage networks from human and mouse . Human and mouse functional linkage networks were obtained from previous work [15] , [16] . The human network incorporates physical and genetic interactions , sequence information ( shared protein domains , transcription factor binding sites ) , and gene expression profiles [15] . The mouse network incorporates physical interaction data , shared phenotype data , phylogenetic profile information , the yeast functional linkage network where orthologs exist , and gene expression information [16] . These functional networks reflect broad functional relationships between genes or proteins and thus are more general than protein-protein interaction networks ( see a detailed discussion in Text S1 , Note 1 , “Implications of using functional linkage vs . physical interaction networks for active subnetwork discovery” ) . While the input data for these networks are largely independent , physical interaction data for mouse was derived from human interactions ( see a detailed discussion in Text S1 , Note 3 , “Independence of the datasets” ) . Conserved active subnetworks between human and mouse were identified by varying the two parameters of the algorithm , the average expression activity ( normalized fold change ) of the network , and the minimum clustering coefficient . This resulted in between 1 and 255 network ( s ) from the most conservative to the most lenient parameter settings , respectively . For example , at a network score cutoff of 0 . 15 ( see Materials and Methods , “Microarray data processing” for fold change normalization ) , and strict clustering coefficient criteria ( >0 . 1 for mouse and >0 . 2 for human ) , we found a total of 255 conserved subnetworks involving 607 genes in each of the two species ( Figure 2A ) . Increasing the clustering coefficient cutoff or increasing the network score threshold enabled the discovery of fewer , but increasingly confident subnetworks ( Figure 2B , Figure S2 ) . To assess the statistical and biological significance of the networks , we performed a network randomization analysis . Specifically , the expression activity scores in both mouse and human were randomly shuffled five times with respect to the gene labels , and the algorithm was then applied to the shuffled expression profiles . Any conserved patterns of these randomized expression data on the functional linkage network should then represent false positives and not biologically relevant conservation . In all randomization experiments , the functional linkage network structure was retained and only gene activities were shuffled , so that we could specifically estimate the conserved expression patterns arising out of clustering of the active genes by random chance . Importantly , we found that while some subnetworks were discovered in various instances of the randomization experiment , far fewer subnetworks were discovered than for the original expression profiles ( Figure 2A ) . For example , at our lenient network score and clustering coefficient cutoffs , we discovered an average of 11 . 4 subnetworks ( standard deviation of 4 ) across five randomization experiments in contrast to the 255 real subnetworks discovered on the original expression data ( Figure 2A ) . Moreover , the average size of the real subnetworks was much larger than the random subnetworks as they contained an average of 22 genes compared to 5 . 7 genes ( standard deviation of 0 . 6 ) across the random trials . This comparison clearly suggests that the subnetworks obtained by our cross-species approach are statistically significant , and are not likely to have been discovered by chance . We also found that the signal to noise ratio , which is the ratio of number of real subnetworks to the average number of random subnetworks , improved as we increased the network score cutoff ( Figure S3 ) and clustering coefficient cutoffs ( Figure S2 ) . This improvement suggests that tuning these parameters is an effective means of isolating high-confidence conserved network signatures for hypotheses generation . We also evaluated the subnetworks in terms of their functional coverage and relevance . The function enrichment of the genes contained in each subnetwork was measured based on significant overlap with biological processes in the Gene Ontology [27] ( see Materials and Methods ) . A large majority of the subnetworks ( 235 of 255 ) were found to be enriched for GO processes , many with suspected involvement in stem cell maintenance and differentiation ( Figure 3 ) . Furthermore , many subnetworks were monochromatic , that is , they contained genes with concordant changes in expression in either stem cells or differentiated cells . Around a third of the subnetworks were consistently more highly expressed in stem cells while approximately half of them were consistently more highly expressed in differentiated cells . As expected , the monochromatic subnetworks active in stem cells were found to play a role in metabolic processes and regulation , biosynthetic processes , cell cycle , DNA repair , and gene transcription and regulation ( Figure 3 ) . On the other hand , the monochromatic subnetworks active in differentiated cells were involved in development and differentiation of various cell types , tissues and organs ( Figure 3 ) . We also noted another interesting class of subnetworks that showed mixed changes in expression , including a combination of up and down-regulated genes , whose patterns matched across species . This class may highlight pathways that require or at least exhibit dramatic imbalances in gene expression to maintain stem cell state . We compared conserved subnetworks discovered by our approach to gene sets obtained from a simple intersection of orthologs on the human and mouse differentially expressed gene lists . One might suggest that a reasonable approach to finding the core conserved modules underlying stem cell pluripotency is to simply analyze the most extreme differentially expressed genes in both species . We attempted this approach by comparing the top 600 differentially expressed genes from mouse and human , which is comparable to the total number of genes contained across our subnetworks . There was relatively low overlap between the gene sets: of the 600 genes , only 36 are up-regulated in the both species while 34 are down-regulated ( Figure S4 ) . This level of agreement is higher than the number expected by chance ( ∼15–20 ) , but certainly not as high as one might expect , suggesting that there are a number of core modules that do not exhibit the most extreme expression changes . The overlap does improve when we consider any genes that show significant changes in expression ( FDR 5% ) : 1367 genes are significantly up-regulated in pluripotent stem cells in both human and mouse while 986 are significantly down-regulated , which reflects an overlap of ∼50% ( Table 1 ) . However , this more lenient cutoff yields thousands of candidate genes to consider , which makes determination of the core conserved modules difficult . Our conserved subnetworks offer a solution to this problem: we find 255 modules containing approximately 600 genes that appear in both the human and mouse subnetworks , including 282 that are differentially expressed and show similar expression patterns . Simultaneous network discovery guided by the combined differential expression data allows us to directly identify the core conserved patterns of expression , even where some of these patterns are subtle but consistent . We were intrigued by the fact that our conserved subnetworks actually contained a significant fraction of genes ( ∼20% ) that showed no evidence of differential expression . By its design ( see Materials and Methods , Algorithm ) , the subnetwork discovery algorithm can include non-differentially expressed genes in identified subnetworks if they connect across highly differentially expressed genes . Briefly , for a given seed gene , the algorithm starts by finding the surrounding functional neighborhood of that seed , which is defined as the set of genes that can be reached within a given path confidence ( the product of linkage weights along the path ) . From this set of genes in the functional neighborhood , the gene that results in the greatest increase in the network activity score is added to the current subnetwork , including any genes required for its connection to the seed . The addition of the corresponding path can potentially bring in non-differentially expressed genes , which may reflect genes that are causally linked to the corresponding subnetwork but whose activity is simply post-transcriptionally regulated [11] . Their activity may be modulated at the protein level which is typical of transduction pathways that control gene expression programs [11] . For example , TEP1 is not differentially expressed but is found in an active subnetwork with many well-characterized stem cells genes like POU5F1 ( Figure S5A ) . TEP1 is involved in telomerase activity [28] and has been shown to be regulated by phosphorylation in breast cancer cells [29] . These examples illustrate the advantages of integrating differential expression data with the broader relationships captured by functional linkage networks in that complete modules can be identified , including genes whose activity is not necessarily transcriptionally regulated . The subnetworks also sometimes contain mixed expression signatures ( both up- and down-regulated genes ) that are conserved across species , highlighting genes in the same pathway that are antagonistic or genes that exhibit different interactions at various stages of development . For example , one conserved network with mixed expression changes was centered about the important extracellular structural protein ostepontin ( also known as secreted phosphoprotein 1 , SPP1 ) ( Figure S5B ) . SPP1 is highly up-regulated in both mouse and human stem cells while its surrounding subnetwork is significantly down-regulated in comparison to differentiated cells in both species . Osteopontin is known to be highly expressed in bone and other cell types like smooth muscle cells , endothelial cells and hematopoietic stem cell niches . The subnetwork captures some well-known interactions of SPP1 in these cells . For example , osteopontin has been shown to be a ligand for CD44 in tumor cells [30] . Pou5f1 has been shown to bind to the preimplantation enhancer element of osteopontin , and thus , the expression of the two proteins is highly correlated in early mouse embryonic development [31] . The induction of osteopontin in immortalized mouse embryonic fibroblasts , in response to TGF-β2 , has been shown to promote the maintenance of undifferentiated human embryonic stem cells [32] . This is attributed to the presence of a TGF-β responsive element in the osteopontin enhancer . Thus , osteopontin likely plays a pivotal role in the maintenance of both human and mouse embryonic stem cells , and this subnetwork supports this idea . The functional linkages of osteopontin in early embryonic cells have not been fully elucidated yet , but this subnetwork suggests that this gene may play a role in the embryonic context since the other genes in the subnetwork show an opposing expression pattern . These interesting cases would not be readily discovered through a simple comparison of differential expression lists across species . To our knowledge , our method is the first attempt to interpret differential expression data by integrating with interaction networks across multiple species . Thus , we further assessed the advantages of simultaneous , cross-species network search as compared to active subnetwork discovery in a single species , which has been the focus of previous methods [5] , [6] , [8] , [9] , [10] , and is the principle behind commonly used analysis tools such as Ingenuity Pathway Analysis ( Ingenuity® Systems , www . ingenuity . com ) . Analogous experiments to those performed on our cross-species algorithm were applied to discover active subnetworks in the mouse functional linkage network alone ( see Materials and Methods ) . Most of the existing approaches did not scale to the complete functional linkage network used by our approach ( Table 2 ) , so we reduced the scale of the mouse functional linkage network by restricting the network to the 50 , 000 highest weight edges to allow for a direct comparison of our approach to other methods in the single-species context . We implemented MATISSE [33] , jActiveModules [5] and Ingenuity ( Ingenuity® Systems , www . ingenuity . com ) on the mouse data and compared with a single-species version of our approach as well as our cross-species algorithm . For methods that do not incorporate weighted edges , we binarized the reduced network . To allow a direct comparison of the number of subnetworks produced by each approach , subnetworks were sorted in descending order by size and overlapping subnetworks were removed when their overlap with larger networks ( in genes ) was greater than 60% . To estimate the significance of the subnetworks identified by each algorithm , we randomized the gene labels in the expression data and ran each algorithm five times on randomized expression data . The number and scores of subnetworks produced by each algorithm were compared with the number and scores of the subnetworks generated from the 5 runs on randomized expression data ( Figure 4 ) . Although our main contribution in this work is the cross-species algorithm , we found a single-species version of our approach performed favorably in comparison to existing approaches ( Figure 4 ) . Specifically , it produced more subnetworks than other approaches on the real expression data while producing far fewer subnetworks on the randomized data ( Figure 4 ) . Surprisingly , we found that 2 of the 3 existing approaches ( Ingenuity and jActiveModules ) produced as many or more networks on the randomized data as on the real data for most score cutoffs ( Figure 4B–C ) . Among the existing methods we evaluated , MATISSE provides the best performance , often reporting 1 . 5–2 fold more real networks at a given score cutoff than on randomized data ( Figure 2A ) . There was significant variation in the size of subnetworks produced across the various approaches , with some producing networks as large as 2000 genes and others producing relatively small subnetworks consisting of less than 10 genes ( Figure S6 ) . The most useful number and size of networks will , of course , depend on the application , but one particularly unique feature of our implementation is that subnetworks are captured at all stages of their growth , thus giving the user to control of the tradeoff between size and significance of the subnetwork in consideration ( see Web Interface section ) . Perhaps the most striking result of our comparison was our finding that any single species approach , including our own , performed much worse than our cross-species subnetwork discovery algorithm . For example , in the single-species setting for mouse , we were able to find 164 subnetworks while discovering an average of 71 ( standard deviation of 7 . 8 ) subnetworks in our randomization experiments under the same setting ( mouse , clustering coefficient threshold = 0 . 1 , network score cutoff = 0 . 3 ) , suggesting an enrichment of approximately 2 . 5-fold ( Figure 4D ) . Using the cross-species approach , we found 234 subnetworks while discovering an average of 9 . 8 ( standard deviation of 4 . 16 ) in our randomization experiments ( parameter setting: mouse and human clustering coefficient thresholds = 0 . 1 and 0 . 2 , network score cutoff = 0 . 15 ) , which represents a 20-fold enrichment ( Figure 2B ) . Thus , not only did we discover more candidate networks in the cross-species setting , but the networks we found were of higher statistical confidence . Similar results were obtained when we applied our single-species approach to the complete functional linkage network ( Figure S7 ) . The improvement in sensitivity and specificity by the cross-species approach is a particularly interesting result because it suggests that simultaneous cross-species network discovery can serve as an effective means of improving the signal-to-noise ratio in network discovery even if one is not necessarily interested in asking questions about conservation across species . More pessimistically , this result suggests that separating biologically relevant active subnetworks from random networks based on a single functional linkage network is a challenging problem . The enhanced performance of the cross-species approach can be attributed to the fact that coordinated expression changes can be reasonably clustered in both species' functional linkage networks . Due to the small-world nature of functional linkage networks ( or protein-protein interaction networks ) [34] , given a large set of genes , subnetworks involving partitions of this set can often be readily found even if these genes do not necessarily play a specific role together . The coherent grouping of genes across species eliminates random aggregation of active genes , and thus , the cross-species approach is able to relax both the network score and clustering coefficient stringency criteria , while still maintaining statistical confidence in the networks . Indeed , when our approach was applied independently to mouse and human data , we found little intersection among the two species' subnetworks: of the genes covered by human ( 305 orthologous clusters ) and mouse subnetworks ( 261 orthologous clusters ) , only 21 were overlapping . In contrast , the cross-species approach discovers around 250 subnetworks covering 607 genes in both mouse and human ( Table 1 ) . We obtained a similar result when comparing to subnetworks derived from another approach , MATISSE , applied to the human and mouse data ( see Text S1 , Note 4 , “Comparison of the overlap of mouse and human subnetworks discovered through MATISSE and neXus” , Table S1 , S2 ) . Thus , in addition to the underlying biological question of conservation of expression signatures , cross-species analysis can serve as an effective noise filter , which is critical for discovering clustered patterns of expression changes in a dense interaction network . The difficulty in identifying subnetworks from a list of genes within a single species has important implications for how the statistical significance of such networks should be assessed . This problem often arises in practice during the interpretation of candidate gene lists . For example , analysis tools such as Ingenuity Pathway Analysis ( Ingenuity® Systems , www . ingenuity . com ) are now being widely used based on the single-species discovery method we evaluated above . The significance of networks identified by such approaches are typically assessed by comparing the network score after optimization to scores that obtained by randomly sampling a similarly sized set of genes . However , as demonstrated above , high-scoring networks are often obtained when search algorithms are applied to randomly selected candidate genes . Put simply , in many protein interaction networks , random lists of genes are much easier to connect than one might expect . Our results suggest that significance should instead be estimated by applying the network search process ( with the same parameters ) to several random candidate genes lists , and evaluating the actual scores in the context of the resulting random score distribution . Using the cross-species network discovery algorithm , we are able to find subnetworks reflecting conserved functional modules between mouse and human pluripotent stem cells . We found many of these subnetworks to be monochromatically active in stem cells or differentiated cells . This was not a prerequisite for network discovery , but reflects that the majority of genes supporting a local process are regulated in the same direction . Monochromatic subnetworks up-regulated in stem cells were our primary focus because these reflected potential candidate processes that are necessary for maintaining a pluripotent , self-renewing stem cell state . One of the most significant conserved subnetworks of this type captures the core pluripotency circuit in embryonic stem cells ( Figure 5A ) . This network recovers associations between important transcription factors such as POU5F1 , NANOG , SOX2 and FGF4 , all of which have been shown to form an important transcriptional circuit in embryonic stem ( ES ) cells , consisting of feed-forward and autoregulatory loops [35] . Chromatin immunoprecipitation experiments have shown that these three proteins exhibit a significant overlap in their binding sites in the genome [35] , [36] . The subnetwork links FGF4 to the core signaling circuitry formed by POU5F1 , SOX2 , and NANOG . FGF4 has been shown to be expressed in the peri-implantation mouse embryo [37] and the SOX2/POU5F1 complex has been shown to activate transcription of FGF4 by binding to an enhancer element [38] . The role of this module has also been studied quite extensively in early embryonic development . FGF4 null mutants in mouse are embryonic lethal due to defective primitive endoderm [39] . The cells of the mouse inner cell mass ( ICM ) show a reciprocal expression pattern of FGF4 ( ligand ) and FGFR2 ( receptor ) . It has been shown that the FGF4 secreted by the epiblast precursor cells is crucial to the differentiation and maintenance of cells of the trophectoderm and extraembryonic endoderm lineages [40] , [41] . Human ESCs show a striking resemblance to mouse epiblast-derived stem cells in terms of morphology and maintenance culture conditions , amongst other characteristics [42] , [43] . Thus , this network highlights a core , conserved module active in the pluripotent cells of both the species , irrespective of the downstream effects on cell signaling and morphology . FGF4 stimulation of ERK1/2 signaling in mouse ES cells has been shown to facilitate lineage commitment [44] . In human ES cells , FGF signaling promotes self-renewal by directly affecting the expression of NANOG [45] , [46] as well as suppressing expression of genes responsible for reversion to an ICM-like state [47] . Another highly significant subnetwork discovered by our approach pertains to the control of cell cycle progression in ES cells ( Figure 5B ) . Both human and mouse ES cells have a very short G1 phase which can be attributed to the constitutively active CDK2/6 [48] , [49] . CCNB1 and MYBL2 are two important cell cycle regulators that are expressed at high levels in undifferentiated ES cells and their expression decreases rapidly upon induction of differentiation [50] . This happens even before loss of the important regulator proteins such as POU5F1 or NANOG can be detected . The conserved subnetwork highlights the role of these two genes in the maintenance of cell cycle progression in ES cells . Knockdown of MYBL2 has been shown to induce polyploidy/aneuploidy in ES cells and CCNB1 is a known target of MYBL2 [51] . B-MYB is also crucial for inner cell mass development in mice embryos [52] . The role of CCNF in embryonic stem cells has not been explored but yeast two hybrid assays have shown that the NLS domain of CCNF can regulate nuclear localization of CCNB1 [53] . Many conserved subnetworks also included genes that are up-regulated during the initiation of differentiation . This supports the idea that the maintenance of ES cell phenotype requires the suppression of differentiation-associated gene expression as well . One interesting example of this phenomenon was highlighted in a third subnetwork discovered by our approach , which was centered on the protein ZIC3 ( Figure 5C ) . ZIC3 has been shown to be required for maintaining pluripotency of mouse embryonic stem cells by suppressing endoderm specification [54] while GLI1 has an important effect on embryonic stem cell proliferation [55] . These two proteins are known to work in coordination for transcriptional activation or repression [56] . Both of these genes code for DNA binding zinc finger proteins and they share and recognize highly conserved zinc finger domains . The down-regulated genes in the subnetwork , namely , WNT5A , FOXF2 and RARB , play important roles in the differentiation of embryonic stem cells [57] , [58] , [59] . It is interesting to observe that these genes have GLI binding sites in their promoter region or cis-regulatory domains , which suggests that GLI1 and ZIC3 could potentially regulate their expression in ES cells [60] , [61] . Also , GLI proteins participate in regulation of Hedgehog signaling , of which RARB and FOXF2 are members , and GLI is also known to regulate the members of WNT family [62] . These functional interactions and coordinated expression strongly suggest ZIC3 and GLI1 might be responsible for suppressing the expression of genes such as FOXF2 , WNT5A and RARB . This network in particular provides an illustrative example of how subnetwork discovery can provide novel testable experimental hypotheses . This hypothesis could be explored experimentally through RNAi knockdown of ZIC3 and GLI1 in embryonic stem cells to check for resultant changes in expression of the other genes in the network . Lim et al . [54] conducted RNAi knockdown of ZIC3 in human and mouse ESCs and saw enhanced expression of endodermal transcripts like SOX17 and PDGFRA . Further experiments could also be used to check for direct binding of ZIC3 and GLI1 to the promoter regions of the differentiation-associated genes . The subnetwork also highlights the striking observation that the gene ZIC1 , despite sharing 69% homology with ZIC3 , does not show the same trend in expression in either mouse or human pluripotent stem cells . While ZIC2 and ZIC3 have been suggested to have partially overlapping or redundant roles in suppressing endoderm in embryonic stem cells , the role of ZIC1 in this context has been not been explored much . Further overexpression studies of this gene could be used to elucidate its exact role in this network . Another interesting subnetwork found by our approach was centered around the seed gene SIRPA . The only gene in the whole subnetwork that is found to be up-regulated in mouse and human pluripotent stem cells is LCK ( Figure 5D ) . LCK is one of the eight SRC family kinase genes , which are known to play crucial roles in regulating signals from a variety of cell receptors , affecting a variety of cellular processes such as differentiation , growth and cell shape [63] . Members of this family , namely Hck and Lck , have been implicated in the maintenance of self-renewal of murine embryonic stem cells [46] . Cyes , along with Hck , have been shown to be regulated by LIF in mouse embryonic stem cells and the expression of their active mutants allows the maintenance of these cells at lower concentrations of LIF [64] . Other studies have also reported the evolutionarily conserved transcriptional co-expression of LCK in human and mouse embryonic stem cells based on transcriptomic studies [65] . LCK has also been shown to induce STAT3 phosphorylation and this is believed to cause transformation of cells having constitutive LCK activity [45] . All of the other genes in the sub-network are down-regulated in ES cells , which may be due to the fact that the expression of SFKs is generally associated to lineage-restricted patterns in the adult , such as , the expression of LCK in T lymphocytes . While the hypotheses suggested by the discovered subnetworks ultimately require experimental follow-up , these examples illustrate that the networks capture many of the well-characterized processes supporting stem cell pluripotency as well as implicating some novel players . In general , the process of active subnetwork discovery can play an important role in interpreting differential expression or other genome-wide data . Active subnetworks , and in particular those that are conserved across species , provide evidence that a whole process or pathway is up/down-regulated , which is more definitive than the type of information provided by a differential expression list , for example . A single highly differentially expressed gene is less compelling than an entire functional module with evidence of differential expression . Furthermore , because the underlying functional linkage networks are based on large collections of genomic data , our approach can potentially identify functional modules that are not yet characterized , but that play a critical role under the conditions being studied . We modified the cross-species network discovery algorithm to discover subnetworks that are markedly different in the expression patterns between the two species ( see Materials and Methods , “Score of a Subnetwork” ) . These subnetworks represent tightly interconnected groups of genes or proteins that are active only in one of the species or where the expression changes are in opposite directions , highlighting places where pluripotent stem cell signaling differs between human and mouse . Through randomization experiments similar to the conserved subnetwork identification approach ( see Materials and Methods ) we found that we were able to find such non-conserved network signatures approximately twice as frequently as on randomized expression profiles ( Figure 6A ) . We note that this is a substantially lower signal-to-noise ratio than for the conserved subnetwork discovery approach , for which we achieved approximately 20-fold improvement over random , suggesting that statistically significant species-specific active subnetworks are harder to discover . This is not surprising given that the relatively frequent appearance of random subnetworks in a single species ( Figure 4D ) , which cannot be easily classified as statistical artifacts or biologically relevant changes across species . The species-specific network discovery problem is not able to take advantage of the noise filtering property of the conserved network search described above . Nevertheless , we find interesting subnetworks which highlight differences between gene expression in mouse and human stem cells . For example , one species-specific subnetwork ( Figure 6B ) recapitulates the well-known difference in BMP signaling between human and mouse embryonic stem cells . Mouse embryonic stem cells require BMP2/BMP4 to induce the expression of Inhibitor of differentiation ( Id ) genes via Smad pathway for self-renewal [66] . Thus , exogenous addition of LIF and BMP4/2 is required to maintain mouse ES cells in culture without differentiation . On the other hand , human ES cells cultured in unconditioned medium exhibit high levels of BMP signaling which causes the cells to differentiate . Mouse epiblast stem cells , like human ES cells , differentiate to trophoectoderm upon BMP4 induction [42] . This needs to be suppressed through an antagonist such as noggin to maintain these cells in an undifferentiated , self-renewing state [46] . The other genes in the subnetwork that show opposite trends in differential expression between human and mouse ES cells are MGP , ACTC1 and ENG . Endoglin ( Eng ) is an accessory receptor for several TGF-β growth factors , including BMP2 , and has been shown to be crucial for embryonic hematopoiesis [67] . Matrix GLA protein ( MGP ) is a small matrix protein that has been shown to have a direct interaction with BMP2 and has been shown to modulate BMP signaling [68] . The potentially disparate role of these genes in mouse and human ES cells can be explored further . To facilitate public access to active cross-species subnetworks identified by our approach , we developed a web-based interface for convenient browsing of conserved and species-specific stem cell expression signatures ( http://csbio . cs . umn . edu/neXus/subnetworks , download subnetworks in raw text from http://csbio . cs . umn . edu/neXus ) . Subnetworks are listed according to their corresponding network seed gene , and when a seed gene is selected , the following information is displayed: the conserved active human and mouse subnetworks , significance of the identified subnetwork based on a comparison to network randomization , expression fold changes and name details of mouse and human genes , and the function enrichments of the genes in respective to human and mouse subnetworks based on the Gene Ontology [27] . The subnetwork generation was automated using neato , a Graphviz graph plotting tool with spring model layouts [69] . The Cytoscape version of the subnetworks are also available on the website , which are linked using Cytoscape Webstart [70] . The gene names in the subnetwork are linked to gene information at the Mouse Genome Informatics ( MGI ) database [71] and GeneCards [72] for mouse and human genes , respectively . Another useful feature of our web-interface is that subnetworks can be interactively expanded based on the cross-species discovery algorithm , which allows for real-time analysis of additional candidate genes that are closely associated with the network of interest . As networks are expanded , a statistical significance score is calculated after each iteration , which allows the user to estimate the potential biological relevance of the network as it is expanded . We have described a scalable approach for discovering conserved active subnetworks across species . Starting from candidate gene lists reflecting parallel differential expression studies in two different species , we are able to search for dense subnetworks with conserved patterns of differential expression . In contrast to previous active subnetwork discovery algorithms , our approach not only extends this idea across species , but also enables application of the approach to functional linkage networks as opposed to sparse protein-protein interaction networks . Functional linkage networks integrate information from a diverse collection of genomic and/or proteomic studies ( including protein-protein interactions ) , and thus offer the potential for more sensitive discovery of active subnetworks , including those which involve previously uncharacterized genes . We applied our approach to a differential expression study between pluripotent mouse and human stem cells versus their differentiated cell types to produce several hundred subnetworks that reflect conserved changes between mouse and human . Network search across species produced specific hypotheses about conserved and differentiated mechanisms of stem cell maintenance , and importantly , demonstrated that such an approach can be an effective means of filtering noise from the active subnetwork discovery problem . We found that identifying statistically significant active subnetworks independently within a single species may be a harder problem than previously appreciated , and we suggest the cross-species approach as one solution to this problem . Despite the success of our approach , there are a number of promising directions for further improvement and broader application of the method . While the approach was successfully applied to relatively dense functional linkage networks for mouse and human , it is a computationally challenging problem , and the algorithm cannot be applied in real-time as it still requires several days to run . Strategies for improving the efficiency of conserved network discovery and more formal selection criteria for the parameters associated with our approach are both useful future directions . Furthermore , the approach can be readily extended to discover conserved subnetworks across more than just two species , which will make another fruitful direction as we begin to accumulate functional genomic data across a broad variety of other model organisms . Finally , although our study focused on the interpretation of candidate gene lists derived from differential expression analyses , the algorithm is general and can be readily applied to interpret lists arising from other genomic screens , including , for example , genome-wide association studies .
249 mouse microarray data samples were obtained from 20 GEO datasets ( Table S3 ) . All the samples had been hybridized to the Affymetrix mouse chip MOE 430 2 . 0 . 132 human microarray data samples were obtained from 12 GEO datasets ( Table S4 ) . All the samples had been hybridized to the Affymetrix human chip HGU 133 plus 2 . 0 . The raw data was normalized using the MAS 5 . 0 algorithm [73] and the average chip intensity was scaled to 500 . The probes set IDs with detection p-values higher than 0 . 4 were termed absent and were filtered out for further analysis , along with the probe set IDs with average intensity lower than 50 . Non-negative matrix factorization ( NMF ) was used to identify major biological classes in the data in both species independently [26] . The algorithm factorizes the expression matrix A into two matrices , W and H . If the expression matrix is of size N X M , the algorithm computes an approximation , where W and H have sizes N X k and k X M , respectively [74] . Here , k represents the number of clusters that the samples can be divided into . Each of the k columns of matrix W defines a metagene in such a way that the entry wij represents the coefficient if gene i in metagene j . Each of the M columns of matrix H depicts the metagene expression profile in different samples such that the entry hij represents the expression level of metagene i in sample j . The accuracy of the classification is evaluated by the value of the cophonetic coefficient . NMF was used to cluster the samples into biologically meaningful sets . As an example , for k = 6 , the mouse samples were clustered into the classes that represented the different levels of pluripotency of the stem cells . The cophonetic coefficient for this classification was 0 . 978 . Similar classification could be achieved for k = 5 in the human gene expression data ( cophonetic coefficient of 0 . 977 ) . As mentioned earlier , the matrix W detected the metagenes representing every cluster of similar samples in the data and , the matrix H gave the expression profile of every sample in the particular metagene . The expression profile of the various samples in the metagene corresponding to the cluster of pluripotent stem cells was used to divide the samples into two major classes , on the basis of the values of the entry hij . Class 1 included the pluripotent ES cells and induced pluripotent stem cells while class 2 represented samples that were in the process of early differentiation or late reprogramming . Differential expression analysis was performed between these two biological classes using Significance Analysis of Microarrays [1] . The results of this differential expression analysis were used as the starting point for subnetwork discovery . The differential expression criteria were set at false discovery rate less than 5% . The results of this differential expression analysis yielded fold changes for significantly differentially expressed genes which was log normalized for both up-regulated and down-regulated genes , separately . The log-ratios were rescaled to ranges from −1 to +1 , where −1 represented the gene which is most down-regulated and +1 represented the most up-regulated gene . The majority of the genes were not significantly differentially expressed; the log-ratio of these genes was set to zero . The normalized expression fold change data can be downloaded from http://csbio . cs . umn . edu/neXus . We used the mouse functional linkage network previously published in [16] with all edges below 0 . 10 confidence set to zero , which resulted in around 2 . 7 million weighted edges among 17868 genes . We obtained the human functional linkage network from [15] ( the “global network” ) and trimmed the network to the highest 6 million weighted edges , which corresponded to a minimum edge weight of 0 . 58 and covered 15806 genes . The algorithm identifies functional modules enriched for active genes in both species under consideration . Conserved active modules are found based on two criteria: ( 1 ) a high degree of clustering in both species' functional linkage networks , and ( 2 ) a high average normalized differential expression fold-change ( network score ) sharing the same sign across species . Because the search space is exponential , a greedy heuristic is applied to expand subnetworks from candidate seed genes . Each candidate network is grown until it fails to meet one of the constraints . This algorithm is implemented in Python and the source code can be downloaded from the supplementary website ( http://csbio . cs . umn . edu/neXus ) ( see Box 1 for pseudocode ) . Each component of the algorithm is described in more detail below .
|
Microarrays are a powerful tool for discovering genes whose expression is associated with a particular biological process or phenotype . Differential expression analysis can often generate a list of several hundred or even thousands of significant genes . While these genes represent real expression differences , the large number of candidates can make the process of hypothesis generation for further experimental studies challenging . Use of complementary datasets such as protein-protein interactions can help filter such candidate lists to genes involved with the most relevant pathways . This approach has been applied successfully by many groups , but to date , no one has developed an approach for discovering active pathways or subnetworks that are conserved across multiple species . We propose an algorithm , neXus ( Network – cross ( X ) -species – Search ) , for cross-species active subnetwork discovery given candidate gene lists from two species and weighted protein-protein interaction networks . We validate our approach on expression studies from human and mouse stem cells . We find many active subnetworks that are conserved across species relevant to stem cell biology as well as other subnetworks that show species-specific behavior . We show that these networks are not likely to have been discovered by chance and discuss several specific cases that reveal potentially novel stem cell biology .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Materials",
"and",
"Methods"
] |
[
"genetics",
"and",
"genomics/genomics",
"genetics",
"and",
"genomics/gene",
"discovery",
"computer",
"science/applications",
"developmental",
"biology/stem",
"cells",
"evolutionary",
"biology/evolutionary",
"and",
"comparative",
"genetics",
"genetics",
"and",
"genomics/functional",
"genomics",
"genetics",
"and",
"genomics/gene",
"expression",
"genetics",
"and",
"genomics/bioinformatics",
"developmental",
"biology/cell",
"differentiation",
"genetics",
"and",
"genomics/gene",
"function",
"computational",
"biology/genomics",
"computational",
"biology",
"computational",
"biology/systems",
"biology"
] |
2010
|
A Scalable Approach for Discovering Conserved Active Subnetworks across Species
|
We previously showed that CD8+ T cells are required for optimal primary immunity to low dose Leishmania major infection . However , it is not known whether immunity induced by low dose infection is durable and whether CD8+ T cells contribute to secondary immunity following recovery from low dose infection . Here , we compared primary and secondary immunity to low and high dose L . major infections and assessed the influence of infectious dose on the quality and magnitude of secondary anti-Leishmania immunity . In addition , we investigated the contribution of CD8+ T cells in secondary anti-Leishmania immunity following recovery from low and high dose infections . We found that the early immune response to low and high dose infections were strikingly different: while low dose infection preferentially induced proliferation and effector cytokine production by CD8+ T cells , high dose infection predominantly induced proliferation and cytokine production by CD4+ T cells . This differential activation of CD4+ and CD8+ T cells by high and low dose infections respectively , was imprinted during in vitro and in vivo recall responses in healed mice . Both low and high dose-infected mice displayed strong infection-induced immunity and were protected against secondary L . major challenge . While depletion of CD4+ cells in mice that healed low and high dose infections abolished resistance to secondary challenge , depletion of CD8+ cells had no effect . Collectively , our results show that although CD8+ T cells are preferentially activated and may contribute to optimal primary anti-Leishmania immunity following low dose infection , they are completely dispensable during secondary immunity .
The spectrum of disease collectively known as leishmaniasis continues to be a major threat to global health in many regions of the world . According to the World Health Organization ( WHO ) estimate , about 15–20 million people are afflicted with the disease and close to 2 million new cases occur annually [1] . Despite intensive research , there is currently no effective licensed vaccine for prevention of human leishmaniasis . This is in part related to lack of proper understanding of the immunobiology of the disease , particularly the factors that regulate the induction , maintenance and loss of protective immunity . Because Leishmania are obligate intracellular parasites , a strong T cell-mediated immunity is critical for effective control of the infection . Indeed , T cell deficient mice are highly susceptible to Leishmania infection , and adoptive transfer of T cells restores resistance in these mice [2] . Although it is widely believed that CD4+ T cells are the key lymphocyte subset that regulates anti-Leishmania immunity , studies utilizing low dose infections show that CD8+ T cells are also important for optimal primary immunity [3] , [4] . Thus , while CD8 deficient mice are still resistant to high dose L . major infection , low dose infection of these mice results in uncontrolled parasite proliferation and impaired IFN-γ response [3] . However , no study has addressed the impact of parasite dose on the magnitude of initial T cell ( both CD4+ and CD8+ ) expansion , and whether this affects the development of secondary anti-Leishmania immunity . In addition , the contribution of CD8+ T cells to low dose infection-induced resistance is not known . In this present study , we compared the primary and secondary immune responses to low and high dose L . major infections in mice . In particular , we investigated whether parasite dose affects the quality ( subset ) and magnitude of the initial ( primary ) T cell expansion and the impact of this on infection-induced ( secondary ) anti-Leishmania immunity . In addition , we also assessed the role of CD8+ T cells in secondary immunity following primary low dose L . major infection . We show that the early immune response to primary low dose L . major infection is dominated by CD8+ T cells while high dose infection preferentially induced proliferation and IFN-γ production by CD4+ T cells . Interestingly , mice that healed their primary low and high dose infections displayed comparable delayed-type hypersensitivity ( DTH ) response and rapid parasite control following secondary L . major challenge . Depletion of CD4+ cells in mice that healed their low and high dose infections led to impaired parasite control following secondary challenge . In contrast , depletion of CD8+ cells had no effect on control of secondary L . major challenge .
All experiments in this study were reviewed and approved by University of Manitoba Animal Care and Use Committee . Protocol #: 12-072 . The University of Manitoba Animal Care and Use Committee adhere to the guidelines and standards stipulated by the Canadian Council for Animal Care . Six to eight weeks old female C57BL/6 mice were purchased either from Charles River Laboratory , St . Constante , Quebec or from the University of Manitoba Central Animal Care Services ( CACS ) breeding facility . Six to eight weeks old female C57BL/6 ( Thy1 . 1 ) mice were purchased from The Jackson Laboratories ( Bar Harbor , ME ) . All mice were maintained in specific-pathogen free environment at the CACS facility . Leishmania major parasites ( MHOM/IL/80/Friedlin ) were grown in Graces' Insect medium ( Invitrogen , Life Technologies , Burlington , Ontario , Canada ) supplemented with 20% heat inactivated fetal bovine serum ( FBS ) , 2 mM L glutamine , 100 U/ml penicillin , 100 µg/ml streptomycin , 25 mM HEPES ( complete parasite medium ) . All media additives were purchased from Invitrogen . Seven day stationary phase promastigotes were used for all infection . Groups of C57BL/6 mice ( 4–6 mice per group ) were infected with 1×103 ( low lose ) or 2×106 ( high dose ) parasites suspended in 50 µl ( footpad infection ) of sterile PBS . For secondary challenge , infected mice were challenged with 5×106 parasites in the contralateral footpads between 12–16 weeks following primary infection when lesions were fully resolved . Lesion development and delayed type hypersensitivity response ( DTH ) following primary and secondary challenge infections , respectively , were determined by measuring the thickness of infected footpads with Vernier caliper ( Fisher Scientific , Ottawa , ON Canada ) . At different times after primary infection or 3 weeks after secondary challenge , mice were sacrificed and parasite burden in the footpads was determined by limiting dilution as previously described [5] . Briefly , the footpads were collected and homogenized in 2 ml complete parasite medium using 15 ml tissue grinders ( VWR , Edmonton , AB , Canada ) . The suspension was then plated in 96-well plates in triplicates at 10-fold serial dilution , incubated for 7 days at 27°C and assessed for parasite growth under a microscope . Draining lymph-node ( dLN ) and spleen cells from healed ( >12 weeks post-infection ) donor ( Thy1 . 2 ) mice infected with low or high dose Leishmania major were labeled with CFSE dye as described previously [6] and transferred into naïve Thy1 . 1 recipient mice by intravenous injection ( 30 million cells per recipient mouse ) . Twenty-four hours after adoptive transfer , recipient mice were infected with 5 million L . major . After 7 days , mice were sacrificed and dLN cells were assessed for proliferation and cytokine ( IFN-γ and TNF ) secretion directly ex vivo by flow cytometry following 5 hr . in vitro stimulation with phorbol myristic acetate ( PMA; 50 ng/ml ) , ionomycin ( 500 ng/ml ) , and brefeldin A ( BFA , 10 µg/ml ) . BMDCs were generated from naïve mice and infected with Leishmania major promastigotes as described previously [7] . At sacrifice , the draining lymph nodes ( dLNs ) were harvested and made into single cell suspensions . In some experiments , the dLN cells were labeled with CFSE dye and cultured in 96-well round bottom plates ( 2×105/well in 200 µl ) in the presence of soluble Leishmania antigen ( SLA , 50 µg/ml ) as described previously [6] . In some experiments , the cells were co-cultured with infected BMDCs for 4 days ( at BMDC:lymph node cell ratio of 1∶100 ) . At the end of the culture period , the cells were stimulated with PMA ( 50 ng/ml ) , ionomycin ( 500 ng/ml ) and BFA , ( 10 µg/ml ) for 4 hours and routinely stained for CD4 , CD3 , CD8 , IFN-γ and TNF and analyzed by flow cytometry . CD4+ and CD8+ T cells were depleted in healed mice 24 hr . before L . major challenge by injection of 500 µg of anti-CD4 ( GK1 . 5 ) and anti-CD8 ( TIB 210 ) monoclonal antibody intraperitoneally . Previous studies from the lab show that this dose of antibody leads to complete depletion ( >98% ) of CD4+ and CD8+ positive cells , respectively , for up to 7 days . High and low dose L . major infected mice were sacrificed at days 7 and 14 weeks post-infection and the draining lymph nodes collected , minced into small pieces and digested with 2 ml RPMI-1640 medium containing 2% FBS , 1 mg/ml collagenase Type 1 A ( Sigma , Oakville ON , Canada ) and 100 µg/ml DNase I ( Roche , Mississauga ON , Canada ) by incubating for 25 minutes at 37°C . The tissues were further disrupted with a tissue grinder and the suspension was filtered through a 70 µm cell strainer ( Roche ) to remove tissue debris , and washed with 10 ml complete medium by centrifuging at 1200 rpm for 5 minutes . The cell pellets were resuspended in 2 ml complete medium , counted , stained directly ex vivo with different flourochrome-conjugated mAbs against CD11c , CD40 , CD86 , CD103 , CD8α and MHC II and analyzed by flow cytometry . High and low dose L . major infected mice were sacrificed at 14 weeks post-infection and single cell suspensions of the draining lymph nodes and spleens were made . The cell were counted and adjusted to 5×106/ml and 100 µl aliquots were stained directly ex vivo with different flourochrome-conjugated mAbs against CD3 , CD4 , CD8 , CD44 and CD62L and analyzed by flow cytometry . Student T test was used to compare mean and standard error of mean ( SEM ) between two groups . In some experiments , nonparametric one-way analysis of variance ( ANOVA ) was used to compare mean and standard deviation ( SD ) of more than two groups . Tukeys test was used where there was significant difference in ANOVA . Differences were considered significant when p<0 . 05 .
We infected mice with low or high dose L . major and at indicated times sacrificed them to determine parasite burden and number of cells in the draining lymph nodes ( dLNs ) . The pattern of lesion development in the footpads was similar in both low and high dose-infected mice although the overall swelling sizes were significantly ( p<0 . 05–0 . 001 ) higher in mice infected with high dose parasites from 2–7 weeks post-infection ( Fig . 1A ) . In addition , high dose-infected mice had significantly ( p<0 . 05–0 . 001 ) more cells in their dLNs at 2 and 3 weeks post-infection than those infected with low dose parasites ( Fig . 1B ) . Beyond 3 weeks , there was no significant difference in the number of cells in the dLNs , despite continued difference in lesion size that lasted up to 8 weeks post-infection ( Fig . 1A ) . Parasite burden in the lesions of high dose-infected mice was also significantly ( p<0 . 05–0 . 001 ) higher at 2 , 3 , 5 and 7 weeks post-infection than in low dose-infected mice but this difference was absent by 9 weeks post-infection ( Fig . 1C ) . These results show that although the size of cutaneous lesion , inflammation and cell recruitment into the dLN are different following high and low dose L . major infections , the pattern and time to lesion resolution and parasite clearance are comparable . CD8+ T cells play important role in optimal immunity to primary low dose L . major infection but are dispensable for immunity high dose infections [3] . To determine the influence of parasite dose on the early expansion and activation of different T cell subsets following L . major infection , we co-cultured CFSE-labeled dLN cells from mice infected with low and high dose L . major ( one week post-infection ) with L . major-infected BMDCs and assessed T cell proliferation and cytokine ( IFN-γ and TNF ) production by flow cytometry . There was a striking difference in proliferation of CD4+ and CD8+ T cells from dLNs of low and high dose-infected mice with high dose infection inducing significantly ( p<0 . 05–0 . 01 ) more CD4+ T cell proliferation compared to low dose infections ( Fig . 2A and B ) . Conversely , low dose infection induced significantly ( p<0 . 05–0 . 01 ) more CD8+ T cell proliferation than high dose infection ( Fig . 2A and B ) . Consistent with this , the percentage ( Fig . 2C and D ) and mean fluorescence intensity ( MFI , Fig . S1A and S1B ) of IFN-γ and TNF-producing CD4+ T cells in dLNs of high dose-infected mice were higher than those of low dose-infected mice . In contrast , the percentage and MFI of IFN-γ and TNF-producing CD8+ T cells were significantly ( p<0 . 01 ) higher in low dose-infected mice ( Fig . 2E and F , Fig . S1C and S1D ) . Consistent with the higher CD4+ T cell response , the percentage and absolute numbers of MHC II+CD11c+ ( dendritic ) cells in dLNs of high dose-infected mice were significantly ( p<0 . 05 ) higher than those of low dose-infected mice ( Fig . 2G ) . Interestingly , the percentage of CD11c+CD103+CD8α+ dendritic cells was higher in the dLNs of low dose than in high dose-infected mice ( Fig . 2H ) . However , there was no significant difference in the expression of key costimulatory molecules ( including CD86 and CD40 ) on dendritic cells from high and low dose-infected mice ( Fig . S2A and S2B ) . Next , we wished to determine whether the early differential expansion of CD4+ and CD8+ T cells subsets by high and low dose L . major infection , respectively , is transient and related to differences in parasite burden at the infection site ( see Fig . 1C ) . Therefore , we assessed T cell responses ( proliferation and cytokine production ) after 12 weeks when lesion is fully resolved and parasite burden in the footpads and dLNs of high and low dose-infected mice are comparable ( see Fig . 1C ) . Flow cytometric analyses show that akin to observations during early infection , the percentage of proliferating CD4+ T cells was significantly ( p<0 . 01 ) higher in high dose-infected mice compared to the low dose infected group ( Fig . 3A and B ) . In contrast , there were significantly ( p<0 . 01 ) more proliferating CD8+ T cells in the low dose-infected mice compared to the high dose-infected mice ( Fig . 3A and B ) . Furthermore , the percentage of proliferating and IFN-γ and TNF-secreting CD4+ T cells in mice infected with high dose parasites were significantly ( p<0 . 05 ) higher than those from low dose-infected mice ( Fig . 3C and D ) . In contrast , mice that healed their low dose infection had significantly ( p<0 . 05 ) higher percentage of proliferating and cytokine- ( IFN-γ and TNF ) secreting CD8+ T cells than those that healed high dose infection ( Fig . 3E and F ) . In addition , and consistent with observations during early infection , the absolute numbers of dendritic cells ( CD11c+ ) expressing MHC II were significantly ( p<0 . 05 ) higher in healed high dose-infected mice compared to those from healed low dose-infected mice ( Fig . 3G ) . Collectively , these results suggest that an early antigen encounter creates an imprint in T cell subset responses that is maintained throughout the course of L . major infection . To confirm that the preferential activation of CD8+ and CD4+ T cell responses by low and high dose infections , respectively , is physiologic and occurs in vivo , we adoptively transferred CFSE-labeled CD3+ cells isolated from Thy1 . 2 mice that healed either low or high dose L . major infection into naïve Thy1 . 1 recipient mice . Recipient mice were challenged 24 hr . later with L . major , sacrificed after one week and donor ( Thy1 . 2+ ) cells from the dLNs were assessed directly ex vivo for proliferation and cytokine ( IFN-γ and TNF ) production by flow cytometry . As shown in Fig . 4 , donor cells from low dose-infected mice contain significantly ( p<0 . 05 ) higher percentage of proliferating ( CFSElo ) CD8+ cells whereas those from high dose-infected mice had significantly ( p<0 . 05 ) higher percentage of proliferating CD4+ cells ( Fig . 4A ) . Similar to the in vitro findings , the percentage of proliferating IFN-γ and TNF-producing CD8+ T cells were significantly ( p<0 . 05 ) higher in cells from low dose-infected donor mice compared to those from high dose-infected donor mice ( Fig . 4B and C ) . In contrast , donor cells from high dose-infected mice had significantly ( p<0 . 05 ) more proliferating and cytokine-producing CD4+ cells than those from low dose-infected mice ( Fig . 4B and C ) . Collectively , these results show that high and low dose L . major infections-induced differential T cell responses occur in vivo and create an imprint that is maintained over time during the course of infection . Although we found that the pattern of lesion development and parasite clearance are similar in mice infected with low and high dose L . major , it is possible that they induce different types or subsets of memory cells , which might affect the quality of secondary anti-Leishmania immunity . Indeed , previous studies show that the quality of memory T cell response is influenced in part by antigen dose [8] , [9] . Interestingly , we found no significant difference in the percentages of CD44+CD62Llo ( effector memory-like , Tem ) or CD44+CD62Lhi ( central memory-like , Tcm ) CD4+ and CD8+ T cells in the dLNs and spleens of mice that healed primary low and high dose L . major infections ( Fig . S3A and S3B ) , suggesting that parasite dose may not influence the quality of anti-Leishmania memory T cell responses . In line with this , when mice that healed high or low dose primary infection were rechallenged with either high ( Fig . 5A and C ) or low ( Fig . 5B and D ) dose L . major , there were no significant differences in either DTH response ( Fig . 5A and B ) or rapid parasite control in the challenged footpads ( Fig . 5C and D ) , suggesting that both low and high dose infections induce qualitatively comparable infection-induced resistance in healed mice . Although we previously reported that CD8+ T cells are important for optimal immunity to primary low dose L . major infection [3] it is not known whether they also contribute to secondary anti-Leishmania ( infection-induced ) immunity . We observed herein a preferential expansion of CD8+ cells during recall responses in mice that healed their low dose L . major infection ( see Figs . 3B and 4B ) , suggesting that CD8+ T cells might also be critical for secondary anti-Leishmania immunity in low dose-infected mice . Therefore , we treated healed low and high dose-infected mice with anti-CD4 or anti-CD8 mAb ( to deplete CD4+ and CD8+ cells , respectively , Fig . 6A ) and challenged them after 24 hr . with either high or low dose L . major . Surprisingly , depletion of CD8+ cells in both high and low dose-infected mice did not affect DTH response ( Fig . 6B ) and rapid parasite control ( Fig . 6C ) following high dose challenge . In contrast , CD4+ T cell depletion in both low and high dose-infected mice resulted in significant ( p<0 . 05 ) impairment in DTH ( Fig . 6B ) and rapid parasite control ( Fig . 6C ) . Similar results were also obtained following low dose L . major challenge of low or high dose healed mice following CD4+ or CD8+ T cell depletion ( Fig . S4A and S4B ) . Collectively , these results indicate that although low dose L . major infection preferentially expands CD8+ T cells , CD4+ T cells are the major players that mediate secondary anti-Leishmania immunity in mice . They further show that although CD8+ T cells are important for optimal immunity to primary low dose infection [3] , they are completely dispensable during a secondary challenge .
Clinical observations and experimental studies suggest that the development of effective cell-mediated immunity is essential for protection against leishmaniasis . However , the lack of a universally approved and effective vaccine against human leishmaniasis suggests that we still do not completely understand the factors that regulate the development of cell-mediated immunity against the disease . Although CD4+ T cells are critical for protective immunity in cutaneous leishmaniasis [10] , CD8+ T cells have also been shown to be essential in certain situations , particularly in low dose infections [3] , [11] . During low dose L . major infection , CD8+ T cells were shown to contribute to lesion resolution and parasite control by producing IFN-γ that augment optimal CD4+ Th1 response [3] , [11] . However , whether CD8+ T cells also contribute to secondary anti-Leishmania immunity following resolution of primary infection is unclear . In addition , no study has investigated the impact of low parasite dose on secondary anti-Leishmania immunity . We show here that the patterns of lesion development and parasite burden were similar in both high and low dose infections , the quality of the immune response was strikingly different . Whereas high dose infection induced strong CD4+ T cell proliferation and Th1 cytokine response , low dose infection predominantly activates CD8+ T cells . Using adoptive transfer of T cells from healed mice into naïve congenic recipients , we also demonstrated that this differential activation of CD4+ and CD8+ T cells by high and low dose infections , respectively , is observed in vivo following L . major challenge . Interestingly , while depletion of CD4+ T cells in mice that healed both high and low dose infections led to loss of immunity following secondary L . major challenge , depletion of CD8+ T cells had no effect . Taken together , the results presented here show that although low dose L . major infection preferentially activates CD8+ T cells that contribute to optimal primary immunity , they are completely dispensable for resolution of secondary L . major challenge . Previous studies showed that CD8+ T cells produce large amounts of IFN-γ following L . major infection and were critical for optimum primary immunity [3] , [11] . In addition , earlier studies suggest that CD8+ T cells are activated following secondary L . major challenge , suggesting that they contribute to secondary anti-Leishmania immunity [12] , [13] . We found that CD8+ cells are completely dispensable for protective secondary anti-Leishmania immunity . Healed mice ( following primary low dose or high dose infections ) depleted of CD8+ T cells before virulent low or high dose rechallenge were as resistant as those treated with control-Ig . This is an agreement with our previous observation which showed that CD8 deficient mice that healed their primary high dose infection are resistant to virulent L . major challenge [3] . However , whether CD8+ T cells are critical for secondary immunity following high or low dose challenge in mice that were exposed to primary low dose infection has never been reported . In this current study , we addressed this question by depleting CD8+ T cells in mice that healed their primary low dose infection before being challenged with either low or high dose virulent parasites . We show that regardless of the dose at both primary and secondary challenges , CD8+ T cells are completely dispensable during secondary anti-Leishmania immunity ( Fig . 6C and D ) . In contrast , depletion of CD4+ T cells led to complete loss of infection-induced immunity . Collectively , these observations suggest that the role of CD8+ T cells may be limited to helping for optimal activation of CD4+ Th1 cells during primary infection . Once effective CD4+ Th1 response is induced , CD8+ T cells are no longer relevant , thus making them dispensable during a secondary response . We believe that differences in animal models , parasite strain and experimental design could account for the discrepancy between our findings and the studies that found a role for CD8+ T cells in secondary immunity . For example , in those studies , splenocytes were first depleted of CD4+ cells and then cultured in vitro for extended period of time before assessing for IFN-γ production by CD8+ T cells [12] , [13] . Such in vitro culture conditions could potentially influence the magnitude of CD8+ T cell responses that otherwise would not be seen in short-term and/or bulk whole cell cultures as performed in our study . Why would high and low dose infections differentially activate CD4+ and CD8+ T cells , respectively ? It is conceivable that this may be related in part to differences in activation threshold for CD4+ and CD8+ T cells . It has been shown that naïve CD4+ T cells require at least 6 hours of contact with APCs presenting their cognate peptides in order to acquire optimum signals leading to activation , proliferation and cytokine production [14] . In contrast , naïve CD8+ T cells require less than 2 hours of antigenic stimulation to acquire enough signals required for their activation , proliferation and cytokine release , suggesting that the requirements for activating CD8+ T cells are less stringent [15] . Hence , low dose infection provides lower antigen availability that favors activation of CD8+ T cells . In contrast , high infectious dose provides high antigen load that could overcome the need for longer contact thus favoring expansion of CD4+ T cells . In addition , high dose infection was associated with strong upregulation of MHC class II molecules on DCs ( Fig . 2G ) , which would potentially favor activation CD4+ T cells . Further more , we found that the percentage of CD103+CD8α+ dendritic cells in the draining lymph nodes of low dose-infected mice were significantly higher than those in high dose-infected mice . Given that CD103+CD8α+ dendritic cells cross present exogenous antigens to CD8+ T cells leading to their activation and effector cytokine response [16] , [17] , it is conceivable that the higher of induction of this subset of dendritic cells contributes to preferential activation of CD8+ T cells following low dose infection . Although differences in co-stimulatory molecules expression has been associated with differences in activation of CD4+ and CD8+ T cells [18] , it is unlikely that these contributed to differential induction of CD4+ and CD8+ T cells in our model system . In line with this , we did not observe any difference in expression of CD40 and CD86 molecules on DCs from the dLNs following low and high dose infections , suggesting that the effect of antigen dose is mostly restricted to TCR-peptide interaction and not on co-stimulation . Comparing the lesion outcome and parasite burden in mice infected with high and low dose L . major infection , we found that although high dose infection was associated with significantly higher lesion size and parasite burden early during infection , disease resolution ( healing ) occurred almost at the same time and this was associated with comparable level of protection following virulent challenge . This finding , which is in agreement with our previous report [3] has important implications in vaccine development as well as vaccination strategies against cutaneous leishmaniasis . Following recovery from natural L . major infection ( which is usually self-resolving ) , a long-term ( sometimes lifelong ) immunity develops to reinfection . This observation is the basis for leishmanization , which is the deliberate inoculation of lesion-derived virulent parasites into hidden parts of the body in order to prevent a more serious visible cutaneous disease . Leishmanization is the oldest and only effective preventive practice against human cutaneous leishmaniasis [19] . The usual practice in leishmanization is to employ a relatively high dose inoculum because it is believed that such high dose is able to induce strong inflammatory and immune responses necessary for protection against subsequent reinfections . As a result , some leishmanized individuals develop large ( sometimes non-healing ) lesions that require medical treatment . In some cases , exacerbated chronic skin disease and/or immunosuppression have been reported [20] . Due to the effectiveness of leishmanization , recent efforts have focused on ways to make the practice safer , including suggestions to include killed parasites in the inoculum [21] or to use genetically engineered attenuated parasites [22] . Whether high doses of parasites during leishmanization ( as is currently practiced ) is required for protection is unclear . We show here that despite inducing comparatively lesser inflammatory responses ( smaller lesion sizes ) , low dose infection induced secondary immunity and protection following virulent L . major challenge comparable to high dose infection . This observation indicates that vaccination with large dose of live virulent parasite is not necessary to achieve protection . They further suggest that leishmanization with low dose inoculum could be a viable alternative practice as it would lead to smaller lesion at the inoculation site that is less prone to ulceration and secondary bacteria infection . In line with this , it has been shown that low dose infection of the highly susceptible BALB/c mice leads to resistance and protection against virulent L . major challenge [23] . Collectively , our study shows that parasite dose critically influence the magnitude of expansion of CD4+ and CD8+ T cells following L . major infection . While low dose infection preferentially activates CD8+ T cells , high dose infection leads to preferential activation of CD4+ T cells . Surprisingly , despite the strong activation of CD8+ T cells and their importance in primary immunity following low dose infection , secondary immunity in mice that healed their low dose L . major infection was completely dependent on CD4+ ( and not CD8+ ) T cells .
|
It is known that CD8+ T cells are important for primary immunity to low dose L . major infection , but whether low dose-induced immunity is long lasting and whether CD8+ T cells are also important for memory immune response to low dose L . major is unknown . We studied whether infectious dose affects primary anti-Leishmania immunity and the contribution of CD8+ T cells in immunity following recovery from low and high dose infections . We found that low and high dose infections preferentially induced proliferation and cytokine production by CD8+ and CD4+ T cells , respectively , during early and late stages of infections . Also , both low and high dose-infected mice were solidly protected against secondary L . major challenge . Depletion of CD4+ cells in mice that healed low and high dose infections abolished resistance to secondary challenge , but depletion of CD8+ cells had no effect . Together , our results show that although CD8+ T cells are selectively activated and contribute to optimal primary immunity after low dose infection , they are not required for secondary immunity . This research further enhances our understanding of the immunobiology of cutaneous leishmaniasis .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"infectious",
"diseases",
"veterinary",
"diseases",
"zoonoses",
"medicine",
"and",
"health",
"sciences",
"leishmaniasis",
"biology",
"and",
"life",
"sciences",
"immunology",
"protozoan",
"infections",
"parasitic",
"diseases",
"veterinary",
"science",
"parasitology"
] |
2014
|
CD8+ T cells Are Preferentially Activated during Primary Low Dose Leishmania major Infection but Are Completely Dispensable during Secondary Anti-Leishmania Immunity
|
It has been proposed that neural noise in the cortex arises from chaotic dynamics in the balanced state: in this model of cortical dynamics , the excitatory and inhibitory inputs to each neuron approximately cancel , and activity is driven by fluctuations of the synaptic inputs around their mean . It remains unclear whether neural networks in the balanced state can perform tasks that are highly sensitive to noise , such as storage of continuous parameters in working memory , while also accounting for the irregular behavior of single neurons . Here we show that continuous parameter working memory can be maintained in the balanced state , in a neural circuit with a simple network architecture . We show analytically that in the limit of an infinite network , the dynamics generated by this architecture are characterized by a continuous set of steady balanced states , allowing for the indefinite storage of a continuous parameter . In finite networks , we show that the chaotic noise drives diffusive motion along the approximate attractor , which gradually degrades the stored memory . We analyze the dynamics and show that the slow diffusive motion induces slowly decaying temporal cross correlations in the activity , which differ substantially from those previously described in the balanced state . We calculate the diffusivity , and show that it is inversely proportional to the system size . For large enough ( but realistic ) neural population sizes , and with suitable tuning of the network connections , the proposed balanced network can sustain continuous parameter values in memory over time scales larger by several orders of magnitude than the single neuron time scale .
The question of whether balanced networks can produce persistent activity has attracted considerable interest in recent years . Several works explored architectures which give rise to slow dynamics in balanced networks , characterized by the coexistence of multiple discrete balanced states [22] . In several recent works multi-stability resulted from the existence of clustered connectivity , and slow transitions were observed between the discrete semi-stable states [23–25] . Other works [8 , 26] demonstrated that a discrete set of semi-stable states can be embedded in a balanced neural network , using a similar construction as employed in the classical Hopfield model of associative memory [27] . A few works have addressed the possibility that balanced neural networks may generate slow persistent activity over a continuous manifold . Such dynamics were demonstrated in simulations of neural networks that included short-term synaptic plasticity [28] , or a derivative-feedback mechanism [29 , 30] . Previous works have not demonstrated the existence of a continuum of steady states in a balanced neural network analytically , and it has remained unclear whether such a continuum can be obtained without evoking additional mechanisms ( such as short-term synaptic plasticity , or derivative-feedback ) . In addition , the influence of the chaotic dynamics on the persistence of stored memory has not been analyzed . These questions are addressed in the present work . Below , we identify an architecture in which slow dynamics are attainable in a simple form of a balanced neural network . We prove analytically the existence of a continuous attractor in our model in the large population limit . In finite networks , we show that the chaotic noise drives diffusive motion along the attractor—leading , among other consequences , to slowly decaying spike cross-correlations . We show that the diffusivity scales inversely with the system size , as predicted previously for continuous attractor networks with intrinsic sources of neuronal stochasticity . With a reasonable number of neurons and suitable tuning , our model network exhibits slow dynamics over a continuous manifold of semi-stable states , while exhibiting single neural dynamics which appear stochastic , as observed in cortical circuits .
Our neural network model is based on the classical balanced network model presented in Refs . [3 , 7 , 8] . This model consists of two distinct populations of binary neurons , one inhibitory and the other excitatory . The recurrent connectivity is random and sparse , with a probability K/N for a connection , where N is the population size ( assumed for simplicity to be the same in both populations ) , K is the average number of connections per neuron from each population , and the connection strength is ∼ 1 / K . For 1 ≪ K ≪ N and over a wide range of parameters , the mean population activity settles to a fixed point ( the balanced state ) where on average the total excitation received by each neuron is approximately canceled by the total inhibition ( to leading order in 1 / K ) , and the neural dynamics are driven by the fluctuations in the input . The single neuron activity appears noisy , neither of the populations is fully activated or deactivated , and the overall network state is chaotic . Despite the nonlinearities involved in the dynamics of each neuron , the population averaged activities in the balanced state are linear functions of the external input [3 , 7] . We exploit this linearity to build a simple system of two balanced networks projecting to each other . The intuition comes from a simple model of a continuous attractor neural network consisting of linear neurons arranged in two populations that mutually inhibit each other , Fig 1A . The linear rate dynamics of this system are given by: τ r ˙ = - r + W r + E , ( 1 ) where E = [E0 , E0] , E0 > 0 is an external input and W = 0 - J - J 0 . ( 2 ) For J = 1 the system has a vanishing eigenvalue , and the fixed points form a continuous line: r1+r2 = E0 . The simple neural architecture of Fig 1A was used as a basis for modeling the dynamics of neural circuits responsible for memory and decision making in the prefrontal cortex [31–35] . In our model , a balanced subnetwork replaces each of the populations of Fig 1A , and the inhibitory population in each subnetwork projects to the excitatory population of the other subnetwork , Fig 1B and Methods . Thus , the model consists of two reciprocally inhibiting balanced neural populations . We consider first a scenario in which the inhibitory connectivity between the two sub-networks is all-to-all . Therefore , the network includes a combination of strong , random synapses within each sub-network and highly structured , weak synapses between the two sub-networks . This scenario lends itself to analytical treatment of finite N effects ( see below ) . Later on , we present results also for an alternative scenario , in which the connections between the two-subnetworks are sparse , random , and strong ( Additional randomness in connectivity and inputs ) . We first examine whether the two-subnetwork architecture can give rise to a continuum of balanced states . The parameters of the network connectivity in our model are summarized in Fig 1 and in Methods . The mutual inhibition between the subnetworks is assumed to be all to all , and the interaction strength is scaled such that the total inhibitory input to each neuron , coming from the opposing subnetwork scales in proportion to K . Similar to the case of a single balanced network [7] , the mean field dynamics of the population averaged activities for N → ∞ and K ≫ 1 are given by: τ i m ˙ i = - m i + H ( - u i / α i ) , ( 3 ) where m i ( t ) = 1 / N ∑ k = 1 N σ i k ( t ) [i = 1 ( 2 ) for the excitatory ( inhibitory ) population of the first subnetwork , and similarly i = 3 , 4 in the second subnetwork] , σ i k ( t ) is the state of neuron k in population i at time t , H ( x ) is the complementary error function , and ui ( αi ) is the mean ( variance ) of the input to a neuron in population i , averaged over the population and over the random connectivity ( Methods ) . This equation is an approximation which becomes exact in the limit K → ∞ . To check whether there exist parameters for which the system has a continuum of balanced states , it is convenient to write the steady state equations of the above dynamics , while making use of the assumption that K is large . In the limit K → ∞ these equations become linear ( Methods ) : m 1 - J E m 2 - J ˜ m 4 + E 0 = 0 , m 1 - J I m 2 = 0 , m 3 - J E m 4 - J ˜ m 2 + E 0 = 0 , m 3 - J I m 4 = 0 . ( 4 ) By choosing the interaction strength between the two subnetworks to be J ˜ = J E - J I , this system becomes singular , and has a continuum of solutions arranged on a line in the mean activities space , which represent a continuum of stable balanced states . Next , we consider the realistic situation in which N is finite in the two-subnetwork model , while still requiring that N ≫ K ≫ 1 . Instead of adding noise to the dynamics of each neuron , we ask whether the chaotic dynamics are sufficient to drive diffusive motion along the approximate attractor . This question is motivated by the fact that diffusive dynamics are observed in model neural networks of intrinsically noisy neurons , with a finite number of neurons [19] . In addition , this question is motivated by evidence of diffusive dynamics underlying continuous parameter working tasks—as observed both in the behavioral data and in its neural correlates in the prefrontal cortex [10 , 11 , 36] . Since the population dynamics are no longer given by Eq 3 , we performed large scale numerical simulations of networks with N ranging between 104 to 15 × 104 ( additional details on the simulations can be found in Methods ) . To simplify the analysis and the presentation , we chose the random weights within each subnetwork such that they precisely mirrored each other , which ensured that the fixed point would be symmetric ( m1 = m3 and m2 = m4 ) . If , alternatively , the connections in each subnetwork are chosen independently , the fixed point deviates slightly from this symmetry plane ( this deviation approaches zero for infinite networks ) . However , all the results described below remain qualitatively valid ( see below , Additional randomness in connectivity and inputs ) . The neural activity observed in our simulations is irregular and individual neurons approximately exhibit exponential ISI distributions similar to those observed in the two population case , although their dynamics are deterministic ( Fig 3 ) . To test whether the network can perform short term memory tasks , we initiated the population activities such that the network state was close to some point along the approximate line attractor . Fig 4A shows the resulting dynamics of the four populations: the activities persisted for a few seconds before decaying towards the symmetric fixed point . Fig 4B shows the projection along the slow direction , X ( t ) ( defined in Eq 17 ) , again revealing the slow decay of the initial state . Fig 4C shows statistics of trajectories that start from two initial positions along the approximate attractor , when J ˜ is tuned to achieve λ−1 ≃ 9 s . The state of the network enables discrimination between the two conditions over a time scale of several seconds . The ability to do so with high confidence is influenced both by λ and the stochasticity of the motion , which we characterize in the following section ( see also Discussion ) . S1 Fig shows the mean square displacement ( MSD ) from the starting point for the same dataset , averaged over all trials . Long after initialization , the population activities fluctuate around the symmetric fixed point , along a line corresponding to the approximate attractor: a projection on the m1 − m3 plane is shown in Fig 4D . Fig 4E demonstrates that X ( t ) exhibits slow diffusive dynamics . To demonstrate that the dynamics are effectively one dimensional , a projection on a perpendicular direction is shown as well . The diffusion along the approximate attractor implies that the population activities are correlated over long time scales , up to order λ−1: Fig 7A shows examples of the population correlation functions Cm , which differ dramatically from those of the single balanced network , Fig 6 ( note the different time scales in the two sets of figures ) . Spike trains generated by single neuron pairs are correlated over long time scales as well , since all neurons in the network are coupled to the collective diffusion along the approximate attractor . However , a reliable observation of the slowly decaying correlation in a single pair might require an unrealistically long recording time . This difficulty can be overcome potentially by considering the simultaneous activity of multiple neurons: for example , we find in our simulations of a network with N = 105 that for 15 minutes of simulated time , a simultaneous recording from ∼50 or more neurons from each population would be sufficient to reliably observe the slow temporal decay of the correlations , Fig 7C , whereas a simultaneous recording from ten neurons over 15 minutes may be insufficient . As demonstrated in Fig 7 ( B ) –7 ( D ) the noise falls as one over the number of measured neurons and as one over the total recording time . Hence , by extrapolating from the results in Fig 7 ( B ) –7 ( D ) , ∼12 hours of recording would be required to obtain a measurable correlation signal from a single pair of neurons . Next , we briefly address the chaotic nature of the noise that drives diffusive motion . Fig 8A shows results from multiple simulations in which the initial network state differed solely by a flip of one neuron in each population ( out of ∼105 neurons ) . All other parameters , including the asynchronous update schedule and the network weights were identical across runs . The time dependence of the variance across different runs is similar to the variance over realizations of an OU process , Fig 8B , with a similar diffusion coefficient as observed in the fit for G ( X , Δt ) , Fig 5 ( C ) and 5 ( E ) . Thus , the different initial conditions are equivalent to different realizations of dynamic noise that drives diffusive motion along the approximate line attractor . In addition to the results described above , we investigate several scenarios in which we introduce additional randomness , either frozen or dynamic . First , we relax the assumption that connections in the two sub-networks precisely mirror each other . This assumption was made above for convenience: the precise identity of the synaptic connections simplifies the numerical analysis since it ensures a precise symmetry of the dynamics around the hyperplane X = 0 . S2 Fig demonstrates that the main conclusions of our analysis remain valid when the connectivity in each sub-network is drawn independently: dynamics are slow along the approximate line attractor , and the diffusion coefficient along the line scales as 1/N with a prefactor which is somewhat larger than the value observed in Fig 5 . The main new feature that arises when the synaptic connections are drawn independently in the two sub-networks , is that the relaxation point of the dynamics along the approximate attractor deviates from the hyperplane X = 0 . The characteristic magnitude of this deviation decays monotonically to zero with increase of the system size N . We note that the mean field description of the dynamics for finite K ≫ 1 and in the limit N → ∞ , is identical to the mean field dynamics associated with the perfectly symmetric scenario . Next , we consider a scenario in which the inhibitory connections between the two sub-networks are random , and follow the same basic architecture as the connections within each sub-network . Therefore , instead of assuming weak all-to-all connections of order K / N , we include random connections of order 1 / K , with a probability K/N for a synaptic connection . In addition , we relax the assumption of mirrored connections in the two sub-networks . In this case it is straightforward to show that the mean-field equations remain identical to those associated with the case of all-to-all connections , in the limit N → ∞ , K → ∞ . Therefore , a continuous set of balanced states can be achieved ( Methods ) . When K is large and finite , the mean-field equations are slightly different in the two scenarios ( Methods ) . The main outcome of this difference is a shift of the unstable fixed points of the dynamics from the planes m1 = m2 = 0 and m3 = m4 = 0 . Consequently , there is a certain degree of activity in both sub-networks even in the unstable fixed points of the dynamics . This is shown in S3 Fig More significantly , S3 Fig demonstrates that in the case of random and sparse connections between the two sub-networks , the dynamics exhibit the same characteristics as in the case of all-to-all connectivity , and can be accurately approximated as an OU process over time scales longer than τ . The coefficient of diffusion D scales linearly with 1/N , with a prefactor which is close to the one observed in S2 Fig . Finally , we explore the effects of stochasticity in the input E0 to the network ( see Fig 1B ) . S4 Fig demonstrates that even when the inputs include a large degree of temporal variability , and the noise injected to all the neurons in the network is highly correlated , the network exhibits slow dynamics along an approximate attractor , with statistics that are qualitatively similar to what we present above .
Several models of decision making circuits in the prefrontal cortex were based on the simple neural architecture of Fig 1A [32] . This network architecture can precisely generate a continuous attractor if the activity of single units is a linear function of their input . While the linear dynamics provide a simple intuition for the principles underlying continuous attractor dynamics in recurrent neural networks [31] , it is more difficult to obtain a continuum of steady states using the above network architecture when single neural responses are nonlinear . Therefore , specifically tuned forms of nonlinearity [33 , 35] , or more elaborate network architectures—still based on mutual inhibition between two or more neural populations have been proposed [33 , 34] . In this context the linear input-output relationship , characterizing the single balanced network of Ref . [3] , is a useful computational feature that facilitates the construction of a continuous attractor network based on the simple architecture of Fig 1A . However , the main motivation for considering the balanced state in this work lies in its ability to account for the irregular spiking of single neurons in cortical circuits . Continuous attractor networks are an important model for maintenance of short-term memory in the brain . The memory is represented by the position along the attractor , and therefore the stochastic motion along the attractor determines the fidelity of memory retention . Since the dynamics of our proposed network are well characterized as an OU process over time scales longer than τ and over a large range of positions along the approximate attractor , it is straightforward to assess how the position along the approximate attractor evolves in time . All aspects of the trajectory can be easily inferred based on the initial state along the approximate attractor , the time interval , and the two parameters which characterize the OU process: λ and D . Similar considerations can be applied also for noisy continuous attractors in which the stochasticity arises from mechanisms other than the chaotic dynamics studied here [19 , 20] . We next discuss how the decay time and the diffusivity depend on the parameters of our model . The decay time λ−1 can be calculated exactly in the limit of N → ∞ , K ≫ 1 ( Fig 2 ) . It is interesting to note that there are competing influences of K on the tuning of the attractor: with increase of K , λ−1 becomes more sensitive to J ˜ . However , when K is reduced , the nullclines ( Fig 2 ) become less linear , causing deviations from the ideal behavior far from the symmetry point . We showed that for K ∼ 103 and τ = 10ms it is possible to achieve persistence over several seconds , while the decay time and diffusion coefficient are approximately constant along a wide range of positions . This requires to tune J ˜ to a relative precision of order 0 . 1% . Since all times scale linearly with the intrinsic time constant , longer persistence times ( or a weaker tuning requirement ) can be achieved if the intrinsic time constants of individual units is longer that the value of 10 ms assumed in our examples . Intrinsic neural persistence or slow synapses could potentially contribute to this goal under more realistic biophysical descriptions of the neural dynamics . Finally , we note that the requirement for precise tuning of the connectivity is a characteristic feature of all continuous attractor models . Several works have proposed ways to achieve tuning through plasticity mechanisms [37 , 38] , or ways to stabilize the dynamics by additional mechanisms such as synaptic adaptation [28 , 39 , 40] or negative derivative feedback [29 , 30] , in order to increase the persistence time . The diffusive motion along the approximate attractor , which is the main focus of this work , poses an additional limitation on the persistence of short term memory . While appropriate readout mechanisms may be able to take into account the systematic drift caused by the decay towards to the symmetry point , random diffusion inherently degrades the information stored in the position along the attractor . With 105 neurons per population , random diffusion over an interval of one second causes a deflection in X with a standard deviation of ∼10−2 . This quantity should be compared with the possible range of X , which is approximately [-0 . 2 , 0 . 2] in our parametrization of the position along the approximate attractor ( we verified that the dynamics are accurately approximated as an OU process over the range [-0 . 1 , 0 . 1] ) . Therefore , with the tuning chosen in Fig 4C , where λ−1 ∼ 10s , and with N = 105 , the limiting factor for discrimination between nearby stimuli after a delay period of order 1 s is the diffusive dynamics along the approximate attractor . Random diffusion in continuous attractor networks of Poisson neurons is also often very significant [19 , 20] . The diffusivity can be suppressed by increasing the number of neurons , increasing the intrinsic time constant of individual neurons and synapses , or by assuming that the firing of individual neurons is sub-Poisson [19 , 41] . Our proposed model for a line of persistent balanced states similarly predicts a significant degree of random diffusion , highlighting the need to better understand how noise influences the retention of continuous parameter memory in cortical circuits . There are several ways in which the random , diffusive component of the motion can potentially be reduced: First , by increasing the number of neurons . We presented results for networks containing ( altogether ) up to 6 × 105 neurons , and it is straightforward to extrapolate our estimates for D to larger networks based on the 1/N dependence of the diffusion coefficient . Second , the diffusion coefficient is expected to decrease significantly if slow synapses participate in the dynamics [19] , or if the intrinsic time constant of the neurons is increased . Third , additional mechanisms such as synaptic adaptation [28 , 40] or derivative feedback [29] may perhaps contribute to a reduction in the diffusivity . Finally , an intriguing possibility is that highly structured and tuned connectivity can yield improved robustness to noise in a balanced state , as hinted by recent results on predictive coding in spiking neural networks [42] .
In our model , two balanced neural subnetworks inhibit each other reciprocally: the inhibitory population in each subnetwork projects to the excitatory population of the other subnetwork , Fig 1B . As in Refs . [3 , 7] , the neurons are binary and are updated asynchronously , at update times that follow Poisson statistics . The mean time interval between updates is τE ( τI ) for neurons in the excitatory ( inhibitory ) populations . In each update of a neuron k from population i , the new state of the neuron σ i k is determined based on the total weighted input to the neuron , σ i k = Θ ( u i k ) , ( 8 ) where Θ is the Heaviside step function , and u i k is the total input to the unit at that time , u i k = ∑ l = 1 4 ∑ j = 1 N l J k l i j σ l j ( t ) + K E 0 l - T k . ( 9 ) Here , Tk is the threshold and E0 is an external input . We chose the external input to be zero for the inhibitory populations and to be positive ( and constant ) for the excitatory populations . Connections within each network are random with a connection probability K/N , where 1 ≪ K ≪ N . Here N is the population size ( chosen to be equal in all populations for simplicity ) and K is the average number of inputs a neuron gets from each population . Connection strengths are: J E E / K , J I E / K , J E I / K and J I I / K according to the identity of the participating neurons . Without loss of generality , we have chosen JEE = JIE = 1 and defined JEI ≡ −JE , JII ≡ −JI . Mutual inhibition is generated either by weak all-to-all connections ( in all figures except for S3 Fig ) , or by strong random and sparse connections ( S3 Fig ) . In the former scenario , synapses of strength - J ˜ K / N connect each inhibitory neuron to all excitatory neurons in the excitatory population of the other subnetwork . In the latter scenario ( S3 Fig ) , connections from each inhibitory population to the excitatory population of the other subnetwork are chosen randomly with a connection probability K/N , and with strength - J ˜ / K . An excitatory feed-forward input K E 0 is fed into both excitatory populations . We denote by ui the mean of u i k over all the neurons k within the population i and over realizations of the connectivity: u 1 = K ( m 1 - J E m 2 - J ˜ m 4 + E 0 ) - T 1 , u 2 = K ( m 1 - J I m 2 ) - T 2 , u 3 = K ( m 3 - J E m 4 - J ˜ m 2 + E 0 ) - T 3 , u 4 = K ( m 3 - J I m 4 ) - T 4 . ( 10 ) Here mi are the population averaged activities . The variance of u i k over all the neurons k within the population i and over realizations of the connectivity is given ( to leading order in K/N ) by: α 1 = m 1 + J E 2 m 2 , α 2 = m 1 + J I 2 m 2 , α 3 = m 3 + J E 2 m 4 , α 4 = m 3 + J I 2 m 4 . ( 11 ) These expressions are obtained in similarity to the derivation of the variances in Ref . [7] . Note that the all-to-all inhibitory connections between the subnetworks contribute only terms of higher order in K/N . In the scenario where the connections between subnetworks are randomly drawn ( S3 Fig ) , the variance of the input to the excitatory neurons includes an additional term , due to the variability of inhibitory synapses from the opposing sub-network . In this scenario α 1 = m 1 + J E 2 m 2 + J ˜ 2 m 4 , α 2 = m 1 + J I 2 m 2 , α 3 = m 3 + J E 2 m 4 + J ˜ 2 m 2 , α 4 = m 3 + J I 2 m 4 . ( 12 ) The mean field equations written below are valid both for all-to-all and for random connections between sub-networks , with the appropriate choice of αi . Our results for networks with finite N are based on large scale numerical simulations . In each simulation the connections were chosen randomly as described in the text , and an asynchronous update schedule was generated by a Poisson process . Parameter values are specified in the legend of Fig 2 in the main text . Averaged population activities were calculated online . The projection along the approximate attractor was defined at each time point as X ( t ) ≡ v 0 T · m ( t ) - m 0 , ( 17 ) where m ( t ) is the measured 4 dimensional averaged population activity , m0 is the vector of mean population activities at the symmetric fixed point , and v0 is the left eigenvector of the linearized dynamics with an eigenvalue close to zero . We chose the following normalization for the corresponding right eigenvector ( see Eq 14 ) : 1 1 / J I - 1 - 1 / J I , ( 18 ) and the normalization of v0 was chosen such that the dot product of the left eigenvector and the right eigenvector equals unity . Measurements of G ( X , Δt ) ( Eq 6 ) were done in the following way: for each value of X we found all the time points for which |X ( ti ) − X| < δ , using a small δ ≃ 10−3 . Then , for each such ti we calculated [X ( ti + Δt ) − X ( ti ) ]2 , and averaged all these values to get G ( X , Δt ) . Subsequently , we averaged over multiple simulations with different quenched noise and update schedules . In the manuscript we present results for G ( 0 , Δt ) , but in similarity to F ( X , Δt ) /X , G ( X , Δt ) was fairly uniform along the approximate attractor . A similar calculation was performed to measure the drift F ( X , Δt ) . In Fig 5 ( C ) –5 ( F ) , results are based on ( 1−2 ) × 103 simulations with random initial conditions , each spanning a simulated time of about 10 seconds . In simulations of the finite N network , we estimated λ from measurements of F ( X , Δt ) near the symmetric fixed point , and tuned J ˜ to obtain λ−1 ≫ τ . In Fig 5 , λ−1 ≃ 2 seconds . To analytically evaluate G ( X , Δt ) ( Eq 6 ) over short time scales , we start by writing the change in the state of the k-th neuron in population i in a short time interval Δt as: σ i k ( t + Δ t ) - σ i k ( t ) = c i k ( t ) Θ i k ( t ) - σ i k ( t ) , ( 22 ) where Θ i k ( t ) is the outcome of an update if it occurs , and c i k ( t ) is a random variable equal to 1 if the i-th neuron was updated between t and t + Δt and to 0 otherwise . The updates occur each τ ms on average , so that c i k ( t ) t = c i k ( t ) 2 t = Δ t τ k , ( 23 ) whereas for i ≠ j and/or k ≠ l , c i k ( t ) c j l ( t ) t = Δ t 2 τ k τ l . ( 24 ) Now the mean squared displacement , G ( X , Δt ) , can be written as: X ( t + Δ t ) - X ( t ) 2 = 1 N 2 ∑ i , j = 1 4 ∑ k , l = 1 N v i 0 v j 0 σ i k ( t + Δ t ) - σ i k ( t ) σ j l ( t + Δ t ) - σ j l ( t ) , ( 25 ) where v i 0 is the i’th component of the left eigenvector of the Jacobian with eigenvalue close to zero . From Eqs 23 and 24 we see that for Δt ≪ τ the contribution of elements with i = j , k = l dominates the sum . To leading order in Δt we have X ( t + Δ t ) - X ( t ) 2 ≃ 1 N 2 ∑ i = 1 4 ∑ k = 1 N v i 0 2 × σ i k ( t + Δ t ) - σ i k ( t ) 2 . ( 26 ) Defining q i ( Δ t ) = 1 N ∑ k = 1 N σ i k ( t + Δ t ) σ i k ( t ) , ( 27 ) we obtain: G ( X , Δ t ) ≃ 2 Δ t N ∑ j = 1 4 ( v j 0 ) 2 - ∂ q j ( t ) ∂ t t → 0 . ( 28 ) To derive an expression for the diffusive dynamics over arbitrary time scales , we start by representing the stochastic linearized dynamics of a single balanced network , near the symmetric fixed point , as a two dimensional stochastic process: δ m ˙ = B 1 δ m + B 2 δ E + ξ , ( 29 ) where δm is the deviation of the mean activities from the fixed point and δE is the deviation of the input from the constant input E0 . Here B1 is a 2 × 2 matrix representing the response to perturbations in m , and B2 is a 2 × 2 matrix representing the response to perturbations in the feedforward input . Both are obtained analytically from a linearization of the mean field dynamics . Finally , ξ is a random process with vanishing mean , whose covariance functions Cξ ( Δt ) are stationary and are yet unspecified: C ξ , i j ( t - t ′ ) ≡ ξ i ( t ) ξ j ( t ′ ) . ( 30 ) Using Eq 29 , it is straightforward to relate Cξ ( t ) to the covariance of the activities ( while assuming constant feedforward input , δ E ¯ = 0 ) : C ξ ( t ) = - d 2 d t 2 C m ( t ) + d d t C m ( t ) B 1 T - B 1 C m ( t ) + B 1 C m ( t ) B 1 T . ( 31 ) Using the measurements of Cm from simulations , we can thus obtain Cξ numerically , using the above equation . In similarity to Cm , Cξ decays to zero over a time scale of order τ . Altogether , Eq 29 describes the stochastic dynamics of a single balanced network close to the balanced state , in response to small fluctuations δE in the feedforward inputs . In the two-subnetwork architecture , each subnetwork is coupled only to the mean activity of the other subnetwork , because of the all-to-all connectivity . More specifically , the mean activity of each subnetwork linearly modulates the external input to the excitatory population of the other subnetwork . Therefore , we can approximate the state of the 4-population network as a stochastic process with the following dynamics: δ m ˙ = A δ m + ξ , ( 32 ) where δm is now a 4 dimensional vector , whose first ( last ) two entries represent the state of the first ( second ) subnetwork , and A is the Jacobian of the full 4 dimensional dynamics around the fixed point ( Eq 38 ) , related in a simple manner also to the matrices B1 , 2 defined above . The correlation matrix of the 4 dimensional noise vector ξ is given by C ˜ ξ = C ξ 0 0 C ξ , ( 33 ) where Cξ is the 2 × 2 noise correlation matrix Eq ( 30 ) evalulated for a single balanced network receiving fixed excitatory input , equal to the mean input to each subnetwork at the symmetric fixed point . Finally , we use this description of the dynamics to predict the statistics of diffusion along the line . We multiply Eq 32 from the left by v0 , the eigenvector with the eigenvalue close to zero , which we denote by λ ( note below that λ < 0 ) : X ˙ = λ X + v 0 T · ξ . ( 34 ) Here , X = v 0 T · δ m . Thus , we obtain using the Wiener—Khintchine theorem the time dependent correlation function of X , C X ( t ) = - 1 2 λ ∫ - ∞ ∞ e λ | t ′ | v 0 T C ˜ ξ ( t - t ′ ) v 0 d t ′ . ( 35 ) Finally , the diffusion over an arbitrary time interval Δt is given by: X ( t + Δ t ) - X ( t ) 2 = 2 C X ( 0 ) - C X ( Δ t ) . ( 36 ) Here we show that when ∂m1/∂m3 = −1 at the symmetric point ( m1 = m3 , m2 = m4 ) , the Jacobian matrix has a vanishing eigenvalue , leading to slow dynamics near the fixed point . We denote: f i , j ≡ ∂ H - u i / α i ∂ m j . ( 37 ) In terms of these quantities , the Jacobian matrix can be written as A ≡ f 1 , 1 - 1 f 1 , 2 0 f 1 , 4 f 2 , 1 / τ ( f 2 , 2 - 1 ) / τ 0 0 0 f 3 , 2 f 3 , 3 - 1 f 3 , 4 0 0 f 4 , 3 / τ ( f 4 , 4 - 1 ) / τ . ( 38 ) At the symmetric fixed point f1 , 1 = f3 , 3 , f2 , 2 = f4 , 4 , f1 , 2 = f3 , 4 , f1 , 4 = f3 , 2 , and f2 , 1 = f4 , 3 . The Jacobian’s eigenvalues at that point are then: λ ± ± = 1 2 ( f 1 , 1 - 1 ) + f 2 , 2 - 1 τ ± 1 2 ( f 1 , 1 - 1 ) - f 2 , 2 - 1 τ 2 + 4 τ f 1 , 2 f 2 , 1 ± 4 τ f 1 , 4 f 2 , 1 . ( 39 ) Next , we approximate the derivative ∂m1/∂m3 at the symmetric point . We use a first order Taylor expansion of the mean field equations to get: δ m 1 = f 1 , 1 δ m 1 + f 1 , 2 δ m 2 + f 1 , 4 δ m 4 , δ m 2 = f 2 , 1 δ m 1 + f 2 , 2 δ m 2 , δ m 3 = f 1 , 4 δ m 2 + f 1 , 1 δ m 3 + f 1 , 2 δ m 4 , δ m 4 = f 2 , 2 δ m 4 + f 2 , 1 δ m 3 . ( 40 ) Here , δmi are the small deviations from the symmetric fixed point . Using these equations we can write δm2 and δm4 as functions of δm1 and δm3: δ m 2 = f 2 , 1 1 - f 2 , 2 δ m 1 ; δ m 4 = f 2 , 1 1 - f 2 , 2 δ m 3 . ( 41 ) Plugging these expressions into the equation for δm1 yields an expression for δm1 as a function of δm3 . The derivative is: ∂ m 1 ∂ m 3 = f 1 , 4 f 2 , 1 f 1 , 1 f 2 , 2 - f 1 , 2 f 2 , 1 + 1 - ( f 1 , 1 + f 2 , 2 ) . ( 42 ) Note that ∂m1/∂m3 = ∂m2/∂m4 . Equating this derivative to −1 ( noting that in this case , ∂m1/∂m3 = ∂m3/∂m1 ) yields: f 1 , 4 = f 1 , 2 f 2 , 1 - 1 + f 1 , 1 + f 2 , 2 - f 1 , 1 f 2 , 2 f 2 , 1 . ( 43 ) By inserting f1 , 4 into Eq 38 we get λ - - = 0 .
|
This work studies the effects of chaotic dynamics , a prominent feature of the balanced state model , on storage of continuous parameters in working memory . We propose a simple model of a balanced network with mutual inhibition , and show that it possesses a continuum of steady states , a commonly proposed mechanism for maintenance of continuous parameter working memory in the brain . We use analytical methods , combined with large scale numerical simulations , to analyze the diffusive dynamics and correlation patterns generated by the chaotic nature of the system . We obtain new conclusions and predictions on irregular activity resulting from the combination of continuous attractor dynamics with the balanced state . These include a prediction of measurable , slowly decaying spike correlations , and a quantification of how persistence depends on the neural population size .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[
"cognitive",
"science",
"medicine",
"and",
"health",
"sciences",
"neural",
"networks",
"nervous",
"system",
"electrophysiology",
"neuroscience",
"learning",
"and",
"memory",
"mathematics",
"algebra",
"cognition",
"computational",
"neuroscience",
"memory",
"neuronal",
"tuning",
"computer",
"and",
"information",
"sciences",
"animal",
"cells",
"chemistry",
"physics",
"mass",
"diffusivity",
"cellular",
"neuroscience",
"cell",
"biology",
"linear",
"algebra",
"anatomy",
"synapses",
"single",
"neuron",
"function",
"neurons",
"physiology",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"physical",
"sciences",
"computational",
"biology",
"chemical",
"physics",
"eigenvalues",
"neurophysiology"
] |
2017
|
Slow diffusive dynamics in a chaotic balanced neural network
|
Atrial fibrillation ( AF ) is the most common cardiac arrhythmia at the clinic . Recent GWAS identified several variants associated with AF , but they account for <10% of heritability . Gene-gene interaction is assumed to account for a significant portion of missing heritability . Among GWAS loci for AF , only three were replicated in the Chinese Han population , including SNP rs2106261 ( G/A substitution ) in ZFHX3 , rs2200733 ( C/T substitution ) near PITX2c , and rs3807989 ( A/G substitution ) in CAV1 . Thus , we analyzed the interaction among these three AF loci . We demonstrated significant interaction between rs2106261 and rs2200733 in three independent populations and combined population with 2 , 020 cases/5 , 315 controls . Compared to non-risk genotype GGCC , two-locus risk genotype AATT showed the highest odds ratio in three independent populations and the combined population ( OR=5 . 36 ( 95% CI 3 . 87-7 . 43 ) , P=8 . 00×10-24 ) . The OR of 5 . 36 for AATT was significantly higher than the combined OR of 3 . 31 for both GGTT and AACC , suggesting a synergistic interaction between rs2106261 and rs2200733 . Relative excess risk due to interaction ( RERI ) analysis also revealed significant interaction between rs2106261 and rs2200733 when exposed two copies of risk alleles ( RERI=2 . 87 , P<1 . 00×10-4 ) or exposed to one additional copy of risk allele ( RERI=1 . 29 , P<1 . 00×10-4 ) . The INTERSNP program identified significant genotypic interaction between rs2106261 and rs2200733 under an additive by additive model ( OR=0 . 85 , 95% CI: 0 . 74-0 . 97 , P=0 . 02 ) . Mechanistically , PITX2c negatively regulates expression of miR-1 , which negatively regulates expression of ZFHX3 , resulting in a positive regulation of ZFHX3 by PITX2c; ZFHX3 positively regulates expression of PITX2C , resulting in a cyclic loop of cross-regulation between ZFHX3 and PITX2c . Both ZFHX3 and PITX2c regulate expression of NPPA , TBX5 and NKX2 . 5 . These results suggest that cyclic cross-regulation of gene expression is a molecular basis for gene-gene interactions involved in genetics of complex disease traits .
Genome-wide association studies ( GWAS ) have been highly successful in identifying common genomic variants that are associated with complex human diseases or traits . However , these common variants have small effects , and in aggregate explain only a small fraction of heritability for most diseases or traits . The major portion of heritability remains missing , and this represents a major dilemma in complex trait genetics referred to as “missing heritability” . Gene-gene interaction has been proposed to be a contributor to the problem of missing heritability . Gene-gene interaction has been long known to have an impact on an organism’s phenotype , for example , the color of a flower in plants and the color of a fly’s eye . However , it has been challenging to detect gene-gene interaction in human GWAS . Moreover , no gene interaction was functionally validated . Considering the potentially large number of gene-gene interaction , identification of true and casual interaction has been proven to be a daunting task . However , without doubt , studies of gene-gene interaction will contribute to the understanding of inheritance , particularly inheritance of important diseases and traits , and provide insights into the biological pathways and molecular mechanisms of disease pathogenesis . Atrial fibrillation ( AF ) is the most common cardiac arrhythmia seen at the clinical setting and accounts for approximately one-third of hospitalizations for cardiac rhythm disturbances [1] . The prevalence of AF is 0 . 4%-1 . 0% in the general population , and increases with age , reaching 8% in people over 80 [1] . A similar prevalence rate of 0 . 77% was found for AF in the Chinese population [2] . AF accounts for 15% of all strokes , worsens heart failure , and independently increases the risk of stroke 5-fold and risk of cardiac death up to 1 . 9-fold [3] . Genetic factors play an important role in the pathogenesis of AF . The heritability of polygenic liability to AF has been estimated to be 0 . 62 [4] . To date , several major GWAS have been reported for common complex AF and identified variants in ten chromosomal loci that were associated with AF . The first GWAS for AF identified significant association between SNP rs2200733 near the PITX2c gene encoding paired-like homeodomain 2 transcript c on chromosome 4q25 and AF in several populations of European ancestry as well as one Hong Kong population [5] . Our group later reported that SNP rs2200733 confers a significant risk in the mainland Chinese Han population , too [6] . Then , two independent GWAS identified significant association between AF and SNPs rs2106261 [7] and rs7193343 [8] , both of which are located in the ZFHX3 gene encoding zinc finger homeobox 3 on chromosome 16q22 . We have found that rs2106261 , but not rs7193343 , showed significant association with AF in the Chinese Han population [9] . Later , a common variant in KCNN3 ( encoding potassium intermediate/small conductance calcium-activated channel , subfamily N , member 3 ) , rs13376333 , was found to be associated with lone AF [10] . However , we found that rs13376333 did not show significant association with AF in the Chinese Han population [9] . Ellinor et al [11] identified six susceptibility loci for AF through meta-GWAS analysis . We have shown that only one SNP , rs3807989 at the CAV1 locus ( encoding caveolin 1 ) among the six loci , were associated with AF in the Chinese Han population [12] . In this study , we studied the gene-gene interaction for three AF loci replicated in the Chinese population , i . e . SNP rs2106261 in ZFHX3 , rs2200733 near PITX2c , and rs3807989 in CAV1 . We provide strong genetic evidence that SNP rs2200733 near PITX2c and rs2106261 in ZFHX3 interact with each other , resulting in a synergistic effect that increases the odds ratios ( ORs ) to risk of AF . Most importantly , we also carried out a series of cellular and molecular studies to identify the molecular mechanisms underlying the gene–gene interaction . We found that PITX2c and ZFHX3 cross-regulate each other’s expression as well as expression of downstream genes such as NPPA ( encoding atrial natriuretic factor or ANF ) , providing a novel molecular basis for their interaction at the molecular genetic level .
We previously reported that among the first three genetic loci for AF identified by GWAS in European ancestry populations , only rs2200733 at the PITX2c locus on 4q25 and rs2106261 in ZFHX3 on 16q22 , but not rs13376333 in KCNN3 , were replicated in the Chinese Han populations [6 , 9] . We , therefore , carried out a deeper study to determine whether there is gene-gene interaction between rs2200733 and rs2106261 . We utilized a case control design which involves three independent populations . The initial association study was carried out with 569 AF patients and 1 , 996 non-AF control samples ( referred to as the Discovery population ) . The positive findings in the Discovery population were validated in two independent replication populations . The first replication population consisted of 641 AF cases and 1 , 692 controls ( referred to as Replication I population ) . The second replication population consisted of 810 cases and 1 , 627 controls ( referred to as Replication II population ) . The clinical characteristics of the three study populations are shown in S1 Table . We first examined the association of AF with each GWAS SNP individually . There was no deviation from the Hardy-Weinberg equilibrium for the two SNPs , rs2200733 and rs2106261 in the control groups of the three populations ( S2 Table ) . As shown in S3 Table , SNP rs2200733 showed highly significant association with AF in the Discovery population with a P value of 1 . 58×10-14 ( OR = 1 . 70 ( 95% CI 1 . 48–1 . 94 ) ) with the T allele as the risk allele . After adjusting for covariates of age and gender with multivariable logistical regression analysis , rs2200733 remained significantly associated with AF ( Padj = 5 . 50×10-13 , OR = 1 . 32 ( 95% CI 1 . 22–1 . 42 ) ) . SNP rs2200733 remained significant association with AF in Replication I population ( Pobs = 1 . 27×10-11 , OR = 1 . 57 ( 95% CI 1 . 38–1 . 79 ) ; Padj = 3 . 17×10-10 , OR = 1 . 27 ( 95% CI 1 . 18–1 . 37 ) ) and Replication II population ( Pobs = 2 . 20×10-10 , OR = 1 . 48 ( 95% CI 1 . 31–1 . 67 ) ; Padj = 7 . 84×10-10 , OR = 1 . 22 ( 95% CI 1 . 14–1 . 29 ) ) . In the combined population , the association between SNP rs2200733 and AF was highly significant ( Pobs = 2 . 83×10-33 , OR = 1 . 57 ( 95% CI 1 . 46–1 . 69 ) ; Padj = 4 . 54×10-29 , OR = 1 . 26 ( 95% CI 1 . 21–1 . 31 ) ) . In addition to analysis of allelic association , we also analyzed genotypic association assuming three different genetic models . As shown in S4 Table , highly significant genotypic associations were detected between SNP rs2200733 and AF in the Discovery population , Replication I population and Replication II population in an additive , dominant , or recessive model . In the combined cohort of the three populations , the genotypic associations between SNP rs2200733 and AF were also highly significant with Padj of 4 . 54×10-29 ( OR = 1 . 61 ( 95% CI 1 . 48–1 . 75 ) ) , 1 . 82×10-23 ( OR = 1 . 36 ( 95% CI 1 . 28–1 . 45 ) ) and 4 . 69×10-16 ( OR = 1 . 37 ( 95% CI 1 . 27–1 . 48 ) ) under an additive , recessive and dominant model , respectively ( S4 Table ) . Similarly , SNP rs2106261 on 16q22 also showed significant allelic and genotypic association with AF in the Discovery population , Replication I population and Replication II population ( S3 and S4 Tables , respectively ) . In the combined population , the allelic association between SNP rs2106261 and AF was highly significant ( Pobs = 6 . 26×10-12 , OR = 1 . 30 ( 95% CI 1 . 21–1 . 40 ) ; Padj = 3 . 03×10-12 , OR = 1 . 16 ( 95% CI 1 . 11–1 . 21 ) ) ( S3 Table ) with the A allele as the risk allele . Genotypic associations were also identified between rs2106261 and AF ( Padj of 3 . 11×10-12 ( OR = 1 . 33 ( 95% CI 1 . 23–1 . 45 ) ) , 1 . 02×10-12 ( OR = 1 . 35 ( 95% CI 1 . 24–1 . 46 ) ) and 1 . 42×10-6 ( OR = 1 . 15 ( 95% CI 1 . 08–1 . 21 ) ) under an additive , recessive and dominant model , respectively ( S4 Table ) . To study the interaction between rs2106261 ( G to A substitution , risk allele = A ) and rs2200733 ( C to T substitution , risk allele = T ) , we first defined the frequencies of nine possible two-locus genotypes ( 32 genotypes: GGCC , GGCT , GGTT , AGCC , AGCT , AGTT , AACC , AACT , AATT ) in cases and controls of the three independent study populations . Then , we used the wild type non-risk GGCC genotype ( non-risk homozygote for both loci ) as baseline or reference , and estimated the OR for each of the eight other genotypes . As shown in Table 1 and Fig 1 and S1 Fig , compared with the GGCC non-risk reference genotype , the double risk homozygous genotype AATT showed a dramatically increased risk for AF with the highest ORs of 4 . 81 ( 95% CI 2 . 88–8 . 04 ) ( Pobs = 3 . 83×10-10 ) and 6 . 64 ( 95% CI 3 . 64–12 . 11 ) ( Padj = 6 . 38×10-10 ) before and after adjustment for covariates of age and gender , respectively , in the Discovery population . This interesting finding was replicated in two independent AF populations with ORs of 4 . 04 ( 95% CI 2 . 23–7 . 32 ) ( Padj = 4 . 34×10-6 , Replication I ) and 5 . 70 ( 95% CI 3 . 34–9 . 71 ) ( Padj = 1 . 58×10-10 , Replication II ) . In the combined cohort of the three populations , AATT increased risk of AF with an OR of 5 . 36 ( 95% CI 3 . 87–7 . 43 ) ( Padj = 8 . 00×10-24 ) ( Table 1 ) . The ORs among the different genotypes were compared for statistical significance using the Breslow-Day test ( S5 Table ) . In all three independent populations as well as the combined population , ORs for genotype AATT ( double risk homozygotes for both rs2106261 and rs2200733 ) were significantly higher than the ORs for each single-risk homozygotes ( GGTT , homozygous risk genotype for rs2106261; AACC , homozygous risk genotype for rs2200733 ) ( Table 1 and S5 Table ) . Moreover , the OR of 6 . 64 for double-risk homozygote AATT was higher than the combined ORs for the two single-risk homozygotes GGTT and AACC together ( 2 . 14+1 . 25 = 3 . 39 ) in the Discovery population ( Table 1 , Fig 1 and S1 Fig ) . Similar findings were observed in the Replication I population ( 4 . 34 vs . 3 . 25 ( 2 . 16+1 . 09 ) ) , the replication II population ( 5 . 70 vs . 3 . 90 ( 2 . 19+1 . 71 ) ) , or the combined cohort ( 5 . 36 vs . 3 . 31 ( 2 . 14+1 . 17 ) ) ( Table 1 , Fig 1 and S1 Fig ) . These data provide genetic evidence for interaction between ZFHX3 variant rs2106261 and PITX2c variant rs2200733 , which generates a synergistic effect that markedly increases the risk of AF . Two other genotypes , GGTT and AGTT , significantly increased risk of AF compared to reference non-risk genotype GGCC , consistently in all three populations ( Padj<0 . 006 after Bonferroni correction ) ( Table 1 , Fig 1 and S1 Fig ) . To substantiate the novel finding of the genetic interaction between rs2106261 and rs2200733 as identified by the analyses above , we carried out functional studies to identify the underlying molecular mechanism of the interaction . The PITX2c gene near rs2200733 has been demonstrated to be an AF gene using mouse models and shown to regulate several genes in the atria [13–15] . Because PITX2c encodes a transcriptional factor , we hypothesized that PITX2c would regulate the expression of ZFHX3 , generating a synergistic effect for gene-gene interaction . To test this hypothesis , we transfected HCT116 cells with a PITX2c-specific siRNA and a negative control siRNA ( NC control ) and then used real-time RT-PCR analysis to measure the expression level of ZFHX3 . As shown in Fig 2 , knockdown of PITX2c expression by siRNA significantly decreased the expression level of ZFHX3 ( P = 4 . 00×10-3 ) ( Fig 2A and 2C ) . In a parallel study , overexpression of a FLAG-tagged PITX2c protein by transfection of a p3×FLAG-PITX2c expression plasmid significantly increased the expression level of ZFHX3 ( P = 0 . 01 ) ( Fig 2B and 2C ) . These studies indicate that PITX2c positively regulates expression of ZFHX3 . To explore the molecular mechanism by which PITX2c regulates ZFHX3 , we searched for a potential PITX2c binding site at the ZFHX3 promoter and regulatory region , but failed to find one . Because PITX2c was shown to negatively regulate the expression of miR-1 ( microRNA 1–1 ) [15] , we hypothesize that PITX2c may regulate expression of ZFHX3 through miR-1 . To test this hypothesis , we transfected HCT116 cells with miR-1 mimics and control microRNA mimics and measured the expression level of ZFHX3 . Both real-time RT-PCR analysis and Western blot analysis showed that miR-1 mimics significantly decreased expression of ZFHX3 at both mRNA ( P = 4 . 00×10-4 ) and protein levels ( P = 6 . 84×10-5 ) , although the effect on the protein level was more robust ( Fig 3A–3C ) . This interesting finding of down-regulation of ZFHX3 by miR-1 was confirmed in another cell line , SW620 at the ZFHX3 mRNA ( P = 0 . 01 ) and protein levels ( P = 4 . 89×10-4 ) ( Fig 3D–3F ) . These results suggest that miR-1 negatively regulates expression of ZFHX3 . To explore the molecular mechanism by which miR-1 regulates ZFHX3 , we performed bioinformatic analysis by searching two databases , DIANA TOOLs and microRNA . org-Target and Expression , and found that the 3’-untranslated region ( 3’-UTR ) of ZFHX3 contained two potential targeting sites for miR-1 ( Fig 3G ) . We cloned each region containing a miR-1 binding site downstream of the firefly luciferase coding region in the pMIR-REPORT luciferase vector , resulting in luciferase reporters pMIR-ZFHX3-3’-UTR-1 ( cloned genomic region: chr16: 72819500 to 72820662 ) and pMIR-ZFHX3–3’-UTR-2 ( cloned genomic region: chr16: 72818241 to 72819390 ) , respectively ( Fig 3H ) . Each reporter was co-transfected with miR-1 mimics ( 100 nM ) into HCT116 cells and luciferase assays were carried out . A schematic diagram shows luciferase reporters containing the potential miR-1 binding site or the related mutated site ( Fig 3H ) . As shown in Fig 3I , miR-1 mimics significantly reduced luciferase activities from pMIR-ZFHX3–3’-UTR-2 , but not that from pMIR-ZFHX3–3’-UTR-1 . Mutation of the miR-1 binding site in pMIR-ZFHX3–3’-UTR-2 from CATTCCA to TGCGAAC abolished the miR-1-mediated reduction of the reporter luciferase activity ( Fig 3I ) . These data suggest that miR-1 negatively regulates expression of ZFHX3 by targeting to the second binding site at the 3’-UTR of ZFHX3 . The AF SNP rs2106261 identified by GWAS is located within the ZFHX3 gene , therefore , we consider ZFHX3 as a strong candidate gene for AF at the chromosome 16q22 locus . As our genetic studies indicate a gene-gene interaction between PITX2c and ZFHX3 , we hypothesized that ZFHX3 may regulate expression of PITX2c . Interestingly , knockdown of ZFHX3 expression by a specific siRNA significantly decreased expression of PITX2c about 2-fold ( P = 5 . 00×10-3 ) ( Fig 4A and 4C ) . Conversely , overexpression of ZFHX3 significantly increased expression of PITX2c by 2 . 96-fold ( P = 2 . 00×10-3 ) ( Fig 4B and 4D ) . Knockdown of ZFHX3 expression by siRNA reduced the transactivation activity from a reporter with a 1 . 5 kb DNA fragment upstream of the PITX2c transcriptional start site fused to the luciferase gene ( PITX2c-PGL3 ) by 1 . 97-fold ( P = 5 . 00×10-3 ) ( Fig 4E ) . Several earlier studies showed that PITX2c regulates the expression of the NPPA gene encoding ANF ( a cardiac protein hormone ) , but conflicting results on either positive regulation or negative regulation were obtained in different studies [13 , 15 , 16] . We tested the regulation of NPPA by PITX2c in HCT116 cells . As showed in Fig 5A , knockdown of PITX2c expression using siRNA significantly reduced expression of NPPA by 60% ( P = 3 . 20×10-4 ) . Overexpression of PITX2c by transfection of HCT116 cells with p3×FLAG-PITX2c significantly increased NPPA expression by 2 . 42 fold ( P = 0 . 01 ) ( Fig 5B ) . Interestingly , we found that ZFHX3 also regulated NPPA expression . As shown in Fig 5A , knockdown of ZFHX3 expression by siRNA significantly decreased expression of NPPA ( P = 4 . 00×10-3 ) . Overexpression of ZFHX3 up-regulated NPPA expression 2 . 36 fold ( P = 0 . 03 ) ( Fig 5B ) . Overexpression of both PITX2c and ZFHX3 dramatically increased expression of NPPA ( P = 0 . 031 ) ( Fig 5B ) . Knockdown of both PITX2c and ZFHX3 also reduced expression of NPPA ( P = 1 . 00×10-3 ) ( Fig 5A ) . It was reported that PITX2c could also regulate expression of other downstream genes including NKX2 . 5 ( encoding NK2 transcription factor related , locus 5 ) , TBX5 ( encoding T-box 5 ) , KCNQ1 ( encoding potassium voltage-gated channel , KQT-like subfamily , member 1 ) , and SCN1B ( encoding sodium channel , voltage-gated , type I , beta subunit ) [13 , 15 , 17 , 18] . As shown in Fig 6 , knockdown of the PITX2c expression by siRNA significantly increased expression of NKX2 . 5 by 3 . 10-fold , TBX5 by 2 . 32-fold , KCNQ1 by 1 . 55-fold , and SCN1B by 1 . 27-fold . Interestingly , knockdown of the ZFHX3 gene by siRNA also significantly increased expression of NKX2 . 5 by 3 . 45-fold and TBX5 by 3 . 23-fold , but decreased expression of SCN1B by 1 . 52-fold and did not affect expression of KCNQ1 ( Fig 6 ) . Co-transfection of both PITX2c siRNA and ZFHX3 siRNA also significantly reduced NKX2 . 5 by 2 . 91-fold , TBX5 by 2 . 42-fold , but did not affect expression of KCNQ1 or SCN1B ( Fig 6 ) . GWAS in European ancestry populations have identified ten genetic loci for AF [5 , 7 , 8 , 10 , 11] . We analyzed these loci in the Chinese Han population for their association with AF . We found that in addition to the ZFHX3 locus and the PITX2c locus reported previously [6 , 9] , one other locus , rs3807989 in CAV1 encoding caveolin-1 , also showed significant association with AF , whereas no significant association was identified for other loci [12] . Therefore , we also analyzed gene-gene interactions between rs2106261 and rs3807989 and between rs2200733 and rs3807989 . The classical gene-gene analysis by comparing ORs for the nine two-locus genotypes did not reveal any significant synergistic effect between rs2106261 and rs3807989 ( Table 2 ) . The OR for the double risk homozygotes for both rs2106261 and rs3807989 ( AAGG ) was 1 . 25 , which is smaller than the product of the ORs ( 1 . 18+1 . 15 ) for each single-risk homozygotes ( AAAA , homozygous risk genotype for rs2106261; GGGG , homozygous risk genotype for rs3807989 ) ( Table 2 ) . These results suggest that there is no interaction between the ZFHX3 locus and the CAV1 locus for AF . Similarly , the OR for the double risk homozygotes for both rs2200733 and rs3807989 ( TTGG ) was 1 . 08 , which is smaller than the product of the ORs ( 1 . 00+0 . 77 ) for each single-risk homozygotes ( TTAA , homozygous risk genotype for rs2200733; CCGG , homozygous risk genotype for rs3807989 ) ( Table 2 ) . These results suggest that there is no interaction between the PITX2c locus and the CAV1 locus for AF . Real-time RT-PCR analysis showed that knockdown of either ZFHX3 or PITX2c increased the expression level of CAV1 ( Fig 7 ) . Similar results were obtained with Western blot analysis ( Fig 7 ) . On the contrary , knockdown of CAV3 did not significantly affect the expression of ZFHX3 or PITX2c ( Fig 8 ) . Together , these data suggest that there is no cyclic cross-regulation between ZFHX3 and CAV1 or between PITX2c and CAV1 . Many gene-gene programs have been developed in recent years , therefore , we also analyzed interaction among SNPs rs2200733/PITX2c , rs2106261/ZFHX3 and rs3807989/CAV1 using RERI and INTERSNP programs . RERI ( relative excess risk due to interaction ) analysis was developed to quantify the extent of synergistic effect by adopting a fundamental measure of additive interaction and relative excess risk due to interaction ( RERI ) [19] . Here we used this strategy to investigate interaction between rs2106261 and rs2200733 in terms of risk alleles A and T in the combined population . The RERI analysis can distinguish the additive effect from the synergistic effect [19] . A significant RERI value higher or lower than 0 is considered to demonstrate a synergistic effect , whereas a non-significant RERI value indicates an additive effect [19] . The results are shown in Table 3 . First , we analyzed the synergistic effect when exposed to one copy of risk alleles at any one locus or both loci ( H1 ) . No significant synergistic effect was observed between rs2106261 and rs2200733 ( RERI = 0 . 22 ( 95% CI -0 . 20–0 . 54 ) , P = 0 . 13; RERI = 0 . 18 ( 95% CI -0 . 29–0 . 52 ) , Padj = 0 . 22 after adjustment of covariates of age and gender ) . Second , we assessed the synergistic effect when exposed to two copies of risk alleles at any one locus or both loci ( H2 ) . A significant synergistic effect was detected between rs2106261 and rs2200733 with a RERI value of 2 . 26 ( 95% CI 1 . 06–3 . 73 ) ( P<1 . 00×10-4; RERI = 2 . 87 ( 1 . 48–4 . 69 ) , Padj<1 . 00×10-4 after adjustment of covariates of age and gender ) . Third , we assessed the synergistic effect when exposed to two copies of risk alleles at one locus and one copy of risk alleles at the other locus ( H3 ) . A significant synergistic effect was detected between rs2106261 and rs2200733 with a RERI value of 0 . 99 ( 95% CI 0 . 29–1 . 79 ) ( P<1 . 00×10-4; RERI = 1 . 29 ( 95% CI 0 . 44–2 . 33 ) , Padj<1 . 00×10-4 after adjustment of covariates of age and gender ) . These results provided statistical genetic evidence for the interaction between rs2106261 and rs2200733 . The RERI analysis did not identify any significant interaction between ZFHX3 SNP rs2106261 and CAV1 variant rs3807989 ( H1: RERI = 0 . 51 ( 95% CI -0 . 54–1 . 12 ) , Padj = 0 . 19; H2: RERI = -0 . 52 ( 95% CI -5 . 64–1 . 41 ) , Padj = 0 . 37; H3: RERI = 0 . 29 ( 95% CI -0 . 61–1 . 06 ) , Padj = 0 . 36 ) ( Table 4 ) . The RERI analysis was also used to analyze the interaction between PITX2c variant rs2200733 and CAV1 variant rs3807989 ( H1: RERI = 0 . 80 ( 95% CI -0 . 02–1 . 27 ) , Padj = 0 . 11; H2: RERI = 1 . 37 ( 95% CI 0 . 24–2 . 71 ) , Padj = 0 . 05; H3: RERI = 0 . 42 ( 95% CI -0 . 33–1 . 12 ) , Padj = 0 . 08 ) . We also analyzed gene-gene interaction using the INTERSNP program[20 , 21] , which can analyze genotypic interactions under additive by additive , additive by dominant , dominant by additive and dominant by dominant terms . For rs2106261 and rs2200733 , nominal significant interaction was found additive × additive after adjusting for age and gender ( OR = 0 . 85 , 95% CI: 0 . 74–0 . 97 , Padj = 0 . 02 ) , although the global test on all interaction terms were not significant ( Table 5 ) . After simplifying the model by removing dominant effects without significant loss of goodness-of-fit of the model ( P = 0 . 11 ) , the additive interaction on a multiplicative OR scale was also significant ( OR = 0 . 85 , 95% CI: 0 . 76–0 . 96 , Padj = 0 . 01 ) ( Table 5 ) . A similar pattern was found for additive × additive interaction between rs2200733 and rs3807989 under models with dominant effects ( OR = 1 . 40 , 95% CI: 1 . 16–1 . 70 , Padj = 1 . 00×10-3 ) and after removing dominant effects ( OR = 1 . 25 , 95% CI: 1 . 06–1 . 48 , Padj = 7 . 00×10-3 ) . No significant genotypic interaction was found for rs2106261 and rs3807989 under any model ( Table 5 ) .
In this study , we show that gene-gene interaction plays an important role in generation of disease phenotype by identifying gene-gene interaction involved in the pathogenesis of a cardiac disorder , AF . We employed a multi-stage case control association design to compare the frequencies of all nine two-locus genotypes from GWAS SNPs rs2106261 in the ZFHX3 gene and rs2200733 close to the PITX2c gene . Our study involves a careful design with a Discovery population consisting of 569 cases and 1 , 996 controls , Replication I population with 641 cases and 1 , 692 controls , and Replication II population composed of 810 cases and 1 , 627 controls . The combined population has 2 , 020 cases and 5 , 315 controls , and is considered to represent a considerably large sample size in the modern population studies for AF . We consider this point as strength of this study . When SNP rs2106261 and rs2200733 were analyzed together , two-locus genotype AATT ( double risk homozygote ) showed the highest odds ratio ( OR ) of 6 . 64 ( 95% CI 3 . 64–12 . 11 ) ( P = 6 . 38×10-10 ) , 4 . 04 ( 95% CI 2 . 23–7 . 32 ) ( P = 4 . 34×10-6 ) , 5 . 70 ( 95% CI 3 . 34–9 . 71 ) ( P = 1 . 58×10-10 ) and 5 . 36 ( 95% CI 3 . 87–7 . 43 ) ( P = 8 . 00×10-24 ) in the Discovery , Replication I , II , and combined population , respectively , when compared to wild type non-risk genotype GGCC . The Breslow-Day test showed that the ORs for AATT were significantly higher than ORs for GGTT or AACC in all populations ( P = 5 . 26×10-5 vs . GGTT and 2 . 94×10-22 vs . AACC in the combined population ) and higher than the combined ORs for both GGTT and AACC ( 5 . 36 vs . 3 . 31 in the combined population ) . We also analyzed gene-gene interaction using the RERI analysis and identified synergistic effects between SNP rs2106261 and rs2200733 when exposed two copies of risk alleles at any one locus or both loci ( H2 ) ( P<1 . 00×10-4 ) or when exposed to two copies of risk alleles at one locus and one copy of risk alleles at the other locus ( H3 ) ( P<1 . 00×10-4 ) ( Table 3 ) . Analysis using the INTERSNP program revealed significant genotypic interaction between SNP rs2106261 and rs2200733 under an additive × additive model , but not under other models ( Table 5 ) . Overall , our studies establish that gene-gene interaction is involved in the pathogenesis of AF . Most importantly , our results suggest that gene-gene interaction accounts for heritability of human disease because it generates synergistic effects that markedly increase disease risk . The present study identifies the interaction between two common GWAS loci for AF . Ritchie et al [22] previously found that the risk alleles of common variants rs2200733 and rs10033464 at the 4q25 PITX2c AF locus could predict whether carriers of rare mutations in SCN5A ( encoding the cardiac sodium channel ) , NPPA , KCNA5 ( encoding potassium voltage-gated channel , shaker-related subfamily , member 5 ) , and NKX2 . 5 ( encoding transcriptional factor NK2 homeobox 5 ) developed AF , suggesting potential interaction between common variants and rare mutation in familial AF . Moreover , Lubitz et al [23] studied AF risk signals within nine GWAS loci and found that there are at least four distinct AF susceptibility signals at the 4q25 AF locus upstream of PITX2c that may increase the risk of AF by 5-fold together . Our cellular and molecular biological studies here on rs2200733/PITX2c and rs2106261/ZFHX3 identify a fundamental molecular mechanism underlying gene-gene interaction . SNP rs2200733 on 4q25 was the first genomic variant for AF identified by GWAS [5] and located 146 kb from the PITX2c gene encoding a paired-like homeodomain transcription factor 2 involved in the asymmetrical development of the heart and other organs [24–26] . Heterozygous knockout PITX2c mice developed atrial arrhythmias ( atrial flutter , atrial tachycardia ) upon programmed stimulation [13] . Kirchhof et al [14] showed that PITX2c is expressed in human and mouse left atria . Isolated hearts from heterozygous PITX2c knockout mice developed AF upon programmed stimulation and showed shortened action potential duration [14] . Chinchilla et al [15] showed that the expression level of PITX2c was decreased in AF patients and that atria-specific , but not ventricle-specific knockout of PITX2c , resulted in differences in action potential amplitude and increased expression of miR-1 . Therefore , all evidence to date strongly suggests that PITX2c should be the causative gene for AF at the 4q25 locus . SNP rs2106261 is located within the ZFHX3 gene . ZFHX3 encodes a transcription factor [27] which contains four homeodomains and seventeen zinc fingers [28] . The ZFHX3 transcription factor appears to regulate myogenic [29] and neuronal differentiation [30] . Although the function of ZFHX3 in cardiac tissue is unknown , it is expressed in mouse hearts [31] . Here we show that PITX2c and ZFHX3 positively cross-regulates each other . PITX2c negatively regulates expression of miR-1 , which negatively regulates expression of ZFHX3 by targeting a miR-1-binding site at the 3’-UTR , resulting in a positive regulation of ZFHX3 by PITX2c ( Fig 9 ) . Interestingly , ZFHX3 positively regulates expression of PITX2c . The net effect is a cyclic loop of cross-regulation between ZFHX3 and PITX2c ( Fig 9 ) . A cyclic loop of cross-regulation of two risk genes for a disease is expected to generate synergistic effects , which further increase disease risk , and therefore provides a novel molecular mechanism for gene-gene interaction . One important future direction is to determine whether this novel mechanism applies to other human disease and to plant and animal phenotypes in general . On the molecular level , how does the cyclic loop of cross-regulation between ZFHX3 and PITX2c generate interaction between the two genes and increase risk of AF ? The expression level of miR-1 was shown to be reduced in human AF patients , which was correlated with up-regulation of potassium channel Kir2 . 1 and increased potassium current IK1 responsible for AF maintenance . PITX2c negatively regulates expression of miR-1 , which increases IK1 , resulting in AF . Decreased miR-1 increases expression of ZFHX3 , which increases expression of PITX2c , further decreases expression of miR-1 and increases risk of AF ( Fig 9 ) . ZFHX3 increases expression of PITX2c , which decreases expression of miR-1 and increases risk of AF ( Fig 9 ) . PITX2c positively regulates expression of ZFHX3 , which further increases expression of PITX2c , and leads to down-regulation of miR-1 expression and increased risk of AF ( Fig 9 ) . In addition to miR-1 , PITX2c and ZFHX3 may regulate NPPA , TBX5 , NKX2 . 5 or other downstream target genes to increase risk of AF ( Fig 9 ) . These results provide novel insights into the roles of gene-gene interaction in the pathogenesis of AF . One other important insight from this study is that not all risk genes for AF interact each other . We have previously shown that genomic variants increase susceptibility of cardiovascular diseases in a population-specific manner . Although some variants increase disease risk in both European ancestry populations and Asian populations , but other variants show significant disease association only in Asian populations [9 , 11 , 32] . For AF , we found that among ten GWAS variants for AF identified in European ancestry populations , only three were associated with risk of AF in the Chinese population , including SNPs rs2106261 in the ZFHX3 gene , rs2200733 at the PITX2c locus , and rs3807989 in the CAV1 gene . Despite the robust gene-gene interaction identified for rs2106261/ZFHX3 gene and rs2200733/PITX2c , we did not identify any significant interaction between rs2106261/ZFHX3 and rs3807989/CAV1 with all three gene-gene interaction programs ( Tables 2 , 4 and 5 ) . For interaction between rs2200733/PITX2c and rs3807989/CAV1 , inconsistent results were obtained . Analysis for the OR for each multi-locus genotype and RERI analysis did not find gene-gene interaction between rs2200733 and rs3807989 , whereas the INTERSNP program found significant interaction under a model of additive by additive . Future studies are needed to reconcile the differences between different programs developed for studying gene-gene interaction . One limitation of the present study is that our statistical analysis was not adjusted for principal components to correct for possible stratification in Chinese samples due to a limited number of SNPs genotyped in the study populations . Genetic interaction may be especially susceptible to small degrees of population stratification , however , this may be unlikely given the replication of the finding in multiple populations . In summary , we have found that gene-gene interaction can generate synergistic effects that markedly increase disease risk , therefore , accounting for a portion of heritability of human disease . Our identification of the gene-gene interaction between SNPs rs2106261 in the ZFHX3 gene and rs2200733 at the PITX2c locus provide significant insights into the pathogenesis of AF . We further show that PITX2c and ZFHX3 positively regulate each other at the molecular level , generating a loop of cross-regulation between PITX2c and ZFHX3 . Our data provide an interesting molecular basis for some gene-gene interaction at the molecular genetic level .
The subjects involved in the present study include AF patients and non-AF controls selected from the GeneID database [6 , 9 , 32–39] . All study subjects are of Han ethnic origin based on self-description . The study was approved by the Ethics Committee of Huazhong University of Science and Technology and the Ethics Committees from local hospitals , and consistent with the guideline in the Declaration of Helsinki . Written informed consent was obtained from the participants . The diagnosis of AF was made by multiple experienced cardiologists and cardiac electrophysiologists using data from 12-lead surface electrocardiograms ( ECGs ) or Holter recordings . The ECG characteristics of AF include the absence of P waves , the presence of rapid oscillations or fibrillatory waves ( F waves ) , and irregular R-R intervals [40–42] . The controls are healthy individuals who do not have AF at the time of physical examinations or from medical records . Details of study subjects and GeneID and preparation of genomic DNA samples were described in S1 Text . Genotyping of SNPs was carried out using High-Resolution Melt ( HRM ) analysis as described previously by us [6 , 9 , 32–39] . HRM genotyping data were validated by direct sequencing analysis of 52 randomly selected study subjects . Primers for genotyping are listed in S6 Table . The HRM genotyping data matched the sequencing data . Details of bioinformatics prediction of miR-1 binding sites were described in S1 Text . Details of plasmids , siRNAs and microRNA mimics were described in S1 Text . HCT116 and SW620 cells were cultured and transfected with plasmid DNA , siRNAs , and microRNA mimics using Lipofectamine 2000 and the Opti-MEM I reduced serum medium as described [43 , 44] . Luciferase activities were measured using the Dual-Glo luciferase assay kit ( Gibco Life Technologies , Gaithersburg , MD , USA ) as described previously by us [43 , 45] . Each experiment was performed in triplicate and repeated at least three times . Details of cell culture and luciferase assays were described in S1 Text . The expression levels of PITX2c , ZFHX3 , NPPA , CAV1 , NKX2 . 5 , TBX5 , KCNQ1 , and SCN1B were measured using real-time RT-PCR analysis with SYBR green I mix as described by us previously [36 , 44] and described in detail in S1 Text . Primers for real-time RT- PCR analysis are listed in S6 Table . Western blot analysis was carried out as described by us previously [43 , 44] and described in detail in S1 Text . The genotyping data for all SNPs are included in S8–S13 Tables . The genotyping data from the control group for each SNP were first tested for the Hardy-Weinberg equilibrium using PLINK1 . 06 ( http://pngu . mgh . harvard . edu ) . If a P value was >0 . 01 , the genotyping data were considered to be in the Hardy-Weinberg equilibrium . Genotypic frequencies in controls were all in Hardy-Weinberg equilibrium ( P>0 . 01 ) . For case-control association analysis , we used Pearson’s 2×2 and 2×3 contingency table χ2 tests as implemented in PLINK1 . 06 ( http://pngu . mgh . harvard . edu ) to compute the P values for allelic and genotypic associations , respectively . The same PLINK1 . 06 program was used to estimate the odds ratio ( OR ) and 95% confidence interval ( CI ) for each association . In order to exclude confounding factors , multivariable logistic regression analysis was performed using SPSS 17 . 0 to adjust for gender and age . For analysis of gene-gene interaction , SNP rs2106261 in ZFHX3 or SNP rs2200733 at the PITX2c locus each has two alleles ( G vs . A for rs2106261; C vs . T for rs2200733 ) . The two SNPs together generate nine different genotypes . We defined the homozygous , non-risk ( or protective ) two-locus genotype GGCC as the reference group , and then estimated the OR of AF for each of the other eight two-locus genotypes GGCT , GGTT , AGCC , AGCT , AGTT , AACC , AACT , and AATT in relation to the reference genotype . The Pearson’s 2×2 contingency table χ2 test was used to compute the nominal P values , ORs , and 95% CIs for each genotypic association using PLINK1 . 06 . The Breslow-Day test was carried out to test whether the ORs between two different genotypes showed a statistically significant difference . Gene-gene interaction was also measured by a relative excess risk due to interaction ( RERI ) analysis [19] . The RERI analysis analyzes was suggested to be more meaningful for disease prevention and intervention in public health [46] , and advocated to be more biologically interpretable compared to that measured on the multiplicative scale [47] . A synergistic effect was defined as the extent of the combined effect of the exposures in excess of the sum of their individual effects [48] . We adopted a fundamental measure of RERI versus additive interaction to quantify the extent of synergistic effect in this study . The original form of RERI was defined as RERI = RR11-RR10-RR01+1 , where subscript 11 , 10 and 01 denote relative risks ( RR ) for doubly-exposed and individually-exposed to each risk factor when treating doubly-unexposed as a reference . When a RERI value equals to 0 , it indicates a perfect additive model . Any significant deviation from 0 indicates a synergistic ( + , positive values ) or antagonistic ( - , negative values ) interaction . In a case-control study , RERI can be calculated by substituting ORs for RRs , yielding RERI = OR11-OR10-OR01+1 . Although simply replacing RRs with ORs would induce an exaggeration problem for ORs [19 , 49] , especially for a high prevalent disease , it is shown that RERI in terms of ORs is a good approximation of RERI in terms of RRs in a disease such as AF with a prevalence rate of 0 . 4%~0 . 8% ) [49] . Under this circumstance , an OR is a good approximation of the RR . The statistical significance of RERI values in terms of ORs was addressed by the 95% confidence intervals based on the “MOVER” method , which utilizes the asymmetric intervals for ORs [19] . Since the SNPs are bi-allelic , it is of interest to explore if the interaction exists ( 1 ) when doubly-exposed to one copy of risk alleles ( i . e . doubly heterozygous genotype ) and ( 2 ) when doubly-exposed to two copies of risk alleles ( i . e . double homozygous risk genotypes ) . In both scenarios , the doubly-unexposed is referred to as the homozygous non-risk genotype ( e . g . GGCC ) . In addition , we tested the interaction when exposed one additional copy of risk alleles given being exposed to one copy of risk alleles . Note that the doubly-unexposed in this scenario is the doubly heterozygous genotype ( e . g . AGCT ) . The P values were estimated by 10 , 000 times of bootstrap sampling . The P value of 0 . 05 or less than and the 95% CI of RERI through zero was considered to show statistical significance . We also conducted a 4 degree of freedom test for genotypic interaction with logistic regression developed by Cordell and Clayton[20] , which was implemented in the software INTERSNP[21] as Logistic Regression test #6 . This model partitions the variance in AF risk into Additive and Dominant terms for each main effect , then into Additive by Additive , Additive by Dominant , Dominant by Additive and Dominant by Dominant terms . The test yielded ORs and 95% CI for each interaction term along with the global P values for the four terms . P values for individual terms were computed using Wald tests . We also used Logistic Regression test #5 in the INTERSNP program to test for additive interaction on a multiplicative ORs scale . In molecular studies with quantitative data , a standard Student’s t-test was used to compare the means between two groups of variables . A P value of 0 . 05 or less was considered to show statistical significance .
|
Gene-gene interaction is assumed to be critical to the pathogenesis of human disease , but its contribution to human disease phenotype needs definitive documentation . Moreover , the underlying molecular mechanism for gene-gene interaction is unknown . Here we use atrial fibrillation ( AF ) as a model to demonstrate that gene-gene interaction plays an important role in disease pathogenesis . Only three of the ten AF loci identified by GWAS in European ancestry populations , including PITX2c , ZFHX3 , and CAV1 , were replicated in the Chinese population and thus selected for gene-gene interaction studies . We show that the PITX2c locus interacts with the ZHFX3 locus to increase the risk of AF . Because gene-gene interaction can generate synergistic effect that markedly increases risk of AF , we conclude that gene-gene interaction accounts for a significant portion of heritability of AF . Mechanistically , PITX2c positively regulates ZHFX3 via miR-1 and ZHFX3 positively regulates PITX2c , which generates a loop of cross-regulation of the two genes . Our study suggests that cyclic cross-regulation of gene expression is a molecular basis for gene-gene interaction involved in disease phenotype .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2015
|
Molecular Basis of Gene-Gene Interaction: Cyclic Cross-Regulation of Gene Expression and Post-GWAS Gene-Gene Interaction Involved in Atrial Fibrillation
|
Whether , when , how , and why increased complexity evolves in biological populations is a longstanding open question . In this work we combine a recently developed method for evolving virtual organisms with an information-theoretic metric of morphological complexity in order to investigate how the complexity of morphologies , which are evolved for locomotion , varies across different environments . We first demonstrate that selection for locomotion results in the evolution of organisms with morphologies that increase in complexity over evolutionary time beyond what would be expected due to random chance . This provides evidence that the increase in complexity observed is a result of a driven rather than a passive trend . In subsequent experiments we demonstrate that morphologies having greater complexity evolve in complex environments , when compared to a simple environment when a cost of complexity is imposed . This suggests that in some niches , evolution may act to complexify the body plans of organisms while in other niches selection favors simpler body plans .
The “arrow of complexity” hypothesis [1] posits that the most complex products of open-ended evolutionary systems tend to increase in complexity over evolutionary time . Whether such a tendency exists is a long standing open question [2]–[6] . While it seems evident that more complex organisms exist today than at the advent of life , simple ( single-celled ) organisms continue to persist in large numbers , so it is clear that evolution does not guarantee complexity must increase . Moreover , loss of complexity has been observed in many species [7]–[9] . This begs the question: under what circumstances will complexity increase or decrease over evolutionary time ? It is likely that particular environmental conditions are more likely to select for increased complexity than others , especially if this complexity comes at a cost . As argued by proponents of embodied cognition , intelligent behavior emerges from the interplay between an organism's nervous system , morphology , and environment [10]–[14] . Therefore , if the ecological niche of a species remains constant and its body plan is evolutionarily constrained , then the neural system must adapt in order to succeed under this particular set of circumstances . This may be investigated experimentally through the use of evolving robots [15] , [16] which stand in for biological organisms . For instance , it has been demonstrated [11] , [17] that the complexity of an evolved neural system depends on the particular morphology it is controlling: in a given task environment certain morphologies can readily succeed with simple neural systems , while other morphologies require the discovery of more complex neural systems , or may prevent success altogether . Another corollary of embodied cognition is that different environments will impose different selection pressures on the nervous systems and/or morphologies of organisms evolving in them . This can be studied by observing how organisms evolve in different environments . For instance , Passy [18] demonstrated that the morphological complexity of benthic colonial diatoms ( measured as their fractal dimension ) is significantly correlated with the variability of the environmental niches in which they are found . However , the biological evidence for a correlation between environmental and morphological complexity is sparse . This is in part because it is difficult to isolate systems where this may be studied effectively and to develop metrics that quantify morphological and environmental complexity . Ideally , it would be desirable to perform controlled investigations in which environmental complexity is under experimental control . Given enough time and resources it may be possible to carry out these investigations directly on living organisms . However , by performing experiments in silico , it is possible to do so with much greater speed and more precise control over experimental conditions . Specifically , by evolving virtual organisms [19] in physically realistic simulations , it is possible to faithfully model the relevant interactions between organisms and their environments . Previously , the evolution of complexity has been investigated in silico using an alternative computational model [20] . In that work , populations of computer programs competed among themselves for the energy required to execute their instructions and gained energy by executing specific logic functions . With their system , Lenski et al . were able to demonstrate how complex functional features may evolve and how these features depend on the programs' environment . However , in that system the programs did not have bodies with which to physically interact with their environment . On the contrary , the evolutionary model employed here evolves embodied virtual organisms with evolutionarily determined body plans in physically realistic simulation environments . This provides a testbed for investigating how environment may influence the complexity of evolving physical morphologies . Using in silico evolution to act on both the morphologies and nervous systems of simulated organisms or robots was first demonstrated by Sims [19] , and has since been followed by a number of other studies ( e . g . [21]–[32] ) . These studies employed a variety of experimental techniques , including different genetic encodings , morphological systems ( such as branching structures or cellular aggregations ) , and evolutionary models . However , by constructing morphologies out of a relatively small number of geometric primitives , all of these studies were severely limited in the complexity of the morphologies which they could evolve , and therefore do not offer good test beds for investigating morphological complexity . Recently , we introduced a new method for evolving virtual organisms that is capable of producing a greater diversity of morphologies than previous systems [33] . By using it to evolve organisms with restricted nervous systems in a variety of environments it was possible to demonstrate how such a system could be used for investigating the relationship between environmental and morphological complexity . Here , the results of [33] are refined and extended to demonstrate that selection for locomotion tends to induce selection pressures favoring more complex morphologies than would be expected solely due to random chance , and is therefore a driven rather than passive trend [3] , [6] , [34] . In subsequent experiments we employ a multi-objective selection mechanism to select for simplicity in addition to behavioral competency . This selection mechanism filters out morphological complexity that arises due to biases in the underlying evolutionary model or because of genetic drift , and only allows for complexity that confers a selective advantage on the simulated organism . Moreover , this selection mechanism acts to impose a cost on complexity as is thought to occur in biological organisms [35] , [36] . Under this regime complex environments tend to induce selection for greater morphological complexity when compared to a simpler environment . This result supports the hypothesis that the environment plays an active role in determining morphological complexity . In this work organisms are evolved in a variety of simulated environments in order to better understand the role of the environment in shaping morphological complexity . While inspired by the above mentioned studies in which the morphologies and controllers of virtual organisms were also evolved [19] , [21]–[32] , the system presented here has several advantages which make it better suited for studying the evolution of morphological complexity . The first advantage relates to the task environments within which organisms evolve . The majority of the studies mentioned above were restricted to evolving for locomotion over flat terrain . While investigating this task has yielded interesting results , it suffers from its simplicity: simple morphologies composed of just a few cuboids or spheres are all that are needed to be successful . Even when more challenging task environments have been explored ( e . g . those investigated in [37] ) , they employed morphologies composed of a small collection of cuboids and therefore the maximum complexity of their evolved morphologies was severely limited . In the current work , a variety of task environments with interesting properties are investigated , and morphologies with greater geometric detail are used , so it is possible to study the evolution of morphological complexity . Another advantage of the current system is the way in which the genetic material that the evolutionary model acts on is encoded . As has been demonstrated in the past [25] , [26] , genetic encodings that simulate development to some extent offer demonstrable benefits over those that do not . This is because such encodings tend to produce regularities and symmetries in the phenotype; such patterns in nature are the inevitable result of biological development , which biases the kinds of phenotypes that biological evolution may act on [38] . For this reason , here we employ a particular form of genetic encoding that produces three-dimensional shapes with regular patterns ( see Methods for more details ) [39] . Each genome generated from this encoding generates a triangular mesh ( trimesh ) that forms the body plan of the virtual organism . Trimeshes allow evolution to craft morphologies with greater geometric detail compared to other systems in which evolution composes a small number of simple three-dimensional shapes together [19] , [21]–[32] ( see Figs . 1 and 2 for examples of morphologies evolved with the current system ) . Finally , populations of these genetic encodings are evolved with a commonly-used evolutionary model which has been demonstrated to be more evolvable than other evolutionary models [40] . The behavior of each virtual organism is simulated in a three-dimensional , physically-realistic virtual environment in order to assess its fitness . Because of the organisms' triangular mesh body plans and the complex environments in which they are evolved , evaluating the fitness of each organism requires considerable time . Moreover , many evolutionary trials were conducted in each of several environments to allow for meaningful statistical analysis . For these reasons all of the experiments were carried out on a 7 . 1 teraflop supercomputing cluster and required a total of over 100 CPU-years of distributed compute time .
It is clear that different environments in this parameterization present the evolutionary system with varying degrees of difficulty , but the question now becomes: how does environment influence the evolution of morphological complexity ? There are many approaches to quantify the complexity of an evolved morphology . Commonly , the variability of part types such as the number of cell types [41] has been used to measure the morphological complexity of biological organisms . But , the parts under consideration may vary in scale from organelles [42] to limbs [43] , and it is unclear what should be considered a part in the current work . More geometric measures describing how space-filling a morphology is could also be employed ( see Text S1 and Figure S2 ) . Alternatively , a morphology's surface area to volume ratio could be measured , or its concavity could be computed ( e . g . by taking the ratio of a morphology's volume to that of its convex hull ) . However each of these measures may be deceived by relatively simple body shapes , such as those that are very flat or contain large , simple concavities ( e . g . a ‘C’ shape ) . Instead , it is useful to think about the complexity of a body plan in information theoretic terms . One commonly used measure of complexity is Shannon's Entropy [44] , which measures the uncertainty of a random variable . Recent work [45] , [46] has demonstrated how Shannon Entropy can be applied to measure the complexity of a 3D object by considering the curvature of the object as a random variable . This means that in order to have higher complexity it is necessary to have more angles ( regions of non-zero curvature ) that can not simply be a repeating pattern , exactly what humans would think of as more complex shapes . And in fact , quantifying the complexity of 3D objects in this way has been shown to strongly correlate with human observers' notions of complexity [46] . In this work , the complexity of an organism's morphology is computed as the quantity which is the morphology's entropy of curvature or , in terminology which may be more familiar to biologists , it is the Shannon diversity [47] of the curvature on the organism's exterior ( see Methods for details ) . Does capture the complexity of evolved morphologies ? To answer this question , is calculated for all 9800 best-of-trial virtual organisms from all environments ( icy and control ) . Out of those 9800 , the five morphologies with the smallest value and the five morphologies with the largest value are selected . Images of these morphologies are shown in Fig . 2 . Looking at these two sets of morphologies , those with high values appear more complex than those with low values . In light of this observation and the previous work in this area it is concluded that successfully captures morphological complexity . Similarly , the concept of entropy may also be applied to characterizing the complexity of an environment . In the current formulation , environments are differentiated by variability in surface friction and terrain elevation . In the flat ground environment both the height of the terrain and the surface friction are uniform throughout , thus conveying zero entropy . On the other hand , in all of the icy environments there is variability in both of these properties . The surface friction is low on the ice blocks , but high on the ground between them . Likewise , the terrain is one height on the blocks and another in the intervening space . Therefore each of the icy environments has non-zero entropies of friction and elevation and so is considered to be more complex than flat ground . However , since each icy environment consists of a uniform series of ice blocks , the relative complexity between these environments is not considered . Armed with these measures , it is now possible to characterize how different environments influence the morphological complexity of evolving organisms . In order to understand the evolutionary pressures which lead to virtual organisms that are more or less morphologically complex , it is interesting to consider how morphological complexity varies over evolutionary time in different environments , and how these changes correspond to variations in fitness . Towards that goal , Fig . 4 depicts the mean morphological complexity and mean displacement of the current best individual over evolutionary time for each of several icy environments along with a corresponding set of control trials . Here it can be seen that morphological complexity tends to increase over time along with fitness . This means that in these environments selection for locomotion corresponds to an increase in complexity . However , it is unclear whether this increase of complexity is the result of a passive or a driven trend [3] , [6] , [34] . Passive trends may result from envelope expansion without any directional bias . For example , if there is a minimum level of complexity necessary for success , but no upper bound , then both the mean and the maximum complexity of the population will increase over time simply due to random variation ( what Stephen Jay Gould famously referred to as a “drunkard's walk” [9] ) . On the other hand , driven trends exhibit a consistent , directional bias . This corresponds to active selection for greater complexity . In this case not only will there be an increase in mean and maximum complexity , but the minimum level of complexity will increase over evolutionary time as well . When looking only at how morphological complexity varies over evolutionary time it is unclear what change in complexity is due to selection pressure from the environment and what change is due to biases towards increasing complexity within the evolutionary model itself and/or the general tendency of evolutionary systems to produce increasing complexity in the absence of selection [48] . In order to separate the influence of these factors it is useful to compare the evolving populations to a neutral shadow model [49] , [50] . For a generational evolutionary model , such as that employed here , a neutral shadow of a given experiment is equivalent to re-running the evolutionary model with the same parameters but with random selection . Fig . 5 shows how the morphological complexity of organisms evolved in flat ground ( black ) , as well as all icy environments ( blue ) , changes over evolutionary time compared to those evolved in 100 independent trials using random selection ( purple ) in which the only preference is for genomes that produce valid morphologies ( so that there exists a morphology for which complexity can be calculated; see Methods ) . It is known that the evolutionary system employed here [40] has an inherent bias to increase genotypic complexity over evolutionary time . The increasing purple curve in Fig . 5 indicates that there exists a bias to produce more complex morphologies over time as well . In fact , random selection alone produces morphologies that are more complex than those selected in any of the environments investigated . However , this comparison is not entirely fair . At any given generation , individuals in the random selection experiments will be the end product of many more reproduction ( mutation and crossover ) events than the corresponding individuals evolved for displacement , because under random selection it is unlikely that any individual will persist in the population for very long . Therefore , individuals in the random selection experiments will have had many more opportunities to increase the complexity of their genomes and hence the complexity of their morphologies . In order to correct for this discrepancy in the number of reproduction events , alternative shadow models are employed . Specifically , neutral shadow models of both the flat ground experiments and a representative icy environment ( spacing 0 . 025 , height 0 . 8 ) are created , which control for the number of reproduction events leading to the individuals in the current population . In each of the 100 independent trials evolving for locomotion in both of these environments , a record of every reproduction event is kept , and alternative shadow models are created for each trial such that they maintain the same rate of reproduction . These shadow models are detailed in Text S1 . All model alternatives have similar complexity curves ( see yellow , green , red and gray lines in Fig . 5 ) indicating that this shadow formulation is robust to whichever alternative is employed . Qualitatively they both show a much slower increase in morphological complexity ( especially early on in evolution ) compared to the experiments selecting for displacement , and so contrary to the naïve shadow model , both flat ground and icy environments select for increased morphological complexity beyond what would be expected in a neutral model . This implies that greater morphological complexity is being actively selected for in these environments: there is a driven trend towards increased morphological complexity . While the results reported so far support the hypothesis that there is a driven trend for increased morphological complexity in all environments , they do not differentiate between the complexities of organisms evolved based on which environment they are evolved in . Specifically , Fig . 5 depicts similar levels of complexity evolving in icy environments as compared to the flat , high friction environment under this regime . In fact , when the morphological complexities of organisms evolved in each of the 49 icy environments are compared with independent sets of trials conducted in the control environment ( see Figs . 4 and S1 ) they do not reflect a consistent relationship between environment and evolved morphological complexity . It is hypothesized that without a cost to becoming more complex the driven trend towards increased morphological complexity will dominate in all of the investigated environments . On the other hand , it is hypothesized that when complexity does come at a cost–as is thought to occur in biological organisms [35] , [36] –there will be greater pressure towards increased morphological complexity in more complex environments . In an an attempt to test this hypothesis , a second set of experiments is conducted which uses Pareto based multi-objective selection [51] , [52] to evolve organisms that can locomote in their given task environment and are as simple as possible , therefore imposing a cost on complexity . As was done for the single-objective experiments , 100 independent trials of a multi-objective model are run in each of the 49 icy environments along with a corresponding 49 independent sets of 100 trials apiece in the high friction , flat ground control environment . By selecting for both maximal displacement and minimal morphological complexity these experiments should evolve organisms that are no more complex than necessary to succeed in their task environment . If indeed more complex environments induce greater selection pressure favoring morphological complexity than simple environments when morphological complexity comes at a cost , then these differences should be observable under this regime . Comparing the results of these multi-objective experiments , we indeed see that more complex environments tend to select for organisms with greater morphological complexity when compared with organisms evolved in the simple , control environment . Figs . 6–8 show how the morphological complexities of organisms evolved in each of the icy environments under multi-objective selection differs from that of organisms evolved in a corresponding set of trials from the control environment . Since selecting a single representative individual from each trial is not as straightforward as in the single-objective case ( see Methods ) , several different techniques are employed to compare the results of these experiments . First , for the final Pareto front of each trial in a given environment , the mean morphological complexity is taken . These means ( 100 from each environment ) are compared to the mean morphological complexity in the final Pareto front of each trial from a corresponding set of trials from the control environment . This comparison is depicted in Fig . 6 . Fig . 7 presents the same comparison except that it considers the organism with median performance on each Pareto front: the organism with equal number of individuals on the front that displace less and more than it ( e . g . the most central point in Fig . 9 ) . Lastly , Fig . 8 shows the same comparison except that it considers the mean complexity of those organisms in the middle half of their respective Pareto fronts . That is , the top quarter of the most complex morphologies ( rightmost three points in Fig . 9 ) and the bottom quarter of most simple morphologies ( leftmost three points in Fig . 9 ) in each front are ignored , and the means are taken across the remaining organisms in each front ( which should reduce the influence of any outliers ) . While some differences can be observed across these plots , the general pattern is largely consistent ( and therefore not an artifact of the particular comparison employed ) : imposing a cost on complexity results in a multitude of icy environments where significantly more complex morphologies evolve compared to the control environment , and many of these differences are observed at the highest significance level ( ) . This corroborates the hypothesis that the more complex environments induce selection pressure for increased morphological complexity beyond what would evolve in a simpler environment when morphological complexity comes at a cost . In the lower right of Figs . 6–8 , where the environments become too difficult to succeed in ( because the organisms get trapped in the large gaps; see Fig . 3 ) , multi-objective selection actually results in the evolution of morphologies that are significantly less complex than those that evolve to locomote on flat ground . The reason for this is that when it is not possible to evolve for greater displacement , the majority of selection bears down on the simplicity objective , and therefore simpler morphologies evolve in these environments under multi-objective selection . This paper has presented a new method for evolving not only the neural systems but also the body plans of virtual organisms . This system differs from previous work by evolving populations of genetic encodings that produce complex morphologies instantiated in virtual environments as triangular meshes . This methodology opens up the possibility of investigating previously unexplored relationships between evolving organisms and their environments in a systematic manner . Here , this system was used to investigate how different environments induce differing selection pressures on morphological complexity . By evolving virtual organisms in a number of different task environments and analyzing how an information theoretic measure of morphological complexity varies over evolutionary time , it was demonstrated that not only do all investigated environments actively induce selection pressure favoring greater complexity above and beyond what would be expected in the absence of selection , but that more complex environments in fact induce selection for more complex morphologies then simple environments when a cost is imposed on morphological complexity . Since it is often thought that complexity does incur a cost in biological organisms [35] , [36] , the differences observed between environments in this regime may be more representative of the selection pressures present in biological systems . These results have illustrated how the environment may influence the complexity of evolving morphologies . Based on the results presented here it is possible that a similar evolutionary dynamic has been partially responsible for the “arrows of complexity” observed among biological organisms . As organisms have come to occupy more complex niches it is likely that these niches have actively selected for increased morphological complexity . Additionally , it should be possible to leverage this property for evolving more complex artifacts with evolutionary computation systems . However , it is not likely that increased environmental complexity will select for increased morphological complexity in every case where such complexity incurs a cost . While this work has demonstrated that such a relationship can exist , future work is needed to clarify this relationship across different environments , tasks , organisms , evolutionary models , and neural systems . A number of simplifications were made here which it may be desirable to relax in future work . By constraining the number of morphological components and using very simple neural architectures it was possible to largely bracket the question of neural complexity and focus on one particular aspect of morphological complexity . However , it may be desirable to investigate how many different forms of complexity evolve as a function of environment . For instance , in a recent study [53] we demonstrated that another measure of complexity: “mechanical complexity” , decreased in the same environments that selected for greater morphological complexity . This result lends support to the notion that various forms of complexity may be inversely correlated as discussed in [54] , and it also suggests that there is likely a trade-off between the various forms of complexity needed to succeed in a given environment , similar to the trade-off between morphological and neural complexity [11] , [17] . To investigate these ideas further it will be interesting to allow for more complex neural architectures , more complex sensorimotor systems , and a greater diversity of materials ( including ‘soft’ materials [55] ) to study how environments may influence the evolution of sensorial , nuerological , motoric , material , mechanical , and morphological complexity of these various systems . By extending the information theoretic ideas used here for quantifying morphological complexity it is hoped that a ‘common currency of bits’ may be used to investigate these complexity trade-offs in a systematic manner .
The morphologies evolved in this work are encoded by Compositional Pattern Producing Networks ( CPPNs ) [39] . CPPNs are a form of artificial neural network ( ANN ) [56] which differ from traditional ANNs in several ways . While each internal node in a traditional ANN typically has the same activation function ( such as a sigmoid or a step function ) , CPPN nodes can take on one of several activation functions from a predefined set . This function set often includes functions that are repetitive , such as or , as well as symmetric functions , such as , thus allowing for motifs seen in natural systems that arise as a result of development: symmetry , repetition , and repetition with variation . Additionally , CPPNs are often used as generative systems to produce other objects of interest , such as images [57] , 3D structures [58] , [59] , robot morphologies [31] , [32] or traditional ANNs themselves[60]–[64] . This is in contrast to the typical , direct application of ANNs as robot control architectures or classifiers . CPPNs act as functions of geometry . Geometric coordinates meaningful to the object being represented are fed as inputs to the CPPN . These input values are passed through the various connections of the CPPN from node to node . Each node aggregates its inputs by taking a weighted sum of the values output by each upstream node ( weights are specific to each connection ) and outputs the result of applying a particular activation function ( specific to that node ) to this weighted sum . By passing the inputs through subsequent nodes the activation functions are composed to produce novel outputs while maintaining features of the different functions ( hence the “compositional” aspect of CPPNs ) . Additionally , since these functions are chosen to have desirable properties present across a wide range of natural systems , as discussed above , CPPNs are capable of directly producing structures which in nature require a developmental process . For a more in-depth description of CPPNs , and further discussion of their ability to act as an abstraction of development , the reader is referred to [39] . In this study CPPNs are evolved via CPPN-NEAT [39] . CPPN-NEAT is an extension of the NeuroEvolution of Augmenting Topologies ( NEAT ) [40] method of neuro-evolution . NEAT is capable of evolving not only connection weights for existing network topologies , but also the network topologies themselves . Its operation is based on a few key ideas . First , the initial population is comprised of minimal networks ( those without any internal or hidden nodes ) , which may then gradually increase in complexity over evolutionary time through structural mutations which add new nodes and links to the network . When a new node or link is created in this manner it is assigned a unique historical marking . These historical markings are inherited during reproduction and allow meaningful crossovers to occur without the use of expensive graph matching procedures . Additionally , these markings are used to divide the population into “species” of similar network topologies . Speciation promotes genotypic diversity and , because competition is primarily intraspecies , novel structural innovations are given time to mature before directly competing with individuals in other species . CPPN-NEAT extends NEAT to evolve CPPNs . Effectively , this means that since nodes are no longer restricted to having sigmoid activation functions , each node contains an additional parameter which specifies its own activation function . When a new node is added to a network it is assigned a random function from a predefined set ( the signed cosine , Gaussian and sigmoid functions are used in the experiments reported here ) . Additionally , the compatibility distance metric used for speciation is modified to incorporate the number of different activation functions between two networks . In all other respects , CPPN-NEAT behaves the same as NEAT . NEAT and CPPN-NEAT have successfully evolved ANNs and CPPNs for a variety tasks [40] , [57] , [59] , [61] , [62] , [65] which makes CPPN-NEAT a good option for evolving the CPPNs used in this study . Moreover , CPPN-NEAT's ability to systematically increase network complexity over evolutionary time as needed should lend itself well to studying how morphologies increase in complexity when evolving inside different environments . For a more thorough description of these algorithms , including additional details of the mechanisms discussed above , please refer to [39] , [40] . While previously [31] , [32] evolving virtual organisms were constructed out of spherical components , the current study employs a voxel-based method to create morphological components out of triangular meshes ( trimeshes ) similar to what is done for the creation of 3D shapes in [59] . This process is illustrated in Fig . 10 , and is explained in detail below . First , A regular grid is placed over a region of 3D space which defines the presence of voxel locations . In the current work this region extends from to 1 ( inclusive ) in each dimension and grid lines are placed at intervals of 0 . 2 . This yields a total of 11 grid lines in each dimension for a total of voxels . A candidate CPPN is iteratively queried with the Cartesian coordinates at every voxel location except for the extrema in each direction . Querying a CPPN at a given location involves resetting all node values , and updating the CPPN for a fixed number of iterations ( in this case 10 ) before the output value is retrieved . This procedure is employed in order to extract consistent output signals from networks with recurrent connections , which may fall into cyclic or chaotic attractors . Previously [32] , it was found that allowing recurrent connections in morphology-generating CPPNs increased their evolvability . Voxel locations that exceed a predefined output threshold ( in this case ) are considered to contain matter , while those that fall below this threshold are considered to be devoid of matter . All voxels lying on one of the extrema ( ) are given output value so that no matter-containing voxel abuts against the boundary of the grid , and therefore guarantees that the final triangular meshes have completely enclosed surfaces . Once the CPPN has been queried for every voxel location , the Marching Cubes algorithm [66] is employed to create triangular meshes from the underlying voxel data . Specifically , an enclosed triangular mesh is created for each connected voxel component which defines the exterior surface of a single physical shape . These triangular meshes are then sent to the physics simulator where they define the exterior surface of a solid object and are imbued with mass ( see Fig . 1 for some examples ) . This is the first instance of physically simulating evolved , rigid body organisms composed of triangular meshes . Since the purpose of this study is to investigate how different task environments affect the shapes of evolved morphologies , a number of simplifications are used in order to concentrate on the physical shapes of the evolved organisms and control for other factors that may influence their performance . From the multiple enclosed trimesh components that could be produced when querying a single CPPN , only one of these ( the largest in terms of number of triangles ) is used in the resulting organism . This single component is copied and reflected across the . The resulting components ( the original and its mirror image ) are then spread apart by meters and a capsule of this length is placed between them such that it connects their two closest points . The two trimesh components each connect to this capsule by means of a hinge joint . These joints are constructed such that one rotates through the organism's coronal plane while the other rotates through its sagittal plane . Reflecting and copying a single component like this ensures that all organisms have the same mechanical degrees of freedom and ensures that the organisms are all bilaterally symmetric ( which should facilitate locomotion ) while at the same time it allows for a very large number of different morphologies due to the flexibility of the CPPN representation and trimesh model . The two mechanical degrees of freedom of each organism are actuated by means of coupled oscillators . Each of the two oscillators is parameterized by several parameters: amplitude , period , and phase shift . These six parameters ( three parameters apiece for each of the two joints ) are directly encoded in the genome of the evolving organisms as floating point values so that the genome is in actuality a CPPN plus a six dimensional floating point array . These floating point values are recombined and mutated in the same manner as CPPN link weights with mutation magnitudes scaled by the range of values for that parameter . Additionally , crossover on these vectors is possible in all instances of sexual reproduction since every individual contains a vector of the same dimensionality . Values for these parameters are constrained to predefined ranges: amplitude , radians ( so that the hinge rotates between and a radians ) , period simulation time steps ( or equivalently of the total evaluation time ) and phase shift periods . Each parameter has a mutation probability of 0 . 1 , which was chosen experimentally . Encoding the control parameters in this fashion is done to keep the controllers as simple as possible so that fitness is primarily dictated by the physical form of the organisms , while at the same time allowing for diverse enough behavior so that the organisms can succeed in the different task environments . The focus of this study is on how environment influences the evolution of morphological complexity in virtual organisms . Towards this aim a simple task is chosen which can be accomplished with more or less difficulty in a variety of environments . Specifically , as in previous work ( e . g . [24] , [25] , [32] , [67] , [68] ) , the task investigated here is to maximize directed displacement in a fixed amount of time , across a range of terrains . A candidate morphology ( triangular mesh ) and accompanying set of control parameters are sent to a physics simulator and allowed to act for a fixed number of simulation time steps . ( In this work simulations are conducted in the Open Dynamics Engine ( http://www . ode . org ) , a widely used open source , physically realistic simulation environment . ) Since trimeshes can be arbitrarily shaped and , unlike spheres , may simultaneously contact the environment at several points , it is necessary to use a much smaller step size than has been used in previous work in order to get physically realistic behavior . Specifically , a step size of 0 . 001 s is used in this work . Because of this smaller step size a proportionally larger number of time steps are needed to achieve the same effective simulation length . Here organisms are evaluated for time steps . In this section , the building blocks of computing the entropy of curvature are presented . The reader is referred to [45] , [46] , [72] , [73] for more in-depth discussions of their theoretical underpinnings . Given a random variable with a probability density function ( PDF ) , entropy is defined as ( 3 ) where is discretized such that where the are specific values of . Following [45] , [46] , the random variable on which is calculated is an approximation of the Gaussian curvature of the points on the surface . ( The Gaussian curvature of a point is the product of the principal curvatures and of that point [72] . ) Since the bodies here are built out of triangular meshes the points at which this curvature is non-zero are precisely the vertices of the triangular mesh . Specifically , for each vertex in a trimesh the angular deficit is calculated as ( 4 ) where is the internal angle at of each triangle of which is a vertex . This angular deficit is directly proportional to the Gaussian curvature of that point [45] , and so here we set for calculating the entropy of curvature . ( The relationship between angular deficit and Gaussian curvature can be derived through application of the Gauss-Bonnet theorem [72]; see [73] for more details . ) Following the calculation of for every vertex , a PDF is estimated by placing the values of into discrete bins of uniform width ( ) and counting the number of samples that fall into each bin . This results in a discrete set of probabilities , and Eqn . 3 can be used to arrive at an estimate of entropy that depends on the chosen , denoted here . ( see Text S1 for further details . ) The source code used to run the experiments reported in this paper is publicly available at https://github . com/jauerb/CPPN_Trimesh Additionally , the data files corresponding to the experiments reported in this paper have been made publicly available at http://dx . doi . org/10 . 6084/m9 . figshare . 858799
|
The evolution of complexity , a central issue of evolutionary theory since Darwin's time , remains a controversial topic . One particular question of interest is how the complexity of an organism's body plan ( morphology ) is influenced by the complexity of the environment in which it evolved . Ideally , it would be desirable to perform investigations on living organisms in which environmental complexity is under experimental control , but our ability to do so in a limited timespan and in a controlled manner is severely constrained . In lieu of such studies , here we employ computer simulations capable of evolving the body plans of virtual organisms to investigate this question in silico . By evolving virtual organisms for locomotion in a variety of environments , we are able to demonstrate that selecting for locomotion causes more complex morphologies to evolve than would be expected solely due to random chance . Moreover , if increased complexity incurs a cost ( as it is thought to do in biology ) , then more complex environments tend to lead to the evolution of more complex body plans than those that evolve in a simpler environment . This result supports the idea that the morphological complexity of organisms is influenced by the complexity of the environments in which they evolve .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Methods"
] |
[
"organismal",
"evolution",
"computer",
"science",
"biology",
"evolutionary",
"biology",
"computerized",
"simulations",
"evolutionary",
"processes"
] |
2014
|
Environmental Influence on the Evolution of Morphological Complexity in Machines
|
Significant progress has been made in defining the central signaling networks in many organisms , but collectively we know little about the downstream targets of these networks and the genes they regulate . To reconstruct the regulatory circuit of calcineurin signal transduction via MoCRZ1 , a Magnaporthe oryzae C2H2 transcription factor activated by calcineurin dephosphorylation , we used a combined approach of chromatin immunoprecipitation - chip ( ChIP-chip ) , coupled with microarray expression studies . One hundred forty genes were identified as being both a direct target of MoCRZ1 and having expression concurrently differentially regulated in a calcium/calcineurin/MoCRZ1 dependent manner . Highly represented were genes involved in calcium signaling , small molecule transport , ion homeostasis , cell wall synthesis/maintenance , and fungal virulence . Of particular note , genes involved in vesicle mediated secretion necessary for establishing host associations , were also found . MoCRZ1 itself was a target , suggesting a previously unreported autoregulation control point . The data also implicated a previously unreported feedback regulation mechanism of calcineurin activity . We propose that calcium/calcineurin regulated signal transduction circuits controlling development and pathogenicity manifest through multiple layers of regulation . We present results from the ChIP-chip and expression analysis along with a refined model of calcium/calcineurin signaling in this important plant pathogen .
Rice blast , caused by the fungal pathogen Magnaporthe oryzae , is a recurrent and devastating problem worldwide [1] . Severe disease outbreaks can destroy upwards of 90% of rice yields for an entire field , region , or country resulting in a dramatic impact on human welfare and regional economies . The process starts when an asexual spore lands on a rice leaf . Given suitable moisture and temperature , the spore germinates with a short germination tube at the tip of which a specialized infection structure , an appressorium , emerges . The formation of an appressorium is essential for successful disease as it facilitates breaching the plant cuticle and cell wall , allowing access to the underlying tissues . Following penetration , the entry peg forms an un-branched hyphal strand that subsequently matures into branched bulbous infection hyphae . The fungus fills the infected cell in what is considered a biotrophic state before invading adjacent cells and switching to a necrotrophic state where it ramifies through the host tissues killing cells . In the final stage , the fungus produces new asexual spores that are spread to neighboring plants [2] , [3] . While knowledge of the core signal pathways regulating each phase of this process continues to resolve , the key determinants controlling environmental perception and cellular response are as yet not fully understood [4] . Specifically , we have little knowledge of the upstream receptors used by the fungus to detect stimuli , nor do we know how downstream factors specifically interact to affect expression of genes deployed during infection related development , establishment of host associations , and invasive growth . Calcium signaling has been implicated in regulating growth and development in M . oryzae including the infection process [5]–[9] . The components of Ca2+ signaling have been studied in many organisms and are relatively well understood . Ca2+ signaling starts when G-protein coupled receptors are activated by an external stimulus . Phospholipase C ( PLC ) is activated to hydrolyze phosphatidyl inositol-1 , 4-bisphosphate ( PIP2 ) into inositol 1 , 4 , 5-triphosphate ( IP3 ) and diacylglycerol . IP3 activates Ca2+ release from intracellular stores into the cytosol . Ca2+ ions bind to and activate calmodulin , which in turn , activates the Ca2+/calmodulin-dependent serine/threonine protein phosphatase calcineurin . Calcineurin is a heterodimer consisting of catalytic ( CNA ) and regulatory ( CNB ) subunits . In fungi , calcineurin mediated Ca2+ signaling has been shown to be required for growth , development , response to stress , and pathogenesis [10] . It was necessary for survival during environmental stresses such as ions ( Mn2+ , Li+ , Na+ ) , high pH , high temperature , ER stress , and prolonged incubation with mating pheromone α-factor in Saccharomyces cerevisiae [11] , [12] . It is essential for growth and virulence of Candida albicans and Cryptococcus neoformans [13]–[16] , and controls the dimorphic transition from mycelia to yeast in Paracoccidioides basiliensis [17] . Effects of gene deletion or chemical inhibition in filamentous fungi typically have pleiotropic effects . For example , a cnaA deletion mutant in Aspergillus fumigatus was viable but severely affected in hyphal morphology , sporulation , conidial architecture , pathogenicity , and invasive growth [18] , [19] . Reduction of calcineurin activity by the immunosuppressant drug cyclosporine A , resulted in reduction of mycelial growth and alteration in hyphal morphology as shown in Neurospora crassa [20] , [21] , A . nidulans [22] , A . oryzae [23] , and M . oryzae [24] . RNA silencing in M . oryzae showed similar effects , specifically a reduction in mycelial growth , sporulation , and appressorium formation in MCNA knock down mutants [8] . Calcineurin functions mainly through the activation of the transcription factor CRZ1 ( Calcineurin Responsive Zinc Finger 1 ) . Upon activation by increased intracellular Ca2+ and calmodulin , calcineurin dephosphorylates CRZ1 leading to its nuclear localization . As a major mediator of calcineurin signaling , crz1 deletion mutants in a variety of fungi showed similar phenotypes as calcineurin mutants [5] , [25]–[28] . However , differences have been noted . CRZ1 in C . albicans was not involved in tolerance to antifungal agents ( fluconazole , terbinafine ) and only slightly affected in virulence , which is in contrast to the calcineurin mutants [28] . On the other hand , CRZ1 is strongly associated with virulence both in human and plant pathogenic fungi [5] , [25] , [26] , [28] , [29] . The B . cinerea CRZ1 ortholog BcCRZ1 was required for growth , conidial and sclerotial development , and full virulence while being dispensable for conidia-derived infection of bean plants [25] . In M . oryzae , the Δmocrz1 deletion mutant showed decreased conidiation and was not able to cause disease when spray inoculated . Mutant conidia were not distinguishable from that of wild type and formed appressoria at a similar level as wild type . Importantly , a significant portion of appressoria failed to penetrate rice sheath tissue . The mutant could colonize and cause disease when the conidia were infiltrated directly into the host tissue , thus bypassing the penetration process and suggesting that MoCRZ1 plays a role in appressorium mediated penetration and establishing a biotrophic association with its host [5] . Comprehensive genome-wide expression analysis in S . cerevisiae identified 163 genes regulated in a calcineurin/CRZ1 dependent manner by the stimulation of Ca2+ or Na+ . These genes were associated with a diverse range of cellular processes including signaling pathways , ion/small molecule transport , cell wall synthesis/maintenance , and vesicular trafficking [30] . In C . albicans , microarray analysis revealed 60 genes to be transcriptionally activated by exogenous Ca2+ treatment through calcineurin/CRZ1 regulation using cnaΔ/Δ and crz1Δ/Δ mutants . Analysis of putative functions revealed that about 60% of these genes were involved in cell wall organization , cellular organization , cellular transport and homeostasis , cell metabolism , and protein fate [28] . Many of the genes regulated through calcineurin signaling in these two species belonged to similar functional groups , although only 9 genes were found to be commonly regulated [28] . To date , no genome-wide study has been conducted to identify regulated genes by direct binding of CRZ1 to promoter regions . Here we report the use of chromatin immunoprecipitation coupled with non-coding region tiling arrays ( ChIP-chip ) analysis and whole-genome expression studies to identify target genes directly regulated by MoCRZ1 . To our knowledge , this is the first report of ChIP-chip technology being applied to filamentous fungi . From this analysis , we can model the Ca2+/calcineurin signaling and control pathways that in-part influence infection related development and establishment of a compatible host association for this devastating plant pathogen . Further , this work reveals divergence within the fungal kingdom of the suites of genes and processes directly regulated by this ubiquitous signaling pathway .
A MoCRZ1::eGFP construct was co-transformed into fungal protoplasts along with a hygromycin resistance conferring vector . Transformants were single spore isolated and screened under the epi-fluorescent microscope . MoCRZ1-eGFP fluorescence was faint and evenly dispersed through the cytosol in mycelia with nuclear localization detected in hyphal tips ( data not shown ) . Nuclear translocation of CRZ1 in response to Ca2+ is a well conserved phenomenon and was shown previously to occur in M . oryzae [5] . Following mycelia treatment with CaCl2 , eGFP fluorescence was localized to the nucleus ( Figure 1A ) , as expected . Addition of the calcineurin inhibitor FK506 completely blocked nuclear accumulation of the fluorescent protein . The MoCRZ1::eGFP over-expression line showed normal growth , appressoria development , and virulence to susceptible rice cultivar Nipponbare ( data not shown ) , and was selected for subsequent ChIP-chip analysis . Experimental design for ChIP-chip analysis is depicted in Figure 1B . CaCl2 treated mycelia were used to enrich MoCRZ1 occupied genomic fragments , while mycelia treated with CaCl2 and FK506 were used as the negative control . Expression of PMC1 ( P-type ATPase ) was analyzed for each sample by RT-PCR as shown in Figure 1C to confirm the effect of each treatment , i . e . , up-regulated in the calcium treated mycelium and blocked by FK506 . PMC1 is a previously described target of MoCRZ1 [5] and was used throughout this study as positive control marker . Enrichment of MoCRZ1 bound DNA fragments in ChIPed fractions when compared to input DNA ( non-ChIPed ) was confirmed by real-time PCR ( Figure 1D ) . As expected , the fold change of PMC1 was 4 . 54±1 . 13 in ChIPed DNA from Ca2+ treated mycelia over input DNA from Ca2+ + FK506 treated mycelia , while that of β-tubulin was 1 . 78±0 . 56 ( Figure 1D ) . Input and IPed DNA was amplified and subsequently re-amplified to amass sufficient DNA ( ∼8 . 5 ug ) for labeling and hybridization to the array . Enrichment of MoCRZ1 bound DNA in the ChIPed fraction was validated by PCR amplification of PMC1 at each step ( data not shown ) . ChIPed DNA labeled with Cy5 was co-hybridized with Cy3 labeled input DNA to the NimbleGen M . oryzae intergenic region specific tiling array ( see Materials and Methods for array description ) . Two complementary approaches were applied to analyze ChIP-chip hybridization data to identify putative MoCRZ1 binding sites and the genes it regulates . Initially , 42 peaks were identified as having a false discovery rate less than 0 . 2 and being common in both biological replications of the Ca2+ treatment but not in the Ca2++FK506 treatment . The 42 peaks were within 1 kb upstream of 37 predicted genes . Following this initial analysis , we used NimbleGen's SignalMap software to manually interrogate the ratio signal tracks across the genome to identify short sequence stretches showing a normal distribution profile of signal intensities upstream of annotated ORFs ( Figure 1B bottom ) . Sequence tracks showing this profile were accepted only if they appeared in both biological replications of the Ca2+ treatment with no or lower signal intensities in the Ca2++FK506 treatment . If the binding signals were located between two divergently transcribed ORFs , both ORFs were regarded as possible candidates . This manual analysis resulted in the identification of 346 genes evenly distributed through the genome with no apparent bias ( Figure 2 , Table S1 ) . Importantly , the 37 genes resulting from the first automated analysis were captured in the set of 346 . Manual analysis produced more putative targets than the automated because SignalMap reports the probe ID with the highest signal intensity . In most cases , a single binding site did not share the identical probe as having the highest signal between biological replicates , thereby disqualifying them from the automated analysis . Data was deposited in the Gene Expression Omnibus ( GEO ) at NCBI under the accession number of GSE18180 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? token=dlarvwwmsoquoxs&acc=GSE18080 ) . Expression microarray analysis was conducted to corroborate genes predicted by ChIP-chip to be regulated by MoCRZ1 . The experimental design is described in Figure 3A . Wild type strain KJ201 was treated with CaCl2 without or with FK506 to identify calcium and calcineurin dependent genes , respectively . MoCRZ1 deletion mutant ( Δmocrz1 ) was also treated with CaCl2 to identify MoCRZ1 regulated genes . Four biological replicates for each of the 4 treatments were selected for hybridization to the Agilent M . oryzae whole genome microarray chip version 2 . Signal intensities from the single channel hybridization were normalized to the average expression level of all probes among the 16 data sets . Pair wise comparison between treatments was conducted as depicted in Figure 3A , in which ( a ) , compared Ca2+ treated/no treatment in wild type strain KJ201 ( CA/CK ) ; ( b ) , Ca2+ treated/Ca2++FK506 treated in KJ201 ( CA/CAFK ) ; ( c ) , Ca2+ treatments in KJ201/Ca2+ treatments in Δmocrz1 ( CA/CRZ ) . Genes were regarded as differentially expressed if their average signal intensity among 4 replicates was above 20 in a minimum of one condition and the expression ratio is greater than 2 fold with P<0 . 05 ( Student's t-test ) . Changes in gene expression of the 346 genes identified from ChIP-chip were analyzed . Of the 346 , we found 309 with expression in each condition , with 121 and 19 genes up- or down- regulated , respectively , in the Ca2+ treated KJ201 condition in at least one comparison ( Table S1 , Figure 3B ) . It was noteworthy that the expression level of some MoCRZ1 target genes was lower in the Ca2+ activated wild type condition than in calcineurin and/or MoCRZ1 defective conditions suggesting that MoCRZ1 can act as repressor . These 140 ( 121+19 ) genes represent those directly bound to and regulated by MoCRZ1 , and form the set used for analyses described below . The full dataset was deposited in NCBI GEO with the accession number of GSE18185 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? token=ttmdlgumsmsoury&acc=GSE18185 ) . SuperSeries GSE18193 combining the ChIP-chip and microarray data were also generated ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? token=tnutnsemyikamlm&acc=GSE18193 ) . The 15 most up-regulated genes in calcium treated wild type strain KJ201compared to that of the Δmocrz1 mutant were selected for real-time RT-PCR to validate expression data . The results in Figure 4 show that each gene is transcriptionally more activated in all three comparison , i . e . , Ca2+ treated vs . untreated control ( a ) , Ca2+ treated vs . Ca2+ + FK506 treatment ( b ) , and Ca2+ treated wild type vs . Δmocrz1 ( c ) . Although the magnitude of fold changes was much higher than that from microarray in most cases , real-time RT-PCR data supported the microarray results . Functional categorization was conducted in two ways . At first , hierarchical classification according to the expression pattern resulted in two groups . This analysis was followed by GO annotation using an InterPro to GO module incorporated in the Comparative Functional Genomics Platform ( Figure 3C and Table S2 , S3 ) [31] . Group I contains 64 genes of which expression was tightly regulated in a Ca2+/calcineurin/MoCRZ1 dependent fasion , i . e . , up-regulated in all three comparisons . Group II comprises 76 genes whose expression was differentially regulated in three comparisons . Twenty four genes of group I and 36 of group II were assigned with GO terms , 14 and 22 to biological process , 20 and 33 to molecular function , 10 and 19 to cellular component , respectively ( Table S2 ) . Second , genes were annotated through literature with their specific functions assigned according to the functional classes of Cyert [32] ( Table 1 ) . Sixty-two of 140 genes identified by both ChIP-chip and microarray analyses could be assigned to one of 7 functional groups ( Table 1 ) . Consistent with the role of MoCRZ1 in providing tolerance to ionic and cell wall stress , genes involved in small molecule transport or ion homeostasis and cell wall synthesis/maintenance were highly represented . Among them was PMC1 , which provides support for the validity of these results . Furthermore , the AAA family of ATPase as a whole was highly represented as direct targets of MoCRZ1 , as well as members of major facilitator superfamily of multidrug-resistance proteins . Considering cell wall synthesis/maintenance genes , a number of GPI-anchored cell surface glycoproteins were captured in addition to the previously known downstream genes like chitin synthase activator ( Chs3 ) and chitin syntase 1 . Small secretory proteins , including effectors and cell wall degrading proteins , are regarded as key molecules acting at the interface between the plant and microbe . Efficient secretion of these proteins is assumed to be essential during the interaction between host and pathogen . Among the targets identified were genes comprising the vesicle mediated secretory pathway , including rhomboid family membrane protein ( MGG_07535 ) , Sso1/2 type SNARE protein ( MGG_04090 ) known to be localized at secretory vesicles from Golgi to plasmamembrane , homocysteine S-methyltransferase ( MGG_04215 ) , golgi apyrase ( MGG_07077 ) , and a protein required for assembly of ER-to-Golgi SNARE complex ( MGG_01489 ) . Proper protein folding in the ER mediated by co-chaperone LHS1 [33] , and efficient Golgi performance involving exocytosis entailing functions of the integral membrane P-type ATPase encoded by MgAPT2 [34] , have been reported to be necessary for protein secretion and biotrophic phase infection in this fungus . A major group of genes found in this study to be regulated by MoCRZ1 are those involved in cellular signaling and transcription . Among them were genes encoding serine/threonine protein kinases ( MGG_04660 , MGG_07287 , MGG_06928 ) , a phosphatase ( MGG_00552 ) , and a Rho guanyl nucleotide exchange factor ( MGG_11178 ) . In addition , genes comprising calcium signaling machinery were also common . Genes encoding annexin ( MGG_06360 ) , lysophospholipase 3 ( MGG_07287 ) , PX domain-containing protein ( MGG_11649 ) , calcineurin binding protein ( CBP1; MGG_03218 ) and the calcineurin temperature suppressor CTS1 ( MGG_01150 ) were identified . Of note , the expression of CBP1 and CTS1 was highly increased in all three comparisons . Binding in the promoter of these two genes and regulation of their transcription strongly suggests a previously uncharacterized level of negative feedback regulating calcineurin activity . MoCRZ1 functions to activate 12 genes from diverse families of transcription factors . Significantly , MoCRZ1 itself was identified , suggesting an autoregulatory role not previously reported . The expression of MoCRZ1 was induced by exogenous calcium treatment , but not altered by the inhibition of calcineurin activity with FK506 . Regarding the fact that calcineurin dephosphorylates MoCRZ1 upon activation by calcium and that FK506 blocks nuclear localization of MoCRZ1 , inactivation of calcineurin regulates function only at post-translational level . This data suggests an additional level of regulation at transcription . Δmocrz1 mutants are defective in post appressoria formation penetration and establishment of biotrophic host association . Appressoria from this mutant background have defects in penetration , however those that successfully penetrated fail to incite disease . We examined MoCRZ1 target genes for previously defined roles in fungal virulence . Of the 140 target genes queried to pathogen-host interactions database ( PHI-base ) version 3 . 1 [35] , [36] , 16 had matches to more than one entry from plant or human pathogens using a stringent cut-off value e<−20 ( Table 2 ) . Three proteins , MoCRZ1 itself [5] , MGG_03530 encoding chitin syntase activator Chs3 [37] and PMC1 [8] were previously characterized to regulate virulence in M . oryzae . Similarly , members of ATPases family , PDE1 [38] and MgAPT2 [34] , are known to be involved in M . oryzae pathogenicity . Two genes encoding serine/threonine protein kinases ( MGG_04660 and MGG_06928 ) had 45 and 27 hits , respectively . Genes involved in drug resistance ( MGG_05723 and MGG_10869 ) , MGG_07230 alpha-1 , 3-mannosyltransferase CMT1 , MGG_07287 lysophospholipase 3 , MGG_05727 ankyrin repeat protein , MGG_03288 bZIP transcription factor , and MGG_09361 homolog of CgDN3 were similarly listed as being involved in fungal virulence . To identify the MoCRZ1 binding motif , the exact binding sequences of MoCRZ1 ( ∼50–1247 bp ) revealed from ChIP-chip studies were retrieved from the promoters of genes in common between ChIP-chip and microarray expression studies and subjected to motif signature analyses ( Figure 5A ) . Initially , 106 sequences from each of the 83 genes differentially regulated in the WT/Δmocrz1 comparison ( Figure 3B ) were analyzed with MEME [39] and MDScan [40] ( Figure 5A ) . There were more sequences than genes as 21 genes had 2 peaks in their promoter region and 2 genes had 3 . Candidate motifs from both algorithms were manually interrogated and enumerated to identify the top 3 candidates , which were subsequently screened against randomly retrieved 106 intergenic sequences with an average length of 509 bp ( Figure 5A ) . The most enriched motif of CAC[AT]GCC was identified in 33 sequences in front of 24 genes , a 16X enrichment in MoCRZ1 bound sequences . The most common motif of TTGNTTG was found in 68 promoter sequences in front of 42 genes with 4X enrichment . Motif TAC[AC]GTA occurred in 22 promoter sequences of 18 genes with 4X enrichment . Fifty-six genes had at least one motif , while all three of these motifs occurred in front of 5 genes including PMC1 , CTS1 , MGG_01494 encoding a cell wall protein , and two genes encoding conserved hypothetical proteins MGG_03539 and MGG_06359 ( Table S4 ) . These motifs were searched against yeast motif database using TOMTOM [41] . The top match for CAC[AT]GCC was MET28 ( p-value = 0 . 0013 ) , while the second match was CRZ1 with significant p-value ( 0 . 0022 ) , showing Crz1p of S . cerevisiae has this motif in its promoters although it was not previously identified as a calcineurin dependent response element ( CDRE ) ( Figure 5B ) . Pbx1b ( p-value = 0 . 00039 ) and Zec ( p-value = 0 . 00045 ) were best two matches for TTGNTTG , while no significant match was returned for TAC[AC]GTA . Binding of MoCRZ1 to the promoter region was confirmed by Electrophoretic Mobility Shift Assay ( EMSA ) . A 209 bp PCR fragment having 1 TTGNTTG and 2 CAC[AT]GCC motifs from the MoCRZ1 promoter region was bound to purified MoCRZ1 protein ( Figure 5C , left panel ) . A 325 bp fragment of CBP1 ( MGG_03218 ) having 1 TTGNTTG and 1 CAC[AT]GCC motifs was also shown to bind to purified MoCRZ1 ( Figure 5C , right panel ) .
Our combined approach identified 140 direct targets of MoCRZ1 whose expression was concurrently regulated in a calcineurin/CRZ1 dependent manner . Sixty-two of these genes could be grouped in the same categories that were used to functionally assign yeast genes [32] . These groups include cell wall synthesis , ion or small molecule transport , vesicle transport , lipid or sterol synthesis , degradative enzymes , and signaling and transcription . This is expected as the Δmocrz1 mutant also exhibited sensitivity to calcium ions and chemicals perturbing cell wall integrity as did the yeast crz1 mutant [5] . However , the large diversity in the suite of individual target genes is compelling . In S . cerevisiae , genome wide expression profiling identified 153 calcium/calcineurin/Crz1p dependent genes [30] . Comparison between these two sets showed that only 15 out of 140 MoCRZ1 targets had 12 yeast orthologs whose expression was regulated by calcium/calcineurin/Crz1p ( Table 3 ) . When we compared our gene list to the 120 A . fumigatus genes whose expression was changed by exposure to calcium for 10 min [26] , only 21 matched to 14 A . fumigatus genes , with only 6 having reciprocal best blast hits ( Table 3 ) . When the same analysis was applied to the 141 A . fumigatus genes whose expression was modulated by AfcrzA , as recently reported by Soriani et al . [43] , 28 MoCRZ1 targets having 15 A . fumigatus orthologs were found with only 3 matching the 14 genes identified previously ( data not shown ) . The observed diversity may reflect divergently evolved molecular features modulated by CRZ1 in each species to cope with its unique environment . This diversity was also reflected in the crz1 mutant phenotypes across the species . For example , crz1 mutants in different species showed a spectrum of ion sensitivity; ΔcrzA of A . fumigatus was more sensitive to Mn2+ , but Δmocrz1 of M . oryzae was not . BcCRZ1 of B . cinerea was dispensable for conidia mediated infection , but MoCRZ1 was necessary . These data suggest that although the core calcium signaling machinery including calmodulin and calcineurin is highly conserved across the species , their mechanism of action has diverged . Similar suggestions have been proposed by Karababa et al . [28] . Kraus and Heitman [44] also made note of species-specific action mechanisms through which calcium signaling involving calmodulin and calcineurin was manifested in three different species , S . cerevisiae , C . albicans , and C . neoformans . Additional evidence supporting this hypothesis is found in the differences in CDRE ( calcineurin-dependent response elements ) sequences [27] , [30] , [45] , [46] . In S . cerevisiae , nucleotide stretches of 5′-GNGGC ( G/T ) CA-3′ was reported as the Crz1p-binding site by in vitro site selection . A similar sequence , 5′-GAGGCTG-3′ , was also identified as a common motif in the upstream 500 bp regions of 40 genes with ≥4 . 0 fold Crz1-dependent expression profile [30] . Two similar sequences , 5′-GTGGCTC-3′ and 5′-GAGGCTC-3′ , were reported as CDREs from the genus Aspergillus in the A . nidulans chsB and A . giganteus afp promoters [47] . A slightly divergent motif , 5′-G ( C/T ) GGT-3′ , was identified as a common regulatory sequence from the 60 upstream regions of calcineurin/Crz1p-dependent genes of C . albicans [28] . In contrast , the motif sequences obtained in this study shows further divergence from that of S . cerevisiae or C . albicans . Although , core hepta nucleotide , 5′-GGCTC-3′ , was found in the probe sequence of 10 genes including PMC1 ( MGG_02487 ) and MGG_03530 ( chitin synthase activator ) , the occurrence of the full S . cerevisiae or C . albicans motif sequences is not enriched/overrepresented among the 106 sequences we analyzed ( data not shown ) . In addition to PMC1 , several other genes implicated in the calcium signaling pathway regulating calcineurin activity were identified . Among them were calcineurin binding protein CBP1 ( MGG_03218 ) and the calcineurin temperature suppressor CTS1 ( MGG_01150 ) , suggesting feedback regulation of calcineurin activity mediated by MoCRZ1 . In addition , MoCRZ1 bound its own promoter to activate expression . CBP1 shows homology to CbpA in A . fumigatus and Cbp1 of C . neoformans , which in turn , are orthologous to RCN1 of S . cerevisiae , an inhibitor of calcineurin called calcipressin . Over-expression of RCN1 in a pmc1 mutant background conferred Ca2+ tolerance by activation of vacuolar Ca2+/H+ exchanger Vcx1p , expression of which was negatively regulated by calcineurin . Expression of RCN1 ( YKL159c ) was regulated by calcineurin and Crz1p , suggesting negative feedback regulation [30] , [48] . However , both stimulatory and inhibitory regulation of calcineurin by RCN1 was reported [49] . RCN1 expression at endogenous level and phosphorylation by GSK-3 kinase , positively regulated calcineurin activity [49] . Degradation of phosphorylated RCN1 is required for precise calcineurin activity in response to changes in Ca2+ concentration [50] . Negative feedback regulation of calcineurin by CbpA was also reported in A . fumigatus , where it down regulates cnaA expression as well as that of downstream genes vcxA and chsA encoding vacuolar Ca2+/H+ exchanger and chitin synthase A , respectively [51] . Expression of CbpA ( Afu2g13060 ) was known to be up-regulated in response to Ca2+ , which was CrzA dependent [26] . This feedback loop seems to extend at least to the level of calcineurin . Roles and action mechanisms of Cbp1 in the calcium/calcineurin signaling pathway also seems to be diverged in a species specific manner . In C . neoformans , Cbp1 does not stimulate or inhibit calcineurin expression , and does not seem to participate in a feedback loop . Taken together , these data lead us to propose that a negative or positive feedback loop , which includes MGG_03218 , regulates the calcium/calcineurin signal transduction pathway in M . oryzae . Phospholipid-binding protein Cts1 ( calcineurin temperature suppressor ) was identified and characterized in C . neoformans as able to restore growth defect at 37°C in calcineurin-deficient strains and to confer resistance to the calcineurin inhibitor FK506 [52] . Δcts1 mutants were synthetically lethal in combination with a calcineurin mutation . However , no direct interaction between Cts1 with either the catalytic or regulatory subunit of calcineurin was reported . With these data , they suggested that Cts1 acted in either parallel pathways or a branched pathway to compensate , at least in part , for the loss of calcineurin function [52] . MGG_01150 , an ortholog of Cts1 , was found to be a direct target of MoCRZ1 . Its calcineurin dependent expression pattern is opposite to that of Cts1 . Unlike the elevated expression of Cts1in calcineurin deficient strains , MGG_01150 expression was activated by Ca2+ treatment , which was blocked by calcineurin inhibitor FK506 and abolished in the Δmocrz1 mutant . Therefore , it is evident that MGG_01150 is a component of the calcineurin signaling pathway in M . oryzae unlike its counterpart in C . neoformans . Our data revealed that MoCRZ1 binds to its own promoter to activate expression in a Ca2+ /calcineurin dependent manner . Therefore , MoCRZ1 regulation appears to occur at the posttranslational and transcriptional levels via activation by calcineurin and positive autoregulation respectively . Calcineurin/CRZ1 dependent expression of CRZ1 was also reported in C . albicans suggesting a common mechanism of regulation across the fungal species [28] . However , expression dynamics of the catalytic ( MCNA: MGG_07456 ) and regulatory ( MCNB: MGG_06933 ) subunit were not altered in the Δmocrz1 mutant compared to wild type in response to Ca2+ treatment ( data not shown ) . This suggests that the proposed feedback regulation does not include direct regulation of calcineurin expression by CRZ1 . Involvement of CRZ1 in fungal virulence has been recently demonstrated in both human and plant pathogenic fungi [5] , [10] , [25] , [26] , [28] , [29] . Signals related to these virulence traits seemed to be transmitted to a diverse range of downstream genes , as 18 genes out of 140 direct targets including MoCRZ1 have been found to be related to fungal virulence in both human and plant pathogens . Gene repertoire ranges from cell wall synthetic enzymes , proteins conferring resistance to antifungal agents encoded by major facilitator type transporter , calcium homeostasis to transcription factors . Three genes ( MoCRZ1 , MGG_03530 encoding chitin synthase activator 3 , and MGG_02487 ) have been functionally characterized in M . oryzae [5] , [8] , [37] . Association of MGG_03530 encoding chitin synthase activator 3 ( Chs3 ) with virulence was found in a T-DNA insertion strain with reduced growth rate on nutrient rich media , reduced conidiation rate with aberrant conidia morphology , and reduced appressorium formation and virulence [37] . Filamentous fungal genomes contain up to 10 chitin synthase genes of 7 classes [53] . Different CHS were regarded as to have functional redundancy in a variety of developmental processes because single mutation of a class I or II gene did not result in a marked phenotype . Therefore , specific roles for individual genes have not yet been clearly assigned and the association with fungal virulence has been controversial . However , several lines of evidence implicate class V myosin-like CHS in virulence in the maize anthracnose pathogen Colletotrichum graminearum [54] and in Fusarium oxysporum , the tomato wilt fungus [55] . One Class III chitinases , BcChs3a of Botrytis cinerea had important roles in virulence , especially on leaf tissue colonization , grape vines and Arabidopsis thaliana [56] . The M . oryzae genome contains 7 predicted chitin synthases , of which the expression of 5 were induced and 1 repressed in response to exogenous calcium treatment ( data not shown ) . Only one ( MGG_01802 encoding class II chitinase ) was directly regulated by MoCRZ1 . Therefore , calcium seems to regulate expression of most chitin synthases in diverse pathways . Two genes encoding small molecule transporting P-type ATPases ( MGG_02487 Ca transporting ATPase and MGG_12922 phospholipid-trasnporting ATPase ) were found to have homologs implicated in fungal pathogenicity . Knock-down of MGG_02487 encoding PMC1 by RNAi technology resulted in no conidiation , growth retardation , and reduced melanization [8] . The association of PMC1 and fungal virulence was not investigated in other fungi . Other evidence on the involvement of Ca2+ transporting ATPase in fungal virulence arises from the study of PMR1 of C . albicans [57] . PMR1 is a Golgi membrane located Ca2+ transporting P-type ATPase , and is known to work cooperatively with PMC1 in the maintenance of cytosolic Ca2+ homeostasis . Capmr1Δ mutant of C . albicans had a weakened cell wall , probably due to the glycosylation defect and showed severely attenuated virulence in a murine model of systemic infection [57] . Two genes encoding Drs2 family of P-type ATPases , PDE1 and MgAPT2 , were functionally characterized to act in appressorium formation and invasive growth [34] , [38] . MgAPT2 was necessary for the normal development of Golgi apparatus that is required for secretion of a subset of extracellular enzymes via exocytosis [34] , [58] . This study is the first of its kind where ChIP-chip technology has been applied to filamentous fungi . The correlation of comprehensive whole genome expression data with results from ChIP-chip have allowed for significant refinement of the predicted targets of MoCRZ1 . This refinement alone allowed for the identification of a predicted signature binding motif for this transcription factor . This study reveals conserved elements of the calcium/calcineurin signaling pathway , as well as elucidates species specific differences that we propose function to regulate the system and allow for responses tailored to biology of the organism . Calcium signaling is a ubiquitous and complicated aspect of cell physiology . This study represents a major advance in our understanding of this pathway in M . oryzae and provides the launching point for the functional characterization of the genes and interactions it implicates . Figure 6 depicts our proposed model resulting from this work and includes our new findings of predictive roles for MoCRZ1 in autoregulation , feedback inhibition , and secretion .
Strains of M . oryzae were maintained on oatmeal agar ( 50 g of oatmeal per liter ) and grown at 22 °C under constant fluorescent light to promote conidiation . Mycelial blocks from actively growing margins of colony were inoculated into complete media ( CM ) liquid media at 25 °C by shaking for 3 days . After thorough washing with sterile distilled water , the mycelia were treated with 200 mM CaCl2 with or without 10 µg/ml FK506 for the indicated time . Mycelia were harvested and immediately frozen with liquid nitrogen and stored at −80°C before use . FK506 ( Sigma , St . Louis , MO ) stock solution in 5 mg/ml was prepared with DMSO , and stored at −20°C until used . Protoplasts generation and transformation were performed following established protocols [59] . Protoplasts were generated from young mycelia grown in complete media with 10 mg/ml Lysing Enzyme ( Sigma , St . Louis , MO ) in 20% sucrose . Protoplasts were harvested by filtration through 4 layers of miracloth ( Calbiochem , Darmstadt , Germany ) , washed twice with STC ( 20% sucrose , 50 mM Tris-HCl , 50 mM CaCl2 , pH 8 . 0 ) followed by centrifugation at 5 , 000 rpm for 15 min at 4°C , and resuspended to 5×107 protoplasts/ml . GFP tagging construct in TOPO cloning vector ( PMoCRZ1::MoCRZ1::GFP ) was co-transformed with pCX63 containing hygromycin resistance cassette by the mediation of 40% polyethyleneglycol . After incubation for 7–10 days at 25°C , hygromycin resistant colonies were transferred to V8 juice agar media . The fluorescing transformants observed under the microscope with epifluorescent optics ( Nikon eclipse 80i , Melville , NY ) were purified through single spore isolation . Nuclear translocalization of MoCRZ1 was observed after treatment with 200 mM CaCl2 with or without 10 µg/ml FK506 for 1 hour at room temperature . Total RNA was isolated from frozen mycelial powder using an Easy-Spin RNA extraction kit ( iNtRON Biotechnology , Seoul , Korea ) . Five micrograms of total RNA was reverse-transcribed into first-strand cDNA by oligo dT priming using the SuperScript first-strand cDNA synthesis kit according to the manufacturer's instructions ( Invitrogen Life Technologies , Carlsbad , CA ) . Resulting cDNA was diluted to 1∶20 with sterile water . Real-time RT-PCR was performed according to the established protocol [59] using iQ SYBR Green Supermix ( Bio-rad , Hercules , CA ) on an iCycler iQ5 Real-Time PCR Detection System ( Bio-rad ) . Fold changes were calculated by 2−ΔΔCt , where ΔΔCt = ( Ctgene of interest-Ctcontrol gene ) test condition- ( Ctgene of interest-Ctcontrol gene ) control condition . Test and control conditions are as same as in Figure 3A , where ( a ) compares expression level between Ca2+ treated vs . no treatment in wild type strain KJ201 , ( b ) Ca2+ vs . Ca2++FK506 in KJ201 , ( c ) Ca2+ treatments in KJ201 vs . in Δmocrz1 . Primer sequences were listed in Table S5 . Young mycelia grown in liquid media were treated with 200 mM CaCl2 with or without 10 µg/ml FK506 for 1 hour with shaking . The harvested mycelia were divided with one part being immediately frozen in liquid nitrogen for future RNA isolation and the other treated with 1% formaldehyde in buffer A ( 0 . 4 M sucrose , 10 mM Tris-HCl , pH 8 . 0 , 1 mM EDTA , 1 mM phenylmethylsulfonyl fluoride , and 1% formaldehyde ) for cross-linking for 20 min . Mycelia were harvested with excess amount of distilled water after stopping cross-link with 0 . 1 M glycine for 10 min , frozen in liquid nitrogen , ground into a fine powder in a chilled mortar and pestle , and stored at −80°C until used . Chromatin immunoprecipitation was conducted according to published procedures with modification [60] , [61] . Nuclear DNA was then isolated from cross-linked mycelia with Plant Nuclear Isolation Kit ( Sigma , St . Louis , MO ) and sheared into fragments by sonication to 200- to 1 , 000-bp with an average size of 500 bp . Sonication was conducted on ice with an amplitude of 10% using 30×30 s pulses ( 30 s between bursts ) using Biorupter ( Cosmo Bio , Tokyo , Japan ) . After pre-clearing nuclear lysates with Salmon sperm/protein A agarose ( Upstate , Temecula , CA ) for 4 hours at 4°C , immunoprecipitations were performed with either 1 µg of rabbit control IgG ( ab46540-1 , Abcam , Cambridge , MA ) or 0 . 5 µl of antiGFP antibody ( ab290 , Abcam ) for overnight at 4°C . A small aliquot of DNA ( 30% ) was saved for input DNA ( input ) . Immunoprecipitated DNA was captured with proteinA agarose beads ( Upstate , Temecula , CA ) for 4 hours , and then washed twice with LNDET buffer ( 0 . 25 M LiCl , 1% NP40 , 1% deoxycholate , 1 mM EDTA ) and twice with TE buffer . The DNAs were reverse-cross linked at 65 °C overnight in elution buffer ( 1% SDS and 0 . 1 M NaHCO3 ) containing 1 mg/ml proteinase K , and purified using PCR purification kit ( Qiagen ) . Real-time PCR was performed with 1 µl each of pulled-down DNA and input DNA as template following the procedures described above . Fold changes for control gene ( β-tubulin ) and putative target gene ( PMC1 ) were calculated by 2−ΔΔCt , where ΔΔCt = ( Ctinput DNA-CtChIPed DNA ) Ca2+ treated sample - ( Ctinput DNA-CtChIPed DNA ) Ca2+/FK506 treated sample . Primer sequences for the promoter region of PMC1 and β-tubulin were listed in Table S5 . For ChIP-chip experiments , 10 µl ChIPed DNA and 10 ng input DNA were amplified using GenomePlex Whole Genome Amplification Kit ( Sigma ) . Amplified DNA was then labeled with Cy3 or Cy5 fluorescent dyes for input or immunoprecipitated DNA , respectively , and hybridized to NimbleGen Magnaporthe grisea promoter tiling arrays according to the manufacturer's instruction ( NimbleGen Systems of Iceland ) . Probes for tiling array were designed to have about 70 nucleotides per 100 bp of promoter and intergenic region based on annotation of M . grisea genome version 5 . After getting peak intensity , peak data files ( . gff ) were generated from the scaled log2-ratio data using NimbleScan . It detects peaks by searching for 4 or more probes whose signals are above the cutoff values using a 500 bp sliding window . The ratio data was then randomized 20 times to evaluate the probability of “false positives” . Each peak was then assigned a false discovery rate ( FDR ) score based on the randomization . Peaks with FDR score ≤0 . 2 were regarded as positive . Mycelia of wild type KJ201 and Δmocrz1 strain were treated with 200 mM CaCl2 with or without 10 µg/ml FK506 , with water as control . Initially , samples were harvested at 0 , 15 , 30 , and 60 min . after treatment . PMC1 expression level was checked by RT-PCR with the highest expression at the 30 min . time point . Four biological replicates of wild type and mutant mycelia were harvested after 30 min . treatment with chemicals , frozen immediately with liquid nitrogen . Total RNA was isolated described as above . After validation of sample quality by RT-PCR , total RNA was sent to Cogenics ( Morrisville , NC ) for hybridization to the Agilent M . grisea whole genome microarray chip version 2 . 0 using the single channel hybridization design . Quality of RNA was determined with Agilent Bioanalyzer . Five hundred nanograms of total RNA was converted into labeled cRNA with nucleotides coupled to fluorescent dye Cy3 using the Quick Amp Kit following the manufacturer's instructions ( Agilent Technologies , Palo Alto , CA ) . After analyzing the quality with Agilent Bioanalyzer , Cy3-labeled cRNA ( 1 . 65 µg ) was hybridized to M . grisea 2 . 0 4×44 k microarrays . The hybridized array was washed and scanned , and the data were extracted from the scanned image using Feature Extraction version 10 . 2 ( Agilent Technologies ) . An error-weighted average signal intensity of two probes within a chip was used for normalization with Lowess normalization module implanted in JMP Genomics software . An average expression of all probes among 16 data sets was used as the baseline . Pairwise comparison between treatments was conducted to get the expression profiles of each probe . Genes were regarded as differentially expressed if their average signal intensity among 4 replicates was above 20 in a minimum of one condition and expression ratio is greater than 2 fold with P<0 . 05 ( Student's t-test ) . The two commonly used motif discovery programs , MEME [39] and MDScan [40] , were used to identify the MoCRZ1 binding motif . Input data consisted of the exact binding sequences retrieved from the promoters of 83 genes with differential expression in the WT/Δmocrz1 comparison ( Figure 3A ) . Candidate motifs from both algorithms were manually interrogated and enumerated to identify the 3 top candidates , and queried against the yeast motif database using TOMTOM [41] . Enrichment was calculated over the 106 background sequence set which was randomly retrieved from intergenic region of the whole genome . Consensus motif sequences were calculated using WebLogo server at http://weblogo . berkeley . edu [62] . Protein expression vector was constructed by ligation of MoCRZ1 cDNA encompassing full ORF into pGEX-6P-1 ( Invitrogen , Carlsbad , CA ) having GST tag at the N terminus . The resulting construct was transformed into the E . coli strain BL21 ( DE3 ) pLysS ( Novagen ) after verifying the cDNA sequences . Induced protein was purified with GST agarose beads ( Sigma ) based on the procedures of Frangioni et al . [63] . Probe DNA was prepared by PCR with Biotin labeled primer at the 5′ end , followed by gel purification . Cold probe was amplified with the same primer sequence without Biotin labeling . Primer sequences were listed in Table S5 . Binding of putative motif sequences to MoCRZ1 protein was performed using LightShift Chemiluminescent EMSA kit following the manufacturer's manual ( PIERCE , Rockford , IL ) . Reaction mixtures containing 10 ng of purified MoCRZ1 and biotin labeled probe were incubated for 20 min . at room temperature . The reactions were electrophoresed on 5% polyacrylamide gel in 0 . 5×TBE , and transferred to a positively charged nylon membrane ( Hybond N+ , GE Healthcare ) . Signals were detected using Chemiluminescent Nucleic Acid Detection Module ( PIERCE ) according to the manufacturer's instruction .
|
All organisms have the innate ability to perceive their environment and respond to it , largely through controlling gene expression . Tailored specificity of a response is primarily achieved through signal cascades involving unique receptors , downstream transcription factors ( proteins that bind to DNA to regulate gene expression ) , and the genes these transcription factors regulate . For fungal plant pathogens , signal transduction cascades are involved in perception of hosts , transgression of physical barriers , suppression or elicitation of host defenses , in vivo nutrient acquisition , and completion of their life cycle . We know that the Ca2+/calcineurin signaling pathway is a central conduit regulating these aspects of the life cycle for fungal pathogens of plants and animals . In this study , we used advanced ChIP-chip and microarray gene expression technologies to identify the genes that the Ca2+/calcineurin responsive transcription factor MoCRZ1 directly binds to and regulates the expression of . Our findings show conservations and divergence in this pathway within the fungal kingdom . It also identifies points of control in the pathway that were previously unidentified . Most importantly , this study implicates this pathway in the establishment of host associations and virulence for the causal agent of rice blast disease , Magnaporthe oryzae , the most important disease of rice worldwide .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"microbiology/plant-biotic",
"interactions",
"plant",
"biology/plant-biotic",
"interactions",
"genetics",
"and",
"genomics/gene",
"expression",
"genetics",
"and",
"genomics/functional",
"genomics",
"infectious",
"diseases/fungal",
"infections",
"genetics",
"and",
"genomics/gene",
"function"
] |
2010
|
Combining ChIP-chip and Expression Profiling to Model the MoCRZ1 Mediated Circuit for Ca2+/Calcineurin Signaling in the Rice Blast Fungus
|
Compensatory mutations between protein residues in physical contact can manifest themselves as statistical couplings between the corresponding columns in a multiple sequence alignment ( MSA ) of the protein family . Conversely , large coupling coefficients predict residue contacts . Methods for de-novo protein structure prediction based on this approach are becoming increasingly reliable . Their main limitation is the strong systematic and statistical noise in the estimation of coupling coefficients , which has so far limited their application to very large protein families . While most research has focused on improving predictions by adding external information , little progress has been made to improve the statistical procedure at the core , because our lack of understanding of the sources of noise poses a major obstacle . First , we show theoretically that the expectation value of the coupling score assuming no coupling is proportional to the product of the square roots of the column entropies , and we propose a simple entropy bias correction ( EntC ) that subtracts out this expectation value . Second , we show that the average product correction ( APC ) includes the correction of the entropy bias , partly explaining its success . Third , we have developed CCMgen , the first method for simulating protein evolution and generating realistic synthetic MSAs with pairwise statistical residue couplings . Fourth , to learn exact statistical models that reliably reproduce observed alignment statistics , we developed CCMpredPy , an implementation of the persistent contrastive divergence ( PCD ) method for exact inference . Fifth , we demonstrate how CCMgen and CCMpredPy can facilitate the development of contact prediction methods by analysing the systematic noise contributions from phylogeny and entropy . Using the entropy bias correction , we can disentangle both sources of noise and find that entropy contributes roughly twice as much noise as phylogeny .
In the course of evolution , proteins are under selective pressure to maintain their function and correspondingly their structure . A possible mechanism to maintain structural integrity is the compensation of deleterious mutations between residue pairs in physical contact , known as compensatory mutations: Upon the mutation of one residue the contacting residue has an increased probability to mutate into a residue that will locally restabilize the protein structure , for instance by regaining a lost interaction between them . In multiple sequence alignments ( MSAs ) of related proteins , this effect leads to correlations between columns of residues in contact among most protein family members [1–4] . Many of these correlations are indirect , though , and arise through transitive chains of contacting residue pairs [5–8] . By applying statistical techniques that can distinguish mere correlation from direct statistical coupling of residue positions [5 , 7 , 9] , many false positive predictions could be eliminated . The adoption of this class of statistical models , known as Markov random fields ( MRFs ) , or Potts models in statistical physics , led to a breakthrough in de-novo ( template-free ) protein structure prediction: The predicted contacts proved sufficiently accurate to be used as spatial restraints to reliably predict protein 3D structures purely from sequence information [10–20] . The requirement for large MSAs for sufficiently precise predictions has limited the applicability of contact-assisted de-novo protein structure prediction , all the more because large protein families are more likely to contain at least one member whose structure has been solved and which can be used as a template for homology modelling . Therefore , most research has focused on making contact prediction reliable enough for medium-sized protein families [20–25] . The background noise effects arising in residue-residue contact prediction have been postulated to arise from three sources [5 , 26–32]: random sampling noise due to the limited number of sequences , phylogenetic noise due to the evolutionary relatedness of sequences in the MSA , and entropic noise or rather bias , which biases high-entropy columns towards higher scores . Unfortunately , the relative contribution and properties of the three different sources of noise are difficult to study in real alignments , mainly because the true values of coupling parameters are not known . In addition , the stochastic noise , entropy-dependent noise and phylogenetic noise cannot be modified independently ( for example by subsampling ) , as these noise sources are indirect , complex consequences of learning on only a limited number of sequences that are statistically dependent on each other according to their phylogenetic relationship . Many correction schemes for removing noise from the matrix of predicted contact scores have been examined [27 , 29 , 30 , 33–36] , and the average product correction ( APC ) [26] came out as a clear winner and is used in almost all recent studies . However , it is widely acknowledged in the field that our limited understanding of what noise effects APC is correcting and why it is so effectively correcting them is severely impeding progress in developing better statistical methods to predict contacting residue pairs . We repeatedly made the experience that a promising extension to the standard MRF model that considerably improved the contact prediction performance before applying the APC was doomed to failure because it inexplicably yielded worse results than the baseline method after applying APC . Based on theoretical considerations ( Material and methods ) , we propose a simple entropy bias correction ( EntC ) that is computed solely from per-column entropies of the input MSA and corrects for entropy-dependent bias without affecting noise from phylogenetic effects . We find that the EntC eliminates nearly as much noise as the APC . The observation that both corrections can be expressed as a product of two factors depending only on each column separately explains partly the success of APC and suggests that it mainly corrects for entropy noise . Whereas the APC is applied as a post-correction to the matrix of predicted contact scores , the EntC can be applied directly on the statistical couplings of the MRF model , prior to computing a contact score and other post-processing treatments . To systematically study the sources of noise limiting the accuracy of contact predictions from MSAs and to facilitate progress in the development of better contact prediction methods , we have developed CCMgen , a method for generating realistic synthetic protein sequence alignments whose residues obey the selection pressures described by a MRF with pairwise statistical couplings between residue positions . For that purpose , CCMgen requires an exact statistical model that will reliably reproduce the empirical alignment statistics , such as single-site , pairwise or even higher-order amino acid frequencies , of the input MSA that was used to learn the MRF model in the first place . A typical strategy to obtain estimates of the MRF model parameters would involve maximizing the logarithm of the likelihood function over all sequences in the MSA . However , the normalization factor in the likelihood function requires to sum 20L terms , where L is the protein length , and methods to optimize the full likelihood are very slow for realistic proteins [5 , 37–41] . The most popular approximation is to maximize the pseudo-likelihood instead of the likelihood , as it can be shown that it converges to the same solution for large numbers of samples and it is fast to compute [42–44] . Even though pseudo-likelihood maximization gives results of the same quality of predicted residue-residue contacts as those using the full likelihood optimization , several studies unveiled that the pseudo-likelihood model is inaccurate and not able to accurately reproduce the empirical alignment statistics [37 , 45] . We provide an implementation of an alternative precise inference technique , persistent contrastive divergence ( PCD ) [46] with our tool CCMpredPy . Compared to pseudo-likelihood maximization , PCD achieves identical precision for contact prediction while the inferred MRF model reproduces empirical marginals much more precisely . The increased quality of the models comes at the expense of longer run times , which are however still practical for even large proteins and alignments using a single desktop computer . High quality MRF models learned with PCD might prove beneficial beyond the purpose of contact prediction when problems require exact model statistics , e . g . when studying mutational effects or designing new protein features using the model energies . Finally , we employ CCMgen in combination with MRF models that have been learned with the PCD algorithm and our entropy bias correction to quantify the relative effect sizes of phylogenetic and entropic bias on the precision of contact prediction . We find that the contribution of entropy noise in contact prediction is on average twice as big as that of phylogenetic noise .
An exactly inferred MRF will reliably reproduce the empirical single-site and pairwise amino acid frequencies , fi ( a ) and fij ( a , b ) for all positions i , j in the MSA and all amino acids a , b ∈ {1 , … , 20} [7 , 47] . Several studies demonstrated that pseudo-likelihood maximization , while being the method of choice for contact prediction , yields models that cannot accurately reproduce the empirical alignment statistics [37 , 39 , 45] . We developed a method that uses an inference technique called persistent contrastive divergence ( PCD ) [46] to learn MRF models that accurately reproduce the empirical alignment statistics . As in the study by Figliuzzi et al . [37] , we computed for all Pfam MSAs in the PSICOV dataset the single-column and paired-column amino acid frequencies as well as covariances , cov ( δ a , x i , δ b , x j ) = f i j ( a , b ) - f i ( a ) f j ( b ) , where δa , x is the Kronecker symbol . We compared these statistics with those from sequences obtained by Markov chain Monte Carlo ( MCMC ) sampling from MRFs that were trained on the Pfam MSAs using either pseudo-likelihood maximization or PCD . We find that the empirical single-site amino acid frequencies are well reproduced by both models . But whereas the empirical pairwise amino acid frequencies and covariances correlate strongly with the corresponding statistics computed from the PCD samples , the correlation is much weaker for samples obtained from pseudo-likelihood MRF models ( Fig 1A and 1B and S1 Fig ) . Furthermore , as in Figliuzzi et al . [37] , we investigated how well the generated MCMC samples reproduce the alignment substructure of the original Pfam alignments with respect to the organisation of subfamilies in sequence space . We projected the protein sequences of the MCMC samples onto the first two principle components obtained from a principal component analysis ( PCA ) of the original Pfam MSA ( for details see S2 Text ) . Again , we find that the alignment substructure described by the grouping of sequences that can be observed in the two-dimensional PCA space , is reproduced more reliably by MCMC samples generated from PCD models than from pseudo-likelihood models ( S2b and S2d Fig ) . It has been argued that for the purpose of predicting residue contacts an approximate model such as those obtained by maximizing the pseudo-likelihood for a MRF is sufficiently accurate to infer the correct topology of the interaction network of residues [45] . Fig 1C shows the mean precision of the predicted contacts from a PCD model and a pseudo-likelihood model versus the number of predictions per columns in the MSA . The precision for one MSA is the fraction of correctly predicted contacting pairs of positions ( i , j ) out of all predicted pairs . The correctly predicted pairs ( i , j ) are those for which the Cβ − Cβ distance in the reference protein structure of the Pfam MSA is below 8Å . Residue pairs that are separated by less than six positions along the protein sequence are not considered for the evaluation as they typically correspond to contacts within secondary structure elements and reflect local geometrical constraints . Indeed , predicted contacts from a PCD model achieve equal precision as predictions from a pseudo-likelihood model . S3 Fig shows further analysis , comparing the APC-corrected contact scores from pseudo-likelihood and PCD models . However , more complex problems such as prediction of mutational effects or generating realistic samples of sequences , require exact model statistics . Several methods have been developed that exactly infer MRF models , such as bmDCA and ACE [5 , 37–41] , but they are computationally intensive which renders them impractical for real proteins . In comparison , our PCD-based CCMpredPy method is only about a magnitude slower than pseudo-likelihood maximization ( Fig 1D ) . A major obstacle for improving the statistical methods for residue-residue contact prediction is our lack of understanding of the sources of noise . The background noise effects have been postulated to arise from at least three sources , whose size and properties are difficult to quantify: phylogenetic , entropic and sampling noise . Phylogenetic noise originates from the violation of the assumption of independence of sequences in the MSA [48] . This assumption has been made by all methods that have been employed for contact prediction so far . To understand the origin of phylogenetic noise , consider the example in Fig 2 . The MSA is composed of two subtrees whose last common ancestor sequences , DSMF and ETMF , had a mutation at the second and first position respectively . All descendants of the first ancestral sequence whose first two residues have not mutated in the meantime will have a DS at first and second position , while all descendants of the other ancestral sequence whose first two residues have not mutated yet will have a ET at those positions . Therefore , pairs DS and ET are more likely than would be expected from the frequencies of D and E in the first column and of S and T in the second column . The first and second position will therefore appear to be statistically coupled even though they are not . Entropic bias describes the tight correlation of the expectation value of the contact score cij between columns i and j of a MSA under the assumption of no coupling between both columns with the product of the square roots of column entropies s i = - ∑ a = 1 20 f i a log f i a: E [ c i j ] ∝ ∼ s i 1 2s j 1 2 . ( 1 ) Put simply , higher column entropies lead to higher expected contact scores cij even if no coupling exists . To understand the origin of this bias , we need a bit of notation . From the MSA we compute coefficients wij ( a , b ) that quantify the statistical coupling between residue a ∈ {1 , … , 20} occurring in column i and residue b ∈ {1 , … , 20} in column j of the same sequence ( Materials and methods ) . A coefficient wij ( a , b ) = 0 . 1 signifies that residue a in column i and residue b in column j in the same sequence is exp ( 0 . 1 ) times more likely to occur than what would be expected if the amino acids in both columns were independent of each other . To predict contacts , we estimate the coupling coefficients wij ( a , b ) , for example by maximizing the pseudo-likelihood , and obtain estimates w ˜ i j ( a , b ) , from which we can calculate a score to predict contacts . The commonly used contact score between columns i and j of a MSA is simply the norm of the 400-dimensional vector w ˜ i j , c i j ≔ ∥ w ˜ i j ∥ = ( ∑ a , b = 1 20 w ˜ i j ( a , b ) 2 ) 1 / 2 . ( 2 ) It sums up the squared coupling coefficients over all possibly coupled amino acid pairs . Let us assume that a MSA has no statistically coupled residue pairs , meaning that the true coupling coefficients are all zero . But the estimation of the coefficients results in errors , which contribute a systematic bias , as we will now see . The regularization of the MRF will ensure that the coupling coefficients wij ( a , b ) for those amino acid pairs ( a , b ) without counts will be zero and will therefore not contribute to the overall contact score cij for this residue pair . For those pairs ( a , b ) with one or more counts , the wij ( a , b ) will be distributed around zero but will rarely be exactly zero , just as fij ( a , b ) is rarely exactly equal to fi ( a ) × fj ( b ) . So each amino acid pair ( a , b ) that occurs at least in one sequence will make a contribution E [ w ˜ i j ( a , b ) 2 ] to the sum in Eq 2 . These contributions to cij stemming from noisy estimates wij ( a , b ) create a bias that will increase with the number of pairs ( a , b ) of bins over which the N counts are distributed . Columns with high entropy tend to disperse the counts of amino acid pairs over more bins ( a , b ) than columns with low entropy . It is shown in Materials and Methods that the expectation value of this bias on cij can be approximated by a term proportional to product of the square roots of the entropies of the two columns . The factorization of the EntC into two factors depending only on each column separately explains partly the success of APC and suggests that it mainly corrects for entropy noise ( Materials and methods ) . Sampling noise on the estimated coupling coefficients would remain , even if we correct for entropic bias and phylogenetic effects , because with a finite sample of sequences we cannot estimate fractions arbitrarily accurately . For example even if the sequences could be assumed to be independent of each other , the probability of an amino acid pair ( a , b ) that has been observed n ≪ N times out of N is only estimated to a relative accuracy of approximately σ / μ = n ( 1 - n / N ) / n ≈ 1 / n , according to the standard deviation of the binomial distribution . More precisely , whereas the entropy bias describes the systematic offset of the contact score cij stemming from the non-zero expectation values E [ w ˜ i j ( a , b ) 2 ] , the sampling noise originates from the variance of the coefficients , var [ w ˜ i j ( a , b ) 2 ] , which is due to the finite number of measurements ( sequences ) N taken . Our workflow to analyse the relative contributions of noise sources is described in Fig 5 . First , we estimate the parameters of a second order MRF model with PCD using CCMpredPy for each of the 150 Pfam MSAs in the PSICOV data set . To obtain models with few but precise constraints , we set coupling parameters to zero for non-contacting residue pairs ( Cβ distance >12Å ) during parameter learning . In a second step , we use CCMgen with the learned model parameters to generate realistic synthetic MSAs of interdependent sequences with pairwise statistical couplings between some positions as they are observed in MSAs between residues in physical contact . CCMgen provides full control over the generation of the synthetic MSAs by allowing us to specify the evolutionary times and phylogeny along which the sequences are sampled . We sample two sets of synthetic MSAs: one set with a star tree topology and the other with a binary tree topology ( Fig 6 ) . Given sufficient evolutionary time , the phylogenetic dependencies between sequences drawn according to the star tree topology should be negligible , whereas sequences drawn along the binary tree are expected to contain stronger interdependencies . Because the accuracy of predictions strongly depends on alignment depth and diversity [49 , 52] , we ensured that the synthetic alignments contain the same number of sequences and have similar diversities as the original Pfam alignments ( for details see Material and methods ) . These provisions justify a direct comparison of the results for sampling sequences along the star and binary topologies . Third , we run CCMpredPy on each of the synthetic MSAs and predict residue-residue contacts by ranking the pairs according to the descending raw contact scores ( Eq 2 ) , or by the APC-corrected contact scores ( Eq 17 ) or by entropy corrected scores ( Eq 18 ) . Since we know the ground truth of which pairs are coupled from the MRF model used for generating the synthetic MSAs , we can use these alignments to investigate and quantify the effect of phylogenetic noise on the precision of residue-residue contact prediction . Fig 7A and 7B plot the mean precision of the predicted contacts from both types of synthetic MSAs versus the number of predictions per columns in the MSA . As expected , the mean precision drops as more predictions are considered and lower ranks are included . Both APC and EntC correction have a huge effect in reducing noise and increasing the precision of predictions . Both corrections give very similar results for MSAs generated with star topology trees , which are not expected to show phylogenetic noise , while the APC performs slightly better than the EntC on MSAs with binary tree topologies . This suggests that the APC corrects out a small part of the phylogenetic noise . This would be plausible because this noise source affects some positions more than others ( Fig 2 ) and would thereby also cause striping , which could be corrected by APC . We estimate the strength of the phylogenetic noise as the drop in precision between the EntC-corrected precisions on MSAs with star topology and EntC-corrected precision on the MSAs with binary tree topology ( Fig 7C ) . The strength of the entropy noise is shown in terms of the drop in precision between the EntC-corrected and uncorrected , raw contact scores , both for the star tree topology and for the binary tree topology . The contribution of entropy noise to the drop in precision is roughly two times larger than that of the phylogenetic noise .
The success of the average product correction ( APC ) ( Eq 17 ) is in part explained by three key insights: First , as we have seen , the entropy bias explains a large part of the noise in residue contact prediction . Second , as shown in Material and Methods , the EntC factorizes over columns , that is , it can be written as a product of two factors , each of which depends only on one column . Third , as we show now , the APC boils down to subtracting from the score cij an approximation to its expectation value under the null model of no couplings , if this expectation value factorizes into two terms , each of which depend only on one column , E [ c i j ] ≈ u i u j . ( 4 ) Taken together , these three insights explain why the APC includes the EntC , which corrects for most of the bias . To demonstrate the third insight , we approximate c i • = ∑ j = 1 L c i j ≈ ∑ j = 1 L E [ c i j ] ≈ ∑ j = 1 L u i u j = u i 〈 u • 〉 ( 5 ) c i • • = ∑ i , j = 1 L c i j ≈ ∑ i , j = 1 L E [ c i j ] ≈ ∑ i , j = 1 L u i u j = 〈 u • 〉 2 , ( 6 ) because the sum over L terms averages out the fluctuations around the expectation value of each term . This approximation is probably the reason why the APC works better on cij than on c i j 2 because for c i j 2 the values in the sum are much more dispersed and dominated by one or a few terms , which renders the above approximation much less accurate . The APC correction is then c i • c • j c • • = u i 〈 u • 〉 〈 u • 〉 u j 〈 u • 〉 2 = u i u j = E [ c i j ] . ( 7 ) Hence the APC subtracts approximately the expectation value from cij if it factorizes over columns . As we have seen , the success of the average product correction ( APC ) ( and other denoising techniques such as LRS [33] ) depends on the specific form of the bias it can correct . The combination of pseudo-likelihood maximization , L2 regularization , and the definition of the contact score as the norm of the coupling vector ∥ w ˜ i j ∥ lead to a factorized form of the entropy , the leading cause of bias to correct for . It is plausible that changing the statistical model , its method of optimization , the regularization , or the contact score will usually result in the entropy bias to not factorize any more . For example , exchanging the L2 regularization by an L1 regularization destroys the factorization property . Therefore , even though the latter regularizer might work better , it can still perform worse after APC because APC does not correct its entropy bias well any more . A potentially very valuable result of this work is therefore the insight into what the APC actually corrects . If we can work out the expectation value of the contact score under the L1 regularization , for example , we could apply the appropriate entropy bias correction specifically for that model and regularization . As another example , consider the following contact score , which uses amino acid pair-specific weights βab to upweight those pairs that are more predictive of contacts than others: c i j ′ = ∑ a , b = 1 20 β a b w i j ( a , b ) 2 . ( 8 ) The expectation value of this score does not factorize into separate terms for i and j any more and therefore the average product correction fails . Similarly , even neural networks would have a hard time to combine the coupling coefficients wij ( a , b ) 2 while learning to subtract the correct expectation value at the same time . This explains why it has been so difficult to improve on the popular combination of L2 regularization and contact score c i j = ∥ w ˜ i j ∥ in combination with the APC . But by subtracting the correct expectation value for each pair ( a , b ) , E [ w i j 2 ( a , b ) ] ≈ N 2 λ w 2 ( N - 1 ) f i a ( 1 - f i a ) f j b ( 1 - f j b ) ( 9 ) we should be able to overcome this roadblock . For instance we can now define a score with weights βab whose expectation value under the null model is near 0 , as it should , s i j = ∑ a , b = 1 20 β a b ( w i j ( a , b ) 2 - E [ w i j 2 ( a , b ) ] ) . ( 10 ) This equation allows for the correction of individual couplings wij ( a , b ) . It could therefore be used to train deep neural networks directly on the EntC-corrected coupling coefficients wij ( a , b ) , combining the advantages of entropy correction with learning directly from the full set of coupling coefficients [21 , 24] instead of only from their EntC-corrected norms ∥wij∥ , as given in Eq 3 . Pseudo-likelihood maximization is the state-of-the-art inference technique for MRF models in contact prediction . Whereas the approximate nature of the model is sufficient for the correct ranking of residue pairs , the model is not exact in a way that it can reliably reproduce the empirical amino acid statistics of the original MSA . We implemented an alternative inference technique for MRFs , known as persistent contrastive divergence ( PCD ) which yields similar precision for predicted contacts but permits learning the fine statistics of the MRF model with higher precision . Even though other accurate model inference methods such as ACE [39] or bmDCA [37] can infer model parameters up to arbitrary precision , they are computationally intensive and their applicability is limited to small proteins . On the PSICOV dataset , our open source Python implementation of the PCD algorithm , CCMpredPy , was only about seven times slower than pseudo-likelihood maximization . ( Its speed is proportional to the number of Markov chains and thereby depends on the required accuracy . ) CCMpredPy might therefore be of use for large-scale studies that require exact models , such as investigating mutational effects or designing new protein features . We developed CCMgen , the first tool for generating realistic MSAs of protein sequences for a given phylogenetic tree whose residues follow the pairwise coupling constraints from a Markov random field model . CCMgen provides full control of parameters that determine the interdependencies between sequences through the specification of the phylogenetic topology and the evolutionary rate of the sampling process . It enables to distinguish different sources of noise observed in alignments and how they affect the performance of residue-residue contact predictions . We believe CCMgen will prove to be useful for improving and validating contact prediction methods . In this study , we demonstrated how CCMgen can be applied to analyse the noise contributions from entropy and phylogeny . Given MRF models learnt on real MSAs , we generated synthetic MSAs with statistically coupled amino acid columns from two types of phylogenetic trees , one in which the sequences are maximally independent ( star topology ) and one in which the statistical dependences are much stronger ( binary tree ) . By predicting contacts from the two types of synthetic alignments and correcting the predicted contacts either with the APC or with our proposed entropy bias correction , we were able to elucidate the effect of phylogenetic and entropic noise on contact prediction accuracy . According to the quantification of noise effects , the most important goal for residue-residue contact prediction is an accurate treatment of entropic bias , as it accounts for roughly twice the amount of correctable noise and is especially important for correctly identifying the strongest evolutionary couplings . However , phylogenetic noise has an important contribution to the predictions and only a fraction of it is probably corrected by the popular average product correction ( APC ) . This result shows that it might be very worthwhile to develop methods for contact prediction and for learning of MRFs that can explicitly take the statistical dependencies of sequences by common descent into account .
To predict contacts between residues , a popular approach is to train a Markov random field ( MRF ) model describing the probability to observe a sequence x = ( x1 , … , xL ) of length L with xi ∈ {1 , … , 20} representing the 20 amino acids , p ( x | v , w ) = 1 Z exp ( ∑ i = 1 L v i ( x i ) + ∑ i < j L w i j ( x i , x j ) ) . ( 11 ) The couplings wij ( a , b ) describe the preference to find amino acid a at position i and b at j in the same sequence in relation to the probability if these positions were independent , as parametrized by the single-column amino acid preferences vi ( a ) . Z is the normalization constant , equal to the sum of the exp function in the numerator summed over all possible 20L sequences . To estimate the parameters vi ( a ) and wij ( a , b ) of the MRF , the logarithm of the likelihood for all sequences in the MSA , equal to the sum over the log-likelihood of each sequence xn , could be maximized: ∑ n = 1 N log p ( x n | v , w ) → max . A regularization term that pushes all parameters towards zero needs to be added to prevent overtraining , most commonly a L2 penalty , R ( w ) = - 1 2 λ ∑ i < j L ∑ a , b = 1 20 w i j ( a , b ) 2 . ( 12 ) But the huge number 20L of terms in Z renders an exact solution infeasible for realistic protein lengths . A number of approximations have been developed for this general class of problems . The approach that has consistently been found to work best for residue contact prediction is the pseudo-likelihood approximation , in which we replace the likelihood with the pseudo-likelihood and maximize the regularized log pseudo-likelihood [42–44] , PL ( v , w ) =∏n=1N∏i:xni≠0Lp ( xni|xn , \i , v , w ) I ( xni≠0 ) PLreg ( v , w ) =∏n=1N∏i:xni≠0L1Znievi ( xni ) +∑i<jLwij ( xni , xnj ) +exp ( R ( w ) ) →v , wmax . ( 13 ) Here , xn , ∖i denotes the vector obtained from xn by removing the i’th component and Z n i = ∑ c = 1 20 exp ( v i ( c ) + ∑ j : j ≠ i L w i j ( c , x n j ) ) is a normalization constant , which can therefore be evaluated easily . The second product runs over all columns i for which xni is not a gap ( represented by a 0 ) . Once the parameters v , w are estimated from a MSA , we can predict contacts for pairs of positions i and j using their statistical couplings . The most widely used score for residue contact prediction simply takes the L2 norm ∥wij∥2 of the 20 × 20-dimensional vector wij with elements wij ( a , b ) ( Eq 2 ) [42 , 43 , 51 , 53 , 54] . In this study , we chose the regularization strength λ = 0 . 2 ( L − 1 ) [51] . Sequences in a MSA do not represent independent draws from a probabilistic model . To reduce the effects of redundant sequences , we employ a popular sequence reweighting strategy that has been found to improve contact prediction performance . Every sequence xn of length L with n ∈ {1 , … , N} in an alignment with N sequences has an associated weight ωn = 1/mn , where mn represents the number of similar sequences: m n = ∑ m = 1 N I ( Id ( x n , x m ) ≥ 0 . 8 ) , ( 14 ) Id ( x n , x m ) = 1 L ∑ i = 1 L I ( x n i = x m i ) . ( 15 ) An identity threshold of 0 . 8 has been used for all analyses . Amino acid counts and frequencies are computed with respect to the sequence weights . For example , f i ( a ) = ∑ n = 1 N ω n I ( x n i =a ) / ∑ n = 1 N ω n ( 16 ) is the weighted fraction of sequences that have an amino acid a in column i . We treat gaps as missing information and not as a 21st character . An example is Eq 13 , where the second product runs over all MSA columns i except those having a gap in sequence n , xni = 0 . This gap treatment leads to very minor changes both to the results and to the equations with respect to treating gaps as 21st character , e . g . the weighted number of sequences ∑ n = 1 N ω n gets replaced by N i = ∑ n = 1 N ω n I ( x n i ≠ 0 ) ( the summed weight of sequences that do not contain a gap at positions i of the MSA ) , or by N i j = ∑ n = 1 N ω n I ( x n i ≠ 0 , x n j ≠ 0 ) . See , for example , Eqs 24 and 31 ( for details see subsection 3 . 7 . 2 of PhD thesis of Susann Vorberg , available from soeding@mpibpc . mpg . de ) . The APC subtracts from each score cij = ∥wij∥2 the product of the average score ci• for row i times the average score cj• for column j divided by the average score c•• over all cells [26]: c i j APC = c i j − c i • c j • c • • . ( 17 ) The APC ensures that the average of the corrected coupling score over each column and over each row is 0 . This can be verified by summing Eq 17 over all i or j . The assumption made is that , since each residue is only in contact with a small fraction of all residues , the mean coupling score over a column or row is dominated by the systematic score bias on all pairs in the column or row rather than by the coupling scores on a small fraction of contacting residues . APC can also be interpreted as an approximation to the first principal component of the raw contact matrix [33] . It therefore removes the highest variability in the raw contact matrix that is assumed to arise from background biases . We define the following entropy bias correction ( EntC ) , which depends solely on the per-column entropies of the MSA from which the MRF was trained: c i j EC = c i j - α s i 1 2s j 1 2 ( 18 ) where α is a coefficient determining the strength of the correction , and s i = - ∑ a = 1 20 f i ( a ) log 2 f i ( a ) ( 19 ) is the entropy of column i . We determine α by analytically minimizing the sum of squares of the corrected off-diagonal coupling scores , ∑ i ≠ j L ( c i j - α s i 1 2s j 1 2 ) 2 → min α , ( 20 ) By setting the derivative to zero we obtain the optimal α value , α = ∑ i ≠ j L c i j s i 1 2s j 1 2 ∑ i ≠ j L s i s j . ( 21 ) We also investigated other correction strategies using entropy statistics computed from the input MSA , such as the joint entropy for pairs of columns or different exponents in Eq 18 . The resulting variations of the entropy correction performed comparably regarding the average correlation with APC as well as precision of contact predictions . We are given an MSA under the model that the sequences evolved under no pair couplings , that is , wij ( a , b ) = 0 for all columns i , j and all amino acids a , b . The square of the coupling score for columns i and j is c i j 2 = ∑ a , b = 1 20 w ˜ i j ( a , b ) 2 , where w ˜ i j ( a , b ) are our estimates of the coupling coefficients learnt by maximizing the regularized pseudo-likelihood PL ( v , w ) in Eq 13 . Our task is to calculate the expectation value of the coupling scores c i j = ( ∑ a , b = 1 20 w ˜ i j ( a , b ) 2 ) 1 / 2 . This expectation value under the null model of no couplings will be subtracted from the score to obtain the entropy-corrected score . For simplicity , we first assume that all sequences are independent draws from an MRF ( with zero pair couplings ) . From Eq 13 we derive the logarithm of the regularized pseudo-likelihood , P L Lreg ( v , w ) = ∑ n = 1 N ∑ i : x n i ≠ 0 L ( v i ( x n i ) + ∑ j : j ≠ i L w i j ( x n i , x n j ) - log Z n i ( v , w ) ) - λ 2 ∑ i ≠ j L ∑ a , b = 1 20 w i j ( a , b ) 2 . ( 22 ) At the local and global optimum , its partial derivatives with respect to the coupling coefficients must vanish: ∂ P L Lreg ∂ w i j ( a , b ) = ∑ n = 1 N I ( x n i =a , x n j =b ) - ∑ n : x n i ≠ 0 N ( 1 Z n i ( v , w ) ∂ Z n i ( v , w ) ∂ w i j ( a , b ) ) - λ w i j ( a , b ) = 0 ∂ P L Lreg ∂ w i j ( a , b ) = nijab - ∑ n : x n i ≠ 0 N p ( x n i =a | x n , \ i , v , w ) I ( x n j =b ) - λ w i j ( a , b ) = 0 , ( 23 ) where n i j a b ≔ ∑ n = 1 N I ( x n i =a , x n j =b ) counts how often a appears in column i at the same time as b in column j . Under the hypothesis that none of the columns is coupled to any other and that the regularization λ is sufficiently strong , the estimated coupling coefficients w ˜ i j ( a , b ) will all be fairly small and scattered around zero . Therefore , the model probabilities p ( x n i =a | x n , \ i , v ˜ , w ˜ ) can be approximated by the empirical frequency f i a ≔ n i a / N = ∑ n = 1 N I ( x n i =a ) / N . Hence Eq 23 reduces to λ w ˜ i j ( a , b ) ≈ n i j a b - N N i f i a ∑ n = 1 N I ( x n i ≠ 0 , x n j =b ) . ( 24 ) Because under the null model gaps at position i occur approximately independently from b at j , ∑ n = 1 N I ( x n i ≠ 0 , x n j =b ) ≈ ( 1 / N ) ∑ n = 1 N I ( x n i ≠ 0 ) × ∑ n = 1 N I ( x n j =b ) = ( N i / N ) N f j b , we obtain λ w ˜ i j ( a , b ) ≈ n i j a b - N f i a f j b . ( 25 ) We now show that the counts nijab are distributed according to a hypergeometric distribution , p ( k =n i j a b | f i a , f j b , N ) = Hypergeom ( k =n i j a b | n , K , N ) . ( 26 ) with k = nijab| , n = Nfia , and K = Nfjb . Suppose you draw n objects ( here: sequences ) without replacement from a set of N objects , and K of these N objects have a certain feature ( here: xnj = b ) while N − K don’t . Then the probability that k out of the n drawn objects have the feature is given by the hypergeometric distribution . In our case , the subset of objects = sequences that is drawn is the set of n = Nfia sequences that have an a in column i . The number nijab of these sequences that also have the feature xnj = b is therefore distributed according to the hypergeomteric distribution . The expectation value for a variable k = nijab is E[nijab] = nK/N = N fia fjb . Therefore , the square of the coupling score cij can be expressed as c i j 2 = | | w i j | | 2 2 = ∑ a , b = 1 20 w ˜ i j ( a , b ) 2 ≈ 1 λ 2 ∑ a , b = 1 20 ( n i j a b - E [ n i j a b ] ) 2 . ( 27 ) The expectation value of the numerator is ( nijab − E[nijab] ) 2 = var[nijab] which is n K N N - K N N - n N - 1 , or , using our notation n = Nfia , and K = Nfjb , E [ c i j 2 ] ≈ 1 λ 2 ∑ a , b = 1 20 N f i a ( 1 - f i a ) N f j b ( 1 - f j b ) N - 1 . E [ c i j 2 ] ≈ N 2 λ 2 ( N - 1 ) ( ∑ a = 1 20 f i a ( 1 - f i a ) ) ( ∑ b = 1 20 f j b ( 1 - f j b ) ) . ( 28 ) Remarkably , the expectation value factorizes into a term depending only on i and one depending only on j . The factorization is in fact the reason why the APC ( Eq 17 ) works so well , since the APC subtracts a product of two terms , c i • / c • • 1 / 2 × c j • / c • • 1 / 2 , one depending only on i and the other only on j . The factors in Eq 28 are highly correlated with the column entropies si ( Fig 8A ) , so that we can write E [ c i j 2 ] ∝ ∼ N 2 λ 2 ( N - 1 ) s i s j . ( 29 ) Finally , the variance of cij is small in comparison to E [ c i j 2 ] because usually many approximately independent terms ( nijab − E[nijab] ) 2 contribute to c i j 2 such that the fluctuations around the expectation value of each such term tend to average each other out . We can therefore approximate the entropic bias as E [ c i j ] = ( E [ c i j 2 ] - var [ c i j ] ) 1 / 2 ≈ E [ c i j 2 ] 1 / 2 ∝ ∼ N 1 / 2 λ s i 1 / 2s j 1 / 2 . ( 30 ) Given that Eq 28 is a more accurate estimate for E [ c i j 2 ] than Eq 30 is for E[cij] , we were expecting better results by predicting contacting residues ( i , j ) based on a ranking by c i j 2 - α u i u j with u i = ∑ a = 1 20 f i a ( 1 - f i a ) than when we ranked by the entropy bias corrected score c i j - α s i 1 / 2 s j 1 / 2 . To our surprise , the entropy bias correction worked slightly better . Investigation of this puzzling result is left for future work . We used a regularization strength proportional to the number of residues L in the MSA , λ = 0 . 2L . Therefore , without sequence weighting , N 1 / 2 λ should be proportional to the α parameter from Eq 21 that defines the optimum strength of the entropy bias . Indeed , Fig 8B shows a tight correlation of α with N / L . We cannot expect the relationship to be strictly linear , because in our theoretical analysis we had assumed that sequences are independent and have a weight of 1 , whereas in the example of Fig 8B the coupling coefficients w ˜ i j ( a , b ) where learned from Pfam MSAs using sequence weighting . While the log likelihood function cannot be efficiently computed because of the exponential complexity of the normalization constant Z , it is possible to approximate its gradient with an approach called contrastive divergence [55] . The gradient of the log likelihood with respect to the couplings wij ( a , b ) can be written as ∂ ∂ w i j ( a , b ) [ ∑ n = 1 N ( ∑ i = 1 L v i ( x i ) + ∑ i < j L w i j ( x i , x j ) ) - log Z ] = N i j q ( x i = a , x j = b ) - N i j p ( x i =a , x j =b | v , w ) , ( 31 ) where N i j = ∑ n = 1 N ω n I ( x n i ≠ 0 , x n j ≠ 0 ) is the summed weight of sequences that have no gap in either column i or j , q ( x i =a , x j =b ) = 1 N i j ∑ n = 1 N ω n I ( x n i =a , x n j =b ) represents the empirically observed pairwise amino acid frequencies that are normalized over a , b ∈ {1 , … , 20} , and p ( xi = a , xj = b|v , w ) corresponds to the model probabilities of the MRF for observing an amino acid pair ( a , b ) at positions i and j . The empirical amino acid counts , given by Nijq ( xi = a , xj = b ) , are constant and need to be computed only once from the alignment . The marginal distributions of the MRF cannot be computed analytically as it involves the normalization constant Z . Markov chain Monte Carlo ( MCMC ) algorithms can be used to generate samples from probability distributions that involve the computation of complex integrals such as the normalization constant Z . Given that the Markov chains run long enough , the equilibrium statistics of the samples will be identical to the true probability distribution statistics . Thus , an estimate of the marginal distribution of the MRF in the gradient in Eq 31 , p ( xi = a , xj = b|v , w ) , can be obtained by simply computing the expected amino acid counts from MCMC samples . However , MCMC methods require many sampling steps to obtain unbiased estimates from the stationary distribution which comes at high computational costs . Hinton suggested contrastive divergence ( CD ) as an approximation to MCMC methods [55] . The idea is simple: instead of starting a Markov chain from a random point and running it until it has reached the stationary distribution , we run C chains in parallel , each being initialized with one of the sequences from the input MSA and we evolve them for only a small number of steps . Obviously the chains do not converge to the stationary distribution in only a few steps and the sequence samples obtained from the current configuration of the chains present biased estimates . The intuition behind CD is that even though the resulting gradient estimate from the biased samples will also be noisy and biased , it points roughly into the same direction as the true gradient of the full likelihood . Therefore the approximate CD gradient should become zero approximately where the true gradient of the likelihood becomes zero . We apply CD and generate sequence samples to estimate the marginal probabilities by evolving Markov chains which have been initialized with randomly selected protein sequences from the original Pfam MSAs for one full step of Gibbs sampling . We set the number of Markov chains to C = max ( 500 , 0 . 1N ) , with N being the number of sequences in the MSA , which seems to give a good trade-off between performance and runtime . Gibbs sampling requires updating at each sampling step all sequence positions xi with i ∈ {1 , … , L} ( L = sequence length ) . For each position , a new amino acid a is chosen according to the conditional probability p ( x i t + 1 =a | x - i t , v , w ) ∝ exp ( v i ( a ) + ∑ j ≠ i L w i j ( a , x j t ) ) . ( 32 ) This Gibbs sampling approach is known to generate samples x0 , … , xt that are distributed according to the model probability in Eq 11 [56 , 57] . Note that we do not update positions representing a gap and we thereby retain the gap structure of the initial sequence . A modification of CD known as persistent contrastive divergence ( PCD ) does not reinitialize the Markov chains at data samples every time a new gradient is computed [46] . Instead , the Markov chains are kept persistent: they are evolved between successive gradient computations . The assumption behind PCD is that the model changes only slowly between parameter updates given a sufficiently small learning rate . Consequently , the Markov chains will not be pushed too far from equilibrium after each update but rather stay close to the stationary distribution [46 , 58 , 59] . Tieleman and others observed that PCD performs better than CD in all practical cases tested , even though CD can be faster in the early stages of learning [46 , 58 , 60] . Therefore we start optimizing the full likelihood with CD and switch to PCD at later stages of learning . CCMpredPy settings for training a MRF with persistent contrastive divergence are listed in S1 Text . MCMC samples for the analysis in Fig 1 have been generated with CCMgen by evolving 10000 Markov chains by repeated Gibbs sampling as described in Eq 32 . The Markov chains , each representing protein sequences of length L ( length of protein in the PSICOV data set ) have been randomly initialized with the 20 amino acids . Since the alignment substructure is strongly impacted by the non-random distribution of gaps in the sequences ( S2a Fig ) , in a second step the gap structure of randomly selected sequences from the original Pfam alignment is copied over ( gaps represented as 21st amino acid ) . Thus , it is ensured that the sampling procedure reproduces the original alignment substructure as closely as possible ( S2b and S2c Fig ) . The number of Gibbs steps before drawing samples was set to 500 . Increasing the number of Gibbs steps to e . g . 1000 does not change the statistics of the MCMC samples , hence we can assume that the Markov chains have reached the equilibrium distribution . CCMgen settings for MCMC sampling are listed in S1 Text . Instead of evolving sequences along a linear path , the MRF model can also be used to sample protein sequences according to an arbitrary phylogenetic tree . CCMgen can simulate the evolution of sequences along any given phylogenetic tree constrained by a MRF model , such as those calculated from CCMpred for example . The user can either supply a phylogenetic tree in Newick format that has been generated by a phylogenetic reconstruction program such as FastTree [61] on a real alignment or choose between two types of idealized trees , a binary and a star-shaped topology . For these idealized trees the user can specify the number of leaf nodes and the total depth of the tree , which is the total number of mutations per position from the sequence at the root to the leaf nodes . The root sequence can either be supplied by the user or be generated by evolving an all-alanine sequence with a number of mutations ( i . e . Gibbs sampling steps according to the MRF as described in Eq 32 ) . Sequences at subsequent child nodes are generated one by one , by duplicating the sequence at the parent node and evolving the respective child node sequences each with a number of mutations proportional to the edge length . The output of CCMgen is a MSA file with the sequences at the leaf nodes of the tree . CCMgen is released as open-source python command-line application . We used the PSICOV data set that was published together with the PSICOV method [49] and which comprises MSAs for 150 Pfam domains with known crystal structures . For each Pfam MSA in the PSICOV set we first removed sequences with more than 75% gaps and columns with more than 50% gaps , similarly as in [50 , 51 , 63] , to reduce the well-known impact of gaps on the analysis .
|
Knowledge about the three-dimensional structure of proteins is key to understanding their function and role in biological processes and diseases . The experimental structure determination techniques , such as X-ray crystallography or electron cryo-microscopy , are labour intensive , time-consuming and expensive . Therefore , complementary computational methods to predict a protein’s structure have become indispensable . Over the last years , immense progress has been made in predicting protein structures from their amino acid sequence by utilizing highly accurate predictions of spatial contacts between amino acid residues as constraints in folding simulations . However , contact prediction methods require large numbers of homologous protein sequences in order to discriminate between signal and noise . A major obstacle preventing progress on the statistical methodology is our limited understanding of the different components of noise that are known to affect the predictions . We provide two tools , CCMpredPy and CCMgen , that can be used to learn highly accurate statistical models for contact prediction and to simulate protein evolution according to the statistical constraints between positions of residues as specified by these models , respectively . We showcase their usefulness by quantifying the relative contribution of noise arising from entropy and phylogeny on the predicted contacts , which will facilitate the improvement of the statistical methodology .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"taxonomy",
"medicine",
"and",
"health",
"sciences",
"immune",
"cells",
"markov",
"models",
"statistics",
"split-decomposition",
"method",
"antigen-presenting",
"cells",
"immunology",
"multiple",
"alignment",
"calculation",
"phylogenetics",
"data",
"management",
"mathematics",
"forecasting",
"phylogenetic",
"analysis",
"thermodynamics",
"research",
"and",
"analysis",
"methods",
"sequence",
"analysis",
"computer",
"and",
"information",
"sciences",
"entropy",
"sequence",
"alignment",
"bioinformatics",
"animal",
"cells",
"mathematical",
"and",
"statistical",
"techniques",
"evolutionary",
"systematics",
"probability",
"theory",
"physics",
"cell",
"biology",
"computational",
"techniques",
"database",
"and",
"informatics",
"methods",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"physical",
"sciences",
"evolutionary",
"biology",
"statistical",
"methods"
] |
2018
|
Synthetic protein alignments by CCMgen quantify noise in residue-residue contact prediction
|
Although there is tremendous interest in understanding the evolutionary roles of horizontal gene transfer ( HGT ) processes that occur during chronic polyclonal infections , to date there have been few studies that directly address this topic . We have characterized multiple HGT events that most likely occurred during polyclonal infection among nasopharyngeal strains of Streptococcus pneumoniae recovered from a child suffering from chronic upper respiratory and middle-ear infections . Whole genome sequencing and comparative genomics were performed on six isolates collected during symptomatic episodes over a period of seven months . From these comparisons we determined that five of the isolates were genetically highly similar and likely represented a dominant lineage . We analyzed all genic and allelic differences among all six isolates and found that all differences tended to occur within contiguous genomic blocks , suggestive of strain evolution by homologous recombination . From these analyses we identified three strains ( two of which were recovered on two different occasions ) that appear to have been derived sequentially , one from the next , each by multiple recombination events . We also identified a fourth strain that contains many of the genomic segments that differentiate the three highly related strains from one another , and have hypothesized that this fourth strain may have served as a donor multiple times in the evolution of the dominant strain line . The variations among the parent , daughter , and grand-daughter recombinant strains collectively cover greater than seven percent of the genome and are grouped into 23 chromosomal clusters . While capturing in vivo HGT , these data support the distributed genome hypothesis and suggest that a single competence event in pneumococci can result in the replacement of DNA at multiple non-adjacent loci .
Horizontal gene transfer ( HGT ) is a fundamental process in bacterial genome evolution [1] . In the context of infections it can provide pathogenic bacteria with ready access to crucial resistance determinants or virulence factors . Analysis of whole genome sequences ( WGS ) of multiple isolates from a single bacterial species have revealed extensive population-wide differences among strains in multiple species [2]–[4] . The differences among strains can occur through small-scale mutations affecting only a few base pairs ( substitutions , deletions , insertions ) or by HGT where DNA segments from hundreds to hundreds of thousands of bases can be incorporated from another organism's genome without the recipient being its offspring . HGT , via transformation , transduction , or conjugation , can lead to the acquisition of entirely new sequences , as well as sequences that are homologous to existing DNA . The transfer of DNA via homologous recombination ( HR ) leads to the replacement of a region of the genome of a recipient cell by the corresponding region from the donor cell [5] . HR can lead to differentiation by incorporation of identical genes containing single nucleotide polymorphisms ( SNPs ) and/or the insertion and/or deletion of entire genes and operons . In addition , HR can also increase similarity by incorporation of closely related regions [6] . Whereas only rarely are HGT events detected in Mycobacterium species [7] , they are frequently observed in naturally transformable species such as Streptococcus pneumoniae where multi-locus sequence typing ( MLST ) and theoretical modeling have indicated that recombination rates are 3 to 10 fold higher than DNA polymerase mutation rates [8] , [9] . These highly variable HGT rates imply that the relative contributions of HGT and point mutations to the genomic diversification process varies greatly amongst species . In addition to the variability in HGT rates , there may also be extensive disparity of the sizes of genome regions that are replaced . While it is generally assumed that HR involves mainly small regions , experimental work and in silico comparison of multiple WGSs of S . agalactiae demonstrated that HR can transfer DNA segments of several hundred kilobases [10] . However , the tempo , pattern and relevance of HGT to bacterial strain evolution within natural chronic infections remains poorly understood , with there being only one published study in Helicobacter pylori [11] . In this regard , S . pneumoniae is ideal for studying the evolution of bacterial genomes in real time because it forms persistent polyclonal biofilms on the mucosal surfaces of the nasopharynx and the middle-ear . Such environments are highly conducive to HGT [12] . S . pneumoniae is a gram-positive bacterium , commonly referred to as pneumococcus , which is causatively associated with severe invasive diseases such as meningitis and bacteremia , as well as with many mucosal diseases including pneumonia , sinusitis , and otitis media ( OM ) [13] . Worldwide , S . pneumoniae is estimated to kill annually one million children under the age of five . In Europe and the USA , S . pneumoniae accounts for at least 30% of all cases of community-acquired pneumonia admitted to hospitals , and has a case fatality rate of 10–30% [14] . Despite its pathogenic potential , S . pneumoniae is a common natural component of the human nasopharyngeal ( NP ) commensal flora . In developed countries virtually every child becomes an NP carrier of S . pneumoniae during the first year of life with a recent study of European day care centers reporting that over 95% of the children were colonized by S . pneumoniae at least once during the study with many children showing evidence of polyclonal infection [15] , [16] . There are 91 S . pneumoniae serotypes and very significant differences with regard to genic ( gene possession ) diversity and disease-inducing phenotypes both within and among serotypes [4] , [17]–[19] . Collectively WGS analyses have support the distributed genome hypothesis ( DGH ) that posits that there are many genic differences that exist among the individual strains that make up a bacterial species ( or infecting population ) . Thus , there exists a species-level supragenome ( pangenome ) that is much larger than the genome of any given strain [2] , [3] [20] , [21] . Previous studies have provided evidence in support of the DGH by demonstrating that fewer than 50% of the total number of S . pneumoniae genes that have been identified are found in any individual strain [4] . Both the intense intra-species competition within S . pneumoniae biofilms [22]–[25] and the natural capacity of S . pneumoniae to undergo transformation by the active uptake of environmental DNA embedded within the extracellular polymeric matrix of these biofilms [26] , have likely driven much of the genic diversification of this species . The DGH postulates that the same mechanisms which promote genomic plasticity at the species level also result in the in situ creation of clouds of related S . pneumoniae strains within chronically infected individuals and that these may act as a potent counterpoint to the host's adaptive immune response [27] .
As part of an influenza vaccine trial , NP sampling was performed on pediatric patients presenting with flu-like respiratory symptoms . Bacteria were recovered , isolated , typed , and frozen from these samples [28] . An 8 month-old child , patient 19 , enrolled in this study had 12 clinic visits due to rhinorrhea and/or ear infections over a 7-month period . These included a visit at enrollment and seven subsequent visits during which nasopharyngeal swabs were obtained for bacterial culture ( Table 1 ) . All of the bacterial strains recovered from this patient were typed as S . pneumoniae and could be divided into one of two MLST types: ST13 or ST2011 . As with other bacterial species , MLST-based analyses of S . pneumoniae strains enables accurate identification using the allelic profiles of seven housekeeping genes that are strongly correlated with , and indicative of , genome-wide degrees of strain variability ( ST13 and ST2011 differ in the sequence of two of these genes , specifically xpt and ddl ) [29] . The two distinct MLSTs identified within patient 19 suggest that the child was infected with at least two divergent S . pneumoniae strains . Interestingly , half of the ST13 strains were identified as being of serotype 14 , and the others were non-typeable ( Table 1 ) . The patient 19 isolates are named by their MLST type ( ST13 or ST2011 ) , followed by the visit number when they were isolated ( v1 through v13 ) . 454 Lifesciences-based pyrosequencing ( without paired end analysis ) was used to sequence six of these isolates ( Table 1 ) . A PCR-based analysis of both of the non-sequenced isolates indicates that they are clones of sequenced isolates as their gene possession profiles are identical ( data not shown ) . The complete genomic sequences of the six sequenced isolates have been deposited in GenBank and are also available at the Strepneumo database http://strepneumo-sybil . igs . umaryland . edu/ . The genomes have an average size of 2 , 070±17 Kb and a GC content of 39% ( Table 1 ) . The Microbial Genome Annotation Tools and Genome Annotation Pipeline from NCBI were used to predict and annotate the coding sequences ( CDSs ) ( http://www . ncbi . nlm . nih . gov/genomes/static/Pipeline . html ) . The average number of CDSs per strain is 2250 ( Table 2 ) . Global comparisons of the WGS of these six isolates revealed that two pairs were essentially identical ( i . e . there are four strains with two isolates each of two of the strains ) despite being sampled 23 days apart ( ST13v12 and ST13v13 ) and ∼5 months apart ( ST13v1 and ST13v10 ) ( Text S1 and Text S2 , respectively ) . Thus , it can be inferred that the first ST13 strain that was isolated persisted for at least 5 months without any detectable evidence that it was an HGT recipient . The WGS of the four genically distinct strains ( ST13v1 , ST2011v4 , ST13v6 , and ST13v12 ) were aligned using the progressive Mauve feature in the MAUVE genome alignment software [30] . To visualize the genomic differences , a similarity plot was generated from this alignment ( Fig . 1A , white areas represent areas of low conservation ) . To assess the phylogenetic relationships among these strains , a maximum likelihood tree was created from the alignment using PHYML [31] as implemented in Recombination Detection Program ( RDP ) [32] ( Fig . 1B ) . Both figures show that strain ST2011v4 is the most distant among the strains isolated from this patient , and that smaller differences also exist among the ST13 strains . To investigate the role of HGT in differentiation of these strains , we created a specially modified version of RDP ( RDP3 ) capable of comparing full-length bacterial genomic sequences and identifying recombination sequences and their breakpoints ( revision 42–2; freely available from http://darwin . uvigo . es/rdp/rdp . html ) . RDP3 implements a variety of published recombination detection methods to determine statistical evidence of recombination [32] . To avoid any sequencing artifacts , all base pairs with a sequencing quality score of less than 40 ( a probability of >1∶104 that they were incorrectly called ) were eliminated from this and all subsequent analyses . The final RDP analysis identified evidence of 16 statistically significant recombination events among the four strains ( Fig 1C , Table 3; with all detected events identified unambiguously by at least five out of seven independent recombination signal detection methods ) . This analysis suggests that 459 Kb of genomic sequence was exchanged among the analyzed strains by recombination . These segments vary in size from 0 . 4 kb to 235 Kb , with a mean size of 28 Kb and a median size of 13 Kb . Comprehensive SNP analyses were used to further investigate the differences amongst these four strains . MAUVE was used to generate a list of all 11470 SNPs ( Table S1 ) . The majority of the SNPs ( 71% ) are identical among the three ST13 strains but variable in relation to the ST2011v4 strain ( rows 3310–11474 in Table S1 ) . Nonetheless , 28% ( 3306 of 11470 ) of the SNPs show differences among one or more of the ST13 strain pairs ( first 3308 rows in Table S1 ) . To determine the relative positions of these 3306 ST13 SNPs , they were sorted based on their chromosomal placement . The sorted list was manually curated to group together SNPs that a ) share the same distribution across strains ( that is , the same strain contains the variable nucleotide ) and b ) are located within an area where there is high concentration of SNPs ( from 4 to 68 SNPs/Kb – as opposed to isolated SNP found at levels <0 . 2SNPs/Kb ) ; such groups are hereafter referred to as neighbor groups {NG} ( Fig . 1C ) . The NG breakpoints were selected as areas that demark the transition between highly conserved regions ( less then 0 . 2 SNPs/Kb ) and divergent regions ( more then 4 SNPs/Kb ) . Ninety five percent of these 3306 ST13 SNPs were organized into 23 distinct chromosomal NGs . Fourteen NGs are larger then 500 bp and nine are smaller then 500 bp ( Tables 4 and 5 respectively , a detailed list of all the SNPs and their organization into NGs is illustrated in the first 3308 rows of Table S1 ) . NG analysis suggests that HGT has led to the exchange of at least 156 Kb between strains ST13v1 and ST13v12 . The analyses with RDP3 ( statistical tool package ) and NG ( manually curated grouping of SNPs based on pattern and localization ) predict very similar recombination events ( Fig 1C , Table 4 ) . The major differences identified by these two methods are in the position of the recombination breakpoints , and thus the size of each event . The NG method is extensively curated and overall leads to the most conservative estimate ( see methods and Text S3 ) . The MAXCHI method used by RDP to infer breakpoint positions identifies breakpoints as the midpoint between the two phylogenetically informative SNPs bounding the borders of identified recombinant regions . As a result the RDP estimated bounds of the recombinant regions are less conservative than the manually curated estimates because they include numerous sites on the 5′ and 3′ ends of the regions that are identical between the identified parental sequences . Regardless of the method selected for analysis , the WGS comparisons suggests that: ( 1 ) four distinct strains were isolated from one patient; ( 2 ) these strains fall into two groups , three ST13 isolates and one ST2011 isolate; ( 3 ) the differences among the ST13 strains are grouped into multiple chromosomal regions , ( 4 ) these regions are the result of multiple recombination events . The RDP3 and NG WGS comparison methods are alignment based , and do not focus on DNA regions that include genic differences ( presence/absence of CDSs ) since these do not align to other sequences . To analyze these differences , we compared all predicted coding sequences ( CDSs ) from the six genomes . These CDSs were organized into 2250 orthologous gene clusters as described in [4] and further divided into 2077 core and 173 distributed gene clusters ( Table 2 ) . A distributed cluster was defined as any orthologous gene cluster not present in all strains , and as such represents one of the genic differences among strains ( complete list in Table S2 ) . There are a total of 126 distributed gene cluster differences between ST2011v4 and the other three strains . These include 37 genes present only in the ST13 strains , and 89 genes present only in the ST2011v4 strain ( Table S2 ) . Among the three ST13 isolates there are only 47 genic differences in total . ST13v1 differs from the other strains by 23 genes ( 18 genes present and 5 genes missing ) , while ST13v12 differs by 24 genes ( 2 genes present and 22 genes missing ) ( Table S2 ) . There were no genic differences between ST13v1 versus ST13v10 or between ST13v12 versus ST13v13 . This is consistent with the WGS comparisons that identified these isolate pairs as being nearly identical , thus corroborating the hypothesis that these clones have persisted in the patient without detectable HGT . A previous study showed that the number of genic difference between pairs of independently isolated S . pneumoniae genomes ranged from 160 to 629 [4] , suggesting that all strains isolated from this patient are more closely related than most independent isolate pairs . For comparison , the difference between a clinical strain ( D39 ) and its lab derivative ( R6 ) was 35 genes [4] , similar to the number of differences between the ST13 strain pairs . Forty six of the 47 ST13 distributed genes are grouped into three of the recombination regions; 17 belong to the type 14 capsule locus within NG5 ( RDP D ) , 5 belong to NG12 ( RDP L ) , and 24 belong to NG14 ( RDP P ) ( Table 4 , and Table S3 provide the annotations for these genes and their relative positions within the recombinant regions ) . The high degree of genic similarity shared by the CDSs within the ST13 strains , and the positioning of the genic differences within the predicted recombination regions complements the results from the WGS comparisons , suggesting that these strains diverged from each other by multiple recombination events . For a population-wide perspective , we quantified the allelic and genic differences of 22 S . pneumoniae strains , including the 6 strains in this study . The WGS of the remaining 16 has been previously published [4] . These comparisons group strains based on either their genic or allelic content , but do not account for their phylogenetic relationships since high recombination rates can abrogate genome-wide phylogenetic signals [33] . The genic distance measured between genomes was defined as the number of distributed gene clusters shared ( both strains contain the gene ) or not shared ( neither strain contains the gene ) by a given strain pair , divided by the total number of distributed gene clusters ( Fig . 2A ) [34] . The allelic distance measure was based on the variation among the core gene clusters [34] ( Fig . 2B ) . These graphs show that the ST13 strains are more closely related then most other isolates collected from independent infections . Note again that R6 is a lab derivative of D39 and therefore these strains are not independent isolates . The only other similarly closely related strains is a pair of serotype 3 ST180 strains ( OXC141 and CGSSp3BS71 ) that show allelic distances comparable to that of the ST13 strains . Importantly , other strains from the same study , all collected in the same hospital in Pittsburgh over the same time period are highly genetically variable ( asterisks in Figs . 2A and 2B ) , demonstrating that the similarity among the ST13 strains isolated from patient 19 is unlikely to be an effect of the isolation locale . These data strongly suggest that the ST13 strains are more similar than would be expected had they been isolated from independent infections . In contrast , the larger distance between ST2011v4 and the ST13 strains suggests that ST2011v4 , or a highly related strain , was acquired during an independent infection . The most likely donor and recombinant strains were selected based on RDP3 predictions , as well as SNP patterns from all four unique patient 19 strains where the recombinant regions between the donor and recombinant strains must be virtually identical . Fig . 3A shows the diagram generated using RDP3 , with NG results superimposed as numbers within the light gray boxes . Dark gray boxes under the WGS schematic are labeled with the name of the most likely donor strain and located under the schematic of the most likely recombinant strain . In a few cases where multiple options for the recombinant strain are probable , they are all represented . Results suggest that for at least 9 recombination events ( red and orange in Fig . 3A ) ST2011v4 is the most likely DNA donor , and ST13v6 and/or ST13v12 are the most likely recombinants . ST2011v4 is identical in many of the recombination segments to one or more ST13 strains ( green in Fig . 3Bi , ii ) . This suggests that either a ) ST2011v4 served as DNA donor for these recombination segments , or b ) an un-sampled strain served as a DNA donor leading to modification in one or more of the ST13 strains ( blue Fig . 3Bi , ii ) in a region where ST2011v4 is identical to a subset of the ST13 isolates ( green Fig 3Bi , ii ) . Importantly , the regions surrounding most recombination segments are virtually identical among all three ST13 strains but variable ( containing many SNPs ) relative to ST2011v4 ( yellow versus pink in Fig . 3Bi , ii ) . Table S3 displays the allelic and genic differences within the recombination fragments , as well as their surrounding areas ( labeled W or S , respectively ) , and Table S1 shows the SNPs surrounding the recombinant region highlighted in yellow ( within rows 3310–11474 ) . Two observations provide compelling evidence for the first option where ST2011v4 acted as a DNA donor to ST13 strains . The first observation is the regional genomic similarity between ST2011v4 and subsets of the ST13 strains in the recombinant region ( green Fig . 3Bi , ii ) . The second observation is the genomic identity among the ST13 strains but not ST2011v14 on the regions surrounding the recombinant fragments ( yellow versus pink in Fig . 3 Bi , ii ) . Moreover , the synteny among all four strains in and around the recombination breakpoints suggests that HR is the most likely operative mechanism . The scenario involving the least number of strains and recombination events that explain the genomic sequences isolated from this patient is illustrated in Fig . 3Ci . Here , ST13v6 evolved from ST13v1 through the acquisition of NGs 2 , 5 , and 12 from ST2011v4 ( ST13v6 and ST2011v4 are identical in these regions yet differ from ST13v1 - Fig . 3Bi ) . Using the most conservative recombination estimates , these three regions add up to 46 . 8 Kb , include 22 distributed genes , and differ by 1079 SNPs ( strain name in orange in Fig . 3A and sizes in Table 4 ) . Subsequently , ST13v12 evolved from ST13v6 through the acquisition of NGs 1 , 3 , 4 , 7 , 8 , and 11 from ST2011v4 ( ST13v12 and ST2011v4 are nearly identical in these regions yet differ from ST13v1 and ST13v6 - Fig . 3Bii ) . Using the most conservative recombination estimates , these six HGT regions sum to 37 . 8 Kb , and differ by 478 SNPs ( strain name in red Fig . 3A and sizes in Table 4 ) . While the scenario illustrated in Fig . 3Ci is the most likely explanation for the evolution of the sequenced strains , we are unable to exclude the possibility that a different , albeit less parsimonious , pattern of HGT might have yielded the observed genetic variation ( Fig . 3Cii ) . In this second scenario , ST13v1 may be a recombinant having arisen from transfer of DNA ( NGs 2 , 5 , and 12 ) from an unknown parental donor into either ST13v6 , or a highly related strain . Collectively , these data place ST13v6 as a genomic intermediate between ST13v1 and ST13v12 , since it shares 3 recombination events in common with ST13v12 ( NGs 2 , 5 , and 12 ) but lacks evidence of additional events ( NGs 1 , 3 , 4 , 7 , 8 , and 11 ) . Notably , this model of recombination events is consistent with the time of isolation of these strains . Not all of the recombination fragments can be explained by the genetic exchanges occurring among the four unique sampled strains . There are 2 recombinant regions ( NGs 9 and 14- RDP H , O and P ) that must have been acquired by ST13v12 from an unsampled donor , as these regions are unique with respect to all of the sequenced strains including ST2011v4 ( purple in Fig . 3A and 3Biii ) . NG 14 ( RDP O and P ) is ∼50 Kb , differs from the other ST13 strains by 24 distributed genes and ∼1200 SNPs ( Fig . 4 ) . Importantly , this region is also unique with respect to the un-sequenced ST13v3 and ST2011v5 strains as shown by PCR-based sequencing of the target regions , demonstrating that these strains cannot have served as donors . NG9 is also unique with respect to all of the other sequenced and unsequenced strains isolated from this patient suggesting , as for NG14 , an origin from an unsampled strain . Further differences among the strains suggest that additional HGT events affecting a smaller number of loci have also occurred during this infection . The 5′ four Kb in NG14 ( RDP N ) varies between ST13v1 and ST13v6 , and the 3′ end of NG8 varies between ST2011v4 and ST13v12 suggesting the occurrence of additional DNA exchange events at these regions . Also , there are an additional three regions where ST2011v4 may have served as a donor ( pink in Fig 3A ) . In NG10 ( RDP I ) , ST13v1 resembles ST2011v4 but not ST13v12 ( data for ST13v6 is missing in this region ) . In NG13 ( RDP M ) , ST13v6 differs from all strains while ST13v1 and ST13v12 resemble ST2011v4 suggesting that ST13v6 underwent an additional HGT event in this region with another donor , and therefore the direct antecedent of ST13v12 incurred an additional change creating the ST13v6 genome . Finally , for the majority of the SNPs in NG6 ( RDP E ) , ST13v6 differs from the other ST13 strains but resembles ST2011v4 suggesting yet another HGT event from a relative of ST2011v4 . In addition to the large ( >500 bp ) recombinant regions , we identified an additional 9 regions that are <500 bp ( Table 5 ) . These nine small NGs ( sNG ) have a variety of SNP patterns ( detailed SNPs in Table S1 ) . Finally , minor differences among the strains also implicate other mutation mechanisms . Differences in the number of repeats in the glucan binding domain of pneumococcal protein A genes is suggestive of DNA polymerase slippage , and differences in restriction endonucleases resemble DNA inversions between S subunits ( Text . S1b and Fig . S1 , respectively ) [35] . Given the polyclonal nature of infection it is not possible to correlate with a high degree of confidence the patient's symptoms observed during any particular visit to the bacterial strain isolated from samples collected at that visit . Nonetheless , it is noteworthy that the appearance of the nontypeable ( NT ) strain ST2011v4 correlates with the beginning of a severe bout of acute otitis media ( diagnosis listed in Table 1 ) . It is therefore conceivable that either this strain itself or sequences horizontally transferred from this strain into the ST13 strain may have had an influence on virulence . To investigate how the genetic differences amongst these strains may have affected their biology we compared the capacities of the ST13 strains to form biofilms . Biofilm-formation is thought to be important for persistence following nasopharyngeal colonization and for the establishment and maintenance of chronic mucosal infections such as otitis media with effusion [36] , [37] . Confocal images of biofilms produced in vitro by the ST13 strains after 1 , 3 , and 5 days of growth show that the two unencapsulated strains ST13v6 and ST13v12 make much more robust biofilms when compared to ST13v1 , the capsular type 14 strain ( Fig . 5 ) . The differences between the ST13 strains are contained within 150 genes: 46 distributed genes and 104 core genes with allelic differences that collectively contain ∼2200 SNPs ( labeled “W” in Table S3 ) . The CDS that differentiate ST13v1 from both ST13v6 and ST13v12 are located within NGs 2 , 5 and 12 ( RDP part of A , D , L ) ( Table 4 ) . NG5 ( RDP D ) on ST13v1 encoded the type 14 capsular genes as well as adjacent allelic core genes ( Table S3 ) . In the corresponding region , the non-typeable ST13v6 , ST13v12 and ST2011v4 have lost the capsular genes yet carry two genes that are not found in ST13v1 . Within NG2 ( part RDP A ) modifications to pneumococcal surface protein A ( PspA ) , a virulence gene that encodes a choline-binding protein associated with resistance to fixation of complement [38] and with binding human lactoferrin [39] , also differentiate ST13v1 from ST13v6/ST13v12 . Of note , within ST13v12 and ST13v13 there were various nucleotide polymorphisms within PspA that were not obviously derived through HGT suggesting that this locus is under strong selective pressure ( Text S1 and Table S3 ) . Within NG12 ( RDP L ) there are three distributed genes in ST13v1 surrounded by 16 allelic core genes , which can be organized into four operons , two of which include only hypothetical proteins . The observed HGT event within this region resulted in ST13v6 and ST13v12 having two genes in this region that is absent in ST13v1 . One of these genes has been annotated as a possible cell surface protein . A species-wide comparison demonstrated that neither of these genes is shared with any of the other 16 sequenced S . pneumoniae strains analyzed in Fig . 2 . The NG14 ( RDP O ) region of ST13v12 is ∼23 Kb smaller than the corresponding regions of ST13v1 and ST13v6 ( Fig . 4 ) . This region in these latter two strains contains 44 genes distributed over at least 4 separate operons . The annotations of these genes suggest that they function in the metabolism and/or transport of amino acids , sugars , zinc and glycerol ( Table S3 ) . The corresponding smaller region in ST13v12 is missing 22 of these genes , carries two other genes ( a beta-galactosidase and a hypothetical protein ) , and differs from the other ST13 strains by over 1000 SNPs in their shared genes .
S . pneumoniae has long been recognized as a major human pathogen with Sir William Osler at the turn of the 20th century referring to pneumococcal pneumonia as “the captain of the men of death” . Shortly thereafter S . pneumoniae was shown to be transformable [40] and this observation led directly to the identification of DNA as the hereditary molecule [41] . However , it has only been in the last decade that HGT has been recognized as a significant virulence trait [27] . In this study we analyzed the WGS of six S . pneumoniae isolates obtained over ∼7-months from a single pediatric patient presenting with nasopharyngeal and middle-ear disease symptoms . We recovered two pairs of virtually identical genomes over the 13 visits ( at visits 1 and 10 and visits 12 and 13 ) strongly suggesting that some strains persisted within the patient during asymptomatic periods . Moreover , the recovery of other divergent strains at visits 4 and 6 suggests that the clones isolated on visits 1 and 10 were present simultaneously with these divergent strains . Additionally , since recombination between a donor and recipient strain is much more likely if both parental strains are present simultaneously , it can be inferred that this was a polyclonal infection . This is not surprising given ample evidence for polyclonal carriage of S . pneumoniae [15] , [16] . Thus , while the available experimental samples ( single strain isolations at each of 8 visits ) did not allow us to fully survey the polyclonal nature of this S . pneumoniae infection , or extract the strains present during non-symptomatic periods , the complete genomes of the strains that we did isolate provide evidence of polyclonality and strain persistence in this patient . Collectively , these strains provide an excellent study set for characterizing in vivo HGT during polyclonal nasopharyngeal S . pneumoniae infection . The very high degree of similarity amongst the three ST13 strains relative to all other S . pneumoniae genomes that have been sequenced suggests they had diversified mainly by HGT in amongst the strains that were present within the studied infection . The most parsimonious explanation for the nucleotide patterns within the recombinant regions suggests that ST2011v4 acted as an extensive DNA donor during both the genesis of ST13v6 from ST13v1 and the subsequent generation of ST13v12 from ST13v6 ( or ST13v6-like strains ) . We were however , unable to exclude the possibility that a different , albeit less parsimonious , pattern of HGT might have yielded the observed genetic variation . Regardless of the actual recombination pathways , it was also very clear that not all of the observed exchanges could be explained by the genetic exchanges between the four sampled strains . In at least two recombination events detectable within the ST13v12 genomes sampled on visits 12 and 13 it is apparent that divergent sequences have been derived from an unsampled donor-parental lineage . Experiments with commensal populations of streptococci in the upper respiratory tract show that a few clones tend to dominate , thus it is not far fetched to suggest that there may have been multiple uncultured and undetected S . pneumoniae strains present within this patient during the study period [42] . The synteny in and around the recombination breakpoints , as well as the absence of phage-related sequences largely rules out transduction and suggests that HR is the most likely operative mechanism . Our detection methods are not able to differentiate between DNA acquired by conjugation or transformation . However , given that the S . pneumoniae are naturally competent , it is most likely that DNA from lysed bacteria enters the cells during bouts of competence , and is incorporated into the genome via homologous recombination . It is possible that each one of these 23 loci that differentiate the ST13 strains resulted from a separate HGT event and that the differences accumulated one at a time over the entire study period . Alternatively , given that at least six loci ( NGs 1 , 3 , 4 , 7 , 8 , and 11 ) were most likely all acquired by ST13v12 from ST2011v4 it is also conceivable that these regions may have been exchanged during the same competence event that supported multiple homologous recombinations . The same simultaneous multiple replacement mechanism could also be used to explain the formation of ST13v6 , which differs by at least three loci ( NGs 2 , 5 , and 12 ) from its most likely predecessor , ST13v1 . The possibility of a single competence event resulting in the replacement of multiple loci warrants further investigation as it has not been previously explored given that S . pneumoniae recombination studies have been limited to a subset of genes in population-wide studies [8] , [33] . If in fact , multiple loci were simultaneously replaced , the selection for one transferred gene or allelic difference would be enough to explain inheritance of the multiple replacements after a competence event . Ultimately , one would expect that the persistence over multiple generations of each of the HGT-acquired loci would depend on its adaptive value . Our whole-genome comparisons indicate that at least 156Kb of S . pneumoniae strain's genomic content was exchanged during multiple HGT events involving multiple potential donors over a seven-month period . Given that the average ST13 genome is ∼2 megabases , this corresponds to ∼7 . 8% of the genome being replaced . Experiments in biofilm-grown S . mutans cells show they were transformed 10- to 600-times more frequently than their planktonic counterparts [43] , suggesting that , for any given polyclonal population of bacteria , rates of recombination could be much higher in the context of a chronic biofilm infection than they would be in an analogous acute infection . The extent of homologous recombination in the S . pneumoniae population , estimated using theoretical models informed by MLST data ( involving 6 or 7 housekeeping genes ) suggests that recombination can generate new alleles ∼3–10 times more frequently than DNA polymerase errors [8] , [9] . Another S . pneumoniae study that investigated recombination breakpoints on the seven MLST housekeeping genes led to the suggestion that some strains may have very high recombination rates -i . e . they are so-called hyper-recombinants [44] . Our whole genome data clearly support the notion that S . pneumoniae evolution is characterized by extremely high rates of recombination . The unprecedented degree of HGT detected here within strains isolated from a single infection is strongly supportive of the distributed genome hypothesis [3] , [27] , [45] . As is the case with highly recombinogenic viral pathogens [46] , the genetic malleability we have detected in S . pneumoniae genomes is possibly a property that this and other related bacterial species have evolved to cope with both the adaptive immunity of individual hosts and the genetic variation that exists within host populations [27] , [47] . As is the case with the continuing debate over the evolutionary value of sexuality [48] , [49] it may ultimately prove quite difficult to precisely enumerate costs and benefits of S . pneumoniae's high rate of recombination .
We obtained six clinical S . pneumoniae isolates from a pediatric patient participating in a vaccine trial at the Children's Hospital of Pittsburgh . The genomes of these strains were sequenced at the Center for Genomic Sciences ( CGS ) using a 454 Life Sciences FLX sequencer . The limitation of this sequencing method is that it may overlook a frame shift mutation when it is present within a homo-nucleotide stretch . As previously described , strains were assembled by the 454 Newbler de novo assembler and prediction of putative coding sequences and gene annotations were done by NCBI using the Microbial Genome Annotation Tools and Genome Annotation Pipeline [4] . The final assemblies have been deposited in GenBank , the accession numbers are: ABWQ for ST13v1 , ADHN for ST2011v4 , ABWB for ST13v6 , ABWA for ST13v10 , ABWU for ST13v12 , and ABWC for ST13v13 . The annotations prefixes are as follows: ST13v1-CGSSp14BS292 , ST2011v4-CGSSpBS455; ST13v6-CGSSpBS457; ST13v10-CGSSpBS458 , ST13v12-CGSSpBS293; and ST13v13-CGSSpBS397 . The multiple contigs from the final assembly of each genome were concatenated into a single fasta file using a combination of the Mauve Contig Mover utility of MAUVE 2 . 3 and manual rearrangements . The single fasta sequence of all 4 genomes was aligned using the progressive Mauve function from the MAUVE 2 . 3 package available at: http://asap . ahabs . wisc . edu/[30] . A mapping of the contigs from the final assembly available in GenBank onto the whole genome alignments is available in Supplementary Table S4 . A plylogenetic tree for the four genotypes from the same patient was constructed using the WGS alignment generated by Mauve , by maximum likelihood ( using Phyml ) [31] with optimal model and parameter selection carried out in RDP3 ( revision 42–2; available from http://darwin . uvigo . es/rdp/rdp . html ) [32] . A complete description of the algorithms used to create the orthologous clusters is given by Hogg et al [3] . SNPs from the whole genome sequence were identified using the tab-delimited SNP file produced by MAUVE 2 . 3 . A specially modified version of RDP3 ( revision 42–2; available from http://darwin . uvigo . es/rdp/rdp . html ) capable of analyzing full-length bacterial genomes was used to identify signals of recombination and characterize specific detectable recombination events . An initial exploratory screen with 2 independent recombination signal detection methods in primary exploratory mode ( RDP and MAXCHI ) [50]–[51]; was followed up with a confirmatory screen with five additional methods ( GENECONV , CHIMAERA , SISCAN , RECSCAN and 3SEQ [32] , [51]–[54] . Other than RECSCAN and SISCAN window size settings being adjusted from their default settings to 10000 nucleotides , RDP window size settings being adjusted to 30 nucleotides , and sequences being analyzed as though linear , default settings were used throughout . Only recombination signals identified by five or more out of seven different recombination detection methods were accepted as evidence of recombination . In all cases the most probable position of recombination breakpoints was inferred with the MAXCHI method ( which is the most accurate breakpoint detection method amongst the seven non-parametric methods implemented in RDP3 ) . Phylogenetic trees were constructed from aligned regions bounded by identified recombination breakpoints . These were compared in RDP3 with phylogenetic trees constructed with the full genome alignment . Recombinant sequences were identified manually as the sequence that showed greatest positional shift with respect to the other sequences analyzed . The RDP3 inferred recombination breakpoints are at the center of the two most terminal SNPs at each of the 5′ and 3′ edges of identified recombinant regions . The NG-inferred recombination breakpoints are at the most terminal SNPs at each of the 5′ and 3′ edges . As a consequence of this , the NG analysis yields a more conservative estimate of the size of recombinant regions then RDP3 by requiring that: 1 ) the vast majority of the SNPs in the recombination fragment have the same distribution pattern across the ST13 strains , and 2 ) the recombination edges exclusively contain a high concentration of SNPs ( in the majority of cases the last three SNPs fitted into a 500 bp region , Table S1 ) . 22 S . pneumoniae strains were compared using genic and allelic difference-based graphs . Genic distances between genomes were calculated as the total commonality between strain subsets of distributed genes divided by the total number of distributed genes . Commonality included the case where both genomes either contained the distributed gene , or did not contain a given distributed gene . The commonality number was then subtracted by one to give the distance metric between two genomes [34] . Allelic distance measures between genomes are directly proportional to the percent identity among all the 1405 core alleles . The distance metrics were used to create a neighbor joining tree using the PHYLIP package ( Version 3 . 69 ) [55] . The Fig tree package ( Version 1 . 3 . 1 ) was used to visualize the tree using a midpoint root ( freely available from http://tree . bio . ed . ac . uk/software/figtree/ ) . The full sequence of the seven house-keeping genes used for MLST typing were obtained from the whole genome sequences of the six sequenced strains and by Ibis T-5000 universal biosensor technology for the two unsequenced strains [56] . The internal fragments required for typing were trimmed as directed by the S . pneumoniae MLST site , and were submitted to this site to determine their ST type ( http://spneumoniae . mlst . net/ ) . The strain serotypes were determined by two methods: ( 1 ) the Pneumotest-Latex kit ( Statens Serum Intitut ) , and ( 2 ) a PCR-based approach [57] . S . pneumoniae biofilms were grown and visualized as previously described [26] . The study was approved by the Children's Hospital of Pittsburgh Human Rights Committee . They recruited healthy children aged 6 to 24 months from the hospital's primary care center and from the community at large . Research personnel informed parents in the primary care center about the study , and advertisements were placed on the radio and in the regional newspaper . Written informed consent was obtained from the parent ( s ) of each enrolled child . They excluded children who had been born prematurely or had a craniofacial abnormality; or who had or were living with persons who had any medical condition placing them at high risk of complications of influenza; or who had a neurologic disorder , a history of tympanostomy tube insertion , hypersensitivity to egg protein or thimerosal , or a febrile illness or severe respiratory illness within the preceding 48 hours [28] .
|
Bacterial infections have long been studied using Koch's postulates wherein the paradigm is that a single clone leads to a given infection . Over the past decade , it has become clear that chronic bacterial infections often do not fit this paradigm . Instead these are associated with the presence of multiple strains or species ( polyclonal ) of bacteria that are organized into highly structured communities , termed biofilms , which can persist in the body and are recalcitrant to antibiotic treatment . In addition , there is extensive evidence that bacteria can incorporate genes from neighboring bacteria into their own genomes . This process can produce new strains and is known as horizontal gene transfer . In this study , we investigated for the first time , the tempo and relevance of gene transfer among bacterial strains of Streptococcus pneumoniae during a naturally occurring chronic childhood infection . We identified extensive gene transfer among multiple infecting strains , by sequencing of isolates recovered sequentially over a seven-month period . This gene transfer may serve as a counterpoint to the host's adaptive immune response and help explain the phenomenon of bacterial persistence , since , as occurs with some chronic viral and parasitic infections , the immune system may become overwhelmed by a set of related strains .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"evolutionary",
"biology/microbial",
"evolution",
"and",
"genomics",
"computer",
"science/applications",
"genetics",
"and",
"genomics/comparative",
"genomics",
"computational",
"biology/comparative",
"sequence",
"analysis",
"infectious",
"diseases/bacterial",
"infections",
"microbiology/medical",
"microbiology"
] |
2010
|
Generation of Genic Diversity among Streptococcus pneumoniae Strains via Horizontal Gene Transfer during a Chronic Polyclonal Pediatric Infection
|
Although evidence suggests that T cells are critical for immunity to malaria , reliable T cell correlates of exposure to and protection from malaria among children living in endemic areas are lacking . We used multiparameter flow cytometry to perform a detailed functional characterization of malaria-specific T cells in 78 four-year-old children enrolled in a longitudinal cohort study in Tororo , Uganda , a highly malaria-endemic region . More than 1800 episodes of malaria were observed in this cohort , with no cases of severe malaria . We quantified production of IFNγ , TNFα , and IL-10 ( alone or in combination ) by malaria-specific T cells , and analyzed the relationship of this response to past and future malaria incidence . CD4+ T cell responses were measurable in nearly all children , with the majority of children having CD4+ T cells producing both IFNγ and IL-10 in response to malaria-infected red blood cells . Frequencies of IFNγ/IL10 co-producing CD4+ T cells , which express the Th1 transcription factor T-bet , were significantly higher in children with ≥2 prior episodes/year compared to children with <2 episodes/year ( P<0 . 001 ) and inversely correlated with duration since malaria ( Rho = −0 . 39 , P<0 . 001 ) . Notably , frequencies of IFNγ/IL10 co-producing cells were not associated with protection from future malaria after controlling for prior malaria incidence . In contrast , children with <2 prior episodes/year were significantly more likely to exhibit antigen-specific production of TNFα without IL-10 ( P = 0 . 003 ) . While TNFα-producing CD4+ T cells were not independently associated with future protection , the absence of cells producing this inflammatory cytokine was associated with the phenotype of asymptomatic infection . Together these data indicate that the functional phenotype of the malaria-specific T cell response is heavily influenced by malaria exposure intensity , with IFNγ/IL10 co-producing CD4+ T cells dominating this response among highly exposed children . These CD4+ T cells may play important modulatory roles in the development of antimalarial immunity .
Clinical immunity to malaria eventually develops in endemic populations , but only after repeated infections with significant morbidity to both individuals and their communities [1] . Studies in regions of high malaria transmission intensity have consistently shown that the incidence of severe disease decreases considerably after the first years of life , but sterile immunity ( i . e . protection against parasitemia ) develops rarely if ever [2] , [3] . Moreover , previously immune individuals may lose protection against symptomatic infection in the absence of continuous exposure [4] , [5] . The reasons underlying the slow acquisition of clinical immunity and the failure to develop sterilizing immunity are unclear , but may include parasite diversity and evasion [6] , age-related differences in immune responses [7]–[12] , and/or host immunoregulatory mechanisms induced by the parasite [13]–[19] . As the incidence of malaria continues to be high in many parts of Africa despite insecticide-treated bednets and artemisinin-based combination therapy [20]–[22] , there is a tremendous need to better understand mechanisms of immunity to malaria in naturally exposed populations . The identification of immunologic correlates of exposure and protection in naturally exposed children would significantly help with the rational design of vaccines and other malaria control interventions . Both CD4+ and CD8+ T cells have been demonstrated to play an important role in protective antimalarial immunity in mouse models [23]–[30] , and experimental challenge models in humans and mice strongly suggest that malaria-specific T cells contribute to protective immunity [31]–[36] . However , the identification of T cell correlates of immunity in field-based studies of naturally exposed humans has proven to be quite challenging . Prior studies employing cross-sectional or prospective cohort designs have found associations between cellular immune responses and protection from future malaria , including IFNγ responses to liver stage [37]–[40] and/or merozoite stage malaria antigens [41]–[44] . However , such studies may be confounded by the level of exposure to malaria-infected mosquitoes , which varies greatly within populations , leading subjects with lower exposure to be miscategorized as “protected” [45] , [46] . Because naturally acquired immunity confers relative rather than absolute protection – manifested by a gradual decline in the incidence of clinical disease - careful quantitative outcome measures are essential , but few population-based studies of natural immunity have included careful measurement of malaria incidence over time . Pathogen-specific T cells exhibit notable functional heterogeneity , largely dependent on the antigen and cytokine microenvironment encountered during activation , and measurement of a single parameter of T cell function ( i . e . IFNγ production ) may overlook others that are more critical for protection [47] . In other parasitic infections such as leishmania [48] , [49] and toxoplasma [50] , the functional phenotype of the CD4+ T cell response correlates with the success or failure to clear the pathogen . Recent observations in individuals naturally exposed to malaria suggest an important role for CD4+ T cell production of TNFα , with or without IFNγ , as a potential immunologic correlate of protection [51] . Conversely , CD4+ T cell production of the regulatory cytokine IL-10 has been implicated in modulating the severity of disease [18] , [52] and may interfere with the development of protective immunity [14] , [42] , [53] . The role of these inflammatory and regulatory cytokines in mediating protective immunity in naturally exposed children , and in determining the balance between immunopathology and chronic repeated infection , remains unknown . In this study we performed a detailed functional characterization of malaria-specific T cell responses among four-year-old children residing in a highly malaria-endemic region to determine whether naturally acquired T cell responses correlate with exposure to and/or protection from malaria . We hypothesized that CD4+ T cells producing the pro-inflammatory cytokines IFNγ and/or TNFα are associated with protection from malaria , and that T cell production of the regulatory cytokine IL-10 may interfere with the acquisition of protection . Our results suggest that the functional phenotype of the malaria-specific T cell response was heavily influenced by prior malaria exposure intensity , with CD4+ T cells co-producing IFNγ and IL10 dominating this response among highly exposed children . However , these IFNγ/IL-10 co-producing cells were not independently associated with protection from future malaria , and may be associated with increased risk .
The study cohort consisted of 78 HIV-uninfected children followed from infancy through 5 years of age ( Table 1 ) . Blood for this study was drawn at four years of age ( range 49–51 months ) , and 92% of children continued to be followed through 5 years of age . A total of 1855 incident cases of malaria were observed in this cohort through 5 years of age . All children were treated promptly with artemisinin-based combination therapy , and despite the strikingly high numbers of malaria episodes , only 4 cases of malaria were deemed “complicated” ( all based on a single convulsion ) . No cases of severe malaria ( including severe anemia ) were observed . Among children with a lower prior incidence of malaria ( <2 episodes per person year ( ppy ) between 1 and 4 years of age , n = 10 ) , 90% lived in town; whereas among children with higher prior malaria incidence ( > = 2 episodes ppy , n = 68 ) , only 7% of children lived in town . This suggests that children with the lowest prior incidence had less exposure to malaria-infected mosquitoes . Episodes of asymptomatic parasitemia were rare in this cohort ( median 1 episode per subject over the entire study period , IQR 0–4 , Table 1 ) and the incidence of malaria declined only slightly in the year following the blood draw ( from 5 . 7 to 5 . 1 episodes ppy ) , suggesting that effective clinical immunity had not yet emerged in most children . One child had symptomatic malaria ( parasitemia with a fever requiring treatment ) at the time of the blood draw , and 17 ( 22% ) had blood smears demonstrating parasitemia . To define the frequency and function of malaria-specific T cell responses , PBMC were stimulated with malaria-infected red blood cells ( iRBC ) and analyzed by flow cytometry for production of IFNγ , IL-10 , and TNFα ( Fig . 1a ) . The median frequency of malaria-specific CD4+ T cell responses producing any of these cytokines , alone or in combination , was 0 . 20% ( IQR 0 . 12%–0 . 35% ) . Among all children , frequencies of CD4+ T cells producing IFNγ ( median 0 . 16% ) and IL-10 ( median 0 . 14% ) were significantly higher than those producing TNFα ( median 0 . 04% , P<0 . 001 , Fig . 1b ) . Production of these two cytokines largely overlapped , with a median of 83% of IL-10-producing cells also making IFNγ , and a median of 71% of IFNγ-producing cells also making IL-10 . Malaria-specific production of IL-2 was tested in a subset of children ( n = 44 ) , but responses were consistently of low magnitude ( median frequency 0 . 02% , data not shown ) . At the time of the assay 17 of the 78 children had positive blood smears; however there was no significant difference in the overall frequency of malaria-specific IFNγ+ ( P = 0 . 20 ) , TNFα+ ( P = 0 . 29 ) , or IL-10+ ( P = 0 . 21 ) CD4+ T cells between children with or without parasitemia . Malaria-specific CD8 T cell responses were not observed in the peripheral blood of any of the 78 children , although this does not exclude their presence in the liver and other tissues as demonstrated by non-human primate studies [54] . The pattern of cytokine production by malaria-specific CD4+ T cells was noted to differ markedly based on children's prior incidence of malaria ( Fig . 2a–c ) . Both IL-10-producing CD4+ T cells and IFNγ-producing CD4+ T cells were present at higher frequencies among children with a higher prior incidence of malaria ( ≥2 episodes ppy ) than among those with a lower prior incidence ( <2 episodes ppy , P<0 . 001 and P = 0 . 02 , respectively , Fig . 2a ) . Most strikingly , CD4+ T cells co-producing IFNγ and IL-10 dominated the response among children with higher prior incidence , but were virtually absent among lower incidence children ( P<0 . 001 , Fig . 2b ) . Production of TNFα followed the opposite pattern , with higher frequencies of TNFα+/IL10− CD4+ T cells observed among children with lower prior incidence than among those with a higher prior incidence ( P = 0 . 003 , Fig . 2b ) . Interestingly , despite these differences in cytokine production profiles , the overall frequency of malaria-specific CD4+ T cells ( i . e . those producing any cytokine ) did not statistically differ between the higher and lower incidence groups ( P = 0 . 13 ) . We also analyzed the relationship of prior malaria incidence with the “composition” of the malaria-specific response ( i . e . the proportion of each cytokine combination amongst the total malaria-specific CD4+ T cell population ) , and found similar results . Among children with <2 episodes ppy , TNFα-producing CD4+ T cells ( including TNFα single-producers and IFNγ/TNFα double producers ) comprised a greater proportion of the malaria-specific response than among children with ≥2 prior episodes ppy , whereas in children with a higher prior malaria incidence , IL-10-producing CD4+ T cells ( including IL-10 single-producers and IFNγ/IL-10 double producers ) comprised a far greater fraction of the malaria-specific response ( P<0 . 001 , Fig . 2c ) . There was no significant difference in the proportion of IFNγ-producing CD4+ T cells between children with higher and lower incidence . These findings suggest that the functional phenotype of the malaria-specific CD4+ T cell response differs according to prior exposure , and that with more prior episodes , the overall response is more regulatory ( IL-10 producing ) and less inflammatory ( TNFα producing ) . While the data above demonstrate that there is a strong relationship between the functional phenotype of malaria-specific CD4+ T cells and prior malaria history , we wished to determine whether this phenotype was influenced by the time elapsed since the most recent malaria episode , the cumulative number of prior malaria episodes , or both , as these parameters are both logically and statistically related ( Spearman's Rho = −0 . 46 , P<0 . 001 ) . We observed a strong inverse correlation between the frequency of IFNγ+/IL-10+/TNFα− CD4+ T cells and the duration since the last episode of malaria ( Spearman's Rho = −0 . 39 , P<0 . 001 , Fig . 3d ) , with more recent malaria associated with a higher frequency of these co-producing cells , as well as a positive correlation with the total cumulative number of prior episodes ( Spearman's Rho = 0 . 23 , P = 0 . 04 , Fig . 3e ) . However , when assessed in a multivariate model , the frequency of malaria-specific IFNγ/IL-10 co-producing CD4+ T cells remained strongly associated with the duration since malaria , whereas the total prior incidence was no longer significant . Similar results were observed for total IL-10 ( Fig . 3a ) and total IFNγ-producing ( Fig . 3b ) populations , and when assessing the duration since last episode of parasitemia ( data not shown ) . Interestingly , the opposite relationship was observed between total TNFα+ producing cells and the duration since last episode of malaria , with more recent malaria associated with a lower frequency of TNFα -producing cells ( Spearman's Rho = 0 . 23 , P = 0 . 041 , Fig . 3c ) . Further , there was no significant correlation between the number of cumulative prior malaria episodes and TNFα+ producing cells . Together these data suggest that recency of malaria infection , rather than the total number of past episodes , exerts a dominant influence on the functional phenotype of malaria-specific CD4+ T cells . Similar findings were obtained when analyzing the “composition” ( i . e . the proportion of responding cells producing IFNγ , TNFα , and/or IL10 ) of the malaria-specific response and duration since last malaria infection . Protection from clinical malaria in naturally exposed individuals can be defined using a number of outcomes , including a delayed time to reinfection [37] , [38] , [41]–[43] , [51] , a decreased incidence of malaria over time [53] , and/or a decreased probability of clinical disease once parasitemic [46] . In all cases , identification of immune correlates of protection is challenging due to the difficulty of distinguishing protection from a lack of exposure to malaria-infected mosquitos [45] , [46] . To address this , we assessed the relationship between malaria-specific T cell functional subsets and protection from malaria , while adjusting for prior malaria ( duration since last episode and/or cumulative number of prior episodes ) as a surrogate measure of exposure intensity . We also evaluated potential associations with the overall prevalence of asymptomatic parasitemia , as clinical immunity to malaria is normally characterized by a transition from symptomatic to asymptomatic disease [3] . In univariate Cox proportional hazards analysis evaluating time to next episode of malaria , a higher frequency of CD4+ T cells producing any IFNγ or IL10 , or the combinations IFNγ+/IL-10+/TNFα− and IFNγ−/IL-10+/TNFα− was associated with a significantly increased hazard of malaria ( Table 2 , left columns ) . However following adjustment for surrogates of exposure intensity ( duration since last episode of malaria and/or cumulative prior malaria episodes ) in a multivariate model , none of these associations remained significant . Similar relationships were observed when we analyzed the total malaria incidence in the year following the assay in a multivariate regression model ( Table 2 , middle columns ) . However , in this analysis both IFNγ+/IL-10+/TNFα− ( IRR 1 . 40 per 10 fold increase , P = 0 . 038 ) and any IL-10-producing CD4+ T cells ( IRR 1 . 41 per 10 fold increase , P = 0 . 039 ) remained independently associated with an increased risk of malaria after controlling for duration since last malaria infection . Nearly identical results were obtained when analyzing the total composition of cytokine producing cells: both the fraction of IFNγ+/IL-10+/TNFα− and any IL10+ cells among all cytokine-producing cells were associated with increased malaria risk ( IRR 1 . 47 , P = 0 . 038 and 1 . 40 , P = 0 . 039 per 50% increase in fraction of responding cells , respectively ) . Together , these data suggest that the dominant population of malaria-specific CD4+ cells , which co-produce IFNγ and IL-10 , are not associated with protection from future malaria , and may in fact be associated with an increased risk of malaria . We next assessed the relationship of TNFα− producing CD4+ T cells with protection . In RTS/S vaccine recipients , malaria-specific CD4+ T cells producing TNFα in the absence of IFNγ or IL-2 have recently been shown to correlate with protection from malaria infection [55] . In our cohort , a greater frequency of malaria-specific CD4+ T cells producing TNFα alone ( IFNγ−/IL-10−/TNFα+ ) was associated with a significantly reduced hazard of developing malaria ( HR 0 . 31 , P = 0 . 015 per 10 fold increase ) and lower prospective incidence ( IRR 0 . 44 , P = 0 . 004 per 10 fold increase ) in univariate analysis , but in multivariate models controlling for duration since malaria and/or cumulative prior malaria episodes , these associations were no longer significant ( Table 2 ) . Interestingly , however , the frequency of malaria-specific CD4+ T cells producing any TNFα was inversely associated with the monthly prevalence of asymptomatic parasitemia , even after controlling for duration since last episode of malaria and/or cumulative prior malaria episodes ( PRR 0 . 41 per 10 fold increase , P = 0 . 011 ) . Thus , the absence of malaria-specific CD4+ T cells producing TNFα may be associated with the phenotype of asymptomatic infection . Although IL10 production by T cells was initially believed to occur predominantly within Th2 and FoxP3+ Treg CD4+ T cell subsets , it is now known that additional subsets , including cells expressing the Th1 master regulator T-bet , produce IL-10 under conditions of continuous antigen exposure [56] , [57] . We assessed transcription factor expression within the dominant population of malaria-specific IFNγ/IL-10 co-producing cells ( Fig . 4a ) and found that these cells uniformly were TBet+ and FoxP3− ( Fig . 4b–c ) . These IFNγ/IL-10 co-producing CD4+ T cells were predominantly of an early effector memory phenotype ( CD45RA− , CCR7− CD27+; Fig . 4d–e ) . CD4+ T cell IFNγ/IL-10 responses to the polyclonal mitogen PMA/Io have previously been shown to correlate with relative protection against severe malaria [52] . We therefore compared the response to iRBC and PMA/Io stimulation , and found a strong correlation between the frequency of IFNγ/IL-10 double producing CD4+ T cells following iRBC or PMA stimulation ( Spearman's Rho = 0 . 88 , P<0 . 001 , Supplemental Fig . S1 ) . As PMA/Io stimulation is thought to induce cytokine production by recently activated cells , these data suggest that this mitogen stimulates cytokine production by malaria-specific T cells that have recently seen their cognate antigen . IL-10 levels measured concurrently in plasma were significantly higher among children with parasitemia at the time of the blood draw compared with children with no parasitemia ( median 30 . 4 pg/ml vs 11 . 4 pg/ml , P = 0 . 0035 ) , consistent with prior reports [58]–[61] . Similar to IL-10 producing CD4+ T cells , plasma IL-10 strongly correlated with recent malaria ( Spearman's Rho = 0 . 30 , P = 0 . 009 , Supplemental Fig . S2a ) . However plasma IL-10 levels did not correlate with the frequency of total IL-10 producing CD4+ T cells ( Spearman's Rho = 0 . 11 , P = 0 . 35 , Supplemental Fig . S2b ) , suggesting that additional cell types , including cells of the myeloid lineage , may contribute to plasma IL-10 levels during malaria infection [19] . Immunomodulation through downregulation of antigen-specific CD4+ T cell proliferative responses has been described in the context of several chronic parasitic infections [62]–[65] , as well as chronic viral infections that result in persistent antigenemia [66] , [67] . We assessed proliferation of malaria-specific CD4+ T cells by measuring CFSE dilution following stimulation with schizont extract ( PfSE ) in a subset of children ( n = 42 ) . A significant inverse correlation was observed between malaria-specific CD4+ T cell proliferation and cumulative prior incidence ( Spearman's Rho = −0 . 39 , P = 0 . 011; Fig . 5a ) , suggesting that heavy antigen exposure may result in a proliferative defect in malaria-specific CD4+ T cells . We also observed an inverse correlation between CD4+ T cell proliferation following PfSE stimulation and the frequency of IFNγ/IL-10 co-producing CD4+ T cells ( Spearman's Rho = −0 . 31 , P = 0 . 049 ) . It has previously been suggested that IFNγ/IL-10 co-producing CD4+ T cells may play an autoregulatory role through suppression of proliferative responses in an IL-10 mediated manner [68] . We therefore assessed whether in vitro IL10 blockade would reverse the observed proliferative defect . The ability of CD4+ T cells to proliferate in response to PfSE was partially restored in 8 of 9 subjects upon blockade of IL-10 receptor alpha ( fold change 1 . 7 , P = 0 . 01 , Fig . 5b–c ) , suggesting that the CD4+ T cell proliferative defect observed in heavily exposed children may be in part due to IL-10 mediated suppression .
In this cohort of young children living in an area of very high transmission intensity in Uganda , very little evidence of clinical immunity had emerged by five years of age . In this setting , the functional phenotype of the malaria-specific CD4+ T cell response was significantly influenced by prior malaria exposure; with less prior malaria , the overall malaria-specific CD4+ T cell response was more inflammatory ( TNFα-producing ) , but with heavier exposure , the overall malaria-specific response was more regulatory ( IL-10 producing ) . To our knowledge , this is the first study to show that Th1 IFNγ/IL-10 co-producing cells constitute the dominant population of CD4+ T cells responding to malaria in heavily exposed children . Moreover , we found no evidence that these IFNγ/IL-10 co-producing cells were associated with protection from future malaria . Interest in IFNγ/IL-10 co-producing Th1 cells has increased in recent years as these cells have been found to be important regulators of the immune response to several infectious , allergic , and autoimmune diseases [18] , [49] , [50] , [52] , [56] , [69] , [70] . In a murine model of Toxoplasma gondii , IFNγ produced by these cells was shown to be required for pathogen eradication , and concomitant production of IL-10 was vital for the resolution of the inflammatory response and to prevent tissue pathology [50] . However , in a murine model of Leishmania major , co-production of IL-10 by Th1 cells prevented pathogen eradication , contributing to chronic infection [49] . These data suggest that IL-10 co-production by Th1 T cells may help prevent immunopathology , but this may come at the cost of chronic pathogen persistence [71] . IL-10 levels are increased during malaria infection [58] , [59] , [61] and this regulatory cytokine is thought to play a key role in dampening proinflammatory responses and preventing the development of severe malarial anemia and cerebral malaria [72] . In mice , Th1 cells were elegantly shown to be the major producer of IL-10 and were critical for limiting the pathology associated with malaria infection [18] . T cell production of IL-10 has also been described in reports of human malaria infection [14] , [52] , [73]–[77] . Plebanski and colleagues described a switch in production from IFNγ to IL-10 in CD4+ T cells from Gambian adults stimulated with altered peptide ligands of the circumsporozoite protein , with an associated suppression of proliferative responses in vitro [14] . T cells co-producing IFNγ/IL-10 following nonspecific PMA/ionomycin stimulation were described in the context of acute malaria infection [73] , and were also more abundant among children with uncomplicated rather than severe malaria [52] , consistent with a role in modulating inflammation . More recently , Gitau and colleagues described malaria-specific co-production of IFNγ and IL-10 following stimulation of CD4+ T cells with a variety of expressed PfEMP variants , although these co-producing cells represented a minor fraction of the total antigen-specific CD4+ T cell response [75] . The potential role that malaria-specific IFNγ/IL-10 co-producing CD4+ T cell cells play in mediating or inhibiting protective immunity in humans has not thus far been investigated [77] . We observed that CD4+ T cells co-producing IFNγ/IL-10 dominate the T cell response to malaria in heavily exposed children , and that the overall frequency and proportion of these cells among malaria-specific T cells was strongly correlated with recent exposure to malaria , more so than cumulative prior exposure . These IFNγ/IL-10 co-producing cells express T-bet , indicating that they have differentiated along the Th1 pathway . The dominance of this functional phenotype among malaria-specific T cells has not previously been reported , and may be related to the unusually high malaria exposure intensity of our cohort , as this cell population was of much lower frequency among children with <2 malaria episodes per year . Further , frequencies of IL-10–producing and IFNγ/IL10 co-producing cells were not associated with protection from future malaria after controlling for recent and/or cumulative prior malaria , but were instead associated with an increased risk of cumulative malaria in the year following the assay , although this may be due to the inability to fully adjust for the level of environmental exposure to malaria using clinical surrogates such as prior malaria incidence . We further observed that heavy malaria exposure was associated with a decreased ability of CD4+ T cells to proliferate in response to malaria antigens , and that this impaired proliferation is partially reversed by IL-10 blockade . These data are consistent with in vitro studies of recently activated IL7R− , CD25− , CD4+ T cells which co-produce IFNγ and IL-10 and limit CD4+ T cell proliferation through IL-10 dependent mechanisms [68] . In addition , prior studies have shown that IL-10 blockade increases malaria-specific IFNγ cytokine production in filaria-coinfected individuals [78] and in cord blood mononuclear cells from neonates born to mothers exposed to malaria [79] . A similar IL10-dependent functional impairment of CD4+ T cells has been described in other infections such as HIV that are characterized by chronic high-level antigen stimulation [80] , [81] . Together , these data are consistent with the hypothesis that IFNγ/IL-10 co-producing CD4+ T cells primarily function to limit the immunopathology associated with malaria infection – including cerebral malaria , anemia , and death - through autoregulation of CD4+ T cell proliferation and cytokine production . A similar role has been attributed to IL-10-producing Th1 cells in other parasitic diseases characterized by heavy continuous antigen exposure [49] , [50] , with evidence that IL-10 produced by Th1 effector cells acts through a negative feedback loop to regulate CD4+ T cell responsiveness , limiting inflammation and tissue pathology at the cost of impaired pathogen clearance [56] , [71] . It is possible that unmeasured confounders , such as helminthic co-infections , may have been unequally represented in the high and low-incidence groups , particularly as the lower incidence children were more likely to reside in town . However routine deworming was performed in all study subjects every 3–6 months , lessening the likelihood that co-infection with helminths explains our findings . Further studies are needed to determine if IL-10-producing Th1 cells contribute to pathogen persistence , and to the failure of humans to develop sterile protective immunity to malaria . In addition , we found that children with the fewest prior episodes of malaria were significantly more likely to have malaria-specific production of TNFα without IL-10 , and that the absence of this inflammatory cytokine was associated with the phenotype of asymptomatic infection . Studies in murine models have shown that TNFα plays an important role in inhibiting the development of hepatic stages of malaria [82] , [83] . Importantly , a recent study of RTS/S vaccine recipients identified antigen-specific CD4+ T cell production of TNFα as a correlate of protection in vaccinees [55] . In contrast to that study , we found no evidence of protection after controlling for prior malaria , though we did observe that asymptomatic infection was inversely associated with the frequency of TNFα producing CD4+ T cells , independent of prior malaria . Together our data suggest that production of this inflammatory cytokine may decrease with increasing cumulative malaria exposure , enabling a transition to asymptomatic infections . A notable strength of this study was the availability of comprehensive malaria clinical histories spanning from early infancy to the time of the immunologic assessment , plus one additional year thereafter , which enabled us to assess for T cell correlates of both exposure to and protection from malaria . Several prior studies have reported correlations between T cell responses or IL-10 production and protection from malaria in naturally exposed children [37] , [42] , [53] , but such studies have generally been unable to adequately account for prior malaria exposure . While we did observe associations , both positive and negative , between malaria-specific CD4+ T cells of varying functional phenotypes and the risk of future malaria , most of these associations were not significant after adjusting for recent or cumulative prior episodes of malaria , surrogates for the level of ongoing exposure to malaria-infected mosquitos . Hence the failure to account for malaria exposure intensity may lead to spurious associations with protection . Although we did not identify T cell phenotypes that were associated with protection from future malaria , this may be related to the young age of children in this cohort , as there was little evidence that clinical immunity had developed prior to 5 years of age . Future longitudinal studies examining responses in older children and adults , incorporating more precise entomological measurements of malaria exposure , are underway . In conclusion , among naturally exposed children living in a high endemicity setting , malaria-specific CD4+ T cells were present in the vast majority of children , and their functional phenotype differed greatly based on the level of prior exposure to malaria , in particular the duration of time since last infection . IFNγ/IL-10 co-producing Th1 cells dominated the CD4+ T cell response to malaria in these heavily exposed children , but were not associated with protection from future infection . These CD4+ T cells may play important immunomodulatory roles in the pathogenesis of malaria in childhood .
Samples for this study were obtained from children enrolled in the Tororo Child Cohort ( TCC ) in Tororo , Uganda , a rural district in south-eastern Uganda with an entomological inoculation rate ( EIR ) estimated at 379 infective bites per person year ( PPY ) in 2012 [20] . Details of this cohort have been described elsewhere , and the sub-study described in this report includes only HIV-uninfected children born to HIV-uninfected mothers [20] , [84]–[87] . Briefly , children in the TCC were enrolled at infancy ( median 5 . 5 months of age ) and followed for all medical problems at a dedicated study clinic open seven days a week . Monthly assessments were done to ensure compliance with study protocols and perform routine blood smears . All children were prophylactically dewormed with mebendazole every 3–6 months per Ugandan Ministry of Health guidelines [88] . Children who presented with a documented fever ( tympanic temperature ≥38 . 0°C ) or history of fever in the previous 24 hours had blood obtained by finger prick for a thick smear . If the thick smear was positive for malaria parasites , the patient was diagnosed with malaria regardless of parasite density , and given artemisinin-based combination therapy for treatment of uncomplicated malaria . Children were followed until 5 years of age unless prematurely withdrawn . Incident episodes of malaria were defined as all febrile episodes accompanied by any parasitemia requiring treatment , but not preceded by another treatment in the prior 14 days [20] . The incidence of malaria was calculated as the number of episodes per person years ( ppy ) at risk . Asymptomatic parasitemia was defined as a positive routine blood smear in the absence of fever that was not followed by the diagnosis of malaria in the subsequent seven days , and was reported as a count outcome as it was measured via monthly surveillance . The period prevalence of asymptomatic parasitemia was calculated as the number of episodes/total months observed . Written informed consent was obtained from the parent or guardian of all study participants . The study protocol was approved by the Uganda National Council of Science and Technology and the institutional review boards of the University of California , San Francisco , Makerere University and the Centers for Disease Control and Prevention . At approximately 4 years of age , ∼6–10 mls of whole blood was obtained from each subject in acid citrate dextrose tubes . Plasma was collected , and peripheral blood mononuclear cells ( PBMC ) were isolated by density gradient centrifugation ( Ficoll-Histopaque; GE Life Sciences ) . PBMC were cryopreserved in liquid nitrogen and shipped to our laboratory in San Francisco for additional studies . Plasmodium falciparum blood-stage 3D7 parasites were grown by standard methods and harvested at 5–10% parasitemia . Red blood cells infected with mature asexual stages were purified magnetically , washed , and cryopreserved in glycerolyte prior to use ( iRBC ) . Uninfected RBCs ( uRBC ) were used as controls . To assess the impact of parasite diversity on T cell responses , responses to iRBCs prepared from 4 distinct Tororo field strains were compared to iRBC prepared from 3D7 . Responses to the 4 field strains were very similar , indicating that parasite diversity does not significantly influence the T cell response magnitude ( Supplemental Fig . S3 ) . Schizont extracts ( PfSE ) for use in proliferation assays [89] were prepared by 3 freeze-thaw cycles of iRBC in liquid N2 for freezing and 37°C water bath for thawing , then resuspended in R10 media and stored at −20°C until use . Thawed PBMC were rested overnight in 10% fetal bovine serum ( Gibco ) and counted prior to stimulation with uRBC , iRBC , media , or phorbol miristate acetate/calcium ionophore ( PMA/Io ) at 1×106 cells/condition . An E:T ratio of 1∶3 was used with uRBC and iRBC [90] . Anti-CD28 and –CD49d were added for costimulation ( 0 . 5 µg/ml , BD Pharmingen ) . Brefeldin-A and Monensin ( BD Pharmingen ) were added at 6 hours of incubation at a final concentration of 10 µg/ml to inhibit cytokine secretion . At 24 hours of incubation , cells were washed , fixed and permeabilized per standard protocols ( Invitrogen/Caltag; Ebioscience fix/perm reagents used for nuclear transcription factor analysis ) . Surface and/or intracellular staining of PBMC was done with standard protocols [91] , [92] using the following antibodies for the primary analysis: Brilliant violet 650-conjugated CD4 ( Biolegend ) , PerCP–conjugated anti-CD3 , APC-H7-conjugated CD8 , PE-Cy7-conjugated IFNγ , PE-conjugated anti-IL-10 , and FITC-conjugated TNFα ( BD Pharmingen ) . Alexa 700-conjugated CD14 and CD19 , APC-conjugated anti-γδ ( Biolegend ) , and Live/dead aqua amine ( Invitrogen ) were included as exclusion gates to reduce unwanted nonspecific antibody binding when measuring antigen-specific T cell populations [93] . Additional experiments utilized Brilliant violet 421-conjugated anti-IL-2 , Brilliant violet 605-conjugated CD45RA , Brilliant violet 710-conjugated CD27 ( Biolegend ) , APC-conjugated CCR7 ( R&D Systems ) ; eFluor 660-conjugated T-bet and FITC-conjugated FoxP3 ( Ebioscience ) . Thawed PBMC were rested for one hour , washed in 10% Human AB media ( Gemini ) , and 3–6×106 PBMC were labeled with 1 ml of 1 . 25 µM 5 , 6-carboxyfluorescein diacetate succinimidyl ester ( CFSE; Molecular Probes ) for seven minutes . CFSE-labeled PBMC were incubated in 96-well , deep-well culture plates ( Nunc , Roskilde , Denmark ) at a density of 106 PBMC per well at a final volume of 1 ml for 7 days . In a subset of patients , CFSE-labeled PBMC were incubated with antigen in the presence of IL-10 receptor alpha chain ( IL-10Rα ) blocking antibody ( clone 37607; R&D Systems ) or IgG1 isotype control antibody at 10 µg/mL . Antigens tested included media , phytohemagglutinin ( PHA; 5 µg/mL; Sigma-Aldrich ) , uRBC , or PfSE at an E:T ratio of 1∶3 schizont equivalents . At day 7 cells were treated with 100 units DNase I ( Invitrogen ) in culture medium at 37°C for 10 min , washed , and stained with surface antibodies ( PerCP–conjugated anti-CD3 , APC-H7-conjugated CD8 ( BD Pharmingen ) , Brilliant violet 650-conjugated CD4 , Alexa 700-conjugated CD14 and CD19 , and APC-conjugated anti-γδ ( Biolegend ) ) before acquisition . Flow cytometry profiles were gated on CD3+ , γδ-negative lymphocytes , and 200 , 000 to 300 , 000 events were collected . Samples were analyzed on an LSR2 three laser flow cytometer ( Becton Dickinson ) with FACSDiva software . Color compensations were performed for each patient's PBMC using beads or samples single stained for each of the fluorochromes used . Data were analyzed using FlowJo ( Tree Star , San Carlos , CA ) and Pestle ( version 1 . 7 ) /SPICE ( version 5 . 3; M . Roederer , Vaccine Research Center , National Institute of Allergy and Infectious Diseases , National Institutes of Health , Bethesda , MD ) . In experiments with CFSE-labeled cells , the ratio of CFSE-lo cells following PfSE stimulation vs uRBC stimulation was calculated and reported as the proliferation fold change . The FlowJo Proliferation Platform provided additional information about the division characteristics of CD4+ T cells . To examine the effect of IL-10 blockade on proliferation , these parameters for CD4+ T cells were generated in samples that had been stimulated with PfSE plus anti-IL-10Rα and compared to the values obtained from samples stimulated with PfSE plus isotype control . Plasma levels of IL-10 were measured by dual Ab sandwich-ELISA kits , according to manufacturer's instruction ( R&D Systems , Minneapolis , MN ) . Each sample was tested in duplicate , and cytokine concentrations were calculated using a standard curve generated from recombinant cytokines . Cytokine values were expressed as picograms ( pg ) per milliliter . All statistical analyses were performed using Prism 4 . 0 ( GraphPad ) , STATA version 12 ( College Station ) , or SPICE v . 5 . 3 ( NIAID ) . Frequencies of malaria-specific cytokine producing T cells ( alone or in combination ) are reported after background subtraction of the frequency of the identically gated population of cells from the same sample stimulated with control . Background-subtracted responses were consider positive if >0 . 01% parent population [94] . Comparisons of cytokine frequencies between prior malaria incidence groups were done using the Wilcoxon rank sum test , and the Wilcoxon signed-rank test was used to compare paired data . Statistical analyses of global cytokine profiles ( pie charts ) were performed by partial permutation tests using the SPICE software [94] . Continuous variables were compared using Spearman correlation . For multivariate regression models , non-normal variables were log-transformed . To allow for nonlinear relationships between clinical exposure variables and immunologic outcomes , we fit linear splines with knots chosen to best represent observed relationships . Associations between immune parameters and time to next malaria episode were evaluated using the Kaplan-Meier product limit formula , and a multivariate cox proportional hazards model was used to adjust for surrogates of malaria exposure found to be associated with these parameters ( duration since last episode of malaria and/or cumulative episodes in the prior 3 years ) . Negative binomial regression was used to estimate associations between immune parameters and the prospective incidence of malaria in the following year ( incidence rate ratios , IRR ) and prevalence of asymptomatic parasitemia during the entire study period ( prevalence rate ratios , PRR ) , adjusting for prior malaria as above . Two-sided p-values were calculated for all test statistics and P<0 . 05 was considered significant .
|
Despite reports of decreasing malaria morbidity across many parts of Africa , the incidence of malaria among children continues to be very high in Uganda , even in the setting of insecticide-treated bednets and artemisinin-based combination therapy . Additional control measures , including a vaccine , are sorely needed in these settings , but progress has been limited by our lack of understanding of immunologic correlates of exposure and protection . T cell responses to malaria are thought to be important for protection in experimental models , but their role in protecting against naturally acquired infection is not clear . In this study , we performed detailed assessments of the malaria-specific T cell response among 4-year-old children living in Tororo , Uganda , an area of high malaria transmission . We found that recent malaria infection induces a malaria-specific immune response dominated by Th1 T cells co-producing IFNγ and IL-10 , and that these cells are not associated with protection from future infection . IFNγ/IL-10 co-producing cells have been described in several parasitic infections and are hypothesized to be important in limiting CD4-mediated pathology , but they may also prevent the development of sterilizing immunity . These observations have important implications for understanding the pathophysiology of malaria in humans and for malaria vaccine development .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[
"medicine",
"infectious",
"diseases",
"adaptive",
"immunity",
"immune",
"cells",
"clinical",
"immunology",
"immunity",
"t",
"cells",
"immunity",
"to",
"infections",
"immune",
"tolerance",
"malaria",
"plasmodium",
"falciparum",
"parasitic",
"diseases",
"immune",
"response"
] |
2014
|
IFNγ/IL-10 Co-producing Cells Dominate the CD4 Response to Malaria in Highly Exposed Children
|
Channel noise is the dominant intrinsic noise source of neurons causing variability in the timing of action potentials and interspike intervals ( ISI ) . Slow adaptation currents are observed in many cells and strongly shape response properties of neurons . These currents are mediated by finite populations of ionic channels and may thus carry a substantial noise component . Here we study the effect of such adaptation noise on the ISI statistics of an integrate-and-fire model neuron by means of analytical techniques and extensive numerical simulations . We contrast this stochastic adaptation with the commonly studied case of a fast fluctuating current noise and a deterministic adaptation current ( corresponding to an infinite population of adaptation channels ) . We derive analytical approximations for the ISI density and ISI serial correlation coefficient for both cases . For fast fluctuations and deterministic adaptation , the ISI density is well approximated by an inverse Gaussian ( IG ) and the ISI correlations are negative . In marked contrast , for stochastic adaptation , the density is more peaked and has a heavier tail than an IG density and the serial correlations are positive . A numerical study of the mixed case where both fast fluctuations and adaptation channel noise are present reveals a smooth transition between the analytically tractable limiting cases . Our conclusions are furthermore supported by numerical simulations of a biophysically more realistic Hodgkin-Huxley type model . Our results could be used to infer the dominant source of noise in neurons from their ISI statistics .
The firing of action potentials of a neuron in vivo is a genuine stochastic process due to the presence of several sources of noise [1] . The spontaneous neural activity ( e . g . the firing of a sensory cell in absence of sensory stimuli ) [2] , [3] as well as the response of neurons to stimuli cannot be understood without taking into account these fluctuations [4] . Moreover , noise can have a positive influence on neural function , e . g . by stochastic resonance [5] , [6] , gain modulation [7] , decorrelation of spiking activity [8] , enhancement of signal detection [9] , or fast transmission of noise coded signals [10] , [11] . For these reasons , reduced stochastic models of neural activity have been suggested [12]–[14] and analytical methods have been developed to calculate the statistics of spontaneous activity and the response to periodic stimuli [15]–[17] . Studying such reduced models allows to relate specific mechanisms with certain statistics of neural firing . Vice versa , analytical expressions for the firing statistics of model neurons may be used to infer unknown physiological details from experimental data . Spike-frequency adaptation is another common feature of neural dynamics that , however , is still poorly understood in the context of stochastic spike generation . Associated adaptation currents which act on time scales ranging from to seconds are ubiquitous throughout the nervous system [18] . Prominent examples of adaptation mechanisms include M-type currents , calcium-gated potassium currents ( ) , and slow inactivation of sodium currents . Functional roles of spike-frequency adaptation include forward masking [19] , high-pass filtering [20]–[22] , and response selectivity [23]–[25] . If the neuron is driven by fast fluctuations , adaptation reveals itself in the interspike interval statistics of neural firing , most prominently in the occurrence of negative correlations among interspike intervals [26]–[31] . These features can be phenomenologically captured in generalized integrate-and-fire ( IF ) models via introduction of a slow inhibitory feedback variable , either acting as a dynamic threshold or as an inhibitory conductance or current [28] , [29] , [32]–[34] or in even more simplified models [35]–[37] . In previous studies on stochastic models with adaptation , fluctuations were considered to be fast , e . g . Poissonian synaptic spike trains passing through fast synapses or a white Gaussian input current representing a mixture of intrinsic fluctuations and background synaptic input . In particular , the dominating intrinsic source of fluctuations is ion channel noise [1] , [38]–[40] . This kind of noise is not only contributed by the fast ionic conductances , which establish the spike generating mechanism , but also by the channels that mediate adaptation currents . If the number of adaptation channels is not too large , the stochastic opening and closing of single channels will contribute a fluctuating component to the adaptation current . This noise contribution , which was so far ignored in the literature , and its impact on the ISI statistics is the subject of our study . Here , we only consider the simplest adaptation channel model which corresponds to an M-type adaptation current . Our results , however , also apply to other sources of noise emerging from a slow adaptation mechanisms as , for instance , slow fluctuations in the case of calcium-gated potassium currents ( see discussion ) . In this paper , we analyze a perfect integrate-and-fire ( PIF ) model in which a population of channels mediate a stochastic adaptation current . We approximate this model by simplified stochastic differential equations ( diffusion approximation ) . For slow adaptation , we are able to show that the latter is equivalent to a PIF neuron driven by a slow external noise . As a surprising consequence , pure adaptation channel noise induces positive ISI correlations in marked contrast to negative ISI correlations evoked by the commonly studied combination of fast noise and deterministic adaptation [28] . Furthermore , the ISI histogram is more peaked and displays a heavier tail than expected for a PIF model with fast current noise . Our results for the PIF ( positive ISI correlations , peaked histograms ) are qualitatively confirmed by extensive simulations of a conductance-based model with stochastic adaptation channels supporting the generality of our findings .
For the theoretical analysis the channel model describing the dynamics of each individual channel could be considerably simplified by a diffusion approximation . As shown in Methods , the dynamics of the finite population of adaptation channels can be described by ( i ) the deterministic adaptation current and ( ii ) additional Gaussian fluctuations with the same filter time as the adaptation dynamics . Together with our integrate-and-fire dynamics for the membrane potential ( Methods , Eq . ( 35 ) ) , we obtained a multi-dimensional Langevin model that approximates the IF model with stochastic ion channels ( Methods , Eq . ( 20 ) , ( 35 ) ) : ( 1a ) ( 1b ) ( 1c ) Put differently , a finite population of slow adaptation channels ( instead of an infinite population and hence a deterministic adaption dynamics ) entails the presence of an additional noise with a correlation time ( time scale of the deterministic adaptation ) and a variance which is inversely proportional to the number of channels . The membrane potential of the PIF model is thus driven by four processes: ( i ) the base current , ( ii ) the white current fluctuations of intensity ( representing an applied current stimulus , channel noise originating from fast sodium or delayed-rectifier potassium currents , or shot-noise synaptic background input ) , ( iii ) the slow noise due to stochasticity of the adaptation dynamics , and ( iv ) the deterministic feedback of the neuron's spike train due to the deterministic part of the adaptation ( see Fig . 2 ) . In Eq . ( 1 ) , the parameter determines the strength of adaptation and is set by the duration of the action potential relative to the mean ISI ( Methods , Eq . ( 41 ) ) . To study the effect of the two different kinds of noise , we focused on two limit cases: In the limit of infinitely many channels , the adapting PIF model is only driven by white noise . In this case , Eq . ( 1 ) reads ( 2a ) ( 2b ) We call this case deterministic adaptation . A dimensionality analysis shows that the ISI statistics are completely determined by the quantities and if one assumes and as constants ( see Methods ) . Thus , decreasing and by some factor and simultaneously increasing by the same factor ( yielding a decreased firing rate ) would , for instance , not alter the statistical properties of the model . In the opposite limit , we considered only the stochasticity of the adaptation current but not the white noise . Setting we obtain ( 3a ) ( 3b ) ( 3c ) We call this case ( and the corresponding model based on individual adaptation channels ) stochastic adaptation . As shown in Methods , this case is determined by the quantities and ( assuming and as constants ) . For instance , decreasing and increasing and by the same factor ( and thereby lowering the firing rate ) would again result in an equivalent model with the same statistical properties . The fraction of open channels performs a random walk with discontinuous jumps . The direction of jumps depends on the presence of spikes , which in turn is affected by ( Fig . 2A ) . The diffusion approximation of and the separation of deterministic and stochastic components are illustrated in Fig . 2B and 2C , respectively . Although the increments of the continuous diffusion process have the same ( Gaussian ) statistics as the original discontinuous process on a time interval larger than the mean ISI , the short-time statistics is rather different ( Fig . 2A , B ) . Therefore , it is not obvious whether the diffusion approximation yields a good approximation to the ISI statistics , and in particular , how this approximation depends on the number of channels and the adaptation time constant . To clarify this issue , we performed both simulations based on individual channels ( “channel model” ) and simulations of Eq . ( 1 ) ( “diffusion model” ) . It turned out , that the diffusion approximation yields a fairly accurate approximation for the shape of the ISI density , the coefficient of variation and the serial ISI correlations even for small channel populations . However , significant deviation were found for higher-order statistics like the skewness and kurtosis of the ISIH ( see next section ) . The calculation of the ISI statistics ( histogram and serial correlations ) of the PIF model with noise and spike-frequency adaptation is generally a hard theoretical problem . Here we put forward several novel approximations for the simple limit cases Eq . ( 2 ) and Eq . ( 3 ) . For typical adaptation time constants that are much larger than the mean ISI we found the ISI histogram in the case of pure white noise ( , Eq . ( 2 ) ) mapping the model to one without adaptation and renormalized base current ( Methods , Eq . ( 52 ) ) . This corresponds to a mean-adaptation approximation [18] , [43]–[47] , because the adaptation variable is time-averaged by the linear filter dynamics in Eq . ( 2b ) for much large than the mean ISI ( ) . However , this approximation cannot account for ISI correlations , because any correlations between ISIs are eliminated in the limit – in fact , the reduced model is a renewal model . For this reason , we developed a novel technique to calculate serial correlations for a PIF neuron with adaptation and white noise driving , which is valid for any time constant ( see Methods ) . In the opposite limit of only adaptation fluctuations ( , Eq . ( 3 ) ) , we could calculate analytically the ISI histogram , the skewness and kurtosis of ISIs as well as the ISI serial correlations by mapping the problem to one without an adaptation variable but a colored noise with renormalized parameters . Specifically , the IF dynamics for only adaptation channel noise reduces to ( 4a ) ( 4b ) where the effective parameters are scaled by a common scaling factor: ( 5 ) with ( 6 ) As before , a spike is fired whenever reaches , whereupon the voltage is reset to . We call this model ( Eq . ( 4 ) – ( 6 ) ) the colored noise approximation . For the perfect integrate-and-fire model driven by a weak colored noise , i . e . for the model described by Eq . ( 4 ) , analytical expressions for the ISI density and the serial correlation coefficient are known [48] . In addition to that , we derived novel analytical expressions for the skewness and kurtosis of the ISIs ( see Methods ) . Interestingly , the scaling factor in Eq . ( 6 ) has a concrete meaning in terms of spike-frequency adaptation: coincides with the degree of adaptation in response to a step increase of the base current ( see Methods , Eq . ( 56 ) ) . We investigated whether our theoretical predictions based on a simple integrate-and-fire model are robust with respect to a more detailed model of the Hodgkin-Huxley type . To this end , we performed simulations of the conductance-based Traub-Miles model with a M-type adaptation current [51] . As in the previous model we separately considered the two cases of white noise input and a slow M-type channel noise to get an intuition of the individual effects on the ISI statistics . Fig . 11 demonstrates that the ISI histograms show essentially the same features as in the PIF model: in the case of white noise input the shape of the ISIH could be well approximated by an inverse Gaussian distribution which was uniquely determined by the firing rate and the CV . In the case of a stochastic M-type current there is a strong disagreement between the ISIH and an inverse Gaussian with the same rate and CV . In particular , ISIHs exhibited again a sharper peak compared to the relatively broad inverse Gaussian . The different ISI statistics for the case of deterministic and stochastic adaptation are analyzed more closely in Fig . 12 . As in the PIF model ( cf . Fig . 5 ) the rescaled skewness and kurtosis are significantly smaller for white noise than for adaptation noise in a wide range of CVs ( Fig . 12A , B ) . This is in accordance with the pronounced peak of the ISIH in the case of stochastic adaptation ( Fig . 11B ) . However , the values are not strictly separated by as in the PIF model . This discrepancy is not surprising , given that the Traub-Miles dynamics with constant input and white noise driving does not exactly yield an inverse Gaussian ISI density but only an approximate one . Importantly , however , the rescaled kurtosis quickly saturates at a finite value in the large limit ( albeit not at unity , Fig . 12C ) . This is markedly different from the case of stochastic adaptation . In this case , the rescaled kurtosis increases strongly as it was observed for the PIF model . In a similar manner , the rescaled skewness also showed this distinct behavior for stochastic vs . deterministic adaptation , although the increase of the rescaled skewness was not as strong as for the rescaled kurtosis ( data not shown ) . A clear distinction between both cases appears in the serial correlations of ISIs ( Fig . 12D ) . Similar as in the PIF model , the case of deterministic adaptation is characterized by negative ISI correlations at lag one , which are strongest at an intermediate time scale . Furthermore , the case of stochastic adaptation exhibits positive correlation coefficients , which show a maximum at an intermediate value of . This is also in line with the PIF model . The correlations decay rapidly with the lag for deterministic adaptation ( Fig . 12E ) and decay exponentially for stochastic adaptation ( Fig . 12F ) . As in the PIF model , the exponential decay is slower for large time constants . Finally , we inspected the case in which both white noise and slow adaptation noise is present ( Fig . 12G , H ) . As in Fig . 10 for the PIF model , we fixed the noise intensity of the white noise and varied the number of adaptation channels . In the Traub-Miles model one finds qualitatively similar curves as in the PIF model . In particular , the serial correlation coefficient at lag one , shows a transition from positive to negative ISI correlations at a certain number of adaptation channels ( Fig . 12H ) . As for the PIF model , this value can be used to define two regimes – one dominated by adaptation noise ( white region ) and another one dominated by white noise ( gray-shaded region ) . In the adaptation-noise dominated regime the parameter is larger than in the white-noise dominated regime ( Fig . 12G ) . The observation that key features of the ISI statistics in the presence of a stochastic adaptation current seem to be conserved across different models suggests a common mechanism underlying these features . As we saw , this mechanism is based upon the fact that a stochastic adaptation current can be effectively described by an independent colored noise . The long-range temporal correlations of this noise naturally yield positive ISI correlations and a slow modulation of the instantaneous spiking frequency . The latter typically involves a large kurtosis due to the increased accumulation of both short and long ISIs . A significant amount of colored noise can effect the kurtosis and the ISI correlations so strongly , that details of the spike generation seem to be of minor importance . Thus , it becomes plausible that the spiking statistics of a rather complex neuron model could be explained by a simple integrate-and-fire model including a stochastic adaptation current .
In this paper , we have studied how a noisy adaptation current shapes the ISI histogram and the correlations between ISIs . In particular , we have compared the case of pure stochastic adaptation with the case of a deterministic adaptation current and an additional white noise current . Using both a perfect IF model that is amenable to analytical calculations and a more detailed Hodgkin-Huxley type model , we found large differences in the ISI statistics depending on whether noise was mediated by the adaptation current or originated from other noise sources with fast dynamics . As regards the ISI histogram , stochasticity in the adaptation leads to pronounced peaks and a heavy tail compared to the case of deterministic adaptation , for which the ISI density is close to an inverse Gaussian . To quantify the shape of ISI histograms we proposed two measures that allow for a simple comparison with an inverse Gaussian probability density that has the same mean and variance . The first one is a rescaled skewness ( involving the third ISI cumulant ) ; the second is a rescaled kurtosis ( involving the fourth ISI cumulant ) . Both quantities possess the property that they assume unity for an inverse Gaussian distribution . If they are larger than unity as in the case of stochastic adaptation the ISI density is more skewed or respectively has a sharper peak and a heavier tail than an inverse Gaussian density with the same variance . If these measures are smaller than one , the ISI histogram tends to be more Gaussian like . Most strikingly , we found that for a stochastic adaptation current the rescaled skewness and kurtosis strongly increase when the time scale separation of adaptation and spiking becomes large ( ) . By contrast , for a deterministic adaptation current the rescaled kurtosis saturates close to one in this limit . Another pronounced difference arises in the ISI correlations . For a deterministic adaptation current and a white noise driving one observes short-range anti-correlations between ISIs as reported previously ( e . g . [29] ) . In contrast , with slow adaptation noise ISIs exhibit long-range positive correlations . In the presence of both types of noise , the serial correlation coefficient changes continuously from positive to negative values when the ratio of white noise to adaptation noise is increased . The two domains might be useful in determining the dominating source of noise from a neural spike train . Interestingly , the perfect integrate-and-fire model augmented with an adaptation mechanism predicted all the features seen in the spiking statistics of the Traub-Miles model with stochastic adaptation and/or white noise input . This indicates the generality and robustness of our findings . It also justified the use of the adapting PIF model as a minimal model for a repetitive firing neuron with spike-frequency adaptation . It seems , that in the suprathreshold regime the details of the spike generator are of minor importance compared to the influence of adaptation and slow noise . By means of the PIF model one can theoretically understand the underlying mechanism leading to the large kurtosis and the positive ISI correlations in the case of stochastic adaptation . This rests upon the fact that slow adaptation noise effectively acts as an independent colored noise with a large correlation time . One can think of the colored noise as a slow external process that slowly modulates the instantaneous firing rate or , equivalently , slowly changes the ISIs in the sequence . Such a sequence of many short ISIs in a row and a few long ISIs gives rise to a large skewness and kurtosis and positive serial correlations . In previous works , slow processes which cause positive ISI correlations were often assumed to originate in the external stimulus [49] , [50] , [52] . Here , we have shown that an intrinsic process , i . e . the fluctuations associated with the stochasticity of adaptation , yields likewise positive ISI correlations . Our finding also provides an alternative explanation of positive ISI correlations in experimental studies [30] , [53] . Moreover , in vivo recordings from a looming-sensitive interneuron in the locust optic lobe have revealed both positive correlations at large firing rates and negative correlations at low firing rates [23] . Because this neuron exhibits pronounced spike-frequency adaptation an intriguingly simple explanation for these observations would be the presence of both fast noise and stochastic adaptation ( corresponding to our mixed case ) . In this case , a large firing rate could indeed lead to a large effective correlation time of the noise associated to the adaptation mechanism and thus to positive ISI correlations . Spike-frequency adaptation has been ascribed to different mechanisms ( see e . g . [18] ) , involving for instance , calcium-dependent potassium currents [42] , slow voltage-dependent M-type currents [41] , [42] and slow recovery from inactivation of sodium currents [54] . Here , we chose the M-current as an example to illustrate the emergence of noise in the adaptation mechanism . In this specific case , it was the finite number of M-type potassium channels that gave rise to slow channel noise . For the other commonly studied adaptation mechanism , the [18] , , we have to deal with two possible sources of noise: the finite number of potassium channels and fluctuations of the local concentration . Proceeding in a similar fashion as for , we would obtain , with the fraction of open potassium channels , obeying ( 12 ) ( 13 ) Here , the Gaussian white noises and approximately represent the channel noise and the concentration fluctuations due to stochastic removal of calcium , respectively . The calcium gating is characterized by the steady-state activation . For simplicity , the increase of calcium caused by an action potential is assumed to be deterministic . Importantly , however , the channel dynamics is fast compared to the slow removal of calcium , i . e . . Following [18] the open probability of the potassium channels adiabatically adjusts to ( i . e . ) and the relationship is roughly linear ( i . e . ) . Thus , we have , where the “channel noise” possesses a correlation time . If this correlation time is much smaller than the mean ISI , the channel noise can be approximately treated as a white noise . But this means , that a PIF neuron with a calcium-dependent instead of can likewise be approximated by Eq . ( 1 ) : the fast channel noise can be included into the white noise term and the slow fluctuations of the calcium concentration assume the role of the slow adaptation noise . Approximating again by a voltage-independent current , the PIF model with would read ( 14 ) ( 15 ) These equations can indeed be put into the form of Eq . ( 1 ) by splitting the deterministic and the noise part of . This illustrates that the main results derived in this paper are not specific to a certain adaptation current , but apply quite generally to any noise associated to the slow dynamics of adaptation . The adaptation currents and have been distinguished with respect to their ability to synchronize coupled neurons [51] and regarding the influence on neural coding [55] . The difference consists in whether the current is activated solely by spikes as in the case of or whether it is also activated by subthreshold voltages as for . For the sake of clarity , we have set the activation function of the M-type adaptation current in the PIF model equal to zero at subthreshold voltages , i . e . between spikes ( Eq . ( 25 ) ) . Thus , the adaptation current in the PIF model , unlike the M-current , was only activated during action potentials . It is , however , easy to show that the results of this paper are unchanged if subthreshold activation is allowed . For simplicity , let us consider the extension that in-between spikes the steady-state activation function is equal to the value , i . e . instead of Eq . ( 25 ) , ( 26 ) ( see Methods ) we have ( 16 ) ( 17 ) This only increases the mean adaptation to ( cf . Eq . ( 37 ) ) . Similarly , the variance changes according to ( cf . Eq . ( 40 ) in Methods ) . As a result , the effective base current is now given by ( 18 ) with the new scaling factor ( 19 ) The colored noise approximation can be carried out in an analogous manner yielding the same result Eq . ( 4 ) ( again with , , but the new scaling factor , Eq . ( 19 ) ) . Thus , it can be expected that in the presence of subthreshold activation of the colored noise effect ( i . e . pronounced peak of ISIH , positive ISI correlations ) in the case of stochastic adaptation is preserved . Furthermore , still serves as the degree of adaptation , i . e . the ratio of steady-state to initial gain when a step current is applied . The analytical calculation of higher-order statistics in the presence of adaptation is a fundamental theoretical problem , which has been largely ignored so far ( for a recent exception see [37] ) . Here , we succeeded to provide explicit expressions for the ISI histogram and their serial correlations for both white noise driving and noise in the adaptation dynamics . This was achieved by analyzing a spike generator and a channel model that are as simple as possible . There are certainly a lot of details that can be modeled in a more realistic way . For instance , it is known that the M-channel kinetics is governed by several time scales and more than two internal states [56] . Furthermore , channels might not be strictly independent , but channel clusters might exhibit cooperative behavior [57] . The latter case , would actually increase the level of channel noise compared to the case of independent channels , i . e . cooperativity would contribute to stochastic adaptation . For many neurons physiological details like the number of ion channels are hard to obtain directly from experiments . Instead given the spike train statistics of a neuron , our study could be useful to judge whether M-channels or other adaptation mechanisms could potentially contribute to the neuronal variability . Furthermore , it is not impossible to think of experiments , in which the number of adaptation channels is reduced ( e . g . by the mild application of a channel blocker ) and thus the effects of stochastic adaptation is affected in a controlled way . Another possibility to test our predictions would be to vary the firing rate of the neuron by increasing or decreasing the input current . In this way , the time scale of spiking would change relative to the time scale of adaptation and , thus , the colored noise effect of adaptation noise could be enhanced or attenuated , respectively . Channel noise can crucially influence neural firing especially in the absence of synaptic input [58] , [59] . This could be particularly relevant for the irregular discharge patterns of certain receptor cells . So far , channel noise has been studied mostly in the context of stochastic and channel gating involved in the spike generation itself . These channels are considered to be fast . Because we were mainly interested in the effect of slow adaptation channels compared to fast fluctuations resulting from fast ion channels or synaptic activity , we lumped all fast noise sources into an unspecified additive white noise source . This is certainly a simplification; e . g . it has been shown in experiments that voltage noise due to channels depends on the mean voltage itself [39] , [40] . More detailed models of the various sources of noise are worth the efforts in future investigations . However , we do not expect that such sophisticated models would change our results qualitatively , because they mainly hinge upon the presence or absence of long time scales . Realistic numbers of M-type channels per neuron are difficult to estimate and the numbers used in this paper must be seen as a tuning parameter for the channel noise intensity . Channel densities of the M-type have been estimated to be of the order of one functional channel per [60] . Assuming a spherical cell with a diameter of one obtains of the order of channels . Thus , the channel numbers used in this study ( – ) seem to be reasonable; and hence the M-current could be a potential source of fluctuations . The diffusion approximation for the stochastic dynamics of ion channel populations ( also known as Langevin or Gaussian approximation ) has been studied by several authors [38] , [61]–[64] ( see also [65] in the context of chemically reacting particles ) . Here , we have shown how one can map the stochastic dynamics of a population of ion channels with negative feedback to the macroscopic current dynamics plus an additive colored noise ( see [38] for a related treatment ) . In other words , the dynamics could be reduced to an analytically accessible Langevin equation for voltage and adaptation . In particular , we investigated the effect of the diffusion approximation on the statistics of interspike intervals and found a fairly good agreement with the channel model , despite the small number of channels . This seems surprising , given that for a typical parameter set – , open probability and – one expects only closing transitions ( between spikes ) and open transitions ( during action potentials ) per millisecond . Apparently , on the time scale of 1 ms the number of transition events is not Gaussian distributed . However , we found that the main effect of the channel noise consists in a slow modulation of the instantaneous firing rate on the large time scale of , whereas high-frequency components of the noise are of minor importance . Thus , on relevant time scales of the order of 10–100 ms the average number of transitions is much larger and a Gaussian approximation seems to be reasonable . Spike-frequency adaptation has been commonly studied with regard to its mean effect on the firing rate [18] , [43]–[46] . It has been shown that these effects can be exhaustingly analyzed using a universal firing rate model [18] . In this paper , however , it became evident that higher-order statistics and fluctuation effects may differ and may be used to distinguish different kinds of noise sources .
To analyze the slow , voltage-dependent adaptation channels in a simple setup we consider a population of independent ion channels that reside in an open or a closed state . For each channel , we thus have the simple reaction kinetics shown in Fig . 13A . For channels , one can either perform independent simulations of one two-state process or one simulation with states where the number of open channels can be increased or decreased by one: ( 20a ) ( 20b ) Here , denotes the number of closed channels . The rates and for the transitions between the closed and open state can be related to the ( voltage-dependent ) kinetics of a typical gating variable by choosing ( 21 ) Therein , is the steady-state open probability of a single channel when the membrane potential is clamped at and sets the time scale of the channel kinetics . Note , that both and are accessible from experiments . The master equation for the open probability of the two-state model reads ( 22 ) or after insertion of Eq . ( 21 ) ( 23 ) which follows the common scheme for gating variables with voltage-dependent kinetics in Hodgkin-Huxley type models . We will use the function in two different versions . In the conductance-based Traub-Miles model , we will use the kinetic scheme with the rate functions given by ( 24 ) Inserting this relation into Eq . ( 21 ) , both rate functions show a quasi sigmoidal dependence on the voltage ( see Fig . 13B ) , such that is appreciably different from zero only during the action potential ( i . e . for mV ) whereas is only finite in the opposite range of a subthreshold membrane voltages . For the integrate-and-fire ( IF ) model we employ a simplified variant that distinguishes only between two values of : one attained when the voltage is subthreshold and another one for the duration of the action potential . Specifically , at a spike event of the IF model , is set to one for a duration of and otherwise it is set to zero ( Fig . 1B ) . Thus , can be expressed as a function of time and the last spike time : ( 25 ) This function and the set of spike times are related by ( 26 ) where denotes the Heaviside function . Hence , in the simplified model , the dependence of the rates on the membrane voltage is substituted by an explicit dependence on time and the most recent spike time : ( 27 ) In the definition of , Eq . ( 25 ) , we require that the duration of the pulse is much smaller than the mean ISI , so that overlaps of two subsequent pulses are unlikely . Note , that for simplicity channels can be activated only during spikes ( ) , but not in between spikes ( ) . However , one can show that the following results are not changed qualitatively if one allows for subthreshold activation ( ) , as observed for M-type potassium currents ( see Discussion ) . Taking the open probability exactly as the fraction of open channels amounts to taking the limit of an infinite population of channels . In the Traub-Miles model or the integrate-and-fire model with deterministic adaptation , this corresponds to adding an adaptation current of the form [18] ( 28 ) In this equation , denotes the maximal conductance ( per unit membrane area ) and constitutes the reversal potential of the adaptation current . For a finite channel population , the fraction of open channels is given by the stochastic quantity , where evolves according to the kinetic scheme Eq . ( 20 ) . In contrast to Eq . ( 28 ) , this gives rise to a stochastic adaptation current ( 29 ) When the channel number is varied , we assume that the maximal conductance per unit membrane area remains constant . Thus , a change of the channel number can be realized either by a variation of the total membrane area or by a change of the channel density in conjunction with a simultaneous scaling of the single channel conductance . Such procedures enable a change of the amount of channel noise , without changing the mean current per unit membrane area . The perfect integrate-and-fire ( PIF ) model [2] constitutes a minimal model for a neuron possessing a stable limit cycle . In this model the subthreshold voltage is determined by the equation ( 35 ) where and are proportional to the base current and adaptation current , respectively; is the membrane capacitance and the scaling factor for the adaptation current reads . Here , we used an effective-time-constant approximation [66] , where we substituted in Eq . ( 29 ) by the average voltage to obtain a voltage-independent adaptation current [18] . The last term in Eq . ( 35 ) represents fast Gaussian input fluctuations of intensity and correlation function ( here and in the following , the angular brackets denote an ensemble average ) . The model Eq . ( 35 ) is complemented by the fire-and-reset rule: upon reaching the threshold a spike is elicited and is reset to , . Because Eq . ( 35 ) is invariant with respect to a constant shift in , we can choose the reset value as the origin , i . e . . The threshold crossing events define the spike times , , …of the PIF model . It is a feature of the PIF model that the firing rate is directly proportional to a constant driving current and independent of the noise . However , even for the slowly varying driving current in Eq . ( 35 ) , one can define an instantaneous firing rate ( 36 ) which can be seen as the slowly varying firing rate that is obtained by averaging the spike train over the time scale of the adaptation . Averaging over time scales much larger than yields a constant firing rate , because the process is stationary . Thus , the ( stationary ) firing rate can be obtained by averaging Eq . ( 36 ) , which gives . On the other hand , the firing rate is related to the mean fraction of open channels by averaging Eq . ( 33 ) : ( 37 ) Solving for yields ( 38 ) where . In the following , denotes the interspike intervals ( ISIs ) , i . e . the intervals between adjacent spikes , and are the corresponding ISIs in units of the adaptation time constant . The statistics of ISIs can be characterized by different measures . A single interval is distributed according to the probability density , i . e . the normalized ISI histogram . The shape of the ISI density can be characterized using the cumulants . The first cumulant equals the mean ISI and the inverse of the firing rate , ( 57 ) The second cumulant equals the variance and is related to the coefficient of variation ( CV ) , which is a measure of ISI variability: ( 58 ) We further consider the third cumulant , which is related to the skewness of the ISI distribution , defined as ( 59 ) and the fourth cumulant , which determines the kurtosis ( or excess ) of the ISI distribution . It is defined as ( 60 ) Roughly speaking , the kurtosis indicates how much of the variability is due to events that strongly deviate from the mean value . For instance , a unimodal ISI density with a heavier tail compared to another ISI density with the same CV , tends to exhibit a larger kurtosis . This is typically accompanied by a more pronounced peak close to the mean value to balance the heavy tail . In this paper , we want to compare the ISI density with an inverse Gaussian probability density serving as a reference statistic . For an inverse Gaussian ISI distribution ( see below ) one observes that the skewness is proportional to the CV and the kurtosis scales like the squared CV . This suggests to introduce rescaled versions of the skewness and the kurtosis as follows: ( 61 ) and ( 62 ) By defining the rescaled skewness and kurtosis in this manner , we obtain measures that are identical to unity for an inverse Gaussian ISI density . For larger ( smaller ) values , the ISI density is respectively more ( less ) skewed and more ( less ) peaked compared to an inverse Gaussian density . This procedure is somewhat analogous to the definition of the CV , for which the Poisson process serves as a reference for the ISI variability with . Furthermore , we consider the correlations among ISIs as quantified by the serial correlation coefficient [67]: ( 63 ) where due to stationarity the expression does not depend on the integer , i . e . on the position in the sequence of ISIs . Alternatively , the averages involved in Eq . ( 63 ) can also be taken along the sequence ( ) . For the adapting PIF model , the two limit cases that are considered in this paper could be reduced to simplified models , for which analytical results are partly known . Firstly , in the case of deterministic adaptation , i . e . , the ISI density can be approximately described by the first-passage-time density of a biased Brownian motion described by Eq . ( 52 ) ( mean adaptation approximation ) . A classical result for this purely white noise driven PIF neuron with a constant drift is that the ISI density is given by the so-called inverse Gaussian [2] ( 64 ) This distribution has a mean ( 65 ) and a CV ( 66 ) Furthermore , by construction we have ( 67 ) The mean adaptation approximation would wrongly predict that the ISIs are uncorrelated . The reason is that in the PIF model driven by only white Gaussian noise any memory of the ISI history is erased upon reset . A better account of ISI correlations is given below . Secondly , the case of stochastic adaptation can be reduced to a PIF neuron driven by a reduced base current and a colored noise with correlation time and variance ( Ornstein-Uhlenbeck process , see Eq . ( 53 ) ) . In the case of a weak colored noise , it is useful to define the small parameter ( 68 ) For an approximation for the ISI density is given by [48] ( 69 ) with , and mean ISI . For , i . e . for ISIs much smaller than the correlation time of the colored noise , the expression for reduces to [50]: ( 70 ) Although this formula captures the ISI density at small ISIs , it is of limited use , because the second and higher ISI moments diverge . Throughout the paper we have therefore used the full expression Eq . ( 69 ) . The mean ISI and the firing rate do not depend on the noise statistics , in fact they are equal to the white noise case , Eq . ( 65 ) , as shown below ( derivation of the ISI cumulants ) . The squared CV can be obtained to second order in [48] using the methods presented below: ( 71 ) with . Similarly , the rescaled skewness and kurtosis are derived for weak colored noise below . Finally , the serial correlation coefficient can be computed analytically for weak noise [48] ( ) : ( 72 ) Here we derive an expression for the serial correlation coefficient of a PIF neuron with deterministic adaptation current and white noise driving . We consider the following subthreshold dynamics ( 73 ) ( 74 ) augmented with the usual reset rule for at the spike threshold . The adaptation variable jumps by an amount at each spiking event . Following [48] , we derive here the ISI cumulants of the colored-noise driven PIF model , Eq . ( 53 ) , in the weak-noise approximation . The ISI cumulants are required to compute the rescaled skewness and kurtosis . The Fokker-Planck equation associated to Eq . ( 53 ) for the time-dependent joint probability density reads ( 91 ) For the description of the ISI density it is necessary to use the initial condition that corresponds to the distribution of upon spiking . The initial condition is for weak noise , , well approximated by ( see [48] ) ( 92 ) The ISI density is equal to the time-dependent probability flux across the threshold line if threshold crossings of trajectories with negative , i . e . crossing from above the threshold , are prohibited . This is achieved by imposing a reflecting boundary on the half line , . For , however , negative are highly unlikely , so that the free process without reflecting boundary generates a flux that is a reasonable approximation of the ISI density . To carry out a weak-noise expansion of Eq . ( 91 ) we change to the variables ( 93 ) In these variables the Fokker-Planck equation for reads ( 94 ) where is given by Eq . ( 68 ) . The density for the ISIs ( in units of ) is approximately the total probability flux across the threshold: ( 95 ) where ( 96 ) Furthermore , we consider the characteristic function , which by means of Eq . ( 95 ) can be expressed as ( 97 ) In the last equation we introduced the function ( 98 ) which arises from the subsequent application of a Laplace and a Fourier transformation to the probability density . The cumulants of the ISI density can be obtained from the characteristic function , see e . g . [69]: ( 99 ) From Eq . ( 97 ) and ( 99 ) it is clear , that all cumulants can be generated from the function . Applying the Laplace and Fourier transformation ( as in Eq . ( 98 ) ) to Eq . ( 94 ) we arrive at an equation for , which reads ( 100 ) with the boundary condition ( 101 ) To solve Eq . ( 100 ) for weak noise , we expand in powers of the small parameter : ( 102 ) Inserting into Eq . ( 100 ) gives a hierarchy of first-order differential equations for the coefficients : ( 103 ) ( 104 ) ( 105 ) The solutions can be obtained order-by-order using the method of characteristics . Here , we report only the first three coefficients : ( 106 ) ( 107 ) ( 108 ) Higher-order coefficients can be calculated in the same way , although the expressions become quite lengthy . To obtain the kurtosis including the first noise-dependent term correctly , one has to calculate up to the eighth order . These straightforward computations can be accomplished by using a computer algebra system . As a result , the cumulants up to fourth order in read ( 109 ) ( 110 ) ( 111 ) ( 112 ) with From the cumulants one finds the rescaled skewness and kurtosis as defined in Eq . ( 61 ) and ( 62 ) : ( 113 ) and ( 114 ) with To test the generality of our findings , we simulated the conductance-based Traub Miles model modified by B . Ermentrout [51] . It is a single compartment model with an additional M-type current , i . e . a slow voltage-dependent potassium current , inducing spike-frequency adaptation . In order to contrast the effects of deterministic versus stochastic adaptation on the firing statistics of the conductance-based model , we simulated two versions with either additive white Gaussian noise or adaptation channel noise . For the first model with fast fluctuating current noise and deterministic adaptation , the membrane potential measured in mV is determined by ( 115 ) where denotes the membrane capacitance , is the base current , and indicates the intensity of Gaussian white noise with correlation function . The deterministic ionic currents are given by the following equations [51]: Sodium current: Potassium current: Leak current: M-type adaptation current: In this model , the adaptation time constant is a voltage-dependent function that we reparameterized such that roughly corresponds to the time constant governing the exponential buildup of during periodic firing at 100 Hz in a simulation of equation ( 123 ) without the current and with . For the second model with adaptation channel noise , the voltage is described by ( 116 ) The currents , and are the same as in the first model . The M-type adaptation current , however , is modeled as a stochastic current where indicates the fraction of open channels . As in the PIF model , we assumed the adaptation channels to be two-state ion channels with the transition rates and . The gating of the adaptation channels was simulated using the Gillespie algorithm [70] , [71] . This algorithm calculates the time interval until the next state transition , determines the reaction type , here channel opening or closing , and updates the number of channels in each possible state accordingly . For a given time step , the number of channels in the open state is then used to calculate the fraction of open channels as well as the stochastic adaptation current . Furthermore , since the transition rates depend on , we restricted the maximal transition time to . In the model with stochastic adaptation current , the maximal channel conductance and the constant base current ( ) were chosen to result in a CV 0 . 6 and a firing rate of 100 Hz for a simulation of ion channels carrying the adaptation current . For the simulation with deterministic adaptation current and additive white noise the base current was adjusted ( see corresponding figure captions ) to yield the same rate while keeping the conductance the same .
|
Neurons of sensory systems encode signals from the environment by sequences of electrical pulses — so-called spikes . This coding of information is fundamentally limited by the presence of intrinsic neural noise . One major noise source is “channel noise” that is generated by the random activity of various types of ion channels in the cell membrane . Slow adaptation currents can also be a source of channel noise . Adaptation currents profoundly shape the signal transmission properties of a neuron by emphasizing fast changes in the stimulus but adapting the spiking frequency to slow stimulus components . Here , we mathematically analyze the effects of both slow adaptation channel noise and fast channel noise on the statistics of spike times in adapting neuron models . Surprisingly , the two noise sources result in qualitatively different distributions and correlations of time intervals between spikes . Our findings add a novel aspect to the function of adaptation currents and can also be used to experimentally distinguish adaptation noise and fast channel noise on the basis of spike sequences .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[
"neuroscience/theoretical",
"neuroscience"
] |
2010
|
How Noisy Adaptation of Neurons Shapes Interspike Interval Histograms and Correlations
|
Senescence is a universal barrier to immortalisation and tumorigenesis . As such , interest in the use of senescence-induction in a therapeutic context has been gaining momentum in the past few years; however , senescence and immortalisation remain underserved areas for drug discovery owing to a lack of robust senescence inducing agents and an incomplete understanding of the signalling events underlying this complex process . In order to address this issue we undertook a large-scale morphological siRNA screen for inducers of senescence phenotypes in the human melanoma cell line A375P . Following rescreen and validation in a second cancer cell line , HCT116 colorectal carcinoma , a panel of 16 of the most robust hits were selected for further validation based on significance and the potential to be targeted by drug-like molecules . Using secondary assays for detection of senescence biomarkers p21 , 53BP1 and senescence associated beta-galactosidase ( SAβGal ) in a panel of HCT116 cell lines carrying cancer-relevant mutations , we show that partial senescence phenotypes can be induced to varying degrees in a context dependent manner , even in the absence of p21 or p53 expression . However , proliferation arrest varied among genetic backgrounds with predominantly toxic effects in p21 null cells , while cells lacking PI3K mutation failed to arrest . Furthermore , we show that the oncogene ECT2 induces partial senescence phenotypes in all mutant backgrounds tested , demonstrating a dependence on activating KRASG13D for growth suppression and a complete senescence response . These results suggest a potential mechanism to target mutant KRAS signalling through ECT2 in cancers that are reliant on activating KRAS mutations and remain refractory to current treatments .
Cellular senescence , often described as the irreversible arrest of cell proliferation , can be induced by a variety of signals [1 , 2] . It is a complex phenotype , consisting of various effector mechanisms including the DNA damage response ( DDR ) , chromatin modification , autophagy and the senescence associated secretory phenotype ( SASP ) [3 , 4] . Understanding collective control of these mechanisms is a priority [5] and screening approaches might help to deconvolute senescence pathways . However , the senescence response involves the expression of many biomarkers linked to these effector mechanisms , including cell cycle inhibitors such as p16 and p21 , components of the SASP [6] and in some contexts senescence-associated heterochromatic foci ( SAHF ) , which are associated with gene repression [7] . There are also some key morphological changes that occur , including a large flattened morphology , enlarged nucleus and expression of senescence associated beta-galactosidase ( SAβGal ) . Expression of these biomarkers is context dependent and few specifically define senescence , therefore various markers should be investigated in parallel to confirm the senescent state [3 , 6 , 8] . Senescence markers and effectors are evident in various premalignant tissues in vivo , consistent with a role for senescence as a universal barrier to tumorigenesis . In mouse models of KRASV12-dependent lung adenocarcinoma and pancreatic ductal carcinoma , p16 and SAβGal were detected in premalignant lesions , but not in the malignant disease [9] . Similarly , oncogenic BRAFV600E overexpression can induce senescence in cultured melanocytes; human nevi ( moles ) , which often carry BRAFV600E , express many senescence markers and much evidence supports a role for senescence in maintaining their arrested state and preventing progression to melanoma [10 , 11] . Senescence signalling also occurs in advanced disease [4] and in response to standard chemo- or radiotherapy [12] , where it may aid therapeutic activity of these agents by acting as a fail-safe mechanism in cases where pro-apoptotic signalling is defective [13] . Furthermore , induction of the SASP by cancer therapeutics may also induce a tumour-directed immune response owing to the release of inflammatory cytokines . Consistent with this , the re-expression of p53 in a p53-deficient mouse model of liver carcinoma expressing oncogenic HrasV12 resulted in senescence , rather than apoptosis , and tumour regression by immune clearance [14] . However , in some contexts SASP components could promote cell proliferation and tumorigenesis [15] . Senescence induction is an attractive concept in cancer research , and the idea of modulating the senescence response for therapeutic benefit , either to enhance current treatments or as a tumour suppressive therapy in its own right , has been gaining momentum over recent years . However , cellular senescence and immortality remain underserved areas for drug discovery owing to a lack of senescence-inducing agents and an incomplete knowledge of the complexity of the underlying signalling events . To address the first issue , we recently reported identification of a novel compound , CB-20903630 , through a “target agnostic” virtual screen . The compound selectively induces a variety of senescence associated phenotypes and G1 blockade in cancer cells [16] . However , given the diversity of senescence triggers , it seems clear that engagement of the response is under distributed control [17] . Therefore , populating senescence effector pathways remains a major aim . Kinome-focused siRNA screening has previously been used successfully to uncover pathways regulating cell immortality [18 , 19] . In the current study , we undertook a large-scale siRNA screen combined with high-throughput imaging in the human melanoma cell line A375P to identify senescence effectors . Our results revealed diversity in levels of senescence engagement between gene targets and cell lines , consistent with recent data [4 , 6] . Using a panel of HCT116 isogenic cell lines carrying common cancer-relevant gene mutations , we show that some senescence phenotypes can be engaged despite the absence of known effectors such as p53 or p21 . However , distinct differences in proliferation arrest engagement are evident in different genetic backgrounds . Finally , we show senescence induction by knock-down of ECT2 in particular is greatly enhanced in the presence of oncogenic KRASG13D . These findings suggest that pro-senescence therapy may be effective in various malignancies , including those harbouring oncogenic RAS mutations , which are common and often refractory to treatment . Furthermore , by populating the signalling pathways that regulate the senescent phenotype with the effectors identified by our screen , a more complete picture of the senescence response can be drawn .
To identify potential senescence-inducing targets for drug discovery and extend our knowledge of senescence signalling in cancer , we initially performed validation of a high-content fluorescence imaging screen focused on senescence phenotypes . Using A375P cells treated with etoposide , we demonstrated that stable growth arrest in this cell line , assessed by colony formation assays and cell growth kinetics after compound washout , was accompanied by induction of a range of senescence markers . These included increased nuclear area , SAβGal and p21 expression , and 53BP1 and H2AX nuclear foci . We next tested an imaging assay based on the nuclear area marker and cell number in a small scale screen of 160 kinase inhibitor compounds by high content fluorescence imaging using the Operetta platform . We have previously reported the ability of some of these compounds to induce senescence in normal fibroblasts [20] . Four of the best hits from this screen were tested in a range of secondary assays , including colony formation assays , confirming the ability of screening based on these markers to identify chemical agents capable of causing a range of senescence phenotypes alongside stable proliferation arrest . These validation experiments are summarised in S2 File . We next used the validated imaging assay to perform a large-scale screen of 10 , 414 gene targets using the Ambion Silencer Select Druggable Genome siRNA library . As quality controls on each screening plate , etoposide ( S2 File ) and an independent siRNA obtained from Qiagen against CDK1 ( which is known to be involved in senescence ) were included . Primary hits in A375P cells , detected by increased nuclear area and reduced proliferation compared to controls [20] , were rescreened and ranked by confidence ( Materials and Methods , S1 Fig ) and a refined hit list of the 24 most robust hits and 16 others predicted to be druggable targets by functional analysis was generated for further validation . Criteria for the assessment of druggable targets included previous success in targeting the encoded protein or another of the same molecular class with a small molecule or drug-like compound , or the presence of a potentially druggable site within the protein structure . Many of the 24 highest-confidence hits were also classed as druggable , increasing our interest in them . The combined validation list of 40 targets showed a range of senescence-like morphologies upon knockdown , and included some weaker hits with fewer siRNAs passing the cut-off point ( S1B Fig ) . These were included from a drug development standpoint and also to test a range of senescence responses in secondary validation assays . The siRNAs selected are given in Table 1 , together with their prioritisation grouping and confidence ranking . Predictably , the majority of these 40 genes are involved in cell cycle regulation , the DNA damage response and maintenance of DNA integrity . However , several more interesting gene families emerged , including ribonucleoside-diphosphate reductase subunit genes RRM1 and RRM2 , calpain cysteine protease family members CAPN11 and CAPN9 and several proteasome subunit family members ( Fig 1 ) . Interestingly , although the Qiagen positive control CDK1 siRNA performed as well as many of the top hits , the Ambion CDK1 siRNAs from the library were not in our top 40 . We continued to use the Qiagen siRNA in further experiments . To further investigate the 40 prioritised siRNAs as senescence inducers , a fresh batch of the same siRNAs was tested for the ability to increase nuclear area and induce SAβGal in A375P cells using a stricter cut-off value of control mean + 3 standard deviations ( SD ) in at least 2/3 replicate wells in 2/3 independent experiments . Hereafter an “increase” will refer to an increase passing this criterion unless otherwise stated . All 40 siRNAs increased nuclear area in 3 independent experiments ( Fig 2A & S2A Fig ) , further validating the primary screen results . However , many gave SAβGal levels below the stricter cut-off , suggesting a partial or atypical senescence response ( Fig 2A & S2C Fig ) . Analysis of these 40 siRNAs was extended to HCT116 colorectal carcinoma cells , chosen for the availability of isogenic derivatives with various oncogenic genotypes , previously used in screening [21] and their ability to express biomarkers of senescence ( S2B and S2D Fig ) . The results with parental HCT116 cells further corroborated those for A375P ( Fig 2A ) . Many of the strongest inducers of senescence in A375P also came out top in HCT116 , while those that were negative for SAβGal in A375P were also negative for increased nuclear area and SAβGal in HCT116 . Based on the SAβGal results for A375P , the nuclear area and SAβGal for HCT116 , and the assessment of druggability , the list was refined to 16 siRNA targets for further analysis . This refined list encompassed many of the strongest hits from the siRNA screen . We continued to include the Qiagen positive control siRNA targeting CDK1 ( Fig 2B ) . Two additional biomarkers of senescence are p21 ( CDKN1A ) and recruitment of 53BP1 to sites of DNA damage ( nuclear foci ) . To further characterise the response of the refined siRNAs we analysed induction of p21 and formation of 53BP1 foci by immunofluorescence in A375P and HCT116 cells transfected with the 16 hit and control siRNAs ( Fig 3 ) . Consistent with the results for nuclear area and SAβGal , all 16 test siRNAs induced expression of p21 greater than the non-targeting control in A375P ( Fig 3A ) . In HCT116 cells , p21 expression was also increased by all test siRNAs except that targeting CIT ( Fig 3C ) . Likewise , CIT siRNA failed to increase SAβGal in this cell line ( Fig 2 ) , and formation of 53BP1 foci ( Fig 3D ) , suggesting that knockdown of CIT does not induce senescence in HCT116 cells . Generally , only a small induction of 53BP1 was evident in both cell lines ( Fig 3B & 3D ) , however many of the top hits for nuclear area and SAβGal expression , including ECT2 , ESPL1 , DDB1 and CDC45L ( Fig 2 , S2 Fig & S3 Fig ) , were also the most positive for both p21 and 53BP1 , suggesting a coordinated senescence response to the knockdown of these mRNAs . To investigate the ability of our top 16 siRNAs to engage senescence signalling in the presence of common oncogenic mutations , we used a panel of HCT116 isogenic cell lines ( Table 2 ) . The panel consisted of HCT116 parental ( which has pre-existing KRASG13D and PIK3CAH1047R heterozygous mutations ) , HCT116 KRAS+/- ( where the mutant KRASG13D allele has been knocked out ) , HCT116 PIK3CA+/- ( where the mutant PIK3CAH1047R allele has been knocked out ) , HCT116 p21 null and HCT116 p53 null lines . Analysis of nuclear area and SAβGal expression in this panel showed mutation-specific variations in response . We also examined proliferation responses to these siRNAs . We defined cell number of 100%-150% of seeding density at day 5 post-transfection as cytostasis , greater than 150% as growth and less than 100% as toxicity . We found 9/16 siRNAs in p53 null HCT116 cells and 4/16 siRNAs in p21 null cells that increased both nuclear area and SAβGal expression . A further 4 siRNAs increased nuclear area but not SAβGal expression in p21 null cells ( Fig 4A , 4C & 4D ) . However , only 1 siRNA ( KIF11 ) induced both phenotypes in addition to proliferation arrest as defined above in p53 null cells . A number of other siRNA did substantially reduce growth relative to scrambled control siRNA transfections , although the 100%-150% cutoff was not met ( Fig 4C ) . In the HCT116 p21 null cells , transfection of the majority of the top 16 siRNAs , with the exception of CIT , resulted in cell numbers well below plating density ( Fig 4D ) , suggesting that many cells died . However , the p21 null cells that remained viable after siRNA transfection expressed the other senescence phenotypes in many cases . It may be the case that this results from selective killing of non-senescent cells , while those that induce senescent phenotypes survive . However , further investigations would be required to determine this . Conversely , the presence of PIK3CAH1047R appeared to be required for a senescence response . Although 10/16 siRNAs induced both nuclear area increase and SAβGal expression in the PIK3CA+/- cells ( Fig 4A & 4F ) , we saw no reduction in proliferation with any of the siRNAs , indicating that true senescence was not induced in response to knock-down of these targets in PIK3CA+/- cells ( Fig 4F ) . These results are summarised in Table 3 , where we classify each siRNA as provoking a senescence response ( increased nuclear area , increased SaβGal and proliferation arrest ) or partial responses , involving some combination but not all of these phenotypes . Overall , most responses were only partial , suggesting that a spectrum of “senescence-like” phenotypes can be induced in cancer cells of different backgrounds . Parental HCT116 cells showed “complete” senescence response to 2 siRNAs: ECT2 and ALDOA . Interestingly , the siRNA targeting ECT2 induced among the strongest increases in nuclear area across all of the isogenic HCT116 lines tested , along with siRNA against ESPL1 ( Fig 4B–4F ) . Knockdown of ECT2/ESPL1 also resulted in increased expression of SAβGal in all cell lines except p21 null cells ( Fig 4A ) . Furthermore , these were the only siRNAs to increase nuclear area and SAβGal expression in the HCT116 KRAS+/- line ( Fig 4A & 4E ) , in contrast to parental and other HCT116 cells , which showed partial senescence responses involving these phenotypes to many of the siRNAs . Comparing responses , a more robust phenotype also including cytostasis was observed for both ECT2 and ALDOA in the presence of mutant KRAS ( parental cells ) , whereas only partial responses were observed in KRAS+/- cells ( Table 3 ) . For ECT2 , both nuclear area and SAβGal expression were induced in KRAS+/- cells , but proliferation arrest did not occur . For ALDOA , only SaβGal induction was observed in KRAS+/- cells . To further investigate these potential KRAS dependent effects on the senescence response , we characterised the SASP of parental and KRAS+/- cells after transfection with these siRNAs ( Fig 4G ) . Mutliplex chemiluminescent ELISA assays were performed to detect nine pro-inflammatory cytokines . The same cut-off of greater than mean + 3SD of scrambled control by at least 2/3 siRNA in 2/3 experiments was used for assignment of increase . ALDOA knockdown caused a complex inflammatory phenotype involving increased levels of GM-CSF , IL2 and IL8 in both cell lines . In the absence of mutant KRAS , IL1β , IL6 , and IL12-p70 were also induced . In parental cells , instead , IFNγ and IL10 were upregulated . In contrast , ECT2 knockdown produced less complex effects . In KRAS+/- cells , GM-CSF and IL8 were induced . In the presence of mutant KRAS , only IFNγ was increased . We additionally performed assays for activated caspase 3/7 in parental HCT116 cells after ECT2 knockdown ( S4A Fig ) . While the positive control CDK1 siRNA increased levels by 1 . 8-fold , ECT2 knockdown did not increase activated caspase 3/7 . Examination of the levels of ECT2 mRNA detected in microarray analysis after transfection confirmed that the message was reduced to 45% of the level of scrambled control in parental cells ( S4B Fig ) . Taken together , these results indicate that ECT2 induces multiple markers of senescence including growth arrest without caspase activation specifically in KRAS mutant cells , while inducing a SASP phenotype that may suggest sensitisation to apoptotic triggers due to the known involvement of IFNγ in death receptor signalling [22] . ECT2 is a guanine nucleotide exchange factor ( GEF ) for the RHO family of GTPases and has been associated with regulation of RAS–MAPK signalling [23] . This and the induction of a more robust senescence phenotype in the presence of mutant KRASG13D led us to focus on ECT2 . To further investigate the relationship between the senescence response to ECT2 down-regulation and the KRASG13D mutation , we tested the spontaneously immortalised non-transformed mammary epithelial cell line MCF10a and its isogenic derivative MCF10a KRASG13D/+ , in which the KRASG13D allele is knocked-in at the endogenous locus . We also analysed an additional colorectal cancer cell line DLD1 , which , like HCT116 , has a KRASG13D mutation , and an isogenic counterpart in which the mutated KRAS allele is knocked out to leave only wild-type KRAS ( DLD1 KRAS+/- ) . Combined results from all repeats of each individual ECT2 siRNA transfection in this focussed cell panel showed significant increases in nuclear area in all cell lines except the MCF10a isogenic pair ( Fig 5A ) . We did see a slight but significant increase in SAβGal expression in parental MCF10a cells with ECT2 knock-down ( Fig 5B ) . SAβGal expression was significantly increased in response to ECT2 knock-down in the presence and absence of KRASG13D in DLD1 cell lines , but the increase did not reach significance in the HCT116 lines; this effect was strongest in the DLD1 parental cells with the KRASG13D mutation ( Fig 5B ) . Expression of p21 ( Fig 5C ) and 53BP1 ( Fig 5D ) was also significantly increased upon ECT2 knockdown in the HCT116 cell lines , confirming a senescence response . Analysing the results according to the cut-off values used previously ( at least 2/3 siRNAs increase marker levels beyond mean of scrambled + 3SD in at least 2/3 experiments ) , we used heat-maps to summarise the expression of all 4 senescence biomarkers in our KRASG13D isogenic cell line panel . In the non-transformed MCF10a isogenic pair ECT2 knock-down did not reproducibly induce the expression of any biomarker , whereas in HCT116 cell lines the increase in nuclear area and expression of SAβGal was detected in at least 2 out of 3 experiments in both the parental and the KRAS+/- cells , as before . High levels of p21 and 53BP1 were detected in HCT116 parental cells , but no positive increase in KRAS+/- cells , suggesting a more robust senescence response in the presence of oncogenic KRASG13D . This was further corroborated in the DLD1 cells ( Fig 5E ) , although no significant induction of p21 or 53BP1 was detected in these cells . Analysis of MCF10a proliferation showed a low level of toxicity as defined previously ( 83% survival ) , while scrambled transfectants grew to only 149% of seeding density . As noted , this mild toxicity occurred without evidence of induction of any senescence markers . KRAS mutant cells were sensitive to transfection ( 47% survival after transfection with scrambled siRNA ) . However , the relative reduction with ECT2 was similar ( 24% of seeding density ) , again without induction of senescence markers ( Fig 5F ) . Therefore , in this cell panel , engagement of senescence in KRAS mutant background also appeared to be specific for cancer cells . Further studies will be required to evaluate the relative sensitivity of a wider range of normal and KRAS mutant cancer cells . It is known that DLD1 are particularly reliant on mutant KRAS signalling for proliferation in soft agar ( surrogate 3D conditions ) compared with standard tissue culture 2D conditions . HCT116 cells display a similar phenotype , although the difference between WT/mut ( parental ) and WT/- ( KRAS +/- ) genotypes is far less pronounced than DLD1 [24] . We therefore analysed the more sensitive DLD1 system to determine whether ECT2 knockdown affects this phenotype . In our hands , DLD1 KRAS+/- cells proliferated poorly in soft agar , although they grew normally on tissue culture plastic as expected ( Fig 6A ) . We examined the link between KRAS and ECT2 in these conditions . Knockdown of KRAS reduced proliferation in DLD1 parental cells grown in 2D and 3D , but had a greater effect in 3D conditions , as did other published KRAS synthetic lethal targets such as PLK1 and GATA2 ( Fig 6B ) [25 , 26] . ECT2 knock-down in DLD1 parental cells reduced proliferation modestly in 2D , but substantially in 3D conditions , to around 10% of the non-targeting control ( Fig 6B ) , consistent with a senescence response in these conditions , which better model growth in vivo . Finally , to investigate the signalling networks underlying the senescence response to ECT2 knock-down in the presence or absence of KRASG13D , we used network modelling to interrogate gene expression profiles of HCT116 parental and KRAS+/- cells transfected with siRNA targeting ECT2 . A directed network built out from ECT2 using the Shortest Paths algorithm in MetaCore from GeneGo showed shared pathways associated with cell cycle regulation and cytoskeletal remodelling , diverging between the two lines at key hubs that included actin , PKC and RAC1 , a RAS superfamily GTPase ( Fig 7 ) . Diverging networks show reduced signalling from RAC1 towards MYC in HCT116 parental cells , leading to down-regulation of CDC25B and cell cycle arrest , consistent with the robust senescence response . In HCT116 KRAS+/- cells , signalling from RAC1 leads to an up-regulation of cell cycle signalling through cyclin D1 to CDK6 . Expression of CDK6 is associated with G1/S progression; this might contribute to a weaker senescence response , although this warrants further investigation .
The concept of inducing senescence for cancer treatment is an attractive one that has gained interest recently . The presence of residual senescence signalling in various advanced tumour types [4] and the detection of senescence biomarkers in response to standard chemo- and radiotherapy regimens [2] suggests that pro-senescence therapy , either alone or in combination with established therapies , may improve outcomes [3] . Furthermore , the presence of senescence biomarkers in pre-malignant or benign lesions suggests that novel agents that are able to induce senescence in tumour cells might have reduced systemic toxicity and fewer side-effects than standard treatments , improving prognosis where senescence can be robustly achieved . To identify and realise the potential of pro-senescence therapies we need to understand the complex nature of senescence signalling , especially in cancer cells , and to identify robust biomarkers of this response . Here we describe the identification and validation of a panel of siRNAs that induce a spectrum of senescence responses in two cancer cell lines , A375P melanoma and HCT116 colorectal carcinoma . Many of the top siRNA hits targeted genes were involved in cell cycle regulation and maintenance of DNA integrity , consistent with the senescence phenotype . Proteasomal and ubiquitin-protein transferase activities also featured strongly , with a number of proteasome family members in the siRNA target list . This is consistent with reduced proteolytic activity and down-regulation of proteasome β-catalytic subunits reported in human primary fibroblasts senescing in vitro and in ageing human tissues [27] . Moreover , the treatment of cultured primary fibroblasts with specific proteasome inhibitors induces a senescence-like phenotype that includes irreversible growth arrest and expression of SAβGal [28] . In addition we found that siRNA knock-down of AURKB induced nuclear size increase and expression of SAβGal in both A375P and HCT116 cells , consistent with and confirming the results of a parallel compound-based screen for senescence effectors in IMR90 human diploid fibroblasts [20] . Furthermore , siRNA knock-down of AURKB , BUB1B , ECT2 , INCENP , KIF11 and CIT induced large nuclear morphologies that were associated with mitotic defects and growth arrest in a large-scale phenotypic screen in HeLa cells [29] . To validate the observed responses , we investigated the expression of two other established senescence biomarkers , senescence effector p21 [8] , and DDR factor 53BP1 , in 16 of the most robust siRNA hits in both A375P and HCT116 cells . DNA damage components are widely used as senescence markers [6 , 30] . The extent of senescence response following gene knockdown depended on both the genetic background and cell type . Some siRNAs , including those targeting DDB1 , PSMA5 and ECT2 , induced a robust senescence response with expression of all 4 biomarkers in both cell types , while others , such as those targeting CIT , induced a weak response involving only increased nuclear area and SAβGal expression in A375P and little to no induction of p21 or 53BP1 in either cell line . This situation is consistent with a partial senescence response where genes might contribute to senescence evasion but are not crucial [31] . Indeed , considering effects on cell proliferation , a large percentage of hits could induce combinations of senescence associated phenotypes in cell lines lacking key normal senescence effectors p53 ( commonly mutated in various cancer types ) or p21 . Senescence in a p21 null background has previously been shown in murine fibroblasts [32 , 33]; and in HCT116 isogenic cell lines the plant tannin gallotannin was shown to induce a senescence response independent of p53 or p21 expression [34] . Similarly neonatal human melanocytes senesce without expressing p53 or p21 , just p16 [35] and benign nevi likewise have p16 and rarely express p21 [36] . However , in our experiments , induction of the full spectrum of markers including growth arrest was limited to KIF11 in the p53 deficient background , while effects on cell number consistent with toxicity were observed in most cases in the p21 null background . It is well known that many existing toxic and targeted agents are capable of causing senescence in a subset of cells , despite the main phenotype caused being cell death , and this may be the most common mode of senescence induction in p21 null cells . The senescent phenotype results from the complex co-operative interaction of various signalling pathways , so it is not surprising that , in the right context , cells with a particular defect can still respond at least partially [37] . Our study therefore highlights the spectrum of “senescence-like” phenotypes that can arise in different mutational contexts . Activating mutations in KRAS have been described in various cancer types including non-small cell lung and pancreatic cancers and in around 50% of colorectal carcinomas [38] . However RAS proteins have appeared undruggable until recently [39] . Thus research has focussed on targeting RAS signalling pathways and downstream effectors [38] including BRAFV600E and PI3Kα ( PIK3CA ) . Ten of our 16 top siRNA hits in HCT116 cells in which PIK3CAH1047R was deleted induced senescence markers , yet none of them arrested proliferation , notably suggesting dependence of the actual arrest upon oncogenic PIK3CA in HCT116 cells . Analysis of the response in the HCT116 isogenic line lacking the activating KRASG13D mutation revealed that only siRNAs targeting ALDOA and ECT2 were capable of inducing SAβGal and nuclear area in this background . ECT2 also reduced proliferation in this background , though not as robustly as the parental cells . ECT2 ( epithelial cell transforming 2 ) is a known oncogene that is upregulated in various cancer types [40] and associated with poor prognosis in glioblastoma [41] and astrocytoma [42] . Prognostic significance of elevated ECT2 mRNA expression has been evaluated and positively correlated with protein expression in a range of cancer types [43 , 44] . Its regulation is relatively understudied . However , ECT2 expression is known to be transcriptionally regulated through the cell cycle and during DNA damage to control mitosis [45 , 46] . Regulation of ECT mRNA stability via FXR1 may also contribute to FXR1 oncogenic effects [47] , while the tumour suppressor effects of miR-223 may be partly mediated by targeting the ECT2 3’UTR [48] . From a therapeutic standpoint , siRNA knock-down of ECT2 in glioblastoma cell lines caused decreased proliferation , migration and invasion [49] , and mice with U251 astrocytoma cell line xenografts expressing ECT2 shRNAs showed significantly greater survival than non-targeting shRNA controls [42] , making it an interesting target . Our druggability assessment for ECT2 showed no known drugs available to target this gene or other family members; however the protein structure is available and revealed potential druggable sites that could be exploited for drug development . It will be of interest in future studies to develop inducible ECT2 knockout systems to more thoroughly investigate the mechanisms and downstream effects of senescence induction when all cells in the population are targeted . ECT2 regulates cell fate in C . elegans through RAS–MAPK signalling via crosstalk from the RHO-1 pathway [23] . Here , knockdown of ECT2 induced a more complete senescence response in HCT116 cells with KRASG13D , but not in MCF10a non-neoplastic breast epithelial cells irrespective of KRASG13D , suggesting a possible tumour-specific effect . Furthermore , SaβGal was significantly induced in DLD1 parental line with KRASG13D in comparison to the KRAS+/- derivative . These links between ECT2 and RAS signalling make this an attractive and novel target worthy of further investigation for pro-senescence therapy of KRAS mutant tumours . Moreover ECT2 was recently identified in a large siRNA screen as being required for the survival and proliferation of KRAS transformed HCT116 cells , but not those lacking mutant KRAS [25]; and knockdown of KRAS , ECT2 and other KRAS synthetic lethal targets PLK1 and GATA2 , in parental DLD1 cells substantially reduced growth in 3D in our hands . We used network analysis to show that knockdown of ECT2 leads to further mRNA changes in a large number of genes closely associated with its signalling pathway . A similar approach was taken by Long and colleagues to infer ECT2 signalling mechanisms in pancreatic cancer [50] . Our findings point to significant downregulation of Rac1 and PKC signalling , both of which are in line with the known signalling functions of ECT2 , in addition to upregulation of p21 and modulation of multiple other cell cycle and cytoskeletal genes such as vimentin and cdc25 . Together , these results are in line with the effects on senescence that we observed in our screening assays . The screen described here provides proof-of-concept of the ability to induce and detect novel senescence effectors in cancer cells , some with actions enhanced by common oncogenic mutations . Consistent with recent observations , a range of senescence responses was detected , highlighting the importance of using multiple phenotypes to measure the extent of the response induced [3 , 6] . One of our most robust hits , ECT2 siRNA , had the ability to engage a senescence response enhanced by mutant KRASG13D and therefore shows promise as a target for senescence induction therapy for a range of cancers harbouring this mutation . However , to realise the full potential of candidate targets and markers identified in screening , appropriate routes to translation need to be established . We have previously investigated plasma markers of human ageing and DNA damage in gastrointestinal adenocarcinoma patients [51] . Furthermore , we have recently initiated two large , prospective , longitudinal studies to evaluate multiple candidate senescence biomarkers including ECT2 ( UK Clinical Research Network trials 12434 and 12435 ) . Ultimately , validation of novel markers and targets in well-defined human cancer populations will be required to accelerate the field of senescence therapeutics .
Transfections were performed as described in S1 File . 5 days after transfection cells were fixed in 4% paraformaldehyde ( PFA ) or 0 . 5% glutaraldehyde ( pH 7 . 2 ) and stained for analysis of nuclear area , p21 , 53BP1 or SAβGal respectively . Images were captured on the Operetta high content imaging system ( PerkinElmer ) at 10x magnification and 9 fields of view per well and data analysed using Harmony software ( PerkinElmer ) and the parameters outlined in table A of S1 File for nuclear area , p21 and 53BP1 . For all analyses nuclei were detected using a modified “find nuclei” algorithm using method B ( individual threshold 0 . 40 , common threshold 0 . 40 , contrast >0 . 10 ) and split factor 4 . 4 . Border objects were excluded to ensure only whole nuclei were analysed and size criteria to detect DAPI stained objects >40μm2 and <4000 μm2 were applied . Images of SAβGal stained cells were captured on the ArrayScan ( Thermo Fisher Scientific ) using the parameters outlined in table B of S1 File . Transfections were performed as described in S1 File . 5 days after transfection cells were fixed with 4% PFA , permeabilised in 0 . 2% Triton x-100 in PBS and blocked in 10% normal goat serum with 1% BSA . Cells were dual immunostained with primary antibodies against p21 and 53BP1 ( 1:100 in 1% BSA in PBS ) and incubated on a rocking platform at 4°C overnight . Alexa Fluor secondary antibodies were combined at 1:200 dilution in 1% BSA in PBS and incubated on the cells for 1hr at room temperature protected from light . Cells were counterstained with 0 . 1μg/ml DAPI dilactate for 5 min at room temperature and stored at 4°C protected from light . Images were captured on the Operetta high content imaging system ( PerkinElmer ) at 10x magnification and 9 fields of view per well and data analysed using Harmony software ( PerkinElmer ) using the parameters outlined in table A of S1 File . For all assays DAPI stained nuclei were defined as described for nuclear area and this population was then selected for analysis of p21 and 53BP1 as described in table A of S1 File . For MCF10a parental and KRASG13D lines images were captured on the ArrayScan ( Thermo Fisher Scientific ) at 20x magnification using the parameters outlined in table B of S1 File . Further information on screen set-up and reagents can be found in S1 File , Supplemental Methods . Transfections were performed as described in S1 File . 5 days after transfection cells were fixed with 0 . 5% glutaraldehyde ( pH 7 . 2 ) ( Agar Scientific ) , washed with PBS containing 1mM MgCl2 solution and stained with 1 mg/mL X-gal solution diluted in β-galactosidase solution ( 0 . 12mM potassium ferricyanide [K3Fe ( CN ) 6] and 0 . 12 mM potassium ferrocyanide [K4Fe ( CN ) 6] in PBS containing 1mM MgCl2 solution , adjusted to pH7 ( A375P melanoma cells ) or pH6 ( all other cell types ) using 0 . 1M citric acid ) overnight at 37°C in an incubator without CO2 . The assay was terminated by washing 3x in PBS when clear positively stained cells could be detected in the etoposide control wells . Cells were counterstained with 0 . 1μg/ml DAPI dilactate at room temperature and stored at 4°C protected from light . Images were captured on the ArrayScan ( Thermo Fisher Scientific ) using the Brightfield Module and parameters outlined in table B of S1 File to detect cytoplasmic SAβGal staining . All chemicals were from Sigma Aldrich unless stated otherwise . DLD1 parental and KRASG13D/- cells were grown in McCoy’s medium with 10% FBS ( Gibco ) . For the 2D component , cells were seeded at 3000 cells per well in 96 well plates and incubated at 37°C in 5% CO2 . Cell proliferation was quantified using Alamar Blue ( 10% v/v; Invitrogen ) after 72 hours . For the 3D component , plates were coated with a 0 . 6% agar–medium mix and allowed to solidify . Cells were suspended in a semisolid 0 . 4% agar–medium mix , plated at 3000 cells per well and topped with 0 . 6% agar–medium . Cell proliferation was assessed by Alamar Blue ( 10% v/v ) after 7 days . siGenome SMARTpool reagents ( Dharmacon ) were reconstituted and diluted to give a final assay concentration of 25 nM in tissue culture treated ( 2D ) or low attachment ( Corning , 3D ) 96 well plates . These were allowed to complex with Lipofectamine RNAimax ( ThermoFisher ) as per the manufacturer’s instructions . DLD1 cells were diluted to a density of 3000 cells per well , layered on to the siRNA–lipid mix and incubated to allow reverse transfection to take place . Plates were then topped up with either medium ( 2D ) or agar–medium mix ( 3D ) and cell proliferation quantified using Alamar Blue ( 10% v/v ) after 72h ( 2D ) or 7 days ( 3D ) . To determine caspase 3/7 levels , the Promega ( Southampton , UK ) Apo-ONE homogenous caspase 3/7 assay was performed , according to the manufacturer’s instructions . Transfections were performed as described in S1 File . 5 days post-transfection , cells were harvested according to the assay instructions . 100μl per well of caspase 3/7 reagent was added to each well and incubated for 4 hours prior to plate read using Safire II plate reader ( Tecan Trading AG , Switzerland ) . Inflammatory markers including several known components of the SASP were analysed using the Mesoscale Discovery ( Rockville , USA ) human 9-plex pro-inflammatory tissue culture kit ( K15007B ) . Transfections were performed as described in S1 File . Tissue culture supernatants were obtained at 5 days post-transfection . Calibrators and controls were prepared according to the manufacturer’s instructions . 25μl supernatants were incubated for 2 hours at room temperature with shaking in assay plates pre-coated with multiplex capture antibody body cocktails . Detection antibodies were prepared according to the manufacturer’s instructions and 25μl per well of 1x detection antibodies were added to assay plates for 2 hours at room temperature with shaking . After binding of detection antibodies , plates were washed 3 times with the supplied wash buffer and read buffer T added . Plates were analysed immediately on Mesoscale Discovery Quickplex SQ120 plate reader . Hits were taken to be genes for which at least 2/3 siRNA caused induction of cytokines by at least mean + 3SD of scrambled control transfections . RNA was labelled and amplified using the one-colour microarray gene expression analysis protocol ( Agilent Technologies , Santa Clara , CA ) , hybridised to Agilent whole human genome 4 x 44k Agilent whole human genome microarrays and incubated for 17h at 60°C in a hybridisation oven . Arrays were washed on a magnetic stirrer using Agilent wash buffers . Slides were scanned on an Agilent microarray scanner at 5μm resolution , PMT at 100% and 10% . The extended dynamic range setting was corrected for saturation . Microarray data were extracted using Agilent Feature Extraction software ( Agilent Technologies , Santa Clara , CA ) . All array data were analysed in GeneSpring for normalisation and statistical analysis ( Agilent Technologies , Santa Clara , CA ) . Intra-array normalisation was carried out using the 75th percentile for each microarray . Significant differences in expression between scrambled and ECT2 transfected cells were determined in both parental and KRAS+/- HCT116 cells using unpaired t-test . One array was prepared for each of the ECT2 siRNAs against 5 samples from scrambled control transfectants in parental cells . In KRAS+/- cells , 2 scrambled control samples were used . IDs with p<0 . 03 were selected for further analysis . Microarray data are available in the Gene Expression Omnibus with accession number GSE100459 . Differentially expressed genes were analysed using the shortest paths algorithm in MetaCore ( Thomson Reuters ) with ECT2 as a seed object . Individual networks were initially built for differentially expressed genes in HCT116 parental and KRAS+/- backgrounds . These were merged to generate Fig 7 .
|
Cellular senescence is an irreversible arrest of cell proliferation . Senescence is understood to be a universal barrier that all cancers must overcome during their evolution . Developing ways to induce senescence in cancer cells is therefore an attractive strategy to identify targets for cancer therapy . However , a lack of understanding of this complex process has meant that little progress has been made in translating senescence induction into the clinic . Here we describe the identification and validation of a panel of inducers of senescence phenotypes from a large-scale siRNA screen . We show that the senescence response can be induced to varying degrees in genetic backgrounds mimicking common cancer mutations , allowing for a prioritisation of approaches . We also show that partial senescence responses can be triggered in cancer cells in the absence of genes considered by some to be essential for a senescence response . However , distinct differences in proliferation arrest are observed different backgrounds . These results advance our understanding of the complexity of senescence biology . This study has identified several potential targets for drug discovery . Interestingly , our results show the potential for therapeutic intervention in some backgrounds which are so far refractory to current treatments .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"transfection",
"medicine",
"and",
"health",
"sciences",
"senescence",
"gene",
"regulation",
"cancer",
"treatment",
"biomarkers",
"oncology",
"physiological",
"processes",
"developmental",
"biology",
"organism",
"development",
"molecular",
"biology",
"techniques",
"research",
"and",
"analysis",
"methods",
"small",
"interfering",
"rnas",
"specimen",
"preparation",
"and",
"treatment",
"staining",
"gene",
"expression",
"molecular",
"biology",
"aging",
"biochemistry",
"rna",
"carcinogenesis",
"cell",
"staining",
"nucleic",
"acids",
"phenotypes",
"physiology",
"genetics",
"biology",
"and",
"life",
"sciences",
"non-coding",
"rna"
] |
2017
|
A ‘synthetic-sickness’ screen for senescence re-engagement targets in mutant cancer backgrounds
|
A multicellular organism is not a monolayer of cells in a flask; it is a complex , spatially structured environment , offering both challenges and opportunities for viruses to thrive . Whereas virus infection dynamics at the host and within-cell levels have been documented , the intermediate between-cell level remains poorly understood . Here , we used flow cytometry to measure the infection status of thousands of individual cells in virus-infected plants . This approach allowed us to determine accurately the number of cells infected by two virus variants in the same host , over space and time as the virus colonizes the host . We found a low overall frequency of cellular infection ( <0 . 3 ) , and few cells were coinfected by both virus variants ( <0 . 1 ) . We then estimated the cellular contagion rate ( R ) , the number of secondary infections per infected cell per day . R ranged from 2 . 43 to values not significantly different from zero , and generally decreased over time . Estimates of the cellular multiplicity of infection ( MOI ) , the number of virions infecting a cell , were low ( <1 . 5 ) . Variance of virus-genotype frequencies increased strongly from leaf to cell levels , in agreement with a low MOI . Finally , there were leaf-dependent differences in the ease with which a leaf could be colonized , and the number of virions effectively colonizing a leaf . The modeling of infection patterns suggests that the aggregation of virus-infected cells plays a key role in limiting spread; matching the observation that cell-to-cell movement of plant viruses can result in patches of infection . Our results show that virus expansion at the between-cell level is restricted , probably due to the host environment and virus infection itself .
For obligate intra-cellular micro-parasites such as viruses , the cell is the fundamental and minimal unit of infection . Important macro-scale phenomena in viral infection – immunity , virulence , transmission , and evolution – all depend on the infection outcome in individual cells . The biochemical and molecular bases of virus infection have received much scrutiny , and in the past decades there also have been major advances in understanding the dynamics at the host and host-population levels . The next great challenge is a unified picture of virus infection dynamics and evolution that integrates different spatiotemporal scales [1] , [2] . However , integration across different spatiotemporal scales effectively has not occurred across the between-cell level due to practical and methodological considerations . At present , there simply is not a coherent picture of infection dynamics at the between-cell level . A number of key issues have not been addressed adequately to date . First , virus replication in an individual cell can be extremely rapid [3] , [4] , as can the advance of infection and long-range movement [5] . However , little is known about the rate at which infection spreads at the cellular level [6] . What will be the number of newly infected cells per infected cell per day , a value we refer to as the cellular contagion rate ( R ) ? Whereas a reproduction ratio estimates the number of cells directly infected by one cell [6] , the contagion rate estimates the total number of newly infected cells occurring per infected cell over a given time period . For Tobacco mosaic virus ( TMV ) infection of Nicotiana benthamiana plants , a low R was estimated ( 0 . 5–0 . 6 cells/cell/d ) , although why this R value was so low was not discussed [7] . Given the rapid replication and spread of viruses , this result is unexpected and it is not at all clear whether other viruses will adhere to similar patterns . Furthermore , a constant R value was assumed in the analysis described in ref . [7] , whereas a time-varying rate may provide more insights into the underlying dynamics [6] . Another important issue is that individual cells can be observed readily in cell culture systems , whereas gross infection patterns in multi-cellular hosts can be observed by means of virus-induced symptoms , molecular methods [8] or by monitoring infection of tagged viruses [5] . However , these methods do not render information on how the number of infected cells in different tissues changes over time . Finally , variation in genotype frequencies has been described only at higher levels of host organization [9]–[11] . By variation in genotype frequencies , we refer to the differences in the abundance of different virus variants , after a cohort of hosts is initially inoculated with a virus population containing two or more variants . How will this variation change from the population to the individual to the organ , and finally , to the cell ? This variation is pivotal to studying the infection dynamics and evolution of viruses . Within-cell interactions between virus genotypes , such as recombination and the complementation of defective virus genotypes , will require that the presence of two genotypes within a host also carry over to the organ and individual cell levels . Whether genotypes carry over will depend on the genetic bottlenecks a virus population passes through when colonizing organs or infecting a cell , respectively . Plant viruses are ideal model systems for studying virus infection at the between-cell level , and therefore infection dynamics at this level are probably best understood in these systems . The targets of primary infection by mechanical inoculation – epidermal cells – can be readily observed in situ [5] , [12]–[14] , allowing for the tracking of cell-to-cell movement [13] . Moreover , two approaches have been developed to determine whether protoplasts – intact cells extracted after degradation of the cell wall – are infected by different plant virus variants , based on fluorophores [7] or nested PCR [9] . Finally , there is an enviable characteristic of plants: their leaves are natural , biologically relevant compartments that can be removed cleanly ( e . g . [15] ) for further study . The development of plant viruses as model systems to study between-cell infection dynamics has led to important insights and the estimation of some key infection parameters . First , as discussed above , a low R has been estimated for TMV [7] . Second , estimates of the cellular multiplicity of infection ( MOI ) have been made for three plant viruses . For TMV , MOI was found to be low ( MOI<2 ) [7] , [16] . Moreover , in this particular case a substantial proportion of cells ( >0 . 1 ) remain uninfected [7] . However , a model-selection-based analysis of the TMV data suggests MOI might in fact be higher , whilst the number of coinfected cells is low due to spatial segregation of the two virus variants [17] . For Cauliflower mosaic virus ( CaMV ) , MOI was reported to vary from 2 to 13 over time , and most cells were infected [9] . Furthermore , for CaMV virion concentrations in vascular tissue are correlated to MOI [18] . For Soil-borne wheat mosaic virus , MOI was estimated during the first rounds of cellular infection in the inoculated leaf , rendering an estimated of 5–6 [12] . Additionally , low level of potyvirus cellular coinfections suggest a low MOI for potyviruses [19] . Finally , for our model system , Tobacco etch virus ( TEV; genus Potyvirus , family Potyviridae ) , the number of infected cells in systemic tissues early in infection depends on the number of primary infection foci , and the number of infected cells does not increase to a frequency greater than 0 . 5 [15] . Important omissions in our understanding of infection dynamics at the between-cell level remain , however . In particular , a comprehensive view of the between-cell level of infection is missing and the tracking of cell-level infection in multiple host organs or compartments has not been reported . We therefore opted to study these dynamics in TEV and devised an experimental setup in which we could measure infection at the cellular level , which was both sensitive and high-throughput . We opted to analyze the presence of viral variants in individual cells using a flow-cytometry-based method [15] , [20] . This approach allows for quantitative measurements of the number of cellular infections for two virus variants in a large number of mesophyll cells , allowing for an analysis of infection dynamics in different host compartments and at different times . This large dataset allowed us to describe the dynamic pattern of the number of infected cells over time , estimate MOI , quantify R , and consider the variation in genotype frequencies at different levels of host organization as a consequence of bottlenecks .
We generated two TEV variants , TEV-BFP and TEV-Venus , which express blue or yellow fluorescent proteins , respectively . Fluorescent markers inserted in the TEV genome can be stable over multiple short rounds of infection [14] , [21] , and we confirmed the integrity of the marker sequences throughout the experiment ( see Materials and Methods ) . Furthermore , the insertion of eGFP – a variant of the fluorescent protein from which BFP and Venus are derived – in the TEV genome has no effect on virus accumulation after 7 days post-inoculation ( dpi ) ( see Materials and Methods ) . Therefore , these marked viruses have biological properties similar to the wild-type virus from which they are derived . We rub-inoculated the third true leaf of Nicotiana tabacum L . cv . Xanthi plants with a 1∶1 mixture of infectious saps ( ground tissue in inoculation buffer ) of the two variants . We then isolated protoplasts [15] , [20] from the third , fifth , sixth , and seventh true leaves at 3 , 5 , 7 , and 10 dpi , with five replicate plants for each time point . We did not analyze the fourth true leaf because under the current experimental conditions this leaf does not show any infection . Flow cytometry was used to determine which cells were uninfected , infected by one or by both virus variants . Using this approach we could quantitatively measure the distribution of cellular infection over space and time , for the two virus variants . The frequency of virus-infected cells was low ( mean ± 1 SD: 0 . 072±0 . 099 ) , with the highest level of infection observed in any one sample being 0 . 424 ( Leaf 7 at 10 dpi ) ( Figure 1A–D ) . The frequency of cells infected by both virus variants was also low ( mean ± 1 SD: 0 . 012±0 . 023 ) , with the highest level of coinfection observed in any sample being 0 . 112 ( Leaf 6 at 7 dpi ) ( Figure 1A–D ) . These low levels of coinfection are in agreement with previous studies on plant RNA viruses [7] , [13] , [19] , and suggest that MOI is low . Few cells were infected in any leaf at 3 dpi , with the greatest number of infections being found in Leaves 3 and 6 . This surprising observation can be explained by the occurrence of limited , relatively slow TEV expansion at the macroscopic level in the inoculated leaf [8] , combined with fast egress ( <2 dpi ) from Leaf 3 to Leaf 6 at high viral doses [15] . Both infection and coinfection appear to increase over time in the different leaves , although Leaf 5 shows very low levels of infection . Infection progresses slower in Leaf 3 than in Leaves 6 and 7 . Leaf 6 becomes infected before Leaf 7 , but the dynamics in these two leaves are otherwise very similar . The frequency of TEV-Venus infected cells was significantly higher than expected for a 1∶1 inoculum ( one-sample t-test: t79 = 4 . 141 , P<0 . 001 ) , although the magnitude of the deviation was small ( mean Laplace point estimator for the frequency of TEV-Venus infected cells ± 1 SD = 0 . 591±0 . 196 ) . This deviation could occur because of a small discrepancy in the inoculum ratio , or a small difference in infectivity or in within-host competitive fitness of the two variants . To confirm that infection levels in Leaf 7 had saturated at 10 dpi , in a separate experiment we also analyzed infection in Leaf 7 at 13 dpi . The observed frequency of virus-infected cells was slightly lower than at 10 dpi , although the difference was not statistically significant ( two-sample t-test: t8 = 1 . 251 , P = 0 . 246 ) . The data therefore suggest that infection levels had saturated in all analyzed leaves by 10 dpi . To visually illustrate patterns of infection , we infected plants with TEV-eGFP and TEV-mCherry [14] under identical conditions . These viruses were used here , instead of TEV-BFP and TEV-Venus , because their fluorescent proteins are more suitable for microscopy . Even when infection appears to have saturated at both the cell and visible fluorescence level , we could see heterogeneities in the distribution of virus infection over the leaf at different spatial levels ( Figure 1E–G ) . We estimated the time-varying cellular contagion rate ( R ) from the data using a simple maximum likelihood method . This analysis was carried out on the total number of infected cells , regardless the virus variants present . For R>0 the number of infected cells increases , whereas for R<0 it decreases . Our estimates of R for individual leaves ( Figure 2A–D ) ranged from 2 . 43 cells/cell/d ( 95% CI: 1 . 80–3 . 39 ) ( Leaf 6 , 3 dpi; ) to values not significantly different from zero ( e . g . , −0 . 327 cells/cell/d ( 95% CI: −0 . 539–0 . 271 ) for Leaf 5 , 7 dpi ) . We do not expect R<0 in this system , since infection is not cleared and the number of infected cells can therefore not decrease . Our approach might slightly overestimate R in individual leaves because of between-leaf transmission , and we therefore also estimated R for pooled data from different leaves ( Figure 2E ) . One disadvantage of this approach is that tissues with high infection levels will most strongly affect R estimates . These estimates of R ( mean [95% CI] ) ranged from 1 . 342 cells/cell/d [0 . 247–1 . 371] , 3 dpi , to 0 . 196 cells/cell/d [0 . 041–0 . 244] , 7 dpi , and were always significantly greater than zero . Overall , values of R appear to be surprisingly low given estimates of the rapid rate of cell-to-cell movement for TEV during initial infection , whilst they are similar to estimates of R for TMV ( 0 . 5–0 . 6 cells/cell/d ) [7] . Low R values may therefore be commonplace in plant RNA viruses , although data from more pathosystems will be needed to confirm this idea . Dolja et al . [5] observed that a primary infection focus starts with a single infected cell and grows to formation with a diameter of ten infected cells within 24 h , and hence cells/cell/d . This calculation is conservative and underestimates R because infection in the first infected cell cannot be observed at t = 0 , and because it only takes into account infection in the epidermal cells . Note that such a high value – which probably far exceeds the number of other cells to which each cell is plasmodesmally connected [7] – is possible because of multiple rounds of replication can occur within a single day [5] . The R values we have measured are therefore extremely low compared to R values found in the inoculated leaf during early infection . We wanted to test whether our understanding of the process that is likely to govern cell-level infection patterns was congruent with our empirical data . Specifically , we wanted to test whether there were leaf-dependent differences in key infection parameters , and whether there was evidence for aggregation of virus-infected cells limiting infection spread . We therefore developed a simple susceptible-infectious ( SI ) model of within-host infection dynamics . Each leaf in a plant represents a physically separated compartment - with its own physiological state - that a virus must colonize [22] . We therefore developed a simple meta-population dynamics model with between-leaf transmission from lower leaves to upper leaves . For the kth leaf , the rate of change of the fraction of infected cells ( Ik ) is: ( 1 ) where β is the within-leaf transmission coefficient ( from cell to cell ) , χ is the between-leaf transmission coefficient and S is the fraction of susceptible cells . Between-leaf transmission depends on the total fraction of infected cells in the leaves below the kth leaf , given that systemic-movement for phloem-transported viruses is towards the apical sink leaves [5] , [22] . Potyvirus infection appears to be marked by the aggregation of infected cells [19] , and given that plant cells will largely retain their respective positions in developed leaves , the perfect mixing assumptions of the SI model will not be met . We therefore included a spatial aggregation factor of infectious units ( i . e . , infected cells ) ψk in the model , such thatBy spatial aggregation of infected cells , we mean that infected cells are likely to be found together and are therefore not randomly distributed in the leaf . The mechanism resulting in the spatial aggregation of infected cells is probably the dependence of plant viruses on cell-to-cell movement for local infection to spread: the spread of virions , or in some cases unencapsidated genomes , from an infected cell to its direct neighbors [5] , [13] . When ψk = 1 there is perfect mixing , whereas when ψk approaches 0 there is maximum aggregation of infected cells [23] , [24] . The model was fitted using maximum likelihood methods , and model selection was performed to ensure the data supported the inclusion of all model parameters ( see Materials and Methods ) . As with the estimates of R , this analysis was carried out on the total number of infected cells and does not distinguish between the two virus variants . The SI meta-population model could describe the data well , clearly capturing the main trends in the data ( Figure 2F ) . Spatial aggregation of infected cells ( ψk ) was indispensable to the model ( Table S2 ) , and parameter estimates varied over leaves; ψk was most pronounced in Leaves 3 and 5 , and much lower in Leaves 6 and 7 ( Figure 2G ) . The between-leaf transmission coefficients ( χk ) for Leaves 5 and 6 were similar , although infection never reaches even moderate levels in Leaf 5 . χ7 was significantly lower than χ6 ( non-overlapping 95% CIs of parameter estimates ) , although the number of infected cells in both leaves reached moderate levels eventually . Parameter estimates therefore suggest that infection dynamics vary for each leaf , even though the overall pattern ( Figure 1A–B ) is similar for Leaves 6 and 7 . The cellular MOI can be estimated from our data , as has been previously done for two plant viruses with a similar experimental setup [7] , [9] . However , estimates of MOI can be influenced by the estimation method [17] . Model selection was therefore performed on a set of nine MOI-predicting models ( see Materials and Methods ) , by testing which Poisson-based model best predicted the relationship between the fractions of uninfected and coinfected cells ( i . e . , those cells infected by both virus variants ) [17] . The models incorporated spatial segregation of virus genotypes , spatial aggregation of infected cells , superinfection exclusion at the cellular level and combinations of all these effects . We could thereby identify the best model to generate MOI estimates ( Tables S3 and S4 ) . The best-supported model incorporated only a leaf-dependent aggregation factor ψ ( Table S4 ) . The MOI and SI model selection results are therefore in good agreement with each other , although estimated ψ values were higher than those obtained from the SI model ( Figure 2G ) , indicating less aggregation ( Figure 3A ) . These two separate model selection procedures therefore confirm the importance of the spatial aggregation of infected cells for understanding TEV infection dynamics at the between-cell level , as might be expected for a virus that spreads by cell-to-cell movement . On the other hand , in a similar model-selection-based analysis for TMV and CaMV MOI , two viruses that also move by cell-to-cell movement , spatial aggregation only marginally improved model fit for both datasets [17] . These two different model-selection results suggest that whether cell-to-cell movement really has an impact on MOI estimation will depend not only on the mechanism of movement . Other factors , such as the number and distribution of initially infected cells , and the frequency of infected cells , also may play an important role . We then derived predictions of MOI using the best-supported model ( Figure 3B ) . As could be expected from the low frequencies of cellular infection and coinfection ( Figure 1A–D ) , the predicted MOIs were low , ranging from 1 . 001 ( Leaf 5 , 3 dpi ) to 1 . 432 ( Leaf 6 , 7 dpi ) . Note that we report the estimated MOI value in infected cells only ( i . e . , mI in Materials and Methods ) , which has a minimum value of 1 . The corresponding range of MOI values calculated over the whole population of infected and uninfected cells ( mT ) is 0 . 002 ( Leaf 5 , 3 dpi ) to 0 . 735 ( Leaf 6 , 7 dpi ) . Although these estimates may seem low intuitively , MOI is assumed to follow a Poisson distribution over cells and some cells can still be infected by two or more virions , even when the mean of the distribution is low ( Figure 3C–E ) . Our estimates of MOI are similar to the low estimates for TMV [7] , [16] , although model-selection-based estimates for the TMV data result in MOI values ranging to somewhat higher values ( up to 2 . 1 ) , due to the predicted occurrence of spatial segregation of virus genotypes [17] . For CaMV much higher MOI values were observed later in infection [9] , but in our system infection levels remain low even then . The experimental data also allow us to consider variation in the frequencies of viral genotypes at different levels of the host: leaf ( Figure 4A–D ) , cells coinfected by both virus variants ( Figure 4E–H ) , all infected cells ( Figure 4I–L ) , but also at the level of the host-plant population ( Figure 4M ) . Variance of TEV-Venus frequencies appears to increase strongly from the plant and leaf levels to the individual cell level ( Figure 4A–M ) . The log-transformed genotype ratios ( TEV-Venus∶TEV-BFP ) in individual cells appear to be independent of the frequency of TEV-Venus in the leaf ( Figure 5A ) , indicating a decoupling of processes occurring at the leaf and coinfected-cell levels . Low estimates of MOI ( Figure 3B ) imply that the virus population entering each cell is subject to a narrow genetic bottleneck . A decoupling of the infection processes at the leaf and cell levels is predicted to occur because very few cells are infected by more than 2 virions ( Figure 3E ) . Hence , for the vast majority of coinfected cells the frequency of virus variants , as represented by the infecting virions , is limited to 1/3 , 1/2 and 2/3 . If our MOI estimates are correct , than stochasticity in the replication process within the cell accounts for high levels of variation . In line with these expectations , we observed high levels of variation in virus variants at the cellular level ( Figure 4E–H ) and a distribution of variants in coinfected cells that is independent of the frequency of virus variants in the leaf ( Figure 5A ) . Note that there are couplings between the leaf and cell-level dynamics ( i . e . , MOI depends on the overall level of infection for the best-supported MOI models; see Materials and Methods ) , but our observations show that not all leaf-level characteristics of the virus population carry over to individual cells . Finally , we estimated the effective population size , Ne , for individual leaves and the whole plant [25] ( see Materials and Methods ) . For the inoculated leaf we obtained a Ne estimate of approximately 100 ( Figure 5B ) , corresponding well to the approximate number of primary infection foci observed . For Leaf 6 , Ne was also estimated to be approximately 100 , although the confidence interval extends to ∞ and there is no evidence for a genetic bottleneck in this leaf . For Leaves 5 and 7 , much lower estimates of Ne were obtained , suggesting that fewer virions infect these leaves and that it is more difficult for the virus to invade these compartments . A wide range of within-host effective population sizes at the leaf level has been reported for different viruses [10] , [11] , [18] , [26] . Here we show a similar range of effective population sizes can occur with a single virus-host combination , probably due to the combined effects of host physiology , anatomy and immunity .
To link infection dynamics at the cell and host levels , we have measured the number of cells infected by two virus variants within individual plants over time and space . We have estimated R ( the cellular contagion rate , expressed as newly infected cells per infected cell per day ) over time for systemic virus infection . A conservative estimate of the maximum value for R is 1 . 4 cells/cell/d on day 3 ( Figure 2E ) , and it falls to just under 0 . 2 cells/cell/d by day seven . These values are comparable to estimates of R for TMV infection of N . benthamiana of 0 . 5–0 . 6 cells/cell/day , although in this instance a constant R was estimated [7] . We can therefore conclude that for our model system , and perhaps more generally for plant RNA viruses , R is very low during systemic infection , suggesting that most cells will transmit virus to one or possibly even zero other cells during infection . Here we have estimated the cellular contagion rate over a period of one day . Given that TEV infection has been reported to expand at a rate of one row of cells every 2 h [5] , it is entirely possible that multiple rounds of infection will occur during one day . Therefore , the reproduction ratio at the cellular level ( i . e . , the number of cells to which one infected cell spreads infection over its lifetime ) is probably similar to , or even lower than our estimates for the cellular contagion rate . These estimates are in principle the aggregated effect of local cell-to-cell movement and long-range systemic movement . What then accounts for these surprisingly low values of R , and are they reconcilable with high replication rates at the molecular level [3] , [4] and fast virus expansion throughout the plant [5] , [15] ? Decreases in cellular replication because the carrying capacity for infection has been reached do not explain these observations: low R values were estimated when infection levels were very low ( e . g . , compare Figures 2C and 2F ) . However , contagion rates at the cellular level can be much higher than those we have observed here: based on other results [5] we also estimate that during expansion in primary infection foci R≈78 cells/cell/d . We have observed early infection in systemically infected leaves that eventually reached high levels ( i . e . , Leaves 6 and 7 ) , and especially in the case of Leaf 7 these infections appears to be initiated by a small number of virions . Hence , ceteris paribus we would have expected high R levels in these leaves as well , and moreover in Leaf 7 R does not reach the same levels as Leaf 6 . These observations implicate two processes in slowing the observed rate of virus expansion at the between-cell level . First , host immune responses , particularly RNA silencing [27] , is very likely to play a role . Moreover , since a specific RNA silencing signal progresses systemically to sink leaves [27] , [28] , we speculate that this may explain why there appear to be lower R levels in Leaf 7 than in Leaf 6 . Second , our experimental approach limits us to analyzing the cells in a leaf as a whole , whereas the analysis of cells in the infection front would result in higher R values . We found striking differences in infection dynamics in different leaves ( Figure 1A–D ) . These differences were also reflected in estimates of parameters for the different models fitted to the data ( Figures 2B , 3A and 5B ) . What can account for the infection dynamics in different leaves ? First , sink-source transitions will play a major role in determining if and to what extent leaves can be colonized , because phloem-transported viruses cannot cross the sink-source boundary in any leaf [22] . This functional boundary separates the basal part of a developing leaf , which is importing photo assimilates , from the distal part that is already exporting them . Furthermore , sink-source transitions may further impact the spatial aggregation of infected cells on a smaller spatial level: sink-source transitions will determine from which classes of phloem the virus can unload , with much less restriction in smaller veins prior to the transition [22] . Hence the distribution of initially infected cells is likely to be more homogeneous – also on small spatial scale – in sink leaves , leading to less spatial aggregation of infected cells . We saw infection only in the basipetal region of Leaf 5 , whereas about half of the surface of Leaf 6 became infected ( Figure 1E ) . Therefore , we think that Leaf 4 has probably completed the sink-source transition , and is almost exclusively exporting photo assimilates , whereas it has not affected much of Leaf 7 . These assertions on the physiological state of these different leafs are strongly supported by measurements of polyamine levels [29] , which are molecular markers of proliferating source tissues . Putrescine and spermidine levels show that for N . tabacum cv . Xanthi of the same development stage as our plants at inoculation , the sink-source transition is virtually complete in Leaf 4 , almost complete in Leaf 5 , and has not yet commenced in Leaf 6 . Note that whereas sink-source transitions probably account for virus aggregation on a large and intermediate scale ( Figure 1E–1F ) , RNA silencing probably impedes infection at all scales , also resulting in aggregation of infected cells on the smallest scales ( Figure 1G ) [27] . Second , crossing from leaves at one side of the plant to the opposite can be hindered by the phloem connections between leaves [22] . Third , as aggregation of infected cells is increased , the rate of virus spread decreases [23] and the plant will have more time to mount an effective response [27] . Consequently , we hypothesize that large effective population sizes can only be achieved if ( i ) the virus can be readily transmitted between two particular leaves , and ( ii ) the subsequent aggregation of infected cells is moderate to low ( e . g . , the virus is not restricted to the basal part of the leaf by the sink-source transition ) , allowing infection to expand beyond the initial point of entry . Based on these other studies , we therefore speculate on what processes can account for the leaf-dependent differences we have observed . Infection progresses relatively slow in Leaf 3 , probably because under the conditions used the virus only expands locally and egresses from this source leaf [8] . Leaf 4 probably never shows any infection in our setup because it has completed the sink-source transition , and is moreover located opposite Leaf 3 ( for leaf positions see Figure 1 ) . Leaf 5 has a relatively high between-leaf transmission , strong aggregation , and a small bottleneck size ( Figures 2 , 3 and 5 ) . Its position directly above the inoculated leaf explains high between-leaves transmission , whilst the nearly complete sink-source transition results in high aggregation , low levels of infection and therefore a de facto genetic bottleneck . In line with this explanation , the highest levels of aggregation were observed in Leaf 5 , suggesting the virus expansion is very constrained in this leaf . Leaf 6 has a high between-leaf transmission due to its position above the inoculated leaf . Moreover , because the sink-source transition is far from complete there are high levels of infection , moderate aggregation and no genetic bottleneck . Finally , Leaf 7 is positioned on the far side of the plant , with respect to the inoculated leaf , and the increasing intensity of host immune responses results in low between-leaf transmission , and hence a genetic bottleneck occurs . However , since the sink-source transition is far from complete , those viruses that do enter the leaf can expand prolifically , resulting in lower estimated levels of aggregation and high infection levels . In summary , we think that plant anatomy and physiology may largely explain the leaf-dependent differences in infection patterns we have observed , although our explanation will require further testing . Our analyses of infection spread and MOI support the idea that aggregation of virus-infected cells is also important for understanding dynamical patterns and therefore low R values . If there is aggregation of virus-infected cells , which is concurrent with potyviruses achieving local spread by cell-to-cell movement , only those cells on the edge of an aggregate can contribute to virus expansion , and even fortuitously situated cells may not actually infect those susceptible cells they are in contact with before neighboring cells do . The limitations on virus spread from an individual cell to its neighboring cells due to the overall rapid spatial spread of the virus is an effect we refer to as “self-shading” . The importance of self-shading in limiting between-hosts spread [23] , [30] , and its implications for virulence evolution [31] , [32] , have been recognized on larger spatial scales . Our results stress the importance of extending these concepts to within-host dynamics , although we anticipate that there will be differences in the between-host and within-host levels . For example , we hypothesize that a high cellular contagion rate may not incur a major cost in our model system; host cells are static and once a tissue has been infected there are no possibilities for further within-host spread , except for phloem loading in a minority of cells . Therefore , we speculate that aggregation and self-shading will , in this case , impose selection for fast viral replication and spread at the within-host level . Our experimental approach consisted of the isolation of protoplasts , followed by measurements on individual cells by flow-cytometry . Advantages of this approach are its amenability to high-throughput , the high sensitivity of the flow-cytometer , and the fact that mesophyll cells – the primary targets of virus replication – can be analyzed . Disadvantages are the fact that sampling is destructive , and hence a time course cannot be analyzed , and the spatial information is lost during protoplast extraction . Compared to other techniques available for analyzing protoplasts [7] , [9] , the approach used here has a much higher throughput . Our approach may have a higher sensitivity than microscopy [7] , although PCR-based methods are probably more sensitive [9] . Another alternative approach to analyze virus infection dynamics would have been microscopy on whole leaves , which renders spatial information and allows for longitudinal analyses [13] . Although this approach works very well in the inoculated leaf [13] , it is not clear how well it would function in systemic leaves , and this is also a lower-throughput method . For a comprehensive analysis such as we have presented , the high-throughput nature of the assay is essential and dictated our choice of experimental approach . For many other virus-host pathosystems , including those that result in disease in animals such as humans , important spatial characteristics of virus-plant pathosystems may be absent . Short-range virus infections can typically be achieved by diffusion of virions instead of cell-to-cell movement , and most host organs will not have the planar anatomy of leaves . However , there are general characteristics of virus-host interactions that suggest infection aggregation may be a very commonplace phenomenon . First , there are many physical barriers to virus expansion , structuring the host environment and naturally favoring aggregation of infected cells . Second , many viruses replicate in a limited number of cell types or tissues , thus leading to spatial aggregation . Third , epithelia are often targets of viral entry and one of the sites of replication , and consist of highly planar structures . Finally , even for free virions , diffusion and virion removal rates will determine at what distance infection tends to spread . Based on our results and these general considerations , we therefore speculate that aggregation of virus-infected cells and self-shading are likely to be key ingredients for cell-level infection dynamics in a broad range of intra-cellular pathogens infecting complex , multi-cellular hosts .
|
A great deal is understood about how a virus infects an individual cell and manages to replicate . Patterns of disease progression in plant and animal hosts , such as virus titers and the appearance of symptoms , have also been described in great detail . On other hand , very little is known about what is happening at the intermediate levels during virus infection . Here , we use flow cytometry , a technique to rapidly measure large numbers of individual cells , to quantify the number of cells infected by a plant virus , in different leaves and at different times . We found that few cells become infected , and only one or two virus particles typically initiated cellular infection . Moreover , viruses from an infected cell will infect only one or two other cells . Therefore , although viruses replicate at astronomical rates within a cell , their rate of spread between individual cells can be much slower .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion"
] |
[
"mutation",
"plant",
"microbiology",
"genetic",
"polymorphism",
"effective",
"population",
"size",
"disease",
"dynamics",
"microbial",
"evolution",
"virology",
"natural",
"selection",
"genetics",
"population",
"dynamics",
"population",
"genetics",
"host-pathogen",
"interaction",
"biology",
"microbiology",
"evolutionary",
"biology",
"population",
"biology",
"evolutionary",
"genetics",
"genetic",
"drift"
] |
2014
|
Within-Host Spatiotemporal Dynamics of Plant Virus Infection at the Cellular Level
|
The “reproductive ground plan” hypothesis ( RGPH ) proposes that reproductive division of labour in social insects had its antecedents in the ancient gene regulatory networks that evolved to regulate the foraging and reproductive phases of their solitary ancestors . Thus , queens express traits that are characteristic of the reproductive phase of solitary insects , whereas workers express traits characteristic of the foraging phase . The RGPH has also been extended to help understand the regulation of age polyethism within the worker caste and more recently to explain differences in the foraging specialisations of individual honey bee workers . Foragers that specialise in collecting proteinaceous pollen are hypothesised to have higher reproductive potential than individuals that preferentially forage for nectar because genes that were ancestrally associated with the reproductive phase are active . We investigated the links between honey bee worker foraging behaviour and reproductive traits by comparing the foraging preferences of a line of workers that has been selected for high rates of worker reproduction with the preferences of wild-type bees . We show that while selection for reproductive behaviour in workers has not altered foraging preferences , the age at onset of foraging of our selected line has been increased . Our findings therefore support the hypothesis that age polyethism is related to the reproductive ground plan , but they cast doubt on recent suggestions that foraging preferences and reproductive traits are pleiotropically linked .
Two of the most challenging questions to students of social insects are the evolutionary origins of the worker caste [e . g . , 1–6] and the regulation of division of labour within the worker caste [e . g . , 7–13] . It has been suggested that the solution to both these puzzles may lie in modifications of the basal reproductive cycle of solitary insect species . West-Eberhard [14] and Gadagkar [15] have argued that in social species , queens and workers have lost different parts of the original reproductive cycle of solitary species . In some solitary insects , females oscillate between an oviposition phase when ovaries are active , and a foraging phase where the female gathers food for her developing larva ( e ) . This oscillation is regulated by cycling levels of juvenile hormone , ecdysteroids , and vitellogenin [16 , 17] . During the reproductive phase , ovaries are active and titres of circulating vitellogenin—the egg protein precursor—are high . During the foraging phase , the insect's ovaries are nonactive , and circulating vitellogenin titres are low . After her larvae have begun to pupate , the adult solitary female may revert to a new ovulatory phase where her ovaries become reactivated and foraging is curtailed . West-Eberhard and Gadagkar propose that in some social wasps like Polistes , workers express only the foraging phase and the reproductive phase has been lost , whereas in queens , the reproductive phase is expressed and the foraging phase has been lost . We call this the original “reproductive ground plan” hypothesis ( original RGPH , Table 1 ) . Hunt and Amdam [18] proposed a modification of this idea , arguing that in Polistes wasps at least , queens and workers arise from two developmental pathways that characterise the bivoltine lifecycle of some solitary species ( bivoltine RGPH , Table 1 ) . They suggest that queens evolved via co-option of the gene regulatory networks that are switched on in late-emerging , second-generation individuals that diapause , and the worker caste from networks switched on in early-emerging first generation individuals that do not enter diapause . West-Eberhard [19] extended her original RGPH [14] from being solely an explanation of the evolution of caste by suggesting that the changes in behavioural phenotype that typically occur as social insect workers mature may also have their antecedents in the reproductive ground plan of their nonsocial ancestors , i . e . , between the reproductive and nonreproductive phases of a solitary female's adult life . This idea stems from the fact that in some social species , young nest-bound workers retain characteristics of solitary females in their reproductive phase: some ovary activation , aggressiveness towards intruders , and in-nest work , rather than foraging . Later in life , workers cease larval feeding and engage in foraging , retaining some features of the nonreproductive phase , such as inactive ovaries . This idea has found support in the honey bee , Apis mellifera , where workers perform nest-bound tasks early in life and start foraging when older [20–22] . Furthermore , young honey bee workers that are engaged in brood care have high levels of circulating vitellogenin and may have some thickening of their ovaries . Older individuals engaged in foraging have low vitellogenin titres , and their ovaries are completely regressed [23] . We call this the modified RGPH ( Table 1 ) . The most recent version of the RGPH argues that in honey bees , division of labour within the foraging population ( rather than between queen and worker castes or between nursing and foraging workers ) also has its antecedents in the gene regulatory network that once regulated the gonotrophic cycle in solitary ancestors of the social bees [9 , 17 , 24 , 25] . Within a honey bee colony's population of foragers , there is variability among individuals for foraging preferences , and this specialisation has a genetic component ( reviewed in [7 , 12 , 13] ) . In particular , some foragers specialise in collecting pollen , whereas others specialise in collecting water or nectar . It is argued that the basal cycles of reproduction and foraging now regulate division of labour between pollen and nectar foragers and drive foraging behaviour [17] . Especially important here are the genes that regulate a bee's degree of attraction to the concentration of sugar in nectar , and the age at which workers make the transition from in-nest tasks to foraging and other tasks external to the nest [17 , 26 , 27] . We refer to this as the forager RGPH ( Table 1 ) . Experimental support for the explanatory power of the forager RGPH in the evolution of foraging specialisation in the honey bee has come from studies of two lines selected by R . E . Page , Jr . and colleagues for high and low pollen hoarding [25 , 28 , 29] . Relative to workers from the low–pollen hoarding line and to unselected workers , workers from the high–pollen hoarding line start foraging early in life , carry pollen more frequently , carry larger pollen loads , have more ovarioles , are more responsive to low concentrations of sucrose , have elevated levels of vitellogenin , and are more likely to show pre-vitellogenic swelling of their ovaries [24 , 25 , 28 , 30–35] . This divergence between the behaviour and reproductive physiology of the two selected lines has been interpreted within the conceptual framework of the forager RGPH as a demonstration of how the “behavioral mechanisms of division of labour evolve from solitary ancestry , and provides an experimental demonstration of the origins of sib-care behavior from maternal reproductive traits” [24] . Thus correlations between reproductive characters such as ovariole number and levels of ovary activation—however slight—with components of foraging behaviour such as preference for pollen over nectar and the age at which foraging commences , are interpreted as showing that division of labour in foraging is mediated pleiotropically by the same gene networks and hormonal cascades that mediate reproductive behaviour in workers [17 , 24 , 25 , 36] . Thus it is argued [17 , 24] that selection for high pollen hoarding has resulted in a line that displays characteristics of the reproductive phase of the solitary insect's life cycle in which they actively seek protein . In contrast , the low–pollen hoarding line is thought to show characteristics that reflect solitary insects in their nonreproductive phase where they seek carbohydrate . We investigated the link between reproduction and reproductive division of labour by studying the foraging behavior of a line of “anarchistic” ( AN ) honey bees that has been selected for worker reproduction and in which about 1/3 of 10-d-old queenright workers show activation of their ovaries and the presence of oocytes in their ovarioles [37–40] . We evaluated the predictions of the various versions of the RGPH ( Table 1 ) by a comparison of the foraging behaviour and reproductive physiology of workers of this line with that of unselected wild-type ( WT ) workers . We found experimental support for the modified RGPH but no support for the forager extension of the original hypothesis .
We performed our experiment in duplicate , once in January 2007 and once in November 2007 . For each replicate , we chose two WT colonies of standard Australian commercial stock ( primarily Apis mellifera ligustica in origin ) headed by open-mated queens and two AN colonies from the line maintained at the University of Sydney via artificial insemination [37 , 40] . Workers were confirmed to be reproductively active in the AN colonies by the presence of drone brood laid by workers . We confined the queen in each of the four colonies on an empty brood comb by means of a cage constructed of queen excluder material that allows workers to pass through , but not the larger queens [41] . The queens were confined to the combs for 2 d , after which we transferred the four egg-laden combs into a single WT colony in a part of the nest in which the host queen could not enter because of a queen excluder . The genotype of the rearing colony is known to have minimal effect on the expression of anarchistic traits [38] , but we used uniform rearing environments to minimise any possible rearing effects . The host colonies reared the eggs to the late pupal stage , whereupon we transferred the brood combs into an incubator at 34 . 5 °C and high relative humidity . Over 2 d , we marked 1 , 000 emerging workers from each of the four source colonies with coloured paints ( Posca Posta Pens , Japan ) on their thorax to identify each worker as to its colony of origin and day of emergence . For each replicate , we transferred the 4 , 000 marked workers into a single , WT colony , unrelated to any of the source colonies nor the rearing colony , comprising six combs of adult bees and brood , and moved the host colony to Pearl Beach , NSW , Australia , for study . During both replicates , pollen and nectar were abundantly available from a variety of native and introduced species , especially Angophora floribunda in January , and An . hispida during November . In both replicates , some marked bees were seen at the entrance to the host colony 5 d after introduction , and we commenced collecting all marked workers that returned to the colony beginning on day 6 to determine the age at which workers of each source commenced foraging . To do this , we reduced the width of the entrance of the nest to 2 cm with a block of wood . Using forceps , we grasped all paint-marked returning foragers and placed them individually in microcentrifuge tubes . We placed the tubes containing the workers on a frozen “blue ice” brick in a polystyrene box to cool them to immobility . Except when foraging was interrupted by rain , we collected foragers from 09:00 to 13:00 , until we had collected approximately the first 100 bees to forage from each source colony . We avoided collecting foragers in the afternoon when workers make their first orientation flights but do not forage at flowers [42] . The day at which the first foraging workers were collected was used as a measure of age at first foraging . This estimate is skewed to the earliest foragers within each group , and is not an average age at first foraging . However , we wished to leave workers to mature further so that we could obtain an unbiased estimate of their foraging preference – for nectar or for pollen . We preferred to use experienced foragers , because we saw almost no pollen in our samples of first-foraging bees ( a few workers carried traces of pollen ) . This was despite the fact that host colony foragers collected large pollen loads . When the workers were 21 d old , we commenced collecting a second set of workers . Most workers are foragers by this age [30 , 42–44] . We used these workers to determine the foraging preference for nectar and pollen of AN and WT workers . Again we collected returning foragers between 09:00 and 13:00 . For each replicate , we collected ten workers derived from each source colony on each of 4 d . After collecting the final set of foragers , we moved the host colony 10 m away from its original location and replaced it with a dummy hive containing a comb of sealed brood . Any experienced foragers that left the colony to forage returned to the dummy hive at the original location . After the host colony had been moved aside for 6 h , we opened it and collected all of the remaining marked bees . We assume that most of the bees in this sample had never foraged . For the sample of foragers collected at age 21–26 d ( on some days , we could not collect foragers because of rain ) , we determined for each bee: ( 1 ) the volume of nectar she carried in her crop; ( 2 ) the concentration of sugar in the solution carried ( if any was present ) , and ( 3 ) the mass of any pollen she carried . To retrieve nectar from the crop of each bee , we gently squeezed her abdomen between thumb and forefinger and caused her to regurgitate the contents of her crop . We drew the contents of the crop into a 50-μl microcapillary tube ( Drummond Scientific ) and then measured the length of the liquid column with a ruler . We then converted this length into a volume in μl [45] . Where more than 2–3 μl were retrieved , it was generally possible to obtain a measure of the mass of dissolved sugars in the regurgitated crop contents using a hand-held refractometer with a range of 0–80 brix ( Meopta , Taiwan ) . To determine the mass of the pollen loads , we scraped the pellet off the bee's basitarsi into a pre-weighed microcentrifuge tube . We then weighed the tube and pellet and deducted the mass of the tube . Where only one pellet was retrieved , we doubled the mass of the single pellet . To understand the effects of ovary activation and number of ovarioles on worker foraging behaviour , we dissected two groups of workers as follows: ( 1 ) All 160 experienced foragers ( i . e . , aged 21–26 d old , 40 from each source colony ) . This sample allowed us to explore relationships between pollen and nectar foraging preference , degree of ovary activation , and ovariole number . ( 2 ) A matching sample of 160 “non”-forager workers , 40 from each source colony , caught from inside the host colony after this had been moved aside , causing foragers to return to the dummy hive . These dissections allowed us to explore how late onset of foraging ( or lack of foraging ) was associated with ovary activation and ovariole number . To dissect workers , we pinned them to a wax plate through the thorax . We then pulled the abdomen apart using fine forceps between tergites 4 and 5 while irrigating with water . We retrieved the ovaries , scored them on a scale of 0–4 for signs of ovary activation , and counted the number of ovarioles [24 , 38] .
Because of a significant genotype x replicate interaction ( F1 , 768 = 99 . 3 , p < 0 . 001 , two-factor ANOVA with fixed effects ) caused by a significant effect of replicate ( p < 0 . 001 ) , we were unable to pool data from the two replicates . However in both replicates , AN workers were significantly older than WT workers on their first foraging flight . In the January replicate , the first workers to forage from the two AN colonies were on average 13 . 4 d old , whereas the WT workers commenced foraging significantly earlier ( p < 0 . 001 , two-tailed t-test ) when 9 . 9 d old ( Figure 1 ) . In the November replicate , the first AN workers to forage were 8 . 3 d old . Again , WT workers commenced foraging significantly ( p < 0 . 001 ) earlier when 7 . 3 d old . Furthermore , of the 439 marked workers present in the replicate 1 colony at the end of the experiment , which most likely had never foraged , 313 were AN and 126 were WT . Similarly in the November replicate , there were 417 workers remaining of which 279 were AN and 138 were WT . Because equal numbers ( 2 , 000 each ) of AN and WT workers were introduced to each of the host colonies , this is suggestive that a larger proportion of mature AN workers refrained from foraging than did WT workers . Under a hypothesis that the number of workers of each genotype expected to not engage in foraging should be equal , there were significantly more AN workers present in the nonforaging populations than there were WT workers ( January , = 79 . 66 , p < 0 . 001; November , = 47 . 68 , p < 0 . 001 ) . There were no significant genotype ( AN versus WT ) by replicate ( January versus November ) interactions ( ANOVA using fixed effects models ) for the three variables measured on mature foragers: nectar volume ( F1 , 136 = 0 . 58 , p = 0 . 45 , excluding workers that carried no nectar ) , nectar concentration ( F1 , 121 = 0 . 008 , p = 0 . 93 , excluding workers that carried no nectar ) and weight of pollen carried ( F1 , 83 = 1 . 42 , p = 0 . 24 , excluding workers that carried no pollen ) . We therefore analysed data as two-factor ANOVAs of genotype and replicate , in which we tested whether AN and WT workers differed in the kinds of forage they collected . There was no significant difference in the foraging preferences of AN or WT workers for any parameter measured ( Figure 2 ) . There were significant replicate effects for nectar volume ( F1 , 136 = 9 . 30 , p = 0 . 003 ) , nectar concentration ( F1 , 121 = 6 . 19 , p < 0 . 014 ) , and weight of pollen carried ( F1 , 83 = 125 . 28 , p < 0 . 001 ) . In both replicates , mature AN workers that had probably never foraged had a significantly higher number of ovarioles than AN individuals of similar age that were collected after foraging ( January , Mann-Whitney U = 2 , 236 . 5 , p = 0 . 005 , November , U = 2 , 385 . 5 , p = 0 . 005 ) ( Figure 3 ) . There was no significant difference between the number of ovarioles in foraging and nonforaging WT workers in January ( U = 3 , 002 . 5 , p = 0 . 4 ) but there was in November ( U = 2 , 545 . 0 , p = 0 . 02 ) . WT workers had significantly more ovarioles than did AN workers in both the forager group ( January , U = 1 , 949 . 5 p < 0 . 001; November , U = 2 , 442 . 0 , p = 0 . 009 ) and among nonforagers ( January , U = 2 , 900 . 0 , p = 0 . 07 , November , U = 2533 . 5 , p = 0 . 02 ) ( Figure 3 ) . For AN workers , the nonforaging group had higher ovary activation scores than the foragers ( January , U = 2 , 209 . 0 , p < 0 . 001; November , U = 2 , 769 . 0 , p < 0 . 001 ) ( Figure 3 ) and immature ova ( ovary activation score 3 ) were observed in a single nonforaging AN worker in January . For WT workers , the nonforaging group had significantly higher activation scores than foragers in January ( U = 2 , 602 . 0 , p < 0 . 001 ) but not in November ( U = 3 , 081 . 0 , p = 0 . 84 ) ( Figure 3 ) . There was a positive correlation between ovary activation score and average number of ovaries among all non-foraging workers ( Spearman correlation , ρ = 0 . 135 , n = 326 , p = 0 . 01 ) . There was no significant association between genotype and whether a worker carried pollen ( January = 0 . 03 , p = 0 . 85; November , = 0 . 64 , p = 0 . 42 ) . The number of ovarioles was uncorrelated with the mass of pollen carried ( r = 0 . 13 , n = 127 , p = 0 . 88 ) or the volume ( r = −0 . 094 , n = 311 , p = 0 . 10 ) or concentration ( r = 0 . 07 , n = 158 , p = 0 . 35 ) of nectar carried . ( These correlations were also not significant on a per replicate basis . ) Ovary activation scores were significantly correlated with nectar concentration ( Spearman's ρ = 0 . 25 , n = 159 , p = 0 . 002 ) , but not with nectar volume ( ρ = 0 . 08 , n = 313 , p = 0 . 14 ) or pollen mass ( ρ = −0 . 051 , n = 127 , p = 0 . 60 ) . There was a significant correlation between the number of ovarioles and ovary activation scores ( ρ = 0 . 29 , n = 311 , p < 0 . 001 ) . These patterns of significance were identical on a per replicate basis .
Our data show that AN workers forage significantly later in life than WT workers , and that AN and WT workers were equally likely to forage for nectar or pollen , foraged for nectar of similar quality , and carried similar-sized pollen loads . Thus , our data support the modified RGPH , which suggests that workers selected for high reproductive rate should have late onset of foraging , but otherwise should not differ in foraging behavior . Our data do not support the forager RGPH , which predicts that AN workers should commence foraging early in life and focus on foraging for proteinaceous pollen ( Table 1 ) [8 , 25 , 36] . We found a positive association between ovary activation and ovariole number . This suggests that a larger number of ovarioles , laid down in the larval stage , increases the probability that the individual will become reproductively active . These findings accord with findings in other honey bee populations [24 , 46 , 47] . There was a positive correlation between ovary activation scores and nectar concentration . We have no explanation for this , but we note that it is contrary to a prediction of the forager RGPH hypothesis , which posits that more reproductive workers should forage for pollen and nectar of low sugar concentration . Independent support for a link between the tendency of honey bee workers to delay or refrain from foraging and their reproductive potential comes from the Cape honey bee ( A . m . capensis ) of South Africa . Uniquely , workers of this subspecies are thelytokous and therefore produce female offspring from their unfertilised eggs [48–50] . Because of this , natural selection strongly favours worker reproduction , because workers have the opportunity to contribute directly to the pool of eggs that are raised as queens [51–54] . Consistent with the modified RGPH ( Table 1 ) [19] , it is easy to identify two distinct kinds of workers in A . m . capensis . Dominant workers do little work but express traits that are indicative of high reproductive potential . Subordinate workers , by contrast , do the majority of the work , but are reproductively inactive [55 , 56] . The observation that workers of a strain selected for high pollen hoarding show increased vitellogenin titres relative to a strain selected for low pollen hoarding has provided important support for the forager RGPH hypothesis [23 , 36] . The level of circulating vitellogenin is a good predictor of reproductive potential in many social insects ( e . g . , [57 , 58] ) . It thus seems logical to postulate a role for vitellogenin in the regulation of reproductive potential in honey bee workers , and that workers with high vitellogenin titres should have higher reproductive potential than those that have low titres [36] . However , nurse workers need to produce large amounts of vitellogenin in order to produce brood food that is fed to larvae [23] . Once workers have commenced foraging , they no longer need to produce brood food , and vitellogenin titres are reduced while juvenile hormone titres increase [59] . We therefore think that vitellogenin titres of honey bee workers , contrary to many other social insects , may not be a reliable predictor of an individual's direct reproductive potential . We also doubt the validity of a general association between reproductive potential and division of labour when foraging , modulated by the production of vitellogenin . Solitary bees like Megachilidae actively forage for pollen and nectar , building and provisioning brood cells , one at the time . Once a cell has been provisioned , the female oviposits on the pollen mass and commences foraging to provision another cell . In these species , there is no nest-bound phase for adults , and foraging and reproductive behaviours are contemporaneous . If honey bees evolved from an ancestor similar to the Megachilidae , then it seems unlikely that the gene networks that regulate alternate life history phases would also regulate foraging . Alternatively , the honey bee's ancestor could have had a life cycle similar to Xylocopini and Ceratinini bees , where a nonreproductive female ( daughter or unrelated female ) remains inside the nest and guards it [60] . In these species , it is the reproductive female who does the foraging and egg laying , the nonreproductive , nest-bound female merely waits to inherit the nest . If honey bees evolved from this kind of bee , then one would predict that nest-bound workers would have the lowest reproductive potential in accordance with our results as well as the modified RGPH . Our study highlights the pitfalls of making general conclusions about the evolution of behaviour from particular selected lines when the underlying genetic mechanisms behind behavior are poorly understood . Page et al . 's line was selected for pollen hoarding and this has affected some reproductive traits and age at first foraging . Our line was selected for reproduction , and this had no effect on foraging preferences and has increased age at first foraging . The divergent results when selecting on different phenotypes is explainable by weak genetic correlations between the traits in question . Thus selection on one trait ( pollen hoarding ) selects on a different component of variation related to onset of foraging and reproductive potential than direct selection for worker reproduction . Hence , the observed correlation between the tendency to forage for pollen and early onset of foraging can simply be an artefact of selection on gene networks unrelated to reproductive potential .
|
In social insects , the evolution of the worker caste and the regulation of reproductive behaviour by workers are poorly understood . Evolution is conservative and often proceeds by adapting an existing gene network to a new function . The “reproductive ground plan” hypothesis ( RGPH ) suggests that social insects evolved their queen and worker castes by modifying a gene network that once regulated the foraging and reproductive phases of solitary ancestors . In this model , queens retain characteristics of insects in their reproductive phase , whereas workers retain characteristics of the foraging phase . Moreover , the foraging behaviour of workers may also be regulated by the same genes that once controlled the switch between foraging and feeding young in the nest . We evaluated the RGPH by studying a line of honey bees selected for high rates of worker reproduction . We show that in this line workers forage late in life and some may never forage , supporting the idea that genes related to reproduction are also related to foraging . However , we found no support for recent suggestions that genes related to reproduction also regulate the foraging behaviour of individual workers: once they start foraging , our highly reproductive workers forage in the same way that unselected workers do .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"evolutionary",
"biology",
"ecology"
] |
2008
|
Effects of Selection for Honey Bee Worker Reproduction on Foraging Traits
|
Rabies continues to pose a significant threat to human and animal health in regions of Indonesia . Indonesia has an extensive network of veterinary diagnostic laboratories and the 8 National laboratories are equipped to undertake diagnostic testing for rabies using the commercially-procured direct fluorescent antibody test ( FAT ) , which is considered the reference ( gold standard ) test . However , many of the Indonesian Provincial diagnostic laboratories do not have a fluorescence microscope required to undertake the FAT . Instead , certain Provincial laboratories continue to screen samples using a chemical stain-based test ( Seller’s stain test , SST ) . This test has low diagnostic sensitivity , with negative SST-tested samples being forwarded to the nearest National laboratory resulting in significant delays for completion of testing and considerable additional costs . This study sought to develop a cost-effective and diagnostically-accurate immunoperoxidase antigen detection ( RIAD ) test for rabies that can be readily and quickly performed by the resource-constrained Provincial laboratories . This would reduce the burden on the National laboratories and allow more rapid diagnoses and implementation of post-exposure prophylaxis . The RIAD test was evaluated using brain smears fixed with acetone or formalin and its performance was validated by comparison with established rabies diagnostic tests used in Indonesia , including the SST and FAT . A proficiency testing panel was distributed between Provincial laboratories to assess the reproducibility of the test . The performance of the RIAD test was improved by using acetone fixation of brain smears rather than formalin fixation such that it was of equivalent accuracy to that of the World Organisation for Animal Health ( OIE ) -recommended FAT , with both tests returning median diagnostic sensitivity and specificity values of 0 . 989 and 0 . 993 , respectively . The RIAD test and FAT had higher diagnostic sensitivity than the SST ( median = 0 . 562 ) . Proficiency testing using a panel of 6 coded samples distributed to 16 laboratories showed that the RIAD test had good reproducibility with an overall agreement of 97% . This study describes the successful development , characterisation and use of a novel RIAD test and its fitness for purpose as a screening test for use in provincial Indonesian veterinary laboratories .
Rabies is a lethal zoonotic viral disease caused by a member of the Lyssavirus genus within the Rhabdoviridae family . Dog bites are responsible for transmission of rabies to humans in 99% of all mortalities and for 90% of post-exposure prophylaxis ( PEP ) globally . Rabies was first reported in Indonesia in 1884 and is thought to be endemic in 24 of the country’s 34 provinces [1] where it causes 150 to 300 human fatalities annually [2] . Indonesian rabies isolates belong to Asian lineage within the classical rabies virus , lyssavirus genotype 1 [3] . Control programs at provincial and district levels are regularly implemented in Indonesia but adequate vaccination coverage has been difficult to achieve . Testing of suspected animal rabies cases is conducted at Indonesian veterinary service laboratories; 8 National and 23 Provincial . The direct fluorescent antibody test ( FAT ) is widely used as the reference test for rabies diagnosis due to its diagnostic performance [4] . The FAT requires expensive fluorescein isothiocyanate ( FITC ) -labeled antibodies and a well-maintained fluorescence microscope . As a result , Indonesia’s well-resourced National laboratories have the capacity to perform the FAT whilst the majority of Provincial laboratories must instead use the less accurate Seller’s stain test ( SST ) for preliminary diagnosis . SST involves chemical staining and microscopic observation for the presence of intra-cytoplasmic intra-neuronal inclusion bodies ( Negri bodies ) to indicate rabies infection [5]; however , SST has low diagnostic sensitivity [6 , 7] . Brain samples that return a negative result using SST are sent to the nearest National laboratory for confirmatory follow-up testing by FAT and by mouse inoculation test ( MIT ) if the FAT returns a negative result ( Fig 1 ) . This results in considerable delays in reporting results and the additional testing of SST-negative samples ( many of which are false negatives ) places an unnecessary burden on the receiving National laboratory . Delays in confirming and reporting on rabies cases can lead to suboptimal bite case management . To improve access to reliable rabies diagnostics , other rabies virus detection tests have been developed for use in settings where laboratory resources are limited . For example , a direct rapid immunohistochemical test ( dRIT ) that uses a cocktail of two biotin-labelled monoclonal antibodies ( mAbs ) to detect rabies antigen in brain has been described [8] . A modification of this dRIT that replaced the biotin-labelled mAbs with a biotin-labelled polyclonal antibody ( pAb ) demonstrated improved performance [8 , 9] . Both tests eliminated the need for expensive fluorescence microscopes; however , the cost of producing primary antibody conjugates and the absence of their commercial supply could prevent widespread use of these tests . In some countries , commercially-produced rapid ( lateral flow ) tests are used in remote veterinary clinics for testing of brain samples from suspect rabid dogs . Test performance can vary depending on the specific commercial supplier [10] and , for countries such as Indonesia where provincial authorities have established veterinary laboratory networks , the use of such field-based tests is contrary to their desired objectives to build sustainable and cost effective capacity in laboratories . This paper describes the development and validation of an alternative pAb-based , indirect , non-fluorescent test that overcomes these supply constraints by replacing the use of biotinylated primary antibodies with an anti-rabies pAb used in combination with a commercially available horseradish peroxidase ( HRP ) -conjugated secondary antibody . In addition , the RIAD test uses light as opposed to fluorescence microscopy and has been produced and tested as a kit making it suitable for use in resource-limited laboratories .
Rabies challenge virus standard ( CVS ) strain nucleoprotein gene was subcloned into pETHb , a derivative of pET50b ( Novagen ) [11] . In brief , rabies virus nucleoprotein ( RABV NP ) gene was amplified by PCR using primers 5’-gaatggatcctacaatggatgccgacaaga and 5’- attcaagcttatgagtcactcgaatatgt and purified by elution from 0 . 8% ( w/v ) agarose gel . The PCR fragment and pETHb were digested with BamHI and HindIII ( Promega ) and pETHb was dephosphorylated with TSAP ( Promega ) . Both insert and vector were gel-purified and ligated with T4 DNA ligase ( Promega ) . Resulting pET-RABV NP was used to transform DH5 alpha Escherichia coli ( Invitrogen ) . The pET-RABV NP construct was purified from positive clones and sequence fidelity was confirmed by Sanger sequencing . The Australian Animal Health Laboratory ( AAHL ) Animal Ethics Committee ( AEC ) is licensed with Agriculture Victoria , Australia , and complies with all the relevant requirements of the Prevention of Cruelty to Animals Act ( 186 ) and the Regulations; and complies with the Australian Code for the care and use of animals for scientific purposes ( 8th Edition 2013 ) . The AAHL AEC approved the use of animals for production of antisera within the Small Animal Facility ( SAF ) at AAHL under AEC protocol number 1401 . Protein samples were solubilized in NuPAGE 1× LDS sample buffer ( Invitrogen ) containing 50 mM dithiothreitol ( DTT , Promega ) ( reducing sample buffer ) by heating at 100°C for 10 min . Samples were loaded onto NuPAGE 4–12% Bis-Tris polyacrylamide gels and run in MOPs buffer ( Invitrogen ) at 200 V for 50 min . Gels were stained for 10 min in 0 . 25% ( w/v ) Coomassie Brilliant blue R250 then destained by washing for 20 min several times in a solution of 5% ( v/v ) acetic acid and 15% ( v/v ) methanol . For immunoblotting , proteins from unstained gels were Western transferred to PVDF membrane ( Pall ) at 200 mA for 1 h in 20 mM N-cyclohexyl-3-aminopropanesulfonic acid ( CAPS ) buffer pH 11 with 10% ( v/v ) methanol then blocked for 1 h in 30 mL 5% ( w/v ) skim milk in TBST ( 20 mM Tris pH 7 . 4 , 150 mM NaCl , 0 . 05% ( v/v ) Tween 20 ) . Primary rabbit antisera ( produced in the SAF at AAHL as described below ) and secondary HRP-conjugated goat anti-rabbit antibody ( Bio-Rad ) were diluted 1:10 , 000 and 1:20 , 000 in TBST , respectively , and applied separately to transfer membranes for 1 h each at room temperature ( RT ) on a rocking platform . Membranes were washed three times for 10 min per wash with 50 mL TBST after incubation with primary and secondary antibodies . Enhanced chemiluminescence using ECL Plus ( Pierce ) was used to detect immunoreactive bands as read using x-ray film or a Typhoon FLA9000 fluorescence scanner ( GE ) . Chemically competent E . coli ( Shuffle , NEB ) were transformed with 10 ng purified pET-RABV NP as per manufacturer’s instructions . An individual colony from a RABV NP-expressing clone grown on selection media was used to inoculate a 10 mL LB broth starter culture containing 100 μg/mL ampicillin ( Sigma ) . The culture was grown for 18 h at 30°C with agitation at 250 rpm and used to inoculate 1 L of LB broth containing 100 μg/mL ampicillin and grown under these conditions for approximately 3 h . When the culture reached an optical density of 0 . 6 at 600 nm , isopropyl-beta-D-thiogalactopyranoside was added to a final concentration of 0 . 4 mM to induce expression of the nucleoprotein and the culture was grown for a further 3 h or overnight at 16°C . Cells were separated from medium by centrifugation at 4 , 000 x g for 10 min at 4°C and cell pellets were stored at -80°C until required for further processing . Cell pellets were thawed on ice then lysed on a rocking platform mixer for 20 min at RT in 50 mL BugBuster Master Mix ( Novagen ) containing protease inhibitors ( P8849 , Sigma ) . RABV NP inclusion bodies were purified from the lysate in diluted BugBuster Master Mix according to the manufacturer’s instructions . Inclusion bodies were resuspended and solubilized by heating to 100°C for 10 min in 750 μL NuPAGE 1× LDS sample reducing buffer . RABV NP concentration was approximated by comparison with bovine serum albumin ( Sigma ) standards resolved by SDS PAGE and stained with Coomassie Brilliant blue R250 . Approximately 100 μg of solubilized RABV NP inclusion bodies was resolved by SDS PAGE on NuPAGE 4–12% Bis-Tris 1 mm x 2D preparative well gels ( Invitrogen ) at 200 V for 50 min . Gels were washed 3 times for 1 min with deionized water then overlaid with ice-cold 0 . 3 M KCl to visualize RABV NP protein bands . Bands were excised using a flexible skin graft knife blade then macerated by extrusion through a Luer lock syringe . RABV NP was isolated by adding one gel volume of phosphate buffered saline ( PBS ) , 0 . 1% ( w/v ) SDS to the gel fragments and passively eluting overnight at RT on a rotating wheel mixer . The gel-buffer slurry was transferred to Microsep 0 . 2 μm Supor membrane centrifugal devices ( Pall ) and centrifuged at 4 , 000 x g for 5 min at RT . Flow-through containing eluted RABV NP was collected and the passive elution process was repeated once for 1 h to isolate any residual RABV NP . Isolated RABV NP fractions were pooled and protein concentration was determined using a BCA protein assay kit ( Pierce ) . RABV NP was concentrated by centrifugation in a 3 kDa MWCO Centrifugal Filter Unit ( Millipore ) . Purification of RABV NP was confirmed by SDS PAGE followed by staining with Coomassie blue and by immunodetection with penta-His monoclonal antibody ( Qiagen ) as per manufacturer’s instructions . Immunogen was prepared as a water-in-oil emulsion of purified RABV NP and CSIRO Triple Adjuvant prepared as described previously [12] . For each rabbit immunized , 600 μL of immunogen was prepared by mixing 90 μL of RABV NP at 1 mg/mL with 54 μL PBS and 96 μl of 3 mg/mL QuilA ( Superfos Biosector ) , 30 mg/mL DEAE-Dextran ( Pharmacia ) . This aqueous phase was added to 360 μL of Montanide ISA 50 V2 ( Seppic ) and emulsified by repeated extrusion through an 18 gauge blunt needle . Two New Zealand White rabbits were immunized by intramuscular injection on three occasions approximately three weeks apart . Each immunization used a total of 75 μg RABV NP in two 0 . 25 mL doses , one dose in each hind leg . Serum samples ( ~1 mL ) were taken prior to immunization to test for background staining . Sera were taken after each immunization and assessed for the presence of RABV NP antibodies by immunoblotting . All brain samples used in this study were existing samples submitted to a National laboratory for testing as part of the Indonesian Government’s control measures for rabies initiative . Brain samples transported in 50% glycerine-saline solution were washed in PBS pH 7 . 4 for 30 min at RT , and tested using SST or the FAT . SST is a rapid method of staining brain tissue smears or sections that incorporates methylene blue and basic fuchsin dyes [13] . The FAT is a direct immunostaining method of acetone-fixed brain smears that utilizes FITC-conjugated anti-rabies antibody to detect viral antigen [14] . Commercially available anti-rabies FITC-conjugated antibody ( Bio-Rad ) was used for the FAT as per manufacturer’s instructions . Smears of brain material were prepared on positively charged ( DAKO ) or ( 3-aminopropyl ) triethoxysilane ( AAS; Sigma ) -coated glass microscope slides . Smears were air dried for 5 min and fixed in 100% acetone ( RIADacetone ) at -20°C for 30 min or neutral buffered formalin ( RIADformalin ) for 30 min at RT . Smears were air dried again then treated with 200 μL of 3% ( v/v ) hydrogen peroxide for 10 min at RT in a humidified chamber . The rabbit primary anti-RABV NP antibody is described above . The secondary antibody was a commercial HRP-labelled anti-rabbit antibody ( Envision Dako ) . All antibody incubations were performed at RT for 45 min in a humidified chamber . Brain smears were incubated sequentially with 200 μL of primary anti-RABV NP antiserum diluted 1:1000 and secondary HRP-labelled anti-rabbit antibody diluted 1:500 . Dilutions were prepared using TBST containing 1% ( w/v ) skim milk ( Australian origin ) . Smears were washed three times for two min per wash with TBST after each of the fixing , blocking and incubation steps . Chromogen was prepared immediately prior to use by adding 5 μL 30% ( v/v ) hydrogen peroxide to 500 μL of 4 mg/mL 3-amino-9-ethylcarbazole ( Sigma ) in N , N , dimethylformamide ( Sigma ) and diluting to 10 mL with 50 mM sodium acetate , pH 5 . Brain smears were incubated in 200 μL of chromogen substrate for 10 min at RT and the reaction was stopped by rinsing once with distilled water . Smears were counterstained with Mayer’s haematoxylin ( Lillie’s modification; Australian Bio Stains ) for 20 s , rinsed once with distilled water followed by TBST then mounted with aqueous mounting medium ( DAKO ) . Smears were viewed using transmitted white light microscopy with x20 or x40 objectives . Brain samples were deemed positive for rabies virus antigen if neuronal cytoplasmic green fluorescence was present in the FAT; neuronal cytoplasmic brick red deposits were seen for RIAD stained samples; or intracytoplasmic inclusion ( Negri ) bodies were detected using SST . In the absence of these indicators the sample was classified as negative . Dog brain samples ( n = 116 ) were tested for exclusion of rabies virus at DIC Bukittinggi . The RIAD tests were assessed using this panel of samples derived from animals involved in human dog bites cases from within Sumatra and adjacent smaller islands that were submitted for diagnosis ( “diagnostic group” ) and another panel of 110 canine brain samples obtained from diagnostic samples submitted to the laboratory from Indonesian Government-administered dog population control activities in areas including Riau Island and districts within Bali that were thought to be free of rabies but which were close to endemic rabies areas ( “survey group” ) . Both the diagnostic group and the survey group of samples were collected over a 2 year period up to 2015 . Each sample was determined to be positive or negative using the SST , FAT or RIAD when assessed blind and in parallel by independent laboratory staff within the Disease Investigation Center ( DIC ) Bukittinggi . Diagnostic sensitivity ( DSe ) and specificity ( DSp ) were estimated for RIAD , FAT and SST using a 3 tests-in-2 population Bayesian latent class model ( LCM ) which allowed for conditional dependence in the sensitivities of RIAD and FAT [15] . Existing Bayesian code ( http://cadms . ucdavis . edu/diagnostictests/2dep1ind3t2p . html ) was modified for the analysis . The populations were the diagnostic and survey groups , where the latter was confined to dogs which had complete results on all 3 assays ( n = 80 ) . A sensitivity dependence ( covariance ) term was incorporated into the model to account for the fact that the RIAD and FAT assays target the same conserved protein of the rabies virus but probably different epitopes . The SST results were assumed to be conditionally independent of RIAD and FAT results because the SST uses a chemical stain to identify the presence of Negri bodies whereas the RIAD and FAT use rabies virus-specific antibodies to identify viral protein; hence , additional covariance terms were considered unnecessary . Two separate models were created for the RIAD test when used on slides fixed with acetone or formalin . A specificity covariance was not considered since all 3 tests gave zero positive results in the survey ( presumed non-infected ) population . Flat priors ( beta 1 , 1 ) were used for DSe and DSp of all 3 tests and prevalence in the diagnostic group . For the survey population , one of the authors ( JA ) believed that there was a 10% chance that the population might be infected but he was 90% sure that if dogs were infected , prevalence would be <1% with a most likely value ( mode ) of 0 . 1% . The latter information equated to a beta ( 1 . 27 , 274 . 82 ) prior . The opinion of JA was supported by two coauthors ( IR , YF ) who have extensive experience with rabies in Indonesia . Hence , prevalence in the survey group was modelled as a mixture distribution with a prevalence = 0% ( point mass of 0 ) with 90% probability and a beta ( 1 . 27 , 274 . 82 ) distribution with 10% probability . Models were run in OpenBUGS 3 . 2 . 3 rev . 1012 [16 , 17] , and results were reported as medians and 95% probability intervals ( PI ) . The difference in DSe of the 3 combinations of test pairs was calculated at each iteration of the model and the step function in OpenBUGS was used to estimate whether the difference in DSe ( e . g . DSe RIAD minus DSe Sellers ) was positive . Briefly , the step function creates a Boolean variable ( 1 if positive , 0 if negative or zero ) for any node ( e . g . DSe ) and the proportion of ones across all iterations can be interpreted as the probability ( P ) that a test has a higher DSe than a comparator test where P = 1 indicates certainty and P = 0 . 5 indicates no difference . The models were initially run for 50 , 000 iterations after the initial 5 , 000 iterates were discarded as burn-in . Model convergence was assessed by examination of history plots and running 2 separate chains from different initial values and plots of model parameters were checked for autocorrelation and thinning was done , if necessary . A sensitivity analysis was done to assess effects of a flat prior ( beta 1 , 1 ) for prevalence in the survey group rather than use of the mixture distribution described in the previous paragraph . To determine the reproducibility of the RIAD test , RIAD test kits containing materials sufficient for 50 tests were delivered to 16 Provincial laboratories for their use against a proficiency test panel of dog brain samples . Kits included rabbit anti-RABV NP polyclonal antiserum pre-diluted in Envision FLEX diluent ( DAKO ) , wash buffer ( TBST ) , antibody diluent , anti-rabbit HRPO-conjugated secondary antibody ( Jackson ) , AAS-coated slides , plastic ware , and positive and negative acetone-fixed dog brain control smears . Tests were performed according to protocols described above . Sixteen Indonesian Provincial veterinary laboratories used the RIAD test kit to assess a panel of unknown positive and negative samples ( samples 1–6 ) derived from the hippocampus of 6 individual dogs . Four canine brain samples were rabies virus-positive ( samples 1 , 2 , 4 and 5 ) and 2 were rabies virus-negative ( samples 3 and 6 ) . The proficiency testing ( PT ) organizing laboratory at DIC Bukittinggi provided all brain samples and determined their disease status using the FAT and RIAD test . Brain smears were prepared from frozen tissue that was thawed and passed 5 to 10 times through an 18 gauge needle . Homogenized brain was fixed in acetone , air-dried and treated with hydrogen peroxide , as described above , then stored at -20°C . Homogeneity of smears was tested at DIC Bukittinggi using the RIAD test prior and subsequent to the PT round . Stability of dog brain smears fixed in acetone had been previously ascertained during the development of the RIAD test and determined to be at least 6 months duration when stored at -20°C . The panel was transported chilled with the above mentioned kit .
RABV NP was abundantly expressed in Shuffle E . coli and was partially purified from bacterial lysates as insoluble inclusion bodies ( Fig 2A ) . Passive gel elution of the ~50 kDa RABV NP protein in the inclusion body preparation from polyacrylamide gels yielded an enriched RABV NP fraction as demonstrated by staining with Coomassie blue ( Fig 2B ) and by immunoblotting with penta-His antibody ( Fig 2C ) . Pre-immune rabbit sera showed no background reactivity to purified RABV NP by immunoblotting ( Fig 2D ) . Serum samples taken approximately 14 days after each immunization showed successful anti-RABV NP seroconversion by immunoblotting against purified RABV NP and successive immunization led to greater anti-RABV NP polyclonal antibody titer ( Fig 2D ) . Rabies virus antigen detected with the RIAD in brain smears appeared as fine to globular red-brown particles within brain smear material ( Fig 3 ) . While much of the antigen was found free within smear material , it was also found within the cytoplasm of neuron cell bodies . The staining pattern and distribution of antigen in RIAD was similar to that found in the FAT . No red-brown particulate staining was seen in negative smears . History and trace plots and based on use of 2 Markov chains indicated that all models converged . However , autocorrelation was evident in 3 plot: DSe for RIADformalin and FAT , and the sensitivity covariance between RIADformalin and FAT . For inferences about parameters , we ran 500 , 000 iterations thinned by 10 to minimize any effects of autocorrelation on estimates of these 3 parameters . The RIADacetone test and FAT results were highly accurate , with both producing identical median DSe and DSp values of 0 . 989 and 0 . 993 , respectively ( Table 1 ) . This was consistent with the empirical finding that there were no discordant test results . The resultant step function also indicated no difference in DSe ( RIADacetone > FAT: P = 0 . 500 ) . In the diagnostic group , the SST returned 44 negative results when both RIADacetone and FAT were positive indicating lack of sensitivity ( median = 0 . 562 ) . The step function showed that the RIADacetone and the FAT had greater DSe than the SST with probability of 100% . Comparisons of RIADformalin , FAT and SST yielded similar findings ( Table 1 ) except that RIADformalin was more sensitive than FAT in the step function analysis ( P = 0 . 867 ) and reflected in a median sensitivity that was 3% higher . As expected , the estimated median prevalence of rabies in the human bite case ( diagnostic ) group in the 2 analyses was similar ( 0 . 869 and 0 . 908 for RIADacetone and RIADformalin , respectively ) . Median estimates of the sensitivity covariances were close to zero ( Table 1 ) indicating that they were of little practical importance . However , because the 95% probability interval for the RIADacetone model excluded zero and the 2 . 5 percentile for RIADformalin model was close to zero , the sensitivity covariance term was left in both models . All tests correctly identified each of 80 canine brain smears obtained from the presumed rabies-free regions as negative for rabies infection; hence step function results for differences in specificities between tests were not evaluated . Median DSp estimates for all tests were between 0 . 990 and 0 . 993 . Use of a flat prior on prevalence ( beta 1 , 1 ) in the survey group rather than an informative mixture prior had minimal effects on test performance characteristics but the median prevalence in the survey group was 4 . 5 times higher ( median = 0 . 009 , 95% PI = 0 . 0003–0 . 044 ) with use of the flat prior , and probability intervals for DSe and DSp of the 3 tests were slighter wider . The PT round showed that most laboratories were proficient in using the new RIADacetone test demonstrating its high reproducibility between participating laboratories ( Table 2 ) .
Rabies is a lethal zoonotic disease that is endemic in 24 provinces situated across numerous islands within the Indonesian archipelago [1] . The disease presents a significant economic burden to the region due to costs associated with diagnosis , treatment and control programs . Rabies infections in humans are often fatal and thus accurate and timely diagnosis is critical when human exposure is suspected , although the risk to humans is mitigated in Indonesia where the policy is to treat all potentially exposed humans . Nevertheless , the seriousness of the disease demands a thorough diagnostic testing regimen that ensures a low probability of obtaining false-negative results . To achieve this the current diagnostic strategy within Indonesia is to confirm negative results obtained using the SST with the FAT and any negative FAT results with the mouse inoculation test ( MIT ) such that a true negative is only confirmed after three tests are performed . The SST was demonstrated to have significantly lower diagnostic sensitivity leading to a high number of false-negative results ( Table 1 ) , but remains a frontline diagnostic test in resource-constrained laboratories due to its simplicity and affordability . This result was expected as the SST stains brain smears to detect the presence of Negri bodies in the cytoplasm of virus infected neurons [4] . This requires skill and patience in searching for the possible presence of Negri bodies compared with immunological detection of widely-dispersed viral antigen in brain smears [6] . For this reason the SST is not officially accepted by the OIE or World Health Organization as a diagnostic test for rabies providing further impetus for its replacement as the routine first line of testing in Provincial laboratories . The SST is inexpensive to perform , however , this possible economic benefit dissipates when a negative result is produced , be it true or false , since all negative samples must then be transported to National laboratories where they are retested using the more expensive FAT and possibly MIT . Therefore a low cost , simple to use , frontline test that has similar diagnostic sensitivity to the FAT would produce less false negative results and remove the need for further testing and the associated cost . The RIADacetone test developed and validated as described herein provides one such test . An alternative approach for resource-limited situations would be to use one of the commercial rapid immunochromatic rabies tests , commonly known as lateral flow devices , that have been developed and marketed in more recent years [10] . However , the diagnostic and analytical sensitivity of a number of rapid tests were reported as ranging from 0% to 100% for field derived samples and 32% for experimentally infected animals [10] . This high variability and the number of reported false negative results for these rapid tests , in association with the cost for countries to procure and arrange for import clearances , means that many countries , including Indonesia , have a preference to build diagnostic capacity within their laboratory network . Within Indonesian animal health laboratories the vast majority of samples submitted for rabies diagnosis are canine brains and in DIC Bukittinggi , which is the designated National rabies reference laboratory , 87% of submitted canine brain specimens from dog bite cases submitted between 2013 and 2015 were positive . Of the 110 canine brain samples collected from apparently rabies-free areas , all samples tested negative by all tests used in our comparison . Specific immunostaining of brain smears with anti-rabies virus nucleoprotein polyclonal antisera was observed . Smears fixed with acetone or formalin were both compatible with the RIAD method; however , acetone fixation produced staining of greater clarity and intensity and produced superior DSe results . However , one advantage of formalin over acetone fixation is that it inactivates rabies virus allowing less restrictive transport and handling procedures . The staining patterns observed were identical between the RIADacetone test and FAT with both demonstrating discrete punctate staining localized to the cytoplasm of rabies infected cells . No significant non-specific or background staining was observed when antisera were used at the optimized dilutions , aiding the interpretation of results . One major advantage of the RIADacetone test over the FAT is that it may be performed in any laboratory where the SST is currently used because only light microscopy is required . It also replaces expensive antibody conjugates with readily available HRP-conjugated secondary antibodies . Each test was comparatively evaluated using the canine brain samples from rabies endemic areas and from areas thought to be free of rabies . Whilst none of the tests returned positive results when used to test samples from presumed rabies-free regions , RIAD testing of formalin-fixed samples from endemic regions resulted in a number of false positive results when compared to the FAT and the RIADacetone test ( Table 3 ) . The RIADacetone test when compared to the SST and FAT was found to be very sensitive and of comparable accuracy to the FAT when testing dog brain samples infected with strains of rabies virus endemic to Indonesia . The FAT is designated by the OIE as the reference test for rabies because it provides “reliable results on fresh specimens…in more than 95–99% of cases” [4] . We interpreted this statement as the DSe but other interpretations are possible ( e . g . DSp or predictive values ) . Because of the vagueness of this wording , we decided not to use priors for the DSe and DSp of the FAT to inform the Bayesian LCM . This decision was appropriate as the median posterior values for DSe and DSp for the FAT and RIADacetone were both approximately 99% and hence consistent with the reported range of values in the OIE chapter . The RIADacetone test in kit form was transferred to the smaller and resource-limited Provincial laboratories for a proficiency testing round to assess its ruggedness . The test demonstrated high reproducibility with overall agreement of 97% . Of the discrepant results , one false negative was reported by one laboratory . In contrast , two false positive results that were reported by two of the participating Provincial laboratories were of less concern . Indonesian government protocols currently require samples to be sent to a National laboratory for confirmation of any negative results associated with human dog bite cases using the FAT and possibly the MIT . Adoption of the RIADacetone test instead of the SST would reduce the number of false negative samples requiring confirmatory testing . Although the evaluation of performance of the RIADacetone test demonstrated its equivalence to the FAT , it is still suggested that Provincial laboratories forward RIADacetone test-negative brain samples to the National laboratories for confirmation . This would be ongoing or until deemed unnecessary through a review of long term data generated by individual Provincial laboratories , especially those that regularly receive samples for rabies testing and hence maintain a high degree of test competency . The RIADacetone test demonstrated diagnostic sensitivity and specificity comparable to the FAT which is the current reference ( gold standard ) test . It showed good reproducibility and was suitable for use in Indonesian Provincial laboratories using their existing equipment which included standard light microscopes . Replacement of the SST and/or FAT with the RIADacetone test would significantly reduce the economic burden associated with rabies virus diagnosis under the current testing regimen in Indonesia .
|
In Indonesia , veterinary diagnostic laboratories conduct tests for rabies on brain samples from animals suspected of being infected with rabies virus . National laboratories use internationally recommended tests for rabies virus that require expensive materials and equipment . Remote and smaller Provincial laboratories use a simpler older-generation chemical stain test that is less costly to perform but is also highly inaccurate resulting in many rabies-infected brains returning false negative test results . Brain samples that give negative results at a Provincial laboratory are then transported to a National laboratory for retesting to confirm the diagnosis . This results in additional costs and time delays and creates the need for a more effective , lower cost rabies test that Provincial laboratories can effectively use . This paper describes the development of one such test that is of comparable accuracy to the internationally recommended test for detecting rabies in brain and does not require expensive equipment to perform .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"and",
"health",
"sciences",
"immune",
"physiology",
"pathology",
"and",
"laboratory",
"medicine",
"pathogens",
"immunology",
"tropical",
"diseases",
"microbiology",
"vertebrates",
"geographical",
"locations",
"animals",
"mammals",
"dogs",
"indonesia",
"viruses",
"rabies",
"rna",
"viruses",
"neglected",
"tropical",
"diseases",
"antibodies",
"veterinary",
"science",
"research",
"facilities",
"research",
"and",
"analysis",
"methods",
"rabies",
"virus",
"immune",
"system",
"proteins",
"infectious",
"diseases",
"lipids",
"veterinary",
"diseases",
"zoonoses",
"proteins",
"fats",
"medical",
"microbiology",
"microbial",
"pathogens",
"research",
"laboratories",
"people",
"and",
"places",
"biochemistry",
"lyssavirus",
"eukaryota",
"asia",
"oceania",
"viral",
"pathogens",
"physiology",
"biology",
"and",
"life",
"sciences",
"government",
"laboratories",
"viral",
"diseases",
"amniotes",
"organisms"
] |
2017
|
Development and validation of an immunoperoxidase antigen detection test for improved diagnosis of rabies in Indonesia
|
SIRT1 is a member of the sirtuin family of NAD+-dependent deacetylases , which couple cellular metabolism to systemic physiology . Although studies in mouse models have defined a central role for SIRT1 in maintaining metabolic health , the molecular mechanisms remain unclear . Here we show that loss of the Drosophila SIRT1 homolog sir2 leads to the age-progressive onset of hyperglycemia , obesity , glucose intolerance , and insulin resistance . Tissue-specific functional studies show that Sir2 is both necessary and sufficient in the fat body ( analogous to the mammalian liver ) to maintain glucose homeostasis and peripheral insulin sensitivity . Transcriptional profiling of sir2 mutants by RNA-seq revealed a major overlap with genes regulated by the nuclear receptor Hepatocyte Nuclear Factor 4 ( HNF4 ) . Consistent with this , Drosophila HNF4 mutants display diabetic phenotypes similar to those of sir2 mutants , and protein levels for dHNF4 are reduced in sir2 mutant animals . We show that Sir2 exerts these effects by deacetylating and stabilizing dHNF4 through protein interactions . Increasing dHNF4 expression in sir2 mutants is sufficient to rescue their insulin signaling defects , defining this nuclear receptor as an important downstream effector of Sir2 signaling . This study demonstrates that the key metabolic activities of SIRT1 have been conserved through evolution , provides a genetic model for functional studies of phenotypes related to type 2 diabetes , and establishes HNF4 as a critical downstream target by which Sir2 maintains metabolic health .
The incidence of complex metabolic disorders has been on the rise for the past three decades , comprising an epidemic of ever-increasing severity . Much of this can be attributed to an increase in the prevalence of type 2 diabetes accompanied by insulin resistance , the development of which is complex and poorly understood . These trends have prompted widespread changes in public policy and a shift in biomedical research toward improving our understanding of the genetic and environmental factors that contribute to insulin resistance and its progression to a more severe disease state . One focus for these studies has been the sirtuin family of NAD+-dependent deacetylases , which play a central role in coupling metabolic state to systemic physiology . Sirtuin activity is dependent upon the availability of NAD+ , an important electron carrier that contributes to cellular redox balance , drives mitochondrial oxidative phosphorylation , and acts as an important enzymatic cofactor [1–4] . The founding member of the sirtuin family , yeast Sir2 , was discovered based on its role in heterochromatin formation [5–7] . Subsequent studies of the mammalian Sir2 homolog , SIRT1 , have defined it as a critical regulator of metabolic homeostasis , acting through multiple protein targets [1–4] . These include Foxo , the nuclear receptors PPARα and LXR , and co-activators such as PGC-1α [8–11] . The multiple downstream targets of SIRT1 , combined with its dependence on NAD+ , establish it as a pivotal energy sensor that couples cellular redox state to metabolic control . Given the complexity of SIRT1 regulation and function , it is not surprising that genetic studies of SIRT1 in mice have been complicated by environmental factors and genetic background effects , occasionally leading to contradictory results [1 , 3] . This has been most evident in the context of aging , where the beneficial effects of SIRT1 remain controversial . In contrast , tissue-specific functional studies of SIRT1 , combined with overexpression experiments and pharmacological activation , have established important roles for this factor in maintaining metabolic homeostasis [1 , 3 , 4] . These include a role in pancreatic beta-cells to promote glucose-stimulated insulin secretion ( GSIS ) and improve glucose tolerance , as well as activities in peripheral tissues that promote insulin sensitivity [12 , 13] . SIRT1 also supports fatty acid oxidation and oxidative phosphorylation in the liver , suppresses hepatic steatosis , and acts in white adipose to suppress lipid accumulation [9] . Conversely , low-level SIRT1 overexpression promotes glucose tolerance , insulin sensitivity , and prevents fatty liver disease , highlighting the beneficial effects of SIRT1 action and supporting the proposal that SIRT1 activation could be of therapeutic value [14–16] . In spite of these advances , however , the molecular mechanisms by which SIRT1 maintains metabolic homeostasis remain unclear [2] . Studies of the Drosophila SIRT1 homolog , Sir2 , have recently begun to provide a better understanding of its roles in systemic physiology . Null mutants for sir2 display increased levels of stored lipid , analogous to the role of SIRT1 in suppressing obesity [17 , 18] . Elevated free glucose levels were also observed in mutant adults , accompanied by starvation sensitivity [18] . Other metabolic functions for sir2 , however , have been based on overexpression experiments and RNAi [18 , 19] . One of these studies reported that glucose levels are reduced in animals with ubiquitous RNAi against sir2 , contradicting their data from mutants [18] . The RNAi studies also resulted in only a two-fold reduction in sir2 expression , leaving it unclear how these results relate to gene function [18 , 19] . These concerns , combined with the importance of genetic background on SIRT1 activities , led us to undertake a detailed metabolic analysis of a transheterozygous combination of sir2 null alleles compared to genetically-matched controls . We show here that loss of sir2 leads to the age-progressive development of obesity , hyperglycemia , glucose intolerance , and insulin resistance . Tissue-specific RNAi and genetic rescue experiments show that Sir2 function is both necessary and sufficient in the fat body to maintain insulin sensitivity . In addition , our studies show that Sir2 maintains insulin signaling through deacetylation and stabilization of the Drosophila ortholog of HNF4A , dHNF4 . Sir2 interacts with dHNF4 , dHNF4 levels are reduced in sir2 mutants , and expressing wild-type dHNF4 restores insulin signaling in a sir2 mutant background . Taken together , our results define dHNF4 as a key downstream target of Sir2 and provide insights into the molecular mechanisms by which Sir2 promotes insulin sensitivity and metabolic health .
Two previously described deletion alleles of sir2 , sir22A-7-11 and sir24 . 5 , were used in transheteroallelic combination and compared to genetically-matched controls for all studies [20 , 21] . As expected , sir2 is not expressed in these mutants as assayed by RNA-seq or northern blot hybridization , consistent with their characterization as null alleles ( S1A and S1B Fig ) . Unless otherwise indicated , adult male flies were used in all experiments , where the age indicated in the figures refers to the number of weeks after eclosion from the pupal case . The sir2 mutants survive to adulthood and develop starvation sensitivity as previously reported ( S1C and S1D Fig ) [18 , 21] . Basic metabolite measurements , however , reveal that they also display increasing metabolic dysfunction with age in the absence of significant effects on feeding rate ( Figs 1 and S1E ) . At one week of age , sir2 mutants have elevated levels of both free and circulating glucose as well as glycogen but no significant change in triglycerides ( TAG ) ( Fig 1A–1C; S1F Fig ) . Elevated glucose and glycogen levels are still present at two weeks of age , but are also accompanied by elevated TAG , which is consistent with the increased lipid levels reported for sir2 mutants ( Fig 1D–1F ) [17 , 18] . In addition to this obesity , mutants at two weeks of age , but not one week , display fasting hyperglycemia , a hallmark of diabetes ( Fig 1G and 1H ) . This is consistent with the results of metabolomic analysis of sir2 mutants at two weeks of age , which revealed increased levels of glycolytic intermediates , including glucose-6-phosphate , dihydroxyacetone phosphate , and lactate ( S2 Fig ) . Alternative glucose metabolites also increase significantly , such as the glucose alcohol sorbitol , which can accumulate to high levels in diabetics and may contribute to neuropathy and nephropathy [22] . An oral glucose tolerance test was used to determine if these age progressive defects in carbohydrate homeostasis can be accounted for by reduced peripheral glucose uptake . In this assay , male flies are fasted overnight and then allowed to consume 10% glucose for approximately one hour , after which they are transferred back to starvation media for either two or four hours . The kinetics with which they clear glucose from their systems is then monitored by performing glucose assays at each time point . The glucose levels in wild-type animals at both one and two weeks of age return to near fasting levels within two hours after glucose feeding ( Fig 1I and 1J ) . Similarly , although sir2 mutants at both two and three weeks of age are hyperglycemic after consuming glucose , they display relatively normal kinetics of subsequent glucose clearance at two weeks of age ( Fig 1I ) . They are , however , clearly glucose intolerant by three weeks of age , as demonstrated by the continued high levels of glucose present after two hours of clearance on starvation media ( Fig 1J ) . Taken together with our previous results , this indicates that sir2 mutants display a progression of symptoms associated with a loss of glycemic control during early adulthood , from elevated levels of free glucose , to fasting hyperglycemia , to glucose intolerance . The development of diabetic phenotypes in sir2 mutants with age could arise from a defect in peripheral insulin signaling . To determine if this is the case , we measured the levels of phosphorylated AKT ( P-AKT ) , a downstream target of the insulin receptor , by western blot analysis of extracts from control and sir2 mutants using a fasting/refeeding paradigm . While sir2 mutants at one week of age respond normally to feeding by increasing their P-AKT levels , this response is reduced by two weeks of age and almost completely absent by three weeks of age ( Fig 2A–2C ) . Consistent with this result , the Foxo target 4EBP is incompletely repressed upon feeding in sir2 mutants as compared to controls at two weeks of age ( S3A Fig ) . A decrease in insulin signaling could be due to a defect in either insulin sensitivity or insulin secretion . As expected , both controls and sir2 mutants at one week of age have increased P-AKT in response to injected insulin , consistent with the activation of insulin signaling in response to dietary glucose ( Fig 2A and 2D ) . In contrast , while control flies at two weeks of age continue to show increasing levels of P-AKT with increasing concentrations of injected insulin , sir2 mutants fail to respond ( Fig 2E ) . This indicates that sir2 mutants are insulin resistant by two weeks of age . We also measured secreted levels of Drosophila insulin-like peptide 2 ( DILP2 ) in control and sir2 mutants to determine if reduced DILP2 secretion could contribute to their defects in insulin signaling [23] . This study revealed that circulating DILP2 increases with age in both fasting and fed controls , and increases approximately two-fold in response to feeding ( Fig 2F and 2G; S3B and S3C Fig ) , consistent with published wild-type responses [23] . Similar responses were seen in sir2 mutants under these conditions at both one and two weeks of age ( Fig 2F and 2G; S3B and S3C Fig ) . Taken together , these results indicate that defects in peripheral insulin sensitivity , but not insulin secretion , can account for the reduced insulin signaling in sir2 mutants . The GAL4/UAS system was used to determine where Sir2 is necessary and/or sufficient to regulate metabolic homeostasis using tissue-specific RNAi and rescue experiments . Ubiquitous expression of a sir2 RNAi construct efficiently eliminates sir2 mRNA as assayed by northern blot hybridization , indicating that this approach provides a strong loss of sir2 function ( S4A Fig ) . Driving the expression of this construct in the fat body , but not the muscles , intestine , insulin producing cells ( IPCs ) , or AKH-producing cells , disrupts insulin signaling and leads to hyperglycemia ( Fig 3A and 3B ) . Consistent with this , tissue-specific expression of a wild-type UAS-sir2 construct in the fat body of sir2 mutants is sufficient to restore insulin signaling in peripheral tissues , with no rescue seen upon expression of sir2 in the muscles or IPCs ( Fig 3C ) . In addition , expression of sir2 in the fat body , but not the muscle or IPCs , is sufficient to rescue the obesity of mutant animals , as reported previously ( S4B Fig ) [18] . Moreover , GAL4-driven expression of sir2 in the fat body of wild-type animals is sufficient to reduce TAG levels , consistent with previous reports of SIRT1 overexpression in mice ( S4C Fig ) [24] . These results define a central role for Sir2 in the fat body to regulate insulin signaling and suppress obesity and hyperglycemia . Given that the fat body performs functions analogous to the mammalian liver and white adipose tissue , these results are consistent with Sirt1 studies in mice and suggest the Drosophila provides a valuable model to determine the molecular mechanisms by which this sirtuin promotes a healthy metabolic state . As a first step to define the mechanisms by which Sir2 maintains metabolic homeostasis , we conducted RNA-seq analysis using quadruplicate RNA samples from control and sir2 mutants at two weeks of age . A total of 400 genes were identified as differentially expressed in sir2 mutants ( ≥1 . 5-fold change , p-value <0 . 05 ) , with 312 genes down-regulated and 88 genes up-regulated ( S1 Table ) . Gene ontology analysis revealed that the down-regulated genes are enriched in pathways related to the metabolic defects in sir2 mutants , including proteolysis , lipolysis , carbohydrate metabolism , and genes involved in redox homeostasis , while many up-regulated genes are involved in Drosophila defense responses ( S5 Fig ) [25] . This could be analogous to the known role for Sirt1 in suppressing adipocyte inflammation and could contribute to the fat body-specific functions for sir2 [26] . In addition , genes that are expressed at high levels in the intestine are enriched in the Sir2 down-regulated gene set , suggesting that this factor plays an important role in this tissue [27] . Because transcription factors are prominent targets of Sirt1 regulation , we compared our RNA-seq dataset from sir2 mutants with similar datasets for Drosophila transcription factors that control metabolism and insulin signaling . A small , but significant overlap is seen with genes regulated by the LXR homolog DHR96 in adults ( 15% of the 136 DHR96-regulated genes; Fig 4A ) , consistent with the known associations between Sirt1 and LXR [11 , 28] . Similarly , we saw a significant overlap between genes that are expressed at reduced levels in sir2 mutants and genes that increase their expression in foxo mutants ( 18% of the 312 sir2-down-regulated genes; Fig 4B ) [29] . This is consistent with the decreased insulin signaling in sir2 mutants as well as the known interactions between mammalian Sirt1 and Foxo [8] . Most remarkably , however , we saw a major overlap with genes regulated by the dHNF4 nuclear receptor , where more than 30% of the genes down-regulated in sir2 mutants are also down-regulated in dHNF4 mutants and nearly 60% of the genes up-regulated in sir2 mutants are up-regulated in dHNF4 mutants ( Fig 4C and 4D ) ( W . Barry and C . S Thummel , manuscript in revision ) . This observation suggests that dHNF4 represents a major downstream target for Sir2 regulation in Drosophila . The simplest explanation for the large overlap between the genes regulated by sir2 and dHNF4 is that dHNF4 protein levels are reduced in sir2 mutants . This is indeed the case as assayed by western blot , with an approximately 3-fold reduction in protein levels by two weeks of age , accompanied by a 1 . 7-fold reduction in dHNF4 mRNA , with more mild effects in younger flies ( Fig 4E; S6A Fig ) . Mammalian HNF4A can be regulated by acetylation , and lysines that are targets for this modification are conserved in Drosophila ( S6B Fig ) [30] . Consistent with this , when FLAG-tagged dHNF4 is immunoprecipitated and the levels of this protein are equalized between sir2 mutants and controls , there is a 3-fold increase in the proportion of immunoprecipitated protein that is acetylated in sir2 mutants ( Fig 4F ) . Sir2 protein is also present in this immunoprecipitate , indicating that these factors interact physically ( Fig 4F ) . Taken together , these results support the model that Sir2 interacts with dHNF4 to direct its deacetylation and maintain its stability . The reduced levels of dHNF4 protein in sir2 mutants combined with the large overlap between the dHNF4 and Sir2-regulated gene sets suggests that Sir2 stabilizes and promotes the function of dHNF4 . Consistent with this , ectopically increasing the levels of dHNF4 protein by crossing two copies of a genomic dHNF4-GFP-FLAG transgene into the sir2 mutant background is sufficient to restore normal insulin signaling responses in these animals ( Fig 4G ) . It is not sufficient , however , to rescue the hyperglycemia and elevated glycogen levels in sir2 mutants ( S6C and S6D Fig ) . We therefore conclude that some , but not all of the diabetic defects observed in sir2 mutants are due to a reduction in dHNF4 levels .
Here we show that sir2 mutants display a range of metabolic defects that parallel those seen in mouse Sirt1 mutants , including hyperglycemia , lipid accumulation , insulin resistance , and glucose intolerance [1–3] . These results suggest that the fundamental metabolic functions of Sirt1 have been conserved through evolution and that further studies in Drosophila can be used to provide insights into its mammalian counterpart . An additional parallel with Sirt1 is seen in our tissue-specific studies , where we show that sir2 function is necessary and sufficient in the fat body to maintain insulin signaling and suppress hyperglycemia and obesity , analogous to the role of Sirt1 in the liver and white adipose [9 , 13 , 24] . These results are also consistent with published studies of insulin sensitivity in Drosophila , which have shown that the fat body is the critical tissue that maintains glucose and lipid homeostasis through its ability to respond properly to insulin signaling [31 , 32] . Our studies also define the dHNF4 nuclear receptor as a major target for Sir2 regulation . Consistent with this , dHNF4 mutants display a range of phenotypes that resemble those of sir2 mutants , including hyperglycemia , obesity , and glucose intolerance [33] ( W . Barry and C . S . Thummel , manuscript in revision ) . As expected , these defects are more severe in dHNF4 loss-of-function mutants , consistent with sir2 mutants only resulting in a partial loss of dHNF4 protein . Sir2 interacts with dHNF4 and appears to stabilize this protein through deacetylation . This is an established mechanism for regulating protein stability , either through changes in target protein conformation that allow ubiquitin ligases to bind prior to proteasomal degradation , or through alternate pathways [34] . Further studies , however , are required to determine if this is a direct protein-protein interaction or part of a higher order complex . Although two papers have shown that mammalian Sirt1 can control HNF4A transcriptional activity through a protein complex , only one gene was identified as a downstream target of this regulation , PEPCK , leaving it unclear if this activity is of functional significance [10 , 30] . Our study suggests that this regulatory connection is far more extensive . The observation that one third of the genes down-regulated in sir2 mutants are also down-regulated in dHNF4 mutants ( including pepck , S6A Fig ) , and most of the genes up-regulated in sir2 mutants are up-regulated in dHNF4 mutants , establishes this nuclear receptor as a major downstream target for Sir2 regulation . It will be interesting to determine if the extent of this regulatory connection has been conserved through evolution . Despite this regulatory control , the over-expression of an HNF4 transgene was only able to partially restore the insulin signaling response and not the defects in carbohydrate homeostasis in sir2 mutants . This lack of complete rescue is not surprising , given that the Sirt1 family targets a large number of transcription factors , histones , and enzymes , providing multiple additional pathways for metabolic regulation . Moreover , the activity or target recognition of dHNF4 may be altered when it is hyperacetylated , in which case merely over-expressing this factor would not fully restore normal function . Future studies can examine more direct targets , both previously characterized and uncharacterized , for their functions in suppressing diabetes downstream of Sir2-dependent regulation . Finally , sir2 mutants represent a new genetic model for studying the age-dependent onset of phenotypes related to type 2 diabetes . We show that newly-eclosed sir2 mutant adults are relatively healthy , with elevated levels of free glucose and glycogen but otherwise normal metabolic functions . Their health , however , progressively worsens with age , with two-week-old sir2 mutants displaying lipid accumulation , fasting hyperglycemia , and reduced insulin signaling accompanied by insulin resistance . This is followed by the onset of glucose intolerance by three weeks of age . Previous studies of type 2 diabetes in Drosophila have relied on dietary models using wild-type animals that are subjected to a high sugar diet [31 , 32] . Although this is a valuable approach to better define the critical role of diet in diabetes onset , it is also clear that the likelihood of developing type 2 diabetes increases with age . The discovery that sir2 mutants display this pathophysiology provides an opportunity to exploit the power of Drosophila genetics to better define the mechanisms that lead to the stepwise onset of metabolic dysfunction associated with diabetes .
Flies were raised at 25°C on media containing 8% yeast , 6% glucose , 3% sucrose , and 1% agar in 1XPBS for all studies , with flies maintained at 18°C for genetic rescue studies . Adult ages are indicated in the figures and text and refer to the time period after eclosion from the pupal case . Males under ad libitum feeding conditions were used for all experiments unless otherwise indicated . For most fasting-re-feeding paradigms , flies were transferred to 1% agar in 1XPBS for 14–18 hours and re-fed on 10% glucose , 1% agar in 1XPBS for two hours . A transheterozygous combination of the sir22A-7-11 and sir24 . 5 deletion alleles was used for all mutant studies [20 , 21] . These alleles , all GAL4 lines ( except for the dilp2-gal4>UAS-dcr2 line ) , and the rescue construct , were outcrossed to a w1118 control strain , which was then used as a genetically-matched control for all experiments where indicated . The RNAi lines for sir2 ( #32481 ) and mCherry ( #35787 ) were obtained from the Bloomington Stock Center . Immunoprecipitation experiments for dHNF4 were performed on lines containing a transgenic genomic construct with dHNF4 carrying GFP and FLAG tags , driven by the endogenous dHNF4 promoter ( Bloomington #38649 ) . This transgene fully rescues dHNF4 mutant defects and was maintained in homozygous sir22A-7-11 or wild-type genetic backgrounds . Samples of five flies each were collected at one or two weeks of age and washed in 1XPBS . For triglycerides , glucose , glycogen , and protein , samples were homogenized in 120 μL of 1XPBS . For fasting glucose measurements , samples were homogenized in 100 μL trehalase buffer , and for ATP assays , samples were homogenized in 100 μL 6M guanidine HCl , 100mM Tris pH 7 . 8 , 4mM EDTA ) . Assays were performed as described [35] . Samples of ten flies were collected under the indicated conditions at one , two , or three weeks of age , and homogenized in 100 μL of RIPA buffer containing 1X protease inhibitors ( Roche cOmplete Mini EDTA-free protease inhibitor tablets ) . For P-AKT westerns , the buffer also contained Calyculin A and okadaic acid . Equivalent amounts of protein were resolved by SDS-PAGE ( 10% acrylamide ) , transferred to PVDF membrane overnight at 4°C , and blocked with 5% BSA prior to immunoblotting . Western blots were probed with antibodies for P-AKT ( 1:1000 , Cell Signaling #4060 ) , pan-AKT ( 1:1000 , Cell Signaling #4691 ) , Tubulin ( 1:5000 , Abcam #ab184613 ) , dHNF4 ( 1:1000–1:2000 , generated by L . Palanker-Musselman ) , Sir2 ( 1:50 , Developmental Studies Hybridoma Bank #p4A10 ) , and pan-acetyl-lysine ( 1:1000 , Cell Signaling #9441 ) . The westerns shown in the figures are representative of at least three biological replicates . Quantification was performed by measuring protein levels using ImageJ software . The values reported represent the mutant or experimental condition normalized to the control , unless otherwise specified . For P-AKT quantification , the ratio of P-AKT levels to total AKT levels was determined in refed sir2 mutants or the experimental condition and controls using ImageJ . The data from the fasted state was not quantified for these studies because the small changes in the basal levels of P-AKT under these conditions ( ranging from undetectable to low levels ) results in large statistical fluctuations that are not meaningful . RNA was isolated from control and sir2 mutants at two weeks of age using Trizol extraction ( Thermo Fisher ) and the Qiagen RNeasy Mini Kit . Library generation ( Illumina TruSeq RNA Sample Preparation Kit v2 with oligo dT selection ) and sequencing ( HiSeq 50 Cycle Single Read Sequencing v3 ) were performed by the High-Throughput Genomics core facility at the University of Utah . The Bioinformatics Core Facility at the University of Utah aligned this dataset to the genome , utilizing the Genome Build DM3 from April 2006 . Cut-offs for significance were Log2 ratio ± 0 . 585 and p-value <0 . 05 ( <0 . 005 in all cases but two ) . RNA-seq data from this study can be accessed at NCBI GEO ( accession number: GSE72947 ) . A standard two-tailed Student’s t-test was used to determine significance on basic metabolite measurements ( Fig 1; S1F Fig ) . GraphPad PRISM 6 software was used to plot data and perform statistical analysis on all other measurements . Pairwise comparison p-values were calculated using a two-tailed Student’s t-test , and multiple comparison p-values were calculated using two-way ANOVA or Bonferroni correction . For metabolomics , p-values reflect a standard two-tailed unpaired t-test after a Welch’s correction for different variances . For the starvation sensitivity experiment , the p-values reflect results from both a Log-Rank Mantel-Cox test as well as a Gehan-Breslow-Wilcoxon test . For gene-regulatory overlaps , the p-values reflect results from chi-square tests . The UAS-sir2 rescue construct was generated by PCR amplification of the sir2 coding region using primers designed to incorporate a KpnI restriction site in the forward primer ( CGCGGGTACCCCAAATGGGTGCGAAGCTGACG ) and an XbaI site in the reverse primer ( CGCGTCTAGAGGCCCTCGGCTACGATTTCGCAG ) . The template for this reaction was cDNA generated from wild-type RNA using the ProtoScript M-MuLV Taq RT-PCR Kit ( NEB ) . The gel-purified PCR product was digested with KpnI and XbaI and inserted into the multiple cloning site of pUAST-attB . This construct was integrated into each of three attP sites that are predicted to be silent ( attP40 , attP2 , attP3 ) using standard methods ( BestGene Inc . ) [36] . We then combined this transgene with our sir22A-7-11 allele in order to study its effect in a transheterozygous sir2 mutant background . Only the rescue line inserted on the third chromosome at attP2 , however , had sufficiently low background levels of sir2 expression to allow us to see changes in triglyceride levels and insulin signaling using tissue-specific GAL4 drivers , as shown in Figs 3C and S4B . Background expression from the UAS-sir2 transgene in all three lines is sufficient to rescue the hyperglycemia of sir2 mutants , preventing us from examining the tissue-specific regulation of this response in Fig 3 . RNA was isolated from samples containing 10–15 flies using Trizol ( Thermo Fisher ) . Males were used for all studies except for the Act>sir2-RNAi experiment in S4A Fig , as only females were obtained from this cross . Northern blot transfers and hybridizations were performed as previously described [37] . Feeding rates were measured by using radioactive media containing ~5 , 000 cpm/μL α-32P-dCTP in 8% yeast , 6% glucose , and 3% sucrose in 1% agar . Male flies at one or two weeks of age were fasted overnight and then allowed to re-feed on the labeled media for two hours , after which they were transferred to unlabeled food for 45 minutes and sorted into samples of five flies on ice . A scintillation counter was used to measure the radioactivity in each sample , and this value was used to determine the relative volume of media consumed . Groups of 197–198 flies of each genotype , at two or three weeks of age , were transferred to fresh food for about 12–24 hours , and then transferred to starvation media . Lethality was monitored every four to eight hours , with surviving flies transferred to fresh starvation media at least once over the course of the experiment . Flies at two or three weeks of age were fasted overnight for 15–18 hours prior to re-feeding on 10% glucose . After one hour , flies were transferred back to starvation media for either two or four hours . Samples were collected at each time point for glucose assays , which were performed as described [35] . Bovine insulin ( Sigma ) was dissolved in 1% acetic acid at 1 mg/mL before dilution to between 0 . 5 nM-100 nM in 1XPBS and 5% food dye . Flies were fasted overnight ( one week old flies ) or 4 hours ( two week old flies ) prior to insulin/dye injection in the thorax , until the dye was visible throughout the head and abdomen . Injections were performed for 10–15 minute intervals followed by an additional 30 minute rest period prior to collection of protein samples for western blot analysis . Concentrations at 1 week of age were 100 nM , and at 2 weeks of age were 0 . 5–1 . 0 nM [31 , 38] . To extract hemolymph , 30 one week old flies were punctured in the thorax between the head and wing junction using a tungsten needle and centrifuged at 9 , 000xg for five minutes through a Zymo-Spin IIIC filter ( Zymo Research ) . These samples were diluted 1:100 in Trehalose buffer and heat treated at 70°C for five minutes . Final dilutions of 1:200 and 1:400 were used for glucose assays . ELISA assays were performed as described on one and two week old flies [23] . Heterozygous control and homozygous sir22A-7-11 mutant lines were established that contained two copies of the transgenic dilp2 construct carrying HA and FLAG tags , driven by the genomic dilp2 promoter in a dilp2 mutant background ( Dilp2-HF ) . Ten flies were collected per sample . Undiluted circulating Dilp2-HF was measured from hemolymph samples and total Dilp2-HF levels were measured at a 1:10 dilution . Samples of fifteen adult males at two weeks of age were snap-frozen in liquid nitrogen and prepared for analysis by gas chromatography-mass spectrometry ( GC/MS ) as described [35] . Each experiment was performed on six independent samples , and each experiment was repeated three times . The data presented reflect the combined replicates from all three experiments , normalized within each experiment , for a total of 17–18 biological replicates per group . In one experimental replicate we failed to detect DHAP , for which there are only 12 biological replicates per group . Samples were collected from control and sir22A-7-11 homozygous lines at one week of age containing two copies of the dHNF4-GFP-FLAG genomic transgene . Ten flies were homogenized in 100 μL homogenization solution consisting of RIPA buffer with protease inhibitors ( Roche cOmplete Mini EDTA-free protease inhibitor tablets ) . Mouse anti-FLAG antibody ( Sigma #F1804 ) was added to this homogenate at a 1:500 dilution and incubated for one hour , rotating at 4°C . A 1:1 mixture of Protein A/Protein G Dynabeads ( Life Technologies ) was washed with 1 mL RIPA three times before being resuspended in homogenization buffer . The equivalent of 10–20 μL of the original volume of washed beads was added to each homogenate and incubated for an additional two hours , rotating at 4°C . Immunoprecipitates were then eluted according to standard procedures in 1X sample buffer with protease inhibitors . 5–7 . 5 μL of the resulting elutes were loaded into a 10% SDS-PAGE gel Proteins were resolved by SDS-PAGE and analyzed on western blots as described above .
|
The sirtuins are a highly conserved family of deacetylases with targets that range from DNA-associated histones to transcription factors . The activities of these enzymes are dependent upon the energetic state of the cell as they utilize the coenzyme NAD+ , an important electron carrier in central metabolic pathways . We have found that loss of the Drosophila homolog of the founding member of the sirtuin family , sir2 , leads to age-progressive metabolic disease with symptoms similar to those of type 2 diabetes . In addition , we show that the Drosophila HNF4 nuclear receptor is deacetylated and stabilized by Sir2 , and that it accounts for a major part of the transcriptional program controlled by Sir2 . This work provides a new genetic model of insulin resistance in Drosophila and establishes HNF4 as a critical downstream target in the Sir2 signaling pathway .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"carbohydrate",
"metabolism",
"invertebrates",
"medicine",
"and",
"health",
"sciences",
"chemical",
"compounds",
"diabetic",
"endocrinology",
"hyperglycemia",
"carbohydrates",
"glucose",
"metabolism",
"animals",
"organic",
"compounds",
"glucose",
"hormones",
"endocrine",
"physiology",
"animal",
"models",
"physiological",
"processes",
"drosophila",
"melanogaster",
"model",
"organisms",
"homeostasis",
"drosophila",
"research",
"and",
"analysis",
"methods",
"insulin",
"lipids",
"fats",
"endocrinology",
"chemistry",
"insects",
"metabolic",
"disorders",
"arthropoda",
"biochemistry",
"organic",
"chemistry",
"physiology",
"monosaccharides",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"insulin",
"signaling",
"metabolism",
"organisms"
] |
2016
|
Sir2 Acts through Hepatocyte Nuclear Factor 4 to maintain insulin Signaling and Metabolic Homeostasis in Drosophila
|
Mass cytometry ( CyTOF ) has greatly expanded the capability of cytometry . It is now easy to generate multiple CyTOF samples in a single study , with each sample containing single-cell measurement on 50 markers for more than hundreds of thousands of cells . Current methods do not adequately address the issues concerning combining multiple samples for subpopulation discovery , and these issues can be quickly and dramatically amplified with increasing number of samples . To overcome this limitation , we developed Partition-Assisted Clustering and Multiple Alignments of Networks ( PAC-MAN ) for the fast automatic identification of cell populations in CyTOF data closely matching that of expert manual-discovery , and for alignments between subpopulations across samples to define dataset-level cellular states . PAC-MAN is computationally efficient , allowing the management of very large CyTOF datasets , which are increasingly common in clinical studies and cancer studies that monitor various tissue samples for each subject .
Analyses of CyTOF data rely on many of the tools and ideas from flow cytometry ( FC ) data analysis , as CyTOF datasets are essentially higher dimensional versions of flow cytometry datasets . Currently , the most widely used method in FC is still human hand-gating , as other methods often fail to extract meaningful subpopulations of cells automatically . In hand-gating , we draw polygons or other enclosures around pockets of cell events on a two-dimensional scatterplot to define subpopulations and cellular states that are observed in the data . This process is painfully time-consuming and requires advance knowledge of the marker panel design , the quality of the staining reagents , and , most importantly , a priori what cell subpopulations to expect to occur in the data . When presented with a new set of marker panels and biological system , the researcher would find it difficult to delineate the cell events , especially in high-dimensional and multi-sample datasets . The inefficient nature of hand-gating in flow cytometry motivated algorithmic development in automatic gating . Perhaps the most popular is flowMeans[1] , which is optimized for FC and can learn subpopulations in FC data[2] in an automated manner; however , it has not been successfully applied to CyTOF data analysis . Currently , most data analysis tools created for flow cytometry data analyses are not easily applicable for high-dimensional datasets[3] . An exception is SPADE , which was developed and optimized specifically for the analysis of CyTOF datasets[3] . flowMeans and SPADE constitute the leading computational methods in cytometry , but as shown later in this work , their performance may become sub-optimal when challenged with large and high-dimensional datasets . There are also other recent clustering-based tools that utilize dimensionality reduction and projections of high-dimensional data , however , these tools do not directly learn the subpopulations for all the cell events , and may be too slow to complete data analysis for an increasing amount of samples . In this study , we address the data analysis challenges in two major steps . First , we propose the partition-assisted clustering ( PAC ) approach , which produces a partition of the k-dimensional space ( k = number of markers ) that captures the essential characteristic of the data distribution . This partitioning methodology is grounded in a strong mathematical framework of partition-based high-dimensional density estimation[4–7] . The mathematical framework offers the guarantee that these partitions approximate the underlying empirical data distribution; this step is faster than the recent k-nearest neighbor-based method [8] and is essential to the scalability of our clustering approach to analyze datasets with many samples . The clustering of cells based on recursive partitioning is then refined by a small number of k-means style iterations before a merging step to produce the final clustering . Secondly , the subpopulations learned separately in multiple different but related datasets can be aligned by marker network structures ( multiple alignments of networks , or MAN ) , making it possible to characterize the relationships of subpopulations across different samples automatically . The ability to do so is critical for monitoring changes in a subpopulation across different conditions . Importantly , in every study , batch effect is present; batch effects shift subpopulation signals so that the means can be different from experiment to experiment . PAC-MAN naturally addresses batch effects in finding the alignments of the same or closely related subpopulations from different samples . PAC-MAN finds homogeneous clusters efficiently with all data points in a scalable fashion and enables the matching of these clusters across different samples to discover cluster relationships in the form of clades .
PAC has two parts: partitioning and post-processing . In the partitioning part of PAC , the data space is recursively divided into smaller hyper-rectangles based on the number of data points in the locality ( Fig 1A ) . The partitioning is accomplished by either Bayesian Sequential Partition ( BSP ) with limited look-ahead ( Fig 1A and 1B ) or Discrepancy Sequential Partition ( DSP ) ( Fig 1A ) ; these are two fast variants of partition-based density estimation methods previously developed by our group [4–7] , with DSP being the fastest . BSP and DSP divide the sample space into hyper-rectangles with uniform density value in each of them . The subsetting of cells according to the partitioning provides a principled way of clustering the cells that reflects the characteristics of the underlying distribution . In particular , each significant mode is captured by a number of closely located rectangles with high-density values ( Fig 1C ) . Although this method allows a fast and unbiased localization of the high-density regions of the data space , we should not use the hyper-rectangles directly to define the final cluster boundaries for two reasons . First , real clusters are likely to be shaped elliptically , therefore , the data points in the corners of a hyper-rectangle are likely to be incorrectly clustered . Second , a real cluster is often split into more than one closely located high-density rectangles . We designed post-processing steps to overcome these limitations: 1 ) a small number of k-means iterations is used to round out the corners of the hyper-rectangles , 2 ) a merging process is implemented to ameliorate the splitting problem , which is inspired by the flowMeans algorithm . The details of post-processing are given in the Materials and Methods . The resulting method is named b-PAC or d-PAC depending on whether the partition is produced by BSP or DSP . An approach to analyze multiple related samples of CyTOF data is to pool all samples into a combined sample before detection of subpopulations . This is a natural approach under the assumptions that there are no significant batch effects or systematic shifts in cell subpopulations across the different samples . However , such assumptions may not hold due to one or more of the following reasons: Could we extract shared information that allows us to interpret cross-sample similarities and differences ? We note that efforts were made to analyze cross-sample relationships in a previous publication [10] , in which the data was carefully collected with barcode reagents in uniform staining , which enable pooling of the data for downstream analysis . Experimentally , it would be difficult to up-scale the barcoding and uniform staining control to a larger number of samples . Furthermore , previous efforts were dependent on down-sampling of the data points , which would significantly affect the clustering results . While it is possible , through careful experimental design and cross-sample controls , to establish uniform staining for a small pooled sample data analysis , there is a need to resolve the above batch effect difficulties for studies that require scalability , such as in the clinical setting in which hundreds of patient blood samples are collected at different times . To ameliorate the difficulties of potential high-dimensional cluster shifts and scalability , we have designed an alternative approach that is effective in the presence of substantial systematic between-sample variation . In this approach , each sample is analyzed separately ( by PAC ) to discover within-sample subpopulations . As an exploration step , we over-partition to capture both large and small subpopulations in high-dimension . The subpopulations from all samples are then compared to each other based on a pairwise dissimilarity measure designed to capture the differences in within-sample distributions ( among the markers ) across two subpopulations . Using this dissimilarity , we perform bottom-up hierarchical clustering of the subpopulations to represent the relationship among the subpopulations . The resulting tree of subpopulations is then used to guide the merging of subpopulations from the same sample , and to establish linkage of related subpopulations from different samples . We note that the design of a dissimilarity measure ( Materials and Methods ) that is not sensitive to systematic sample-to-sample variation is a novel aspect of our approach . The merging of subpopulations from the same sample is also important , as it offers a way to consolidate any over-partitioning that may have occurred during the initial PAC analysis of each sample . We emphasize that , as with the usage of all statistical methods , the user must utilize samples or datasets that are considered as good as possible and that the sample comparisons make biological sense; interpretation of the analysis results rely on the researchers to collect data with validated reagents for all samples . In general , sensible data would come from 1 ) samples that are carefully prepared to not include contamination of cells from other tissues , 2 ) cytometry panel with validated markers that enable the observation of known , coherent cell subpopulations in the tissue samples ( important for determining the number of PAC clusters to explore in the partition step to control for aggressive over-partitioning ) , 3 ) successful execution of standard cytometry experiment protocol , and 4 ) collection of data to achieve enough cell events ( important for building stable network structures ) . These steps would ensure the reproducibility of PAC-MAN data analysis . In addition , any novel subpopulation discovery or difference between samples observed should be validated with downstream experiments ( perhaps using low-dimensional flow cytometry and sorting methods ) . Appropriate initialization of clustering is very important for eventually finding the optimal clustering labels; PAC works well because the implicit density estimation procedure yields rational centers to learn the modes of sample subpopulations . When tested on the hand-gated CyTOF data on the bone marrow sample in [11] , compared to kmeans alone , PAC gives lower total sums of squares and higher F-measures in the subpopulations ( Fig 2A and 2B ) . In the comparison to kmeans , we utilized random kmeans initialization by Lloyd ( and Forgy ) , which uses random initialization , and also kmeans++ initialization , which uses a more advanced initialization [12 , 13] . The process of rational initialization also helps PAC to converge in 50 iterations ( Fig 3 ) in post-processing , whereas k-means performs very poorly even after 5000 iterations ( Fig 4 ) . Through the lens of t-SNE plots ( Fig 4 ) , the PAC results are more similar to the hand-gating results , while the k-means , flowMeans , and SPADE clustering results perform poorly . In flowMeans , several large subpopulations are merged . SPADE’s separation of points is inconsistent and highly heterogeneous , probably due to its down-sampling nature . On the other hand , by inspection , PAC obtains similar separation for both the major and minor subpopulations as the hand-gating results . In the systematic simulation study , we challenged the methods with different datasets with varying number of dimensions , number of subpopulations , and separation between the subpopulations . The F-measure and p-measures for the PAC methods are consistently equal or higher than that of flowMeans and SPADE ( Table 1 and S2A Fig ) . For some higher dimension cases in which the subpopulation separation is relatively small , SPADE failed to cluster . In addition , we observe that flowMeans gives inconsistent F-measures for similar datasets ( Table 1 ) , which may be due to the convergence of k-means to a local minimum without a rational initialization . Next , we tested the methods based on published hand-gated cytometry datasets to see how similar the estimated subpopulations are to those obtained by human experts . We applied the methods on the hematopoietic stem cell transplant and Normal Donors datasets from the FlowCAP challenges[2] and on the subset of gated mouse bone marrow CyTOF dataset ( Dataset 9 ) recently published[11] . The gating strategy of the CyTOF dataset is provided in S1 Fig . The dataset and expert gating strategy are the same as described earlier[14] . Note that in the flow cytometry data , the computed F-measures are slightly lower than that reported in FlowCAP; this is due to the difference in the definition of F-measures . Overall , the PAC outperforms flowMeans and SPADE by consistently obtaining higher F-measures ( Table 1 ) . In particular , in the CyTOF data example , PAC generated significantly higher F-measures ( greater than 0 . 82 ) than flowMeans and SPADE ( 0 . 59 and 0 . 53 , respectively ) . In addition , PAC gives higher overall subpopulation-specific purities ( S2B Fig and S1 Table ) . These results indicate that PAC gives consistently good results for both low and high-dimensional datasets . Furthermore , PAC results match human hand-gating results very well . The t-SNE ‘islands’ in the plots are well-colored by the PAC methods , demonstrating that both major and minor/rare subpopulations are captured . The consistency between PAC-MAN results and hand-gating results in this large data set confirms the practical utility of the methodology . We use t-SNE plots heavily for visualization because t-SNE is a great visualization tool . It is reasonable to ask whether one can obtain good subpopulations by performing cluster analysis on the low-dimensional data points output by t-SNE . Currently , this alternative approach is computationally expensive and not scalable as existing t-SNE implementations cannot be scaled to millions of high-dimensional points , restricting this analysis approach to only hundred of thousands of points in practice . In the downstream , hierarchical clustering or kmeans clustering could be applied; however , hierarchical clustering is very expensive due to the maintenance of a distance matrix during calculations ( cannot be easily performed for data with more than thousands of points ) , while kmeans clustering does not give satisfactory results ( S7 Fig ) due to the ‘flattened’ geometry of the high-dimensional points in the t-SNE embedding . Thus , embedding is good for visualization but it is not supposed to capture all information of clusters efficiently . In CyTOF data analysis , we recommend performing PAC methods on the dataset , and utilize t-SNE plots to visualize the clustering results with a subset of points for confirmation . It is natural to analyze samples separately then combine the subpopulation features for downstream analysis in the multiple samples setting . However , we need to resolve the batch effects . Two distinct subpopulations could overlap in the combined/pooled sample , such as in the case when the data came from two generations of CyTOF instruments ( newer instrument elevates the signals ) . On the other hand , in cases with changing means , two subpopulations can evolve together such that their means change slightly , but enough to shadow each other when samples are merged prior to clustering . We introduce Multiple Alignments of Networks to resolve the management issue surrounding the organization of homogeneous clusters found in the PAC step ( Fig 5 ) . First , we consider the overlapping scenario ( Fig 6A ) . When viewed together in the merged sample , the right subpopulation from sample 1 overlaps with the left subpopulation in sample 2 ( Fig 6B left panel ) . There is no way to use expression level alone to delineate the two overlapping subpopulations ( Fig 6B right panel ) . By learning more subpopulations using PAC , there are some hints that multiple subpopulations are present ( Fig 6C ) . Despite these hints , it would not be possible to say whether the shadowed subpopulations relate in any way to other distinct subpopulations . PAC-MAN resolves the overlapping issue by analyzing the samples separately ( Fig 7 ) . In the case in which we do not know a priori the number of true subpopulations , we learn three subpopulations per sample ( Fig 7A ) . The network structures of the subpopulations discovered are presented in Fig 7B and 7C and we see that the third subpopulations from the two samples share the same network structures , while the first two subpopulations of the two samples differ by only one edge; these respective networks are clustered together in the dendrogram ( Fig 8A right panel; subpopulation indexes are suffixes on the dendrogram ) . By utilizing the networks , the clades that represent the same and/or similar subpopulations of cells can be established . Clustering by network structures alone resolves the points in the data ( Fig 8A , left panel ) . In contrast , alignment by marker ( gene expression ) levels cannot resolve the batch effect ( Fig 8B ) . Next we consider the case with dynamic evolution of subpopulations that models the treatment-control and perturbation studies . The interesting information is in tracking how subpopulations change over the course of the experiment . In the simulation , we have generated two subpopulations that nearly converge in mean expression profile over the time course ( Fig 9 ) . The researcher could lose the dynamic information if they were to combine the samples for clustering analysis . As in the previous case , we could use PAC to learn several subpopulations per sample ( Fig 10 ) . Then , with the assumption that there are two evolving clusters from data exploration , we align the subpopulations to construct clades of same and/or similar subpopulations ( Fig 11 left panel ) based on the network structural information ( S3 Fig ) . With network and expression level information in the alignment process , the two subpopulations or clades can be resolved naturally ( Fig 11 right panel ) . With networks in hand , we could further characterize the relationships between subpopulations across samples . However , the alignment process needs to work well for true linkage to be established . We could align by network alone , by expression ( or marker ) means , or both . Fig 11 presents these alternatives in comparison . By using all the subpopulation networks , the results still contain subsets of misplaced cells ( 11 left panel ) . This is because small clusters of cells have noisy underlying covariance structure; therefore , the networks cannot be accurately inferred . These structural inaccuracies negatively impact the network clustering . The ( mean ) marker level approach also does not work well ( Fig 11 center panel ) due to the subpopulation mean shifts across samples . On the other hand , the sequential approach works well ( Fig 11 right panel ) . In the sequential approach , larger ( >1000 ) subpopulations’ networks are utilized for the initial alignment process . Next , the smaller subpopulations , which have noisy covariance , are merged with the closest larger , aligned subpopulations . Thus , more subpopulations could be discovered upstream ( in PAC ) , and the network alignment would work similarly as the smaller subpopulations , which could be fragments of a distribution , do not impact the alignment process ( S4A and S4B Fig ) . Moreover , in the network inference step , unimportant edges can negatively impact the alignment process ( S4C Fig ) in the network-alone case . Biologically , this means that edges that do not constrain or define the cellular state should not be utilized in the alignment of cellular states . Effectively , the threshold placed on the number of edges in the network inference controls for the importance of the edges . Thus , the combined alignment approach works well and allows moderate over-saturation of cellular states to be discovered in the PAC step so that no advance knowledge of the exact number of subpopulations is necessary . It is important to note that we have not utilized high-dimensional mutual information for network structure inference , which is computationally intensive . It may be possible that there exist complex relationships between more than two markers that could yield different network structures for two subpopulations that otherwise would have the same network structure . However , in our analysis of cytometry data , pairwise mutual information with downstream processing yields robust characterization of the cellular state relationships between subpopulations . We use the recently published mouse tissue dataset [11] to illustrate the multi-sample data analysis pipeline . The processed dataset contains a total of more than 13 million cell events in 10 different tissue samples , and 39 markers per event ( S2 Table ) . The original research results centered on subpopulations discovered from hand-gating the bone marrow tissue data to find ‘landmark’ subpopulations; the rest of the data points were clustered to the most similar landmark subpopulations . While this enables the exploration of the overall landscape from the perspective of bone marrow cell types within an acceptable time frame , a significant amount of useful information from the data remains hidden; a larger dataset would make it infeasible to analyze by manual gating and existing computational tools to learn the relationships of the cellular states among all samples . In addition , a natural question is how well do the bone marrow cell types represent the whole immune system ? In contrast to the one-sample perspective , using d-PAC-MAN , the fastest approach by our comparison results , we can perform subpopulation discovery for each sample automatically and then align the subpopulations across samples to establish dataset-level cellular states . On a standard Core i7-44880 3 . 40GHz PC computer , the single-thread data analysis process with all data points and optimization takes about two hours to complete , which is much faster than alternative methods . With multi-threading and parallel processing , the data analysis procedure can be completed very quickly . As mentioned earlier , PAC results for the bone marrow subsetted data from this dataset matches closely to that of the hand-gated results . This accuracy provides confidence for applying PAC to the rest of the dataset . Figs 12 and 13 show the t-SNE plots for subpopulation discovered ( top panel of each sample ) and the representative subpopulation established ( bottom panel of each sample ) for the entire dataset . In the PAC discovery step , we learn 50 subpopulations per sample without advance knowledge of how many subpopulations are present . This moderate over-partitioning of the data samples leads to a moderate heterogeneity in the t-SNE plots . From tests , we have found that learning 2–3 times the expected number of subpopulations in the sample works well; it is important to emphasize that aggressive over-partitioning is suboptimal because it creates very small subpopulations that have unstable covariance structures , which removes these small clusters data points from network alignment . Next , the networks are inferred for the larger subpopulations ( with number of cell events greater than 1000 ) , and the networks are aligned for all the tissue samples . To choose the optimal number of total subpopulations to output , we perform the elbow point test at this step , in which we calculated the within cluster standard deviations while varying the number of subpopulations outputted for the entire dataset . The elbow point rests at 130 clusters ( S8 Fig ) , and we outputted 130 representative subpopulations , also called clades , for the entire dataset to account for the traditional immunological cellular states and sample-specific cellular states present . Within samples , the subpopulations that cluster together by network structure are aggregated . The smaller subpopulations ( <1 , 000 cells each , not involved in network alignment ) are either merged to the closest larger subpopulation or establish their own sample-specific subpopulation by expression alignment . We attempt to assign these very small subpopulations back with larger clades by grouping all subpopulations within each sample into 5 expression-level clusters ( using cluster centroids ) , and thus we kept the larger subpopulations and a maximum of 4 minor sample-specific subpopulations for each tissue sample . Subpopulations with less than 100 cell events were discarded . The representative subpopulations ( 143 total including sample-specific minor subpopulations ) follow the approximate distribution of the cell events on the t-SNE plots and the aggregating effect cleans up the heterogeneities due to over-partitioning in the PAC step . The cell type clades are the representative subpopulations for the entire dataset , and they could either be present across samples or in one sample alone . Their distribution is visualized by a heatmap ( Fig 14 ) . While the bone marrow sample contains many cell types , only a subset of them are directly aligned to cell types in other samples , which means using the bone marrow data as the reference point leaves much information unlocked in the dataset . Therefore , the data suggests that the bone marrow cell types are not adequate in representing all cell types in the immune system . The cell types in the blood and spleen samples have various alignments with cell types in other samples . The lymph node samples share many clades likely due to the connection through the lymphatic vessels; the small intestine and colon samples also share many clades , probably due to closeness in location and biological function . Nevertheless , the results show that the tissue samples do not share exactly the same clades , suggesting that the immune system cells have different states in different organs . On the other hand , the thymus sample has few clades shared with other samples , which may be due to its functional specificity . PAC-MAN style analysis can be applied to align the tissue subpopulations by their means instead of network similarities ( S5 Fig ) . As done previously , 143 overall representative clades ( 130 network clades + 13 minor sample-specific subpopulations ) were outputted . The same aggregating effect is observed ( S5A Fig ) , and this is due to the organization from dataset-level variation in the means . Comparing to the network alignment , the means linkage approach has more subpopulations per sample; the subpopulation proportion heatmap ( S5B Fig ) shows more linking . Although the bone marrow sample subpopulations co-occur in the same clades slightly more with other sample subpopulations , this sample does not co-occur with many clades in the dataset . Thus , a PAC-MAN style analysis with means linkage also harvests additional information from the entire dataset . In general , the means alignment approach gives many more clades per sample than that of the network alignment PAC-MAN approach . In fact , the network approach has 88 linkages while the means approach has 270 linkages . The linkage plot ( S6A Fig ) shows that the low linkages occur slightly more frequently for the network approach . One consequence is that the network approach aggregates PAC subpopulations within sample more frequently; for instance , in the thymus sample , the network approach yields 13 clades ( and 2 minor sample-specific subpopulations ) while the means approach yields 39 clades . After aggregating , the clade sizes ( with unique participants per sample ) are plotted ( S6B Fig ) . The network approach tends to find fewer linkages , as more clades have sizes of less than 4 , while the means approach has more clades than the network approach with clade sizes greater than 4 . The network approach is more conservative due to the additional constraints from network structures . Conventionally , in the cytometry field , only the means are considered in the definition of cellular states . The network alignment is more stringent in the establishment of linkages; the network PAC-MAN approach defines cellular states with the additional information from network structures , and it has the effect of constraining the number of linkages between samples while finding linkages for subpopulations that are distant in their means . Further studies are needed to combine the information from both the marker level and network structures to organize the cellular states discovered in cytometry datasets , for example , through a weighted score based on the means and network alignments . In this study , we demonstrated that the covariance and network structures built from subpopulations are valuable and can be utilized to organize data-level cellular state relationships . To further characterize the cell types , we annotate the clades within each sample using the top network hub markers , which constrain the cellular states . The full network structure annotation , along with average expression profiles , is presented in S3 Table . The clade information is presented in the ClusterID column . The annotations for cells across different samples but within the same clades share hub markers . For example , in clade 1 for the blood and bone marrow samples , the cells share the hub markers Ly6C and CD11b . In the bone marrow sample , one important set of subpopulations is the hematopoietic stem cell subpopulations . One such subpopulation is present as clade 33 with the annotation F4/80 . CD16/32 . Sca1 . cKit and is about 1 . 18 percent in the bone marrow sample . Clade 33 is only present in the bone marrow sample , indicating that the PAC-MAN pipeline defines this as a sample-specific and coherent subpopulation using dataset-level variation . The thymus contains a large subpopulation clade 124 ( 84 . 07 percent ) that is characterized as CD5 . CD43 . CD3 . CD4 , suggesting it to be the maturing T-cell subpopulation . PAC-MAN generates both the clade and subpopulation signal ( or expression ) information . Fig 14 visualizes the occurrence and proportions of representative subpopulations in the dataset . To understand the expression levels of the markers for the subpopulation , a heatmap is constructed ( Fig 15 and S14 Fig ) . In high-dimension , the subpopulations can form regions in which similar cellular states are next to each other . Do subpopulations belonging to the same clade occupy the same region ? In addition , what is the spatial spread of subpopulations belonging to the same clade ? To visualize the clade relationships between subpopulations in the dataset , we construct the constellation plot ( Fig 16 ) . First , the centroids of the discovered subpopulations are inputted into a t-SNE visualization processing , which projects and separates the centroids onto a 2D plane . Next , the clades are color-coded such that 1 ) grey color indicates sample-specific clade and 2 ) non-grey colors indicate clades with multiple sample representation . Finally , we group the subpopulations in each clade by drawing lines to connect the closest clade subpopulation on the 2D plane , analogous to the visualization of stars by constellation nomenclature . The constellation plot is useful in looking at the spread of the clades in relation to other subpopulations . For example , clade 10 , which contains subpopulations that are CD45+CD3+CD5+CD8+ , and clade 8 , which contains subpopulations that are CD45+CD3+CD5+CD4+ , are T cells ( S14 Fig ) ; these two clade groups exist next to each other in the constellation plot , but they do not overlap . Clade 2 is in a region that contains CD45+CD19+B220+ subpopulations , which signify B cells . Furthermore , within each clade , the subpopulation networks are similar and contain similar hub genes . For instances , clades 2 and 8 represent data-level subsets of T cells and B cells , respectively; clades 2 and 8’s networks are presented in Figs 17 and 18 . Each clade has its unique network structures and a set of hub markers . Overall , in this analysis , we observe that clades defined by signal levels and network structures tend to occupy defined regions in high-dimensional space . Certainly , not all cell types are present in all tissue samples , and those immune cell subsets that are similar enough to be in the same clade may differ due to their tissue-specific , local environmental factors . We have presented the PAC-MAN data analysis pipeline . This pipeline was designed to remove major roadblocks in the utilization of existing and future CyTOF datasets . First , we established a quick and accurate clustering method that closely matches expert gating results; second , we demonstrated the management of multiple samples by handling mean shifts and batch effects across samples . We demonstrated that the inter-marker relationship in the form of mutual information networks is extremely useful in defining cellular states . The alignment of network structures allows researchers to find relationships between cells across samples without resorting to pooling of all data points . PAC-MAN allows the cytometry field to harvest information from the increasing amount of CyTOF data available . It is important to standardize multi-sample data analysis with automation so that discoveries based on multi-sample CyTOF datasets from different laboratories do not depend on the experts’ manual gating strategies and the grouping of subpopulations that is constrained by non-systematic computations . Furthermore , due to PAC-MAN’s generality , this pipeline can be utilized to analyze large datasets of high-dimension beyond the cytometry field .
There are two partition methods implemented in the comparison study: d-PAC and b-PAC . The results are similar , with d-PAC being the faster algorithm . Fig 1A illustrates this recursive process . d-PAC is based on the discrepancy density estimation ( DSP ) [7] . Discrepancy , which is widely used in the analysis of Quasi-Monte Carlo methods , is a metric for the uniformity of points within a rectangle . DSP partitions the density space recursively until the uniformity of points within each rectangle is higher than some pre-specified threshold . The dimension and the cut point of each partition are chosen to approximately maximize the gap in uniformity of two adjacent rectangles . BSP + LL is an approximation inference algorithm for Bayesian sequential partitioning density estimation ( BSP ) [5] . It borrows ideas from Limited-Look-ahead Optional Pólya Tree ( LL-OPT ) , an approximate inference algorithm for Optional Pólya Tree[6] . The original inference algorithm for BSP looks at one level ahead ( i . e . looking at the possible cut points one level deeper ) when computing the sampling probability for the next partition . It then uses resampling to prune away bad samples . Instead of looking at one level ahead , BSP + LL looks at h levels ahead ( h > 1 ) when computing the sampling probabilities for the next partition and does not do resampling ( Fig 1B ) . In other words , it compensates the loss from not performing resampling with more accurate sampling probabilities . For simplicity , ‘BSP + LL’ is shortened to ‘BSP’ in the rest of the article . We use the F-measure for comparison of clustering results to ground truth ( known in simulated data , or provided by hand-gating in real data ) . This measure is computed by regarding a clustering result as a series of decisions , one for each pair of data points . A true positive decision assigns two points that are in the same class ( i . e . same class according to ground truth ) to the same cluster , while a true negative decision assigns two points in different classes to different clusters . The F-measure is defined as the harmonic mean of the precision and recall . Precision P and recall R are defined as: P=TPTP+FP ( 2 ) R=TPTP+FN ( 3 ) where TP is the total number of true positives , FP is the total number of false positives and FN is the total number of false negatives . F-measure ranges from 0 to 1 . The higher the measure , the more similar the estimated cluster result is to the ground truth . This definition of F-measure is different than that of FlowCAP challenge[2] . The use of co-assignment of labels in this definition is a more accurate way to compute the true positives and negatives . Most of the existing measurements for clustering accuracy aim at measuring the overall accuracy of the entire datasets , i . e . comparing with the ground truth over all clusters . However , we are also interested in analyzing how well a clustering result matches the ground truth within a certain class . Specifically , consider a population with K classes in the ground truth: {C1 , C2 , … , CK} . We construct a class-specific index called the purity measure , or p-measure for short , to measure how well our clustering result matches the ground truth . This index is computed as follows: If we were to match a big cluster with a small class , even though the overlapping may be large , S1 would still be low since we have divided the score by the size of the cluster in S1 . In addition , we are interested in knowing how many points in Ck are clustered together by Lik , which is measured by S2 . After PAC , the discovered subpopulations typically have enough cells for the estimation of mutual information . This enables the construction of networks as the basis for cell type characterization . In these networks , the nodes represent the markers monitored in the experiment , while the edges represent a correlation/mutual information dependence relationship between the marker levels . Computationally , it is not good to directly use the mutual information networks constructed this way to organize the subpopulations downstream . The distance measure used to characterize the networks could potentially give the same score for different network structures . Thus , it is necessary to threshold the network edges based on the strength of mutual information to filter out the noisy and miscellaneous edges . In this work , these subpopulation-specific networks are constructed using the MRNET network inference algorithm in the Parmigene [15] R package . The algorithm is based on mutual information ranking , and outputs significant edges connecting the markers . The top d edges ( d is set to be 1x the number of markers in all examples ) are used to define a network for the subpopulation . This process enables a careful calculation of the distance measure . For each pair of subpopulation networks , we calculate a network distance , which is defined as follows . If G1 and G2 are two networks , let S be the set of shared edges and A be union of the of the edges in the two networks , then we define Similarity ( G1 , G2 ) =|S||A| ( 6 ) where |⋅| denotes the size of a set . This is known as the Jaccard coefficient of the two graphs . The Jaccard distance , or 1- Jaccard coefficient , is then obtained . This is a representation of the dissimilarity between each pair of networks; the Jaccard dissimilarity is the measure used for the downstream hierarchical clustering . We perform agglomerative clustering of the pool of subpopulations from all samples . This clustering procedure greedily links networks that are the closest in Jaccard dissimilarity , and yields a dendrogram describing the distance relationship between all the subpopulations . We cut the dendrogram to obtain the k clades of subpopulations . Subpopulations from the same sample and falling into the same clade are then merged into a single subpopulation ( Fig 5 ) . This merging step has the effect of consolidating the moderate over-partitioning in the PAC step . No merging is performed for subpopulations from different samples sharing the same clade . In this way , we obtain k clades of subpopulations , with each clade containing no more than one subpopulation from each sample . We regard the subpopulations within each clade as being linked across samples . In the above computation , only subpopulations with enough cells to define a stable covariance are used for network alignment via the Jaccard distance; the rest of the cell events from very small subpopulations are then merged with the closet clade by marker profile via distance of mean marker signals . If the small subpopulations are distant from the defined clades , then a new sample-specific clade is created for these small subpopulations . To efficiently find the practical number of clades to output for PAC-MAN , we utilize the elbow point analysis approach . Initially in the PAC step , the sample points are clustered into 2–3 times the expected number of sample subpopulations expected by the researcher . Next , we calculate the within-cluster errors , or distance from the subpopulation centroid , for each cluster in all samples , and we obtain the within-cluster errors for all sample . This calculation is performed for a range of numbers of clades in MAN . Loess smoothing is applied to the average within-cluster errors over the numbers of clades , and the researcher determines the location of the elbow point , which is then inputted into the final network alignment . To visualize the cellular state distribution in high-dimension , we construct the constellation plot . On the constellation plot , we observe two layers of information: the distribution of the clusters by expression level projection and the network similarities . By building the network structures and performing structural alignments , we remove extraneous connectivity between subpopulations that may appear close together in ‘expression space’ by grouping only subpopulations with strong network structural similarity . Those subpopulations that are in different clades but are close together on the constellation plot can be sample-specific subpopulations worth validating by future sorting and characterization experiments; these subpopulations are coherent clusters by expression and their network structures are different from those of other subpopulations . In the constellation plot construction , Barnes-Hut t-SNE with default setting ( perplexity of 30 and 1000 iterations ) was run on the centroids ( of expression/measurement signal ) of the discovered clade subpopulations for the entire dataset after PAC-MAN; t-SNE plot projects and separates the centroids in two dimensions . Next , the clades are color-coded such that 1 ) grey color indicates sample-specific clade and 2 ) non-grey colors indicate clades with multiple sample representation . The subpopulations in each clade are grouped by lines connecting the closest clade subpopulation , analogous to the visualization of stars by constellation nomenclature . Relative Euclidean distances ( in the t-SNE embedding ) between subpopulations and clade centers are utilized to prune away subpopulations that are far away within clades . For clades containing three or more subpopulations , the distances to clade centroids for each clade on the t-SNE plane are used as thresholds , and subpopulations that are more than twice ( threshold constant multiplier ) the average distance to their clade centroid are pruned . For clades with only two subpopulations , the distances between the subpopulations for each two-subpopulation clade are calculated; the mean of these distances for the two-subpopulation clades is used as a global threshold . Any two-subpopulation clade with separate larger than twice ( threshold constant multiplier ) this global threshold is pruned . The researcher also controls a maximum global separation threshold , and the pruning procedure uses the minimum of the thresholds to determine the pruning of clade subpopulations . All pruned away subpopulations are given new clade designation ( S9 Fig ) . To annotate the cellular states , we first apply PAC-MAN to learn the dataset-level subpopulation/clade labels . Next , these labels are used to learn the representative/clade networks . The top hubs ( i . e . the most connected nodes ) in these networks are used for annotation . This approach has biological significance in that important markers in a cellular state are often central to the underlying marker network , which is analogous to important genes in gene regulatory networks; these important markers have many connections with other markers . If the connections were broken , the cell would be perturbed and potentially driven to other states . To run t-SNE [16] a dimensionality reduction visualization tool , we utilized the scripts published here ( https://lvdmaaten . github . io/tsne/ ) . Default settings were used . To run SPADE , we first converted the simulated data to fcs format using Broad Institute’s free CSVtoFCS online tool in GenePattern[17] ( http://www . broadinstitute . org/cancer/software/genepattern# ) . Next , we carried out the tests using the SPADE package in Bioconductor R[18] ( https://github . com/nolanlab/spade ) . To run flowMeans , we carried out the tests using the flowMeans package in Bioconductor R[1] ( https://bioconductor . org/packages/release/bioc/html/flowMeans . html ) . In the comparisons , we selected only cases that work for all methods to make the tests as fair as possible . To calculate the mutual information of the subpopulations , we use the infotheo R package ( https://cran . r-project . org/web/packages/infotheo/index . html ) . To run network inference , we use the mrnet algorithm in the parmigne R package [15] . ( https://cran . r-project . org/web/packages/parmigene/index . html ) . The PAC R package can be accessed at: https://cran . r-project . org/web/packages/PAC/index . html To compare the clustering methods , we generated simulated data from Gaussian Mixture Model varying dimension , the number of mixture components , mean , and covariance . The dimensions range from 5 to 50 . The number of mixture components is varied along each dimension . The mean of each component was generated uniformly from a d-dimensional hypercube; we generated datasets using hypercube of different sizes , but kept all the other attributes the same . The covariance matrices were generated as AAT , where A is a random matrix whose elements were independently drawn from the standard normal distribution . The sizes of the simulated dataset range from 100k to 200k . The simulated data are provided as ( Datasets 1–6 ) . Datasets 1–6 are for the PAC part . Dataset 1 contains data with 5 dimensions; Dataset 2 contains data with 10 dimensions; Dataset 3 contains data with 20 dimensions; Dataset 4 contains data with 35 dimensions; Dataset 5 contains data with 40 dimensions; and Dataset 6 contains data with 50 dimensions . The ground truth labels are included as separate sheets in each dataset . When applying flowMeans , SPADE , and the PAC to the data , we preset the desired number of subpopulations to that in the data to allow for direct comparisons . Two data files were downloaded from the FlowCAP challenges[2] . One data file is from the Hematopoietic stem cell transplant ( HSCT ) data set; it has 9 , 936 cell events with 6 markers , and human gating found 5 subpopulations . Another data file is from the Normal Donors ( ND ) data set; it has 60 , 418 cell events with 12 markers , and human gating found 8 subpopulations . The files are the first ( ‘001’ ) of each dataset . These data files were all 1 ) compensated , meaning that the spectral overlap is accounted for , 2 ) transformed into linear space , and 3 ) pre-gated to remove irrelevant events . We used the data files without any further transformation and filtering . When applying flowMeans , SPADE , and the PAC to the data , we preset the desired number of subpopulations to that in the data to allow for direct comparisons . Human gated mass cytometry data was obtained by gating for the conventional immunology cell types using the mouse bone marrow data recently published[11] . The expert gating strategy is provided as S1 Fig . The gated sample subset contains 64 , 639 cell events with 39 markers and 24 subpopulations and it is provided as Dataset 9 . To test the performance of different analysis methods , the data was first transformed using the asinh ( x/5 ) function , which is the transformation used prior to hand-gating analysis; For SPADE analysis , we utilize the asinh ( x/5 ) option in the SPADE commands . The post-clustering results from flowMeans , SPADE , b-PAC , and d-PAC were then subsetted using the indexes of gated cell events . These subsetted results are compared to the hand-gated results . To test the linking of subpopulations , we generated simulated data from multivariate Gaussian with preset signal levels and randomly generated positive definite covariance matrices . There are two cases , batch effect and dynamic . Each simulated sample file has five dimensions , with two of these varying in levels; these are the dimensions that are visualized . Dataset 7 contains the data for general batch effects case and Dataset 8 contains the data for dynamic effects case . The ground truth labels are included as separate sheets in each dataset . The researcher preprocesses the data to 1 ) normalize the values to normalization bead signals , 2 ) de-barcode the samples if multiple barcoded samples were stained and ran together , and 3 ) pre-gate to remove irrelevant cells and debris to clean up the data[9 , 19] . Gene expressions look like log-normal distributions[20]; given the lognormal nature of the values , the hyperbolic arcsine transform is applied to the data matrix to bring the measured marker levels ( estimation of expression values ) close to normality , while preserving all data points . Often , researchers use the asinh ( x/5 ) transformation , and we use the same transformation for the CyTOF datasets analyzed in this study . In the Spitzer et al . , 2015 dataset[11] , three mouse strains were grown , and total leukocytes were collected from different tissues: thymus , spleen , small intestine , mesenteric lymph node , lung , liver , inguinal lymph node , colon , bone marrow , and blood . In each experiment , 39 expression markers were monitored . The authors used the C57BL6 mouse strain as the reference[11]; the data was downloaded from Cytobank , and we performed our analysis on the reference strain . First , all individual samples were filtered by taking the top 95% of cells based on DNA content and then the top 95% of cells based on cisplatin: DNA content allows the extraction of good-quality cells and cisplatin level ( low ) allows the extraction of live cells . Overall , the top 90% of cell events were extracted . The filtered samples were then transformed by the hyperbolic arcsine ( x/5 ) function , and merged as a single file , which contains 13 , 236 , 927 cell events and 39 markers per event ( S2 Table ) . Using PAC-MAN , we obtained 50 subpopulations in each sample , then , using elbow point analysis , we output 130 clades for the entire dataset . The 130 clades account for the traditional immune subpopulations and sample-specific subpopulations , which may include resident immune cells that are unique to certain tissues . In the network alignment step , smaller PAC subpopulations ( <1 , 000 cells ) are left out because they may not have stable covariance and network structures . We attempt to assign the left-out small subpopulations back to the dataset: hierarchical clustering of the cluster centroids ( marker signals or expression level ) was performed , and we limit the total number of unique small sample-specific subpopulation by generating 5 “expression” clades per sample in the clustering ( the larger subpopulations with a maximum of four sample-specific minor subpopulations that have less than 1 , 000 cells ) . Subsequently , any clade with less than 100 cells was discarded . Subpopulation proportion heatmap was plotted to visualize the subpopulation-specificities and relationships across the samples . Network annotation was performed using the hub markers of each representative subpopulation in each sample . Finally , we plotted the expression heatmap for all the clades and the constellation plot to visualize the cross-sample clade relationships .
|
Recently , the cytometry field has experienced rapid advancement in the development of mass cytometry ( CyTOF ) . CyTOF enables a significant increase in the ability to monitor 50 or more cellular markers for millions of cells at the single-cell level . Initial studies with CyTOF focused on few samples , in which expert manual discovery of cell types were acceptable . As the technology matures , it is now feasible to collect more samples , which enables systematic studies of cell types across multiple samples . However , the statistical and computational issues surrounding multi-sample analysis have not been previously examined in detail . Furthermore , it was not clear how the data analysis could be scaled for hundreds of samples , such as those in clinical studies . In this work , we present a scalable analysis pipeline that is grounded in strong statistical foundation . Partition-Assisted Clustering ( PAC ) offers fast and accurate clustering and Multiple Alignments of Networks ( MAN ) utilizes network structures learned from each homogeneous cluster to organize the data into data-set level clusters . PAC-MAN thus enables the analysis of a large CyTOF dataset that was previously too large to be analyzed systematically; this pipeline can be extended to the analysis of similarly large or larger datasets .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Materials",
"and",
"methods"
] |
[
"flow",
"cytometry",
"medicine",
"and",
"health",
"sciences",
"split-decomposition",
"method",
"random",
"variables",
"covariance",
"simulation",
"and",
"modeling",
"bone",
"marrow",
"cells",
"multiple",
"alignment",
"calculation",
"lymph",
"nodes",
"mathematics",
"lymphatic",
"system",
"cytometry",
"research",
"and",
"analysis",
"methods",
"computer",
"and",
"information",
"sciences",
"animal",
"cells",
"probability",
"theory",
"spectrophotometry",
"cytophotometry",
"cell",
"biology",
"computational",
"techniques",
"data",
"visualization",
"anatomy",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"physical",
"sciences",
"spectrum",
"analysis",
"techniques"
] |
2017
|
Scalable multi-sample single-cell data analysis by Partition-Assisted Clustering and Multiple Alignments of Networks
|
Cathepsin-like enzymes have been identified as potential targets for drug or vaccine development in many parasites , as their functions appear to be essential in a variety of important biological processes within the host , such as molting , cuticle remodeling , embryogenesis , feeding and immune evasion . Functional analysis of Caenorhabditis elegans cathepsin L ( Ce-cpl-1 ) and cathepsin Z ( Ce-cpz-1 ) has established that both genes are required for early embryogenesis , with Ce-cpl-1 having a role in regulating in part the processing of yolk proteins . Ce-cpz-1 also has an important role during molting . RNA interference assays have allowed us to verify whether the functions of the orthologous filarial genes in Brugia malayi adult female worms are similar . Treatment of B . malayi adult female worms with Bm-cpl-1 , Bm-cpl-5 , which belong to group Ia of the filarial cpl gene family , or Bm-cpz-1 dsRNA resulted in decreased numbers of secreted microfilariae in vitro . In addition , analysis of the intrauterine progeny of the Bm-cpl-5 or Bm-cpl Pro dsRNA- and siRNA-treated worms revealed a clear disruption in the process of embryogenesis resulting in structural abnormalities in embryos and a varied differential development of embryonic stages . Our studies suggest that these filarial cathepsin-like cysteine proteases are likely to be functional orthologs of the C . elegans genes . This functional conservation may thus allow for a more thorough investigation of their distinct functions and their development as potential drug targets .
Human lymphatic filariasis ( LF ) , caused by the filarial parasites Brugia malayi , Brugia timori and Wuchereria bancrofti , infects 120 million people worldwide , of which 40 million people show chronic disease symptoms ( www . globalnetwork . org ) [1] . The disease is estimated to be responsible for 5 . 5 million DALYs , and is the second leading cause of permanent and long-term disability worldwide [2] . A further one billion people ( 18% of the world's population ) are at risk of infection ( www . globalnetwork . org ) . The Global Programme to Eliminate Lymphatic Filariasis ( GPELF ) aims to use mass drug administration ( MDA ) to interrupt transmission and to reduce morbidity , with annual doses of a multi-drug regimen for at least five years ( www . filariasis . org ) [3] . The African Programme for Onchocerciasis Control aims to establish by 2010 community-based sustainable ivermectin treatments of 50 million people in 19 African countries having meso- and hyper-endemicity ( www . who . int/apoc/en/ ) . However , it appears unlikely that any of these MDA regimens will be sufficient to eliminate LF or onchocerciasis in all endemic areas [2] . Numerous technical challenges threaten the success of these eradication programs [2] , [4] , including incomplete efficacy of available drugs against adult filarial worms [5] , [6] , severe drug toxicity in people with heavy loiasis infections [3] , and the risk that filarial worms will develop resistance to the drugs available for MDA [7] , [8] . Ivermectin ( IVM ) resistance has been reported worldwide in several other parasitic nematodes [9] . The emergence of drug resistant strains of the O . volvulus parasite was initially suggested by reports of patients with onchocerciasis who failed to respond to IVM treatment [10] , [11] and a recent report from Ghana has provided the first direct proof of IVM resistance in O . volvulus populations [12] . A number of studies have associated IVM resistance with genetic markers ( reviewed in [13] ) , while a recent , more extensive study using worms taken from individuals before and after repeated IVM treatments , has demonstrated that IVM causes genetic selection on O . volvulus [14] . At present there are no alternative drugs for IVM for use in the onchocerciasis MDA programs that can reduce Mf or kill adult worms and no vaccines are available [15] . Vaccines that decrease transmission would complement MDA programs and may be necessary for complete elimination [13] . Notably , antibiotic treatment targeting Wolbachia endosymbionts of filarial nematodes is emerging as an alternative drug treatment for filariasis . Doxycycline treatment has been shown to result in an almost complete loss of Mf and macrofilaricidal effects in LF and in sterilization of adult worms and macrofilaricidal effects in O . volvulus ( reviewed in [16] , [17] ) . However , the period of treatment with doxycycline at present ( between 4–8 weeks ) is not applicable to mass treatment strategies , due to both the logistical difficulties and contra-indication in children under eight and pregnant women [16] , [17] , as a result antibiotic treatment regimens require further refinement . Therefore , a major priority has to be the identification of new drugs with strong activity against adult filarial worms ( macrofilaricidal ) which have new classes of chemistry , new molecular targets , and novel modes of action . Recent breakthroughs in genomics and chemistry make macrofilaricidal drug development more feasible , and accordingly a high priority goal with the WHO/TDR and the Bill & Melinda Gates Foundation [8] , [18] , [19] , [20] , [21] . RNA interference ( RNAi ) was first described in Caenorhabditis elegans where it was shown to spread systemically throughout the whole organism [22] and is widely used to identify gene function and has been developed for high-throughput genomics [23] , [24] , [25] . This powerful reverse genetics mechanism thus provides an invaluable tool which could be transferred to gene function studies and novel drug discovery in filarial nematodes . Moreover , it potentially provides an unprecedented opportunity to identify pre-validated drug targets after efficient mining of nematode genomic databases and RNAi genome wide C . elegans databases . [26] , [27][28][29] . RNAi has been successfully demonstrated in a number of parasitic nematodes ( reviewed in [30] , [31] ) , including filarial nematodes [32] , [33] , [34] , [35] , [36] . However , although RNAi has been demonstrated in parasitic nematodes , its application as a tool for high-throughput functional genomic screening for identification of essential parasite genes has not yet been achieved due to variable success in transferring the technology from C . elegans to parasitic nematodes and problems with inconsistent results and poor reproducibility [30] , [37] . While a wide range of target genes have been screened using RNAi in parasitic nematodes , only a small number of these genes have been selected due to their potential as drug targets [31] . Filarial proteases have been recognized as potential drug targets [33] , [38] . RNAi targeting the cathepsin L- and Z-like cysteine proteases ( CPL & CPZ respectively ) has clearly validated their essential role during O . volvulus L3 molting [33] . The function ( s ) of CPLs in filarial nematodes during embryogenesis , however , was only predicted indirectly by immunoelectron microscopy ( IEM ) [39] , [40] , while detailed studies of the CPLs in C . elegans [40] , [41] have provided more direct proof of their function in the free living nematodes . However , their predicted essential function during development may not always be conserved in both filarial nematodes and C . elegans , as has been shown for the roles of CPL and CPZ in the molting of O . volvulus and C . elegans [33] . Therefore it would be beneficial to be able to directly and reliably assess their essential function ( s ) using RNAi technology in filarial adult worms , in particular during embryogenesis . Here we demonstrate the use of double-stranded RNA ( dsRNA ) -mediated silencing to study the possible function of several of the cathepsin-like enzymes in B . malayi .
Adult female B . malayi , collected from the peritoneal cavities of infected jirds ( Meriones unguiculatus ) at day 80–85 post-infection , were kindly provided by the NIAID/NIH Filariasis Research Reagent Repository Center ( Athens , GA; www . filariasiscenter . org ) . Worms were washed once in normal culture medium ( CM; RPMI-1640 , 100 U/ml penicillin , 100 µg/ml streptomycin , 2 mM L-glutamine , 2 . 5 µg/ml amphotericin B , and 25 mM HEPES ( Sigma , St Louis , MO ) ) preheated to 37°C . Individual worms were transferred into 1 ml of normal CM in 48 well culture plates ( Corning Inc . Life Sciences , Lowell , MA ) and cultured overnight at 37°C under 5% CO2 to ensure the absence of any contaminating microorganisms . Release of Mf was measured after overnight culture . Viable and motile worms , which were secreting motile Mf and were in the middle of the Mf release distribution , were selected for RNAi treatment . The GenBank accession numbers of the targeted B . malayi cathepsin-like cysteine protease ( CP ) genes and the gene-specific primers used to amplify the target genes are listed in Table 1 . Fragments corresponding to cDNA regions of the B . malayi cathepsin-like genes; Bm-cpl-1 , Bm-cpl-5 , and Bm-cpz , were amplified by PCR using gene-specific primers designed against the published sequences [39] ( Table 1 ) . A smaller fragment corresponding to pro-region of the Bm-cpl genes ( Bm-cpl Pro ) , designed to knock-down the three B . malayi cpl genes; Bm-cpl-1 , Bm-cpl-4 and Bm-cpl-5 ( group Ia of the filarial cathepsin L-like cysteine protease gene family [39] ) , which have identical pro-region sequences , was amplified using gene-specific primers ( Table 1 ) . A fragment corresponding to an intronic sequence within intron 2 of the O . volvulus cpz gene sequence Ov-cpz-int2 ( position 621-1132 , GenBank accession no . AY591516 ) , was amplified as a negative control as described previously [33] . The sequence was checked for putative microRNAs ( miRNA ) using the miRBase database ( http://microrna . sanger . ac . uk/sequences/index . shtml ) and by folding it using Mfold ( http://frontend . bioinfo . rpi . edu/applications/mfold/cgi-bin/rna-form1 . cgi ) . These analyses were used to determine if Ov-cpz-int2 contains any strong hairpins with a delta G≤−25 kcal/mol which might indicate a unique miRNA which could potentially initiate miRNA silencing at the translational level [42] , [43] . Both methods have shown that the Ov-cpz-int2 sequence has no miRNA sequences ( C . Poole&L . McReynolds , New England Biolabs , personal communication ) that could have induced silencing of the target gene using a different pathway to dsRNA-induced silencing . E . coli β-lactamase ( Ec-bla ) dsRNA ( GenBank accession no . NC_010862 , position 24517-24770 ) [31] was also used as a negative control . The PCR fragments were then sub-cloned into pCR4-TOPO vector ( Invitrogen , Carlsbad , CA ) or pBluescript SK vector ( Stratagene , La Jolla , CA ) according to the manufacturer's instructions . Clones were confirmed by sequencing both strands . Plasmids were then purified using the Rapid Plasmid Miniprep System ( Marligen Biosciences Inc . , Ijamsville , MD ) and were used as templates for RNA in vitro transcription . For RNA transcription , cDNA template from the purified plasmid was amplified with M13 forward and M13 reverse primers ( Invitrogen ) and then used with either T3 and T7 RNA polymerase for the single-stranded sense or antisense RNA synthesis using the MEGAscript high yield transcription kit ( Ambion Inc . , Austin , TX ) . Large quantities of dsRNA were prepared as previously described [44] . Integrity of dsRNA was checked by standard agarose gel electrophoresis . The final size ( bp ) of dsRNA is approximately 120 bp larger than the amplified gene-specific fragment due to vector linker sequence between the T3 and T7 priming sites and inserted gene fragment . Short-interfering RNA ( siRNA ) corresponding to the specific target was produced by digesting transcribed dsRNA with RNase III ( Ambion ) according to the manufacturer's instructions . Undigested and partially RNaseIII digested material was removed using a siRNA purification unit ( Ambion ) . The siRNA was quantified by measuring absorbance and the concentration calculated according to the manufacturer's instructions . The penetration of both dsRNA and siRNA into adult female B . malayi was followed using fluorescently-labeled RNA . dsRNA and siRNA were fluorescently labeled with cyber red , Cy3 ( Cy3-RNA ) , using the Silencer siRNA labeling kit ( Ambion ) according to the manufacturer's instructions . B . malayi adult female worms , in groups of 4 , were soaked in normal CM containing Cy3-RNA ( 0 . 01 mg/ml ) for 24–72 h . Fluorescence was visualized using a Zeiss Axiovert fluorescence microscope using the rhodamine filter set , using emission 590 nm . RNAi treatment of B . malayi adult females was carried out by soaking with dsRNA or siRNA . dsRNA preparations were dialyzed for 3 h in D-tube Maxi dialysis tubes , 12–14 kDa cut-off ( Novagen , EMD Biosciences , Inc . , Madison , WI ) , at 37°C under 5% CO2 , against normal culture medium before incubation with B . malayi adult female worms . Following 24 h culture in normal culture medium at 37°C in a 5% CO2 incubator , viable and motile worms were transferred into 48-well plates in groups of 3–4 adult females per well . Each group was cultured for upto 3 d in 1 ml of normal CM containing 1 . 5–2 mg/ml dsRNA or 5 µM ( 45 µg/ml ) siRNA , with a daily change of media containing dsRNA . Following RNAi treatment , worms were transferred to normal CM containing 10% heat-inactivated fetal calf serum and cultured for an additional 2 d , with daily change of media . Every 24 h throughout the experiment adult female phenotypes were monitored microscopically , and the number and the phenotype of the secreted progeny , microfilariae ( Mf ) , pre-microfilariae ( p-mf ) , embryos and eggs , were recorded . To compare progeny release after RNAi treatment , release was expressed as a reduction in release in comparison to the release in the dsRNA-free medium control group . At the end of the experiment ( 2 d after RNAi treatment ) worms were collected for either embrograms , where intrauterine progeny were recovered from individual female worms , or RNA extraction for quantitative real-time PCR ( qRT-PCR ) analysis . Worms were gently homogenized in 0 . 5 ml normal culture medium for 2–3 min , and the resulting suspension was examined microscopically to determine the relative proportions of progeny at different stages of development . RNAi treatment with dsRNA was repeated using dialysis tubes instead of 48-well plates , as previously demonstrated by Aboobaker et al ( 2003 ) [34] using a modified protocol . Worms were incubated in dsRNA for 4 d in dialysis tubes ( D-tube Maxi dialysis tubes , 12–14 kDa cut-off ) and dialyzed against normal culture medium , which allowed for a longer incubation time in dsRNA and no further handling of the worms during the incubation period . Bm-cpl-1 , Bm-cpl-5 , Bm-cpl Pro and Bm-cpz dsRNAs were used as the test RNAi treatments , with negative controls being either culture medium containing RNA storage buffer ( medium control ) , dsRNA corresponding to an O . volvulus intronic sequence Ov-cpz-int2 or E . coli β-lactamase ( Ec-bla ) ( negative control ) . Each experiment was repeated at least three times . Loss of specific transcripts following RNAi treatment were examined by real-time quantitative RT-PCR ( qRT-PCR ) . Primers used were designed to prevent re-amplification from dsRNA ( Table 1 ) and to amplify all three of the B . malayi filarial group Ia cathepsin L-like cysteine protease genes; Bm-cpl-1 , Bm-cpl-4 and Bm-cpl-5 ( Bm-cpl ) . Our initial attempts to design dsRNA-mediated silencing confirmation primers to distinguish the three Bm-cpl genes were unsuccessful; Bm-cpl-4 and Bm-cpl-5 are 94% identical , while Bm-cpl-1 is 83% identical to Bm-cpl-4 and Bm-cpl-5 . Moreover , because of the strong similarity , we assumed that dsRNA-mediated silencing with Bm-cpl-5 dsRNA will cross-target also Bm-cpl-4 and Bm-cpl-1 . However , it will not be able to cross-target the other five phylogenetically distinct cpl genes that belong to group Ic [39] . This was further established by comparing the predicted siRNA in these three transcripts to those of group Ic using the Protein Lounge siRNA Database ( www . proteinlounge . com ) ; none of which were similar to those predicted for Bm-cpl-1 , -4 and -5 . B . malayi β-tubulin ( Bm-tub-1; GenBank accession no . AY705382 ) was used as the endogenous control gene . At the end of the experiment ( 2 d after RNAi treatment ) , groups of four adult female worms were removed from culture , washed in PBS , and flash frozen in liquid N2 . Frozen worms were homogenized in Trizol reagent ( Invitrogen ) and RNA was extracted as previously described [40] . First strand cDNA was generated using the SuperScript III first strand cDNA synthesis kit ( Invitrogen ) and priming with oligo ( dT ) 20 . The specific cDNA fragments were then amplified by real-time PCR using QuantiTect SYBR Green PCR kit ( Qiagen Inc , Valencia , CA ) and the ABI Prism 7700 Sequence Detection System ( Applied Biosystems , Foster City , CA ) . The PCR conditions used were 50°C for 2 min , 95°C for 15 min , followed by 40 cycles of 94°C for 15 s , 52–58°C for 30 s , 72°C for 30 s . The relative amount of test amplicon in each experiment was determined by using the comparative CT method normalizing against the endogenous control gene ( Bm-tub-1 ) , as described in the ABI PRISM Sequence Detection System User Bulletin No2 ( Applied Biosystems ) . The value of the medium control group was set to 100% and the relative reduction of the RNAi treated groups was calculated and expressed as a percentage reduction in comparison to the control group . Comparison between the groups in RNA interference experiments were analyzed using the two-tailed non-parametric Mann-Whitney U-test . A P value of <0 . 05 was considered statistically significant .
To determine the possible function ( s ) of the B . malayi cathepsin-like cysteine protease genes during embryogenesis of B . malayi , RNAi was carried out to selectively interfere with Bm cathepsin-like cysteine protease gene expression; Bm-cpz or Bm-cpl . Microfilarial release was determined prior to RNAi treatment to provide a baseline reading and shows that grouped worms were evenly distributed in terms of Mf release ( Fig . 1A ) . Daily microfilarial release was determined from control and RNAi treated adult female worms which were cultured for 18 h in the presence of gene-specific dsRNA ( 2 mg/ml ) and then cultured for an additional 48 h in dsRNA-free culture medium . In comparison to the control worms and the worms treated with a negative control dsRNA ( Ov-cpz-int2 ) , adult worms treated with dsRNAs corresponding to the cathepsin-like cysteine protease genes , Bm-cpl-1 , Bm-cpl-5 and Bm-cpz , all showed a reduction in the release of Mf into the culture medium ( Fig . 1B & C ) . This reduction in the secretion of Mf corresponded to a 71 . 4% ( Bm-cpl-1 ) , 92 . 8% ( Bm-cpl-5 ) and 61 . 3% ( Bm-cpz ) , inhibition ( average ) of Mf release 48 h after RNAi treatment . The reduction in the release of Mf after treatment with Bm-cpl-5 dsRNA was significantly different from both the control worms ( P = 0 . 026; 48 h after RNAi ) and the negative control ( Ov-cpz-int2 ) treated worms ( P = 0 . 002 and 0 . 041; 24 h and 48 h after RNAi respectively ) ( Fig . 1 ) . This demonstrates a persistent effect of dsRNA-mediated silencing as worms had been cultured in dsRNA-free medium for a further 24–48 h following RNAi treatment . As we saw the most significant reductions in the release of Mf after treatment with dsRNA corresponding to Bm-cpl-5 ( Fig . 1B & C ) we decided to focus on optimization of the RNAi technique and analyzing the effect on embryogenesis using dsRNA corresponding to Bm-cpl-5 and to the identical pro-region of Bm-cpl-1 , Bm-cpl-4 and Bm-cpl-5; Bm-cpl Pro , which was designed to target all three Bm-cpl transcripts . In order to optimize and utilize the RNAi soaking technique in B . malayi to target the cathepsin-like cysteine protease genes it was important to demonstrate uptake of the in vitro transcribed dsRNA and siRNA by adult female B . malayi . FITC-labeled Bm-tub-1 dsRNA ( approx . 300 bp ) has previously been shown to be taken up successfully by adult female B . malayi after soaking for 18 h with 0 . 08 mg/ml dsRNA [34] . Cy3-labeled Bm-cpl RNAs , both dsRNA and siRNA , were taken up by adult worms in vitro after soaking for 24–72 h at 0 . 01 mg/ml ( Fig . 2 ) . Uptake of Cy3-labeled RNA was clearly seen after 24 h incubation with Cy3-labeled RNAs corresponding to a fragment of Bm-cpl-5 . The dsRNA Bm-cpl-5 fragment appeared in the mouth , esophagus and intestine after 24 h and in the cuticular and hypodermal regions after 48 h and 72 h ( Fig . 2A ) . Uptake of the Cy3-labeled RNAs appeared to be affected by the size of the dsRNA . While the dsRNA Bm-cpl-5 fragment ( approx . 800 bp ) was taken up by the adult worms and was seen in the hypodermal regions only after 48 h , the siRNA appeared in these regions earlier and was clearly already seen after 24 h ( Fig . 2B ) . In addition , a smaller dsRNA corresponding to the pro-region of Bm-cpl; Bm-cpl Pro ( approx . 400 bp ) gave a more intense and diffuse uptake staining pattern after 72 h when compared to the uptake of the larger Bm-cpl-5 fragment ( Fig . 2C ) . The smaller Bm-cpl Pro Cy3-labeled dsRNA and the Cy3-labeled siRNAs were also found along the length of the uterus . It is important to note that while the dsRNAs were added at the same concentration ( wt/vol ) , they have different molarities due to their different lengths , Bm-cpl-5; approx . 800 bp , 0 . 019 µM , Bm-cpl Pro; approx . 400 bp , 0 . 038 µM , siRNA; approx 13 . 5 bp , 1 . 1 µM . This uptake was consistent in all worms examined . We also examined whether the use of a lipid carrier ( lipofectin ) could improve the uptake of dsRNA . Lipofectin was not toxic to B . malayi adult female worms , however , lipofectin mixed with Bm-cpl-5 dsRNA did not improve the penetration of dsRNA ( data not shown ) . We next examined in more detail the effect of RNAi selectively carried out to interfere with Bm-cpl gene expression using both Bm-cpl-5 and Bm-cpl Pro . Microfilarial release was determined daily from worms cultured in the presence of gene-specific dsRNA ( 1 . 5 mg/ml , Bm-cpl-5; 2 . 82 µM , Bm-cpl Pro; 5 . 63 µM ) or their corresponding siRNA ( 5 µM ) for 3 d , and then cultured for a further 2 d in dsRNA-free culture medium . In comparison to the control worms , adult worms treated with dsRNA or siRNA corresponding to the Bm-cpl genes showed a persistent and significant reduction in the release of Mf into the culture medium after 1 , 3 and 5 days ( P<0 . 05 for all ) ( Fig . 3 ) . Reduction in Mf release after treatment with Bm-cpl-5 dsRNA was equivalent to that shown in Fig . 1 . Although treatment with the negative control dsRNA , Ov-cpz-int2 , did result in a reduction in the release of Mf this reduction was not significantly different from the medium control ( P>0 . 05 ) ( Fig . 3 ) . Reduction in the release of Mf was seen most rapidly after treatment with Bm-cpl-5 dsRNA , however , by day 5 both Bm-cpl-5 and Bm-cpl Pro dsRNA treatment had equivalent reductions in Mf release . Importantly , reductions in Mf release were also seen after treatment with siRNA corresponding to the Bm-cpl-5 and Bm-cpl Pro genes ( Fig . 3 ) . We also noted that in the groups treated with either Bm-cpl-5 or Bm-cpl Pro dsRNA the majority of Mf which were released from day 1 onwards were immotile and granulated ( approx . 90–95% ) suggesting that the Mf were dead ( data not shown ) . RNAi treatment with Bm-cpl-5 and Bm-cpl Pro genes had dramatic effects on intrauterine embryogram profiles ( Fig . 4 ) . Intrauterine progeny were examined 2 d after RNAi treatment and expressed as the relative proportions of progeny at different stages of development; eggs , developing embryos , pre-microfilariae ( pre-Mf ) and Mf . In comparison to the percentage of intrauterine pre-Mf in both the medium control ( 21 . 2% ) and the negative control , Ov-cpz-int2 ( 22 . 1% ) adult female worms treated with Bm-cpl-5 and Bm-cpl Pro dsRNA or siRNA showed significant reductions in pre-microfilariae ( pre-Mf ) ; Bm-cpl-5 dsRNA; 6 . 8% ( Control; P = 0 . 002 , Ov-cpz-int2; P = 0 . 004 ) , Bm-cpl-5 siRNA; 13 . 2% ( Control; P = 0 . 0037 , Ov-cpz-int2; ns ) , Bm-cpl Pro dsRNA; 9 . 0% ( Control; P = 0 . 0011 , Ov-cpz-int2; P = 0 . 028 ) , and Bm-cpl Pro siRNA; 11 . 6% ( Control; P = 0 . 0003 , Ov-cpz-int2; P = 0 . 016 ) . While the numbers of pre-Mf were reduced , the percentage of embryos within the uterine progeny were significantly increased in comparison to medium control ( 49 . 1% ) and the negative control , Ov-cpz-int2 ( 51 . 3% ) ; Bm-cpl-5 dsRNA; 64 . 2% ( Control; P = 0 . 002 , Ov-cpz-int2; P = 0 . 048 ) , Bm-cpl-5 siRNA; 64 . 5% ( Control; P = 0 . 0037 , Ov-cpz-int2; P = 0 . 042 ) , Bm-cpl Pro dsRNA; 64 . 0% ( Control; P = 0 . 0011 , Ov-cpz-int2; P = 0 . 028 ) , and Bm-cpl Pro siRNA; 60 . 2% ( Control; P = 0 . 0003 , Ov-cpz-int2; ns ) . No changes were observed in the proportions of Mf or eggs within the uterus . Phenotypic changes and structural abnormalities in developing embryos were also observed following RNAi treatment ( Fig . 5 ) . Treatment with either Bm-cpl-5 dsRNA ( Fig . 5C ) or Bm-cpl Pro dsRNA ( Fig . 5D ) resulted in malformed intrauterine embryos in comparison to both the medium control ( Fig . 5A ) and negative control ( Fig . 5B ) . Embryos from treated worms appeared to be not fully developed within the eggshell , leading to space between the embryo and eggshell . Effects on embryonic viability were also observed following RNAi treatment with Bm-cpl suggesting that the eggs and embryos released from Bm-cpl treated B . malayi were less viable ( 18–22% viable ) than those from controls as demonstrated by MTT viability staining [45] ( data not shown ) . RNAi treatment was repeated using dialysis tubes instead of the culture plate system . Worms were incubated in dsRNA for 4 d in dialysis tubes and dialyzed against normal culture medium , which allowed for a longer incubation time in dsRNA with the total amount of dsRNA required being much less . RNAi treatment targeting the B . malayi cathepsin-like cysteine protease group Ia genes using dialysis tubes resulted in similar phenotypic effects to those seen in the culture plate system ( data not shown ) . RNAi in dialysis tubes has the potential to facilitate higher-throughput RNAi screens for selected genes . RNAi treatment targeting Bm-cpl-5 and Bm-cpl Pro using both dsRNA and siRNA resulted in a specific reduction in Bm-cpl transcript level ( Fig . 6 ) . Analysis by qRT-PCR on RNA isolated from the RNAi treated worms , which had identical phenotypes to those shown in Figs . 1&3 , showed that the Bm-cpl transcript levels were reduced by 51 . 0% in Bm-cpl-5 dsRNA treated worms and by 48 . 9% in the siRNA treated worms , these reductions were significantly different from both the medium control ( dsRNA; P = 0 . 0006 , siRNA; P<0 . 0001 ) and the negative control ( dsRNA; P = 0 . 0033 , siRNA; P<0 . 0001 ) . The Bm-cpl transcript levels were also reduced in Bm-cpl Pro treated worms; 66 . 5% and 37 . 4% in dsRNA and siRNA treated worms respectively , with significant differences from both the medium control ( dsRNA; P = 0 . 0003 , siRNA; P = 0 . 002 ) and the negative control ( dsRNA; P = 0 . 0016 , siRNA; P = 0 . 0082 ) . The percent reduction was normalized using a tubulin ( Bm-tub-1 ) transcript . In comparison , the Bm-cpl-3 ( 89% identical to Bm-cpl-2 and -7 ) and Bm-cpl-6 transcripts that belong to the Ic subgroup of the B . malayi cysteine protease protein family were not reduced in the Bm-cpl Pro treated worms ( data not shown ) . These data demonstrate that the inhibition of Mf secretion in vitro , death of Mf , and changes in intrauterine progeny following dsRNA-mediated silencing is likely associated with reductions in the Bm-cpl gene specific transcript levels belonging to group Ia .
Cysteine proteases play important roles in both intracellular and extracellular processes which are important in both development and survival . Cathepsin-like enzymes have been identified as potential targets for drug or vaccine development in many parasites , including filarial nematodes [33] , [38] , due to as their potential essential roles in feeding [46] , [47][48] , molting [33] , [49] , embryogenesis [40] , [44] and immune functions [50] ( reviewed in [51] ) . Here we describe the use of RNAi techniques in B . malayi adult females to directly assess the function ( s ) of the cysteine protease ( CP ) genes that belong to the CPZ and the group Ia of Bm-CPL protein families . In filarial nematodes the gene family of group Ia cathepsin L-like protease enzymes ( Bm-cpl: Bm-cpl-1 , Bm-cpl-4 and Bm-cpl-5 ) has been shown to be associated with larval molting and remodeling of the cuticle and eggshell [39] . Antibodies raised to Ov-CPL-1 ( 64–65% identity with Bm-CPL-1 , -4 and -5 ) and Bp-CPL-4 ( 77 , 91 and 94% identity with Bm-CPL-1 , -5 and -4 , respectively ) recognized the reproductive system in B . malayi similarly [39] ( S . Lustigman , unpublished data ) , and this localization was similar to that of Ce-CPL-1 [40] . The putative functions of the group Ic cathepsin L-like protease enzymes ( Bm-cpl-2 , Bm-cpl-3 , Bm-cpl-6 , Bm-cpl-7 and Bm-cpl-8 ) are still unknown . In C . elegans , Ce-cpl-1 was shown to be essential for embryogenesis [40] , while Ce-cpz-1 has a function during embryogenesis but is not essential and also has an important role during molting [44] . The C . elegans Ce-cpl-1 knock-out mutant has an embryonic lethal phenotype [40] . RNAi targeting of Ce-cpl-1 produced an early embryonic lethal phenotype with 95–100% of F1 embryos arresting with only 100–200 cells following dsRNA injection and ∼92% of F1 embryos arresting after soaking of L4 in dsRNA [40] . Interestingly , the H . contortus cpl-1 gene can rescue the C . elegans cpl-1 RNAi effect , suggesting that the parasite CPL is an orthologue of C . elegans CPL-1 and that CPL-1 is functionally conserved in parasitic nematode species [52] . This is further demonstrated in O . volvulus , where Ov-cpl-1 can rescue the C . elegans Ce-cpl-1 mutant , using a rescue construct containing the Ce-cpl-1 promoter , the full length Ov-cpl-1 cDNA from the first ATG to the stop codon and the Ce-cpl-1 3′UTR ( S . Hashmi , unpublished data ) . However , despite high sequence similarity the predicted essential function of a gene is not always essentially conserved in both filarial nematodes and C . elegans , as has been shown for the different roles of CPL and CPZ in the molting of O . volvulus and C . elegans [33] . dsRNA-mediated silencing of the Bm-cpl group Ia genes in B . malayi adult females leads to phenotypic changes in embryos , where embryos from Bm-cpl treated worms were not fully developed within the eggshell , leading to space between the embryo and eggshell . Embryonic viability following RNAi treatment with Bm-cpl was also affected . The role ( s ) of cysteine proteases during embryo development have been shown to include an essential role in yolk processing . Cysteine proteases are known to be involved in yolk degradation during invertebrate embryonic development [53] , [54] , [55] and Ce-CPL-1 has been shown to play an essential role in yolk protein processing during embryonic development , where a loss of CPL-1 activity in the Ce-cpl-1 mutant leads to formation of enlarged cytoplasmic yolk vesicles and embryonic lethality [41] . Whether the Bm-cpl cysteine proteases are also involved with yolk protein processing would be interesting to investigate . Immunolocalization of the native Bm-CPLs indirectly suggests that they do localize to yolk vesicles within the developing embryonic stages [39] but further co-localization studies are required to confirm these observations . Also of relevance would be to investigate the co-localization of the native Bm-CPLs with their endogenous cysteine protease inhibitors , such as the cystatins [56] . The activities of cathepsin-like cysteine proteases are controlled by their specific endogenous protein inhibitors [57] , [58] . In C . elegans , co-localization of CPL-1 and CPZ-1 enzymes with both endogenous CP inhibitors and yolk proteins has been clearly demonstrated and data indicates that CPL-1 and CPZ-1 and their putative inhibitor , Ce-CPI-2a , play a role during oogenesis and fertilization in C . elegans [59] . The O . volvulus CPL-1 and CPZ are also localized in the same regions as the endogenous inhibitor , Ov-CPI-2 [40] , [49] , implying that Ov-CPI-2 may regulate both enzymes during O . volvulus development . In addition , both enzymes are essential for third to fourth stage larva molting as demonstrated by RNAi [33] . dsRNA-mediated silencing of the Bm-cpl genes suggests that some of the B . malayi CPs such as the Bm-CPL group Ia enzymes may function during embryonic development and show similar function ( s ) to C . elegans CPL-1 , their most phylogenetically related CP , therefore suggesting conserved essential function ( s ) between these filarial nematode and C . elegans CPs . As CP inhibitors targeting parasite CPs have been proposed as therapeutics in both protozoan [60] , [61] , [62] , [63] and metazoan [64] , [65] parasites , CPs involved in embryogenesis clearly provide a valuable putative target if it results in a block in embryogenesis which , in filarial nematodes , would , in essence , lead to sterility of the adult worm . Therefore , it would be invaluable to determine specific cysteine protease inhibitors which would target the endogenous Bm-CPL group Ia enzymes and moreover , could be used as therapeutic agents against the parasite . Methodologies , such as synthetic combinatorial library analysis [66] , specific chemical library screening [67] and small molecule affinity fingerprinting [68] are available to identify such specific inhibitors . In order to elucidate the function ( s ) of the cysteine proteases of B . malayi we have developed and improved the reverse genetics RNAi techniques which have been used previously in filarial nematodes . RNAi has been demonstrated successfully in parasitic helminths and could provide an invaluable tool in determining gene function and identification of drug and vaccine targets . Using RNAi to selectively target the potential drug targets; the cathepsin-like cysteine protease genes , in adult female B . malayi we have demonstrated that these genes can be affected; dsRNA-mediated silencing of the Bm-cpl genes led to significant reductions in the number of Mf released by adult female nematodes . While Bm-cpl-1 , Bm-cpl-5 and Bm-cpz all showed this persistent phenotype following dsRNA-mediated silencing the most significant reductions in the release of Mf were observed with dsRNA corresponding to Bm-cpl-5 therefore we focused on optimization of the RNAi technique using fragments corresponding to Bm-cpl-5 and the 100% conserved pro-region of Bm-cpl-1 , Bm-cpl-4 and Bm-cpl-5 , designed to target all three of the B . malayi filarial group Ia cathepsin L-like cysteine protease transcripts [39] , which we named Bm-cpl . Treatment of adult worms with dsRNA or siRNA corresponding to the Bm-cpl genes resulted in a persistent and significant reduction in the release of Mf , the majority of which were dead , and changes in intrauterine progeny . At the same time , these treatments also resulted in a specific reduction in the Bm-cpl transcript levels . While we can not directly correlate the observed phenotypic effects with the specific reduction in Bm-cpl transcript levels it is tempting to speculate that the effects on embryonic development and Mf production are due to the specific knock-down of the Bm-cpl genes using the dsRNA-silencing pathway . Interestingly , while RNAi treatment using both dsRNA and siRNA were effective in specific gene silencing , the uptake studies have suggested that smaller fragments of dsRNA and siRNAs are able to penetrate adult female B . malayi worms more efficiently than longer dsRNAs , pointing to the possibility that the cuticle barrier in the female worm is more yielding to the penetration of smaller fragments of dsRNA and siRNA . Therefore , the efficiency of RNAi may be dependent on the dsRNA fragment size . It is also important to note that due to RNA fragment size differences the molarity of each dsRNA tested was different ( dsRNA range; 400–800 bp , siRNA; ∼13 . 5 bp , at 1 . 5 mg/ml; dsRNA range; 2 . 82–5 . 63 µM , siRNA; 166 . 83 µM , at 2 . 0 mg/ml; dsRNA range; 3 . 75–7 . 51 µM , siRNA; 222 . 44 µM ) . However , despite these molar differences , RNAi using longer dsRNA was still effective in this in vitro screening system . dsRNA molarity has been shown to affect the efficiency of RNAi targeting of actin in L . sigmodontis ( Ls-act ) with 3 . 5 µM giving the most consistent reductions in transcript , while higher molar concentrations of dsRNA ( 17 . 5 and 35 µM ) resulted in increased levels of heat shock protein 60 ( Ls-hsp60 ) suggesting the worms were stressed [35] . Therefore dsRNA molarity might still be an important consideration to keep in mind when optimizing RNAi techniques . While successful siRNA application has been reported in T . colubriformis [69] and S . mansoni [70] , [71] it has not previously been documented for filarial nematodes . We have shown for the first time that RNAi treatment using siRNA can also be effective in filarial nematodes and show the same phenotypic changes and reduction in transcript levels as RNAi treatment with fragments corresponding to the same region of dsRNA . This provides an additional RNAi strategy for future projects aimed at assessing the function ( s ) of genes in filarial worms and the discovery of novel drug targets . This is important as many of the inconsistencies with RNAi in parasitic helminths could be due to differences in the RNAi pathways between these helminths and free-living C . elegans . Analyses of genome databases , including H . contortus [30] and B . malayi [72] , have shown that while putative orthologues of many genes required for RNAi are present , other genes which are essential for the recognition and systemic spread of dsRNA in C . elegans , such as rde-4 , sid and rsd genes , appear to be absent . The absence of these genes could explain the problems associated with RNAi in parasitic helminths , although it could be that these genes are not required or alternative pathways exist . However , although the genes for dsRNA processing appear to be missing , the machinery for processing siRNA appears to be present . Therefore , the use of siRNA vs . dsRNA may provide a more robust RNAi assay for some of the target transcripts . We have also tested an alternative method for delivering the dsRNA molecules to adult B . malayi , which are notoriously difficult to maintain in culture for extended periods [73] , by culturing worms in dialysis tubes in the presence of dsRNA . This technique provides a way of minimizing the handling of nematodes and has the potential to facilitate higher-throughput RNAi screens . Dialysis of the dsRNA prior to RNAi treatment was also important in reducing the non-specific off-target effects which can occur following RNAi [74] . In filarial parasites treatment with control dsRNA , resulted in a 24 . 7–49 . 8% reduction in molting of O . volvulus L3 larvae [32] , [33] , and resulted in reduced motility in B . malayi adult worms [34] . Off-target effects following RNAi treatment have also been demonstrated in S . mansoni , these including non-specific changes in mRNA levels , changes in cercariae distribution and reduction in sporocyst length following treatment with control dsRNA [75] , [76] . We noted some non-specific toxic effects of unrelated dsRNA; for example the RNAi with the negative control dsRNA we selected to use , Ov-cpz-int2 , resulted in 14 . 9% reduction in the release of Mf 48 h after RNAi . However , this was not accompanied by any indirect knock-down in target gene transcription . E . coli β-lactamase ( Ec-bla ) dsRNA ( GenBank accession no . NC_010862 , position 24517-24770 ) [43] was also used as a negative control and also showed some off target effects ( 1 . 8% , 10 . 5% and 26 . 0% reduction in Mf release on day 1 , 3 and 5 ) . Regardless , the effect with the gene specific RNAi was significant . These off-target effects could , in part , be due to an innate immune response mounted in the nematode in response to foreign RNA . Viral RNAs are recognized by the innate immune response through toll-like receptor ( TLR ) 3 for dsRNA and TLR7 or TLR8 for ssRNA [77] . In addition , RNA silencing serves as an innate anti-viral mechanism in plants , fungi and animals [78] , [79] , and has been shown to be involved in C . elegans [80] . RNAi silencing in nematodes may have evolved to protect them against viruses and it could be speculated as a reason to why nematodes appear to be free from viruses . Another possibility is that the off-target effects could be due to the generation of microRNAs ( miRNAs ) from the dsRNA control sequence used . Intronic sequences have been shown to potentially also contain functional miRNAs [81] , [82] . However , searching the intron sequence used in this study ( Ov-cpz-int2 ) shows that it contains no miRNA regions so this appears unlikely . In order to effectively assess the essential function ( s ) of filarial genes using RNAi it is important to keep in mind the unpredictability associated with RNAi and to take care to optimize the conditions according to the target gene of interest . Suppression of gene expression following RNAi treatment , using current methodologies , appears to be more effective for some genes rather than others [83] , [84] . This could be due to experimental parameters such as dsRNA preparations , length of dsRNA , and regions of dsRNA amplified , or parasite factors including abundance and location of the target gene transcript . As seen in C . elegans the life-cycle stage may also be important; RNAi is often not efficient in the treated worms ( P0 ) , and the phenotypic effects are only obvious in the progeny ( F1 ) , after the target mRNA that is produced in the F1 generation is also degraded [85] . Our studies , unfortunately , can only measure the phenotypic and genotypic outcomes in treated ( P0 ) adult female worms . When the steady-state transcript levels in the adult worms are high it can be difficult to observe significant reduction even though the phenotypic release of F1 progeny is affected . One potential solution will be , in the future , to also measure the transcript levels of the targeted mRNA in the F1 progeny within the uterus or secreted into culture . In order to be able to obtain information as similar as possible to C . elegans the culture conditions in parasitic nematodes need to be optimized in order to maintain parasites for longer periods of time and , ideally , to allow for their subsequent development into the next life-cycle stages . The effects of RNAi treatment may also need to be studied directly in vivo or after monitoring the development of the F1 following treatment of adult worms in vitro then passing the secreted Mf through the full life-cycle in vivo . It may also be possible to perform RNAi directly on early developmental stages such as eggs and embryos . Using uptake studies we can demonstrate that dsRNA is taken up into eggs and embryos isolated from the uterine contents of adult female B . malayi ( L . Ford&S . Lustigman , unpublished data ) . In order to maximally utilize RNAi in parasitic helminths as a tool for determining gene function and for high-throughput screening of potential drug and vaccine candidates , various problems need to be addressed and optimization of both the long-term culturing conditions and RNAi techniques is required . By modifying both ours and published RNAi techniques we have shown persistent dsRNA-mediated silencing and demonstrated improved RNAi techniques . In conclusion , RNAi assays have allowed us to examine the function ( s ) of the cathepsin-like cysteine protease filarial group Ia genes in B . malayi adult female worms and have demonstrated that they have function ( s ) in embryogenesis and development which are similar to their orthologous genes in C . elegans . Thus , validating the potential of the filarial cysteine proteases as promising drug targets for filarial chemotherapy .
|
Filarial nematodes are an important group of human pathogens , causing lymphatic filariasis and onchocerciasis , and infecting around 150 million people throughout the tropics with more than 1 . 5 billion at risk of infection . Control of filariasis currently relies on mass drug administration ( MDA ) programs using drugs which principally target the microfilarial life-cycle stage . These control programs are facing major challenges , including the absence of a drug with macrofilaricidal or permanent sterilizing activity , and the possibility of the development of drug-resistance against the drugs available . Cysteine proteases are essential enzymes which play important roles in a wide range of cellular processes , and the cathepsin-like cysteine proteases have been identified as potential targets for drug or vaccine development in many parasites . Here we have studied the function of several of the cathepsin-like enzymes in the filarial nematode , B . malayi , and demonstrate that these cysteine proteases are involved in the development of embryos , show similar functions to their counterparts in C . elegans , and therefore , provide an important target for future drug development targeted to eliminate filariasis .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"infectious",
"diseases/helminth",
"infections",
"molecular",
"biology",
"genetics",
"and",
"genomics/functional",
"genomics"
] |
2009
|
Functional Analysis of the Cathepsin-Like Cysteine Protease Genes in Adult Brugia malayi Using RNA Interference
|
Although neocortical connectivity is remarkably stereotyped , the abundance of some wiring motifs varies greatly between cortical areas . To examine if regional wiring differences represent functional adaptations , we have used optogenetic raster stimulation to map the laminar distribution of GABAergic interneurons providing inhibition to pyramidal cells in layer 2/3 ( L2/3 ) of adult mouse barrel cortex during sensory deprivation and recovery . Whisker trimming caused large , motif-specific changes in inhibitory synaptic connectivity: ascending inhibition from deep layers 4 and 5 was attenuated to 20%–45% of baseline , whereas inhibition from superficial layers remained stable ( L2/3 ) or increased moderately ( L1 ) . The principal mechanism of deprivation-induced plasticity was motif-specific changes in inhibitory-to-excitatory connection probabilities; the strengths of extant connections were left unaltered . Whisker regrowth restored the original balance of inhibition from deep and superficial layers . Targeted , reversible modifications of specific inhibitory wiring motifs thus contribute to the adaptive remodeling of cortical circuits .
Neocortex has a similar multilayered histology throughout [1] , [2] , and different cortical areas are able to adapt , depending on their inputs , to the normal function of other regions [3] . This versatility may reflect the existence of a “canonical” information-processing architecture , underpinned by stereotyped patterns of excitatory connectivity [2] , [4] . The organization of inhibitory neocortical circuits also obeys principles of some generality . A recent survey of inhibitory-to-excitatory wiring patterns in primary motor ( M1 ) , somatosensory ( S1 ) , and visual cortex ( V1 ) of the mouse uncovered 25 interlaminar connection motifs common to all three regions [5] . Whereas most of these motifs were found at comparable frequencies in all cortical areas , the abundance of four motifs varied widely: ascending inhibition from layer 5B ( L5B ) to L2/3 and L4 , as well as from L6 to L5B , was prominent in V1 and S1 , but not in M1; descending inhibition from L4 to L5A featured notably in S1 . These motifs may therefore represent adaptations of a common blueprint to region-specific information-processing demands . This interpretation raises several questions . Is the presence of a specific wiring motif linked to the particular type of input a cortical area receives ? In other words , does the motif change when the type of input changes ? If so , is plasticity limited to a critical developmental period , or does the capacity to adapt persist into adulthood ? And how motif-specific is the change ? Are variable wiring motifs inserted or removed on demand , akin to plug-in devices that add new functionalities , or are circuits reconfigured more broadly ? To answer these questions , we have analyzed and compared the laminar organization of inhibitory inputs to pyramidal neurons in L2/3 of adult mouse barrel cortex ( S1 ) under physiological conditions , during sensory deprivation ( whisker trimming ) , and after recovery ( whisker regrowth ) . In agreement with recent observations in visual cortex [6]–[9] , we found that sensory deprivation of adult barrel cortex induced changes in inhibitory circuits . Importantly , the nature of these changes was not only an overall reduction in cortical inhibition , as had been inferred from the decrease of inhibitory neuron spine and bouton numbers observed earlier . Instead , inhibitory connections from particular cortical layers underwent large , reversible , motif-specific , and sometimes antagonistic adjustments . Individual inhibitory network motifs are thus altered selectively and independently to adapt a cortical area to functional change .
Barrel cortex of adult mice was deprived of sensory input by trimming whisker rows A , B , D , and E every other day for 2–3 wk . Animals were aged 8–11 wk at the beginning of the manipulation and 10–14 wk at the time of analysis . To exclude potential confounds due to deprivation-induced changes in the optical excitability of ChR2-expressing interneurons , we compared the light responses of ChR2-positive cells in deprived barrel-related columns and nondeprived cortex . Our analysis concentrated on L5B , which exhibited the largest functional changes associated with deprivation ( see below ) . We distinguished between fast-spiking and the two principal types of non-fast-spiking interneurons in layers 1 and 5 , termed accommodating and nonaccommodating ( or regular-spiking ) cells ( Figure 2A ) , and examined six measures of light sensitivity: the probability of evoking at least one action potential per pulse during optical pulse trains , which were delivered at 0 . 2 or 5 Hz , at power levels of 0 . 5 or 1 . 8 mW per pulse ( Figure 2B , C ) ; the average number of spikes evoked per 1 . 8 mW pulse during a 0 . 2 Hz train ( Figure 2D ) ; and the decay of light responsiveness as a function of the distance of the stimulating light beam from the soma ( Figure 2E , F ) . Fast-spiking and non-fast-spiking interneurons behaved statistically indistinguishably in all six measures , which were all unperturbed by sensory deprivation . The passive membrane properties of the postsynaptic L2/3 pyramidal cells ( Figure S1 ) were also unchanged in intact and deprived cortex , and the reliability of optically evoked transmission from all layers always exceeded the threshold for detecting a connection by a comfortable margin of safety ( Figure 2G ) . Any experimentally induced changes in the structure of the inhibitory input maps we record therefore reflect changes in synaptic connectivity . In barrel-related columns representing trimmed whisker rows A , B , D , and E , sensory deprivation was linked to a conspicuous loss of ascending inhibitory inputs to L2/3 pyramidal cells from deeper layers ( Figure 3A and Figure S2 ) . The normalized inhibitory charge flow from layers 4 , 5A , and 5B decreased to 43% , 44% , and 19% , respectively , of control conditions ( Figure 3B ) . Overall , the relative and absolute contributions of different cortical layers to the inhibitory charge flow of L2/3 pyramidal cells changed in a manner consistent with targeted adjustments of selected connections ( Figure 3B , C ) . The most striking example of motif-specific plasticity was antagonistic changes occurring simultaneously to different inhibitory connections in the same barrel-related column: while inhibition from L5B weakened 5-fold in deprived columns , inhibition from L1 nearly doubled ( Figure 3B , C ) . After whiskers had been allowed to regrow for 3 mo , the inhibitory charge flow from all cortical layers returned to baseline values ( Figure 3 ) . Full whisker regrowth restored the original balance between L5B- and L1-derived inhibition , by strengthening the former and weakening the latter ( Figure 3B , C ) . The antagonistic relationship between L1- and L5-derived inhibition held even during a transient stage of overcompensation when whisker regrowth was partial; at this stage , the aggregate strengths of inhibitory inputs from the two layers overshot their targets in opposite directions ( Figure 3B , C ) . Sparing a single central whisker from deprivation can cause excitatory circuits to expand , so that signals from the spared whisker now activate surrounding deprived barrel-related columns [28] , [29] . We examined if inhibitory circuits in columns representing intact whiskers similarly expanded into or retracted from deprived cortical areas . If this were the case , the inhibitory input distributions of whisker-related column rows B and D , which in our deprivation protocol neighbor the spared row C , would be expected to become asymmetric . No evidence for this type of territorial reorganization was found: neither the horizontal reach of inhibitory connections into deprived versus nondeprived barrel-related columns ( Figure 4A ) , nor the number of inputs from these columns ( Figure 4B , C ) , differed . We also analyzed the horizontal inhibitory input distribution of L2/3 pyramidal neurons residing in the spared barrel-related columns of row C . Trimming the principal whiskers associated with adjacent barrel-related columns altered neither the horizontal spread ( Figure 4D ) nor the number of locations in deprived columns ( Figure 4E , F ) that gave rise to IPSCs in spared columns: the profile of horizontal inhibitory connections from deprived to spared columns was the same as that between deprived columns ( Figure 4D ) . Our regime of sensory deprivation thus selectively altered the vertical ( laminar ) but not the horizontal ( columnar ) organization of inhibitory circuits . Remarkably , deprivation-induced changes in vertical inhibitory connectivity also affected the spared barrel-related columns of row C . As in deprived columns , the number of home column inputs decreased significantly ( Figure 4E ) , but the detailed pattern of laminar reorganization differed subtly . Spared and deprived whisker columns suffered an equally sharp drop of inhibitory charge flow from the thalamorecipient layers 4 and 5A ( Figure 5A ) . However , some inhibition from L5B was preserved in spared columns ( Figure 5A ) , and the antagonistic increase of L1-derived inhibition was lacking ( Figure 5B , C ) . Several mechanisms could generate these adaptations , singly or in combination . Elaboration or retraction of inhibitory terminals could alter the number of pyramidal cells contacted by one interneuron ( a change in connection probability ) or the number of synapses between one interneuron and one pyramidal cell ( a change in connection strength ) . Differences in connection strength could also arise if synaptic release probability or quantal size were modulated . A formal , though remote , possibility is that the number of interneurons themselves might change during deprivation . In adult mice with intact or fully regrown whiskers , identically sized majorities of pyramidal cells in L2/3 were targeted by L5B interneurons ( 19/23 , or 82 . 6% , of cells in control conditions; 14/17 , or 82 . 4% , of cells after 3 mo of whisker regrowth ) . By contrast , in deprived barrel-related columns approximately one half ( 11/23 ) of L2/3 pyramidal cells lacked any detectable input from layer 5B ( Figure 6 ) . Where connections from L5B remained after deprivation , their numbers were severely , selectively , and reversibly depleted ( Figure 6 and Figure S2 ) . The average number of connected locations in L5B dipped from 8 . 0±9 . 9 in columns representing intact whiskers to 1 . 7±2 . 2 in deprived columns; input numbers returned to 8 . 5±9 . 5 and 5 . 9±6 . 6 , respectively , during and after whisker regrowth ( means ± 1 SD; p = 0 . 002; ANOVA ) . Most of the few surviving sources of inhibition attributed to L5B arose from stimulation sites that straddled the border to L5A ( Figure 3A and Figure S2 ) , raising the possibility that the depletion of inhibitory connections from L5B was virtually complete . In columns representing intact whiskers , in contrast , presynaptically connected interneurons populated the full depth of L5B ( Figure 3A ) . IPSCs could be elicited from 13 . 4%±16 . 0% of L5B stimulation spots in control cortex and from 15 . 3%±17 . 3% in previously deprived columns after whisker regrowth , but only from 3 . 3%±4 . 4% of all L5B locations in deafferented columns ( means ± 1 SD; p = 0 . 023; ANOVA ) . Inhibition from layers 4 and 5A underwent qualitatively similar but less extensive changes . Absolute input numbers , as well as the fractions of connected locations , fell during sensory deprivation but rebounded fully when sensory input was restored ( Figure 6 ) . Connections from L1 followed a trend opposite to that of connections from deep layers . The average number of inhibitory inputs from L1 rose from 7 . 2±3 . 3 in control barrel cortex to 9 . 2±3 . 1 in deprived whisker columns , and returned to 5 . 5±3 . 1 and 8 . 5±3 . 5 during and after whisker regrowth , respectively ( means ± 1 SD; p<0 . 001; ANOVA; Figure 6 ) . These profound and antagonistic changes to four translaminar wiring motifs occurred against a backdrop of stable inhibitory-to-excitatory connectivity in the home layer . The number of inhibitory inputs from L2/3 fluctuated only marginally between a minimum of 23 . 2±9 . 7 after 1 mo of whisker regrowth and a maximum of 27 . 1±8 . 8 after 3 mo of regrowth ( Figure 6 ) . In contrast to the antagonistic relationship between inhibitory inputs from superficial and deep cortical layers in deprived barrel-related columns ( Figure 6 and Figure 7 ) , translaminar inhibitory connections in the spared whisker-related row C of deprived cortex became uniformly sparse . In deep cortical layers , the number of connected locations dropped significantly: from 17 . 2±8 . 5 to 6 . 2±3 . 6 in L4 , from 5 . 0±4 . 5 to 1 . 8±2 . 8 in L5A , and from 5 . 1±6 . 5 to 2 . 6±5 . 0 ( means ± 1 SD; p<0 . 05; t test; Figure 7 ) . The percentages of stimulation spots from which IPSCs could be elicited decreased from 34 . 7% to 11 . 4% in L4 , from 23 . 9% to 8 . 2% in L5A , and from 10 . 2% to 4 . 8% in L5B ( p<0 . 05 , t test ) . Even in L1 the number of connected locations fell slightly ( Figure 7 ) , in keeping with the lack of a deprivation-induced surge in inhibitory charge flow from that layer ( Figure 5B , C ) . Our experimental manipulations altered neither the optical excitability of ChR2-expressing interneurons ( Figure 2B–G ) nor the reliability of optically evoked synaptic transmission ( Figure 2H ) . The rearranged inhibitory input maps of L2/3 pyramidal cells in deprived cortex must therefore reflect changes in the number or subclass distribution of presynaptic interneurons , or changes in connection probabilities between these neurons and their postsynaptic targets . Immunohistochemistry ruled out the first mechanism: neither the densities of ChR2-expressing interneurons in the plastic layers 1 , 4 , and 5 , nor the distributions of the major subpopulations of parvalbumin- and somatostatin-positive cells , changed ( Figure 8A , B ) . The measured variations in the number and locations of sites where IPSCs could be stimulated are thus indicative of changes in connection probabilities . To search for structural correlates of these functional changes , we analyzed the wiring motif undergoing the largest deprivation-induced change: ascending inhibition from L5B ( Figure 3 and Figure 6 ) . Forty-nine interneurons in L5B were filled with neurobiotin; of these , 11 cells showed high-contrast axonal staining in upper cortical layers ( Figure 8C , D ) . All of these 11 cells were non-fast-spiking interneurons of the accommodating ( n = 6 ) or regular-spiking ( nonaccommodating ) type ( n = 5 ) , consistent with the notion that non-fast-spiking Martinotti cells are the principal conduits of L5-to-L2/3 inhibition [30] , [31] . The neurobiotin-filled axons of L5B interneurons extending into L2/3 carried fewer varicosities per unit length in deprived cortex than did their counterparts in control conditions , as would be expected if presynaptic terminals were eliminated following sensory deprivation ( Figure 8D–F ) . There were no statistically significant differences between the volumes , surface areas , and maximal diameters of varicosities in deprived and control conditions; qualitatively , varicosities in deprived cortex even appeared somewhat larger than in intact cortex ( Figure 8G ) . The fractional loss of inhibitory varicosities after whisker trimming was , however , smaller than the fractional reduction in inhibitory charge flow from L5B ( compare Figure 3B , Figure 6 , and Figure 8F ) . This apparent mismatch may be accounted for in several ways . First , it is conceivable that varicosities remain visible after deprivation but the associated synapses have fallen silent . Second , connections with pyramidal cells may represent only a fraction of all synapses formed by L5B interneurons in L2/3 . If only synapses with pyramidal cells are eliminated after deprivation , an aggregate varicosity count will underestimate the magnitude of this change . Third , whole axonal branches might be retracted . Although axonal pruning is considered an unlikely mechanism of lesion-induced inhibitory plasticity in adult visual cortex [7] , it remains a formal possibility in barrel cortex . Interneurons are thought to innervate each of their postsynaptic targets via multiple boutons ( typically ∼15 ) [30] , [32] , [33] . Whisker deprivation might cause some of these boutons to be eliminated at random . The loss of a measurable connection would then simply result from the stochastic depletion of all synapses between two locations . If boutons were indeed silenced or pruned in this shotgun manner , surviving connections would be expected to suffer partial bouton losses and , therefore , be weaker than those in control conditions . Surprisingly , this was not the case: sensory deprivation left the strengths of extant connections from deep layers , measured as the mean integrated current per IPSC , virtually unchanged ( Figure 9 and Figure 10 ) . Importantly , the coefficients of variation of the individual IPSC amplitudes also remained unchanged ( Figure S3 ) . It is therefore implausible that different subsets of inhibitory synapses underwent large but opposite adjustments that canceled one another in the average . Of course , optical stimulus-locked IPSCs may represent compound events if multiple presynaptically connected interneurons are activated simultaneously . Because the mean integrated current depends on the number and the individual strengths of all contributing synapses , it cannot be equated with the strength of a monosynaptic connection . In several instances , however , the two variables of synapse number and average synapse strength could be disambiguated by considering changes in charge transfer in the context of simultaneously occurring changes in the number of connected locations . For example , sensory deprivation greatly reduced the number of connections from layers 4 and 5 but left the charge transfer per remaining connection unchanged . If it is reasonable to assume , in light of this general trend toward synapse elimination , that persisting connections will be made through a constant or smaller rather than a larger number of synapses , then the average strength of these synapses must remain level or increase after deprivation . Our morphometric finding that L5B-derived axonal varicosities in deafferented whisker columns retained their pre-deprivation size or even expanded slightly ( Figure 8G ) reinforces this conclusion , as bouton size and synaptic strength tend to be tightly correlated [34] . An analogous argument applies to L1 of the spared whisker-related row C . The charge transfer per IPSC in these barrel-related columns increased significantly after whisker trimming , while the number of connected locations fell marginally ( Figure 7 and Figure 10 ) . This constellation of changes indicates that the average L1-derived synapse gained in strength . In L1 of deprived columns , in contrast , both the average charge transfer per IPSC and the number of connected locations rose insignificantly . The statistically significant increase in laminar charge flow from L1 after whisker trimming ( Figure 3 ) thus remains an unresolved consequence of combined increases in monosynaptic connection probability and monosynaptic connection strength . A history of deprivation had profound aftereffects on the strengths of inhibitory inputs originating in superficial layers 1 and 2/3 . The inhibitory charge flow per IPSC from these layers increased marginally upon sensory deprivation , echoing similar changes during the critical period [35] , but plummeted during subsequent whisker regrowth ( Figure 9 and Figure S4 ) . In contrast to the rapid and complete resurgence of inhibitory input numbers upon sensory restoration ( Figure 6 ) , the recovery of connection strengths was delayed and partial , even after 3 mo of whisker regrowth ( Figure 9 and Figure S4 ) . Although the causes and significance of this hysteretic effect are currently unknown , the phenomenon provides clear further evidence that sensory plasticity operates by tuning probabilities and strengths of synaptic connections independently of each other .
The adaptations documented here lay bare four remarkable features of experience-dependent plasticity of inhibitory connections . First , extensive changes take place in adult neocortex , long after the critical period for refining neuronal connections has closed [29] . Second , different wiring motifs are altered selectively and independently of one another . The most compelling illustration of this principle is the see-saw relationship between L1- and L5B-derived inhibition in deprived cortex: upon whisker trimming and regrowth , connections to a common postsynaptic target , the L2/3 pyramidal neuron , undergo simultaneous but opposite functional changes ( Figure 3 ) . Third , adjustments of inhibitory connection probabilities are fully reversible upon sensory restoration , even in cases where entire connections appear to have been lost during deprivation ( Figure 3 and Figure 6 ) . The removal of a measurable connection may thus not entail the physical retraction of axonal and/or dendritic branches , as is the case in critical period plasticity [36] , [37] , but rather the shutdown of transmission between synaptic partners that remain in close apposition [7] , [38] . Fourth , probability and strength of a connection are independent dimensions for functional adjustment ( Figure 6 , Figure 7 , Figure 9 , and Figure 10 ) . Although the experimental settings and analytical approaches differ , it is instructive to compare our present findings with those of previous reports of experience-dependent inhibitory plasticity in adult neocortex [6]–[9] , [39] . With one exception [39] , all of these studies have examined the dynamics of structural changes in visual cortex following retinal lesions or monocular deprivation . Chronic imaging of the dendritic and axonal arbors or fluorescently tagged synapses of L1 and L2/3 interneurons revealed a seemingly general , rapid , and lasting loss of dendritic spines or branch tips [6] , [7] , axon terminals [7] , and gephyrin-labeled postsynaptic puncta [8] , [9] after deprivation . These morphological changes were taken to indicate a broad adaptive downscaling of inhibition . Our direct measurements of activity-induced functional changes in inhibitory-to-excitatory connections across the entire depth of somatosensory cortex paint a more differentiated picture . Although we do find a general decrease in the number of cortical locations providing inhibitory inputs to L2/3 pyramidal neurons ( from 67 . 7±17 . 0 in control columns to 46 . 4±19 . 4 in whisker-deprived columns; means ± 1 SD; p<0 . 001 , t test; Figure 3B ) , the scale and specificity of cortical remodelling become apparent only when individual wiring motifs are disentangled and analyzed separately ( Figure 3 ) . From hindsight , hints of motif-specific plasticity are already evident in some earlier studies of excitatory [21] , [40] , [41] and even inhibitory cortical connections . For example , although the net elimination rate of boutons originating from L2/3 interneurons was found to increase after visual deprivation , inhibitory synapses onto L2/3 pyramidal cells—as opposed to those targeting dendrites of layer 5 pyramidal cells—appeared exempt from elimination [6] , [7] . These results are consistent with the stability of home layer-derived inhibition in our hands ( Figure 3 , Figure 6 , and Figure 9 ) . Another example is the distinct behavior of different populations of inhibitory axons in superficial layers of barrel cortex . After whisker plucking , some axons in deprived barrel-related columns sprout , while others in the same column suffer bouton losses [39] . In light of our observations it is likely that the sprouting axons derive from L1 interneurons , while axons suffering bouton losses originate from interneurons in deep layers that target the apical dendrites of L2/3 pyramidal cells ( Figure 3B ) . The adaptive changes displayed by different inhibitory circuit motifs offer some clues to the possible roles of these motifs in normal cortical function . Interneurons in L2/3 are thought to be driven by L2/3 excitatory neurons and the thalamocortical input layers 4 and 5A [27] , [35] . The amount of feed-forward inhibition these neurons impose on L2/3 pyramidal cells is thus expected to scale with the excitatory drive to the column: the smaller the intensity of sensory stimulation , the smaller the inhibitory counterforce generated by the local feed-forward circuit ( Figure 11 ) [35] . This autoregulatory feature may obviate the need for plasticity of L2/3-derived inhibitory connections and explain their relative stability in the face of our experimental perturbations ( Figure 3 and Figure 6 ) . The inhibitory L5B-to-L2/3 motif , in contrast , is all but deleted from deprived whisker columns ( Figure 3 and Figure 6 ) . This striking adaptation suggests that the inhibition imposed by L5B interneurons on L2/3 pyramidal cells would prove excessive if the connection were left in place unaltered . The loss of significant excitatory drive from nonprincipal whiskers may explain why the same adaptation occurs , albeit to a lesser extent , also in spared columns that border deprived cortical tissue ( Figure 5 and Figure 7 ) . Most L5B interneurons—including Martinotti cells , the putative mediators of ascending translaminar inhibition [30] , [31]—lack an autoregulatory mechanism that couples their activity directly to the intensity of sensory stimulation [42] . Instead , L5B interneurons are likely to integrate signals from L2/3 networks spanning more than one cortical column [43] and relay this information back to L2/3 in the form of recurrent inhibition [31] . This type of supralinear feedback inhibition [31] , [44] , triggered by spontaneous activity or sensory input to nearby columns representing intact whiskers , would be expected to extinguish any residual sensory signals reaching columns whose associated principal whiskers have been trimmed . Unplugging the inhibitory feedback connection may be necessary to enable these columns to process feeble thalamic input ( Figure 11 ) . Qualitatively similar considerations may apply to inhibition originating in layers 4 and 5A . L5A and septal regions of L4 form part of a cortico-thalamo-cortical loop involving the posterior medial nucleus ( POM ) [45]–[47] . Residual sensory input , relayed via POM to inhibitory neurons in layers 4 ( septum ) and 5A , might cause excessive inhibition in deprived barrel-related columns . The adaptations we observe may help to rebalance excitation and inhibition ( Figure 11 ) . Inhibitory interneurons in L1 , in contrast , are likely mediators of top-down control of cortical areas by hierarchically higher regions [48]–[50] . The increase in L1-derived inhibition after whisker trimming ( Figure 3 , Figure 6 , and Figure 9 ) could represent an adaptation to the emergence of aberrant spontaneous activity in columns that are no longer properly modulated by sensory signals . Enhanced top-down inhibition may be needed to suppress this activity or prevent it from spilling into other sensory or sensorimotor structures ( Figure 11 ) [51] . Even when surrounded by deprived cortex , columns with spared principal whiskers are exempt from this adaptation ( Figure 5 and Figure 7 ) , presumably because their main sensory afferent remains intact . Plasticity that is wiring motif– , cell type– [40] , [41] , or input-specific [21] illustrates the modularity of neural organization more directly than did earlier , statistical analyses of neuronal connectivity [5] , [52]–[54] . Scrutiny of the connection matrices of compact nervous systems [52] , cortical areas [53] , or local microcircuits [54] revealed that some patterns of interconnectivity are vastly overrepresented in comparison to their expected frequencies in random graphs . These so-called network motifs [52] have been proposed to represent functional modules , which are cascaded differently in circuits with different information-processing capabilities . Our demonstration that individual inhibitory motifs are selectively altered to accommodate new function lends credence to this proposal .
All procedures complied with the UK Animals ( Scientific Procedures ) Act 1986 . Experimental animals were knock-in mice homozygous for R26::CAG-lox-STOP-lox-ChR2-EGFP responder and Gad2::CreERT2 driver transgenes at both targeted loci [5] . Following tamoxifen induction of Cre recombinase activity , these animals express channelrhodopsin-2 ( ChR2; GenBank accession number AF461397 , [13] , [14] ) comprehensively in all main subclasses of GABAergic interneurons defined cytochemically [5] . Mice were maintained in top-open cages on a 12 h light/dark cycle and fed a custom diet based on Teklad 2018 , but with vitamin A levels elevated to 100 IU/g ( Harlan Laboratories ) . At 8–10 wk of age , whiskers in rows A , B , D , and E on the right side of the snout were trimmed every other day for 2–3 wk . Whisker trimming was performed under transient anesthesia induced by subcutaneous ( s . c . ) injection of ∼20 µl of a 3∶5 mixture of ketamine ( 100 mg/ml; Fort Dodge ) and medetomidin ( 1 mg/ml; Pfizer ) and reversed by s . c . injection of 15–20 µl atipamezole ( 5 mg/ml; Pfizer ) . Control animals underwent the same anesthetic regimen as did whisker-trimmed animals . Slices of acutely deprived somatosensory cortex were harvested no later than 36 h after the last whisker trimming session . To examine the effects of recovery from deprivation , whiskers were allowed to regrow for 4–5 wk ( 1-mo regrowth ) or 12–14 wk ( 3-mo regrowth ) . Starting at 6–8 d before slices were cut , mice were injected intraperitoneally ( i . p . ) on 5 consecutive days with 0 . 3–0 . 5 mg 4-OH-tamoxifen ( Sigma-Aldrich ) , which was dissolved in sterile sunflower oil at 5 mg/ml . Experiments were performed on mice 2–4 d after the last 4-OH-tamoxifen injection . Animals were anesthetized by i . p . injection of 150 µl of a 3∶5 mixture of ketamine ( 100 mg/ml; Fort Dodge ) and medetomidin ( 1 mg/ml; Pfizer ) and perfused cardially with ice-cold solution containing ( in mM ) : 2 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 10 glucose , 240 sucrose , 0 . 5 CaCl2 , 7 MgCl2 , pH 7 . 4 , 320 mOsm . The brain was recovered into perfusion solution , and 310-µm slices of the left primary somatosensory cortex ( S1 ) were cut on a Leica VT1200S vibratome . The cutting plane was oriented across whisker barrel rows , maintaining angles of 45° with the midline and nearly 90° with the cortical surface [21] , [55] . Slices were incubated in the dark for 1 h at 34°C and subsequently maintained , shielded from light , at 25°C in modified artificial cerebrospinal fluid ( aCSF ) containing ( in mM ) : 125 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 25 glucose , 1 . 25 CaCl2 , 2 MgCl2 , pH 7 . 4 , 315 mOsm . Recordings were performed at room temperature in aCSF containing ( in mM ) : 125 NaCl , 3 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 25 glucose , 1 . 25 CaCl2 , 1 MgCl2 , pH 7 . 4 , 310 mOsm . All extracellular solutions were bubbled with 95% O2/5% CO2 . Whole-cell recordings were obtained from cells in barrel-related columns corresponding to whisker rows A through E . Patch pipettes had tip resistances of 4–6 MΩ . The internal solution for voltage-clamp recordings from pyramidal cells contained ( in mM ) : 110 CsOH , 110 gluconic acid , 0 . 2 EGTA , 30 Hepes , 2 MgATP , 0 . 3 Na2GTP , 4 NaCl , 5 QX-314-Br , 0 . 2% neurobiotin , pH 7 . 25 , 274 mOsm . The internal solution for current-clamp recordings from interneurons contained ( in mM ) : 120 K-gluconate , 10 KCl , 10 Hepes , 4 MgATP , 0 . 3 Na2GTP , 10 phosphocreatine , 0 . 2% neurobiotin . Signals were amplified and low-pass-filtered at 2 kHz by a Multiclamp 700a amplifier ( Molecular Devices ) and digitized at 5–10 kHz ( Digidata 1440 , Molecular Devices ) . Three criteria were used to distinguish fast-spiking ( fs ) from non-fast-spiking ( non-fs ) interneurons . Fs neurons ( i ) attained firing rates >90 Hz during a 1 , 000-ms depolarizing current step , ( ii ) exhibited a ratio of >0 . 7 of the average interspike interval ( ISI ) at the beginning and end of the depolarizing current step ( averages of 3 ISIs each ) , and ( iii ) displayed a spike width of ≤1 ms at half-maximal amplitude . Cells that met all three criteria were classified as fs and cells that failed all three criteria as non-fs . Optical stimulation experiments were performed on a Zeiss Axioskop 2FS microscope . A 40× , 0 . 8 NA water immersion objective with DIC optics was used for electrode placement and a 10× , 0 . 3 NA water immersion objective , without DIC optics , for optical stimulation . The output of a continuous-wave solid-state laser with a maximum power of 325 mW at 473 nm ( LRS-473-AH-300-10 , Laserglow ) was digitally switched and intensity-modulated by an acousto-optic deflector ( IntraAction model ASN-802832 with ME-802 driver ) , positioned by a pair of galvanometric mirrors ( GSI Lumonics VM500 with MiniSAX servo controllers ) , and merged with the epi-illumination path of the microscope via custom-built optics . Light pulses carried 0 . 5–1 . 8 mW of optical power at the exit pupil of the objective . To generate maps of inhibitory inputs , a virtual instrument written in LabVIEW 8 . 5 delivered focused stimulation light pulses ( spot size 3–5 µm , 20 ms duration ) at intervals of 680 ms to 60-µm grids encompassing 14×20 locations in pseudorandomized order . Animals ( n = 2 in each condition ) were perfused with phosphate-buffered saline ( PBS , pH 7 . 4 ) containing 4% ( w/v ) paraformaldehyde ( PFA ) and 0 . 2% ( v/v ) picric acid under general ketamine-medetomidine anesthesia . The brain was removed , incubated for 24 h in perfusion solution , and infiltrated with 30% ( w/v ) sucrose in PBS for at least 24 h . Coronal sections of 50 µm were cut on a Leica SM 2000R sliding microtome . The sections were rinsed three times in Tris-buffered saline ( TBS , Sigma ) , three times in TBS containing 3% ( w/v ) Triton X-100 ( TBS-T ) , and once for 1 h in TBS-T containing 20% ( v/v ) horse serum ( Vector Labs ) and then incubated for 48 h at 4°C in TBS-T containing 1% horse serum and combinations of the following primary antibodies: anti-GFP ( chicken , 1∶500 , AbCam ) , anti-parvalbumin ( mouse , 1∶2 , 000 , Swant ) , or anti-somatostatin ( rabbit , 1∶500 , Millipore ) . The sections were rinsed 4 times in TBS and stained in TBS-T containing 1% horse serum and Alexa488- and Alexa546-labeled secondary antibodies ( Invitrogen ) . After four rinses in TBS , the slices were mounted in VectaShield ( Vector Labs ) and imaged on a Leica TCS SP5 confocal microscope . To estimate the density of presynaptic varicosities along axon segments , the neurobiotin concentration in the internal solution for current-clamp recordings was raised to 1% . Patch configurations were converted to outside-out after neurobiotin infusion to allow the plasma membrane to reseal . Slices were incubated in modified aCSF for 1 h and then overnight in PBS containing 4% ( w/v ) PFA and 0 . 2% ( v/v ) picric acid . The slices were rinsed in TBS and stained in TBS-T containing 1% ( v/v ) horse serum , 4 µg/ml Alexa546-labeled streptavidin ( Invitrogen ) , and 0 . 0001% DAPI ( Sigma ) for 12–24 h . After four rinses in TBS , the slices were mounted in VectaShield ( Vector Labs ) and imaged on a Leica TCS SP5 confocal microscope . Interneurons were identified from their spiking responses to step-current pulses during the neurobiotin infusion as well as on the basis of morphological criteria after filling ( dendrites with beaded appearance; absence of mushroom spines; absence of a prominent apical dendrite ) . The axonal arbors of 11 out of 49 neurobiotin-filled neurons ( five cells in control conditions , and six cells in deprived conditions ) showed high-contrast axonal labeling in L2/3 that could be traced back to the filled soma in L5B; these arbors were chosen for morphometric analysis . Ten L2/3 axon segments in each condition , ranging in length from 125 to 2 , 063 µm , were reconstructed manually using the freeware Neuromantic [56] . Varicosities were identified as focal swellings that appeared larger and brighter than the neighboring stretches of axon and could , due to their size and brightness and the axial resolution of the microscope , also be seen in at least two adjacent confocal image planes [57] . Objects with a maximum diameter of 2 µm were counted as single varicosites if no constriction of the circumference was evident; the rare objects whose maximum diameter exceeded 2 µm were counted as two varicosities . Data were analyzed as described [5] , using Igor 6 ( Wavemetrics ) and SPSS 17 ( IBM ) . Briefly , maps of inhibitory inputs were constructed from electrophysiological signals recorded during 8–10 sweeps of the stimulation grid . IPSCs were identified by three criteria . First , the amplitude of the upward deflection in the averaged trace had to exceed 3 times the average standard deviation of current fluctuations in the absence of an optical stimulus ( rms noise ) . Our conclusions are robust under different choices of threshold ( 2 and 4 rms noise ) . In exceptional cases , factors of 2 . 5–5 rms noise were applied to isolated maps to compensate for unusually low or high baseline activity in the recordings . Second , IPSCs had to reach half-maximal amplitude within 5–70 ms after optical stimulus onset . Third , IPSCs had to occur in at least three of the 8–10 sweeps and exhibit a temporal jitter of less than ±10 ms . The presynaptic sources of IPSCs were allocated to individual cortical layers , which were identified by differences in shading and cell density [5] , [20]–[22] . The strength of each synaptic input was measured by integrating the recorded current over a 100-ms interval , beginning at 5 ms before the rising IPSC reached its half-maximal amplitude; this measure thus represents the charge transfer per IPSC ( Figure 9 and Figure 10 ) . The contribution of a layer to the total amount of inhibition received by a target cell ( Figure 3B and Figure 5B ) was quantified as a percentage , which was obtained by calculating the product of the number of IPSCs originating from that layer and their average charge transfer ( Figure 3C and Figure 5C ) and normalizing this value to the total inhibitory charge flow of the cell . Differences between multiple experimental conditions were analyzed by one-way ANOVA , which was followed by pairwise Bonferroni-corrected t test . Pairwise hypotheses were evaluated by t test . To visualize inhibitory input maps and their rearrangements following whisker trimming and regrowth , input strengths ( the normalized charge flowing during 100 ms after IPSC onset ) were coded in normalized gray scale ( Figure 3A and Figure 5A; see Figure S2 for nonnormalized maps ) . For comparisons across experimental conditions , input maps were scaled and aligned to barrel septa in the horizontal dimension and to the L1–L2/3 and L5A–L5B borders in the vertical dimension ( Figure 3A , Figure 5A , and Figure S2 ) .
|
Many natural and engineered networks contain recurring patterns of local connectivity . Although these so-called network motifs are thought to have functional significance , direct tests of the idea that network topology reflects function remain scarce . We have performed such a test in the area of mammalian neocortex that is devoted to the sensory representation of touch . To this end , we equipped inhibitory interneurons in the mouse with light-activated ion channels that allowed us to stimulate interneuron activity optically and record light-evoked inhibitory currents in their postsynaptic partners , thereby revealing maps of connectivity . We find that excitatory pyramidal cells in layer 2/3 of primary somatosensory cortex receive inhibition from GABAergic interneurons located in different cortical layers , with a characteristic balance of inhibitory connections from deep and superficial layers . Trimming the whiskers to remove sensory input in adult animals alters this balance—inhibitory connections from deep cortical layers are depleted , while inhibitory connections from superficial layers are augmented . These changes revert when the whiskers regrow , restoring the original balance between wiring motifs . This see-saw relationship between deep and superficial inhibition demonstrates that mature cortical circuits adapt to functional change by selectively altering specific network motifs .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"synapses",
"neurotransmitters",
"sensory",
"systems",
"biology",
"neuroscience",
"neurophysiology"
] |
2014
|
Experience-Dependent Rewiring of Specific Inhibitory Connections in Adult Neocortex
|
Human endogenous retroviruses ( HERVs ) are remnants of ancient infectious agents that have integrated into the human genome . Under normal circumstances , HERVs are functionally defective or controlled by host factors . In HIV-1-infected individuals , intracellular defense mechanisms are compromised . We hypothesized that HIV-1 infection would remove or alter controls on HERV activity . Expression of HERV could potentially stimulate a T cell response to HERV antigens , and in regions of HIV-1/HERV similarity , these T cells could be cross-reactive . We determined that the levels of HERV production in HIV-1-positive individuals exceed those of HIV-1-negative controls . To investigate the impact of HERV activity on specific immunity , we examined T cell responses to HERV peptides in 29 HIV-1-positive and 13 HIV-1-negative study participants . We report T cell responses to peptides derived from regions of HERV detected by ELISPOT analysis in the HIV-1-positive study participants . We show an inverse correlation between anti-HERV T cell responses and HIV-1 plasma viral load . In HIV-1-positive individuals , we demonstrate that HERV-specific T cells are capable of killing cells presenting their cognate peptide . These data indicate that HIV-1 infection leads to HERV expression and stimulation of a HERV-specific CD8+ T cell response . HERV-specific CD8+ T cells have characteristics consistent with an important role in the response to HIV-1 infection: a phenotype similar to that of T cells responding to an effectively controlled virus ( cytomegalovirus ) , an inverse correlation with HIV-1 plasma viral load , and the ability to lyse cells presenting their target peptide . These characteristics suggest that elicitation of anti-HERV-specific immune responses is a novel approach to immunotherapeutic vaccination . As endogenous retroviral sequences are fixed in the human genome , they provide a stable target , and HERV-specific T cells could recognize a cell infected by any HIV-1 viral variant . HERV-specific immunity is an important new avenue for investigation in HIV-1 pathogenesis and vaccine design .
Human endogenous retroviruses ( HERV ) are the remnants of ancient infectious agents that successfully entered the germ-line , established a truce with the host , and now make up 8 . 29% of the human genome [1 , 2] . Most HERVs are normally quiescent , either because the sequences are truncated and full-length transcription does not occur , or because intracellular defenses such as the APOBEC3 proteins keep activity in check [3] . However , two laboratories have recently reconstructed infectious “hybrids” from one family of endogenous retroviral sequences in the human genome , indicating that significant protein coding capacity and activity potential still exist for these endogenous retroviruses [4 , 5] . When HIV-1 infects a permissive cell , integration occurs within a genomic context of endogenous retroviruses . In HIV-1-infected cells , the virus initiates numerous changes in the cellular environment to enhance its own expression , which also affect the endogenous retroviruses in the genome . Intracellular defense mechanisms are compromised by proteins like Vif , which works against cellular APOBEC proteins , helping to establish a productive HIV infection [6 , 7] . Additionally , some HERVs have sequences that are recognized by HIV-1 Rev , providing nuclear export for HERV transcripts in HIV-1-infected cells [8] . RNA derived from HERVs is detectable in the plasma of HIV-1-infected individuals [9 , 10] . HIV-1-infected cells are normally recognized and killed by cytotoxic CD8+ T cells [11 , 12] . The proteasome collects and degrades proteins produced within infected cells , then human leukocyte antigen ( HLA ) class I molecules present peptides on the surface of infected cells [13] . This route of antigen presentation brings a number of antigens , both viral and self , to the surface of infected cells . Following the breaking of tolerance , the recognition of self antigens by CD8+ T cells is possible , especially in the setting of neoplastic transformation [14 , 15] . This suggests that HLA presentation of HERV antigens on the surface of HIV-1-infected cells could prime a CD8+ T cell response against HERV . In the present study , we detected evidence of HERV production in HIV-1-positive individuals that significantly exceeded that in HIV-1-negative controls , in agreement with previous studies [9 , 10] . Our aim was to measure the CD8+ T cell response against HERV . We describe and characterize CD8+ T cell responses against HERV antigens in 29 HIV-1-positive study participants and 13 HIV-1-negative and three hepatitis C positive ( HCV+ ) controls . Finally , we discuss the implications of these responses for the progression of HIV-1 infection .
Individuals were selected from participants in the UCSF OPTIONS cohort study [16] . The study was approved by the local institutional review board ( UCSF CHR ) and individuals gave written informed consent . Peripheral blood mononuclear cell ( PBMC ) samples from HIV-1 study participants were obtained from donated buffy coats . PBMCs from HCV+ individuals were obtained from the University of Toronto under their IRB . Studies were performed on cryopreserved PBMCs . Plasma samples ( 1 ml ) were centrifuged at 2 , 000g and filtered ( 0 . 2 μm ) prior to RNA collection to remove remaining cellular contaminants . High speed centrifugation ( 288 , 244g for 2 h at 4 °C in a Beckman SW41 rotor ) was used to pellet particles for RNA isolation with Trizol reagent ( Invitrogen ) . Samples were pre-treated with DNAse to eliminate genomic DNA contamination as a source of amplified HERV sequences . Reverse transcriptase ( RT ) -PCR was performed with cloned AMV RT ( Invitrogen ) on samples along with control amplifications without RT enzyme . As a calibration standard , cellular transcript expression of HERV and the housekeeping gene β-actin was measured in cDNA prepared from 2 . 5 × 106 HIV-negative donor PBMCs . Quantification standards were prepared by serial dilution of the cellular cDNA . Quantitative PCR with primers specific for the transcripts of interest was performed on all samples with the ABI Prism 7900HT Sequence Detection System ( Applied Biosystems ) using SYBR-Green detection . Primer sequence design and physical data were derived from Primer3 ( http://frodo . wi . mit . edu/cgi-bin/primer3/primer3_www . cgi ) . All amplifications were performed under the following conditions: 94 °C ( 3 min ) ; 36 cycles of 94 °C ( 10 s ) , 63 °C ( 20 s ) , 72 °C ( 25 s ) ; 95 °C ( 15 s ) , 60 °C ( 15 s ) , 95 °C ( 15 s ) . PCR was performed using the ABI PRISM 7900HT Sequence Detection System ( Applied Biosystems ) . Expression levels are presented as percentages relative to PBMC-derived standards and represent the means of triplicate reactions . Gel electrophoresis and melting point analysis of PCR products were used to confirm product purity and amplicon size . Selection of candidate HERV peptides was based on translated HERV protein sequence data compiled from the NCBI databases , Retrosearch [17] , and HERVd [18] . HIV-1 peptides were designed from the sequences of known HIV-1 epitopes listed in the Los Alamos National Laboratory HIV immunology database [19] . Antigenic regions of HERV insertions were assigned an HLA restriction with epitope prediction software [20 , 21] or based on the HLA restriction of corresponding regions of HIV-1 proteins . ELISPOT analysis was performed as previously described [22] . Equivalent antigen concentrations were used for HIV-1 and HERV peptides . Spot totals for duplicate wells were averaged , and all spot numbers were normalized to numbers of IFN-γ spot-forming units ( SFU ) per 1 × 106 PBMCs . Spot values from medium control wells were subtracted to determine responses to each peptide . PBMCs from an HIV-1-infected individual were stimulated with or without the peptides HIV VY10 , HERV-L IQ10 , or a cytomegalovirus ( CMV ) pool ( Becton Dickinson ) for 6 h with anti-CD28 and brefeldin A . The cells were stained with fluorophore-conjugated antibodies to CD3 , CD4 , CD8 , CCR7 , CD27 , CD28 , CD45RA , interferon-γ , IL-2 , and TNF-α to determine phenotype and function and an amine dye to discriminate between live and dead cells ( see also Text S1 and Figure S1 ) . Data were acquired with a LSR-II system ( Becton Dickinson ) . At least 100 , 000 events were collected and analyzed with FlowJo software ( TreeStar ) . The SPICE software was used to assist in the organization and presentation of multicolor flow data . Cryopreserved PBMCs from two study participants who responded to the HERV-L IQ10 peptide were stimulated for 7 d with peptides or pools of each antigen . Autologous , irradiated , peptide-pulsed feeder cells were used to restimulate for an additional 7 d . Cells were tested for their ability to lyse peptide-pulsed , autologous , Epstein-Barr virus ( EBV ) -transformed B cell lines by measuring the percentage of specific 51Cr release . The Mann-Whitney and Spearman rank tests were performed using GraphPad Prism version 4 . 00 for Windows ( GraphPad Software ) . p-Values less than 0 . 05 were considered significant for all tests .
We quantified HERV RNA in plasma from 16 untreated HIV-1-positive primary infection individuals and four HIV-1-negative volunteers . We detected significantly greater levels of HERV transcripts in the plasma of most HIV-1-positive individuals compared to controls ( Figure 1 , Mann-Whitney , p = 0 . 0160 ) . Although HIV-1 and endogenous retroviruses are phylogenetically distant [23] , we identified several regions of clustered and distributed amino acid identity in RT and Gag elements . Sequence comparison within a well-conserved protein like RT showed amino acid identities that were both distributed and concentrated in short , contiguous regions ( Figure 2A ) . Although alignments of the entire amino acid sequence were not possible with less well-conserved proteins , short , contiguous regions of amino acid sequence identity were still present ( Figure 2B ) . Because of the possibility of both a cross-reactive and independent T cell response to HERV in HIV-1 infection , we sought to measure the CD8+ T cell response to a number of HERV epitopes . We manufactured peptides to test for T cell reactivity , based upon the analysis of HERV sequence with epitope prediction programs [20 , 21] and based on similarity to known CD8+ T cell epitopes in HIV-1 [19] . A subset of the peptides shared more amino acids in common with HIV-1 ( ≥4 amino acids ) , and others were unique to HERV ( defined as ≤ 3 amino acids in common ) ( Figure 2B; Table S1 ) . We tested PBMCs from HIV-1-positive and -negative individuals for HERV- and HIV-1-specific T cell responses in 29 HIV-1-positive study participants from the OPTIONS cohort of primary HIV-1 infection at UCSF [24] and in 13 low-risk HIV-1-negative controls . Specific interferon-γ responses were detected to HERV peptides in HIV-1-infected individuals but not in HIV-1-negative controls ( Figure 3 , Mann-Whitney , p < 0 . 001 ) . As expected , PBMCs from HIV-1-positive individuals also recognized HIV-1-specific peptides . There was no statistical difference in the mean frequency of responding cells specific for HERV peptides with similarity to HIV-1 sequences and those unique to HERVs ( Mann-Whitney , p = 0 . 1025 ) . In addition to HIV-1-negative controls , we also tested three HCV+ , HIV-1 negative controls . T cell responses were not detected in response to HERV peptides in HCV+ controls ( Figure 3 ) . Individual T cell responses to each peptide tested for each study participant summarized in this figure are shown in detail in Figure S3 . In a cross-sectional analysis of the cohort of HIV-1-positive study participants , five individuals recognized the unique HERV peptide HERV-L IQ10 with variable magnitudes of response ( Figure 4A ) . In the responder with the highest T cell response magnitude ( OP562 ) , a peptide titration assay was performed ( Figure 4B ) . We measured HERV and HIV-1 T cell responses in three study participants in longitudinal series , including OP562 , who naturally contained HIV-1 viremia without antiretroviral therapy over the duration of our longitudinal analysis . The unique HERV peptide HERV-L IQ10 stimulated T cell responses in all three individuals , demonstrating persistent , independent HERV-specific T cell responses at high magnitude ( Figure 4C ) . In two of the individuals tested ( OP747 and OP841 ) , highly active antiretroviral therapy ( HAART ) was initiated , with subsequent declines in HIV-1 plasma viral load and the level of T cell responses to the HERV-L IQ10 peptide . We also compared responses to HIV-1 and HERV peptides in longitudinal series with a similar pair of peptides . As the peptides HIV Nef LG13 and HERV-H LI13 shared five amino acids , these responses could reflect a level of cross-reactivity . For OP562 , responses to both HIV-1 and HERV were not detectable by week 18 , but emerged by week 63 of HIV-1 infection ( Figure 4D , upper panel ) . Responses to the HIV Nef LG13 and HERV-H LI13 were detected in another study participant , OP747 ( Figure 4D , lower panel ) . To address potential cross-reactivity of HERV- and HIV-1-specific T cells in other study participants , we compared responses to an HLA-A2-restricted HIV-1 peptide HIV RT VL9 with responses to a HERV-L peptide , HERV-L II9 . The HERV-L II9 peptide is classified as a unique HERV peptide for this study because it shares only three amino acids with its closest corresponding peptide in HIV-1 , HIV RT VL9 ( see Table 1 ) . To test the effect of amino acid replacements in the HIV-1 peptide that increased the amino acid sequence similarity to the HERV peptide , we included in this analysis a number of intermediate sequence variant peptides , in which selected amino acids in the HIV-1 peptide were replaced with the corresponding amino acid from the HERV sequence ( Figure 4E ) . One individual ( OP478 ) responded to the HERV peptide , but not to the HIV-1 peptide or any of the intermediate sequence variants . Two individuals who responded to the HIV-1 peptide and the sequence variant peptides ( to varying degrees ) did not respond to the HERV peptide . To qualitatively compare HERV-specific CD8+ T cells with those specific for other viruses , we determined the phenotype and function of HIV-1- , HERV- , and CMV-specific T cells from OP562 and OP841 , who responded to the three viruses . For this analysis , we selected HIV-1 and HERV peptides ( HERV-L IQ10 and HIV RT VY10 ) with only two amino acids in common , minimizing potential cross-reactivity . Upon stimulation with respective HERV , HIV-1 , and CMV peptides , we ascertained the phenotypes of those cells having a cytokine production profile that were associated with degranulation ( Figure 5A; Text S1 ) . The HIV-1-specific T cells of both study participants were skewed towards CD45RA− , whereas CD8+ T cells responding to the HERV peptide had a greater percentage of the terminally differentiated cells ( CCR7−CD45RA+ ) ( Figure 5B , left panels ) . In one study participant ( OP562 ) , HERV- and CMV-specific populations shared a lower percentage of CD28−CD27+ CD8+ T cells compared to their HIV-1-specific counterparts ( Figure 5B , upper right panel ) . In contrast , in the other study participant ( OP841 ) , HERV- and CMV-specific populations shared a higher percentage of CD28−CD27+ CD8+ T cells ( Figure 5B , lower right panel ) . Overall , the phenotype of the HERV-specific CD8+ T cells more closely resembled the phenotype of CMV-specific than HIV-1-specific T cells . As these data suggest possible functionality of HERV specific T cells , we measured the relationship of these responses to HIV-1 viral load within the cohort . For the untreated time points available for 20 study participants , HERV-specific T cell responses were significantly inversely correlated with HIV-1 plasma viral load by Spearman non-parametric correlation analysis and linear regression ( Spearman , two-tailed , r = −0 . 49 , p = 0 . 03; linear regression r2 = 0 . 39 , p = 0 . 003; Figure 6 ) . Because the ability to control viral load by eliminating infected cells depends on killing , we measured the ability of CD8+ T cells specific for the unique HERV peptide HERV-L IQ10 to kill autologous B cells presenting their target peptide . We peptide stimulated PBMCs from two individuals ( OP562 and OP841 ) to enrich for responsive CD8+ T cells . After a 2-wk peptide stimulation , we used the 51Cr-release assay to measure the ability of the enriched CD8+ T cells to kill EBV-transformed B cell targets presenting cognate peptide . CD8+ T cells enriched by stimulation with HERV peptide were able to kill B cell targets presenting their cognate peptide but did not lyse targets loaded with a non-cognate or no peptide ( Figure 7 ) . Similar treatment of PBMCs from HIV-1-negative study participants did not produce HERV-specific effectors capable of killing peptide-pulsed targets ( Figure S2 ) .
Upon infection of a cell , viruses like HIV-1 alter the cellular environment to favor virus production . Viral proteins act as chaperones to modify controls on RNA trafficking into and out of the nucleus [25 , 26] . Additionally , cellular suppression mechanisms active against viral transcripts , but not acting at the reverse transcription step , are compromised [6 , 7 , 27] . In this altered cellular environment , HERV transcripts could enter translational pathways from which they are normally excluded . While only HERV-K elements contain full-length open reading frames for all proteins , many other HERV insertions in different families retain some form of protein- or peptide-coding capacity relevant for producing a CD8+ T cell response [17 , 28] . Even HERV elements with premature stop codons that prevent the production of full-length proteins could still produce peptide fragments capable of being presented on HLA molecules , exposing the immune system to HERV antigens . Our study has identified T cell responses to HERV epitopes in HIV-1-positive individuals . Five participants in the study cohort responded in varying degrees to the unique HERV peptide HERV-L IQ10 , and in the subset of those individuals tested longitudinally , those responses persisted over time . This T cell response was titratable over a range of peptide concentrations . Interestingly , the HERV-L IQ10 peptide originates from an open reading frame in the same region ( within 31 kbp ) of another HERV-L insertion containing a polymorphism associated with improved control of HIV-1 viral load set point [29 , 30] . While Fellay et al . point out that the effects of the HERV polymorphism on HIV-1 viral load set point are not genetically distinguishable from the effects of HLA-B*5701 in their study , they speculate about a possible antisense mechanism of HIV-1 viral control . The authors also point out that the HERV polymorphism results in an amino acid substitution in one of the predicted encoded proteins . Immunologic mechanisms of control via CD8+ T cell recognition of HERV epitopes are another plausible mechanism for the observed improvement of control over HIV-1 viral load set point . Both full-length , nearly intact viral insertions such as HERV-K , but also older , more heavily disrupted insertions , such as HERV-L , could stimulate important HERV-specific T cell responses . Since recognition of proteins by T cells occurs by presentation of short peptide fragments , distantly related viruses can still have epitopic regions in common with shared amino acids , even in less well-conserved proteins or when sub-full-length protein transcripts are produced by the cell . Short regions of similarity between HERV and HIV-1 peptide sequences could lead to a level of cross-reactivity for T cell receptors recognizing similar epitopes . Our data demonstrate parallel dynamics of T cell responses for peptides representing regions of similarity between HIV-1 and different HERV . The studies described here established limits to the cross-reactivity by demonstrating T cell responses to different variants of HIV-1 peptide sequences that do not overlap with T cell responses to HERV epitope peptides , but the extent and broader implications of this potential cross reactivity merit further study . Having identified CD8+ T cell responses to HERV epitopes in HIV-1 infection , it was important to determine the relevancy of these cells in the cellular immune response to HIV-1 infection . One study participant ( OP562 ) was able to control HIV-1 viral load without HAART over the duration of our longitudinal study , and he had the highest observed magnitude of response to a unique HERV peptide . For the entire cohort in a cross-sectional component of the study , the level of T cell responses to HERV was inversely correlated to HIV-1 plasma viral load . Given these data , suggestive of a role for HERV-specific T cells in controlling HIV-1 infection , it was important to compare the phenotype of HERV-specific T cells to those specific for other viruses . Previous studies of T cells in chronic viral infections of humans have identified differences of phenotype and function between virus-specific CD8+ T cells , which have led to the notion that HIV-1-specific CD8+ T cells are deficient in their maturation , and potentially in function [31 , 32] . We detected skewing towards CD45RA− for the HIV-1-specific T cells in both study participants analyzed , as has been previously observed for HIV-1 specific CD8+ T cells [32] . CD8+ T cells responding to the HERV peptide had a greater percentage of CCR7−CD45RA+ cells , presumed to be terminally differentiated cells that are associated with improved viral control in HIV-1-specific T cells [33] . Like the HERV-specific CD8+ T cells , CMV-specific CD8+ T cells in our data and in previous studies did not have a phenotypic profile skewed towards CD45RA− [33] . Additionally , HERV-specific CD8+ T cells had a lower percentage of CD28−CD27+ CD8+ T cells compared to their HIV-1-specific counterparts in one study participant , as reported in previous studies [31] . In the other study participant , both HERV- and CMV-specific CD8+ T cells shared a higher percentage of CD28−CD27+ CD8+ T cells compared to their HIV-1-specific counterparts . CD28 expression on the cell surface modulates responsiveness to co-stimulation , and can change according to the activation level of a T cell [34] . Higher levels of CD28 on the T cell surface increase the sensitivity of that T cell to B7 co-stimulation by antigen presenting cells , but engagement of the CD28 receptor leads to its down-regulation in a negative feedback loop [35] . Overall , our data are consistent with previous phenotypic and functional observations for HIV-1-specific CD8+ T cells , and indicate that HERV-specific CD8+ T cells more closely resemble CD8+ T cells induced in controlled chronic viral infections such as CMV . Controlling HIV-1 viral load also requires killing of infected target cells . HERV antigen production by HIV-1-infected cells would make them a potential target for HERV-specific CD8+ T cells . However , as HERV-specific CD8+ T cells are specific for self-antigens , CD8+ cells responding to HERV peptides may be impaired in their ability to kill targets . Our data show that HERV-specific CD8+ T cells specifically kill B cell targets presenting HERV peptides . Cells capable of killing B cell targets presenting HERV peptides were not detectable in the PBMCs of HIV-1-negative study participants . The T cell responses against HERV peptides in HIV-1-positive individuals appear to be driven by HIV-1 . With effective HIV-1 suppression by HAART , responses to unique HERV peptides and those similar to HIV-1 both undergo a decline . However , chronic viral infections can lead to dysregulation of the CD8+ T cell response [36] . To rule out the effects of chronic viral infection in generating HERV CD8+ T cell responses , we analyzed responses in HCV+ controls . The lack of HERV-specific T cell responses in HCV+ study participants indicates that HERV-specific CD8+ T cell responses are present in HIV-1 but not other chronic viral infections such as HCV . Our data have important implications for redefining the breadth of epitopes that make up what is considered the “HIV-1-specific” CD8+ T cell response . CD8+ T cell depletion in rhesus macaques has been used to demonstrate the importance of CD8+ T cells in viral control [37] . Of necessity , the depletion was performed regardless of specificity of the CD8+ T cells . Thus , CD8+ T cells with specificities for epitopes other than those derived from the virus itself could have also been depleted in these experiments and be important for viral control . Our results demonstrate HERV-specific T cell responses in individuals with primary HIV-1 infection . While these responses could play a role in HIV-1 pathogenesis by attacking any cell presenting HERV epitopes , it is interesting to consider the potential benefit of these responses to the control of HIV-1 . An inverse correlation between anti-HERV T cell responses and HIV-1 plasma viral load demonstrates a role for HERV CD8+ T cell responses in helping to contain HIV-1 viremia . Additionally , the phenotype of HERV-specific CD8+ T cells more closely resembles that of CD8+ T cells generated in the setting of a viral infection with good immunologic control . Our results suggest that HERV-specific CD8+ T cells , both those cross-reactive with HIV-1 epitopes and those specific for their HERV targets , should be included in the repertoire of cells capable of controlling HIV-1 replication . Because one of the greatest challenges for the immune system in effectively and durably controlling retroviral replication is responding to rapidly arising viral variants , it is interesting to speculate on the potential of HERV immunity to aid in broad spectrum and long-term immune control of HIV-1 . HERVs are genome-encoded elements with the same sequences present in every cell . An anti-HERV-specific immune response could target any HIV-1-infected cell irrespective of HIV-1 viral sequence . HERV could provide an effective surrogate target for the immune response to eliminate HIV-1-infected cells and bears investigation as candidates for inclusion in a new type of HIV-1 vaccine . In summary , we show HERV-specific immune responses in HIV-1 infection . Manipulation of these responses could play an important role in HIV-1 immunotherapeutics , or augment HIV-1 vaccine strategies .
HERV-L ORF accession numbers from Retrosearch [17] are 162563 , 162568 , 162604 , 162605 , and 162613 . The National Center for Biotechnology Information ( http://www . ncbi . nlm . nih . gov/ ) accession numbers for the proteins discussed in this paper are HERV-H ( gi|44887889 ) ; HERV-K ( gi|52001472 , gi|5802821 , gi|75766508 , gi|67782351 ) ; and HERV-W ( gi|52000737 ) .
|
The human genome contains a number of remnants or fossils of ancient viral infections referred to as human endogenous retroviruses ( HERV ) . Like fossils , these HERV are considered to be dead or inert in most cases . However , we demonstrate that T cells in the human immune system respond to HERV when a person is infected with the human immunodeficiency virus ( HIV ) . The T cells responding to HERV share characteristics with T cells that effectively control cytomegalovirus , a common chronic viral infection . T cells responding to HERV can also kill target cells carrying HERV protein . For some HIV-positive people , the strength of their response against HERV is related to having a lower HIV viral load . This study has important implications for new directions in HIV vaccine research . One of the key obstacles to creating an effective HIV vaccine is overcoming the ability of some of the viral variants produced when HIV replicates to evade the immune responses that the body mounts to control infections . If T cells that recognize HERV can stably target HIV-infected cells , they could be an important factor in controlling HIV infection .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion",
"Supporting",
"Information"
] |
[
"homo",
"(human)",
"immunology"
] |
2007
|
T Cell Responses to Human Endogenous Retroviruses in HIV-1 Infection
|
Inflammatory caspase-11/4/5 recognize cytosolic LPS from invading Gram-negative bacteria and induce pyroptosis and cytokine release , forming rapid innate antibacterial defenses . Since extracellular or vacuole-constrained bacteria are thought to rarely access the cytoplasm , how their LPS are exposed to the cytosolic sensors is a critical event for pathogen recognition . Hemolysin is a pore-forming bacterial toxin , which was generally accepted to rupture cell membrane , leading to cell lysis . Whether and how hemolysin participates in non-canonical inflammasome signaling remains undiscovered . Here , we show that hemolysin-overexpressed enterobacteria triggered significantly increased caspase-4 activation in human intestinal epithelial cell lines . Hemolysin promoted LPS cytosolic delivery from extracellular bacteria through dynamin-dependent endocytosis . Further , we revealed that hemolysin was largely associated with bacterial outer membrane vesicles ( OMVs ) and induced rupture of OMV-containing vacuoles , subsequently increasing LPS exposure to the cytosolic sensor . Accordingly , overexpression of hemolysin promoted caspase-11 dependent IL-18 secretion and gut inflammation in mice , which was associated with restricting bacterial colonization in vivo . Together , our work reveals a concept that hemolysin promotes noncanonical inflammasome activation via liberating OMVs for cytosolic LPS sensing , which offers insights into innate immune surveillance of dysregulated hemolysin via caspase-11/4 in intestinal antibacterial defenses .
The host innate immune system can sense invading bacteria by detecting pathogen-associated molecular patterns ( PAMPs ) [1] . Lipopolysaccharide ( LPS ) , a component of the outer cell membrane of Gram-negative bacteria , is one of the strongest immune activators [2] . Extracellular and endocytosed LPS is recognized by the transmembrane protein Toll-like receptor 4 ( TLR4 ) , leading to gene transcriptional regulation in response to infection [3] . Recent studies showed that host can detect LPS in the cytosol via a second LPS receptor , caspase-11 in mice and caspase-4/5 in humans [4–6] . Caspase-11/4/5 directly binds cytosolic LPS [6] , leading to its own activation , which thus cleaves gasdermin D to induce pyroptotic cell death and activate non-canonical activation of NLRP3 to release interleukin-1β ( IL-1β ) or IL-18 [7–8] . Therefore , compartmentalization of LPS receptors within cells allows host to respond differentially and sequentially to LPS at distinct subcellular locales , which function in concert to constitute host noncanonical inflammasome defenses . Caspase-11/4/5 , as cytosolic sensors , only recognize LPS that has entered the host cell cytoplasm; however , the mechanism by which LPS from invading bacteria gains access to the cytosolic sensors remains unclear . For intracellular bacteria , although some bacteria such as Burkholderia [9] are cytoplasm-residing and easily expose LPS to the cytosolic sensors , many other bacteria such as Salmonella typhimurium [10] or Legionella pneumophila [11] are predominantly constrained in pathogen-containing vacuoles ( PCVs ) , probably masking LPS from cytosolic innate sensing . In addition to vacuolar bacteria , many extracellular bacteria , including Escherichia coli [12] , Vibrio cholerae [13] , Citrobacter rodentium [14] , and Haemophilus influenzae [15] are thought to rarely access the cytoplasm , but induce caspase-11 dependent pyroptosis and cytokine release in cells . Thus , LPS entry into cell cytoplasm is a critical event for recognition of vacuolar or extracellular bacteria by non-canonical inflammasome . Accumulating data suggest vacuolar bacteria may shed their LPS from endosome into the cytosol [16] . Lipopolysaccharide-binding protein ( LBP ) is also implicated in facilitating intracellular LPS delivery [17] . Alternatively , mouse guanylate-binding protein 2 ( mGBP2 ) induces lysis of PCVs and promotes LPS leakage into the cytoplasm [14 , 18–19] . Recently , Vijay A . K . Rathinam and his colleagues explored how non-canonical inflammasomes detect extracellular Gram-negative bacteria . Briefly , the outer membrane vesicles ( OMVs ) of extracellular bacteria enter cells by dynamin-dependent endocytosis , enabling LPS to access the cytosol by escaping from early endosomes [20] . However , how OMVs gain access from early endosomes to the cytosol remains unknown . Hemolysin belongs to the pore-forming protein family , rupturing the cell membrane and leading to cell lysis at high doses [21] . Recently , hemolysin was found to participate in modulating cell death pathways at sublytic concentrations [21–23] , representing more sophisticated toxin activity in contrast to outright pore-forming function . Uropathogenic Escherichia coli ( UPEC ) isolate CP9 activates caspase-3/7 and stimulates rapid cell apoptotic death in vitro; this phenotype was lost in a ΔhlyA mutant [24] , indicating the involvement of hemolysin in the cell apoptosis pathway . Recently , increasing evidence suggests that hemolysin promotes activation of inflammasome signals during infection . For example , entero-hemolysin of enterohemorrhagic E . coli ( EHEC ) O157:H7 triggered mature IL-1β secretion in human macrophages [25] , and α-hemolysin of UPEC CFT073 mediated NLRP3-dependent IL-1β secretion in mouse macrophages [26] . Strikingly , overexpression of hemolysin in UPEC UT189 activated significantly increased caspase-4 dependent cell death and IL-1α release than the controls [27] . This is the first evidence demonstrating the relevance of hemolysin in caspase-4 activation , indicating that hemolysin might contribute to non-canonical inflammasome activation . In this study , we first demonstrated that hemolysin in various enterobacteria significantly promoted caspase-4 dependent pyroptosis and IL-18 secretion in human intestinal epithelial cell lines . Further , we provided insights into the mechanism of hemolysin-mediated increase in the sensitivity of non-canonical inflammasome to invading bacteria . We showed that hemolysin internalizes into cells via binding to OMVs and promotes rupture of OMV-containing vesicles , thereby releasing OMV-derived LPS into the cytoplasm and eventually triggering significant activation of non-canonical inflammasomes in cells . Oral infection of mice showed that abnormal expression of hemolysin in vivo alerts the immune system and induces caspase-11-dependent enterocyte pyroptosis and IL-18 secretion , which significantly constrains bacterial infection in the gut . Collectively , our results reveal that overproduced hemolysin enables OMV-mediated LPS cytosolic delivery for caspase-11/4 sensing , which alarms intestinal innate immune surveillance in vivo , providing insights into the manipulation of non-canonical inflammasome signals by invading bacteria .
To screen for bacterial factors involved in regulating non-canonical inflammasome activation , a gene-defined mutant library of Edwardsiella tarda ( E . tarda ) , an enteric pathogen infecting hosts from fish to human [28–29] , was used to identify mutants that induced significantly increased pyroprosis in HeLa cells . Compared to the wild-type strain ( EIB202 ) , one of the mutants ( 0909I ) greatly increased LDH release ( Figs S1A and 1A ) and IL-18 secretion ( Fig 1B ) , accompanied by significantly induced caspase-4 activity ( Fig 1C ) in HeLa . To explore whether 0909I promotes caspase-4-dependent non-canonical inflammasome activation in human intestinal epithelial cell lines . , we extended the bacterial infection experiments to the wild-type and Caspase-4-/- Caco-2 and HT-29 cells . Robust pyroptosis ( Figs 1D and S1B ) and significantly increased LDH release ( Figs 1E and S1C ) and IL-18 secretion ( Figs 1F and S1D ) were detected in wild-type cells infected with 0909I compared to those infected with EIB202 , which were abrogated in Caspase4-/- cells . These data indicate that E . tarda mutant 0909I promotes caspase-4 dependent inflammasome activation in non-phagocyte cells . Next , bioinformatics analysis revealed that the transposon insert site within 0909I is located upstream of a non-RTX hemolysin-encoding gene , ethA [29] . In agreement with the upregulated transcription level of ethA ( S2A Fig ) , 0909I showed higher EthA expression ( S2B Fig ) and bacterial hemolytic activity ( S2C Fig ) than EIB202 , which were abolished in the strain of 0909IΔethA . Further , deletion of ethA significantly impaired the ability of 0909I to increase caspase-4 activation in Caco-2 ( Fig 1D–1F ) and HT-29 cells ( S1B–S1D Fig ) . These data suggest that 0909I promotes non-canonical inflammasome activation by increasing hemolysin expression . To explore whether hemolysin also upregulates non-canonical inflammasome activation in other enteric bacteria , we expanded the investigation to the best-known RTX hemolysin , HlyA in UPEC , enterohemorrhagic E . coli ( EHEC ) , enteropathogenic E . coli ( EPEC ) and E . coli K12 . Similarly , the E . coli strains , containing HlyA expression plasmid , showed significantly higher hemolytic activity ( S2D Fig ) , which elicited a higher level of caspase-4 dependent LDH release ( Fig 1G–1J ) and IL-18 maturation ( S3A–S3D Fig ) than their corresponding wild-type strains . Together , these results suggest that hemolysin plays critical roles in promoting caspase-4 activation during Gram-negative bacterial infection . Host immunity senses bacterial LPS in the cytosol via caspase-11/4/5 [4–6] . Because extracellular LPS cannot simply diffuse across the membrane , and most Gram-negative bacteria are not cytosolic , delivery of LPS from bacteria to the cytoplasm is thus very critical for non-canonical inflammasome activation . Because our data demonstrated that hemolysin increased caspase-4 activation in intestinal epithelial cells , it is reasonable to speculate that hemolysin may promote cytosolic release of bacterial LPS during infection . We extracted the cytosol of uninfected or infected Caco2 cells using digitonin and assessed LPS levels in a limulus amebocyte lysate ( LAL ) assay . Digitonin is commonly used to isolate cytosol [30] , and the use of an extremely low concentration of digitonin ( 0 . 005% ) for a very short duration allows extraction of cytosol devoid of plasma membrane , early and late endosomes , and lysosomes [20] . The LAL assay showed that LPS was present in the cytosol of infected cells , but not in uninfected cells ( Fig 2A and 2B ) . Significantly higher levels of cytosolic LPS was detected in 0909I-infected cells than in EIB202-infected cells , and it was greatly reduced by deletion of ethA in 0909I ( Fig 2B ) . These data demonstrate that hemolysin promotes LPS delivery into the cytoplasm upon E . tarda infection . Edwardsiella tarda is an intracellular pathogen that replicates within a E . tarda-containing vacuole ( ECV ) [20 , 31–32] and has no direct access to the cytoplasm . mGBP2-induced vacuole destabilization was reported to release vacuolar bacteria for cytosolic LPS recognition [14] . However , no bacteria were recovered by agar plating in the cytosol extracted from E . tarda-infected cells ( S4A Fig ) , suggesting that hemolysin did not promote vacuole-constrained E . tarda to enter the cytosol . Simultaneously , comparative CFUs were detected in the residual fraction between 0909I , EIB202 , and 0909IΔethA ( S4A Fig ) , indicating that hemolysin did not affect cellular uptake of E . tarda . Further , to discriminate that hemolysin promotes LPS release to the cytoplasm from vacuolar or extracellular bacteria , three endocytosis inhibitors , cytochalasin D ( CD ) , 5- ( N-ethhyl-n-isopropil ) -amiloride ( EIPA ) , and dynasore ( Dyn ) were used to inhibit the internalization of E . tarda in Caco-2 cells . Clearly , CD and EIPA inhibited the internalization of 0909I ( S4B Fig ) , but did not reduce either cytoplasmic LPS release ( Fig 2C ) or caspase-4 activation ( Figs 2D and S5A ) , suggesting that hemolysin-mediated LPS delivery predominantly depends on extracellular bacteria . Unexpectedly , dynasore similarly inhibited cellular uptake of E . tarda ( S4B Fig ) , but significantly suppressed LPS delivery ( Fig 2C ) and sensing by cytosolic caspase-4 ( Figs 2D and S5A ) , indicating that dynasore may block a pathway that delivers LPS from extracellular bacteria to the cytoplasm . Next , we investigated the effects of CD and Dyn in E . coli-infected Caco-2 cells . Dyn significantly repressed LPS cytosolic release ( Fig 2E ) , cell death ( Fig 2F ) and IL-18 secretion ( S5B Fig ) in cells , while CD showed no influence on them . These results suggest that hemolysin promotes LPS cytosolic release from extracellular bacteria through a dynamin-dependent endocytosis process . Outer membrane vesicles ( OMVs ) are spherical , bilayered nanostructures constitutively released by growing bacteria [33] . The association of bacterial toxins with OMVs protects them from inactivation or degradation during infection , representing a highly efficient mechanism of bacteria modulating host defenses [34] . Previous studies have reported that bacterial hemolysins , such as HlyA in E . coli was delivered into cells via association with OMVs in a dynamin-dependent manner [22 , 35–37] . To dissect the role of E . tarda hemolysin in promoting caspase-4 inflammasome activation , we first explored the localization of EthA in this bacterium . Here , blebbing and shedding of OMVs were observed in growing E . tarda ( S6A Fig ) and produced in supernatants over time ( S6B Fig ) . We fractionated the bacterial culture into pellets , OMV-free supernatants and OMVs . Notably , over 60% of the total EthA was detected in the fraction of OMVs ( Fig 3A ) , indicating that OMV-associated EthA is the major form of this toxin in E . tarda . In accordance with the increased hemolytic activity in 0909I ( S2C Fig ) , significantly higher levels of EthA and hemolytic activity were detected in 0909I OMVs than in EIB202 or 0909IΔethA OMVs ( Fig 3B ) . In contrast , when subject to proteinase K ( PK ) digestion , in which proteins inside OMVs are protected from degradation , OMV-associated EthA was completely degraded , which agrees with the decreased hemolytic activity ( Fig 3B ) . These data suggest that EthA is exposed on the exterior of E . tarda OMVs , in accord with the localization of HlyA in E . coli [22 , 36] . A recent study suggested that OMVs were responsible for delivering LPS into the cytosol and activating caspase-11 inflammasome [20] . Because E . tarda hemolysin was largely associated with OMVs and critically involved in caspase-4-dependent inflammasome activation , it raises the possibility that E . tarda employs hemolysin-associated OMVs to activate the non-canonical inflammasome during infection . Subsequently , purified E . tarda OMVs were incubated with wild-type or Caspase-4-/- cells . OMVs internalized and formed obvious specks within cells ( S6C Fig ) at comparable level between EIB202 , 0909I , and 0909IΔethA , which were significantly inhibited by treatment with Dyn , but not with CD or EIPA ( Fig 3C ) . Importantly , according to the higher hemolytic activity ( Fig 3B ) , 0909I OMVs induced obvious cell pyroptosis ( Fig 3D ) , and a higher level of LDH release ( Fig 3F ) than EIB202 OMVs in wild-type cells , and this difference was counteracted in Caspase-4-/- cells ( Fig 3E ) . Further , hemolysin depletion by deleting ethA ( Fig 3E ) or pretreating OMVs with PK ( Fig 3D ) remarkably reduced 0909I OMV-induced caspase-4 activation . Simultaneously , treatment with dynasore to inhibit cellular uptake of 0909I OMVs reduced LDH release in wild-type cells ( Fig 3E ) . Similarly , purified E . coli HlyA+ OMVs significantly increased cell death than wild-type OMVs ( Fig 3F ) . These data suggest that hemolysin promotes OMV-mediated caspase-4 inflammasome activation in Gram-negative bacteria . As LPS-enriched OMVs are internalized via endocytosis and restrained within endosomes [38] , it raises the question of how vacuole-contained OMVs achieve cytosolic localization of LPS . Because hemolysin is a type of bacterial pore-forming toxins , that leads to membrane permeabilization [21] , a possible explanation could be that hemolysin induces lysis of OMV-containing vacuoles and facilitates cytosolic exposure of LPS . Galectin-3 is a β-galactoside binding protein , that is specifically recruited to disrupted pathogen-containing vacuoles [39] . We assessed the recruitment of GFP-tagged galectin-3 in cells incubated with OMVs . Indeed , 0909I OMVs induced more galectin-3 specks than EIB202 OMVs ( Figs S7A and 4A ) , and galectin-3 showed clearly association with OMV specks within wild-type cells ( Fig 4B ) . In contrast , removal of hemolysin from 0909I OMVs by proteinase K degradation or deleting ethA significantly decreased the intracellular galectin specks in wild-type cells ( Fig 4A ) . These data suggest that hemolysin promotes rupture of OMV-containing vacuoles . Next , we explored the contribution of hemolysin to damaging the membrane of OMV-containing vacuoles during E . tarda infection . As expected , 0909I triggered significantly increased signal of galectin aggregation in DMSO-treated cells , compared to EIB202 or 0909IΔethA ( S7B and S7C Fig ) . In the presence of EIPA , which inhibited the internalization of bacteria , but not OMVs into cells , 0909I induced obvious cytoplasmic galectin specks ( S7B and S7C Fig ) , indicating that 0909I-induced galectin aggregation is mainly associated with internalized OMVs rather than bacteria . Further , 0909I trigged a significant increase in cytoplasmic galectin signal , compared to EIB202 or 0909I ΔethA , which was greatly suppressed by pretreating the cells with Dyn ( S7B and S7C Fig ) . These results suggest that hemolysin contributes to the destruction of OMV-residing vesicles during bacterial infection . Because hemolysin triggered significant membrane rupture of OMV-containing vacuoles within cells , we evaluated whether vesicle lysis facilitates LPS exposure to cytosolic sensors . We extracted the cytosol and quantified the cytosolic LPS in Caspase-4-/- cells upon incubation with purified OMVs . Although EIB202 , 0909I , and 0909IΔethA OMVs showed comparable LPS contents ( Fig 4C ) and uptake efficiencies ( Fig 3C and 3D ) , significantly increased cytosolic LPS was detected in the cytosol of 0909I OMV-incubated cells than in the controls . This difference was remarkably reduced by either eliminating OMV-bound hemolysin or damping OMV internalization ( Fig 4D ) . Furthermore , purified OMVs from E . coli strains were incubated with Caspase-4-/- cells for the cytosolic LPS assay . Accordingly , HlyA+ OMVs induced more LPS release into the cytoplasm than their correspondent wild-type OMVs ( Fig 4E ) , indicating that hemolysin promotes cytosolic release of OMV-LPS . Collectively , these data suggest that hemolysin-mediated membrane lysis of OMV-containing vacuoles represents an important means to liberate OMV-LPS for cytosolic sensing in Gram-negative bacteria . As hemolysin was verified to promote non-canonical caspase-4 inflammasome activation in intestinal epithelium cell lines , we further assessed the involvement of hemolysin in noncanonical inflammasome activation in vivo . C57BL/6 wild-type mice were orally infected with E . tarda strains . Compared to EIB202 , 0909I showed significantly reduced bacterial burdens in the colon , cecum and lumen ( Fig 5A ) , but not at the systemic sites ( S8A Fig ) , indicating that that over-expressed hemolysin restricts E . tarda colonization in the mouse gut . Subsequently , it is interesting to explore the in vivo relevance of hemolysin-mediated gut infection restriction with non-canonical inflammasome activation . Caspase-11 is the murine ortholog of human caspase-4 , which was reported to mediate IL-18 secretion and enterocyte pyroptosis [40–42] during intestinal infection . Notably , caspase-11 depletion counteracted the superiority of bacterial loads in wild-type mice infected with 0909I over EIB202 ( Fig 5A ) , suggesting the critical requirement of caspase-11 by hemolysin-triggered gut infection limitation . Further , we demonstrated that 0909I induced remarkably higher mucosal and serum IL-18 level than EIB202 in wild-type mice ( Fig 5B and 5C ) . Tissue pathology analysis revealed that 0909I evoked prominent intestinal inflammation in wild-type mice , typically featured by focal filtration of inflammatory cells and epithelia cell shedding ( S8B and S8C Fig ) . Moreover , intraperitoneally injection of 0909I OMVs caused a stronger IL-1β and IL-18 release compared with 0909IΔethA OMVs in wild-type mice ( Fig 5D and 5E ) . In contrast , these phenotypes were largely neutralized in Caspase-11-/- mice ( Fig 5B–5E ) . Together , these results indicate that dysregulation of hemolysin in vivo alerts caspase-11 dependent intestinal defenses via OMVs and restricts bacterial gut infection .
Sensing of LPS in the cytosol by inflammatory caspase-11/4/5 has emerged as a central event of innate immune responses during Gram-negative bacterial infections [4–5 , 7] . Vanaja et al . reported that OMVs of extracellular Gram-negative bacteria can deliver LPS into the host cell cytosol from early endosomes; however , the mechanism of LPS translocation remains unclear . Biological membrane characteristics inherent to OMVs may permit them to fuse with endosomal membranes , leading to LPS cytosolic access [20] . Recently , it was reported that mGBPs were involved in OMV-dependent non-canonical inflammasome activation [44–45] . Mechanistically , mGBPs didn’t promote the entry of OMVs into the cytosol , but directly target cytosolic OMVs and facilitate the interaction of LPS with caspase-11 . However , GBPs are different in mice and human due to a loss of immunity-related GTPases ( IRGs ) in humans , and thus hGBP2 might not have the same action as mGBP2 [46] . In addition to host factors , whether bacterial factors , such as OMV-associated bacterial components , are involved in promoting non-canonical inflammasome activation remains unknown . Here , we demonstrate that hemolysin binds OMVs and promotes the lysis of OMV-residing vesicles , which facilitates cytosolic release of OMV-LPS and eventually triggers significant non-canonical inflammasome signals ( Fig 6 ) . Our results suggest that hemolysin represents a biologically important mechanism for releasing endosome-constrained OMV-LPS to cytosolic sensors . In addition to hemolysin , whether other pore-forming proteins produced by bacteria [47] play roles in liberating pathogen-associated molecular patterns for immune detection requires further examination . Outer membrane vesicles are extruded from the surface of living bacteria and may entrap some of the underlying periplasm [33] . In addition to outer membrane proteins , LPS , phospholipids , and periplasmic constituents , OMVs deliver various bacterial cargos , including toxins intracellularly and further in specific compartments to modulate host defenses during infection [34] . Thus , OMVs offer bacterial products specific access to otherwise inaccessible cellular or tissue compartments . Particularly , OMVs enable the cytosolic localization of LPS for caspase-11 sensing [20] , which is thought to be an important mechanism in extracellular bacteria . In this work , in addition to E . coli strains , E . tarda , a typical vacuolar bacterium [48] was also found to employ OMVs to activate caspase-4 with the aid of overproduced hemolysin . These results indicate that OMV-mediated non-canonical inflammasome activation represents a more generalized mechanism for bacteria than previously thought . Moreover , because OMVs were linked to the translocation of diverse bacterial factors into host cells , it will be interesting to determine whether hemolysin also increases cytosolic recognition of other ligands such as DNA or flagellin by the AIM2 [49] or NLRC4 [50] inflammasome . Hemolysin , as a class of pore-forming toxins , is known to disrupt the membrane of eukaryotic cells at high doses in vitro , which potentially facilitates bacterial invasion and dissemination during infection [21] . HlyA is the prototype member of the hemolysin family [21] . Only a very low percentage of nonpathogenic E . coli harbors HlyA , while ∼50% of UPEC strains harbor and express HlyA , presenting a close correlation between increased clinic E . coli pathogenesis with HlyA expression [21 , 23 , 51] . In recent years , because it has been hypothesized that hemolysin is secreted at sublytic concentrations in vivo , there is increasing interest in understanding the more subtle effects of hemolysin on host cellular processes aside from outright lysis , including RhoA activation [52] , Akt signaling [53] , cell death pathway [22 , 54] and protease degradation [55] . Strikingly , Escherichia coli α-hemolysin toxin was reported to inhibit IL-1β secretion induced by Rho GTPase-activating toxin CNF1 , thereby promoting bacterial stability in the blood [56] . Here , our study adds new findings to the scenario of hemolysin-mediated biological functions by identifying that abnormal expression of hemolysin in bacteria promotes host immune recognition of replicating pathogens via non-canonical inflammasomes and restricts bacterial colonization in vivo ( Fig 6 ) . Because hemolytic proteins exhibit multilayered or even antagonistic roles in modulating host responses to infections , it is not surprising that bacteria have evolved complicated mechanisms to fine-tune the synthesis , maturation , and transport of hemolysin [21 , 23 , 51] . Thus , studies of the spatiotemporal expression of hemolysin in vivo will be essential for understanding both host defense and bacterial survival . Inflammasomes are activation platforms for inflammatory caspases , leading to pyroptosis and secretion of IL-1α , IL-1β , and IL-18 [15] . In contrast to the canonical inflammasome , caspase-11/4/5 mediated non-canonical inflammasome is activated by the cytosolic LPS , however , whether caspase-11/4/5 can directly cleave IL-1β or IL-18 remains controversial [41–42] . Given that the actions of inflammasomes have been well-studied in myeloid cells , accumulating evidence highlights their role in non-professional immune cells as an important antibacterial defense mechanism [43] . Recent studies demonstrated that epithelium utilize both canonical and non-canonical inflammasomes to promote mucosal defense against enteric bacterial pathogens [40–42] . These independent studies suggest that pyroptosis might lead to extrusion of infected enterocytes and promote bacterial infection in vivo even in the absence of IL-1α , IL-1β , and IL-18 . In our study , hemolysin-overexpressed E . tarda significantly promoted caspase-11-dependent IL-18 secretion and gut inflammation in orally-infected mice . Furthermore , intraperitoneal injection of hemolysin-rich OMVs induced higher level of serum IL-18 and IL-1β in mice than the control OMVs . These results suggest that hemolysin-associated OMVs are , if not fully , at least partially responsible for initiating Caspase-11 mediated immune defenses in vivo . Further insights into whether hemolysin-associated OMVs induces enterocyte pyroptosis and further recruits immune cells , ultimately contributing to pathogen clearance from the host , require future studies .
The animal trials in this study were performed according to the Chinese Regulations of Laboratory Animals—The Guidelines for the Care of Laboratory Animals ( Ministry of Science and Technology of People's Republic of China ) and Laboratory Animal-Requirements of Environment and Housing Facilities ( GB 14925–2010 , National Laboratory Animal Standardization Technical Committee ) . The license number associated with their research protocol was 20170912–08 , which was approved by The Laboratory Animal Ethical Committee of East China University of Science and Technology . All surgery was performed under carbon dioxide anesthesia , and all efforts were made to minimize suffering . 0909I E . tarda was screened from E . tarda gene-defined mutant library . 0909IΔethA was constructed by unmarked gene deletion . Hemolysin-overexpressing E . coli strains were constructed by introducing into the wild-type strain ( EIB202 , CCTCC M208068 ) , the plasmid of pSF4000-hlyBACD [55] into UPEC UT189 , EHEC EDL933 [57] , EPEC ( E . coli O26 ) [58] and K12 , respectively . Hemolytic activity was quantified using a microtiter assay developed previously [59] with slight modifications . Indicated bacterial strains were cultured for 18–20 h and then pelleted , resuspended in PBS , and standardized at 600 nm to an OD of approximately 1 . 0 . Next , 100 μL bacterial suspensions or OMV preparations were mixed with equal volumes of sheep erythrocytes in a 96-well plate . The plate was incubated at 37°C with gentle shaking for the indicated time periods and erythrocytes were pelleted by centrifugation ( 400 × g , 5 min ) . Clear supernatants were transferred to a fresh plate and absorbance at 540 nm was measured using a microplate reader ( Dynex Technologies , Chantilly , VA , USA ) . Suspension buffer containing 0 . 1% Triton X-100 was served as a total lysis control . The percentage of hemolysis was calculated as follows: % hemolysis = ( OD540 of samples—OD540 of background ) / ( OD540 of total—OD540 of background ) × 100 . RNA was extracted using an RNA isolation kit ( Tiangen , Beijing , China ) . One microgram of each RNA sample was used for cDNA synthesis with the FastKing One Step RT-PCR Kit ( Tiangen ) and quantitative real-time PCR ( RT-qPCR ) was performed on an FTC-200 detector ( Funglyn Biotech , Shanghai , China ) using SuperReal PreMix Plus ( SYBR Green ) ( Tiangen ) . The gene expression of bacterial ethA was evaluated for three biological replicates , and the data for each sample were expressed relative to the expression level of the 16S gene by using the 2-ΔΔCT method . sgRNA ( Oligo1: GACCGGGTCATCTCTGGCGTACTCC; Oligo2: AAACGGAGTACGCCAGA GATGACCC ) was designed ( http://zifit . partners . org ) and cloned into LentiCRISPR v2 ( AddGene , Cambridge , MA , USA; 52961 ) , containing puromycin resistance . Lentiviral particles were prepared in HEK239T cells as previously described [31] . HT-29 or Caco-2 cells were prepared in approximately 30% density and infected with lentiviral particles containing 10 μg/mL polybrene for 24 h . Transfected cells were selected with puromycin ( 2 μg/mL ) . Cells were diluted to 5 cells/mL and seeded to 96-well plates , followed by an expansion period to establish a new clonal cell line . The caspase-4-/- cell lines were confirmed by western blotting using anti-caspase-4 antibody . Wild-type or Caspase-4-/- Caco-2 , HT-29 and HeLa cells ( wild-type cells were from ATCC; F . Shao’s Lab provided caspase-4 deficient HeLa cells , other deficient cells were established by our Lab ) were seeded and grown to a density of ~4×105 cells per well in 12-well plates or ~105 cells per well in 24-well plates . For bacterial infection , cells were incubated with indicated E . tarda or E . coli strains at a MOI of 25 or 50 , respectively . Infection was initiated by centrifuging the plate at 600 ×g for 10 min . After 50-min incubation at 35°C for E . tarda strains or 37°C for E . coli strains , the plates were washed once with PBS and then transferred into fresh medium containing 100 μg/mL gentamicin ( Gm ) to kill extracellular bacteria . The time point after antibiotic treatment was recorded as 0 h after infection . For OMV infection , cells were seeded and grown to a density of ~105 cells per well in 24-well plates , and then incubated with purified 50 μg OMVs . Cell supernatants were harvested at the indicated time points for subsequent assays . Aliquots of cellular supernatants were transferred into 96-well plates ( round bottom ) and centrifuged at 1000 ×g for 5 min . The supernatants were transferred to another 96-well plate ( flat bottom ) , and the plate was subjected to the cytotoxicity assay using a CytoTox 96 assay kit ( G1780 , Promega , Madison , WI , USA ) or an ELISA kit ( eBioscience , San Diego , CA , USA ) according to the manufacturer’s protocol . Each sample was tested in triplicate . Cytotoxicity was normalized to Triton X-100 treatment ( 100% of control ) , and LDH release from uninfected/untreated cells was used for background subtraction . OMVs were purified from E . tarda strains as described previously with minor modifications [51] . Briefly , the bacterial strains were grown in 3000 mL of DMEM till OD600 of ~1 . 5 and the bacteria-free supernatant was collected by centrifugation at 5000 ×g for 10 min at 4°C . This supernatant was further filtered through a 0 . 45 μm filter and concentrated using a spin concentrator ( GE Healthcare , Little Chalfont , UK; molecular weight cut off = 30 kDa ) . Subsequently , OMVs were pelleted by ultracentrifugation at 284100 ×g for 1 . 5 h at 4°C in a Beckman NVTTM65 rotor ( Brea , CA , USA ) . Isolated OMVs were resuspended in PBS , transferred to the bottom of a 13-mL ultracentrifugation tube ( Beckman Coulter ) and adjusted to 45% OptiPrep ( Sigma-Aldrich , St . Louis , MO , USA ) in a final volume of 2 mL . Different OptiPrep/PBS layers were sequentially added as follows: 2 mL of 40% , 2 mL of 35% , 2 mL of 30% , 2 mL of 25% and 1 mL of 20% . Gradients were centrifuged ( 284000 ×g , 16 h , 4°C ) in a Beckman NVTTM65 rotor and the fractions of equal volumes ( 1 mL ) were removed sequentially from the top . After removing the OptiPrep , OMVs were resuspended in 1000 μL sterile DMEM without phenol red . The protein content of purified OMVs was assessed by modified Bradford protein assay kit ( Shenggong Biotech , Shanghai , China ) . OMV-free supernatants or OMVs , treated with proteinase K ( Sigma ) ( 5 μg/mL , 5 min ) or not were separated by SDS-PAGE and immunoblotted with antibodies against OmpA , EthA , or RNAP . HeLa cells were seeded and grown to a density of ~7 × 104 cells per well in 24-well plates . Cells were pretreated with dyn ( 80 μM ) or CD ( 1 μg/mL ) for 30 min at 37°C and incubated with 20 μg OMVs for 4 h . Cells were then washed , fixed , and quenched , and permeabilized/blocked with PBS containing 1 mg/mL saponin 10% and goat serum . OMVs were stained with anti-OmpA antibody and Alexa Fluor 488-conjugated goat anti-rabbit IgG . Actin was counterstained with TRITC phalloidin ( Yeasen Biotech , Shanghai , China ) and nuclei with DAPI ( Beyotime , Jiangsu , China ) . Fixed samples were observed under a confocal microscope ( Nikon , Tokyo , Japan; A1R ) . The eukaryotic expression plasmid for GFP tagged galectin-3 was constructed and introduced into HeLa cells by lentivirus transfection [51] . Briefly , HeLa cells were seeded in 24-well plates at a density of 5 × 104 cells per well in antibiotic-free medium and transfected with lentiviral particles for 24 h . Next , positively-transfected cells were pooled via treating cells with 1 . 5 μg/ mL of puromycin . The pooled cells were seeded and grown to a density of ~105 cells per well in 24-well plates . For bacterial infection , transfected cells were preincubated with EIPA ( 30 μM ) or Dyn ( 80 μM ) for 1 h before infection ( DMSO pre-treatment as a control ) and then challenged with different E . tarda strains at an MOI of 50 . After 2 . 5 h-incubation at 35°C , the specks of galectin recruitment were observed under a confocal microscope ( Nikon , A1R ) . For OMV infection , cells were incubated with purified OMVs at 50 μg/105 cells . After 16-h incubation , the cells were washed twice with sterile PBS and fixed with 4% paraformaldehyde ( PFA ) at 25°C for 2 h , then washed in PBS and permeabilized with Triton X-100 ( 0 . 1% in PBS , 10 min at 25°C ) , and blocked in 5% bovine serum albumin . After overnight incubation with OmpA antibody , the cells were incubated with secondary fluorescent antibody for 1 h and DAPI was used for nuclear counterstaining . Fixed samples were observed under a confocal microscope ( Nikon , A1R ) Subcellular cell fractions were extracted by a digitonin-based fractionation with modifications [30] . Briefly , Caspase-4-/- HeLa or Caco2 cells post incubation with indicated strain or purified OMVs were washed with sterile cold PBS 6 times on a platform shaker on ice to remove attached bacteria or OMVs . Subsequently , 250 or 100 μL of 0 . 005% digitonin extraction buffer was added to the bacteria-incubated cells in 12-well plates or OMV-incubated cells in 24-well plates , respectively . After 8 min , the supernatants were collected by centrifugation as the fraction containing cytosol , and the residuals were resuspended in 250 or 100 μL of 0 . 1% CHAPS buffer , respectively , as the fraction containing cell membrane , organelles and nucleus . Cytosol and residual fractions were subjected to the Limulus Amebocyte Lysate ( LAL ) assay ( Associates of Cape Cod , East Falmouth , MA , USA ) according to the manufacturer’s instructions to quantify LPS . C57BL/6J wild-type and Caspase-11-/- mice from Jackson Lab ( 6–8 weeks old ) were bred under specific pathogen-free conditions . For oral infections , water and food were withdrawn 4 h before per os ( p . o . ) treatment with 20 mg/100 μL streptomycin per mouse . Afterward , animals were supplied with water and food ad libitum . At 20 h after streptomycin treatment , water and food were withdrawn again for 4 h before the mice were orally infected with 5 × 107 CFU/g of EIB202 or 0909I suspension in 200 μL PBS , or treated with sterile PBS ( control ) . Thereafter , drinking water ad libitum was offered immediately and food 2 h post-infection . At the indicated times points , mice were sacrificed and the tissue samples from the intestinal tracts , kidneys , spleens , and livers were removed for analysis . For OMV infection , ten to 12-week-old C57BL/6 and Caspase-11-/- mice were primed with intraperitoneal ( i . p . ) administration of 200 μg of poly ( I:C ) ( high molecular weight; InvitroGen ) for 6 hr prior to i . p . injection with 80 μg of purified OMVs . Cytokine levels in the serum were analyzed at 6 hr post-OMV injection [20] . Collected tissues or organs were homogenized in PBS ( pH 7 . 4 ) and the dilutions were plated on Deoxycholate Hydrogen Sulfide Lactose ( DHL ) agar plates for CFU counting . The cecum and colon were collected in 10% neutral-buffered formalin for histological analyses . Tissue pathology was blindly scored by two researchers using hematoxylin and eosin-stained sections ( 6 μm ) . The scoring criteria for submucosal edema , PMN infiltration into the lamina propria , goblet cell loss and epithelial integrity was conducted as previously described by Barthel et al . [60] . In addition , inflammatory focal infiltration ( IFI ) within cross-section was scored at five levels ( 0 , none; 1 , inflammation occurrence , but without apparent IFI; 2 , with apparent IFI; 3 , more than one to three IFIs; 4 , with more than three IFIs ) . The cumulative scoring range was 0–17 . Cecal and colonic tissues were removed from mice 24 h after infection with E . tarda strains . Tissues were washed free of luminal contents and then incubated in DMEM ( supplemented with penicillin and streptomycin at 1% and streptomycin at 1 mg/mL ) for 6 h , and the supernatants , along with the serum samples , were collected for IL-18 detection by ELISA following the manufacturer’s protocols . Rabbit anti-Na+/K+ ATPase ( 1:1000; 3010S; Cell Signaling Technology ) , mouse anti-EEA1 ( 1:1000; ab70521; Abcam ) ; rabbit anti-Rab7 ( 1;1000; ab137029; Abcam ) ; rabbit anti-GAPDH ( 1:1000; 5174S; Cell Signaling Technology ) , rabbit anti-E . tarda OmpA polyclonal antibody ( 1:500; custom-made; Genscript Biotech , Piscataway , NJ , USA ) , rabbit anti-E . tarda EthA polyclonal antibody ( 1: 1000; custom-made; Genscript ) , and mouse anti-RNAP monoclonal antibody ( 1:5000; Santa Cruz Biotechnology ) were used for western blotting or immunofluorescence .
|
Sensing of lipopolysaccharide ( LPS ) in the cytosol triggers non-canonical inflammasome-mediated innate responses . Recent work revealed that bacterial outer membrane vesicles ( OMVs ) enables LPS to access the cytosol for extracellular bacteria . However , since intracellular OMVs are generally constrained in endosomes , how OMV-derived LPS gain access to the cytosol remains unknown . Here , we reported that hemolysin largely bound with OMVs and entered cells through dynamin-dependent endocytosis . Intracellular hemolysin significantly impaired OMVs-constrained vacuole integrity and increased OMV-derived LPS exposure to the cytosolic sensor , which promoted non-canonical inflammasome activation and restricted bacterial gut infections . This work reveals that dysregulated hemolysin promotes non-canonical inflammasome activation and alerts host immune recognition , providing insights into the more sophisticated biological functions of hemolysin upon infection .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"medicine",
"and",
"health",
"sciences",
"vacuoles",
"pathology",
"and",
"laboratory",
"medicine",
"hela",
"cells",
"pathogens",
"immunology",
"biological",
"cultures",
"microbiology",
"physiological",
"processes",
"inflammasomes",
"cell",
"cultures",
"caco-2",
"cells",
"cellular",
"structures",
"and",
"organelles",
"bacterial",
"pathogens",
"research",
"and",
"analysis",
"methods",
"immune",
"system",
"proteins",
"proteins",
"medical",
"microbiology",
"microbial",
"pathogens",
"cell",
"lines",
"cytoplasm",
"biochemistry",
"cell",
"biology",
"physiology",
"secretion",
"biology",
"and",
"life",
"sciences",
"cultured",
"tumor",
"cells",
"cytosol"
] |
2018
|
Dysregulated hemolysin liberates bacterial outer membrane vesicles for cytosolic lipopolysaccharide sensing
|
The Ku heterodimer associates with the Saccharomyces cerevisiae telomere , where it impacts several aspects of telomere structure and function . Although Ku avidly binds DNA ends via a preformed channel , its ability to associate with telomeres via this mechanism could be challenged by factors known to bind directly to the chromosome terminus . This has led to uncertainty as to whether Ku itself binds directly to telomeric ends and whether end association is crucial for Ku's telomeric functions . To address these questions , we constructed DNA end binding–defective Ku heterodimers by altering amino acid residues in Ku70 and Ku80 that were predicted to contact DNA . These mutants continued to associate with their known telomere-related partners , such as Sir4 , a factor required for telomeric silencing , and TLC1 , the RNA component of telomerase . Despite these interactions , we found that the Ku mutants had markedly reduced association with telomeric chromatin and null-like deficiencies for telomere end protection , length regulation , and silencing functions . In contrast to Ku null strains , the DNA end binding defective Ku mutants resulted in increased , rather than markedly decreased , imprecise end-joining proficiency at an induced double-strand break . This result further supports that it was the specific loss of Ku's telomere end binding that resulted in telomeric defects rather than global loss of Ku's functions . The extensive telomere defects observed in these mutants lead us to propose that Ku is an integral component of the terminal telomeric cap , where it promotes a specific architecture that is central to telomere function and maintenance .
Eukaryotic telomeres are comprised of repetitive G-rich sequence arrays and a host of proteins that associate with the repeats directly or as part of a complex [1] , [2] . These proteins maintain telomeres by facilitating the replication of telomeric DNA and by protecting the natural chromosomal ends from end-joining and resection activities associated with the repair of DNA double-strand breaks ( DSBs ) . Perturbation of the normal structure and function of the telomere , through the loss of telomere sequence or telomere-associated proteins , can result in genomic instability , checkpoint activation , and cellular senescence or apoptosis [3] . Understanding the structural relationship of telomere-associated proteins with telomeric DNA is integral to our comprehension of the maintenance , structure , and function of telomeres . To date , proteins have been shown to associate with telomeric DNA via at least one of three mechanisms . Some telomere-associated proteins , such as Rap1 in Saccharomyces cerevisiae [4] and TRF1 and TRF2 in mammalian cells [5] , utilize a myb/homeodomain to bind directly and with high affinity to double-stranded ( ds ) telomeric DNA sequences . Other proteins , such as Cdc13 in S . cerevisiae [6] and the TEBPα/β heterodimer in Oxytricha nova [7] , utilize one or more oligonucleotide-oligosaccharide binding ( OB ) folds to associate avidly and specifically with single-stranded ( ss ) G-rich telomeric DNA . Because telomeres terminate with G-rich overhangs , ss telomeric DNA binding proteins are localized to the chromosome end . The third mechanism by which proteins associate with telomeres is via protein-protein interactions . For example , in S . cerevisiae , Sir3 and Sir4 are recruited to telomeric repeat-containing chromatin via Rap1 [8] , and in mammalian cells , TIN2 localizes to ds telomeric repeats via its interaction with TRF1 and TRF2 [9] , [10] . Proteins capable of binding telomeric DNA directly also can employ this mode of association . For example , POT1 , a protein that binds directly to the G-overhang through its OB folds [11] , also associates with ds telomeric repeats through a series of protein-protein interactions terminating with TRF1 [12] . Thus , the ability to bind to telomeric DNA directly does not preclude telomeric association via protein-protein interactions . In contrast to the above examples , the mechanism for the evolutionarily conserved Ku heterodimer's association to the telomere is uncertain . Comprised of the Ku70 and Ku80 subunits , Ku loads onto DNA ends via a preformed ring consisting of an expansive base and narrow bridge that encircles the DNA [13] , [14] . Ku binds DNA ends in a sequence-independent manner through a limited number of contacts with the sugar phosphate backbone [14] . Even though Ku binds DNA in a sequence independent manner , it does load onto DNA ends in a specific orientation . The loading face is comprised predominantly of Ku80 , while the lagging face consists mostly of Ku70 . This orientation is possibly dictated by steric and electrostatic features present on the lagging face that impede the DNA's access to this end of the DNA binding channel . Given its high affinity for DNA ends , Ku could localize to telomeric chromatin through direct DNA end binding [15] . In vitro assays have shown that Ku does associate with a telomeric DNA substrate in this way , as long as it is allowed access to the DNA before Cdc13 is added [16] . However , Cdc13 is associated with telomeres throughout the cell cycle in vivo [17] . Thus , the means by which telomeric end binding by Ku might occur in vivo in the context of Cdc13 and other telomeric binding proteins that avidly bind either ds or ss telomeric DNA remains unclear . In addition , the telomeres of many species are thought to exist in a higher order structure known as a t-loop , where the terminal G-strand 3′ overhang invades proximal ds telomeric DNA to form a d-loop , thereby concealing the DNA end [18] . In fact , one proposed function of this structure is to prevent Ku's association with the chromosome's terminus , in turn , preventing telomeres from engaging in Ku-dependent , nonhomologous end-joining ( NHEJ ) [19] . In addition to binding DNA , Ku70 and Ku80 each have N terminal α/β domains which lie laterally to the DNA binding channel and are thought to mediate Ku's interaction with other factors [14] . Ku has been shown to bind or co-purify with telomere-associated factors such as Sir4 in budding yeast and RAP1 , TRF1 , and TRF2 in mammalian cells [20]–[23] . In the case of Sir4 , this has been shown to require residues in the yeast Ku80 α/β domain . Also , Ku's interaction with TLC1 , the RNA component of telomerase , is disrupted by a mutation ( yku80-135i ) that maps to the Ku80 α/β domain [24] . Therefore , Ku might also associate with telomeres via its interaction with telomeric factors . Consistent with this , the association of budding yeast Ku80 with subtelomeric chromatin is reduced in the absence of Sir4 [25] . Thus , the possibility exists that Ku is recruited to telomeric repeats through these interactions , independent of its DNA end binding activity . Defining how Ku associates with telomeres is crucial for understanding its non-overlapping roles in a myriad of telomeric processes . For example , in S . cerevisiae , Ku is required for the localization of Est2 , the catalytic subunit of yeast telomerase , to telomeres in G1 [17] . This recruitment depends upon a direct interaction between Ku and a 48-nt stem-loop in TLC1 [24] . Ku also protects the telomeric 5′ strand from resection by Exo1 [26] , [27] , a 5′-3′ exonuclease and flap-endonuclease involved in a variety of DNA repair processes including resection of DSBs [28] , [29] . Studies with the Yku80 separation-of-function mutant yku80-135i , however , indicate that Ku can regulate telomere length independently of its end protection function [24] . In addition to , and distinct from , its role in telomere replication , Ku aids in the formation of silent telomeric chromatin [30] , which results in transcriptional repression of adjacent genes , a phenomenon referred to as telomeric silencing or telomere position effect [31] . Ku promotes the recruitment of Sir4 to sub-telomeric DNA , [25] , [32] , [33] and separation-of-function mutants have demonstrated that Ku's interaction with Sir4 , while required for Ku's role in telomeric silencing , is dispensable for Ku's role in telomere end protection and length homeostasis [34] . Yeast Ku also is required for the normal localization of telomeres at the nuclear periphery via two pathways that are independent of Sir4 and silent chromatin [35] , [36] . One of these pathways is S-phase specific and relies on Ku's interaction with TLC1 and telomerase's interaction with the integral nuclear membrane protein Mps3 [37] . Although perinuclear tethering can promote telomeric silencing , Ku's influence on silencing can occur , at least at some telomeres , independently of its effects on tethering . This is supported by the finding that the perinuclear localization of truncated versions of Tel VI-R and Tel VII-L is maintained in a yku70-Δ background ( i . e . , the tethering of these telomeres is Ku-independent ) , whereas the telomeric silencing of Tel VI-R ADE2 and Tel VII-L URA3 reporters is lost ( i . e . , the silencing of these telomeres is Ku-dependent ) [35] , [38] . In addition to telomeres , Ku localizes to DSBs , where it plays a crucial role in NHEJ . Separation-of-function mutations that result in isolated NHEJ or telomeric defects suggest that Ku plays distinct roles at telomeric versus broken ends [20] , [24] , [34] , [36] , [39] , [40] . For example , mutations in helix 5 of the yeast Ku70 α/β domain impact selectively on NHEJ , whereas mutations in helix 5 of the yeast Ku80 α/β domain impact selectively on telomeric silencing [40] . The structural position of NHEJ versus telomeric function-specific residues and the polarity with which the heterodimer loads onto DNA ends led us to propose a ‘two-face’ model for how Ku's NHEJ and telomeric functions are spatially organized [40] . In this model , we proposed that Ku's outward face , juxtaposed with the DNA terminus when bound to a broken end , is dedicated to its NHEJ functions , whereas its inward face , which would be oriented toward telomeric chromatin when bound to a telomeric end , is dedicated to Ku's telomeric functions . This model assumes , however , that Ku loads directly onto telomeric ends and that Ku's telomeric functions are dependent upon this mode of binding . To determine whether Ku associates with telomeres and mediates one or more of its telomeric roles via end binding , we generated DNA end binding defective S . cerevisiae Ku heterodimers and examined their association with telomeric chromatin and their impact on telomere function . We found that the loss of DNA end binding activity greatly reduced Ku's association with telomeres and rendered Ku deficient in its ability to protect telomeres from nucleolytic processing , to maintain telomere length , and to contribute to the formation of silent telomeric chromatin . Therefore , Ku must access and directly associate with telomeric ends in order for telomeres to adopt their proper architecture and perform their genome-protective functions .
To determine whether DNA end binding is required for Ku's association with telomeres and its telomeric functions , we sought to create S . cerevisiae Ku heterodimers that were specifically defective in their ability to bind DNA ends by mutagenizing residues that interact with DNA . We used the crystal structure of human Ku bound to DNA ( PDB 1JEY ) [14] and multiple sequence alignments containing yeast and human Ku70 ( Yku70 or hKu70 ) and Ku80 ( Yku80 or hKu80 ) subunits to identify potential DNA contacting residues in yeast Ku . Twenty-four residues in human Ku , distributed between the two subunits , contact DNA in the crystal structure . Because of poor sequence conservation between most of these residues and the aligning residues in yeast , we limited single amino acid substitutions to a subset of five residues that ranked in the top 20th percentile of our previous analyses of evolutionarily conserved amino acid residues of Ku70 and Ku80 [40] ( Figure 1 ) . We also targeted Yku70-R456 , the yeast residue that aligned with hKu70-R444 , which was in close proximity to DNA in the human Ku-DNA co-crystal and highly conserved throughout evolution . In addition to these single amino acid substitutions , we targeted residues that corresponded to small runs of DNA contacting residues in each of the human subunits ( Yku70-IMFQ266-269 and Yku80-SKKDS400-404 which correspond to hKu70-ALSR255-258 and hKu80-YDKRA397-401 ) . The impact of the mutations on DNA end binding was determined first by individually expressing each mutant allele in yeast lacking the corresponding wild type genomic copy and assaying DNA end binding activity in whole cell extracts ( WCE ) using an in vitro electrophoretic mobility shift assay ( EMSA ) [41] . In this assay , a 1000-fold molar excess of non-labeled closed circular DNA inhibits nonspecific binding to the 198 bp radiolabeled DNA fragment , resulting in an electrophoretic mobility shift of the radiolabeled fragment that is dependent on Ku binding to the DNA ends [41] ( see also Figure 2A ) . As sought after , we saw significantly reduced end binding activity in two of the single missense mutant strains examined , yku70-K422E and yku70-R456E . Although Ku is known to bind DNA ends independently of other factors , we also examined purified recombinant WT and Yku70-R456E/Yku80 ( Figure S1 ) heterodimers . We observed a 50- to 100-fold reduction in the DNA end binding affinity of the mutant ( Figure 2B ) , confirming that the targeting indeed affected Ku's inherent ability to bind DNA ends . As shown in Figure 1B , the yku70-IMFQ266-269QDEY and yku80-SKKDS400-404DEEDD mutations ( referred to hereafter as yku70-p1 and yku80-p1 , respectively ) targeted amino acids on opposing sides of the DNA binding channel . Whereas expression of yku70-p1 with YKU80 or yku80-p1 with YKU70 did not result in a severe deficiency in DNA end binding , expression of both panel mutant alleles resulted in undetectable DNA end binding activity in the WCE EMSA ( Figure 2A and Figure S2 ) . These results suggest that there are synergistic contributions to Ku's stable association with DNA ends mediated by amino acids on each side ( i . e . , leading and lagging ) of the DNA binding channel . Prior to utilizing the alleles identified as being DNA end binding defective for in vivo analyses , we needed to determine whether the loss of DNA end binding was the indirect result of markedly reduced steady state protein levels or loss of heterodimer formation . Using epitope tagged versions of Yku70 and Yku80 , we found that the mutations did not substantially alter steady state protein levels ( Figure 2C , input ) or the ability of the mutant subunits to form heterodimers ( Figure 2C , α-FLAG IP ) . Anti-FLAG immunoprecipitates from yku70-R456EFLAG , yku70-p1FLAG , and yku80-p1myc18 single mutant WCE contained WT levels of both subunits , despite the presence of the mutation . The levels of Yku70-K422EFLAG and Yku80myc18 , and Yku70-p1FLAG and Yku80-p1myc18 were only slightly reduced in the yku70-K422EFLAG YKU80myc18 single mutant , and yku70-p1FLAG yku80-p1myc18 double mutant strains' input and anti-FLAG IPs . Thus , the yku70-K422E and yku70-R456E single mutants and the combination of the yku70-p1 and yku80-p1 panel mutant alleles appeared to directly impact Ku's ability to bind DNA ends . As with heterodimerization , the interpretation of our in vivo findings would be confounded if the mutations conferring a DNA end binding defect also disrupted Ku's ability to associate with the telomeric factor Sir4 or the telomerase subunit TLC1 . Yeast Ku interacts with Sir4 via the Yku80 subunit [40]; therefore , we examined whether the Yku80-Sir4 interaction was retained in the DNA end binding defective mutants using a previously reported Yku80-Sir4 yeast two-hybrid system with a LEU2 reporter [20] . Consistent with the prior observation that the Yku80-Sir4 yeast two-hybrid interaction was independent of Yku70 [20] , we found that the presence of yku70-R456E , yku70-p1 or yku70-K422E instead of YKU70 did not impair the association between Yku80 and Sir4 , as evidenced by growth of strains expressing these alleles , the YKU80 BAIT , and the SIR4 PREY on media lacking leucine ( Figure 3A , compare strains 3 and 6 with 1 , and 11 with 9 ) . Significantly , the substitution of yku80-p1 for YKU80 on the BAIT plasmid in YKU70 or yku70-p1 cells did not inhibit growth in the absence of leucine ( Figure 3A , compare strains 7 and 8 with 1 ) . These results suggest that the mutations did not hinder the association between Yku80 and Sir4 . We also assayed the in vivo association of TLC1 with the Yku70-R456E Yku80 and Yku70-p1 Yku80-p1 Ku heterodimers . To do so , we immunoprecipitated mutant or WT Ku from asynchronous cultures expressing myc or FLAG tagged-Yku80 and performed real-time , reverse transcriptase polymerase chain reaction ( RT-PCR ) to detect TLC1 RNA . The yku70-R456E strain exhibited a level of TLC1 enrichment comparable to the YKU70 strain in both myc- and FLAG-tagged YKU80 backgrounds ( Figure 3B; P = 0 . 51 and 0 . 20 , respectively ) . Enrichment for TLC1 was not seen in the ‘no tag’ control or in a tlc1-Δ48 strain , which is defective for interaction with Yku80 [17] ( P = 0 . 007 ) . Likewise , actin RNA was not enriched in any of the reactions , demonstrating specificity of the co-immunoprecipitation . TLC1 was enriched similarly in yku70-p1 yku80-p1FLAG anti-FLAG IPs ( P = 0 . 17 ) . Thus , both of these DNA end binding defective Ku heterodimers were capable of binding TLC1 in vivo . The above analyses indicated that we had succeeded in introducing mutations into Ku that markedly impacted on DNA end binding without substantially altering protein stability , heterodimerization or interactions with known telomeric factors [40] . Therefore , we proceeded to address whether selective loss of Ku's DNA end binding activity altered its association with telomeric chromatin in vivo by performing chromatin immunoprecipitation ( ChIP ) assays . We found an 80–86% decrease in the amount of telomeric DNA that immunoprecipitated with Yku80myc in the yku70-R456E strain relative to the amount immunoprecipitated in the YKU70 strain ( Figure 4A and Figure S3A ) . This decrease was not due to decreased levels of Yku80myc or decreased immunoprecipitation efficiency , as protein levels appeared equal in both the input and α-myc IP samples ( Figure 4B and Figure S3B ) . We also performed ChIP experiments with the yku70-p1 yku80-p1 double panel mutant strain ( Figure S3A ) , which showed undetectable DNA end binding in the WCE EMSA ( Figure 2A ) . The immunoprecipitates isolated from the double panel mutant contained less than 1% of the telomeric DNA isolated from WT immunoprecipitates . However , we could not rule out the possibility that this was due to the reduced immunoprecipitation efficiency that was observed in this mutant under these conditions ( Figure S3B ) . Thus , we conclude that , at the very least , the vast majority of Ku's localization to telomeres requires that it bind directly to telomeric ends . An important role of Ku at telomeres is to protect the C-rich telomeric strand from exonucleolytic degradation by Exo1 . The mechanism by which this is accomplished has not been directly studied; however , the simplest model posits that Ku sterically hinders Exo1's access and/or progression by loading directly onto telomeric ends . By examining the telomeric G-overhangs in the DNA end binding defective strains , we could now test this model directly . To do this , XhoI-digested genomic DNA prepared from asynchronous cultures was hybridized with a C-rich radiolabeled telomeric probe and subjected to native gel electrophoresis as previously described [34] ( Figure 5A , native ) . The signal was quantified and normalized to the total telomeric G-strand DNA signal , which was obtained by in-gel DNA denaturation followed by hybridization with the same probe ( Figure 5A , denatured , and 5B ) . Two of the DNA end binding defective Ku mutants , yku70-R456E and yku70-p1 yku80-p1 , exhibited readily detectable ss G-rich DNA signals similar to a yku70-Δ strain . Thus , a loss of DNA end binding correlated with a loss of telomere end protection . In apparent contradiction to this correlation , the yku70-K422E mutant strain , which was also defective in DNA end binding in the WCE EMSA , exhibited only a slight increase in ss G-rich DNA , comparable to strains expressing yku70-p1 or yku80-p1 ( Figure 5 ) . This could indicate that Ku's end binding activity is , in fact , not required for Ku's telomere end protection function . However , yku70-K422E single mutants also appeared similar to WT in all additional assays performed ( telomere length , telomeric silencing , synthetic lethality with tlc1-Δ , temperature sensitivity , and imprecise NHEJ; Figure S4 and data not shown ) . It was possible that this mutant protein was unfolded in vitro or differentially sensitive to the WCE EMSA conditions . An alternative explanation was that the Yku70-K422E mutant protein , with its mutated residue on the lagging face of the DNA binding channel ( Figure 1 ) , was stabilized at telomeric ends in vivo . We reasoned that if this was the case then the combination of the yku70-K422E allele with the yku80-p1 allele , whose mutated residues are on the loading face of the DNA binding channel ( Figure 1 ) , might destabilize the mutant Yku70-K422E-containing Ku's association with DNA ends in vivo and give rise to defective telomere end protection . Indeed , we found a synergistic increase in the ss G-rich DNA signal in the yku70-K422E yku80-p1 double mutant strain ( Figure 5 ) . An increase was not observed , however , when yku70-K422E was combined with the yku80-8 allele , which contains a mutated residue on Ku's loading face . This mutation impairs Ku's ability to bind Sir4 [40] , but it does not affect Ku's ability to bind DNA ends ( Figure S5 ) . These findings suggest that the ability of Ku to protect telomeres from exonucleolytic resection requires that Ku bind directly to the telomeric end . As described earlier , a second important telomeric function of Ku is in the maintenance of telomere length . Whereas it is known that this requires Ku's interaction with TLC1 , it is unknown whether it requires Ku's direct association with telomeric ends . To address this question , we determined the length of telomeres present in yku70-R456E and yku70-p1 yku80-p1 mutant strains . We found that the telomeres in these mutants were nearly as short as those in yku70-Δ and yku80-Δ null strains ( Figure 6A ) . In contrast , the telomeres in the yku70-p1 and yku80-p1 single mutants were of WT length or only slightly shortened . Additionally , as illustrated in the yku70-R456E mutant , the degree of shortening was much greater than in a tlc1-Δ48 mutant , in which Ku can no longer bind TLC1 . Moreover , the telomeres were no shorter in a yku70-R456E tlc1-Δ48 double mutant than in a yku70-R456E single mutant ( Figure 6B ) . Therefore , and consistent with the TLC1 co-immunoprecipitation results ( Figure 3B ) , the shortening observed in these DNA end binding defective Ku mutants was due to a defect downstream of TLC1 association . These data suggest that Ku must load onto and stably associate with the telomere end in order to mediate its role in telomere length regulation , yet it was possible that the telomere length defect in the DNA end binding defective strains was due to the increased single-strandedness of the G-rich telomeric strand . If the failure to protect telomeric ends was the sole contributing factor to the shortened telomeres present in the end binding defective Ku mutants , then the deletion of EXO1 should fully restore the length of the telomeres . We found that , although the deletion of EXO1 fully complemented the defect in end protection of DNA end binding defective mutants ( Figure S6 ) , it did not fully restore telomeres to a WT length ( Figure 6C ) . Thus , Ku's ability to bind telomeric ends contributes directly to the maintenance of telomere length beyond simple telomeric end protection . Whereas we predicted Ku's association with the telomeric end would be required for its role in telomere end protection and length regulation , a role for Ku's DNA end binding in the formation of subtelomeric heterochromatin was not as clearly anticipated [30] , [42] , [43] . In the formation of subtelomeric chromatin , Ku is recruited to the centromere-proximal region of subtelomeric DNA in a Sir4-dependent manner [25] . Therefore , the DNA end binding activity of Ku may not be required for Ku's ability to promote the silencing of subtelomeric reporters . To address this question , we utilized a strain containing a URA3 telomeric reporter gene , which is subject to Ku-dependent silencing [43] , to examine whether the DNA end binding mutations resulted in telomeric silencing defects . WT strains grow poorly on medium lacking uracil ( and robustly on 5-FOA medium ) because of the spread of telomeric heterochromatin and secondary transcriptional repression of the adjacent URA3 gene . Cells with a silencing deficiency , such as yku70-Δ , display the opposite growth pattern – robust growth in the absence of uracil and minimal growth in the presence of 5-FOA . As in the previous assays of Ku's telomere end protection and length maintenance functions , the yku70-p1 yku80-p1 double mutant strain exhibited a markedly defective phenotype , whereas the single panel mutant strains , yku70-p1 and yku80-p1 , displayed a phenotype similar to that of WT ( Figure 7A ) . The silencing defect in the yku70-R456E mutant was not as severe as the defect observed in the yku70-Δ or double panel mutant strain , but was clearly evident by the larger colony size and greater colony number on the plate lacking uracil ( Figure 7A ) . This partial silencing defect was also evident when we examined the expression of a second telomeric reporter gene in this strain , ADE2 . The reversible nature of telomeric silencing results in variegation , such that WT colonies are mostly red ( due to ADE2 repression ) with some white sectors ( due to ADE2 expression ) . However , colonies from strains that are deficient in telomeric silencing are white in appearance . Utilizing this reporter , we found that the double-panel mutant and yku70-Δ strains were white , hence , defective for telomeric silencing ( Figure 7A ) . The yku70-R456E strain exhibited a very slight variegated appearance , suggesting that it was not completely defective in silencing; however , it was clearly dysfunctional when compared to WT , yku70-p1 , and yku80-p1 strains . Rap1 plays a central role in the establishment of telomeric silencing by binding ds telomeric repeats and recruiting the essential silencing proteins Sir3 and Sir4 to chromatin [8] . Because excess degradation of the telomeric sequence would result in the loss of Rap1 binding sites , the telomeric silencing defect of the DNA end binding defective Ku mutant strains could be an indirect consequence of their defect in telomere end protection . To determine if such was the case , we repeated the silencing assay in an exo1-Δ background . Deletion of EXO1 restored the level of ss , G-rich DNA in the DNA end binding defective mutants to near WT levels ( Figure S6 ) , which was further reflected by suppression of the temperature sensitive phenotype ( Figure 7B ) . Despite restoring ds telomeric DNA , we found that deleting EXO1 only slightly restored silencing , as evidenced by the slightly decreased growth of yku70-R456E exo1-Δ and yku70-p1 yku80-p1 exo1-Δ strains on medium lacking uracil relative to their corresponding EXO1 strains ( compare Figure 7B to 7A ) . Similarly , the deletion of EXO1 led to the appearance of very faint pink colonies in the DNA end binding defective mutants ( Figure 7B ) . Altogether , these results indicate that the DNA end binding activity of Ku ( and , hence , its direct association with telomeric DNA ) is required for its role in telomeric silencing . In each of the telomeric function assays performed , the yku70-R456E and yku70-p1 yku80-p1 mutants exhibited loss of function similar to that of yku70-Δ or yku80-Δ strains; however , the interaction assays ( e . g . , for heterodimerization , TLC1 interaction , and Yku80-Sir4 yeast two-hybrid interaction ) argued against a global null phenotype . This was further revealed in an assay for imprecise NHEJ in which the DNA end binding defective mutants could be distinguished clearly from Ku-deficient strains . The yeast strain used in this assay contains an HO endonuclease cleavage site at the mating type ( MAT ) locus but lacks HML and HMR donor sequences , which are required for the normal homology directed repair of HO-cleaved MAT [44] . The expression of HO endonuclease , and , hence , the induction of a DSB , is under the control of a galactose-inducible promoter . Because of the absence of a homologous donor sequence , only NHEJ can repair a DSB created by HO at MAT in this strain . When cells are plated on galactose-containing medium , HO is constitutively expressed and available to re-cleave any accurately repaired cleavage sites , precluding cell growth . However , when the HO cleavage site is altered by an imprecise end-joining event , a colony can form . This imprecise NHEJ occurs in approximately 0 . 1% of plated WT cells [44] . Previous studies have demonstrated that colony survival in the presence of constitutively expressed HO ( i . e . , imprecise NHEJ ) is reduced to 0 . 0001% of cells plated in the absence of Ku [44]; therefore , we expected to observe little colony formation when the DNA end binding defective Ku mutations were examined in this assay . However , unlike the yku70-Δ strain , the yku70-R456E strain gave rise to ample colonies when grown on galactose ( Figure 8A , +HO ) . This strain also was engineered to test whether the yku70-R456E mutation impacted cell viability in the absence of TLC1 ( and , therefore , telomerase ) , which requires Ku's telomere end protection function [40] . In contrast to the result of the imprecise NHEJ assay , the effect of yku70-R456E mutation on survival in the absence of telomerase was similar to that of a yku70-Δ mutation because both resulted in synthetic lethality when combined with a tlc1-Δ mutation ( Figure 8A , tlc1-Δ ) . Thus , the seemingly WT behavior of the yku70-R456E mutant in the HO assay ( versus the more null-like behavior in the telomere function assays mentioned above ) was not due to a difference in the general requirement for Ku in the different strain backgrounds . We noted that the yku70-R456E mutants appeared to grow more robustly than WT when plated as serial dilutions on galactose-containing medium ( Figure 8A , +HO ) . To confirm this , as well as to test the yku70-p1 yku80-p1 double panel mutant for this phenotype , we repeated the assay in a strain containing an intact chromosomal copy of TLC1 . We consistently found that both DNA end binding defective mutants displayed greater than WT number of colonies on galactose-containing medium ( Figure 8B ) . The yku70-Δ mutant , in contrast , showed minimal growth on galactose . Additionally , as with the telomeric assays , yku70-p1 and yku80-p1 single mutant strains behaved like WT with comparable levels of survival ( Figure 8B ) . Therefore , the phenotype of the DNA end binding defective mutants was different not only from the yku70-Δ strain , but also from WT . Taken together , these results provide phenotypic evidence that argue against the telomeric phenotypes simply being a reflection of a generalized Ku deficiency .
We found that the singular loss of Ku's ability to bind DNA ends had dramatic effects on its association with telomeric chromatin and its telomeric functions . These results demonstrate that Ku predominantly associates with telomeres via direct DNA end binding and that DNA end binding is required for Ku to carry out its various telomeric functions . Thus , Ku is an integral component of the telomere cap . So , what is involved in this integral association between Ku and telomeric DNA ? The phenotypes seen with the DNA end binding defective mutants and the position of these mutations within the Ku heterodimer provide clues that address this question . There appear to be at least two distinct steps involved: the initial loading of Ku onto the telomeric end , and the stabilization of Ku at telomeres . For example , the yku70-R456E mutant contains a single amino acid substitution on the loading side of the central channel ( Figure 1 ) . More telomeric chromatin is associated with Yku70-R456E Yku80 heterodimers ( Figure 4 ) than one would expect based upon in vitro DNA end binding assays with recombinant protein ( Figure 2B ) . This is likely due to the differences in substrates in the two assays – naked DNA fragments versus telomeres , and these data suggest that , once this mutant Ku heterodimer is able to load onto telomeric ends in vivo , a fraction can be stabilized . In vitro , this mutant protein is not as stably bound to DNA as WT , as suggested by the excessive smearing seen in the EMSA ( Figure 2B ) . The synergistic loss of telomeric function and synergistic increase in imprecise NHEJ observed with the yku70-K422E yku80-p1 double versus the yku70-K422E and yku80-p1 single mutants ( Figure 5 and Figure S4 ) also points to factors influencing the stabilization of Ku's binding to DNA ends in vivo relative to the in vitro assays . The loss of Ku's ability to bind DNA ends affected each of the telomeric functions examined in this study . Least surprisingly , loss of DNA end binding resulted in a telomere end protection defect , with an EXO1-dependent increase in G-overhang DNA as observed in yku-Δ strains ( Figure 5 and Figure S6 ) . These data are consistent with Ku's DNA end binding contributing to end protection through direct steric hindrance of Exo1 . However , the mechanism of Ku-mediated telomere end protection is likely to be more complex . Despite the association of Yku80 ( and presumably Yku70 ) with telomeres throughout the cell cycle [17] , its role in the protection of de novo telomeres is limited to G1 [45] , a time when the Cdc13-Stn1-Ten1 ( CST ) complex is dispensable [46] . This may be the case at native telomeres as well [45] . Thus , Ku or telomeres may be altered as cells progress through S phase , after which telomeres no longer require Ku , but rather depend on the CST complex , as well as Rap1 , Rif1 and Rif2 , for their end protection [45] , [46] . That end binding is required for Ku's role in telomere end protection is also consistent with the observation that Ku is required to prevent the initiation of telomere resection [45] . Ku has been shown to repress recombination near telomeres . It has been proposed that this repression is mediated through Ku's promotion of a telomere fold-back architecture via interactions with the subtelomeric core X element [47] , [48] . Such an architecture may also serve to protect telomeres from end resection . Such a structure is likely to unfold with progression through S phase and telomere replication , perhaps limiting its role to G1 . This opening of the fold-back structure is consistent with the greater amount of sub-telomeric DNA found in Yku80 ChIPs from cells in G1 compared to the amount seen after cells have entered S phase [17] . Determining whether the DNA end binding Ku mutants retain their interaction with core X and/or repress recombination near telomeres will be important toward our understanding of the significance of Ku – core X interactions . In contrast to a defect in telomere end protection , the defect in telomeric silencing observed in the DNA end binding defective mutants ( Figure 7A ) was not necessarily expected for two reasons . First , Ku does not impact on Sir4′s association with telomeric repeat-associated chromatin , but rather on its association with subtelomeric nucleosomal chromatin [33] . Second , the association of Ku with subtelomeric chromatin is diminished in the absence of Sir4 , suggesting that Ku is recruited to the subtelomere via its association with Sir4 [25] . How , then , might Ku's association with the very end affect subtelomeric chromatin ? The answer to this question may also relate to a role for Ku in mediating a fold-back structure at telomeres and the importance of this structure for the formation or maintenance of silent telomeric chromatin [47]–[50] . We note that , although , the truncated Tel V-R and Tel VII-L telomeres used in our silencing assays lack core X elements , Ku has been shown to associate with non-core X sequences near telomeres and has core X-independent effects on recombination repression [47] , [48] . Ku also contributes to the tethering of some telomeres to the nuclear periphery , a phenomenon that contributes to telomeric silencing [43] , [51] . Thus , it is possible ( and indeed likely ) that the Ku DNA end binding mutations , which alter Ku's association with telomeres , also impact on the localization of some telomeres to the nuclear envelope . Therefore , might the effects these mutations have on telomeric silencing be a consequence of loss of telomere perinuclear localization ? Although we can not eliminate this possibility at those native telomeres that are subjected to both Ku-mediated localization and silencing , the perinuclear anchoring of the truncated Tel V-R and Tel VII-L telomeres used in our telomeric silencing assays is not Ku-dependent [38] , [52] . Therefore , the results from this assay reveal that the silencing defects observed in DNA end binding defective mutants are independent of Ku's role in nuclear tethering of telomeres . The short telomere phenotype seen in Ku DNA end binding defective mutants suggests that the association of Ku with telomeric ends is also essential for Ku's ability to positively regulate telomere length . Ku-TLC1 association was not significantly diminished in the DNA end binding defective mutants ( Figure 3B ) , arguing against this simple explanation for the short telomeres in these mutants ( Figure 6A ) . Moreover , yku70-R456E mutants had telomeres that were significantly shorter than the telomeres present in tlc1-Δ48 mutants ( Figure 6B ) . These results indicate that binding TLC1 ( and presumably telomerase ) alone is insufficient and that end binding is essential for Ku-mediated telomere length regulation . Deletion of EXO1 , while suppressing the increased ss G-strand telomeric DNA , had only minor effects on the short telomere phenotype , arguing against the loss of end protection in the end binding defective mutants as the primary contributor to the loss of telomere length ( Figure 6C ) . Therefore , we conclude that the direct association of Ku with the very end is required for Ku to contribute to telomere length maintenance . By targeting residues in Ku that were predicted to contact DNA , we generated Ku heterodimers that were defective for DNA end binding yet proficient for TLC1 association in vivo . These mutants are the reverse complement of the yku80-135i mutant , which is defective for its ability to bind TLC1 but fully competent for binding DNA ends [24] . These data suggest that Ku's association with TLC1 does not involve the central channel and that Ku has separate determinants for binding DNA ends and TLC1 . Whether Ku can simultaneously bind telomeric DNA and TLC1 RNA has yet to be determined and would have implications for molecular models of how Ku influences telomere length regulation . We observed that strains that were defective in DNA end binding did not phenocopy Ku-deficient strains in an assay for imprecise end-joining ( Figure 8 ) . This result supports the conclusion that the telomere defects in these strains were not due to a generalized deficiency in function . Quantification of survival under constant DSB-inducing conditions , and subsequent sequencing of the repaired junctions , revealed that the DNA end binding defective mutants exhibited an increase in survival relative to WT and that the repaired junctions were distinct from junctions observed in WT and null strains ( Figure 8 and data not shown ) . Thus , these mutants exhibit a novel phenotype . This likely relates to the fact that , in contrast to yku70-Δ or yku80-Δ strains , Ku is still present in these mutants; in contrast to WT strains , in which Ku is also present , the Ku that is present exhibits markedly reduced binding to DSBs . How the presence of these mutant proteins might result in more efficient re-ligation through an error-prone end-joining pathway is unknown . One possibility is that they have retained not only their interactions with telomeric factors ( such as Sir4 and TLC1 as we have shown ) , but also with NHEJ factors . Consequently , DNA end binding defective Ku heterodimers might titrate one or more factors away from DSBs , which results in an increase in Ku-independent error-prone NHEJ . Our findings show that Ku must access and load onto telomeric ends in order for it to participate in its various telomeric functions in S . cerevisiae . It will be interesting to determine whether Ku's end binding property is also required at human telomeres . Ku in human cells has been shown to prevent the rapid loss of telomeric repeats and the generation of so-called t-circles ( extrachromosomal circular DNA containing telomeric repeats ) [53] . Perhaps in human cells , Ku plays a role in stabilizing the t-loop structure , similar to its proposed role in maintaining the protective fold-back architecture of yeast telomeres .
The strains and plasmids used in this study are described in Table S1 and Table S2 , respectively . Strains lacking YKU70 and/or YKU80 were constructed using one-step allele replacement with the indicated selection marker . YAB285 , YAB438 and YAB439 were constructed by integrating yku70-R456E , yku70-p1 and yku80-p1 into YTSF79 [17] by pop-in/pop-out technique . Mutations were introduced into YKU70 and YKU80 containing plasmids using oligonucleotide single-stranded mutagenesis as described [40] . For EMSA using yeast WCEs , cell lysates were prepared , and EMSA was performed using a modification of a previously published protocol [41] . Five ml cultures were lysed with glass beads in lysis buffer [420 mM KCl , 42 mM HEPES , 4 . 2 mM EDTA , 0 . 1 mM dithiothreitol ( DTT ) ] . Approximately 10–15 µg of protein from each cell lysate was incubated with ( 1 . 5 ng ) 32P end-labeled Bgl II – Nru I fragment ( 198 bp ) of pcDNA3 . 1 ( Invitrogen ) for 15 minutes at room temperature , along with 1 µg unlabeled closed circular pcDNA3 . 1 . The samples were loaded onto a 5% polyacrylamide gel in Tris-glycine buffer and run at 30 mA for approximately 45 minutes . The gel was dried and exposed to film or a phosphorimager screen and analyzed using a Storm865 imaging system ( Molecular Dynamics ) . Quantitation of the phosphorimager image was done using ImageQuant software . For EMSA using recombinant Ku , binding reactions were prepared containing 20 mM Tris-HCl pH 8 , 50 mM NaCl , 1 mM DTT , 5 mM MgCl2 , 0 . 5 mM EDTA , 100 µg/ml BSA , 11% glycerol , 1 . 5 ng of 32P- labeled DNA ( as used for the yeast WCE EMSA ) , and the indicated amounts of purified recombinant Ku . Reactions were incubated for 15 minutes at room temperature . The reactions were loaded onto a 5% polyacrylamide gel , which was prerun at 40 mA for 30 minutes . Electrophoresis was carried out in Tris-glycine buffer at 40 mA for 100 minutes . The gel was dried and analyzed as above . Ku70 , Ku70-R456E and Ku80 were individually subcloned into pFastBac HT A ( Invitrogen ) opened with NcoI and NotI . An N-terminal 6xHis tag followed by a linker region and a TEV protease recognition sequence precedes the multiple cloning site . Individual recombinant baculoviruses expressing either Ku70 , Ku70-R456E or Ku80 were made from the bacmids and used to infect Sf9 cells by the Baylor Baculovirus Core . Sf9 cell stocks ( Invitrogen ) were maintained as suspension cultures in spinner vessels prior to infection . Sf9 cells were co-infected at a MOI of 1 for each recombinant virus for 48 hours in a 1 liter spinner vessel in 500 ml of Grace's Insect Medium ( Invitrogen ) + 10% FBS at a density of 1 . 0×106 cells/ml . At 48 hours post-infection , the cells were centrifuged at 800× g for 10 minutes , washed one time in PBS , and the cell pellets were immediately placed in a −80°C freezer for storage . Cells were disrupted by the addition of lysis buffer ( 50 mM phosphate buffer pH 8 . 0 , 300 mM NaCl , 0 . 05% Tween 20 , 5 mM β-mercaptoethanol , 10 mM imidazole ) . The lysates were incubated on ice for 30–60 minutes and then sonicated on ice using a Misonix sonicator probe . The lysate was cleared by ultracentrifugation at 26K rpm , 4°C for 15 minutes and then incubated with Ni-NTA agarose ( Qiagen ) rotating for 2 hours at 4°C . The resin was then washed twice in buffer ( 50 mM phosphate buffer pH 8 . 0 , 300 mM NaCl , 5 mM β-mercaptoethanol , 20 mM imidazole ) . Recombinant Ku was eluted in buffer ( 50 mM phosphate buffer pH 8 . 0 , 300 mM NaCl , 5 mM β-mercaptoethanol , 250 mM imidazole ) . The Ku containing eluates were identified by SDS-polyacrylamide gel electrophoresis ( PAGE ) and Coomassie stain , then combined and concentrated on an 50NMDL Amicon filter ( Millipore ) . The Ku retentate was then put over a Superdex 200 16/60 sizing column and Ku dimer containing eluates were concentrated on a 50NMDL Amicon filter ( Millipore ) . The purity of the sample was verified by SDS-PAGE and silver staining . To remove the N-terminal 6xHis tag , Ku was incubated at 16°C overnight with AcTEV protease ( Invitrogen ) in buffer [50 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 0 . 5 mM EDTA , 1 mM DTT] . The digested sample was diluted 10-fold with buffer ( 50 mM Tris pH 8 and 300 mM NaCl ) and then loaded again onto a Ni-NTA column . The Ku containing fractions were combined and concentrated via a 50NMDL Amicon filter ( Millipore ) in 50 mM Tris-HCl pH 8 . 0 , 300 mM NaCl . Storage buffer ( 0 . 5 mM EDTA , 1 mM DTT , 40% glycerol ) was added and Ku was then aliquoted , flash frozen and stored at −80°C . YAB226 ( yku70-Δ yku80-Δ ) was co-transformed with FLAG-tagged versions of YKU70 , yku70-p1 , yku70-R456E , or yku70-K422E and myc-tagged versions of YKU80 or yku80-p1 . Fifty ml asynchronous cultures were lysed with glass beads in TMG-200 ( 10 mM Tris-HCl , pH 8 . 0 , 1 mM MgCl2 , 10% [v/v] glycerol , 0 . 1 mM DTT , 0 . 1 mM EDTA plus 200 mM NaCl ) . Forty µl of anti-FLAG agarose ( Sigma A220 ) was added to 150 µg of protein extract . After 3 hours of incubation at 4°C , the beads were washed in TMG-200 plus 0 . 5% Tween and resuspended in 40 µl TMG . Input and IP samples were subjected to SDS-PAGE on 10% gels and analyzed by western blotting with anti-myc ( 9E10 , Sigma M4439 ) and anti-PGK ( Molecular Probes A6457 ) primary antibodies and IRDye 800CW conjugated goat anti-mouse secondary antibody ( LiCor ) . Fluorescence was visualized using the LiCor Odyssey® Infared Imaging System . After quantitation was performed using ImageQuant software ( Molecular Dynamics ) , the membrane was stripped and re-probed with anti-FLAG ( M2 , Sigma F3165 ) . YAB327 is a yku70-Δ derivative of EGY48 [20] , which was grown on 5-FOA-containing media to select against the presence of the URA3-containing LacZ reporter plasmid . YAB327 was transformed with YKU70 ( pVL1874 ) , yku70-R456E ( pAB558 ) , yku70-p1 ( pAB608 ) , yku70-K422E ( pAB710 ) or pRS416 and then a pEG202-based BAIT plasmid containing YKU80 ( pEGku80 ) or yku80-p1 ( pAB599 ) and a pJG4-5-based PREY vector containing full-length Sir4 ( pB42AD::SIR4 ) . Interaction between the BAIT and PREY constructs was determined by growth on Gal –Leu plates . Procedures were performed using modified versions of described protocols [17] , [54] , [55] . Fifty mL cultures ( A600 = 0 . 8–1 . 0 ) were lysed in TMG-50 using silica beads , and extracts containing approximately 4 mg protein were prepared . For samples immunoprecipitated with anti-myc antibody ( 9E10 , Sigma M4439 ) , Protein G beads ( EMD Chemicals ) were added after they were incubated overnight at 4°C . Both FLAG conjugated beads ( Sigma A220 ) and Protein G beads were washed in TMG-50 plus 0 . 5% Tween before being resuspended in 50 µl TMG-50 . The resuspended beads ( and 50 µl aliquots of the inputs ) were added to 350 çl of Proteinase K solution and incubated at 37°C for 30 minutes . Following phenol:chloroform extraction and ethanol precipitation , nucleic acids were resuspended in 100 µl RNase free water . The RNA was subjected to RNA cleanup and on-column DNase I treatment using the RNeasy kit ( Qiagen ) . Two µg of input RNA ( 11 µl of IP RNA ) was used as a template for cDNA synthesis using the Flex cDNA synthesis kit ( Quanta Biosciences ) . Quantitative PCR was performed on 5 µl of the cDNA synthesis reaction using the Perfecta SYBR Green FastMix , ROX ( Quanta Biosciences ) on an ABI7300 real time PCR system ( Applied Biosystems ) . Actin and TLC1 primers used were the same as described [54] . ABI7300 software was used to determine threshold and CT values . CT values were averaged and plotted using MS Excel . Percent input was determined by multiplying 100 by 2 ( Adjusted input CT – IP CT ) , where adjusted input CT = Input CT - log2 ( dilution factor ) . Fifty ml cultures of YTSF23 ( YKU70 YKU80 ) [17] , YTSF79 ( YKU70 YKU80myc18 ) [17] , YAB285 ( yku70-R456E YKU80myc18 ) , YAB438 ( yku70-p1 YKU80myc18 ) and YAB439 ( yku70-p1 yku80-p1myc18 ) were grown at 28°C to an A600 = 0 . 8–1 . 0 . Formaldehyde crosslinking was carried out , as described by Aparicio [56] . Cell pellets were washed with 1 M sorbitol , resuspended in approximately 5 ml zymolyase buffer ( 1 M sorbitol , 50 mM Tris , 10 mM β-ME ) , and incubated with 125 µl zymolyase ( 50 mg/ml 20T zymolyase ) for 1 hour at 28°C , as described [57] . Spheroplasts were pelleted and resuspended in approximately 500 µl NP-S buffer [57] before brief ( 2 minutes ) glass bead lysis to further disrupt cells . The pellets were re-suspended in FA-lysis buffer [56] and sonicated with a Misonix Sonicator 3000 ( power setting 5 , 90 sec , 3 cycles ) . Total protein levels were adjusted so that equal amounts of protein were present in each sample . Five µl of 9E10 ( Sigma ) were added to IP samples and incubated overnight at 4°C . Protein G agarose beads ( Calbiochem IP04 ) were equilibrated with lysis buffer , added to IP samples , and incubated at 4°C for 3 hours . Beads were washed at room temperature according to Aparacio , et al . [56] . Crosslinks were reversed , and the DNA was ethanol precipitated . The DNA fragments ( ∼500 bp ) were re-suspended in 50 µl of TE and brought up to 300 µl with a 0 . 4N NaOH , 10 mM EDTA solution . The samples were heated to 100°C for 10 minutes before being blotted onto a Hybond XL nylon membrane ( Amersham ) using a dot blot manifold ( Schleicher & Schuell ) . The blot was probed with a telomere-specific , end-labeled oligonucleotide ( AB766 ) or a randomly labeled 3 . 2 kb DNA fragment containing TyB sequence excised from plasmid pAB126 . After excess probe was washed away , the blot was exposed to phosphorimager screen for quantitation . To monitor for efficiency of immunoprecipitation , 50 µl aliquots of the Input and IP samples were run on a 10% polyacrylamide gel , transferred to a nitrocellulose membrane , and probed with anti-myc ( 9E10 , Sigma M4439 ) and anti-PGK ( Molecular Probes A6457 ) primary antibodies and IRDye 800CW conjugated goat anti-mouse secondary antibody ( LiCor ) . Southern blot analysis of telomeres and telomeric G-strand overhang assays were done as described previously [34] , [58] . Plasmids containing WT or mutant yku70 and yku80 alleles or empty vector controls were co-transformed into YAB219 ( yku70-Δ yku80-Δ ) and YAB353 ( yku70-Δ yku80-Δ exo1-Δ ) . Cultures were grown to mid-log phase in –Trp –Leu media . Five-fold serial dilutions were spotted onto –Trp –Leu , –Trp –Leu –Ura , and –Trp –Leu low ade plates , to monitor the expression of URA3 and ADE2 . Growth was determined after 2 to 4 days at 28°C . Temperature sensitivity was determined by plating the serial dilutions onto –Trp –Leu plates and incubating at 37°C for 2 to 4 days . Plasmids containing yku70 and/or yku80 alleles or empty vector controls were transformed into YAB199 ( yku70-Δ ) , YAB198 ( yku70-Δ ) or YAB273 ( yku70-Δ yku80-Δ ) . Five-fold serial dilutions of asynchronous cultures were spotted onto –Trp and –Trp Gal plates ( for YAB198 and YAB199 ) –Trp –Leu and –Trp –Leu Gal plates ( YAB273 ) . For single colony experiments , asynchronous cultures were diluted and 100 µl were spread onto these sets of plates . Colonies were counted after 2 to 4 days of incubation at 28°C . A student's T-test was used to determine whether the differences in values shown on bar graphs were statistically significant .
|
The telomeric cap modulates telomere replication and prevents natural chromosome ends from being processed as DNA double-strand breaks ( DSBs ) . In multiple species , including budding yeast , a detailed picture exists of the factors that comprise the telomeric cap and how they associate with telomeric DNA . It is less clear where to place Ku , a conserved heterodimer involved in multiple aspects of telomere biology and DSB repair . Although Ku avidly binds DNA ends , its access to telomeric ends might be restricted by telomere binding proteins and/or higher-order telomere structure . Ku might also be recruited to telomeres via its telomere-associated binding partners . Here , we address whether Ku loads directly onto telomeric ends and whether direct DNA binding is crucial for its telomeric functions . Using structure-guided mutagenesis , we generated end binding–defective yeast Ku heterodimers that retained the ability to associate with Ku's known telomeric binding partners . These end binding–defective heterodimers showed a dramatic reduction in telomere association and were defective for all of Ku's telomeric functions . Our findings indicate that Ku is indeed a component of the telomere cap and that its loading onto telomeric ends is crucial for its telomeric functions and , perhaps , a specific telomere architecture .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"telomeres",
"cell",
"biology",
"chromosome",
"biology",
"biology",
"molecular",
"cell",
"biology"
] |
2011
|
Ku Must Load Directly onto the Chromosome End in Order to Mediate Its Telomeric Functions
|
Allosteric modulators are ligands for proteins that exert their effects via a different binding site than the natural ( orthosteric ) ligand site and hence form a conceptually distinct class of ligands for a target of interest . Here , the physicochemical and structural features of a large set of allosteric and non-allosteric ligands from the ChEMBL database of bioactive molecules are analyzed . In general allosteric modulators are relatively smaller , more lipophilic and more rigid compounds , though large differences exist between different targets and target classes . Furthermore , there are differences in the distribution of targets that bind these allosteric modulators . Allosteric modulators are over-represented in membrane receptors , ligand-gated ion channels and nuclear receptor targets , but are underrepresented in enzymes ( primarily proteases and kinases ) . Moreover , allosteric modulators tend to bind to their targets with a slightly lower potency ( 5 . 96 log units versus 6 . 66 log units , p<0 . 01 ) . However , this lower absolute affinity is compensated by their lower molecular weight and more lipophilic nature , leading to similar binding efficiency and surface efficiency indices . Subsequently a series of classifier models are trained , initially target class independent models followed by finer-grained target ( architecture/functional class ) based models using the target hierarchy of the ChEMBL database . Applications of these insights include the selection of likely allosteric modulators from existing compound collections , the design of novel chemical libraries biased towards allosteric regulators and the selection of targets potentially likely to yield allosteric modulators on screening . All data sets used in the paper are available for download .
The generation of drug-like lead and candidate molecules against a specific molecular target remains a major challenge in drug discovery . We are now in a position to partially understand the factors behind this , and they fall into two basic themes – 1 ) the diversity and size of the set of compounds used in the initial screen , and 2 ) the physicochemical properties of the binding site of the target , which may contain obligate features that are incompatible to binding molecules with drug-like properties [1]–[7] . There are now a large number of ‘tantalizing targets’ , those that have strong biological rationale ( for example genetic validation ) , but are currently outside the reach of the development of novel small molecule therapies . One strategy to avoid the issues of factor 2 ) above is to consider the development of allosteric regulators , which may have better , or at least differentiated physicochemical properties or advantages in selectivity and so forth [8]–[11] . The concept of allosterism has received ample attention in literature , yet the term is used relatively loosely , the current work starts by defining the definition of allosterism [8] , [12]–[17] . Allosteric modulators are ligands for a biological target that exert their effect on this target via a mechanism that is not located at the molecular site of action of those ligands that are the natural ligands or substrates for this protein . Hence the term ‘allosteric modulator’ covers a very broad spectrum of compounds and it depends on the context and function of the protein in question what effect allosteric modulators truly have . Thus , while some papers have previously been published classifying allosteric modulators as a separate class of ligands in general , here it is argued that the physicochemical properties of the molecules depend equally on the target in question [11] , [18] . For example if the target is a signaling protein ( e . g . a G protein-coupled receptor ( GPCR ) ) which naturally signals in response to ligand binding , an allosteric modulator can induce , inhibit , increase , or decrease this signal while still allowing the natural ligand to bind to the receptor ( albeit with modified thermodynamic and kinetic parameters ) . In some cases the allosteric modulator can even prevent the natural ligand from binding through a conformational shift . Similarly , in the case of an enzyme , an allosteric modulator can increase , decrease , or block enzyme catalytic activity . In the case of proteins with multiple functions and active sites , categorizing ligands as allosteric versus orthosteric can be problematic . For example , Figure 1 shows cyclin-dependant kinase 2 ( CDK2 ) involved in cell cycle control and known to have multiple binding sites for which multiple inhibitor types exist [19] . Firstly , several inhibitors are known to inhibit the protein via the ATP binding site ( which is commonly referred to as orthosteric inhibition , type I inhibition ) . Hence both ligands in competition with the ATP-binding site and proteins in competition with the substrate to be phosphorylated could be deemed orthosteric but differ significantly in their physicochemical properties . However , in literature the latter group is also classified as allosteric inhibitors . Moreover , one naturally occurring inhibitor of CDK2 , cyclin-dependent kinase inhibitor 1B or cyclin-dependent kinase inhibitor p27 , binds to the complex of CDK2 – cyclin A and protrudes into the ATP binding site [20] . Conversely , several small molecule classes have been identified that inhibit protein kinases in an allosteric manner . Type II inhibitors occupy the nucleotide-binding pocket and extend into the allosteric pocket , stabilizing the enzymatic inactive conformation ( DFG out ) , whereas type III inhibitors bind and occupy an allosteric pocket . Additionally , there are type IV inhibitors that are covalent inhibitors targeting reactive proximal cysteine residues [19] . Finally , a fifth class ( Type V ) of inhibitors have also been discovered for CDK2 . These are non-ATP competitive , but the binding pocket has been shown to differ from that of known type II and III inhibitors [21] . The inhibitors have been shown to bind near the C-helix , which is involved in the interaction of CDK2 with cyclins A and E [22] , [23] . Binding of these ligands also disrupts the protein – protein interaction ( PPI ) between CDK2 and cyclin , confirming the potential of allosteric modulators to disrupt PPIs . Hence CDK2 is home to multiple binding sites to which multiple sets of ligands/substrates can bind in different ways . These ligands can be orthosteric ( ATP-competitive ) , substrate competitive peptidomimetic molecules ( non-competitive with regard to ATP , allosteric ) or non ATP-competitive small molecules ( allosteric ) , and can be further subdivided based on the mechanism of action . For these reasons , and because the approach here relies on retrieving allosteric papers , the term non-allosteric ( rather than orthosteric ) is used to describe other ligands binding to the same protein than those retrieved in the here described allosteric dataset . The definition of allosteric follows herein the target in question and general agreement in the literature; hence any observations are relative to this agreement in literature . There are a number of targets for which one can make similar distinctions with different forms of classification ( e . g . in the kinase case one can define the ATP-competitive ligands as allosteric and only define the peptidomimetic ligands to be orthosteric ) . However , as the current results are derived from and based on medicinal chemistry literature it is chosen to follow this literature . Please see Case study 4 for further details on applying the here described methods to Kinase targets . As discussed above , the differences in binding site properties relative to the substrate/agonist/antagonist site are potentially attractive for operational drug discovery reasons . Allosteric modulators can hit targets with natural ligands that are outside classic oral drug-like space ( e . g . class B GPCRs ) , or are difficult to hit with specificity with regard to paralogs ( e . g . class C GPRCs ) , or can even be used to distort protein-protein interactions [24]–[26] . In all of these cases allosteric modulators can allow modulation of these targets by small molecules using well-established medicinal chemistry and drug delivery strategies . Furthermore , allosteric modulators are interesting from a physiological viewpoint , as they provide a way to modulate natural regulation ( amplify a naturally regulated response ) rather than completely inhibit or continuously activate proteins . Orthosteric drugs activate or inhibit a protein in a dose dependent manner . Yet allosteric drugs can differ , while their concentration in the body is dose dependent , their effect can be dictated only by concentration but can also be dictated by concentration in combination with physiological signaling and feedback loops [15] . Finally , in GPCR signaling allosteric modulators have been shown to possess other advantages over orthosteric ligands due to functional selectivity displayed by these allosteric ligands . Functional selectivity is expected to lead to greater selectivity and safety of drugs targeting GPCRs [27] . However , there are also less favorable characteristics of allosteric modulators making them less suitable as drugs . By definition allosteric modulators inhibit non-competitively and often via a secondary binding pocket . Hence the shape and pharmacophoric properties of such a pocket are not necessarily as highly conserved across paralogs and orthologs , as a catalytic/substrate site would be . The former site will usually not be under the same selective evolutionary pressure for protein function as the latter [28] . In the case of viral inhibitors or any other systems where rapid genetic mutation and selection is possible ( e . g . anti-fungals , anti-bacterials and anti-cancer therapeutic areas ) , the use of allosteric modulators might lead to easier onset of resistance by point mutations . This is empirically the case of the non-nucleoside reverse transcriptase inhibitors ( NNRTI ) used in the treatment of infections with the Human Immunodeficiency Virus ( HIV ) . NNRTIs are well known for a quick onset of ( cross ) resistance [29] . Moreover , they are only effective on the HIV-1 subtype and not on the closely related HIV-2 subtype ( 61% identical when comparing HIV-1 strain M with HIV-2 strain A ) . In HIV-2 the allosteric pocket cannot be formed due to the presence of substitutions native to HIV-2 , which lead to NNRTI resistance in HIV-1 . Conversely , non-allosteric inhibitors are effective on both strains due to their similarity to the natural ligands [29] , [30] . Public resources like ChEMBL [31] , Pubchem [32] , BindingDB [33] , and Drugbank [34] have transformed many parts of drug discovery . The availability of the data enables new research into signaling processes and the ligand – target bioactivity space [35]–[37] . For example , computational models can be developed using existing compound structure and activity data , and used to predict potential activities for other compounds . Hence this data opens the door for new applications like in silico side effect prediction , personalized medicine and rational design of polypharmacological drugs [38]–[40] . However the presence of multiple binding sites and binding modes potentially confuses and frustrates model development and validation in cases where multiple binding sites exist . Consequently the ability to distinguish between mode of action and systematic characterization of these compounds could potentially prove invaluable in drug discovery . In this work a top down analysis of allosteric modulators in the ChEMBL database was applied . Sets of ligands from papers in ChEMBL-14 were classified as being either allosteric , or non-allosteric ( or presumed orthosteric ) based on keywords , which were identified in both title and/or abstract . From the resulting papers the primary target was identified and then the compounds associated with this target were retrieved . The resulting sets of ligands ( allosteric and non-allosteric ) are information dense ( containing annotated target information , bioactivity , and the source documents ) . This information is subsequently exploited to study the allosteric concept over all bioactivities in ChEMBL , but also on a per target basis . Finally trends describing the chemistry , targets and bioactivity of compounds annotated to be allosteric are extracted
Allosterism has been reported in the ChEMBL database since the first indexed papers in 1980 ( although the concept has been around in literature since the 1960's ) [12] , [13] . In total 987 unique documents were retrieved that together form the allosteric set ( after manual curation for the case studies this number rises to 1 , 002 ) . Likewise a non-allosteric set was retrieved , this set consisted of the documents that were not pulled in the first set and included the same restraints as applied to the allosteric set ( see Methods ) . Finally a balanced non-allosteric set was derived from the full non-allosteric set to better perform unbiased classification . This balanced set was more similar in raw size and target distribution to the allosteric set ( Table 1 ) . The allosteric records made up only a small fraction of the total records ( around 3–4% of the total , Figure 2 ) . However a trend was seen that the number of allosteric records have been increasing since the early 90's with a peak in 2009–2010 . Possibly this increase was caused by the recent focus on allosteric modulation of GPCRs [14] , [15] , [17] , [41] , [42] . While the total number of allosteric records in 2012 was lower , this was likely caused by the fact that ChEMBL-14 does not contain an entire years' worth of 2012 publications . The full datasets are available for download on www . gjpvanwesten . nl/allosterism or ftp . ebi . ac . uk/pub/databases/chembl/Allosterism as are lists of all identified allosteric and non-allosteric activity_ids in ChEMBL-14 . The next obvious question was: what targets are amenable to allosteric modulation ? This information could be useful in assessing the likelihood of finding an allosteric modulator for related targets , and can also be input to screening or assay strategies . Ideally this information leads to insights how theses targets differ from the targets preferentially interacting with non-allosteric modulators . Since allosteric modulators are sometimes a secondary approach when non-allosteric modulation is infeasible or impossible , the expectation would be that the target distribution is different . Recently , Li et al . published work where they studied targets that can be allosterically modulated [43] . Yet , their work was limited to targets with known crystal structures , and hence would suffer from a systematic bias to simpler globular proteins . Here this was taken a step further investigating all allosterically modulated literature targets that are retrieved from ChEMBL . The targets in ChEMBL are classified within a hierarchy , in which level 1 ( L1 ) denotes the protein type ( e . g . ‘Membrane Receptor’ or ‘Enzyme’ ) , L2 further narrows the protein family ( e . g . Class A GPCR known as ‘7TM1’ ) and so forth down to individual proteins ( supporting Figure S1 ) [31] . Distinct differences were identified in the distribution of target classes when the total number of bioactivity measurements retrieved per target class was considered ( L2 target distribution for both data sets , Figure 3 ) . While the major target classes known from medicinal chemistry literature were represented in both sets ( e . g . class A GPCRs and Proteases ) their distribution differed between sets , moreover there were major differences [4] . For instance class C GPCRs were enriched among the allosteric set , as were the Nuclear Receptors and the Ligand-Gated Ion Channels . For class C GPCRs it has traditionally been difficult to obtain selectivity using non-allosteric ligands as these ligands tend to be very small [41] . The tight structure-activity relationships observed , centered around very ligand efficient recognition of the natural effector ligand do not allow much opportunity for variation in the receptor sequence and consequently in synthetic ligands biding this site ( e . g . Metabotropic glutamate receptors , GABAB receptors , etc . ) . However it has previously been shown that selectivity can be obtained using allosteric modulation and this course of action has been pursued in the literature and was hence represented in the data set [17] , [41] , [44] . The overrepresentation of GPCRs was expected as it has previously been shown that GPCRs are targets typically readily accessible to allosteric modulation [16] , [42] , [45] . A similar plot has been created for the L1 target class , which can be found in the supporting information ( supporting Figure S2 ) . Similar to the target-based overview of allosteric versus non-allosteric compounds , the chemical properties of both classes of compounds were investigated to highlight differences ( Figure 4A ) . The two most important observations were that historically identified allosteric modulators tend to fall within a much more narrower range of molecular weight ( but are a subset of non-allosteric compounds rather than distinctly separated from non-allosteric compounds ) and secondly that allosteric modulators adhered slightly better to Lipinski's rule of 5 ( 75% versus 66% ) . Yet the important observation here was that the literature does not contain much information about allosteric modulators that are far from drug-like space . However , the relative scarcity of non drug-like allosteric modulators does not mean that these are not possible ( e . g . the peptidomimetic kinase inhibitors ) . A similar observation has been made by Wang et al . yet some examples of allosteric modulators outside drug-like space were retrieved here , contrary to their work [11] . One possible explanation for this lack of non-drug-like allosteric modulators could be based on the bioactivity statistics of allosteric modulators ( see below ) . The differences between allosteric modulators and non-allosteric modulators were further explored in Figure 4B where normalized activity was also included ( based on a negative log value of IC50 , EC50 , Ki and Kd values ) . Overlap was observed in the high affinity locations shared by allosteric and non-allosteric ligands in a scatter plot showing compound fractional polar surface area and molecular solubility . Yet non-allosteric compounds also showed high affinity at fractional polar surface and molecular solubility values outside the values preferred by the allosteric compounds . From these observations it was concluded that the allosteric modulators in literature form a more restricted range subset ( in the sense of physicochemical properties ) from the overall set of compounds . Combined , these results demonstrate that allosteric compounds are not distinct from non-allosteric compounds , however , given historical data , they appear to form a subset of the broad non-allosteric compounds ( or medicinal chemistry derived compounds ) . The results also showed that allosteric compounds on average had a larger similarity between allosteric sets binding different target classes than between non-allosteric compounds binding different target classes ( when considering physicochemical properties ) . The differences were further demonstrated using a case study where the chemical differences are relatively large between the two sets . As touched upon in the introduction , the desirability of allosteric modulators for a certain target is not only governed by physiological or pharmaceutical demands . There are cases where orthosteric modulation is not feasible for the development of orally active small molecule drugs . Example cases are the class B GPCRs for which the natural effectors are polypeptide ligands of typical length ranging 30 to 40 residues [25] , [46] . There are many functionally and genetically validated links to pathology for this target class , and a number of drugs are available ( some examples are iv/sc dosed - Calcitonin ( Miacalcin ) , Exendin-4 ( Exenatide ) , and PTH ( Forteo ) ) [25] , [47]–[49] . This target class was represented approximately equal in both the allosteric and non-allosteric data set ( 0 . 3% of the allosteric and 0 . 6% of the non-allosteric papers ) . While no large differences were apparent in the target distribution , the physicochemical properties of compounds annotated as allosteric modulators differed from those annotated as non-allosteric modulators . Figure 5 summarizes some of the findings for the class B GPCRs as retrieved from the data set . A figure with all 68 descriptors used ( supporting Table S1 ) is also available ( supporting Figure S3 ) . In addition all data is available in tab delimited text format on www . gjpvanwesten . nl/allosterism or ftp . ebi . ac . uk/pub/databases/chembl/Allosterism . Here a limited figure is displayed for reasons of clarity . Differences in physicochemical properties were found for allosteric and non-allosteric class B ligands ( Figure 5 ) . The non-allosteric ( peptide like ) ligands were very large ( Mwt range 334 Da to 3591 Da for 95% of the data ) whereas those ligands annotated to be allosteric modulators were ‘classical’ small molecules ( Mwt between 305 Da and 569 Da for 95% of the data ) . Hence , differences were observed in properties related to size like: the number of chains or the number of hydrogen bond acceptors . However , when corrected for the size of the ligands , the differences were less distinct ( e . g . carbon fraction of the total atoms ) . Interestingly the allosteric ligands were more rigid as indicated by a higher sp2 hybridized carbon fraction , lower sp3 hybridized carbon fraction , higher aromatic bonds fraction , and higher rigidity index ( see methods for a further explanation of the rigidity index ) . Allosteric ligands tended to pass the Lipinski rule of five ( 60% ) and were more drug-like , whereas non-allosteric ligands were less prone to pass Lipinski's rule ( 30% ) and were not drug-like ( Figure 5 ) . Finally , the average formal charge for allosteric ligands was slightly negative and slightly positive for non-allosteric ligands . Similar charts have been created for all other significantly populated target classes ( L2 ) and can be found on www . gjpvanwesten . nl/allosterism or ftp . ebi . ac . uk/pub/databases/chembl/Allosterism . Secondary to physicochemical properties , substructures that are overrepresented in either the allosteric ligands or the non-allosteric ligands for a target class are of interest . Hence for each target class all present substructures ( using circular fingerprints FCFP_6 ) were retrieved and their frequency in the allosteric and non-allosteric sets were compared against the background of the combined ( full ) set . Substructures were then sorted based on the enrichment score ( supporting Table S2 , Table S3 , and Table S4 ) . The results were in correspondence with what would be expected considering the natural ligands for these receptors and the observations from Figures 4 and 5 . Substructures ranking high based on their allosteric score were quite specific , and tended to be aromatic . Conversely , substructures ranking very low based on their allosteric preference were small , frequently occurring and mainly introducing polarity . Interestingly , substructures scoring high based on their non-allosteric score included protein backbone like structures . The full set for all L2 target classes is available as a download from www . gjpvanwesten . nl/allosterism or ftp . ebi . ac . uk/pub/databases/chembl/Allosterism . Protein targets and chemical properties of ligands in the allosteric set and the non-allosteric set were the point of focus in the above text . Now the differences between the bioactivity of allosteric compounds and the bioactivity of non-allosteric compounds are summarized . Considered were: potency ( affinity ) , the number of targets that compounds from both groups have been tested on , the number of targets compounds from both groups were active on , the Ligand Efficiency ( LE ) [50] , and a number of other efficiency indices ( Binding Efficiency Index ( BEI ) , Surface Efficiency Index ( SEI ) , Normalized Surface Efficiency Index ( NSEI ) , etc . [51] , [52] ( Table 2 ) The median potency was lower for allosteric modulators than for non-allosteric modulators ( 5 . 96 log units versus 6 . 66 log units , p<0 . 01 ) . Moreover , a lower fraction of the compounds was considered ‘active’ ( 33% versus 47% ) , with activity being defined operationally as potency better than micromolar ( 6 log units ) or annotated ‘active’ in the source data . Likewise a higher fraction was inactive ( 39% versus 29% , less than 6 log units or annotated ‘inactive’ in the source data ) . Allosteric modulation is a process that cannot be explained by only ligand affinity ( the dynamics are much more complicated and the reader is referred to a number of reviews ) [53] , [54] , yet the current findings with regard to affinity are discussed here given the importance of this measurement in drug discovery . Several possible explanations for the observed differences can be considered . Firstly , it is known that metabolites can be allosteric regulators and these metabolites can be present locally at very high concentrations and can hence exert their effect with a relatively low potency . Table 2 could implicate that the data set reflects the presence of metabolites in our dataset , annotated as allosteric ligands . High concentration metabolites would not need micromolar affinity when they are present at a millimolar concentration locally [8] , [55] , [56] . Secondly , another explanation can be that the optimization of high affinity allosteric binders is more challenging given the more constrained chemical characteristics that allosteric modulators display compared to non-allosteric modulators . However , there are two more likely but also more complex potential explanations for the observed lower affinity as will be described below . A third explanation for the observed lower affinity could be derived from observations in the field of GPCRs . The current work is not the first to observe a lower affinity for allosteric modulators compared to non-allosteric interactions , in particular in the field of GPCRs this has been observed before [54] . While GPCRs are a complex modeling system given the baseline presence of both an orthosteric ( natural ligand ) and allosteric ( G protein ) binding site in all GPCRs , there are some observations that can perhaps be translated to a more general view of allosterism . It has been shown that allosteric interactions have a direct effect on the affinity of non-allosteric ligands ( orthosteric in GPCRs ) [54] . Given that affinity is defined as the ratio of ligand association to ligand disassociation rates , allosteric modulators directly affect the non-allosteric ( dis ) association rate . However , the allosteric interaction between two sites has been shown to be reciprocal [54] , hence the affinity of allosteric modulators is influenced by the affinity of non-allosteric modulators . As such the observation of the lower affinity of allosteric modulators might be a product of the dominant usage of radio-ligand binding assays ( as follows ) . Typically radio-ligand binding assays are set up using a well-known ligand , a radioactive molecule is synthesized based on this ligand and the binding of uncharacterized molecules is explored through their effect on the radio-ligand . Given that the radio-ligand is usually a well-known ligand , it is often a ligand with a reasonably high affinity . Hence this high affinity effect might influence the observed affinity of allosteric ligands due to the reciprocal nature between the binding site of an allosteric ligand and a non-allosteric ligand . When comparing competitive inhibition between two non-allosteric ligands ( radio ligand and unknown molecule ) this effect will likely not be present . While this explanation is funded on observations from the field of GPCRs , it should be noted that in this field allosteric modulation has arguable been the most intensely explored . Another observation from the field of GPCRs is that ligand efficacy does not necessarily correlate to ligand affinity . There is documented evidence in literature wherein the ligand with the best affinity does not display the best efficacy [57] , [58] . It has been hypothesized that this discrepancy can partially be explained through the concept of binding kinetics . For a number of GPCRs it has been found that efficacy is better explained when receptor residence time or disassociation rate is considered ( the most efficacious ligands are shown to be the ligands with longer residence time ) than when only affinity is considered [57]–[59] . In the case of allosteric modulators a similar principle might apply . Indeed , cases in which allosteric modulators modify binding kinetics of non-allosteric ligands have been described in literature [60] , [61] . Given that we observe here that allosteric modulators tend to be relatively small and lipophilic molecules one can expect de-solvation to play a major role in binding kinetics . Hence these molecules might display a baseline longer residence time than non-allosteric molecules due to their physicochemical properties . However , further research and experimental evidence is required to confirm or reject this hypothesis . While any classification into ‘active’ or ‘inactive’ is based on a cut-off , the observations here regarding affinity illustrate a larger issue . In screening efforts cut-offs are important to retrieve interesting ligands . If the median potency of allosteric modulators is lower than that of non-allosteric modulators ( corroborated by the tendency of allosteric ligands to be smaller , to be more lipophilic , and to possess less hydrogen bonding potential ) this could very well lead to possible allosteric modulators being missed in screening efforts . The general threshold for activity in primary screening is 10 µM to find compounds that are shown to have a median activity of 6 . 66 log units in ChEMBL . Hence , the implication would be that any screening effort for allosteric modulators ( median activity of 5 . 96 log units ) would need to be more sensitive or at least have the definition of ‘active’ adapted to conform to our observations . Moreover , given the reciprocal nature of the effects that allosteric and non-allosteric sites have on each other , it would be recommended to not use a single radio-ligand if one is aiming to find new allosteric modulators . A better choice is to use a spectrum of assays with different radio labeled ligands as has also been suggested by May et al . [54] . That said , allosteric compounds were found to have similar but slightly higher median binding efficiency indices ( LE , BEI , SEI , NSEI ) , this difference was likely caused by the fact that allosteric modulators tend to be smaller than non-allosteric modulators . This potentially indicates on average smaller , less polar binding sites for allosteric versus non-allosteric classes [43] . Moreover , we observed that allosteric modulators tend to have been annotated to a lower number of targets ( 2 versus 3 ) but this difference is marginal . Additionally , the median number of targets a compound is active on is shown to be 1 ( average 1 . 43 ) for the non-allosteric set , in line with the findings of Hu and Bajorath [62] , but the values are median 0 ( average 1 . 40 ) for the allosteric set . In conclusion , allosteric modulators were found to be able to modulate targets with low affinity but high efficiency . In addition , the data did not show allosteric modulators to be inherently promiscuous binders – at least as inferable from the distribution of assays reported in ChEMBL – , rather there was a trend for allosteric compounds to be less promiscuous than non-allosteric modulators , which is also seen in previous work [43] . While the potential of the current data set is demonstrated by comparing the allosteric and non-allosteric set , this analysis is by no means exhaustive . Similar analyses can be performed comparing different allosteric sets or for instance comparing class C GPCR ligands from the allosteric set with the class A GPCRs of the non-allosteric set ( comparing two different sets of trans-membrane domain binding ligands ) . Moreover , it should be noted that further research is required to determine if the lower binding affinity observed results from database bias or if this is an intrinsic property of allosteric modulators ( and if so , what the cause is of this observation ) . Above it was shown that there are chemical differences between allosteric ligands for a certain target class and non-allosteric ligands for that same target class . In some cases these differences were large ( as in the case of class B GPCRs ) whereas in other cases the differences appeared to be smaller ( as in the case of class A GPCRs ) . These chemical distinctions were used to train a classification model that would be able to predict if a compound would likely be an allosteric modulator or a non-allosteric modulator for a given target based on the physicochemical properties . These models were created on the balanced set to avoid a large bias in classifier predictions ( Table 1 ) . Non-balanced models have also been trained and data is available in the supplementary information . The use of ( circular ) fingerprints in the full ( non-target specific ) models was sidestepped for several reasons; firstly these models should have a large applicability domain and should hence not be limited to certain chemical motifs . Secondly , ( chemical ) sampling bias of specific historical target classes was to be avoided . Thirdly , the large chemical diversity would probably make those features that are predictive very generic ( as shown in the class B GPCR case study for substructures negatively associated with allosteric modulators , supporting Table S2 , S3 , S4 ) . Finally the improvement of circular fingerprints to the models was marginal ( on average 5% as calculated by the average of the used parameters , supporting Table S5 ) . Hence circular fingerprints were only used in more congeneric chemical sets ( e . g . target specific ) [63] . Models were judged by recall of allosteric modulators ( Sensitivity ( sens ) ) ; recall of non-allosteric modulators ( Specificity ( spec ) ) ; precision for allosteric modulators ( Positive predictive value ( PPV ) ) ; precision for non-allosteric modulators ( Negative predictive value ( NPV ) ) ; and Matthews correlation coefficient ( MCC ) . These were all 0 for a non-predictive/random model and 1 for an ideal model with the MCC also potentially being -1 for an ideal inverse model ( see Methods for further details ) . Table 3 shows a selection of the results for allosteric classification models ( each trained on 70% of the data and externally validated on the remaining 30% ) . For the full table see supporting Table S6 , here we limited ourselves to a single page for reasons of clarity . Different models on data sets grouped by class L0 ( protein binding compounds ) , L1 ( first level classification ) , and L2 ( second level classification ) have been trained . Figure 6 shows the out-of-bag ROC curve and external validation for the L0 model . For all groups models were able to classify a compound as allosteric modulator or non-allosteric modulator of a given target class with good accuracy , yet model performance improved when sets became more specific ( limited to a target class ) . These models provide a useful tool for the elucidation of the mechanism of action for compounds identified in primary HTS screening efforts . Second to being able to predict if a compound will or will not be an allosteric modulator , it is also of interest to find out what properties are important to make this distinction . Given in Table 3 are the three most important properties that were correlated with the ‘allosteric’ class and the three most important properties that were correlated with the ‘non-allosteric’ class for each classification model . These properties allow the further investigation into what differentiates allosteric from non-allosteric compounds . While in most cases allosteric modulators were more lipophilic and non-allosteric compounds were associated with a higher polar surface area this was not always the case . Examples were the Transient Receptor Potential Channels ( TRP ) and Voltage Gated Ion Channels ( VGC ) target classes ( L2 target class , ion channels ) , part of the Ion Channel ( L1 target class ) . Here allosteric ligands had a larger polar surface area ( TRP ) or larger polar solvent accessible surface area ( SASA ) ( VGC ) . Conversely non-allosteric ligands were more rigid ( TRP ) . No explanation for this observation is currently available but possibly , in the case of these two ion channels , the uncompetitive binders could bind near the ion channel itself and hence resemble these ions that are transported by these proteins rather than resembling the natural regulators ( which is Voltage in the case of VGC and can be diverse in the case of TRP ) . Note that this observation was absent for the Ligand Gated Ion Channels ( LGIC ) where the allosteric modulators seem to correspond more to what we observe in other protein classes ( double bonds are favorable and solubility/positive atom fraction are not favorable ) . For the full table containing the results of all classification models trained on all targets in levels 0–2 ( including class ‘undefined’ models ) see supporting Table S6 . In the final section the potential of the data set is demonstrated using three further different case studies . To illustrate possible applications of the data set , the classification models were applied to a number of previously studied targets for which a range of allosteric inhibitors has been published . The first of these targets is the viral enzyme HIV-1 reverse transcriptase ( HIV-RT ) , for which substantial SAR data and several approved drugs are well established [64] , [65] . A relevant drug target in the treatment of HIV , this target will fall into ‘Enzyme’ L1 target class and is not further defined on lower target class levels due to the sparseness of other related proteins in version 14 of ChEMBL . Importantly , both allosteric and non-allosteric drugs have been successfully developed as therapeutics , and many co-crystal structures reported clarifying the binding sites of various compound classes , making this an ideal target case . Furthermore rational design and random screening have been used to extensively study the protein . Before training models the molecules were clustered based on FCFP_6 fingerprints . As can be expected there were some misclassifications in the dataset . Known allosteric compounds were in the non-allosteric training set ( sharing scaffolds with known allosteric inhibitors ) . Moreover , a number of compounds in the allosteric training set were noted to be non-allosteric compounds ( nucleotide like structures ) and vice versa . Capturing this unannotated , or tacit knowledge within a field is challenging , and highlights some issues with data-mining the literature where ad hoc vocabularies and conventions are used; however , it also highlights the opportunity and added value for further curation . The clusters containing these compounds were reclassified based on the information in the original publications and subsequently a model was trained ( Table 4 ) . The model performed well with a sensitivity of 0 . 89 , specificity of 0 . 88 , PPV of 0 . 92 , NPV of 0 . 84 and MCC of 0 . 76 and was hence interpreted ( Figure 7 ) . The HIV-RT allosterism model showed the three most important descriptors for non-allosteric compounds to be fraction of Oxygen atoms as a part of all atoms ( for instance the presence of a ribose moiety or a number of phospho groups contributes to this descriptor ) , a larger polar surface area and a larger fraction of atoms that are H-bond acceptors . Conversely , the following parameters were found to be predictive for allosteric ligands: a larger fraction of the bonds should be aromatic , the fraction of bonds that are ring-bonds should be higher and the distribution coefficient ( LogD ) should be higher ( for a top 20 list see supporting Table S7 ) . These results demonstrate that the here-published data set is a suitable starting point to create a model that can differentiate between likely non-allosteric and likely allosteric ligands for a specific target . However , after further data set curation this approach can lead to a well performing model that can reliably differentiate between these classes . This approach to developing a predictive method for allosterism is however not limited to enzymes as is shown in the following examples . Like HIV-RT , the class A GPCR adenosine receptors form a highly validated and important drug target , where both agonists and antagonists have a therapeutic potential . Moreover , there is now structural data for this GPCR target . Adenosine receptors are relevant targets in the treatment of diabetes and Parkinson's disease [66] . Allosteric modulation of the adenosine receptors has anticipated advantages over orthosteric modulation as it is expected to increase tissue specific selectivity and enable modulation of receptors present in the brain [66] . Moreover , class A GPCRs make up a large fraction of the targets present in ChEMBL . This is due to their high relative tractability , the historical research effort on this class , the large size ( ca . 300 family members in the human genome ) , and linkage to many important diseases [4] . However , unlike HIV-RT no allosteric modulators of adenosine receptors have yet been launched as drugs . One compound , T-62 , was under evaluation for the treatment of chronic pain but crashed out in phase 2 trials [67] . Moreover , there is a preclinical body of work that demonstrates allosteric modulation for these drug targets and hence they were chosen to be included here as a case study . Different from the HIV-RT case study is that here a group of closely related proteins is used rather than a single target . Hence it is shown that the current data set can also be used to capture properties that distinguish allosteric modulators for a family of targets . Again some manual curation was needed before moving to model training . The main finding was the paper by Narlawar et al . [68] . This paper describes bitopic ligands that possess both allosteric and non-allosteric domains . The compounds were marked as allosteric due to the keywords noted in the abstract , yet the large non-allosteric part of the ligands ( including a ribose moiety ) deteriorates model performance . Similarly a number of ligands described by Jacobson et al . were included in the allosteric set as the abstract mentions that only some compounds appeared to bind at an allosteric site , yet the majority of the 78 compounds were non-allosteric , hence these were also cleaned [69] . The adenosine receptor allosteric modulator model performed well ( sens 0 . 94; spec 0 . 97 , PPV 0 . 66; NPV 1 . 00; and MCC 0 . 77; Figure 7 ) , although the lower PPV lead us to believe further curation might improve model performance . The model was then interpreted . Allosteric ligands had a higher fraction of aromatic bonds , a higher LogD , and a higher average bond length compared to non-allosteric ligands . Whereas non-allosteric ligands had a higher heteroatom fraction and a larger polar surface area compared to allosteric ligands . Yet there was an interesting distinction with the HIV-RT models . The structures of known adenosine ligands ( both allosteric and non-allosteric ) are much more conserved than those of HIV-RT ligands . Hence structural features ( in this case FCFP_6 substructures ) were much more important in model creation compared to generic physicochemical properties ( for example a xanthine scaffold was found to be correlated with non-allosteric modulators , supporting Table S8 ) . Three substructures were shown to have high importance values in model creation ( meaning that model quality significantly decreased by leaving them out of the descriptor set ) . A fourth and final case study presented in this paper is protein Kinase-B ( PKB ) /Akt 1 . This enzyme target is relevant in oncology as it plays an important role in cellular survival pathways by inhibiting apoptotic processes [70] , [71] . PKB differs from the previous targets as two different classes of allosteric modulators have appeared in the literature . As touched upon in the introduction , allosteric modulators of kinases can be small molecules that act for instance by shifting the balance of protein dynamics ( e . g . locking a protein in an inactive conformation ) . However in the case of kinases where orthosteric modulators are defined as ATP-competitive , allosteric modulators can also be compounds that resemble the substrate of the kinase and hence be peptides ( protein like compounds ) . In the current case study the allosteric modulators hence make up two major classes , one of which are large peptide like compounds . As such Protein Kinase B is an interesting target that forms the inverse of the class B GPCRs mentioned above . The non-allosteric modulators in this case were all ATP-competitive and it was hypothesized that this class forms a group that is more similar chemically than the allosteric modulators . Given the clear distinction between allosteric modulators that are peptidomimetic and small molecule allosteric modulators , the chosen course of action was to train the model using a three-class model rather than a binary classification model . The model had good predictivity ( sens 0 . 96; spec 0 . 94; PPV 0 . 71; NPV 0 . 99; and MCC 0 . 86; Figure 7 ) ; the added third class , ‘allosteric biological’ , was predicted very well with recall 1 . 00 and predictive value 1 . 00 . As expected , properties mostly related to size ( Molecular polar surface area , volume ) were correlated with the biological allosteric modulators as is the ChEMBL calculated molecular class ‘biological’ . The physicochemical properties mostly correlated with small molecule allosteric modulators were number of chain assemblies , ringbond fraction , carbon fraction , and number of sp2 hybridized carbons . Additionally the ChEMBL calculated molecular class ‘small molecule’ was correlated to small molecule allosteric modulators . Interestingly , properties Lipinski pass , aromatic bonds frac , ringbonds frac , and LogD were also correlating with non-allosteric modulators ( contrary to the trends observed in other targets ) . This is likely due to the fact that ‘small molecule allosteric modulators’ and ‘small molecule ATP competitive modulators’ more closely resemble each other than they do the ‘biological allosteric modulators’ in terms of physicochemical properties . Moreover the non-allosteric/ATP-competitive set contained a number of drugs , which are highly optimized structures . Yet , LogD , and ringbonds fraction correlated to both the allosteric and non-allosteric small molecule classes . Conversely , negative atom fraction and number of hydrogen bond acceptors were correlated with only non-allosteric compounds ( likely due to the need for ATP-competitive compounds to also resemble parts of ATP ) , but this effect was less pronounced . Also in this case study ( similar to HIV RT ) sub-structural features were observed to be very important . Moreover , in the biological allosteric modulators class protein/peptide backbone fragments were appearing as important in combination with charged arginine side chains . Inversely , in the case of small molecule allosteric modulators the important substructures mostly contain aromatic rings . For a longer list see supporting Table S9 . In the case studies the potential of the data set identified and provided in this paper is demonstrated . The dataset is shown to be a solid starting point for allosteric focused drug discovery towards existing targets or towards new targets . With modest further curation highly predictive models could be obtained . While it is outside the scope of this paper to provide a case study on all potentially interesting protein targets , possible other examples included in the set are ( but not limited to ) : Kinesin EG5 [72]–[74] , Alcohol dehydrogenases ( e . g . Isocitrate dehydrogenase 1 and 2 ( ICDH ) ) [75] , and class C GPCRs [26] . The models obtained here trained on the full allosteric modulator set should have a broad domain of applicability due to their generic nature ( physicochemical properties were used as descriptors ) . Hence it is expected that these models are not limited to certain known chemical motifs as would be the case when using circular fingerprints . While also outside the scope of the current paper , the authors would very much welcome a prospective validation of the models . It should be noted that these models are solely classifying between ‘a likely allosteric interaction’ and ‘a likely non-allosteric interaction’ . Hence the models cannot be used to predict the affinity of ligands on certain targets , but are able to predict the likely type of interaction for a given interaction . As such these models should ideally be combined with dedicated bioactivity models that can predict the affinity of molecules on a certain target and not replace them . Hence the allosteric classifiers can be used as a secondary filter when selecting compounds from a chemical vendor to be tested experimentally . The authors feel that other potential applications could be the following: Firstly , creation of allosteric focused libraries based on known chemical properties of allosteric modulators , these libraries can be further sub divided on target type ( e . g . Class A GPCR or Protein Kinase ) . Secondly , determination of interaction type of hits retrieved from HTS screening ( allosteric or non-allosteric ) . The authors are very open for potential collaborative projects to experimentally validate the approach as described here . Hence the authors would urge readers to contact them when they are interested in a specific set of allosteric modulators . As stated in the introduction , the term allosteric modulator is a very broad definition directly depending on the target ( class ) in question . Despite the presence of peptidic ligands and very diverse chemistry , there are some general conclusions that can be drawn from the current work . Allosteric modulators tend to be more rigid and lipophilic structures compared to the background set . This is in line with their mode of action via binding in distinct structural locations of proteins rather than catalytic or agonist sites . Yet the magnitude of these changes in physicochemical properties depends on the target in question and the non-allosteric ligands . Moreover , it is observed that allosteric modulators are constrained to a narrower structure activity window than are non-allosteric modulators . When the physicochemical properties of allosteric modulators are compared to all ligands for a target , the allosteric modulators are often a subset of the non-allosteric ligands . Secondly , it is observed that allosteric modulators are interesting drugs for several reasons . They tend to adhere better to Lipinski's rule of 5 , making them good candidates for oral formulation . This could indicate that , if allosteric hits are identified for a target , allosteric ligands are more developable then non-allosteric ligands . Moreover , a trend is observed that allosteric modulators are less promiscuous than non-allosteric modulators . Thirdly , the absolute potency for allosteric modulators is observed to be lower , while their binding efficiency and surface ligand efficiency is similar . Some potential causes are discussed here , but before a qualitative statement can be made about this observation further research is required . However this observation does call for the adaptation of screening assays to pick up the lower affinity compounds . In conclusion , the differences between non-allosteric and allosteric modulators for a given target are usually such that it is not straightforward to turn a non-allosteric compound into an allosteric compound or vice versa . Yet it is these chemical differences that allow the creation of classification models that can distinguish between allosteric and non-allosteric modulators . These models are shown to perform better if the target definition is more concise , yet even without these constraints already predictive models were constructed . Hence non-allosteric and allosteric inhibition of a single target can be considered different target classes overall . The work performed here should lead to improvement of bioactivity models by providing tools to incorporate binding mode as a descriptor for compounds and hence reducing the noise present in a data set . While the authors have demonstrated in the current paper how the dataset can be used as a starting point for allosteric drug design , full manual curation of the dataset is at the moment infeasible . Hence the authors encourage everybody who encounters an error or misclassification in this data set to contact them so that curation can take place via crowdsourcing and the quality of this data in ChEMBL can increase .
The data set was obtained from ChEMBL version 14 [31] . For the allosteric set , abstracts and titles of journal articles were searched for keywords ( supporting Table S10 ) . For hits both PubMed ID and citation information ( primary author , year , journal , volume , and starting page ) were kept . From these retrieved records the primary target ( based on bioactivity annotation frequency for targets considered in the document ) was included along with all compounds annotated on this primary target . As a final step duplicate compounds were removed for each target ID . Herein a distinction was made in the quality of the bioactivity measurement , best measurements ( e . g . pKi ) were favored over lower quality measurements ( e . g . activity comment ‘active’ ) . The background set was retrieved in a similar fashion , but here all document IDs that were not part of the allosteric set were kept . Finally , the balanced non-allosteric set was retrieved from the full non-allosteric set by keeping a random percentage of bioactivities from each L2 target class which was roughly equal in size to the number of bioactivities present in the allosteric set . All data is available on www . gjpvanwesten . nl/allosterism or ftp . ebi . ac . uk/pub/databases/chembl/Allosterism , see supporting Figure S4 for details . Compounds were standardized , charged at a pH of 7 . 4 , salts were removed and 2D and 3D coordinates were calculated . All of this was done in Molsoft ICM version 3 . 7-2d [76] . Volume , Polar Surface Area , Molecular weight , and drugLikeness were calculated in Molsoft ICM , carbon hybridization states were calculated using the Perl molecular toolkit in Pipeline Pilot [76] , [77] . For partition coefficient ( LogP ) calculations it has been shown that consensus methods perform well [78] , hence the used LogP value was the average of AlogP calculated in Pipeline Pilot , logP according to Molsoft ICM and ACD LogP [76] , [77] , [79] . Similarly LogD was the average of the Pipeline pilot module and ACD LogD , finally solubility was the average of the pipeline pilot calculator and Molsoft ICM value . The remaining compound physicochemical descriptors were calculated in Pipeline Pilot using the chemistry component collection [77] . The Lipinski Pass/Fail class was calculated allowing no violations . For the individual case studies additional FCFP_6 descriptors were used , on these Bayesian feature selection from Pipeline Pilot was applied to transfer them into a 512 bits fixed bitstring [77] , [80] . Finally , the rigidity index was an estimation of compound rigidity that was calculated as follows: ( AromaticBonds fraction ) + ( 1-RotatableBonds fraction ) +Aliphatic Ringbonds fraction+ ( 1-SingleBonds fraction ) +DoubleBonds fraction+TripleBonds fraction+BridgeBonds fraction ) /7; Target information from ChEMBL ( Uniprot ID , target classification ) was kept as it was defined in ChEMBL . However , when target classification levels were unpopulated the value was replaced with ‘Undefined’ . Models were trained in Pipeline Pilot using the ‘Random Forest’ component . This component uses R-Statistics ( version 2 . 15 . 0 ) and the ‘forest’ package [81] , [82] . For variable importance selection permutation based selection and Gini importance without scaling were used , as recommended by Strobl et al . [83] , [84] . Important variables were selected based on Pareto optimization of both importance values and class correlation values ( e . g . correlation with ‘Allosteric’ class ) . Validation was performed using 5 different metrics these were: sensitivity ( allosteric recall , the fraction of true positives of the total number of allosteric compounds ) , specificity ( non-allosteric recall , the fraction of true negatives of the total number of non-allosteric compounds ) , positive predictive value ( allosteric precision , the fraction of true positives of the total number of compounds predicted to be allosteric modulators ) , negative predictive value ( non-allosteric precision , the fraction of true negatives of the total number of compounds predicted to be non-allosteric modulators ) , and the Matthews correlation coefficient [85] . Given a confusion matrix were A represents an allosteric modulator classification and B represents non-allosteric modulator classification , Sensitivity is class A recall , and specificity is class B recall , whereas positive predictive value is class A precision and negative predictive value is class B precision ( Table 5 ) . For the MCC equation ( 1 ) was used; herein the numerator is the product of the correctly predicted data points minus the product of the incorrectly predicted data points . The denominator is formed by the square root ( 2-classes ) of the total product of all possible sums of correct and incorrectly predicted data points . ( 1 ) Note that false negatives are missed class A predictions and false positives are missed class B predictions . Hence this can be rewritten as follows: Numerator: ( 2 ) Denominator: ( 3 ) In the case of the three-class model ( ternary classification ) these calculations were modified to represent the three-class confusion matrix . Assume class A to be allosteric modulators , class B to be non-allosteric modulators and class C to be biological allosteric modulators . Sensitivity remains the fraction of true positives of the total number of allosteric compounds ( here class A recall ) , specificity remains the fraction of true negatives of the total number of non-allosteric compounds ( here class B recall ) , positive predictive value remains the fraction of true positive of the total number of compounds predicted to be allosteric modulators ( here class A precision ) , and negative predictive value remains the fraction of true negatives of the total number of compounds predicted to be non-allosteric modulators ( class B precision ) . Additionally a class C recall ( the fraction of true allosteric-biological predictions of the total number of allosteric-biological compounds ) and precision ( the fraction of true allosteric-biologicals of the total number of compounds predicted to be allosteric biologicals ) are introduced . It should also be noted that the baseline values for a random model in a ternary classification model are expected to be around 0 . 33 ( 33% correctly predicted compared to 66% incorrectly predicted ) . This is lower than the value of 0 . 50 ( 50% correct prediction and 50% incorrect prediction ) for a binary model . Hence values were scaled to be directly comparable between the two model types . Equation ( 1 ) was again used for the MCC but adapted to the ternary matrix ( Table 6 ) ; the product of the correctly predicted data points minus the product of the incorrectly predicted data points forms the numerator . The denominator is formed by the cube root ( 3-classes ) of the total product of all possible sums of correctly and incorrectly predicted data points . The following types are defined: AB+AC = AX ( Missed class A predictions ) BA+BC = BX ( Missed class B predictions ) CA+CB = CX ( Missed class C predictions ) Hence the MCC can be written as follows: Numerator: ( 4 ) Denominator: ( 5 ) ( 6 ) The MCC still produces values between 1 ( perfect prediction ) , 0 ( random prediction ) and −1 ( anti correlation ) and need not be scaled , contrary to the recall values and predictive values as the full confusion matrix is considered in absolute numbers when calculating the MCC . Substructures were obtained using pharmacophore feature class based circular fingerprints ( FCFP_6 ) [63] , [80] . For all present substructures , substructure frequencies were obtained from the full data set ( background frequency ) , the allosteric set per L2 target ( allosteric frequency ) , and the non-allosteric set per L2 target ( non-allosteric frequency ) . These frequencies were normalized per set ( substructure frequency as a fraction of the total substructures per set ) to prevent a biased ranking . Subsequently all substructures were ranked based on their normalized frequency . Enrichment was calculated based on the logarithm of the normalized ranks quotient ( between allosteric and background or between non-allosteric and background ) . These final scores were ranked to obtain the final scored rank . 4 supporting figures ( Figure S1 , S2 , S3 , S4 ) and 10 supporting tables ( Table S1 , S2 , S3 , S4 , S5 , S6 , S7 , S8 , S9 , S10 ) that further support the findings are available online . In addition , the datasets , further chemical analyses ( per target level ) , physicochemical property histograms ( for L0 , L1 , and L2 ) , all model training and validation reports , and delimited text files are available online: www . gjpvanwesten . nl/allosterism or ftp . ebi . ac . uk/pub/databases/chembl/Allosterism .
|
The physicochemistry and topography of ligand binding sites is generally conserved amongst related proteins , however , comparisons of the pharmacology of related targets ( and even the same target ) are often confounded by the existence of multiple , distinct , binding sites within the same protein . Importantly , these multiple binding sites can have ‘druggability’ or selectivity properties , and can therefore offer attractive novel approaches to develop new therapeutic agents . In this paper , sets of known ligands binding to the same target are classified as being either allosteric ( binding at a site that is non-competitive for a natural ligand/substrate ) or non-allosteric ( binding at the same site as a natural substrate ) , it is demonstrated that there are differences in the profiles of ligands discovered empirically against these sites . Finally predictive models are developed with several useful applications in drug discovery .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Methods"
] |
[
"biochemistry",
"computer",
"and",
"information",
"sciences",
"medicinal",
"chemistry",
"computer",
"modeling",
"biology",
"and",
"life",
"sciences",
"chemistry",
"physical",
"sciences",
"chemical",
"biology"
] |
2014
|
Chemical, Target, and Bioactive Properties of Allosteric Modulation
|
Aedes aegypti is a cosmopolite mosquito , vector of arboviruses . The worldwide studies of its insecticide resistance have demonstrated a strong loss of susceptibility to pyrethroids , the major class of insecticide used for vector control . French overseas territories such as French Guiana ( South America ) , Guadeloupe islands ( Lesser Antilles ) as well as New Caledonia ( Pacific Ocean ) , have encountered such resistance . We initiated a research program on the pyrethroid resistance in French Guiana , Guadeloupe and New Caledonia . Aedes aegypti populations were tested for their deltamethrin resistance level then screened by an improved microarray developed to specifically study metabolic resistance mechanisms . Cytochrome P450 genes were implicated in conferring resistance . CYP6BB2 , CYP6M11 , CYP6N12 , CYP9J9 , CYP9J10 and CCE3 genes were upregulated in the resistant populations and were common to other populations at a regional scale . The implication of these genes in resistance phenomenon is therefore strongly suggested . Other genes from detoxification pathways were also differentially regulated . Screening for target site mutations on the voltage-gated sodium channel gene demonstrated the presence of I1016 and C1534 . This study highlighted the presence of a common set of differentially up-regulated detoxifying genes , mainly cytochrome P450 genes in all three populations . GUA and GUY populations shared a higher number of those genes compared to CAL . Two kdr mutations well known to be associated to pyrethroid resistance were also detected in those two populations but not in CAL . Different selective pressures and genetic backgrounds can explain such differences . These results are also compared with those obtained from other parts of the world and are discussed in the context of integrative research on vector competence .
Aedes ( Stegomyia ) aegypti ( Linnaeus , 1762 ) is a mosquito species of high medical importance due to its widespread distribution and ability to transmit a variety of arboviruses . For decades its control has involved mechanical elimination of breeding sites as well as larvicidal applications and adulticide spatial spraying operations . However , the efficacy of these insecticide treatments has been reduced due to the development of resistance in this species . French overseas territories such as French Guiana , Martinique and Guadeloupe ( French Territories in the Americas , FTAs ) , and New Caledonia ( West Pacific ) have all experienced insecticide resistance in Ae . aegypti populations [1–4] over the course of vector control programmatic changes . Since the 1940s , all the territories that once used organochlorine ( OC ) , organosphosphate ( OP ) , pyrethroids ( PY ) and bioinsecticides insecticides successfully have observed the development of vector resistance to the majority of them with the exception of bio-insecticides . Since the prohibition of the sale and use of many biocide products by the European Community ( EC ) , the FTAs are facing a dilemma in their vector control strategies . Despite the fact that pyrethroids have the highest level of resistance in Ae . aegypti , they are the only insecticide family authorised for mosquito control . Vector control strategies depend on the decisions of local authorities , which differ from one territory to another . The French Guianan program of surveillance and management of dengue cases conducts Ae . aegypti density reduction throughout the year which is intensified during outbreaks . Vector control activities include both indoor and outdoor spatial spraying of deltamethrin ( PY ) against adults and the removal of breeding sites or their treatment with Bacillus thuringiensis var . israelensis ( Bti ) based larvicides . Deltamethrin is also used routinely for pest mosquito management . In contrast , the territory of Guadeloupe limits the use of insecticides to only during dengue and other arbovirosis epidemics and focuses on larval elimination during non-epidemic periods . In New Caledonia , where EC regulations do not apply , the local government conducts regular monitoring of insecticide resistance that has led to a switch from deltamethrin to malathion ( OP ) in zones where PY resistance was detected in 2003 . Adulticides are sprayed in a 100 meter-radius around dengue , Chikungunya or Zika fever cases , which is coupled with mosquito source reduction and community awareness raising campaigns . Pyrethroid resistance is explained by two main mechanisms . One is the alteration of the membrane voltage-gated sodium channel inducing insensitivity to the insecticide and absence of knock-down effect ( kdr ) . This mechanism confers cross-resistance to PYs and DDT [5] . The second recognized mechanism is an overproduction of detoxifying enzymes , also called metabolic resistance . These enzymes are naturally involved in the degradation or transformation of toxic compounds into non-toxic products before elimination or sequestration into the insect body . Three large enzyme families , the cytochrome P450 monooxygenases ( P450s ) , Glutathione S-transferases ( GSTs ) and carboxy/cholinesterases ( CCEs ) have been implicated in the metabolism of insecticides in insects . Throughout the species’ range , Ae . aegypti populations have shown strong resistance to pyrethroid , carbamate and organophosphate insecticides correlated with elevated activities of detoxifying enzymes and kdr mutations [6–11] . Monitoring insecticide resistance in mosquito populations and understanding the mechanisms involved is a prerequisite to efficient implementation of vector control strategies . Transcriptomic tools such as the “Aedes detox chip” [12] are useful to identify detoxification gene regulation in insecticide resistant populations . However , while some of these genes are related to insecticide resistance , others are a response to environmental and evolutionary factors . The study of several mosquito populations could identify a common pattern of genes associated with pyrethroid resistance and eliminate those associated with environmental factors . Therefore , the objective of this work is to obtain and compare data on pyrethroid resistance and the underlying resistance mechanisms in three Ae . aegypti populations collected in geographically distinct French overseas territories .
Mosquito blood feeding is done on mice . We hold experimental authorization under the agreement number B973-02-01 delivered by "la préfecture de Guyane" and renewed on June 6th , 2015 . Experimental project was re-approved by the ethical committee CETEA Institut Pasteur ( n° 89 ) , report number 2015–0010 issued on May 18th , 2015 . Aedes aegypti mosquitoes were collected at the larval stage in Cayenne ( French Guiana , GUY ) , Baie Mahault ( Guadeloupe , GUA ) and Noumea ( New Caledonia , CAL ) ( S1 Fig ) . Discarded freezers , tyres and plant pots were the targeted breeding sites . Adults of the F0 generation were reared at each partner facility and allowed to mate . Females were provided a blood meal from mice and eggs of the F1 generation were produced . These eggs were shipped to the Institute Pasteur in French Guiana . Adults of the F2 generation were then obtained by rearing immature stages from the three populations concurrently under insectary conditions in French Guiana ( 28 ± 1°C , 70 ± 10% RH , 12:12h photoperiod ) . New Orleans ( NO ) susceptible strain was reared under the same conditions at the same time and used as a reference in all experiments . Insecticide resistance bioassays were carried out once per population through tarsal contact tests using filter papers impregnated with technical grade deltamethrin ( CAS # 52918-63-5 ) ( Sigma-Aldrich , St Louis , MO , USA ) at the diagnostic dose of 0 . 06% as published in Jirakanjanakit et al . [13] . Filter papers were impregnated following the WHO protocol [14] using acetone solutions of insecticide and silicone oil as the carrier . Impregnation was conducted by dropping 2 mL of a solution containing technical grade chemical dissolved in acetone and silicone oil evenly onto each paper ( 12 x 15 cm ) . Concentrations were expressed in w/v percentage of the active ingredient in silicone oil . The papers were dried for 24 hours before the test . All the mosquito populations were exposed to the same batch of impregnated papers during one week . Following the WHO tube test protocol [15 , 16] , four batches of 25 non-blood fed females ( 2–5 days old ) were introduced into holding tubes for one hour then transferred into the exposure tube and placed vertically for one hour . Knocked-down and dead mosquitoes were recorded after this time ( 1h KD ) before being transferred back to the holding tubes . A mosquito was recorded as knocked-down if it was lying on its back or side and was unable to instigate flight after a gentle tap . Mortality was recorded 24 hours after exposure ( 24h M ) . Controls , made of only acetone and silicone oil delivered to the filter paper , were performed as mentioned above with a total of two batches of 25 non-blood fed females ( 2–5 days old ) per replicate . All replicates were conducted at 27 ± 2°C and a relative humidity of 60 ± 10% . A ten percent sugar solution on soaked cotton balls was provided to the females during the 24 hours observation period . Three day old females were exposed to 0 . 06% deltamethrin impregnated filter papers in the WHO tubes as described above . Up to 100 specimens surviving after 48 hours of observation were anesthetized on ice , dropped into RNAlater ( Qiagen , Redwood city , CA , USA ) and preserved following manufacturer instructions . Samples were shipped by express mail to the Institute Pasteur de Montevideo for microarray experiments . Other specimens were separated into surviving ( resistant ) , dead ( susceptible ) and non-exposed ( control ) mosquitoes and stored dry prior to kdr genotyping . DNAzol ( Life Technologies , Gaithersburg , MD , USA ) was used for total DNA isolation of resistant and susceptible individuals prior to sequencing the fragment encompassing exon 20 and exon 21 of the sodium channel voltage dependent gene ( Nav ) . Primers used to amplify this fragment were AaNaA 5’-ACAATGTGGATCGCTTCCC-3’and AaNaB 5’-TGGACAAAAGCAAGGCTAAG-3’ ( 8 ) . The primers amplify a fragment of approximately 472 bp which includes an intron of approximately 28 bp . For PCR reactions , 100 ng of DNA were used as a template , and samples were incubated for 3 min at 95°C followed by 30 amplification cycles of 30 sec at 94°C , 30 sec at 60°C and 1 min at 72°C , with a final extension of 5 min at 72°C . PCR products were loaded on a 4% gel made of low-melting temperature agarose ( NuSieve , Lonza , Rockland , ME , USA ) and migrated to separate fragments with introns of two sizes . Fragments were isolated , purified , directly sequenced , and sequences were aligned by using ClustalW [17] . Based on these sequences , we designed primers and probes designed to develop an Allelic discrimination assay using the Taqman technology ( Applied Biosystems , Foster City , CA , USA ) . The allelic discrimination assay was composed of two standard oligonucleotides V1016I SNP-F-GCT-AAC-CGA-CAA-ATT-GTT-TCC-C and V1016I SNP-R- CAG-CGA-GGA-TGA-ACC-GAA-AT . Each probe consists of a 5’ reporter dye , a 3’ non fluorescence quencher and a minor groove binder at the 3’end . The probe V1016-PV- CAC-AGG-TAC-TTA-ACC-TTT-T was labeled with 6-Vic dye fluorescence at the 5’ end for the detection of the wild-type allele whereas the probe I1016-PF-CAC-AGA-TAC-TTA-ACC-TTT-TC was labeled with FAM dye fluorescence at the 5’ end for the detection of the mutant allele . In addition , genotyping position 1534 on the domain III segment 6 was performed according to Yanola et al . [18] . The two standard oligonucleotides were F1534C SNP-F- GAT-GAT-GAC-ACC-GAT-GAA-CAG-ATC and F1534C SNP-R- CGA-GAC-CAA-CAT-CTA-GTA-CCT . Each probe consists of a 5’ reporter dye , a 3’ non fluorescence quencher and a minor groove binder at the 3’end . The probe F1534-PV- AAC-GAC-CCG-AAG-ATG-A was labeled with 6-Vic dye fluorescence at the 5’ end for the detection of the wild-type allele whereas the probe C1534-PF-ACG-ACC-CGC-AGA-TGA was labeled with FAM dye fluorescence at the 5’ end for the detection of the mutant allele . For the allelic discrimination assay , DNA from individual adult mosquitoes was extracted with the Purelink Genomic DNA extraction kit ( Invitrogen , Carlsbad , CA , USA ) according to the manufacturer's instructions , with the following modification: cell lysis was performed manually with a sterile piston after addition of PureLink Genomic Digestion Buffer . DNA was isolated and suspended in 100 μl of PureLink Genomic Elution Buffer ( from the kit ) and stored at -20°C . The TaqMan reaction contained 12 . 5 μL of 2X TaqMan Universal Master Mix II ( Life technologies , Gaithersburg , MD , USA ) , 1 . 44 μM of each primer , 0 . 4 μM of each probe and 3 μL of genomic DNA ( 20 ng ) made up to 25 μL with sterile water . The assay was performed under discrimination allele settings using an StepOnePlus Real-Time PCR System ( Life technologies , Gaithersburg , MD , USA ) , under the following thermocycling conditions: 10 min at 95°C , 45 cycles of 95°C for 15 sec and 60°C for 1 min . Data were analyzed by the StepOne Software version 2 . 1 . Relationships between phenotypes and putative resistant genotypes were tested using a Fisher’s test for each population with the MASS package in R 3 . 2 . 1 ( R Development Core Team , Vienna , Austria ) . Linkage disequilibrium was tested with GENEPOP v4 . 2 [19 , 20] .
One hundred females were exposed to deltamethrin 0 . 06% for one hour and left in observation for 24 hours . The percentage of knocked-down mosquitoes after one hour exposure were 29% , 37% and 100% for GUY , GUA and CAL populations respectively , demonstrating the presence of knocked-down resistance in GUY and GUA but none in CAL . Twenty-four-hour mortalities were then , respectively , 26% , 42% and 96% . Thus , All three populations were resistant to deltamethrin according to the WHO criteria [15] . GUY and GUA populations can be characterized as highly resistant whereas CAL has a lower level of resistance . A total of 58 sequences from 32 individuals were analysed ( Table 1 ) . Haplotypes A and B were observed as characterized by Martins et al . [8] in each population including the NO reference strain . Only the GUA population has an evident unbalanced proportion with a frequency 0 . 92 for haplotypes A . In position 1016 of the sequence , valine , related to deltamethrin susceptibility , was the only amino-acid found in NO ( N = 6 ) , the susceptible reference strain , and in the CAL population ( N = 5 ) . Both valine and isoleucine , related to resistance , were found in GUY and GUA populations . We were unable to detect V1016G in our sequences . In addition , two mutation points described in Ae . aegypti were checked: S989P , described in mosquitoes in a deltamethrin selected strain from Thailand [22] and I1011M/V were described in Latin America [8 , 23] . In position 1011 of the amino-acid sequence , methionine was absent in the GUA population , found in GUY population at low frequency ( freqmet = 0 . 11 ) , and in equivalent proportion with isoleucine in the CAL population ( freqmet = 0 . 40 ) ( Table 1 ) . Even if several studies [24 , 25] have related I1011M/V mutation to resistance to cypermethrin or permethrin , others did not demonstrate such association with pyrethroid and particularly deltamethrin resistance [8 , 23 , 26] . However , amino-acid changes in position 1016 and 1534 are of higher interest for resistance phenotype , considered alone or in combination [8 , 26 , 27] . These are the reasons why we focused on genotyping V1016I and F1534C loci . In addition to the 26 individuals fully genotyped by sequencing , 146 individuals were genotyped for the 1016 codon using Allelic Discrimination Assay , making a total of 172 individuals genotyped for this locus . A total of 161 individuals were also genotyped for the 1534 codon . Each individual was classified as resistant or susceptible according to if they were alive or dead 48 hours after their exposure to deltamethrin ( see above ) . The CAL population exhibited only the wild-type ( homozygous susceptible ) genotype at the two loci regardless of the phenotypic class . This observation is in accordance with the absence of kd resistance observed in the bioassays . The resistant allele , 1016I , was recorded in higher proportion in GUY and GUA resistant mosquitoes , 92% and 88% respectively , than in susceptible ones , 65% and 79% , respectively . The allele 1534C was fixed in the GUY population regardless of the phenotype , while in GUA it exhibited frequencies of 88% in resistant mosquitoes and 77% in susceptible ones . Considering the resistance allele recessive [23 , 25 , 28] , the association between the phenotype and genotype was tested using Fisher’s test . Only V1016I in GUY population was significantly associated with resistant phenotypes ( p = 0 . 0018 ) . Significant linkage disequilibrium was observed between the two loci among all populations ( p<0 . 001 ) . Based of the description of Linss et al . [29] , four alleles ‘1016V + 1534F’ called S , ‘1016V + 1534C’ ( 1534 kdr ) called R1 , ‘1016I + 1534 C’ referred to as R2 ( 1016 kdr+1534 kdr ) and ‘1016I + 1534 F’ referred to as R3 ( 1016 kdr ) were identified . Allele frequencies are shown in S2 Table . According to the alleles , the genotypes were named SS , SR1 , R1R1 , R1R2 , R1R3 and R2R2 . Genotype distributions in each population were different . As described above , the CAL population was genotypically fully susceptible . In contrast , the R2R2 genotype was observed in both GUY and GUA populations . However , R1R2 and R1R1 were observed in the GUY population while a majority of SR1 and SS were observed in the GUA population . R1R2 and R2R3 were scarcely represented in the GUA population , which was more diversified than the two others ( Fig 1 ) . Differences in gene expression of the GUY , GUA and CAL pyrethroid-resistant strains and the NO susceptible strain were assessed using a 15K ‘Ae . aegypti detox chip plus’ microarray platform containing 3746 unique genes . Using an arbitrary cut-off of fold change >2-fold in either direction and a t-test P-value of less than 0 . 001 after multiple testing correction , 63 ( 1 . 3% ) genes were differentially transcribed between the GUY and NO strains ( 41 up regulated and 22 down regulated ) , 76 genes ( 1 . 5% ) were differentially transcribed between the GUA and NO strains ( 48 up regulated and 28 down regulated ) and 50 genes ( 1 . 5% ) were differentially transcribed between the CAL and NO strains ( 39 up regulated and 11 down regulated ) . Of the 123 ( 2 . 51% ) differentially regulated genes , 17 ( 2 . 5% ) were in the three resistant populations: 12 . 2% were up regulated and 2 . 4% were down-regulated . In addition , 23 ( 18 . 7% ) were common to GUY and GUA strains , 4 ( 3 . 3% ) were common to GUY and CAL strains , 5 ( 4 . 1% ) were common to GUA and CAL strains . Almost 16% of the differentially expressed genes are not annotated as ‘conserved hypothetical proteins’ or are specified as ‘metadata non available’ in Vectorbase ( S3 Table ) . The expression data from these microarray experiments can be accessed at Vectorbase ( E-MTAB-4022 , http://www . vectorbase . org ) . A total of 27 genes from the cytochrome P450 family ( CYP450 ) , three from carboxylesterases ( CCE ) and 2 from Glutathione S-transferases ( GST ) were differentially regulated ( Figs 2 and 3 ) . One CYP450 gene was down regulated in each studied population . Nine genes from CYP9J , two from CYP6Z and two from CYP6M families were observed . Those families are important in detoxifying pyrethroids in mosquitoes . Among the genes common to the three studied populations , one was from CCE ( CCE3 ) and four were CYP450 ( CYP12F7 , CYP9J10 and CYP9J27 , CYP6BB2 ) . CY6BB2 is noticeable with up-regulations respectively of 33 . 83 , 13 . 47 and 14 . 19 fold vs the NO population in GUY , GUA and CAL populations . The CAL population is tolerant to deltamethrin , and apart from the five detoxification genes common to all populations , only four others are differentially expressed with only one common to GUY and none to the GUA population . CYP9J28 , CYP9J27 and CYP6M9 were also up-regulated up to 12 . 78 to 20 . 48 fold compared to the NO strain . GUY and GUA populations were highly resistant to deltamethrin and had 10 additional differentially expressed genes in common ( Fig 2 ) . Thirty-seven genes producing enzymes , which may play a role in the detoxification pathway as regulators or product transformations ( redox enzymes ) were differentially expressed in all three populations ( Fig 3 ) . Among them , 19 are from the serine protease family including 13 down-regulated genes and seven up-regulated . In the GUY population , seven are down-regulated and two up-regulated; in the GUA population , 12 are down-regulated and 6 up-regulated; in the CAL population , three are down-regulated and four up-regulated ( Fig 3 ) . Serine protease inhibitors are only up-regulated in the CAL population . Overall , 12 genes annotated as “trypsin” or “trypsin precursor” were down-regulated in the three populations . Two of those were common to the three studied populations; five were down-regulated from 9 . 58 to 20 . 71 fold in GUY and GUA populations; five were observed only in the GUA population . Two other serine proteases were over-expressed in the three populations . Seven genes coding for phosphatases , phosphorylases , kinases and cytochrome b5 were up- or down-regulated ( Fig 3 ) . Those enzymes are known to play a role in regulating Cyp450 enzymes . Nine of those enzymes were down-regulated above 5-fold in both GUY and GUA populations whereas only two are differentially regulated in the CAL population . Finally , four of the five redox related genes code for aldehyde oxidase , glutathione peroxidase and thioredoxin peroxidase . They were mainly observed as up-regulated genes in GUY and GUA populations ( Fig 3 ) . Overall , detoxification profiles were more alike between GUY and GUA populations than with CAL populations .
The three Ae . aegypti populations studied are geographically isolated and are from areas with different vector control histories . Therefore , disparity between these areas was observed: high deltamethrin resistance levels were found in GUY and GUA compared to a lower level in CAL . While French Guiana and Guadeloupe have used organophosphates for a long time before switching to deltamethrin more recently , New Caledonia has managed the development of resistance by alternating and reducing the use of insecticides . In addition , Ae . aegypti resistance to deltamethrin in New Caledonia is only found in urban environments , which suggests a link with public health , domestic and gardening use of insecticides , as opposed to agricultural use of insecticides as the case in Guadeloupe . In French Guiana , agriculture is not so extensive which also suggests a limited impact on the development of AE . AEGYPTI resistance in that territory . On account of the different strategies of insecticide use , resistance mechanism patterns are slightly different from one site to another but there is however a consistent underlying core . Nav sequencing demonstrated the presence of haplotypes A and B [8] in each population with a lesser extent in Guadeloupe samples . In addition , S989P found in Asia associated with V1016G was not recorded in our study . I1011M wild type was found mainly in the FTA and in equivalent proportion in CAL . These two mutations were not further investigated due to scarce evidence of their impact on deltamethrin resistance . S989P was associated with V1016G mutation in Southeast Asia populations [30 , 31] . I1011M was associated with cypermethrin resistance in Brazil [25] and to permethrin under laboratory condition [24] . However , evidence of a strong relation with pyrethroid resistance and particularly deltamethrin can be discussed in Latin America [8 , 23 , 26] . In addition , no kdr phenotype was observed in CAL where the highest proportion of 1011M was observed . We then hypothesized that resistance in CAL is due to metabolic resistance . We then focused on V1016I and F1534C mutations , which are well known to be linked alone or in combination to either deltamethrin or permethrin resistance . In contrast , GUY and GUA populations have alleles of resistance at the 1016 and 1534 loci . In fact , target site mutations are multiple and likely linked with pyrethroid resistant populations [7 , 8 , 30 , 32] even if they are involved in the interaction with pyrethroids [33] . The effect of each mutation , their synergy or antagonism in Ae . aegypti resistant phenotypes , has not yet been fully investigated . Recent publications attempted to identify the role of each mutation in vitro by using the Xenope oocysts expression model [24] . However , neither S989P , I1011M nor V1016I were related independently to a modification in deltamethrin linkage [24] . F1534C was related to such modification of type I pyrethroid and not to type II . S989P has always been reported in relation with V1016G and may just enhance the effect of this latter mutation . On the other hand , I1011M was related to cypermethrin resistance ( type II ) in Ceara , Brazil [25] and V1016I was associated to pyrethroid resistance , including in the present work [6 , 23 , 27] . This mutation is in fact followed-up as a marker of resistance [34–36] . Even if F1534C was associated with permethrin resistance and not deltamethrin [28 , 30] , Brito et al . [27] demonstrated that double mutants V1016I and F1534C have an enhanced survival rate when exposed to deltamethrin . It is then of primary importance to study mutations at the sodium channel gene scale to identify the real impact of these mutations alone or in combination in pyrethroid resistant phenotypes and eventually model their interaction with the molecules . Metabolic resistance appears to be associated with the activities of CCE , CYP9J and CYP6Z , both CYP450 enzyme families and GSTs [37 , 38] . CCE and CYP9J enzymes are able to degrade the insecticide whereas CYP6Z such as CYP6Z8 seems to degrade subsequent products [38] . In the present study , 11 genes from CYP9J and CYP6Z were up-regulated concurrently with eight others from CYP6 which are regularly found in pyrethroid resistant populations , three CCE and two GSTs . Using populations from distinct geographical areas and different levels of deltamethrin resistance revealed the presence of five common genes , including four CYP450s ( Fig 2 ) . Whilst close patterns of expression were observed in GUY and GUA populations , they were different from the CAL population . Those results may be associated with the large difference in resistance levels , vector control histories , impact of agricultural practices and different genetic backgrounds . In addition , 21 over-expressed genes found in the present study were also observed in 12 other natural or pyrethroid selected AE . AEGYPTI populations distributed worldwide and analysed with the “Aedes detox” microarray [12 , 26 , 39 , 40] or by others methods ( different microarray and quantitative PCR ) [41 , 42] ( Table 2 ) . CYP6BB2 , CYP6Z6 , CYP6M11 , CYP9J23 , CYP9J9v1 , CYP9J9v2 , CYP9J28 , CYP6CB1 , CYP9J27 , CYP9J10 , GSTE4 were found in six to nine Ae . aegypti populations in South America and Asia , enhancing their implication in pyrethroid resistance ( Table 2 ) . However , only a few functional studies have shown their role in metabolizing pyrethroids [38 , 41 , 43] . CYP6BB2 and CYP9M6 were identified as degrading permethrin in Kasai et al . [41] . However , none of our studied population exhibited the CYP9J32 up-regulated gene found in Mexico , Thailand and Vietnam . CYP9J32 along with CYP9J24 , CYP9J26 , and CYP9J28 are the four enzymes for which the role in degrading permethrin and deltamethrin was demonstrated [43] . CAL population did not show knock-down resistance phenotype ( 100% knocked-down mosquito after 1h exposure ) , we then hypothesized the implication of metabolic resistance to explain only 96% mortality after 24h observation . Among nine of the differentially expressed genes in this population , three ( CYP6BB2 , CYP9M6 , CYP9J28 ) have been proved to degrade pyrethroids . We can assume that those particular genes play a key role in early resistance development of deltamethrin resistance in CAL population . GUY and GUA demonstrate a larger panel of up-regulated detoxifying genes alongside amino-acid changes in the sodium-channel gene , most likely in response to a higher exposure to insecticides , particularly pyrethroids . Even if detoxification enzymes are considered to be the core of insecticide degradation , other enzymes may play a role in the phenotype of resistance . The serine protease family is related to many processes in insects: digestion [44] , interference with penetration and multiplication of the virus in the mosquito ( trypsins ) [45 , 46] and also insecticide resistance [47–49] . These enzymes are also studied in the development of novel insecticides In our study , we observed that trypsins are significantly down-regulated . Transcripts AAEL013629-RA , AAEL010196-RA , AAEL010203-RA , AAEL013714-RA , and AAEL010195-RA coding for trypsins , and AAEL013712-RA coding for a Precursor of 5G1 trypsin inhibitor were also down-regulated in the pyrethroid resistant Ae . aegypti population called Cayman [39] AAEL008079-RA , which was down-regulated in the GUY , GUA and CAL populations , was also down-regulated in the deltamethrin-resistant lab strain CUBA-Delta but up-regulated in the CAYMAN population . Over-expression of two particular trypsins ( >4 . 68-fold vs susceptible ) and chymotrypsin ( >4 . 89-fold vs susceptible ) in deltamethrin-resistant Culex pipiens pallens populations was shown by using microarray and their role in degrading deltamethrin was demonstrated ( 50 ) . Therefore , opposite results are observed for the same gene family within the Culicidae family . Serine proteases are key enzymes in mosquito metabolism and play a role in pyrethroid degradation but their role in the detoxification process must be explored along with its impact on vector competence . Other enzymes in the detoxification pathway may be of interest in the study of vector competence variability . Indeed , CYP450 enzymes are membrane-bound hemoproteins , which require the activity of NADPH P450 oxidoreductase and sometimes cytochrome b5 to function . Enzyme activity is also regulated by phosphatases , kinases and phosphorylases [50] . In addition , oxido-reduction enzymes are also involved in processing the products of detoxifying enzymes . The presence of up- or down-regulation of those transcripts is then not surprising in our study and are often over transcribed in pyrethroid resistant populations [26 , 47 , 51] . Investigating insecticide detoxification pathways is in itself important but the importance is elevated as because some of those enzyme families could mediate arbovirus infection in mosquito cells [52 , 53] . Particular attention should be paid to these interactions to better understand vector competence and disease dynamics . We chose to study geographically distinct Ae . aegypti populations with different levels of resistance to identify which mechanisms may be common and therefore important for resistance development . Amongst the diversity and complexity of resistance , it is essential to identify key markers in order to develop easy-to-use diagnostic tools . Neither the geographic location , resistance levels nor historical use of insecticides can fully explain the distribution of one or another mechanism of resistance . It is then essential to integrate all of the factors that may impact resistance in vivo and also in vitro to get the best picture , the most reliable data in order to develop predictive tools . In this study , a set of genes were highlighted and which can serve as a basis for further studies on the factors which may induce , develop and maintain the expression of those genes . The role that enzymes may play alone or in association with resistance phenotypes and the consequences of over-expression of those genes on other physiological and cellular processes should then be investigated .
|
Aedes aegypti is vector of Dengue , Chikungunya and Zika viruses , all causing emerging or re-emerging diseases worldwide . Fighting these diseases relies on the control of the vector . Therefore , insecticides have been extensively used worldwide , resulting in the development of insecticide resistance . In the French overseas territories , resistance to pyrethroids has been monitored for many years with high levels in the South American French territories . We then investigated the mechanisms underlying this resistance in populations from French Guiana , Guadeloupe and New Caledonia . Transcription levels of detoxification genes were measured and alongside screening for target site mutations . Upregulation of cytochrome P450 genes and carboxylesterases were observed in all three populations . Mutations related to pyrethroid resistance in position 1016 and 1534 of the voltage-gated sodium channel gene were also observed . French Guiana and Guadeloupe populations presented a closer profile of resistance mechanisms whereas the New Caledonia population had a more restricted profile . Such differences can be explained by different vector control practices , regional insecticide uses and genetic backgrounds . These results are also compared with others obtained from other parts of the world and are discussed with the perspective of integrative research on vector competence .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[] |
2015
|
Deltamethrin Resistance Mechanisms in Aedes aegypti Populations from Three French Overseas Territories Worldwide
|
Legionella pneumophila , the causative agent of Legionnaires' disease , invades and replicates within macrophages and protozoan cells inside a vacuole . The type IVB Icm/Dot secretion system is necessary for the translocation of effector proteins that modulate vesicle trafficking pathways in the host cell , thus avoiding phagosome-lysosome fusion . The Legionella VipA effector was previously identified by its ability to interfere with organelle trafficking in the Multivesicular Body ( MVB ) pathway when ectopically expressed in yeast . In this study , we show that VipA binds actin in vitro and directly polymerizes microfilaments without the requirement of additional proteins , displaying properties distinct from other bacterial actin nucleators . Microscopy studies revealed that fluorescently tagged VipA variants localize to puncta in eukaryotic cells . In yeast these puncta are associated with actin-rich regions and components of the Multivesicular Body pathway such as endosomes and the MVB-associated protein Bro1 . During macrophage infection , native translocated VipA associated with actin patches and early endosomes . When ectopically expressed in mammalian cells , VipA-GFP displayed a similar distribution ruling out the requirement of additional effectors for binding to its eukaryotic targets . Interestingly , a mutant form of VipA , VipA-1 , that does not interfere with organelle trafficking is also defective in actin binding as well as association with early endosomes and shows a homogeneous cytosolic localization . These results show that the ability of VipA to bind actin is related to its association with a specific subcellular location as well as its role in modulating organelle trafficking pathways . VipA constitutes a novel type of actin nucleator that may contribute to the intracellular lifestyle of Legionella by altering cytoskeleton dynamics to target host cell pathways .
The gram-negative bacterium Legionella pneumophila is the causative agent of a severe type of pneumonia known as Legionnaires' disease [1] . Infection of mammalian alveolar macrophages is believed to be largely accidental and occurs after inhalation of aerosols originating from contaminated water sources , where the intracellular pathogen thrives within its natural protozoan hosts . After phagocytosis , Legionella replicates in a specialized vacuole that avoids the endocytic pathway and supports bacterial replication [2]–[4] . The type IVB Icm/Dot ( Intracellular Multiplication/Defective Organelle Trafficking ) translocation system is essential for the processes that prevent the phago-lysosome fusion [5]–[11] . Effector proteins injected into the host cell by the Icm/Dot T4BSS are presumed to modify trafficking pathways thus avoiding bacterial degradation and promoting the formation of a replication-competent Legionella-containing vacuole ( LCV ) . Although to date approximately 300 Icm/Dot substrates have been identified , the functions of most effectors remain unknown . With few exceptions , the absence of one effector does not impair intracellular growth , an occurrence believed to be due to functional redundancy among effectors and/or host cell targets proteins or pathways . In fact , a recent study has demonstrated that a simultaneous deletion of 31% of known Legionella Type IVB substrates minimally affected intracellular growth in mouse macrophages [12] . Therefore , classical forward and reverse bacterial genetics have been mostly ineffective in determining the functions and contributions of effectors to intracellular events during infection . Thus alternative approaches such as bioinformatics and biochemistry have been used to elucidate the roles of translocated effectors in a small number of cases . The conserved organelle trafficking pathways between Saccharomyces cerevisiae and higher eukaryotes , its amenability for genetic manipulation and extensive library of mutants and strains expressing fluorescently-tagged proteins make yeast an attractive model for studying the functions of pathogen effectors . A successful strategy relied on the ectopic expression of Legionella genes in S . cerevisiae and screening for mistrafficking of proteins to the vacuole ( Vacuolar Protein Sorting/VPS ) . One of the Legionella effectors leading to a Vps− phenotype is the VipA protein ( VPS inhibitor protein A ) . We previously showed that in yeast the VipA-derived interference in organelle trafficking was related to the Multivesicular Body ( MVB ) Pathway , and that VipA is a bona fide Icm/Dot substrate translocated into macrophages [13] . In this work we sought to functionally characterize VipA using in vivo and in vitro approaches . We found that VipA binds actin and directly enhances its polymerization in vitro . During infection of macrophages , translocated VipA localizes in punctate structures that associate with filamentous actin and early endosomes . Similarly , hybrid VipA-GFP and VipA-mCherry localize in puncta when ectopically expressed in both mammalian and yeast cells . In budding yeast the puncta are often associated with the vacuole membrane , reminiscent of the pre-vacuolar compartment seen in vpsE mutants defective for MVB formation , and co-localize with actin-rich structures/organelles , such as the bud-neck , endosomes or cortical actin patches . In mammalian cells , VipA-EGFP puncta co-localize with early endosomes and actin patches , but not other components of the endosomal or secretory pathways . In contrast , a VipA mutant ( VipA-1 ) that no longer causes a Vps− phenotype in yeast displays a homogeneous cytosolic distribution and is defective in actin binding in vitro . These results show that Legionella effector VipA associates with components of the endocytic pathway and that this function is linked to its modulation of actin dynamics , suggesting a mechanism of interference with host cell organelle trafficking pathways during infection .
L . pneumophila , S . cerevisiae and E . coli strains used in this work are listed in Table S1 . Bacterial strains were grown as previously described [14] . Deletion of vipA in strain LPIF3 was generated by natural transformation of KS79 [15] with a DNA fragment containing PCR products of a kanamycin resistance cassette and flanking regions of vipA . The PCR products were synthesized by long-flanking homology PCR [16] using primers IF04 , IF05 , IF06 and IF07 ( see Table S2 ) . Plasmids and oligonucleotides used in this study are listed in Table S2 . For expression of GFP tagged proteins in mammalian cells , plasmids pIF203 and pIF213 were constructed . PCR amplification of vipA and vipA-1 mutant was carried out using as template chromosomal DNA from strain JR32 and plasmid DNA pET15b-vipA-1 , respectively . Oligonucleotides IF02 and IF03 allowed the introduction of an XhoI site and a Kozak consensus sequence immediately upstream the 5′ end of the gene , and a BamHI site at the 3′end . PCR products were sub-cloned at the same sites of pEGFP-N1 ( Clontech ) generating translational fusions of the vipA alleles to EGFP under the transcriptional control of the CMV I/E promoter . For expression of GFP fusion proteins in yeast , plasmids pIF206 and pIF209 were generated by PCR amplification of vipA and vipA-1 , respectively , using the same templates as above and oligonucleotides IF08 and IF09 . The PCR products were digested with BamHI and HindIII and inserted into the same sites of pKS84 [15] . In order to construct similar fusions to mCherry , a derivative of pKS84 was engineered in which the URA3 marker was substituted by LEU2 and GFP replaced by mCherry . The exchange of the marker was made by PCR amplification from pACT2-1 ( Clontech ) with oligos IF37 and IF38 , digestion of the product with SalI and XmaI and subcloning in the same sites of pKS84 . Insertion of mCherry was then carried out by subcloning a PCR product obtained by amplification from pXDC50 with oligos IF39 and IF40 and digestion with HindIII and SalI , leading to a Pgal-mCherry fusion in plasmid pIF215 . Two DNA fragments containing vipA or vipA-1 were removed from pIF206 or pIF209 ( see above ) by digestion with BamHI and HindIII and inserted in the same sites of pIF215 , yielding pIF216 and pIF217 carrying the vipA-mCherry fusions . For expression of his-tagged proteins in E . coli , wild-type or mutant vipA-1 alleles were amplified using primers NSP22 and NSP23 , containing NdeI and BamHI sites . The PCR products were digested with these enzymes and inserted in the same sites of pET15b ( Novagen ) , and plasmids pET15b-VipA and pET15b-VipA-1 were obtained . Plasmid pNSvipA was constructed by amplifying vipA with oligos NSP29 and NSP30 , both containing SalI sites . The PCR product was digested with this enzyme and cloned in the SalI site of pNS00 [13] . Mutagenesis was performed using the GPS Linker Scanning System according to the manufacturer's instructions ( New England Biolabs ) and plasmid pNSvipA used as a template ( see above ) . The transposition reaction was performed with 1 . 5 µg pNSvipA incubated with 20 ng pGPS4 and 1 µl TnsABC . Dilutions of the reaction were used to transform ElectroTen Blue E . coli ( Stratagene ) by electroporation and plated on LB+Carbenicillin+Chloramphenicol . Plasmid DNA was prepared from approximately 25000 scraped colonies , transformed into the yeast strain NSY01 and transformants plated on SC-ura/fructose . Invertase overlay was performed on yeast transformants as previously described [13] and roughly 14% of colonies were white , indicating a restoration of the Vps+ phenotype . 100 of these colonies were isolated , grown overnight in liquid SC-ura/fructose and plasmid DNA extracted . Plasmid pools were digested with PmeI to excise the CmR marker , the plasmid backbone gel purified , self-ligated and transformed into E . coli . Plasmid DNA was extracted from 352 colonies and transformed into NSY01 . Roughly 30% of the colonies were white on the Invertase Overlay , from which 40 were grown overnight in liquid SC-ura/fructose , plasmid DNA prepared and sequenced to map and identify the linker insertion . For detection of VipA in L . pneumophila , strains were grown overnight in AYE medium with the appropriate antibiotics . Bacteria ( approximately 2 . 8×108 cells ) were harvested by centrifugation , sample buffer was added and after SDS-PAGE the proteins were immunoblotted with affinity-purified rabbit polyclonal antibody raised against recombinant His6-VipA ( essentially as described by Zhu et al . [17] ) . For detection of GFP or GFP fusions in yeast , strains were grown on SC-ura+Fructose or +Galactose plates for three days at 30°C and several colonies were picked and grown overnight in identical same liquid medium . An amount equivalent to OD600 3 in a total volume of 40 µl of SDS-Loading Buffer ( or OD600 1 in 15 µl in the case of the strain expressing GFP ) was boiled for 5 minutes and run on SDS-PAGE . Transfer and immunoblotting were performed as above using a rabbit polyclonal antibody to GFP at a 1∶1000 dilution . For immunoblots of translocated VipA , differentiated THP-1 cells were infected with L . penumophila strains for 3 hr at an MOI = 50 . Cells were lysed by incubation with PBS+Triton X 2% for 15 min at 4°C . Lysates were centrifuged at 13 . 000 rpm for 15 min , and pellet resuspended in the same solution . E . coli BL21 ( DE3 ) strains carrying vipA expressing plasmids ( see Table S2 ) were grown overnight and backdiluted 1∶50 . At an OD600∼0 . 6 , IPTG ( isopropyl β-D-1-thiogalactopyranoside ) was added to a final concentration of 1 mM and growth was continued for 2 hours . For pull-down assays , pellets were resuspended in 20 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 1 mM EDTA , 100 µg . ml−1 PMSF+Protease Inhibitor Cocktail ( Sigma ) , 1 mg . ml−1 lysozyme and incubated for 30 min on ice . Bacteria were lysed using 3 passages in a French Press , lysates were centrifuged at 16000 rpm and supernatants were transferred to tubes containing Ni-NTA beads ( QIAgen ) or loaded on HisTrap FF Columns ( GE Healthcare ) connected to a AKTA FPLC system ( GE Healthcare ) , and eluted with Imidazole gradients . Acanthamoeba castellanii were cultured in PYG medium at 28°C without agitation . CHO FcγRII and THP-1 cells were grown in DMEM or RPMI , respectively , supplemented with 2 mM glutamine and 10% heat-inactivated fetal bovine serum , at 37°C in a 5% CO2 incubator . Differentiation of monocytes was accomplished 3 days after the addition of 1 ng . ml−1 of PMA ( phorbol 12-myristate 13-acetate ) to the medium . CHO-FcγRII cells [18] were grown , transfected , fixed and permeabilized for immunofluorescence as described previously [15] . Actin or DNA staining was carried out by incubating cells with , respectively , Rhodamine-phalloidin ( Sigma , 5 µg . ml−1 ) or DAPI ( 100 µg . ml−1 ) during 30 min . Infection of THP-1 monocyte-like cells was carried out as described above using an Multiplicity of Infection of 50 , and cells processed for immunofluorescence ( 14 ) . For rhodamine-dextran endocytosis assays , the compound was added at 4 hr post-infection , for 1 hr and chased for 10 , 30 , 120 and 240 min , at 1 mg . ml-1 . Mouse monoclonal primary antibodies used were α-EEA1 ( BD Biosciences ) , α-LAMP1 ( UH1 ) , α-ALIX ( Biolegend ) , and αKDEL ( Santa Cruz Biotecnology ) . Secondary antibodies were goat anti-mouse Alexa-594 or FITC-labeled ( Invitrogen ) , or α-rabbit Alexa-647 ( Invitrogen ) , -TRITC or -FITC ( Sigma ) . The αVipA antibody used in immunofluorescence assays was previously affinity-purified ( see above ) . Microscopy was carried out on Laser Scanner Confocal Microscopes ( Zeiss LSM710 , Leica SP5 II STED-CW or Leica TCS SP2 ) . Quantitative analysis of colocalization was performed by calculating the Manders overlap coefficient , corresponding to the fraction of green pixels ( VipA-EGFP signal ) that overlap with red pixels in relation to the total green pixels [19] . For this purpose , signal intensities for each cell ( n>15 for each antibody ) were adjusted in ImageJ and the coefficients determined with the plugin JACoP [20] . Statistical significance was determined with unpaired t test , and p values obtained are indicated ( ** , p<0 . 01; *** , p<0 . 001 ) . S . cerevisiae cells expressing GFP or mCherry fusion proteins were grown in plates with SC-Ura/-Leu/-Ura-Leu supplemented with 2% fructose at 30°C for 3 days . Several colonies were inoculated in identical liquid medium supplemented with 2% galactose and grown overnight . The next day the cultures were diluted in the same medium to an OD600 = 0 . 3 , and grown to an OD600 = 0 . 5 . FM4-64 staining was performed essentially as previously described [15] . Cells were mounted on agarose pads and visualized on a Laser Scanner Confocal Microscope ( Zeiss LSM710 ) . For pull-down assays , U937 monocyte post-nuclear supernatants ( PNS ) were prepared by harvesting 2×108 U937 cells grown in suspension in RPMI+Glu+10%FBS . Cells were lysed by addition of 10 ml 50 mM Tris-Hcl pH 8 , 150 mM NaCl , 0 . 1 mM EDTA , 0 . 5% NP-40 , 1 mM DTT , 0 . 1 mM NaVO4 , 100 µg . ml−1 PMSF+Protease Inhibitor Cocktail ( Sigma ) . Lysis was accomplished by 1 hour incubation on ice and centrifugation at 4000 rpm for 30 min . 10 ml U937 PNS were mixed with 100 µl Ni-NTA beads loaded with approximately 100 µg his-tagged bait proteins , incubated at 4°C and washed with His buffer ( 20 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 10% Glycerol , 100 µg . ml−1 PMSF ) +40 mM Imidazole . Beads were applied to a BioRad column , washed again and eluted with HisBuffer+500 mM Imidazole . Eluates were collected , mixed with Protein Sample Buffer and analyzed by SDS-PAGE followed by either Coomassie staining or Western blot . Differential protein bands in eluates that were absent in His-FabI and Beads only pull-downs , and did not react with antibody against polyHistidine were excised , digested with trypsin and analyzed by Liquid Chromatography/Mass Spectrometry ( LC/MS ) . For Western blotting 1∶5000 dilutions of monoclonal antibody against either actin or polyHistidine ( Sigma ) were used . G-actin and Pyrene-labeled G-actin ( Cytoskeleton , >99% pure ) were prepared as indicated by the manufacturer and kept in G-actin buffer ( Tris-Cl 5 mM , CaCl2 0 . 2 mM , ATP 0 . 2 mM , DTT 1 mM ) . Before polymerization , conversion to Mg-actin was accomplished by addition of EGTA to 0 . 2 mM and MgCl2 to 50 µM and incubation for 2 min at room temperature . Polymerization assays were made on black bottom microplates and fluorescence read in a microplate reader equipped with an injector ( Infinite M200 , Tecan ) . Values were obtained using an excitation wavelength of 365 nm and emission of 407 nm , and recorded at 10 sec intervals . Data were collected with Magellan software v6 . 4 ( Tecan ) and then processed in Excel ( Microsoft ) . Reactions contained 2 µM actin ( 10% Pyrene-labeled ) , and were initiated by adding KMEI Polymerization Buffer ( Imidazole 10 mM pH 7 , KCl 50 mM , MgCl2 1 mM , EGTA 1 mM ) . Purified His6-VipA or His6-VipA-1 in G-Mg buffer ( G-actin buffer with 0 . 1 mM MgCl2 instead of Cacl2 ) were added as indicated . To measure elongation , actin seeds were prepared as follows ( adapted from [21] ) . Actin was resuspended in G-actin buffer to a final concentration of 15 µM , converted to Mg-actin for 20 min and polymerization induced by addition of KMEI Polymerization buffer ( see above ) and incubation for 2 hr at room temperature . To stabilize filaments , actin was diluted to 5 µM in the presence of 5 µM of phalloidin for 5 min . The polymerized actin was recovered after centrifugation for 20 min at 90 , 000 rpm and 4°C in a TLA100 rotor and these protected filaments ressuspended in G-actin buffer and used in elongation assays with 1 µM G-actin . Filament concentration and rate of monomer addition were calculated as described previously [22] . Swissprot IDs of proteins: Legionella pneumophila VipA , Q5C8M7; Chlamydia trachomatis TARP , Q6GX35; Salmonella enterica serovar typhimurium SipC , Q56020 , and SipA , P0CL52; Vibrio parahaemolyticus VopL , B9A807 , and Vibrio cholerae VopF , B5AN40; Rickettsia conorii Sca2 , Q92JF7; Saccharomyces cerevisiae Bro1 , P48582; Homo sapiens Alix , Q8WUM4 .
In previous work aimed at identifying new effectors [13] , disruption of organelle trafficking was observed when VipA was ectopically expressed in yeast , raising the possibility that an interaction between VipA and host cell proteins could be involved in the process . We sought to identify eukaryotic binding partners for VipA in macrophages by in vitro assays using purified VipA as bait to pull-down proteins from U937 human monocyte-like cell extracts . For this purpose , a recombinant version of VipA carrying a 6-histidine tag at its N-terminal region ( His6-VipA ) was constructed in plasmid pET15b ( Novagen ) and purified by Ni-NTA affinity chromatography . A post-nuclear supernatant was prepared from U937 cells and incubated with Ni-NTA agarose beads , preloaded either with His6-VipA or an irrelevant protein , His6-FabI , the L . pneumophila enoyl acyl CoA reductase protein , used as negative control for the interaction as it is not translocated into the host cell [23] . Bound proteins were eluted with 500 mM imidazole , separated by SDS-PAGE and analyzed by Coomassie staining ( Figure 1A ) . A 42-kDa band that co-eluted with VipA but was absent in the FabI eluate and did not react with an antibody against polyHistidine ( data not shown ) was excised and identified as β-actin by Liquid Chromatography/Mass Spectrometry ( LC/MS ) . This result was confirmed with a Western-blot using a monoclonal antibody against actin ( Figure 1B ) . To test if the interaction between VipA and actin was direct , additional pull-down assays were performed using His6-VipA and increasing concentrations of purified monomeric G-actin ( Figure 1C ) . The results show that actin bound to VipA directly , without the requirement of any additional host factor . The finding that VipA interacted directly with actin in vitro suggested a mechanistic basis for the ability of the protein to affect vesicle trafficking pathways in yeast , as actin is involved in numerous membrane trafficking processes [24] . To test if the ability of VipA to disrupt trafficking depended on its interaction with actin , we generated linker mutant alleles of vipA using the transposon-based GPS system ( New England Biolabs ) . This procedure ultimately introduces 15-bp insertions in the target gene , of which 1/3 result in the insertion of a stop codon and 2/3 in the in-frame insertion of five amino acids . A plasmid harboring vipA was mutagenized and the library of mutants ( approximately 25000 colonies ) transformed into yeast strain NSY01 , used as a reporter for screening Vacuolar Protein Sorting defects ( Vps− phenotype ) [13] . This strain produces the hybrid protein Carboxypeptidase Y-Invertase , which travels to the vacuole in wild-type yeast but can be missorted to the cell surface if vacuolar protein trafficking is disrupted , which is the case when wild-type VipA is ectopically expressed . The excreted enzyme hydrolyses sucrose present in the medium that can be detected by the brown color of the colonies in a particular screening medium ( Vps− phenotype; see [25] ) . In this work , we were interested in isolating vipA mutants that no longer disrupted vacuolar traffic , thus leading to the formation of white/Vps+ colonies . These colonies were isolated from transformants of the vipA linker library , and the corresponding plasmids sequenced to map the insertions . Five different in-frame insertions were identified disrupting the ability of the protein to cause a Vps− phenotype ( Figure 2A ) . All five VipA mutant proteins were stably expressed in yeast at levels identical to wild-type ( data not shown ) . Interestingly , all insertions mapped between codons 67 and 98 , just upstream of a sequence encoding a predicted coiled-coil domain located between residues 137 and 200 . Sequence analysis of this region and additional investigation of the entire protein sequence did not reveal further homologies to known domains or motifs , although we noted a high occurrence of prolines in the C-terminal region of VipA . To further understand the importance of the N-terminal region , one of these mutant proteins ( VipA-1 , insertion at residue 67 ) was expressed and purified as a his-tagged fusion protein ( His6-VipA-1 ) , displaying expression and solubility levels similar to wild-type VipA . Its ability to bind actin was tested as above in pull-down assays using U937 cell extracts , where a significant decrease in affinity for monomeric actin was observed ( Figure 2B ) . Taken together , these results show that the ability of VipA to disrupt vacuolar trafficking in yeast and its ability to bind actin are linked . This result is inconsistent with a non-specific association of actin and VipA . To find out whether VipA had any effect on the assembly of actin filaments we used pyrene-actin polymerization assays . Fluorescence of pyrene-actin increases significantly when G-actin monomers are incorporated into a filament , permitting polymerization to be measured in real-time . In reactions containing 2 µM actin ( 10% pyrene-labeled ) , His6-VipA stimulated actin polymerization at nanomolar concentrations in a dose-dependent manner ( Figure 3A ) . Saturation occurred at approximately 100 nM VipA , a concentration that increased the actin polymerization rate by approximately 3-fold ( Figure 3A , right panel ) . This effect on microfilament polymerization is not as potent as the nucleation of actin by the Arp2/3 complex activated by WASP-VCA ( Figure 3B ) , or the mouse formin mDia2 ( Figure 3C ) . Also , in contrast to the case of other bacterial effectors , His6-VipA does not activate Arp2/3-mediated actin polymerization , as the effect seen in the presence of inactive Arp2/3 and VipA is additive of the two individual effects . The VipA-1 mutant displayed only a small effect on the assembly of actin filaments , in the same order of magnitude of inactive Arp2/3 , even at high concentrations ( Figure 3B ) . Polymerization of actin occurs in two phases with distinct kinetics . The initial and limiting step , nucleation , occurs very slowly and leads to the formation of actin trimers , whereas the subsequent elongation of the filament takes place at a much higher rate . The observed shorter duration of the initial lag phase in the presence of VipA was consistent with a role in enhancing actin nucleation . To find out if VipA was also affecting elongation , we carried out polymerization assays with actin seeds . In these assays all polymerization occurs from short preformed actin filaments , and thus increases in fluorescence are not due to de novo actin nucleation . As shown in Figure 3C VipA enhanced slightly but reproducibly the elongation rate of the small filaments . Cytochalasin D , a well-characterized filament barbed-end capper used here as a negative control , lead to a complete halt in elongation . The effect mediated by VipA during filament elongation could not , however , account for the overall increase in actin polymerization observed above ( Figures 3A and B ) , which indicates VipA is acting predominantly as a nucleator . Taken together , the data obtained with the in vitro actin polymerization assays shows that VipA is able to enhance actin polymerization without the requirement of additional proteins . Moreover , these results indicate the effector uses a mechanism that favors nucleation and moderately increases the rate of addition of monomers during elongation . The L . pneumophila genome encodes approximately 300 effector proteins that are translocated to the host cell during infection . Surprisingly the vast majority of effector genes are dispensable for intracellular replication and yield no obvious phenotype when deleted . In order to determine if VipA is essential for infection of host cells , the vipA gene was deleted from Legionella strain KS79 . The wild-type vipA allele was substituted by a kanamycin-resistance cassette and the absence of VipA protein in the resulting strain LPIF3 was confirmed by Western immunoblot using a polyclonal anti-VipA antibody ( Figure S1A ) . Infection of THP-1 macrophages and the amoeba A . castellanii was carried out with Legionella strains KS79 and LPIF3 harboring plasmid pXDC31 ( Table S2 ) , in which GFP expression is driven by the IPTG-inducible Ptac promoter . Thus , intracellular bacterial replication can be followed by real time monitoring of GFP fluorescence measurements [26] . In addition , the translocation-defective dotA mutant strain containing the same plasmid was used as a negative control for intracellular growth . Replication in A . castellanii was not affected by deletion of vipA , and identical results were obtained using THP-1 macrophages as hosts ( Figure S1B; data not shown ) . As a more subtle effect in the initial phase of phagocytosis may be difficult to detect in this assay , gentamicin protection assays were carried out to measure entry into host cells . Similarly , no significant difference was observed between the number of intracellular bacteria from wild-type or vipA strains ( data not shown ) . To gain further insight into the function of VipA we analysed its subcellular localization in eukaryotic cells . For this purpose initial studies were performed in S . cerevisiae , in which the Vps− phenotype had led to vipA identification . S . cerevisiae strain NSY01 ( see above ) was transformed with plasmids carrying C-terminal GFP fusions of either wild-type VipA or mutant VipA-1 under the control of the galactose-inducible Pgal promoter . In the resulting strains , expression of the ∼70 kDa recombinant proteins in the presence of the inducing sugar was confirmed by Western-blot of the cell extracts using an anti-GFP antibody ( Figure 4A ) . Laser Scanning Confocal Microscopy analysis revealed the localization of VipA-GFP in puncta in yeast cells ( Figure 4B ) , whereas the VipA-1-GFP mutant protein exhibited homogeneous cytosolic distribution , similar to GFP alone . Interestingly , approximately half of the VipA puncta were associated with the mother-bud neck in dividing cells , a site containing the cytokinetic ring and enriched in actin filaments , which is consistent with the ability of VipA to interact with actin in vitro ( see above ) . Thus , the association of VipA with other actin-rich structures was investigated . In S . cerevisiae , visible F-actin structures consist of patches , cables and rings . Patches are highly motile punctate structures that form at the cell cortex , mediate endocytosis and contain numerous actin-associated proteins . After endosome internalization they transition to a phase of rapid movement that depends on their transport along actin cables away from the cell membrane [27]–[29] . Actin cables consist of bundles of F-actin aligned along the mother-bud axis and serve as tracks for the movement of secretory vesicles , mitochondria , Golgi , and vacuoles from the mother cell to the growing bud [30] , [31] . Association of VipA-mCherry with actin in patches and cables was assessed by colocalization studies with two actin markers: Abp1-GFP , which binds F-actin patches in endosomal sites , and Abp140-GFP , which associates with microfilaments in both patches and cables [27] , [32] , [33] . The distribution of VipA-mCherry and VipA-1-mCherry was identical to the observed above with the VipA-GFP and VipA-1-GFP fusions , respectively ( Figure 4C ) . Colocalization between VipA and actin structures was observed in the case of Abp1-containing endosomes ( Figure 4C , left panel ) and for Abp140-associated patches , but not cables ( Figure 4C , right panel ) . The presence and distribution of these markers was not affected by the presence of either VipA or VipA-1-mCherry . These results show that the puncta formed by VipA in yeast are located in sites containing a high array of actin filaments , namely the cytokinetic ring and cortical patches , and that the vipA-1 mutation that decreased the affinity of VipA for actin in vitro also abolished its targeting to these locations . The interference of VipA in the formation of the MVB in yeast [13] strongly hinted its interaction with one or more components of this pathway . MVBs are present in both the Endocytic and Secretory/Biosynthetic pathways and result from the invagination and budding of the endosome membrane , and eventually fuse and deliver their cargo to the vacuole/lysosome . We initially tested an interaction with the yeast vacuole using the styryl dye FM4-64 , which stains the vacuole membrane after being internalized by endocytosis [34] . Staining of cells producing VipA-GFP revealed that approximately 70% of the VipA puncta were associated with the vacuole membrane and co-localized with an abnormal pre-vacuolar compartment observed previously in strains expressing VipA ( Figure 5A; [13] ) . This structure is identical to the one characteristic of class E vps mutants , where defective MVB formation leads to the inability of endocytosed cargo to fuse completely with the vacuole [35] , [36] . This failure is caused mainly by defects in packaging of the cargo into intralumenal vesicles of MVBs , which is carried out by the sequential action of proteins composing the ESCRT 0-III complexes ( Endosomal Sorting Complex Required for Transport ) . One of the proteins involved in these late steps of the MVB pathway is Bro1 , a cytoplasmic yeast protein that transiently associates with endosomes , where it is required for the formation of intralumenal vesicles . We assessed VipA-mCherry colocalization with Bro1-GFP and verified that approximately 37% of VipA puncta colocalized with Bro1-GFP ( Figure 5B ) , confirming an association of the effector with this protein component of the MVB . Taken together , these results show that VipA associates with components of the yeast MVB pathway , namely Bro1-containing endosomes and the vacuole . Moreover , they suggest that the previously observed effect of VipA on mistrafficking of vacuolar proteins could be due to a defect on late steps on the MVB pathway wherein fusion of endosomes with the vacuole occurs . In order to assess the subcellular localization of VipA in host cells under physiological conditions , infection of THP-1 monocyte-like cells was carried out , cells fixed at several time points post-infection , processed for immunofluorescence using a polyclonal anti-VipA antibody and analysed using Laser Scanning Confocal Microscopy . After infection with wild-type strain L . pneumophila JR32 , VipA was found in diverse structures inside the host cell , which varied in size from puncta to larger formations ( Figures 6 and 7 ) . VipA was not associated with the LCV at any time point from 30 min to 14 hr after uptake ( Figure 6A and B; data not shown ) . To assess the localization of the VipA-1 mutant , L . pneumophila vipA null-mutant background strains were constructed carrying plasmids encoding IPTG-inducible copies of either the wild-type or the mutant vipA allele ( respectively , ΔvipA pMMB207c-Ptac-vipA+ or ΔvipA pMMB207c-Ptac-vipA-1 ) . Infection with the strain carrying Ptac-vipA+ led to a similar distribution of the effector as observed in JR32 , although the protein was observed in the host cell earlier after uptake . This localization was lost in the vipA-1 mutant , in which the effector was translocated in levels similar to the wild-type ( Figure 6C ) and homogeneously distributed in the host cell cytosol ( Figures 6A and 7A ) . In order to examine the colocalization of translocated VipA during the course of infection with endosomes and actin , infected cells were stained with respectively , an anti-EEA-1 antibody and rhodamine-phalloidin , a fluorescently-labeled protein that specifically binds F-actin . Colocalization was observed with both markers and increased over time , reaching the highest values at 8 hours post-infection ( Figures 6B and 7A; data not shown ) . To determine the degree of colocalization with these two cell components , quantitative analysis was performed by calculating the Manders overlap coefficient , a commonly used approach for quantifying colocalization in fluorescence microscopy [19] . Briefly , in this experiment it corresponds to the fraction of blue pixels ( VipA signal ) that overlaps with green or red pixels ( EEA-1/Early Endosome and Rhodamine-phalloidin/F-actin , respectively ) ( see Materials and Methods ) . As shown in Figure 7B , at 8 hr post-infection an average of 23% of VipA colocalized with EEA-1 and 40% with actin filaments , being these values approximately 10% higher in the case of the strain carrying Ptac-vipA+ . Notably , in some cells more than 80% of VipA was associated with EEA-1 or actin ( Figure 7B ) , often simultaneously with both ( see also enlarged areas of Figure 7A ) , suggesting a dynamic interaction among these components . We addressed the possible interference of VipA in host cell endocytic trafficking in several ways . Firstly , we looked for a defect in endocytic internalization of rhodamine-dextran in THP-1 macrophages previously infected with L . pneumophila wild-type or ΔvipA . In this experiment , we compared the total rhodamine fluorescence in both types of infected cells , and we investigated possible delays in trafficking of rhodamine-dextran by visualizing colocalization of this dye with anti-EEA-1 antibody over time . However , in neither case a significant difference was apparent , suggesting that endocytic internalization and early endosome trafficking is not being affected by the presence of VipA ( data not shown ) . Additionally , the number and morphology of EEA-1 positive Early Endosomes were identical in these infected cells . These observations demonstrate that during macrophage infection , translocated VipA binds actin filaments and early endosomes in the host cell . To our knowledge , no other currently identified Legionella effector displays either of these properties . The previous macrophage infection experiments did not exclude the possibility that additional Legionella effectors secreted during infection were required for the correct localization of VipA upon translocation into the host cell . Thus , in order to analyse VipA subcellular localization in the absence of other Legionella proteins , we expressed it ectopically in mammalian CHO cells and examined its association with microfilaments and early endosomes . For this purpose , a VipA-EGFP fusion protein was constructed and its expression placed under the control of the CMV promoter in a derivative of plasmid pEGFP-N1 ( Clontech ) . The resulting plasmid was transfected into CHO FcγRII cells and these were fixed and processed after 48 hr . Similarly to what was observed in yeast and during infection , VipA-EGFP was distributed in puncta in transiently transfected CHO cells , whereas the mutant VipA-1-GFP showed homogeneous cytosolic distribution ( Figure S2 ) . To analyse its association with microfilaments , these cells were stained with Rhodamine-phalloidin . As shown in Figure S2 , the degree of colocalization between VipA and actin was difficult to ascertain due to the large array of stress fibers present in CHO cells . Therefore , further visualization was made after the disassembly of stress fibers with Cytochalasin D , an actin depolymerizing agent [37] , [38] . As observed in CHO cells expressing GFP , incubation with 10 µM Cytochalasin D disrupted the actin cytoskeleton , changing the organization of actin in the cell from a filament network to focal accumulations ( Figure S2 , bottom panel ) . The same actin reorganization occurred when VipA-GFP expressing cells were treated with cytochalasin , but interestingly the VipA puncta , which did not disassemble but appear to have coalesced into larger structures , clearly colocalized with the enduring small cortical actin foci ( see colocalization plot in Figure S2 ) . Actin rearrangements upon cytochalasin treatment were also observed in cells expressing the VipA-1 mutant fused to GFP , but the cytosolic distribution of VipA-1-GFP did not alter with cytochalasin treatment . The localization of VipA-EGFP with early endosomes was also tested by immunofluorescence assays with an anti-EEA-1 antibody ( Figure S3A ) . The degree of colocalization with this marker was on average approximately 35% , whereas in cells expressing the VipA-1 mutant the overlap was only 6% ( Figure S3E; p<0 . 001 , unpaired t test ) , and these values are similar to the ones obtained during infection ( Figure 7 ) . VipA was shown in this study ( see above ) to colocalize with components of the yeast MVB pathway , namely Bro1 containing endosomes and the vacuole . The protein Alix ( also known as AIP-1 ) is a component of the MVB and a human counterpart of yeast Bro1 and interestingly was shown to be involved in the assembly of microfilaments , constituting a novel link between the MVB and the actin cytoskeleton [39] . To investigate if VipA associated with other components of the endosomal/MVB pathway in addition to early endosomes , we carried out immunofluorescence experiments in transiently transfected CHO cells and tested colocalization of VipA-EGFP with LAMP-1 and Alix . As shown in Figures S3 C and D , some signal overlap was observed with both lysosomes and Alix . However , the average colocalization was relatively low ( 15% ) and not distinguishable from the values obtained with the mutant form of VipA ( Figure S3E ) . In addition , similar results were obtained when colocalization was tested with the ER marker KDEL ( Figures S3B and E ) , ruling out an association with the secretory pathway . Taken together , these observations demonstrate that VipA associates with cortical F-actin patches and early endosomes in mammalian cells , and this occurs independently of other secreted Legionella effectors . Moreover , ectopically expressed VipA-EGFP does not colocalize significantly with later components of the Endosomal/MVB Pathway , such as lysosomes or Alix , or with the ER .
The L . pneumophila effector VipA was identified in previous studies due to its ability to interfere with organelle trafficking in the yeast Multivesicular Body ( MVB ) Pathway [13] . In this work we have shown that VipA binds actin and is able to polymerize microfilaments in vitro without the requirement of additional bacterial or eukaryotic factors ( Figure 3 ) . During human macrophage infection , translocated VipA associates with actin filaments , as well as with early endosomes ( Figure 7 ) . This subcellular localization was also verified when VipA-GFP or VipA-mCherry were ectopically expressed in S . cerevisiae and mammalian CHO cells , showing further that this distribution is independent of other bacterial effectors ( Figures 4 , 5 , S2 and S3 ) . In addition , a mutation in VipA that abolished the interference of the protein in yeast vacuolar trafficking led simultaneously to a decreased affinity and ability to polymerize actin in vitro ( Figures 2 , 3 ) and the ability to associate specifically with the host cell targets ( Figures 4–7 , S2 and S3 ) , indicating a function of VipA linking the actin cytoskeleton and the MVB pathway . Both these functions are new amongst the pool of L . pneumophila effectors characterized to date . Many extra- and intracellular pathogens target host actin as a means to produce a successful infection . This often involves the hijacking of the Arp2/3 complex by recruiting or mimicking nucleation promoting factors ( NPFs ) by a diverse array of mechanisms that serve different purposes . Examples include the movement of Listeria monocytogenes and Shigella within the host cytosol and cell to cell spread by the formation of actin comet tails mediated by the effectors ActA [40] and IcsA , respectively [41] , or the formation of actin pedestals in pathogenic E . coli EPEC and EHEC through the action of the effector Tir [42] , [43] . However , the number of known bacterial effectors able to polymerize actin directly is reduced . Salmonella enterica serovar Typhimurium SipC nucleates actin with an efficiency identical to the Arp2/3 complex , and is also able to bundle and crosslink actin filaments [44] , and in Chlamydia the protein TARP forms long unbranched actin filaments that similarly to SipC seem to facilitate the internalization of the pathogen [45] . Contrary to SipC and TARP , Vibrio parahaemolyticus VopL and VopF polymerize actin much more potently than activated Arp2/3 , leading to the formation of stress fibers [46] or actin-rich filoform formation in infected cells , respectively [47] . More recently , Rickettsia rickettsii Sca2 was found to function as eukaryotic formin mimic and to be involved in actin tail formation [48] . Several observations support a role of VipA as an actin nucleator instead of alternative activities that would also increase actin polymerization , such as decreasing the actin critical concentration , increasing the elongation rate or severing microfilaments . In addition to decreasing the initial polymerization lag phase , VipA increases the actin polymerization rate , which would not happen if its effect was in decreasing the actin critical concentration ( as is the case , for instance , of Salmonella SipA [49]; Figure 3A ) . The enhancement of actin polymerization is also not due to an effect on the elongation rate , which was observable but weak in assays with actin seeds ( Figure 3D ) . Furthermore , a possible filament severing activity of VipA was not visible in Transmission Electron Microscopy experiments ( data not shown ) or assays with actin seeds . When compared to the above mentioned bacterial and eukaryotic actin nucleators , His6-VipA has lower activity , which can be due to several reasons . One possibility is that the VipA protein used in our assays may not be fully active . In fact , like many actin-binding proteins , VipA may contain auto-regulatory regions that inhibit its activity or it may need to be activated by additional bacterial or eukaryotic factor ( s ) absent in our in vitro experiments . We cannot also dismiss the possibility that the presence of the histidine tag may be partially hindering the protein's function . Secondly , the unusually high number of effectors found in Legionella so far ( >300 ) and their observed functional redundancy raises the possibility that additional effectors may act in concert with VipA in actin polymerization . Another explanation could be that the low activity of VipA is tailored to its predicted function in endosomal trafficking during infection . Rather than causing major alterations in the host cell microfilament network , as was proposed for VopL function , VipA may play a more subtle role where its activity is enough to allow interference with membrane traffic but not sufficient to disturb overall cell actin homeostasis . The primary sequence of VipA contains a central coiled-coil region and a C-terminal proline-rich region ( Figure 2A ) . Both motifs are typical mediators of interaction with other proteins , and Pro-rich regions in particular are involved in binding to profilin . However , no typical actin-binding motifs such as the WASP-homology 2 ( WH2 ) domain or Formin-Homology domain 2 ( FH2 ) are present ( reviewed in [50] , [51] ) . The absence of either of these actin-binding motifs in VipA raises the possibility of a novel molecular mechanism of actin assembly . The analysis of bacterial effectors that influence host cell actin dynamics has provided valuable information to understand how eukaryotic NPFs work , as many share similar modes of action and interacting partners . Thus , future functional studies of VipA may not only widen the knowledge concerning virulence factors targeting actin but may also contribute to a broader comprehension of actin dynamics in the eukaryotic cell . Although present in all sequenced genomes of Legionella pneumophila strains ( Philadelphia-1 , Corby , Lens , Paris , Alcoy , 570-CO-H and 130b ) , similarly to most Legionella effectors studied to date VipA is neither required for intracellular replication nor uptake ( Figure S1B; data not shown ) . However , in vivo colocalization experiments in yeast and mammalian cells provided some clues to its function in the eukaryotic cell . First , in both types of cells VipA forms punctate structures that localize in actin-rich regions . In yeast they often associate with the bud-neck , and with endosomal and cortical actin patches markers Abp1 and Abp140 , respectively ( Figure 4 ) . In addition , VipA also colocalized with the MVB-associated protein Bro1 and the perivacuolar compartment observed in mutants defective in MVB formation that results from improper fusion of the endocytosed cargo with the vacuole ( Figure 5 ) . These observations are consistent with the fact that VipA causes defects in this pathway [13] . Similarly , during macrophage infection translocated VipA associates with host cell early endosomes and F-actin , often with the three components forming large agglomerates ( Figure 7 ) . Moreover , and in contrast to many characterized Icm/Dot substrates , the effector is not found in the LCV ( Figure 6 ) . To rule out the possibility that other T4SS substrates could mediate VipA subcellular localization during infection , we analysed its distribution when ectopically expressed in CHO cells and , therefore , in the absence other Legionella proteins . In concordance with the previous results , we observed VipA-EGFP association with cortical actin foci ( Figure S2 ) and with the early endosome marker EEA-1 , demonstrating that VipA does not require other bacterial factors to bind its eukaryotic targets . In addition , no significant colocalization was seen with later components of the Endosomal/MVB Pathway ( MVB component Alix and the lysosome ) or the Secretory Pathway ( ER; Figure S3 ) . The endocytic pathway is responsible for the uptake of particles and converges in early endosomes , where cargo sorting occurs and molecules targeted for degradation in lysosomes are internalized to intraluminal vesicles originating the MVB . This organelle subsequently travels to and fuses with late endosomes and eventually lysosomes . Formation of the MVB requires the Endosomal Complex Required for Transport ( ESCRT ) , and proteins belonging to the Alix/Bro1 family are important components of the MVB , interacting with ESCRT components to control the fission and fusion events taking place in the endosome lumen in mammalian and yeast cells . In addition , Alix associates with actin and several actin-binding proteins and is involved in cytoskeleton assembly , being to date the only eukaryotic link between actin dynamics and MVB formation [52] . In this work , we found that Legionella VipA also connects the two processes . The wild-type protein binds actin and promotes growth of filaments in vitro , associates with host cell early endosomes and actin filaments and leads to defects in the MVB pathway when ectopically expressed in yeast . In contrast , the VipA-1 mutant is unable to bind or polymerize actin , no longer affects trafficking and displays a homogeneous cytosolic distribution . This shows that the ability of VipA to bind actin is related to its association with a specific subcellular location as well as its role in modulating organelle trafficking . In this context , our results are consistent with a model where VipA , through its dual function of regulating actin dynamics and binding to endosomal organelles , may play a role in altering vesicle trafficking in order to enable the pathogen to escape degradation . A possibility is a role in helping isolating the LCV from the endocytic pathway , a process mediated by Icm/Dot substrates [53] . This hypothesis is sustained by the observation of large agglomerates of endosomal vesicles , F-actin and VipA apart from the LCV during macrophage infection ( Figure 7 ) . To our knowledge , VipA is the only bacterial effector that binds early endosomes and is implicated in actin dynamics in the host cell . Thus , further characterization of its mode of action will undoubtedly shed new light on the mechanisms employed not only by L . pneumophila but also by other pathogens to manipulate host cell pathways during infection .
|
Legionella pneumophila is a facultative intracellular bacterium that can cause an often fatal type of pneumonia known as Legionnaires' disease . In nature , L . pneumophila is found in both fresh water and soil where it parasitizes free-living protists . Upon inhalation of contaminated aerosols , L . pneumophila invades and replicates in alveolar macrophages , leading to inflammation and development of the disease . Legionella uses a type IVB secretion system to translocate effector proteins into the host cell that modify its trafficking pathways and prevent fusion of the newly formed phagosome with the lysosome . One of these effectors is VipA , which , when expressed in yeast interferes with the Multivesicular Body ( MVB ) pathway . We found that VipA protein binds actin and nucleates its polymerization without additional host factors . VipA localizes in puncta in eukaryotic cells , and these colocalize with actin-rich regions and endosomes . We demonstrate that the ability to disrupt the MVB is associated with the capacity to bind actin . Thus VipA may contribute to the intracellular lifestyle of L . pneumophila by targeting the cytoskeleton in order to disrupt normal vacuolar trafficking pathways in host cells .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"cellular",
"structures",
"cell",
"biology",
"biology",
"microbiology",
"molecular",
"cell",
"biology"
] |
2012
|
The Legionella pneumophila Effector VipA Is an Actin Nucleator That Alters Host Cell Organelle Trafficking
|
Typical Martsolf syndrome is characterized by congenital cataracts , postnatal microcephaly , developmental delay , hypotonia , short stature and biallelic hypomorphic mutations in either RAB3GAP1 or RAB3GAP2 . Genetic analysis of 85 unrelated “mutation negative” probands with Martsolf or Martsolf-like syndromes identified two individuals with different homozygous null mutations in ITPA , the gene encoding inosine triphosphate pyrophosphatase ( ITPase ) . Both probands were from multiplex families with a consistent , lethal and highly distinctive disorder; a Martsolf-like syndrome with infantile-onset dilated cardiomyopathy . Severe ITPase-deficiency has been previously reported with infantile epileptic encephalopathy ( MIM 616647 ) . ITPase acts to prevent incorporation of inosine bases ( rI/dI ) into RNA and DNA . In Itpa-null cells dI was undetectable in genomic DNA . dI could be identified at a low level in mtDNA without detectable mitochondrial genome instability , mtDNA depletion or biochemical dysfunction of the mitochondria . rI accumulation was detectable in proband-derived lymphoblastoid RNA . In Itpa-null mouse embryos rI was detectable in the brain and kidney with the highest level seen in the embryonic heart ( rI at 1 in 385 bases ) . Transcriptome and proteome analysis in mutant cells revealed no major differences with controls . The rate of transcription and the total amount of cellular RNA also appeared normal . rI accumulation in RNA–and by implication rI production—correlates with the severity of organ dysfunction in ITPase deficiency but the basis of the cellulopathy remains cryptic . While we cannot exclude cumulative minor effects , there are no major anomalies in the production , processing , stability and/or translation of mRNA .
It is 40 years since two brothers were reported with severely delayed neurocognitive development , spasticity , postnatal microcephaly , short stature , congenital cataracts and primary hypogonadism[1] , characterising a disorder that is now termed Martsolf syndrome ( MIM 212720 ) . Warburg Micro syndrome ( MIM 600118 , 614225 , 615222 , 615663 ) is an overlapping condition that was described in 1993 , which also has microphthalmia/microcornea , retinal dystrophy , optic nerve atrophy and intracranial malformations as clinical features[2] . 60% of cases referred to us with a diagnosis of Warburg Micro syndrome have loss-of-function mutations in either RAB3GAP1 , RAB3GAP2 , RAB18 or TBC1D20[3–6] . 44% of Martsolf syndrome cases have mutations in RAB3GAP1 or RAB3GAP2 , which perturb but do not completely abolish the expression or function of the encoded protein[7 , 8] . The relatively high proportion of unexplained cases in both syndromes indicates that there are likely to be more disease loci and/or causative genetic mechanisms to be discovered . Infantile-onset dilated cardiomyopathy ( iDCM ) is a rare , aetiologically heterogeneous disorder that may present as acute , commonly lethal , and with cardiogenic shock [9] . Isolated iDCM may be caused by genetically determined primary abnormalities of heart muscle ( sarcomere , Z-disc , desmosomes etc ) while iDCM as a component of a multisystem disorder is most commonly secondary to an inborn error of metabolism ( Table 1 ) [10–12] with the prognosis being dependent on the underlying cause . Early genetic testing is recommended in iDCM as it may help direct the clinical management[13] . Here we report two families with a very distinctive clinical presentation of lethal iDCM and Martsolf-like syndrome associated with homozygous null mutations in ITPA which encodes inosine triphosphate pyrophosphatase ( ITPase ) . ITPase is an enzyme that functions to prevent incorporation of inosine bases ( rI/dI ) into RNA and DNA by scavenging ITP/dITP in the cell . An autosomal recessive partial deficiency of inosine triphosphate pyrophosphatase ( ITPase ) has been recognised since the late 1960’s via accumulation of inosine triphosphate ( ITP ) in erythrocytes[14] . This is a relatively common trait that is clinically asymptomatic although it may influence sensitivity to certain drugs[15] . The trait is caused by hypomorphic mutations in ITPA ( the gene encoding ITPase ) which affect splicing and/or protein stability[16] . Biallelic loss-of-function mutations in ITPA have recently been reported as the cause of an early infantile encephalopathy ( EIEE35 , MIM #616647 ) [17] . We present data testing and refuting various hypotheses ( summarised below in Fig 6 ) regarding the molecular consequences of ITPase deficiency on the genome , transcriptome and proteome .
In Family 4911 ( Fig 1A ) a maternal uncle ( 4911 V:5 ) and aunt ( 4911 V:7 ) of the proband ( VI:3 ) had been described in a clinical paper as Martsolf syndrome with a previously unreported association with an early-onset cardiomyopathy[18] . The proband in the present study , their nephew , died at the age of 2 years . No post mortem examination was carried out and the exact cause of his death could not be confirmed . Prior to his demise he had been noted to have postnatal onset microcephaly with severe delay in all aspects of his development . He had bilateral cataracts diagnosed at the age of 13 months . Generalised seizures began at the age of 14 months . He was noted to have small genitalia . A clinical diagnosis of Martsolf syndrome was made and he had a negative screen for RAB3GAP1 , RAB3GAP2 , RAB18 and TBC1D20 . His elder brother , two maternal uncles and a maternal aunt all had a very similar pattern of problems ( Table 2 ) and all had died in early childhood with evidence of cardiac failure[18] . Serial echocardiograms in 4911 VI:3 had shown persistent but mild dilation of his left ventricle and he is assumed to have died as a result of the progression of his cardiac disease . In family 5196 ( Fig 1B ) the affected proband ( 5196 III:3 ) was a girl who died at the age of 4 years . She had previously been clinically diagnosed by an experienced clinical geneticist as having Martsolf syndrome on the basis of profound developmental delay , failure to thrive , microcephaly , seizures and congenital cataracts . Screening of the known Martsolf syndrome and Warburg Micro genes was negative . She presented in severe cardiac failure and died shortly after this . She had not previously been suspected of having any cardiac disease . In addition to the known anomalies , a post mortem examination revealed marked dilation of the left ventricle with increased trabeculation and mild fibroelastosis ( Fig 2A–2C ) . Fatty infiltration was noted of the right ventricle ( Fig 2B ) . Neuropathology showed cerebellar atrophy ( Fig 2F ) , microgliosis of dentate and olivary nuclei , vacuolation of white matter ( Fig 2E ) with scattered axonal spheroids ( Fig 2H ) and gliosis of the hippocampus . 5196 III:3 had two maternal uncles who died in infancy ( 5196 II:3 and 5196 II:4; Fig 1B ) who were suspected of having the same disorder although no clinical details were available from either individual . In an effort to identify additional causative genes for Martsolf and Martsolf-like syndromes we used whole-exome sequencing ( WES ) of affected individuals from consanguineous multiplex families ( Family 4911 VI:3; Family 5196 III:3 ) and their healthy parents ( 4911 V:1 & 4911 V:2; 5196 II:1 & 5196 II:2 ) ( Fig 1 ) . These UK families had no known shared relatives but both had recent Pakistani ancestry . The alignment , read depth and estimated heterozygous SNP detection sensitivity of each individual is given in S1 Table[19] . Sequence variants in the probands were filtered using minor allele frequency ( MAF ) of < 0 . 001 , plausibly deleterious consequence and bi-allelic inheritance . The WES data from the unaffected parents was used primarily to confirm biallelic inheritance in their children and to exclude shared homozygous variants . The percentage coverage of the WES capture sequences >15X for the unaffected parents 4911 V:1 and 4911 V:2 was significantly lower than expected ( S1 Table ) but was sufficient for analysis of the candidate high impact variants in their affected son 4911 VI:3 . In 4911 VI:3 , ten rare homozygous variants ( Table 3 ) were reviewed by manual assessment of read quality using IGV2 . 3 software[20] and by Sanger sequencing of selected variants for their segregation in the family . Of these only a nonsense mutation , c . 452G>A , p . Trp151* ( rs200086262 ) in ITPA ( NM_033453 , MIM 147520 ) segregated within the family in a manner consistent for an autosomal recessive disease-causing mutation ( Fig 1A ) . This variant has been previously identified as disease associated[17] and is present in gnomAD ( genome Aggregation Database ) with a minor allele frequency of 0 . 0058% . In the affected individual from Family 5196 ( III:3 ) , homozygosity for an apparently unique 40bp deletion spanning the splice donor site of ITPA exon 7 was detected on WES . Subsequent Sanger sequencing confirmed this to be chr20 hg19 g . 3202531-3202570del; c . 456_488+7del . This deletion is likely to have been microhomology-mediated as a nine base pair perfect repeat is present at the 5’ end of the deleted region and the genomic region immediately 3’ to the breakpoint ( Fig 1D ) . Both parents ( 5196 II:1 and 5196 II:2 ) were heterozygous for this mutation . Western blotting of lysates from lymphoblastoid cell lines ( LCLs ) from 5196 III:3 and her mother showed that ITPA protein was completely absent in the cells derived from the affected girl ( Fig 1E ) . We were not able to identify other plausibly causative genotypes in 5196 III:3 in any known developmental disease genes using our previously described DDG2P diagnostic pipeline [21] Sanger sequencing of ITPA in the remaining members of the cohort of 85 “mutation negative” families[8] revealed no further plausibly disease-associated mutations . The primers used for this analysis are given in S2 Table . ITPA encodes inosine triphosphate pyrophosphatase ( ITPase ) which hydrolyzes both inosine triphosphate ( rI ) and deoxyinosine triphosphate ( dI ) [22 , 23] . Its major function is thought to be to ensure the exclusion of these “non-canonical” purines from RNA and DNA in order to avoid transcript and genome instability . Complete deficiency of ITPase in all tissues would thus be predicted to result in an increase in the incorporation of rI and dI into RNA and DNA respectively . To test this we first purified cellular RNA from a lymphoblastoid cell line ( LCLs ) from 5196 III:3 . This RNA was digested to single nucleotides and analysed using a combination of HPLC and mass spectrometry ( HPLC/MS ) . Using this approach we found that rI was present in RNA at a level of 725±158 SEM nucleotides of rI per 106 nucleotides of AMP in 5196 III:3 , a significantly higher level than in RNA from LCLs derived from either her father ( 17±11 SEM rI:rA x 106 ) or her mother ( 71±60 SEM rI:rA x 106 ) ( Fig 3A ) . This equates to approximately one rI base in every 5500 bases of RNA from the null LCL . We generated Itpa-null mouse embryonic stem ( ES ) cells using CRISPR/Cas9 genome editing [24 , 25] ( Fig 4A , primers encoding the guide RNAs are detailed in S2 Table ) . In these cells , rI was detectable in RNA at 1889 nucleotides rI per 106 nucleotides AMP ± 295 SEM ( Fig 3B ) . To determine if there was a correlation between the level of rI in different tissues in vivo and the organs affected in the human disease , we generated mice heterozygous for Itpa null alleles using direct cytoplasmic injection of Cas9 mRNA and guide RNAs into zygotes . Heterozygous animals were crossed to generate Itpa-null embryos and wild-type littermate embryos . Both genotyping and Western blot analysis ( Fig 4A ) were used to confirm the null status of each embryo ( Fig 4A ) . As in previously reported targeted inactivation of Itpa , we found reduced body size ( with a proportionate reduction in heart size ) in Itpa-null embryos[26] and no other obvious morphological differences compared to wild-type controls ( Fig 4B ) . The level of rI in RNA from Itpa-null hearts ( 10382 nucleotides IMP per 106 nucleotides AMP ± 2008 SD ) was significantly higher than in either Itpa-null brain or kidney ( p<0 . 05 and p<0 . 01 respectively , student’s t-test ) and equated to approximately one rI for every 385 bases of RNA ( Fig 3C ) . rI was present at very low levels in RNA derived from control tissues . The bacterial endonuclease V ( nfi/EndoV ) cleaves DNA at dI bases creating nicks in the dsDNA[27] . Digestion of genomic DNA from control and Itpa-null ES cell lines using EndoV ( New England Biolabs ) followed by alkaline-gel electrophoresis revealed no measurable difference in migration between the samples ( Fig 3D ) . However , a small but reproducible increase in the EndoV sensitivity of mtDNA from Itpa-null ES cells as compared to that in controls was seen ( Fig 3E ) . To assess whether this increased inosine incorporation was associated with increased instability of the mitochondrial genome ( mtDNA ) , we used quantitative PCR ( qPCR ) to compare levels of mtDNA to levels of nuclear DNA . We also used long-range PCR ( LR-PCR ) of mtDNA to look for any increase in the frequency of mtDNA rearrangements . Neither assay showed any differences between Itpa-null cells and controls ( Fig 3F and 3G ) , or between Itpa-null tissues and controls ( S1A and S1B Fig ) . We used Ion Torrent sequencing to detect base substitutions and MinION sequencing to detect large-scale mtDNA rearrangements amplified from control and Itpa-null kidneys and hearts . No differences between wild type and Itpa-null kidney or heart were detected ( S1C and S1D Fig ) . To assess secondary effects of low-level dI incorporation on genome stability , a commercial comet assay kit ( Trevigen ) was used . No increase in DNA strand breaks could be detected in 5196 III:3 compared to 5196 II:2 LCLs , or in Itpa-null compared to wild-type ES cells ( Fig 3H and 3I ) . To assess whether ITPase deficiency had any effect on mitochondrial function , we carried out metabolic tracer analysis on the ES cells using 13C5-glutamine , and conducted functional histopathology on tissue samples . In the tracer experiments , 13C-incorporation into fumarate and citrate was analysed ( S1E Fig ) . The ( m+4 ) isotopologues of both metabolites indicated that ITPA-loss did not affect normal oxidative TCA cycle function and a low level of reductive carboxylation of oxoglutarate , as measured by citrate ( m+5 ) , was again minimally altered upon loss of ITPA . Functional histopathology on the tissue samples was adapted from analyses used in a clinical diagnostic setting . Samples were reacted for cytochrome c oxidase ( COX ) and succinate dehydrogenase ( SDH ) activities , with sequential COX-SDH histocytochemical analyses carried out in order to identify low level , focal COX-deficiency . No differences were seen between control and Itpa-null tissues and no COX-deficient cells were identified in Itpa-null heart ( Fig 4C ) . We compared the transcriptome of control and Itpa-null mouse hearts using the Affymetrix MTA1 . 0 expression microarray . There was very strong concordance between transcript levels in control and Itpa-null samples when all loci were examined together ( Fig 5A and 5B ) or when a subset of loci that have been implicated in dilated cardiomyopathy in mice or humans were examined separately ( Fig 5E ) . When specific cardiac disease genes were examined using ddPCR , modest reductions could be observed in Itpa-null heart tissue ( Fig 5G ) . However it was not possible to determine if reductions of this magnitude would significantly alter cardiomyocyte function and , more importantly , we could not distinguish whether these changes in transcript levels were primary effects or secondary to an early disease process in heart . A generalized effect on transcription produced by increased inosine incorporation into RNA in Itpa-null cells might not be identified on transcriptome analysis if its effects on individual transcripts are proportionate . Therefore , in order to assess any changes in transcription rate or transcript stability , we labelled RNA transcribed over the course of 30 minutes with the ribonucleotide analogue 4-thiouridine ( 4sU ) . 4sU incorporation was assayed immediately after labelling , providing a measure of transcription rate , and then at subsequent time points , at which any changes in the rate of RNA turnover would be revealed ( S2A Fig ) . 4sU incorporation was assayed by biotinylation of its thiol group and then quantification of biotin using a fluorescence-based kit . When RNA was harvested immediately following treatment , 4sU incorporation appeared lower in Itpa-null cells than in control cells . However , reduced incorporation was not seen when an alternative assay utilizing 5-Ethynyl Uridine was used ( S2C Fig ) . No differences in the 4sU content of control and Itpa-null cells were observed at the later time points , suggesting that inosine incorporation does not affect RNA stability globally ( S2A and S2B Fig ) . Label-free quantitative mass spectrometry was performed to examine the whole proteome in control and Itpa-null mouse heart tissue . Apart from the absence of ITPase , no significant differences were detectable on inspection of either the whole dataset or the dilated cardiomyopathy-associated proteins specifically ( Fig 5C and 5E ) .
Bi-allelic loss-of-function mutations were recently reported in ITPA in seven affected individuals from four families with early-infantile encephalopathy , a distinctive pattern of white matter disease evident on brain MR imaging , microcephaly and progressive neurological disease[17] . While no measurement of rI/dI incorporation into RNA or DNA was presented from these cases , the clinical and genetic evidence for causation was compelling in this group of children . Here we have shown that a Martsolf-like syndrome with iDCM , is an allelic disorder . There is also evidence of phenotypic overlap between the disorders as one of the seven affected individuals reported by Kevelam et al . [17] had iDCM and three had early onset cataracts . Taken together with the existing mouse genetic data [26 , 28] , these data strongly support an essential role for ITPase activity in development and maintenance of brain , eye and heart function in mammals . Since 2015 there have been no further reports of severe ITPase deficiency . The severity , the distinctive phenotype and the increasing use of whole exome sequencing in clinical diagnostics make it unlikely that this would be missed . This suggests that ITPase deficiency is genuinely very rare . In gnomAD ( November 2018 ) there are 57 individuals heterozygous for 25 different loss-of-function ITPA alleles . These variants have a combined MAF of 0 . 0003 indicating a minimum carrier frequency of ~1:1672 , which , assuming random mating , would give a minimum expected birth incidence of ~1:11 million for biallelic LOF alleles in ITPA . Interestingly , the c . 452G>A;p . Trp151Ter variant shows evidence of a founder effect in Finland with a carrier rate of 1 in 1200 but this would still predict a minimum birth prevalence of < 1 in 5 million . This presumably explains why both families we have identified are consanguineous . There are obvious candidate mechanisms for a cellulopathy associated with ITPase null state ( summarised in Fig 6 ) . First , instability of the nuclear genome induced by dITP incorporation into DNA ( as seen in E coli[27] ) ; second , instability of the mitochondrial genome via the same mechanism; third , inhibition of RNA polymerase II by rI ( previously demonstrated in vitro[29] ) ; fourth , instability of mature transcripts through EndoV-mediated degradation of rI-enriched mRNA; finally , induction of energy deficiency state due to biochemical perturbation of mitochondrial function . In this paper we have attempted to address each of these and failed to observe any single major effect . The differential incorporation of inosine bases between DNA and RNA is interesting . This may reflect the evolution of efficient DNA surveillance and repair mechanisms to deal with deamination of adenosine bases to form dI with the steady state for inosine in DNA being ~1 per 106 nucleotides[30] . This would suggest that even moderate increases in the incorporation of dITP into DNA in ITPase null cells are likely to be below the limit of detection for the assays used here . In this regard it is significant that we could detect low-level dI incorporation into the mitochondrial genome . Importantly we could detect no effect on either the quantity or structural integrity of mtDNA from hearts from Itpa-null embryos as compared to controls ( S1 Fig ) . The lack of evidence for a DNA-based mechanism taken together with the correlation of rI incorporation in RNA with organ severity suggested to us that there may be a transcriptomic mechanism of disease . Enzymatic A-to-I editing in RNA is used to “recode” specific transcripts in a highly regulated manner[31] and this may explain why there is no rI-induced repair system for RNA . However , human EndoV[30] is capable of cleaving RNA at inosine bases[32 , 33] and over-incorporation of rI could lead to a generalized instability of the transcriptome . We found evidence of a reduction in the transcript abundance of some longer mRNA extracted from Itpa-null mouse embryonic heart . This effect is difficult to interpret given that it is a relatively minor change and is plausibly a secondary effect of the disease process rather than due to rI-induced RNA instability . One way to address this problem would be to create animals who are null for both Itpa and Endov and thus determine if loss of the ribonuclease activity would rescue any or all of the Itpa phenotypes . Although we have not reported the details here , our preliminary work using Itpa/EndoV double KO mouse embryos suggested no reduction in rI incorporation into RNA . It seems probable that the major disease mechanism in severe ITPase deficiency related to either inosine base production or rI incorporation . It is not clear why the heart , brain and developing eye are more sensitive to the perturbation . The modest reductions in RNA levels in the mouse orthologs of two of the known genetic causes of cardiomyopathy in humans ( Fig 5G; Ttn and Ryr2 ) are interesting but difficult to interpret . There no detectable generalised effect on the transcriptome , even for very long transcripts , and thus these reductions in specific transcripts are more likely to represent an early marker of cardiomyocyte disfunction rather than a primary pathogenic mechanism . A major challenge in studying the cellular basis of ITPA-associated disease is the large number of possible consequences of altering the composition of the cellular nucleotide pool . These include intracellular signalling , post-translational modification and energy production in addition to those detailed in Fig 6 . That fact that ITPA-null cells grow at a normal rate with normal morphology may indicate that the perturbation me be individually subtle but collectively have catastrophic consequences in vulnerable tissues such as the brain and heart . A clear understanding of the disease mechanism is important as it may lead to therapies that will ameliorate the progressive cardiac and neurological effects of this rare but important disease .
Our cohort consists of DNA samples from 85 families , submitted by referring clinicians for research screening for Warburg Micro syndrome , Martsolf and Martsolf-like syndromes[8] . Affected individuals in this cohort are negative for causative variants in the coding sequences of RAB3GAP1 , RAB3GAP2 , RAB18 and TBC1D20 , the genes previously associated with these disorders . Informed , written consent has been obtained from both participating families . The consent process and molecular analysis used protocols approved by the Scotland A Multicentre Research Ethics Committee ( 04:MRE00/19; The Genetic Basis of Brain Growth and Development ) in the UK . The mouse work was done under a UK Government Home Office animal licence: 60/4424 . The work was overseen by the University of Edinburgh Animal Welfare and Ethical Review Body ( AWERB ) . DNAs from Families 4911 and 5196 ( nuclear trios ) were enriched for exonic sequence using kits indicated in S1 Table and sequenced using Illumina HiSeq technology as described previously[34] . Sequence reads were aligned to the GRCh37 human genome reference assembly with BWA mem 0 . 7 . 10[35] . Duplicate reads were marked with Picard MarkDuplicates 1 . 126 . Reads were re-aligned around indels and base quality scores re-calibrated with GATK 3 . 3[36] . Single nucleotide variants and small indels were called with GATK 3 . 3 HaplotypeCaller on each sample and GenotypeGVCFs to produce a raw variant call set . Variants were annotated using the Ensembl Variant Effect Predictor[37] . Statistics for alignment , read depth and estimated heterozygous SNP detection sensitivity for each individual are listed in S1 Table . Sequence variants were filtered using minor allele frequency ( MAF ) of < 0 . 001 , plausibly deleterious consequence and bi-allelic inheritance . A screen for plausibly disease associated genotypes associated with known developmental disorder genes was performed using the DDG2P pipeline as previously described[21] . PCR amplification of the coding exons of ITPA and intron-exon boundaries was carried out using flanking primers with M13 tags to facilitate later sequencing ( see S2 Table ) . Primers were designed using ExonPrimer software on the basis of the reference sequence NM_033453 . Sequencing reactions were carried out with BigDye Terminator 3 . 1 reagents ( Applied Biosystems ) , according to manufacturer's instructions . Sequencing data was analysed with Mutation Surveyor software ( SoftGenetics ) . Cells were maintained under 5%CO2 at 37°C . Lymphoblastoid cell lines ( LCLs ) were maintained in suspension in RPMI1640 media ( Gibco ) supplemented with 10% foetal calf serum , 1mM oxalocacetate , 0 . 45mM pyruvate , 0 . 03% glutamine , 1% penicillin/streptomycin and 0 . 2 I . U/ml insulin . Embryonic stem ( ES ) cells were maintained in adherent culture in GMEM supplemented with 10% foetal calf serum , 0 . 1mM non-essential amino acids , 2mM L-Glutamine , 1mM sodium pyruvate and 106 units/L LIF . Rabbit polyclonal antibody raised to full-length ITPA was obtained from Millipore . Rabbit polyclonal antibody raised to an N-terminal portion of ITPA was obtained from LSBio . Goat polyclonal antibody raised to β-tubulin was obtained from Abcam . Cells were lysed in a buffer containing 0 . 5% ( v/v ) Nonidet P-40 in a solution of 150mM NaCl , 10mM EDTA and 50mM Tris-HCl ( pH = 7 . 5 ) to which a protease inhibitor cocktail ( Roche ) was added . Tissue samples were lysed directly in 1x NuPage LDS Sample Buffer ( Thermo Fisher ) containing 5% β-mercaptoethanol . SDS-PAGE and Western blotting were carried out according to standard methods . ECL 2 Western blotting substrate ( Pierce ) was used to produce chemiluminescent signal , HyperFilm ECL ( General Electric ) was developed using a Konica Minolta SRX-101A . Cellular RNA and DNA were purified using RNAeasy ( Qiagen ) and BACC2 ( GE Healthcare ) kits respectively . mtDNA isolation was performed using a mitochondrial DNA isolation kit ( Abcam ) . For analysis of nucleic acid composition by mass spectrometry , digestion to single nucleotides was carried out . A combination of either 50μg/ml RNAseA ( RNA ) or 20U/ml DNAseI ( DNA ) ( Roche Diagnostics ) respectively and 80U/ml NucleaseP1 ( Sigma ) was used as previously described[38] . Both purification and digestion was carried out in the presence of a 20μM concentration of the adenosine deaminase ( ADA ) inhibitor deoxycoformycin ( DCF ) ( Sigma ) . Digestions were carried out in a buffer containing 1 . 8mM ZnCl2 and 16mM NaOAc , pH = 6 . 8 at 37°C overnight . Nucleases were then removed with 10 , 000 MW cut-off spin columns ( Amicon ) . Samples were loaded onto a ZIC-pHILIC column using a Dionex RSLCnano HPLC and the eluate was applied to a Q Exactive mass spectrometer in negative mode . The instrument was operated in tSIM mode and data were quantified using XCalibur 2 . 0 software . For analysis of genomic and mitochondrial DNA composition by Endov-digestion and alkaline-gel electrophoresis , DNA samples were treated with 10 U of Endonuclease V ( NEB ) with the supplied buffer for 2 hours at 37°C . DNA strands were separated by incubation at 55°C in loading buffer containing 3% Ficoll ( type 400 ) and 300mM NaOH . Samples were separated on agarose gels ( 50mM NaOH , 1mM EDTA ) with a solution of 50mM NaOH , 1mM EDTA used as running buffer . After electrophoresis , gels were neutralized and stained with SYBR Gold ( Invitrogen ) . Alkaline comet assays were carried out using the Trevigen CometAssay electrophoresis kit according to manufacturer’s instructions . Briefly , cells were embedded into low melting agarose on comet slides and incubated in lysis solution overnight in the dark at 4°C . They were incubated in a solution of 300 mM NaOH , 1 mM EDTA for 30 minutes at room temperature , then electrophoresed in this solution for 30 min at 21 volts at 4°C in the dark . Comet slides were immersed in 70% ethanol for 10 min at room temperature and dried at 37°C for 15 minutes . They were then stained with 1x SYBR gold in TE buffer ( pH 7 . 5 ) for 30 minutes at room temperature , dried for an additional 15 minutes at 37°C and visualized with a Zeiss Axioskop 2 epifluorescence microscope with a 10x objective . Data were analysed with CaspLab 1 . 2 . 3 software . 10ng DNA samples were amplified using the TaKaRa LA Taq polymerase mix with primers flanking nucleotide positions 272 and 16283 on the mouse mitochondrial genome ( NC_005089; see S2 Table ) . A long PCR template program was used as follows: 94°C– 2 minutes , 35 cycles of 94°C– 30 seconds and 65°C– 16 minutes followed by a final extension of 72°C– 16 minutes . For quantitative PCR ( qPCR ) of mitochondrial ( mtDNA ) and genomic DNA ( gDNA ) , DNA preparations ( retaining both species ) were made from cell and tissue samples using Viagen reagent ( Viagen Biotech ) according to manufacturer’s instructions . For analysis of gene expression by qPCR , RNA was extracted using Trizol reagent together with an RNeasy mini kit ( Qiagen ) according to manufacturer’s instructions . Purified RNA was used immediately as a template for production of cDNA using a First Strand cDNA Synthesis Kit for RT-PCR ( AMV ) ( Roche ) . qPCR analysis was carried out on a LightCycler 480 ( Roche ) . Amplification from mtDNA and gDNA was carried out using pairs of primers designed to amplify from the mtCO1 locus of the mitochondrial genome and the Gapdh locus of the autosomal genome . Amplification from cDNA was carried out using primers designed to amplify from TTN , RYR2 , TNNT2 and GAPDH cDNAs . Amplification from NKX2-5 was carried out using commercial TaqMan probes ( Mm01309813_s1_Nkx2-5 ) . PCR amplification with unlabelled primers was quantified through binding of specific mono color hydrolysis probes ( Roche ) . Data were analyzed using LightCycler 480 software version 1 . 5 . 0 ( SP4 ) ( Roche ) . Primers were designed using the Universal ProbeLibrary Assay Design Center and are listed in S2 Table . Droplet Digital PCR ( ddPCR ) reactions were carried out according to manufacturer's instructions ( Biorad ) . In each reaction , cDNAs were combined with a VIC-labeled TaqMan control probe , mouse GAPD or eukaryotic 18S ( Life Technologies ) . Primers specific for target genes are as above . Droplets were generated using a Biorad QX200 or QX200AutoDG droplet generator , PCRs were carried out using a C1000 Touch Thermal Cycler , and droplets were analyzed on a QX100 Droplet Reader . The data were analyzed using Quantasoft software ( QuantaLife ) . Tissue samples from Itpa-null embryos and littermate controls ( e16 . 5-e18 . 5 ) were frozen in liquid nitrogen-cooled isopentane prior to sectioning . Cryostat sections were stained for individual activities of COX and SDH and also for sequential COX/SDH activity . Briefly , sections were reacted for 45 min at 37°C with COX reaction media ( 4 mM diaminobenzidine tetrahydrochloride , 100 μM , cytochrome c and 20 μg/ml catalase in 0 . 2 M phosphate buffer , pH 7 . 0 ) and 40 min at 37°C with SDH media ( 1 . 5 mM nitroblue tetrazolium , 1 mM sodium azide , 200 μM phenazine methosulphate , 130 mM sodium succinate , in 0 . 2 M phosphate buffer , pH 7 . 0 ) . RNA was extracted from e16 . 5 mouse hearts using Trizol reagent together with an RNeasy mini kit ( Qiagen ) according to manufacturer’s instructions . RNA quality was assessed using an Agilent Bioanalyser instrument and Total RNA nano . RNA integrity numbers ( RIN ) were ≥9 . 1 for all samples . Transcriptome analysis was carried out by Aros Applied Biotechnology A/S using the Affymetrix MTA1 . 0 microarray . Data were analysed using Affymetrix Transcriptome Analysis Console 3 . 0 and custom R scripts . Protein was extracted from embryonic mouse hearts in a buffer containing 8M Urea , 75mM NaCl and 50mM Tris , pH = 8 . 4 by sonication at 0–4°C in a Bioruptor device ( Diagenode ) together with silica beads . Protein concentrations were quantified using a BCA assay ( Pierce ) and then 100μg of each sample was subjected to in-solution tryptic digest . Samples were loaded onto a C18 column using a Dionex RSLC Nano HPLC and the eluate was applied to a Q Exactive mass spectrometer . The data were quantified using XCalibur 2 . 0 software . Itpa-null mouse ES cells were generated using CRISPR/Cas9 genome editing [24 , 25] . Paired guide RNA ( gRNA ) sequences were selected using the online CRISPR design tool ( http://crispr . mit . edu/ ) . Oligonucleotides encoding these sequences ( S2 Table ) were annealed and ligated into pX461 and pX462 plasmids ( Addgene ) . Recombinant plasmids were verified by direct sequencing . For each targeted locus , the E14 ES cells were transduced with 1μg of each vector using the Neon system ( Life Technologies ) according to manufacturer's instructions . Cells were allowed to recover for 24 hours , then treated for 24h with 1 μg ml−1 puromycin in order to select for cells containing the px462 construct . To select single cells also containing the px461 construct , fluorescence activated cell sorting into 96-well plates was carried out using a FACSJazz instrument ( BD Biosciences ) . Clonal cell lines were analysed by direct sequencing of targeted alleles and by Western blotting . Sequencing primers are shown in S2 Table . To facilitate sequence analysis , PCR products were cloned into pENTR/D-TOPO vectors prior to sequencing . Cytoplasmic zygotic injection of wild-type Cas9 mRNA together with in vitro transcribed gRNAs was used to generate Itpa-null mouse embryos . This approach was also used to produce heterozygous-null animals used to establish transgenic mouse lines . The plasmid vectors described above were used as a template for PCR amplification together with forward primers incorporating T7 promoter sequences and a universal reverse primer ( see S2 Table ) . RNA was synthesised using a HiScribe T7 High Yield RNA Synthesis Kit ( New England Biolabs ) according to manufacturer’s instructions . DNA preparations were made from tissue samples using Viagen reagent ( Viagen Biotech ) . Genotyping was carried out by direct sequencing of targeted alleles and by Western blotting as above . Following initial genotyping of Itpa-null animals produced by crossing heterozygous-null founders , subsequent genotyping of transgenic lines was conducted through PCR analysis . E16 . 5 mouse embryos were mounted in 1% agarose , dehydrated in methanol and then cleared overnight in a solution containing 1 part Benzyl Alcohol and 2 parts Benzyl Benzoate . Imaging was conducted with a Bioptonics OPT Scanner 3001 ( Bioptonics , UK ) using brightfield analysis to detect tissue autofluorescence for capture of anatomical and signal data ( wavelengths: excitation at 425 nm , emission: 475 nm ) . The resulting data were reconstructed using Bioptonics proprietary software ( Bioptonics , MRC Technology , Edinburgh , UK ) , automatically thresholded to remove background signal , then merged into a single 3D image output using Bioptonics Viewer software . Measurements of internal chest cavity diameter , maximum heart diameter , cardiac wall cross-sectional area and total heart cross-sectional area were taken for five embryos per genotype .
|
Nucleotide triphosphate bases containing inosine , ITP and dITP , are continually produced within the cell as a consequence of various essential biosynthetic reactions . The enzyme inosine triphosphate pyrophosphatase ( ITPase ) scavenges ITP and dITP to prevent their incorporation into RNA and DNA . Here we describe two unrelated families with complete loss of ITPase function as a consequence of disruptive mutations affecting both alleles of ITPA , the gene that encodes this protein . Both of the families have a very distinctive and severe combination of clinical problems , most notably a failure of heart muscle that was lethal in infancy or early childhood . They also have features that are reminiscent of another rare genetic disorder affecting the brain and the eyes called Martsolf syndrome . We could not detect any evidence of dITP accumulation in double-stranded DNA from the nucleus in cells from the affected individuals . A low but detectable level of inosine was present in the circular double-stranded DNA present in mitochondria but this did not have any obvious detrimental effect . The inosine accumulation in RNA was detectable in the patient cells . We made both cellular and animal models that were completely deficient in ITPase . Using these reagents we could show that the highest level of inosine accumulation into RNA was seen in the embryonic mouse heart . In this tissue more than 1 in 400 bases in all RNA in the cell was inosine . In normal tissues inosine is almost undetectable using very sensitive assays . The inosine accumulation did not seem to be having a global effect on the balance of RNA molecules or proteins .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods",
"and",
"models"
] |
[
"glycosylamines",
"medicine",
"and",
"health",
"sciences",
"mitochondrial",
"dna",
"cardiovascular",
"anatomy",
"computational",
"biology",
"developmental",
"biology",
"inosine",
"genome",
"analysis",
"forms",
"of",
"dna",
"energy-producing",
"organelles",
"mammalian",
"genomics",
"dna",
"bioenergetics",
"mitochondria",
"embryos",
"cellular",
"structures",
"and",
"organelles",
"molecular",
"biology",
"techniques",
"research",
"and",
"analysis",
"methods",
"embryology",
"artificial",
"gene",
"amplification",
"and",
"extension",
"molecular",
"biology",
"animal",
"genomics",
"biochemistry",
"anatomy",
"nucleic",
"acids",
"cell",
"biology",
"polymerase",
"chain",
"reaction",
"nucleosides",
"genetics",
"transcriptome",
"analysis",
"biology",
"and",
"life",
"sciences",
"genomics",
"glycobiology",
"heart"
] |
2019
|
ITPase deficiency causes a Martsolf-like syndrome with a lethal infantile dilated cardiomyopathy
|
As a neutrophilic bacterium , Helicobacter pylori is growth deficient under extreme acidic conditions . The gastric pathogen is equipped with an acid survival kit , regulating urease activity by a pH-gated urea channel , opening below pH 6 . 5 . After overcoming acid stress , the bacterium’s multiplication site is situated at the gastric mucosa with near neutral pH . The pathogen exhibits exceptional genetic variability , mainly due to its capability of natural transformation , termed competence . Using single cell analysis , we show here that competence is highly regulated in H . pylori . DNA uptake complex activity was reversibly shut down below pH 6 . 5 . pH values above 6 . 5 opened a competence window , in which competence development was triggered by the combination of pH increase and oxidative stress . In contrast , addition of sublethal concentrations of the DNA-damaging agents ciprofloxacin or mitomycin C did not trigger competence development under our conditions . An oxygen-sensitive mutant lacking superoxide dismutase ( sodB ) displayed a higher competent fraction of cells than the wild type under comparable conditions . In addition , the sodB mutant was dependent on adenine for growth in broth and turned into non-cultivable coccoid forms in its absence , indicating that adenine had radical quenching capacity . Quantification of periplasmically located DNA in competent wild type cells revealed outstanding median imported DNA amounts of around 350 kb per cell within 10 min of import , with maximally a chromosomal equivalent ( 1 . 6 Mb ) in individual cells , far exceeding previous amounts detected in other Gram-negative bacteria . We conclude that the pathogen’s high genetic diversity is a consequence of its enormous DNA uptake capacity , triggered by intrinsic and extrinsic oxidative stress once a neutral pH at the site of chronic host colonization allows competence development .
The human stomach is a hostile niche . After overcoming the acid barrier , bacterial colonization at the gastric mucosa represents a constant battle with the host immune system . In particular , the pathogen is confronted with ROS-mediated oxidative stress produced by NADPH oxidases that release superoxide ( O2- ) [1] . Progression of disease towards carcinogenesis is also accompanied by DNA damage of host cells [2 , 3] . Helicobacter pylori are genetically highly variable , microaerobic Gram-negative bacteria that successfully colonize the human gastric mucosa of half the world’s population , with variable prevalence in different human populations [4] . The bacterium causes lifelong gastritis and is also associated with mucosal-associated lymphoma and gastric adenocarcinoma [5] . H . pylori is a neutrophilic organism , but highly adapted to acid survival . For this purpose , the pathogen has a potent urease enzyme , which is tightly regulated via substrate accessibility by a pH-regulated channel [6 , 7] . The microaerobic pathogen is also equipped with a set of oxidative stress enzymes , including catalase ( KatA ) , alkyl hydroperoxide reductase ( AhpC ) and superoxide dismutase ( SodB ) , which are important for host colonization [8–10] . H . pylori is not only exceptional regarding its colonization capacity of a harsh environment , but also regarding its enormous genetic diversity , crucially based on horizontal gene transfer and high recombination frequency [11] . In H . pylori horizontal gene transfer can be accomplished by natural transformation [12] , i . e . the capacity to take up naked DNA from the environment . However , also direct exchange of DNA between H . pylori cells by conjugation-like mechanisms were shown to occur [13–15] and were discussed to play a yet underestimated role in genetic diversity of the gastric pathogen [16] . In contrast to all other known bacteria , DNA uptake in H . pylori during natural transformation is established by a type IV secretion system that is encoded by two separate operons , comB2-B4 and comB6-B10 [12 , 17] . By tracking fluorescent DNA during uptake , we previously showed that DNA uptake in H . pylori is a two-step process [18] . The ComB system mediated transport of external dsDNA over the outer membrane into the periplasm . Our data also indicated that the conserved inner membrane localized ComEC channel , previously described by [19 , 20] is involved in subsequent transport of probably ssDNA into the cytoplasm of H . pylori [18] . Monitoring fluorescent DNA during natural transformation was subsequently applied to other Gram-negative bacteria , like Vibrio and Neisseria [21 , 22] . A maximum of 40 kb of imported DNA was detected in the bacterial periplasmic space of Neisseria , limited by the binding capacity of the periplasmic DNA-binding protein ComE ( ComEA homologue ) [21] , which is absent in H . pylori [23] . In most bacteria , the capacity for natural transformation is tightly regulated , by timely restricting competence phase and/or by preferential uptake of DNA from related species . Quorum sensing for induction of competent state enables Bacillus subtilis , Streptococcus pneumoniae and Vibrio cholerae to increase the chance of import of DNA from siblings [24–26] . In Streptococcus pneumoniae , natural transformation was additionally triggered by antibiotics or DNA-damaging agents [27] . In Vibrio competence is also regulated by the presence of the natural habitat marker chitin and regulatory circuits appeared to be interconnected [28] . In Neisseria species-specific DNA uptake sequences [29] recognized by an outer membrane receptor [30] select for DNA of related bacteria . A DNA uptake sequence also facilitated DNA uptake in Haemophilus influenza [31] , with variants of this sequence and core bases recently detailed by a next generation sequencing approach [32] . Furthermore , H . influenzae was suggested to restrict competence state to conditions of starvation [33] . A previous study implicated also sequence bias to occur in H . pylori natural transformation [34]–probably at the level of recombination . However , single cell analysis clearly showed that H . pylori did not discriminate between own and foreign DNA at the level of DNA uptake [18] , which confirmed previous predictions of absence of redundant sequence motifs in the gastric pathogen [35] . In addition , natural transformation was long thought to occur constitutively in H . pylori . However , multiple competence phases and strain-dependent differences in natural transformation were described [36] . A study suggested that DNA damage exerted by the fluoroquinolone ciprofloxacin triggered natural transformation [37] . Interestingly , increased transformation rate was observed for H . pylori grown on agar plates incubated at slightly increased pH or different atmospheric composition [38] . Using a single cell approach , tracking fluorescent DNA in competent bacteria , we show here that H . pylori tightly regulates its competence state . A pH above 6 . 5 opened a window of opportunity , in which competence development was triggered by further increase in pH in combination with oxidative stress . Furthermore , single cell quantification of imported DNA provided evidence of a stunning DNA uptake activity , providing an explanation for the pathogen’s outstanding genetic diversity .
In previous studies H . pylori competence was usually quantified by monitoring the expression of a resistance marker after contact with external DNA [36 , 39] . This assay is not only dependent on the competent state but also on downstream effects like timely expression of the cassette and survival of the microaerobic bacterium under selection conditions . In addition to resistance marker expression , the transcription of a subset of genes essential for natural transformation was characterized in other studies [37 , 38] . Here we directly monitor active DNA uptake by covalently labelling λ DNA with a Cy3 fluorophore and microscopically tracking DNA import in single cells . We showed previously that active DNA uptake of covalently modified DNA , such as Cy3-labelled λ DNA , occurred over the outer membrane at distinct locations and that DNA was trapped in the periplasm [18] . In this study , we wanted to use the assay for direct visualization of the competent state of H . pylori . Therefore , we first checked that the fraction of cells with incorporated fluorescent DNA , i . e . with at least one DNA focus , corresponding to active DNA uptake activity over the outer membrane , correlated with transformation rate . Thus , in parallel to the fluorescent assay counting cells with active DNA uptake activity , a second aliquot of the same cell suspension was incubated with a 1 , 022 kb DNA fragment harboring the rpsL ( A128G ) mutation for streptomycin resistance in order to measure natural transformation rate . We took cells suspensions from different growth phases before , during and after competence development as described below . As expected , the number of cells with one or more DNA foci correlated exponentially with the log number of transformants . Transformation rate ranged between 2 x 10−6 and 2 x 10-1 relative to total CFU ( Fig 1 ) . The lower limit of counted bacteria with active DNA transport was 0 . 2% ( n ≥ 500 cells ) . When 10% cells with DNA foci were detected , transformation rate was around 1% ( 10-2 ) . At a competent cell fraction of approximately 30% we observed a saturation of transformation rate of around 10−1 in H . pylori N6 . As control and previously published for H . pylori strain J99 [18] , N6 mutants with a defect in one of the essential proteins of the ComB system ( comB4 or comB6 , [12 , 17] ) were impaired for natural transformation and displayed absence of active DNA uptake . These data reveal an enormous range of competence status in H . pylori spanning an order of log 5 . The dynamics were controlled by variation of the amount of cells expressing active DNA uptake complexes . Hence , our assay of direct measurement of active DNA uptake was proven suitable for monitoring competence development . Please note that highly transformable bacteria ( transformation rate 10−1 to 10−2 tested using the streptomycin resistance point mutation marker ) showed reduced transformation rate , when other resistance markers were used . When these cells were transformed in parallel with a chloramphenicol resistance cassette flanked by approximately 500 bp of homologous up- and downstream regions , 7 to 170-fold ( median of 13-fold , n = 8 ) reduced transformation rates were observed . As expected , the transformation rate of the replicative vector pILL2157 [40] was log 3 . 8–6 . 4 less ( median of log 5 . 9 , n = 6 ) compared to the point mutation marker . Homologous recombination can occur in very short homologous regions [39] . Therefore , the chance for integrating a point mutation marker as compared to a complete resistance cassette is more likely , considering genetic barriers , in particular , restriction modification systems . A replicative plasmid has to reconstitute completely in the cytoplasm after single strand DNA uptake . Thus , the likeliness for full reconstitution of a plasmid is even less , compared to successful recombination of a resistance cassette with homology regions . Therefore , the data indicate that these highly transformable bacteria had substantial and to some extent variable genetic barriers that reduced the integrity of incoming DNA . These comprise , in particular , restriction modification systems , which were previously studied [41] . It also confirmed that the point mutation marker most faithfully detected DNA uptake . For H . pylori the term “panmictic population structure” was used to express its enormous genetic variability [42] . We wondered if DNA import activity was also unusually high . For this purpose , we exploited the fact that H . pylori imports DNA independent of its source [18] and used DNA of the bacteriophage λ as standard for a 48 . 5 kb molecule . As mentioned above , we covalently labelled λ DNA with Cy3 . In order to visualize the DNA molecules in a spread out way , a drop of DNA was added on an agarose pad and spread over the surface under concurrent air flow . Subsequently , the pad was sealed by a cover slip and mounted for microscopy . Under our experimental conditions , we expected that the chance for fully spread out λ DNA molecules is low . Hence , total fluorescence intensity of single Cy3 λ DNA molecules with a minimal length of ½ λ DNA ( equals 8 . 6 μm , [43] , Fig 2A ) was determined from at least n = 120 molecules . This cut-off minimized the risk to quantify eventual breakage products and can maximally lead to a theoretical error of a factor two . However , the standard deviation of the measured fluorescence intensities was between 20 . 2% and 21 . 3% , confirming that we indeed measured mainly complete single molecules . Bleaching controls and linearity tests with different exposure times revealed high stability of the Cy3 fluorophore and guaranteed that bleaching was negligible under the experimental conditions ( S1 Appendix , Fig A ) . Microaerobically grown competent H . pylori were challenged with the same lot of Cy3 λ DNA ( 1 μg/ml ) in TSB-FBS for 10 min under microaerobic conditions at 37°C before DNaseI treatment for 5 min was performed ( Fig 2B ) . The amount of imported DNA was extraordinarily high in H . pylori cells . For strain J99 , most of the cells had imported more than 250 kb ( Fig 2C ) . We calculated a median value of 108 kb per DNA focus and 351 kb per cell . The maximum DNA amount per focus was 758 kb and 1 . 6 Mb per cell . Nearly 20% of all cells harbored 4 or more locally distinct DNA foci ( Fig 2D ) . Hence , individual cells were capable of import of a chromosomal equivalent within short periods of time . The values for strain N6 were somewhat lower but still impressive . Here , most of the cells imported more than 50 kb , with a median of 124 kb . The median DNA amount in single foci was 55 kb . Maximum import of 392 kb per focus and 507 kb per cell were observed in strain N6 . Thus , in contrast to other Gram-negative bacteria [21] , the periplasm of H . pylori had a manifold higher maximal capacity for imported DNA . We analysed whether the DNA packaging density within single foci was also outstanding . Assuming that the fluorescence intensity in the analyzed regions of interest ( ROIs ) roughly correlated with the localization of DNA , we estimated the packaging density of periplasmic DNA by measuring DNA focus area . As a result , most of the imported DNA was packed in an area of 0 . 1–0 . 5 μm2 per λ molecule ( S1 Appendix , Fig B ) . Minimally , we detected an area of 0 . 04 μm2 in J99 . Assuming a spherical dimension of a DNA focus , this maximum package density corresponded to 0 . 006 μm3 per λ molecule , a DNA density of around 100-fold less than established in the icosahedron capsid of bacteriophage λ with an inner diameter of 55 nm [44] . Since in other bacteria , DNA uptake was suggested to putatively play a role as nutrient supply , in particular in purine acquisition [45] , we checked whether H . pylori competent bacteria can use external DNA as a purine source . For this purpose , H . pylori was grown in TSB-FBS and it was checked that at least 50% of the cells exhibited competence before the medium was exchanged for minimal medium without any purine source but with a supplementation of 10 μg/ml of DNA ( H . pylori N6 genomic DNA or salmon sperm DNA ) . As control , growth in minimal medium was restored by addition of at least 5 μg/ml adenine ( corresponding to 37 μM ) and higher adenine levels ( 50 μg/ml ) did not lead to enhanced growth ( S1 Appendix , Fig C ) . However , when we added genomic DNA of H . pylori or of salmon sperm , the competent bacteria were not able to grow in minimal medium lacking adenine . A substantial fraction of bacteria was still capable to import DNA after 18–24 h of incubation in minimal medium as checked by transient detection of periplasmic DNA that was non-covalently stained by YOYO-1 and loss of signal within 60 min of incubation in the presence of external DNase ( S1 Appendix , Fig D ) . Using a mutant capable of DNA import into the periplasm but deficient of DNA import into the cytoplasm ( ΔcomEC ) , the YOYO-1 signal had been shown to be rather stable in the periplasm of this mutant , thus , suggesting that complete loss of YOYO-1 signal in the wild type was due to import of DNA into the cytoplasm [18] . In addition , import of DNA into the cytoplasm was indirectly shown by transformability of the cells after incubation in minimal medium for 18–24 h using the rpsL ( A128G ) point mutation marker ( transformation rate between 2 x 10−2 and 1 . 4 x 10−1 ) . The results indicate that DNA was imported but could not complement for purine deficiency . H . pylori encounters different pH values in its habitat , the human stomach . We tested the pH dependency of DNA uptake . The pH of BB-FBS was titrated at 37°C with HCl or NaOH to values between 5 . 0 and 7 . 5 in 0 . 5 unit steps and pH stability of these media was confirmed to be within ± 0 . 1 pH units . In order to keep the pH constant , the 10 min incubation was performed at 37°C under aerobic atmosphere , since higher CO2 concentrations used for microaerobic conditions led to pH decrease ( see below ) . We started from a competent cell suspension exhibiting 63% ± 7 . 4% competent cells , which we defined as 100% ( Fig 3 ) . These values were obtained by incubation of the cells under our standard DNA uptake conditions in TBS-FBS at 37°C , except that the incubation was performed under aerobic conditions ( when TSB-FBS exhibits pH 7 . 5 ) . Changing the medium to BB-FBS with a pH of 7 . 5 demonstrated that medium did not influence the fraction of competent cells . As illustrated in Fig 3 , the activity of DNA uptake was highly sensitive towards slightly acidic pH values . Below pH 6 . 5 , DNA uptake complexes shut down activity and were practically inactive at pH 5 . 5 . Interestingly , preincubation at pH 5 . 0 for 10 min and subsequent shift to pH 7 . 5 reactivated DNA uptake in H . pylori ( Fig 3 ) . Hence , we conclude that a neutral pH is a prerequisite of DNA uptake activity in H . pylori . In addition , the pH dependent regulation was tight , impairing DNA uptake at slight acidic conditions . Therefore , DNA uptake activity most likely occurs under neutral conditions , e . g . when H . pylori is in direct contact with the gastric epithelial cells . We next monitored competence development , defined by the fraction of cells with active DNA uptake . Cells were grown microaerobically in BB-FBS overnight to an optical density of OD600 0 . 19 ± 0 . 05 ( t0 ) , i . e . conditions of exponential growth phase . At this low OD600 H . pylori hardly exhibited any active DNA uptake complex ( 2 . 2 ± 2 . 86% ) . Between an optical density of OD600 0 . 39 ± 0 . 05 ( t1 ) and 0 . 7 ± 0 . 11 ( t2 ) , around 50% of the cells switched into the competent state ( Fig 4 ) . Concomitantly , the pH of the medium rose by 0 . 2 units during this incubation period ( Fig 4 ) . In order to establish the parameters triggering competence development , we changed conditions at time point t0 and monitored alterations of the amount of the competent fraction at time points t1 and t2 . Exchange of supernatant by fresh medium at t0 before the switch into the competent state reduced the fraction of competent cells at time point t2 ( when the cells reached an OD of ~0 . 7 , Fig 4 , lower panel ) . Exchange of medium with fresh medium containing DNA-damaging compounds 0 . 125 μg/ml of ciprofloxacin or 0 . 025 μg/ml of mitomycin C , concentrations , which slowed down growth , did not lead to enhanced competence development . Instead , these conditions were similar to exchange with fresh medium without supplements ( Fig 4 , lower panel ) . Interestingly , addition of 0 . 5 mM adenine at t0 inhibited the switch into competence state ( competent fraction decrease of -36 . 3 ± 11 . 2% at t2 ) , while the addition of 0 . 5 mM glutamine had no effect ( Fig 4 , lower part ) . Note that lower concentrations of adenine ( 0 . 05 mM ) , which were sufficient for purine complementation in minimal medium , had no effect on competence development ( S1 Appendix , Fig E ) . This indicated that the effect of adenine on competence development was distinct from that of purine deficiency . A stimulating effect on competence development was observed by lowering incubation temperature by 5°C to non-optimal growth conditions ( competence fraction increase of +37 . 2 ± 16 . 6% at t1 and +14 . 3 ± 17 . 3% at t2 relative to control cells ) . Compared to the control , the above mentioned conditions did not alter pH during incubation relative to the control . Exchange of medium by fresh BB-FBS accounted on average for -0 . 03 ± 0 . 03 pH units ( maximum -0 . 07 ) at t2 relative to the control , while -0 . 01 ± 0 . 03 pH units deviation ( maximum -0 . 05 ) were observed for addition of adenine and -0 . 07 ± 0 . 03 pH units ( maximum -0 . 15 ) upon lowering the temperature for 5°C . These small pH deviations have unlikely caused the observed effects on competence development , in particular , because lowering the temperature by 5°C resulted in an increase rather than a decrease of the amount of competent cells at t2 . The most pronounced effect on competence development was detected upon exposure to aerobic conditions ( competent fraction increase of +67 ± 14% within the much shorter time period of 1 h compared to ~3–4 hours at time point t1 ( Fig 4 , lower panel , black bar ) ) . Under the latter condition , we did observe a significant rise in pH , probably due to loss of CO2 from the medium upon switch from microaerobic to aerobic conditions ( initial culture pH at t0 of 6 . 7 ± 0 . 06 and a final pH of 7 . 2 ± 0 . 06 after 1 h of aerobic incubation , see also below ) . In order to get an idea of the kinetics of competence development under aerobic conditions , samples were analyzed after 15 , 30 , 60 and 120 min ( Fig 5 ) , starting with a competent fraction of 2 . 2 ± 1 . 7% . H . pylori cannot grow under normal atmospheric conditions but survival was observed at least 2 h without loss of CFU ( log CFU/ml at t0 was 8 . 2 ± 0 . 1 and at t2h 8 . 0 ± 0 . 7; n = 5 ) . After 30 min of exposure to oxidative stress , already 38 . 9 ± 8 . 1% of the cells were capable of import of external DNA . After 60 min this fraction increased to 65 . 6 ± 14 . 8% and reached a maximum of 81 . 9 ± 7 . 9% after 120 min . We observed that upon 30 min of oxidative stress most of the competent cells exhibited more than one DNA focus , with over 20% of the competent cells harboring more than four visually distinct locations of active DNA uptake after 2 h of oxidative stress . This confirmed enormous DNA uptake capacity . Selected samples were measured for transformation rate , using the point mutation marker rpsL ( A128G ) . We confirmed integration of the marker into the chromosome by recombination in agreement with the data in Fig 1 . In order to dissect the role of pH and oxidative stress in triggering competence development , non-competent cells were exposed to aerobic conditions in BB-FBS for 1 h establishing various initial pH . As observed before , 64 . 3 ± 13 . 1% of the cells developed competence in non-titrated BB-FBS , with an initial culture pH at t0 of 6 . 7 ± 0 . 06 and a final pH of 7 . 2 ± 0 . 06 after 1 h of aerobic incubation ( Fig 6 ) . When the initial pH was titrated to 6 . 2 ± 0 . 09 before exposure to aerobic conditions and reached a final value of 6 . 6 ± 0 . 06 after 1 h , competence development was substantially inhibited ( only 6 . 4 ± 7 . 5% competent cells ) . These results suggested that slight acidic pH prevented competence development . Furthermore , when we exchanged the medium with fresh BB-FBS titrated to an initial pH of 7 . 5 and incubated microaerobically for 1 h ( corresponding to pH 7 . 2 ± 0 . 04 after equilibration of CO2 ) , a significantly lower fraction of 33 . 3 ± 3 . 5% switched into the competence state compared to control cells under aerobic atmosphere within 1 hour ( Fig 6 ) . Note that the initial pH under microaerobic conditions was higher ( pH 7 . 2 after CO2 equilibration ) than in case of the aerobic condition ( pH 6 . 7 rising to 7 . 2 during 1 h ) . From these data we conclude that competence development was not only pH-dependent but also triggered by the level of oxidative stress . Consistently , when oxidative stress was further reduced by using an atmosphere of 1% O2 , 10% CO2 and 89% N2 and the same pH medium as for the microaerobic condition , the fraction of competent bacteria was further reduced ( 23 . 3 ± 2 . 5% , Fig 6 ) . Likewise , using TSB-FBS as a different growth medium , which had a slightly higher pH value ( pH 6 . 8 ± 0 . 04 without H . pylori preincubation ) than BB-FBS ( pH 6 . 5 ± 0 . 04 without H . pylori preincubation ) , competence development under microaerobic conditions occurred earlier and was also inhibited by 0 . 5 mM adenine . At OD600 of 0 . 18 ± 0 . 05 ( corresponding to t0 ) already 30 . 3% ± 15 . 1% of the cells were competent in TSB-FBS compared to 11 . 4% ± 8 . 3% in the presence of adenine ( OD600 of 0 . 20 ± 0 . 07 ) . At t1 ( OD600 of 0 . 37 ± 0 . 05 ) the majority of bacteria ( 62 . 6% ± 6 . 2% ) had switched into the competent state , whereas a reduced fraction of 37% ± 5 . 7% displayed competence when adenine was present ( OD600 of 0 . 35 ± 0 . 03 ) . Growth rates were comparable in both media . A final experiment was conducted confirming that pH was a prerequisite of competence development and fine-tuned by the degree of oxidative stress . In this experiment , we titrated BB-FBS to distinct initial pH values ( pH 6 . 6 , 6 . 8 , 6 . 9 and 7 . 0 after equilibration of CO2 under microaerobic conditions ) and exchanged the medium of a non-competent cell suspension at t0 with these fresh media for further microaerobic incubation ( S1 Appendix , Fig F ) . Note that the overall competence development under microaerobic conditions is slower than under aerobic conditions , confirming the results depicted in Fig 6 . As observed for the control cells , the relative pH increased for about 0 . 2 units ± 0 . 05 during microaerobic growth over 6 hours . We observed that the increase of competent cells rose with increasing initial pH values ( S1 Appendix , Fig F ) . For each pH condition , addition of 0 . 5 mM adenine inhibited competence development as observed before . The absolute effect of adenine inhibition was less with rising pH , putatively indicating that sensitivity towards oxidative stress might be increased with rising pH . Taking together , the combined results are consistent with a competence window opening at a pH above 6 . 5 . Subsequently , the combination of pH and oxidative stress appeared to define the kinetics of competence development in H . pylori . We further wanted to confirm that oxidative stress is indeed part of the trigger for competence development . Therefore , we intended to pinpoint the effect of adenine as radical quencher and constructed an oxygen-sensitive mutant of H . pylori . The oxidative stress enzyme superoxide dismutase ( sodB ) detoxifies superoxide radicals [46] . A mutant lacking SodB is confronted with higher oxidative stress because of oxygen radicals produced during microaerobic respiration . Likewise , sodB mutants of H . pylori were oxygen-sensitive and defective in mice colonization [10] . We constructed a mutant lacking SodB in H . pylori N6 . Resuscitation of transformants at common microaerobic atmosphere ( 5% O2 , 10% CO2 , 85% N2 ) failed . However , transformants were obtained on Columbia blood agar at lower oxygen tension ( 1% O2 , 10% CO2 , 89% N2 ) . Also subculturing sodB mutants on Columbia blood agar under 5% oxygen were unsuccessful , suggesting that indeed the mutant exhibited an oxygen-sensitive phenotype . In liquid culture ( BB-FBS ) , where cells are dispersed and oxygen stress might be increased the mutant ceased to grow after some doublings and transformed into non-cultivable coccoid forms even at reduced oxygen level of 1% ( Fig 7A ) . When we added 0 . 5 mM of adenine to BB-FBS , growth was partially complemented; the sodB mutant maintained its rod-shaped morphology and was re-cultivable ( Fig 7B ) . After 18–24 h in BB-FBS at 1% oxygen , CFU were log 6 . 7 ± 0 . 5 in the absence of adenine and partially restored to 8 . 4 ± 0 . 3 in the presence of adenine ( Fig 7C ) , which is consistent with the microscopic observation of coccoid formation in the absence of adenine . Complementation of the sodB deletion was performed by integration of sodB into the fliP locus ( sodBcompl ) . The fliP locus was previously used for insertion of a gentamycin cassette for the creation of a non-motile mutant and proven not to interfere with competence [18] . CFU of the wild type and the sodBcompl in the absence or presence of adenine was similar ranging between log 8 . 5 and 8 . 9 ( Fig 7C ) . Competence development occurred at lower OD values in the sodB mutant than in the wild type under similar conditions ( Fig 7D ) . The sodBcompl mutant showed an intermediate competence phenotype between the deletion mutant and the wild type , indicating that the deletion of sodB at its native locus was only partially complemented by insertion of sodB in the fliP locus . As shown before , addition of adenine resulted in a decrease of the competent fraction of cells . Note that consistently , the wild type also displayed lower competence development at 1% oxygen ( Fig 7 ) as compared to 5% oxygen ( Fig 4 ) . The data support the following conclusions: ( i ) competence development during growth seemed to be triggered by oxidative stress and , ( ii ) , adenine appeared to have an oxygen-protective role , probably as radical quencher . Taking together , the data indicate that the opportunity for competence development was established at pH values above 6 . 5 . Within this window , the combination of pH and oxidative stress triggered competence development . This might support the hypothesis that intrinsic ( caused by respiration ) as well as extrinsic oxidative stress ( caused by oxygen tension and/or host inflammatory response ) contribute to the impressive genetic diversity of the pathogen , colonizing close to the neutral gastric mucosa .
We observed that competence development occurred in liquid medium with increasing optical density . Exchange of the medium by fresh medium before the switch to the competent state inhibited competence development . A regulation by quorum sensing , however is not likely , as a similar effect was observed by adding adenine , conditions during which a putative autoinducer is not eliminated ( Fig 4 ) . In order to understand the effect of adenine ( or fresh medium ) on inhibition of competence development , we characterized a mutant deficient in superoxide dismutase , an important enzyme for oxidative stress defense [10] . The mutant was only able to grow at reduced oxygen levels . In BB-FBS , we observed that addition of adenine was crucial for prolonged growth and survival of the mutant , which turned into coccoids in its absence ( Fig 7 ) . Since superoxide dismutase is crucial for superoxide radical inactivation [46] , this suggested that the effect of adenine on competence development in the wild type probably results from its radical quenching capacity [48] . Also , fresh medium might restore quenching capacity , which is depleted during growth . Note that the pH in the presence of adenine or after exchange of the medium by fresh medium at t0 was comparable to that of the control . Moreover , the sodB mutant displayed enhanced competence under similar conditions compared to the wild type ( Fig 7 ) , which supports the hypothesis that oxidative stress triggered competence development . The sodBcompl mutant showed an intermediate competence phenotype between wild type and deletion mutant , indicating that the deletion of sodB at the native locus could only partially be complemented . Likewise , the most pronounced competence development was observed under aerobic conditions . Competence developed quite rapidly; within only 30 min of exposure , nearly 40% of the cells had switched into competent state and after 2 hours this fraction had increased to over 80% , with 20% of the competent cells expressing more than four distinct DNA uptake locations ( Fig 5 ) . Concomitantly , a pH elevation of around 0 . 5 units by loss of dissolved CO2 from the medium upon shift from microaerobic to aerobic atmosphere was observed , with implications discussed below . We also tested other putative quencher molecules , like vitamin C and vitamin E . However , initial experiments showed that at those concentrations , at which competence development was inhibited , growth was drastically reduced , too . Therefore , these experiments were not followed up and data were not included in this manuscript , since substantial growth inhibition is expected to indifferently reduce protein biosynthesis . It might hint at the fact that quenchers can also impair the function of the respiratory chain . Instead , we showed by three independent experiments , ( i ) addition of adenine , ( ii ) variation of oxygen level and ( iii ) competence development in an oxygen-sensitive mutant that oxidative stress might play a crucial role in natural transformation . We showed competence development in BB-FBS under normal microaerobic growth conditions without external stress ( Fig 4 ) . From the overall data , we are tempted to conclude that ( i ) a slight increase in pH by approximately 0 . 2 units ( above pH 6 . 5 ) in combination with ( ii ) decreasing quenching capacity of the medium led to competence development . It was suggested that the presence of antibiotics triggered natural transformation in H . pylori [37] , as shown for Streptococcus and Legionella [27 , 53] . In the study of Dorer and collegues the presence of ciprofloxacin resulted in a four-fold increase in transformation rate . This increase is rather marginal , considering dynamics of log 5 demonstrated in this study or around log 2 in the study of Moore et al . [38] . We tested ciprofloxacin and also mitomycin C under our atmosphere- , pH- , temperature- and growth phase controlled conditions . Fluoroquinoles are thought to generate double strand breaks in DNA [54 , 55] . Mitomycin C crosslinks to DNA and causes DNA damage [56 , 57] . We chose sublethal concentrations that slow down growth . Additionally , we tested higher concentrations and confirmed lack of competence induction as shown previously [37] . However , sublethal intermediate concentrations of either ciprofloxacin or mitomycin C also did not show any impact on competence development under our conditions ( Fig 4 ) . In contrast , our results demonstrated that small variations in atmosphere , external pH and temperature largely influence competence development . This might be an important reason for the relatively high variation in published transformation rates in different laboratories using various media and gas-generation systems . Furthermore , alterations of these parameters during processing of the sample within the experimental setup , including a drop of temperature have to be taken into account in order to prevent biased results . In B . subtilis the term “competence window” was used to define a time period , during which the likelihood for switching into competence state is increased by rising the basal level of the master competence regulator [58 , 59] . Our data suggest that in H . pylori a “competence window” is opened and closed by external pH . In case the pH reached near neutral values ( > pH 6 . 5 ) , competence development was triggered by the combination of pH increase and oxidative stress . The kinetics of competence development was modifiable by oxygen tension ( Fig 6 ) or by the addition of the radical quencher adenine ( Fig 4 ) under comparable pH conditions . Drop in temperature to non-optimal growth conditions also led to increase in competence development , confirming that competence development was not purely regulated by external pH . The temperature effect might be explained by an imbalance of metabolic processes , putatively also leading to sub-optimal ( oxidative ? ) stress defense . From our data we conclude that DNA uptake in H . pylori is dispensable and eventually even deleterious under acidic conditions . In contrast , the process appears to be limited to neutral pH values . From transcriptome data it was suggested that H . pylori colonizes an acidic niche [60] . However , that study focused on early colonization ( 5–10 days after infection ) of H . pylori in gerbils and not only surface epithelial cells with attached bacteria but also gastric mucus was collected as pool samples . Since the majority of H . pylori is thought to reside in the mucus layer , where slight acidic pH values are likely to occur , detection of acid induced genes may be explainable . However , these data do not necessarily reflect pH in direct contact with the host . Measurements using pH microelectrodes showed that H . pylori is confronted with a pH gradient lining from the near neutral gastric mucosa to the extreme acidic pH in the stomach lumen [61 , 62] . Therefore , our data suggest that DNA uptake is most active in direct contact with the host , a site with near neutral pH . In early stages of H . pylori gastritis , the bacterium is preferentially localized at the antrum; with continued inflammation , gastrin producing cells are lost in the antrum , probably leading to decrease in acid secretion by the parietal cells and spread of bacteria and inflammation to the corpus [63] . Likewise , the pH value of the gastric mucosa in patients with ulcers was higher than for a control group [64] . At the sites of chronic severe inflammation , increased oxidative stress due to ROS production by the host might be expected [65] . We observed that , in contrast to B . subtilis , the majority of the H . pylori population acquired the competent state and growth of the bacterium was not interrupted . In B . subtilis natural transformation probably enables the bacterium to survive and adapt to changing unfavorable conditions . In this respect , transiently interrupting growth might serve for increased survival and enhanced persistence . Our data suggest that natural transformation in H . pylori is established under the most favorable conditions within the hostile stomach , namely at the multiplication site close to the near neutral mucosa . Triggering chronic gastritis , ROS-production by the host cell inflammatory system will be a daily reality for the gastric pathogen . In conclusion , we showed for the first time that during microaerobic growth neutral pH values defined a competence window in which DNA uptake capacity was triggered by a combination of pH and oxidative stress . In consequence , this led to the import of staggering amounts of DNA into the majority of cells within the population . Concomitantly , we detected high transformation rates under these conditions . Hence , our data deliver explanations for the observed extraordinary genetic diversity of this microaerobic pathogen , persisting lifelong at a near neutral multiplication site under conditions of chronic inflammation .
H . pylori strains N6 [74] and J99 [75] were grown either on Columbia blood agar base ( Oxoid ) supplemented with 5% defibrinated sheep blood ( Oxoid , Heidelberg , Germany ) or in liquid culture ( shaking at 140 rpm ) using tryptic soy broth ( TSB , Becton Dickinson , USA ) or brucella broth ( BB , Becton Dickinson ) with 5% fetal calf serum ( PAN-Biotech GmbH , Aidenbach , Germany ) , in the following referred to as TSB-FBS or BB-FBS , respectively . Antibiotics were used at the following final concentrations per milliliter: 12 . 5 μg vancomycin , 0 . 31 μg polymyxin B , 6 . 25 μg trimethoprim , and 2 . 5 μg amphotericin B . If applicable , kanamycin or chloramphenicol was added at 20 μg/mL or 8 μg/ml , respectively . Plates were incubated at 37°C in a microaerobic incubator ( Binder , Tuttlingen , Germany ) at 1% or 5% O2 , 10% CO2 and 85% N2 . Atmosphere for liquid cultures was established in jars by evacuating air and refilling the volume with defined gas mixture ( 1% or 5% O2 , 10% CO2 , rest N2 ) . The minimal medium was composed according to [76] but lacked the sole purine source adenine , which was added at concentrations of 5 or 50 μg/ml ( corresponding to 0 . 037 or 0 . 37 mM , respectively ) . Instead of addition of single amino acids , 100 x MEM non-essential amino acids solution ( Life Technologies ) was used as a 20-fold stock solution and 50 x MEM amino acids solution ( Life Technologies ) as a 10-fold stock solution , matching the amino acid concentrations of the original minimal medium except for 66 mg/l asparagine , 210 mg/l histidine and 180 mg/l tyrosine . Glutamine was added separately according to the original protocol . The pH of the medium was adjusted to 7 . 4 . For supplementation of the medium with 10 μg/ml DNA , either DNA of salmon sperm ( Sigma Aldrich ) or purified chromosomal H . pylori N6 DNA was used . Oligonucleotides used in this study are depicted in S1 Appendix , Tab A . Genomic DNA of a streptomycin resistant mutant of H . pylori strain 26695 , bearing an A128G mutation in rpsL was used as template for the amplification of a 1 , 022-bp rpsL ( A128G ) PCR fragment using oligos H3 and H6 . N6 comB4 and N6 comB6 were constructed by transformation of a PCR fragment amplified on genomic DNA from the respective J99 mutants published previously [18] and selected on 20 μg/ml kanamycin . Approximately 500-bp flanking regions of sodB ( HP0389; ORF definition of 26695 ) were amplified using oligos H167/H168 ( 5′ region including start codon of sodB ) and H169/H170 ( 3′ region including 39 bp of sodB ) and genomic DNA of N6 as template [77] . The nonpolar kanamycin resistance cassette was amplified from pUC18ΔK2 [78] using oligos H13 and H14 . Since the oligo H168 contains 22 bp , which are reverse complementary to H13 , and H169 contains 21 bp , which are reverse complementary to H14 , the nonpolar kanamycin cassette was inserted between the flanking regions by fusion PCR . This led to a fusion PCR fragment , suitable for nonpolar inactivation of sodB in N6 by direct transformation and selection on 20 μg/mL kanamycin . For complementation of the sodB mutant , a 1 . 2 kb fragment comprising sodB including the annotated transcriptional start site [79] and approximately 500 bp 5’ region was amplified using H167 and H177 and genomic DNA of N6 as template . The gentamicin cassette was amplified using H11 and H12 from pUC1813apra [80] . Approximately 600-bp flanking regions of the insertion locus fliP were amplified using oligos H7/H178 ( 5’ region ) and H9/H10 ( 3’ region ) . Since the oligo H178 contains a 20 bp reverse complementary region to H167 , the oligo H177 a 21 bp reverse complementary region to H11 and the oligo H9 a 22 bp reverse complementary region to H12 , all four fragments were fused by PCR using oligos H7 and H10 , resulting in an approximately 3 . 2 kb fragment . Correct allelic exchange of all mutants was checked by PCR and sequencing . λ DNA ( dam- and dcm-; Thermo Fisher Scientific , USA ) was covalently fluorescently labeled applying Mirus Label IT Cy3 ( MoBiTec GmbH , Goettingen , Germany ) according to the manufacturer’s protocol using a 1:1 ( volume: weight ) ratio of Label IT reagent to nucleic acid . λ DNA was non-covalently labeled with the fluorescent dye YOYO-1 ( Invitrogen , USA ) at a bp:dye ratio of 1:50 in phosphate-buffered saline a pH 7 . 2 . YOYO-1 specifically stains dsDNA by intercalation at bp:dye ratios of >8 and is virtually non-fluorescent in aqueous solutions [81] . One microgram per ml of labeled DNA was added to a H . pylori cell suspension , which had been centrifuged and resuspended in TSB-FBS ( 100 μl OD600 ∼0 . 2–0 . 8 ) . After incubation for 10 min at 37°C under microaerobic atmosphere , the suspension was centrifuged , resuspended in 25–50 μl TSB-FBS and incubated for 5 min at 37°C in the presence of 10 units of DNaseI ( Roche , Basel , Switzerland ) . For quantification of fluorescence intensity from single λ DNA molecules , a drop of freshly labelled Cy3 λ DNA diluted in TSB was spread out on a 1 . 5% low melting agarose pad by air flow and immediately sealed with a cover slip . In parallel , H . pylori cells were incubated for 10 min in the presence of the same lot of Cy3 λ DNA . Fluorescence microscopy was performed using a Zeiss Axio Observer Z1 microscope with a plan apochromatic 63x/1 . 4 objective and differential interference contrast ( DIC ) . Cy3 was visualised using a metal halide light source ( HXP120C ) and a filter set with excitation at 550 nm ( bandwidth 25 nm ) , emission at 605 nm ( bandwidth 70 nm ) and a dichroic beamsplitter at 570 nm . YOYO-1 that intercalated into dsDNA was detected with a filter set providing excitation at 470 ± 20 nm and emission at 525 ± 25 nm ( dichroic beamsplitter at 495 nm ) . Images were acquired with the 12-bit monochromatic AxioCam MRm camera , using exposure times between 60 ms and 2 . 5 s . Intensity values of Cy3 were checked for linearity between different exposure times and bleaching was found negligible ( S1 Appendix , Fig A ) . Regions of interest ( ROIs ) were set manually and analyzed by ZEN software ( blue edition ) version 2 . 0 . 0 . 0 . In order to determine the fluorescence of one single λ DNA molecule , fluorescence intensity was determined from spread out DNA molecules with a minimal length of ½ of maximal size of λ DNA ( 8 . 6 μm , [43] ) and defined as one molecule . Fluorescence intensity in single DNA foci and in whole cells were determined , compared to fluorescence intensity of λ DNA molecules and expressed as imported kb . The cut-off value for the analysis of DNA foci was set to a signal to background ratio of ≥ 2 . For the determination of transformation rate , bacterial cells grown in liquid culture were centrifuged and resuspended in 100 μl TSB-FBS to OD600 ∼1 . Subsequently , 1 μg/ml of a 1 , 022-bp rpsL ( A128G ) PCR fragment was added and incubated for 1 h at 37°C under microaerobic atmosphere . Cells were centrifuged , resuspended in 50 μl TSB-FBS in the presence of 10 U of DNaseI ( Roche , Basel , Switzerland ) and directly spotted on Columbia blood agar . After incubation under microaerobic conditions at 37°C for 20 ± 4 h , bacteria were harvested in TSB-FBS . Total colony forming units as well as the number of streptomycin resistant transformants were determined according to the serial dilution method using Columbia blood agar with and without 20 μg/mL streptomycin and a selection time of 4–5 days . As alternative resistance marker , the chloramphenicol cassette of pILL2157 [40] was used within the context of this replicative plasmid or amplified using oligos H108 and H109 and fused to approximately 500 bp 5’ region ( oligos H110 and H111 ) and 3’ region ( oligos H112 and H113 ) of nucT in order to create a 1886 bp PCR fusion fragment ( H110/H113 ) for homologous recombination into the nucT locus . Cells were inoculated at OD600 ~0 . 003 and grown over night under microaerobic conditions in BB-FBS to time point t0 , at which the culture had reached approximately OD600 between 0 . 15 and 0 . 25 . At t0 either one of the different conditions was established; ( i ) one of the effectors ( see below ) was added or ( ii ) , the culture was temperature downshifted by 5°C or ( iii ) , the culture was exposed to aerobic conditions at 37°C or ( iv ) , the medium was exchanged for fresh medium with or without supplementation of either one of the DNA-damaging agents 0 . 125 μg/ml ciprofloxacin or 0 . 025 μg/ml mitomycin C . It was guaranteed that the procedure for changing the condition was completed within 10 min . For exchanging the medium , cells were centrifuged at 9000 x g for 2 min at room temperature and immediately resuspended in the same volume of prewarmed ( 37°C ) fresh medium with or without DNA-damaging agents . At t1 ( 3–4 h after t0 ) and t2 ( 6–8 h after t0 ) a test aliquot was taken from the cultures and checked for the number of cells harboring DNA foci as mentioned above . For aerobic exposure , samples were taken after 15 , 30 , 60 and 120 min , respectively . If indicated , pH was measured using a temperature-controlled pH electrode with an accuracy of 0 . 01 pH units ( pH/ATC , Denver Instrument , New York , USA ) . In order not to disturb competence development , parallel culture aliquots served for pH measurements .
|
Natural transformation , i . e . the capacity to take up DNA from the environment , is one of the crucial means for horizontal gene transfer and genetic diversity in bacteria . The human gastric pathogen Helicobacter pylori is confronted with acid stress before entering its multiplication site , the gastric mucosa . The bacterium causes lifelong chronic gastritis and is perfectly adapted to the human host , crucially by displaying unusual genetic diversity . Using a single cell approach and well-controlled conditions , we show here that the amount of imported DNA in competent H . pylori is outstanding , far exceeding previous measurement with other Gram-negative bacteria . Furthermore , DNA uptake activity was tightly regulated and limited to pH above 6 . 5 , conditions thought to be met in close contact with the gastric mucosa . In addition , we show that within this pH competence window , competence development was triggered by an increase in pH in combination with the level of oxidative stress . Our data provide explanations for the extraordinary high genetic diversity , often referred to as genome plasticity of this unusual microaerobic pathogen .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"bacteriology",
"medicine",
"and",
"health",
"sciences",
"pathology",
"and",
"laboratory",
"medicine",
"chemical",
"compounds",
"oxidative",
"stress",
"oxygen",
"pathogens",
"microbiology",
"nucleotides",
"organic",
"compounds",
"purines",
"helicobacter",
"bacterial",
"genetics",
"bioassays",
"and",
"physiological",
"analysis",
"dna",
"microbial",
"genetics",
"bacteria",
"bacterial",
"pathogens",
"research",
"and",
"analysis",
"methods",
"medical",
"microbiology",
"microbial",
"pathogens",
"chemistry",
"adenine",
"biochemistry",
"chemical",
"elements",
"helicobacter",
"pylori",
"cell",
"biology",
"nucleic",
"acids",
"organic",
"chemistry",
"genetics",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"organisms",
"fluorescence",
"competition"
] |
2016
|
Genetic Diversity as Consequence of a Microaerobic and Neutrophilic Lifestyle
|
4C-Seq has proven to be a powerful technique to identify genome-wide interactions with a single locus of interest ( or “bait” ) that can be important for gene regulation . However , analysis of 4C-Seq data is complicated by the many biases inherent to the technique . An important consideration when dealing with 4C-Seq data is the differences in resolution of signal across the genome that result from differences in 3D distance separation from the bait . This leads to the highest signal in the region immediately surrounding the bait and increasingly lower signals in far-cis and trans . Another important aspect of 4C-Seq experiments is the resolution , which is greatly influenced by the choice of restriction enzyme and the frequency at which it can cut the genome . Thus , it is important that a 4C-Seq analysis method is flexible enough to analyze data generated using different enzymes and to identify interactions across the entire genome . Current methods for 4C-Seq analysis only identify interactions in regions near the bait or in regions located in far-cis and trans , but no method comprehensively analyzes 4C signals of different length scales . In addition , some methods also fail in experiments where chromatin fragments are generated using frequent cutter restriction enzymes . Here , we describe 4C-ker , a Hidden-Markov Model based pipeline that identifies regions throughout the genome that interact with the 4C bait locus . In addition , we incorporate methods for the identification of differential interactions in multiple 4C-seq datasets collected from different genotypes or experimental conditions . Adaptive window sizes are used to correct for differences in signal coverage in near-bait regions , far-cis and trans chromosomes . Using several datasets , we demonstrate that 4C-ker outperforms all existing 4C-Seq pipelines in its ability to reproducibly identify interaction domains at all genomic ranges with different resolution enzymes .
Understanding the 3D organization of the genome and the intricacies of chromatin dynamics has been the focus of studies aimed at characterizing gene regulation in physiological processes and disease states [1 , 2] . Microscopy based studies provided the first snapshots of nuclear organization , revealing that individual chromosomes occupy distinct territories with little intermingling between them [3 , 4] . The development of chromosome conformation capture ( 3C ) transformed the field of nuclear organization enabling identification of chromatin interactions at the molecular level and at the same time opening the door to high-throughput , genome-wide techniques [5] . Hi-C , for example , captures all pairwise interactions in the nucleus and has revealed that chromosomes segregate into two distinct spatial compartments ( A and B ) depending on their transcriptional and epigenetic status [6] . These compartments are further subdivided into Topological Associated Domains ( TADs ) , which are highly self-interacting megabase scale structures [7–9] . To probe interactions between regulatory elements using Hi-C requires a depth of sequencing that for many labs is cost-prohibitive [10] . 5C can circumvent these issues , but the interaction analysis is limited to the portion of the genome for which primers are designed [11] . Circular chromosome conformation capture combined with massive parallel sequencing ( 4C-Seq ) is currently the best option for obtaining the highest resolution interaction signal for a particular region of interest . In 4C-Seq , an inverse PCR step allows for the identification of all possible genome wide interactions from a single viewpoint ( the “bait” ) and an assessment of the frequencies at which these occur . The sequencing coverage obtained by 4C near the bait region is extremely high and therefore enables precise characterization and quantification of regulatory interactions [12 , 13] . By focusing on one locus at a time and thus only the interactions that this locus is engaged in , 4C can reproducibly identify long-range interactions on cis and trans chromosomes [14] . For example , 4C was used to demonstrate that genes controlled by common transcription factors tend to occupy the same nuclear space even when located on different chromosomes [15 , 16] . There are many inherent biases specific to the 4C technique that have made detecting meaningful and reproducible interactions challenging . First , in accordance with the chromosome territory model , the majority of 4C signal is located on the bait chromosome . Secondly , coverage and signal strength are highest in the region around the bait and this decreases along the chromosome as a function of linear distance from the bait . Third , the restriction enzyme used for the first digest in the experiment is an important determinant of the resolution at which interactions can be detected . Finally , as with most PCR-based techniques , 4C data includes PCR artifacts that manifest as a large accumulation of reads in particular locations . Current methods of analysis have addressed some of these issues , however there are still many hurdles to overcome . Specifically , existing methods do not properly account for the differences in 4C signal coverage across the genome and therefore they are only able to either identify interactions in ( i ) regions where the signal is highest , i . e . , near the bait or ( ii ) regions of low 4C signal ( far-cis and trans ) . Thus there is no method that comprehensively identifies interactions across the genome . In addition , most methods were developed and tested using datasets generated with 6bp cutters and we show that they do not perform well with 4bp cutter generated libraries . The goal of 4C-ker is to address these weaknesses by: 1 ) identifying domains that interact most frequently with the bait across the genome in a given population of cells , and 2 ) detecting quantitative differences in 4C-Seq signal between conditions . Here we use a Hidden Markov Model to account for the polymer nature of chromatin , in which adjacent regions share a similar probability of interacting with the bait . In addition , to account for the variation in signal captured at different 3D distances we use a window-based approach . To determine the window size of analysis , we adapted a k-th nearest neighbor approach to account for the decrease in 4C-Seq coverage along cis and trans chromosomes . We used 4C-ker to analyze several publically available 4C-Seq datasets as well as data generated in our own lab and compared this with other published methods . Our results demonstrate that 4C-ker can correct for multiple 4C-Seq biases and reproducibly detect genome wide interactions from the bait viewpoint . Importantly , 4C-ker is the only tool that can identify interactions with regions in near and far-cis as well as trans .
We developed 4C-ker to identify genome-wide interactions generated by 4C-Seq data and to quantitatively examine differences in interaction frequencies between conditions . The main components of the 4C-ker method are outlined in Fig 1 . First , 4C-Seq reads are mapped to a reduced genome consisting of unique sequences adjacent to all primary restriction enzyme sites in the genome . Mapping to a reduced genome helps to remove spurious ligation events that do not result from crosslinking . The analysis of 4C-Seq is typically performed separately for cis and trans chromosomes because of the large differences in signal in these distinct locations . Additionally , we present the option for focusing the analysis on the region surrounding the bait , where 4C-Seq signal and resolution are highest . A window-based approach is applied in order to take into account of differences in signal strength at different 3D distances and the dynamic nature and variability of chromosome interactions in a population of cells . One of the most challenging aspects of 4C-Seq is determining the window size at which the data should be analyzed . Adaptive window sizes that depend on the distance to the bait can adjust for differences in coverage of 4C-Seq signal in regions near the bait , far-cis and trans . 4C signal is generally higher around the bait region and decreases in far-cis and trans . We developed a kth nearest neighbor method to build overlapping windows of adaptive sizes based on the 4C-Seq signal coverage of a given dataset at each location in the genome . The aim of using this approach is to obtain a similar number of observed fragments for each window . Therefore the size of each window is determined by the amount of signal detected in the region . This will result in small windows near the bait and other regions where there is high coverage , versus larger windows further away from the bait where there is low coverage . An example of this is shown in S1A Fig . To correct for PCR amplification bias , the counts at observed fragments within each window that exceed the 75th quantile for that window are limited to this value . The window counts across all samples are subsequently normalized using DESeq2 [17] to correct for sequencing depth . For the cis chromosome , the linear distance from the bait to the mid-point of each window is used to correct for the inverse relationship between counts and linear separation from the bait . The counts and distances are log transformed and are used as inputs ( observed states ) for the Hidden Markov Model ( HMM ) . A separate model is used for cis and trans chromosomes ( in the latter there is no effect of linear distance from the bait ) . A three-state HMM is used to partition the genome into windows that interact with the bait at ( 1 ) high frequency , ( 2 ) low frequency and ( 3 ) those that do not interact . Use of overlapping windows , allows us to more precisely define the regions of high interaction ( for more detailed explanation of the workflow , refer to the methods section ) . The resulting parameters for the model show higher probabilities for transitioning to the same state , correctly accounting for the polymeric nature of DNA on chromosomal interactions ( Fig 1 , 3-state HMM ) . Since we use overlapping windows , consecutive windows that are detected as highly interacting can be stitched to form a ‘domain’ of high interaction with the bait . Domains that are found as interacting at a high frequency in at least one sample can be used for downstream quantitative analysis . Furthermore DESeq2 designed for differential analysis of counts-based sequencing data can be used to quantitatively compare interactions across conditions . The 4C-ker pipeline is available as an R package along with the domains of interactions identified for all the datasets analyzed in this study ( github . com/rr1859/R . 4Cker ) . 4C-Seq is commonly used to identify regulatory interactions that occur in close linear proximity to the bait . Therefore , we provide an option to focus the analysis only in this region , where the highest resolution interactions are identified . An important aspect of 4C-Seq library preparation is the choice of restriction enzyme used to digest cross-linked chromatin as the genome-wide frequency of enzyme recognition sites determines the resolution of the experiment . Therefore , 4bp cutters such as DpnII or NlaIII , which cut the genome more frequently , provide a higher resolution profile of 4C interactions compared to 6bp cutters like HindIII ( S1B Fig ) [18] . To ensure that our method works with both types of restriction enzymes , we tested it using several datasets generated from our lab as well as all publically available datasets for which replicates are available that passed our stringent quality controls . See methods section and S1 Table for details ) . The near-bait analysis was restricted to 10MB around the bait for 6bp cutters and 2MB for 4bp cutters as these are the regions that contain the highest 4C signal in each case . As can be seen in Fig 2A and 2B , raw 4C-Seq signal is highest near the bait and decreases with increasing linear distance . As 4C-ker corrects for this decrease in signal it is able to detect interactions across the entire region analyzed . In addition , due to the adaptive windows , the size of interactions detected are smallest near the bait where coverage is highest and larger in regions further away from the bait where coverage is lower ( Fig 2C and 2D ) . The resolution of domains identified by 4C-ker can be conveniently adjusted by using different values for the number of observed fragments used to generate the adaptive windows ( S2A Fig ) . Here we used values ranging from 3–10 in the 2MB region around the bait . As the value of ‘k’ increases , we observe a consistent increase in the size of the domains as well as increased similarity between replicates ( see methods section for details ) . The parameter k can be adjusted by the user depending on the biological question that is being addressed . For example , if the aim of the study is to identify interactions between enhancers and promoters , we suggest k = 3–5 . In order to identify larger domains that coincide with broad regions encompassing chromatin with similar histone modifications , setting k = 10 is a suitable choice . To assess the performance of 4C-ker we used existing methods to analyze the same datasets . There are currently four publicly available methods to detect significant interactions using 4C-Seq datasets ( fourSig , Splinter et al , r3CSeq and FourCSeq ) . Details of how we implemented these algorithms for comparison with the 4C-ker pipeline can be found in the methods section . Although the method developed by van der Werken et al ( 4cseqpipe ) [13] does not identify significant interactions , it provides a good visualization tool for 4C-Seq signal near the bait ( Fig 2A and 2B ) . The fourSig approach generates windows based on restriction enzyme fragments and compares the counts within each window against a random background distribution [19] . As fourSig does not take account of the impact of distance on 4C-Seq signal , it identifies most of this region as large interacting domains and this results in a high similarity index between replicates ( S2B and S2C Fig ) . However , in contradiction to decreasing resolution of 4C-Seq signal with increased separation from the bait , the size of the domains identified by fourSig are largest near the bait and these decrease with increasing separation from the bait ( Fig 2C and 2D ) . The method described by Splinter et al [20] , referred to here as the ‘de Laat method’ excludes the 2MB region around the bait and only calls interactions in the rest of the genome based on enrichment of binary coverage in a given window , compared to a local background . As such , the de Laat method does not identify any interactions with 4bp cutters ( Fig 2B and 2D ) . Moreover , using the 6bp cutter datasets it only identifies interactions in 2 out of 7 datasets in the 10MB region ( S2B Fig ) . Together these findings reflect the limitations of this method in detecting 4C-Seq interactions in the region with highest coverage , where the majority of important regulatory interactions occur . The r3CSeq method uses reverse cumulative fitted values of the power law normalization and a background scaling method to correct for interactions near the bait [21] . This approach also provides the option to detect interactions at the fragment level or at the window level . In most datasets r3CSeq only identifies significant interaction near the bait as shown in Fig 2A and 2C and 2D and therefore have a high similarity index between replicates ( S2B and S2C Fig ) . Although interactions further from the bait are identified ( Fig 2B ) , they are not reproducible as measured by the similarity index ( S2C Fig ) . The FourCSeq pipeline only has the option to analyze interactions at the fragment level . It is based on the DESeq2 method with an additional function that corrects for the effect of linear distance from the bait [22] . This method failed to identify any significant interactions for any of the datasets analyzed . If interactions between regulatory elements are being analyzed , the majority will be identified in the region near the bait . Therefore , it is important that a 4C-Seq analysis can properly identify these interactions . Here we show that 4C-ker outperforms other methods and identifies interactions that correctly reflects the nature of high-resolution 4C-Seq signal in this region . We next used 4C-ker for analysis of the entire bait chromosome using the same fourteen datasets described above . Due to lower 4C-Seq signal in regions distant from the bait ( far-cis ) the correlation between replicates decreases compared to near-bait regions ( S3A Fig ) . This difference is more pronounced with 4bp cutter generated datasets . A potential explanation for this difference is that when 4bp cutters are used 4C-Seq coverage in windows distant from the bait decreases at a much faster rate than when using 6bp cutters ( S3B Fig ) . Based on these results , it is clear that when designing a 4C-Seq experiment , the biological question should determine the choice of primary restriction enzyme . For example , to detect long-range interactions in cis and trans it seems preferential to use a 6bp cutter to achieve a more reproducible 4C profile . On the other hand , for characterization of short-range regulatory interactions , 4bp cutters provide a high-resolution map of near-bait interactions , as previously shown [18 , 23 , 24] . With adaptive window sizes and consideration of distance separation from the bait , 4C-ker is able to reproducibly identify domains of interaction across the whole cis chromosome . As expected , interacting domains proximal to the bait are smaller in line with the fact that increased 4C-Seq signal allows for generation of smaller windows of analysis . In contrast , in regions located distal to the bait where the 4C-Seq signal is reduced , the window sizes for analysis are increased and 4C-ker identifies larger interacting domains ( Fig 3A ) . In comparison to other methods , 4C-ker identifies the most reproducible interactions for all 4 bp cutter datasets and for 5 out of 7 6bp cutter datasets ( Fig 3B ) . For 6bp cutter datasets , the domains identified by 4C-ker are comparable in size to the de Laat method ( Fig 3C ) . For the 4bp cutter datasets the other methods identify smaller domains compared to 4C-ker , however they do not achieve high similarity indices between replicates . Although the de Laat method and fourSig perform comparably to 4C-ker when analyzing 6bp-generated data they fail to do so with 4bp datasets . To validate interactions identified by 4C-ker we used the Igh Cγ1 HindIII dataset and performed 3D-FISH to analyze interactions with Igh . We selected three bacterial artificial chromosome ( BAC ) probes that hybridize to high , low and non-interacting regions in close proximity to each other ( 4–7Mb ) , but separated from Igh by ~70Mb ( S4A Fig ) . Of note , the selected non interacting region is in closer linear distance to Igh , and the highest interacting region is furthest away . Using differentially labeled BAC probes for these regions in conjunction with an Igh specific probe we found that in accordance with the 4C-ker output , the BAC in the high interacting domain is in closer spatial proximity to Igh than the BACs in the low and non-interacting domains ( S4A Fig and Fig 3C ) . It is of note that all other methods failed to identify the region that we validated by FISH as a significant interaction . According to the chromosome territory model most interactions occur between loci on the same chromosome . As such , inter-chromosomal interactions occur at low frequency . However , unlike other 3C-based techniques , 4C-Seq can still detect these interactions . Nonetheless , since at least 40% of the signal is on the cis chromosome , the rest is spread over all trans chromosomes and is thus significantly reduced . As a result the 4C signal is less reproducible compared to interactions on the bait chromosome ( S3A Fig ) . 4C-ker and the de Laat method outperform fourSig and r3CSeq in identifying trans interactions . Both 4C-ker and the de Laat method identify equivalently sized interaction domains across all fourteen ( 6bp and 4bp cutter ) datasets ( Fig 3D and 3E ) . In most cases 4C-ker outperforms the de Laat method in identifying reproducible interactions from 4bp cutter experiments , while the reverse is true for most 6bp cutter experiments . One useful application of 4C-Seq is a quantitative comparison of interactions from a particular viewpoint across conditions or cell types . The highly interacting domains identified by 4C-ker for several conditions can be merged to generate a list of “Dataset-specific Interacting Domains” ( DIDs ) . These domains represent regions that are interacting with the bait in at least one of the conditions . In general , 4C-Seq counts follow a negative binomial distribution , which is suitable for differential DESeq2 analysis . We use raw counts for the adaptive windows that fall within DIDs and normalize these using DESeq2 to correct for sequencing depth . To identify differential interactions between conditions we use an FDR adjusted p-value of < 0 . 05 . To test this approach , we compared two datasets generated with NlaIII digested 4C template that have a bait on the Eβ enhancer of Tcrb in double negative ( DN ) and immature B ( ImmB ) cells . DIDs were generated for the bait chromosome ( chromosome 6 ) for these two cell types . In Fig 4A , the normalized values for windows within DIDs are plotted across the entire bait chromosome . Windows that are significantly different in the two cell-types are represented as larger filled circles . It is clear from Fig 4A that the majority of differentially interacting regions are concentrated near the bait . This can be seen in detail for the interaction of the Eβ enhancer with the 5’ end of the Tcrb gene ( Fig 4B ) . In DN cells this locus is in a contracted conformation which brings distal Vβ genes into contact with the proximal DJCβ region for V ( D ) J recombination [25] . In contrast , the locus does not recombine in B cells and is not in a contracted form and the Vβ genes are found in less frequent contact with the bait . Interestingly , we found a differentially interacting DID in far-cis containing the Cd69 gene ( Fig 4C ) , which is a known T cell marker and interacts more frequently with Eβ in DN cells compared to Immature B cells . This is expected since both Cd69 and Tcrb are active in T cells and it has been shown that transcriptionally active regions come into frequent contact [15 , 16] . Thus , the DIDs determined by 4C-ker can be used to detect quantitative interactions that correlate with functional processes . The ability to detect reproducible long-range interactions with 4C-ker enables us to assess the properties of these regions . Based on nuclear organization principles described by 3C-based studies [6 , 15 , 16] we validated 4C-ker domains by assessing if they preferentially contact regions with the same transcriptional and epigenetic status as the bait . For this , we used 4C data generated with the Eβ enhancer bait in DN T cells and immature B cells . Using ATAC-Seq [38] , a technique that identifies accessible regions of chromatin , we find that as expected , the enhancer is active in T cells and inactive in B cells ( Fig 5A ) . Conversely , a bait on the MiEκ enhancer of Igk is active in B cells and inactive in T cells ( Fig 5A ) . Using 4C-ker we identified the highly interacting domains with each bait across the two cell types . Since we used NlaIII to generate the template we restricted the analysis to the bait chromosome . We then asked if the 4C interacting domains are enriched for ATAC-Seq peaks . Here , we define enrichment as the ratio of the sum of the size of ATAC-Seq peaks within interacting regions to those within a background generated by randomly repositioning these domains along the chromosome . In T cells , where the Eβ enhancer is active , we found a higher enrichment of ATAC-Seq peaks in 4C interacting domains compared to B cells ( Fig 5B ) . The opposite is observed with the MiEκ 4C bait in B cells , where the enhancer is active and enrichment of ATAC-Seq peaks in 4C interacting domains is higher compared to T cells ( Fig 5B ) . Thus , in line with previous studies using both HiC and 4C-Seq [15 , 16] , active regions of the genome preferentially contact other active regions while inactive regions contact other inactive regions , and this pattern is consistent across lineages . To determine the relationship between transcriptional status and accessibility , we next integrated RNA-Seq data with the output from 4C-ker . We first confirmed the transcriptional activity of both enhancers across lineages , as demonstrated by the transcriptional activity of the Tcrb and Igk loci that are controlled by their respective enhancers ( Fig 5C ) . The active Eβ enhancer selectively directs transcription of Tcrb in T cells , while MiEκ contributes to the high levels of Igk transcription that is found only in the B cell lineage . Next we compared the expression values of genes within the interacting domains across the different cell types . The genes within the Eβ-interacting domains in T cells show a higher transcriptional activity compared to genes within Eβ-interacting domains in B cells ( Fig 5D ) . The same is observed in genes within MiEκ-interacting domains in B versus T cells ( Fig 5D ) . Again , these results are in agreement with Hi-C studies , which show that regions with similar transcriptional activity occupy the same space in the nucleus [6 , 15 , 16] .
Here we describe 4C-ker , a 4C-Seq analysis framework , that is unique in its ability to reproducibly detect short and long range-interactions on the same and across different chromosomes from a single viewpoint . Unlike other 4C-Seq pipelines , 4C-ker takes into account difference in coverage in regions proximal to the bait , far-cis and trans . As summarized in Table 1 , 4C-ker outperforms all other methods in regions near the bait and in far-cis and performs comparably to the de Laat method for analysis of trans interactions . Moreover , all other tested methods fail to detect a region we defined as highly interacting using 4C-ker that we subsequently validated by FISH . Finally , 4C-ker also has the option to perform differential analysis of cis interactions . 4C-Seq can be used as an unbiased approach to identify short-range regulatory interactions that occur with the bait as well as long-range interactions that can provide insights into the global organization of chromatin in the nucleus . With 4C-ker , we validated long-range interactions from enhancer viewpoints and analyzed the epigenetic and transcriptional properties of interacting domains ( where the validation includes reproducibility of results and experimental validation with FISH ) . This enabled us to demonstrate that the domains that 4C-ker calls have biological significance: active regions preferentially associate with active regions and inactive regions preferentially associate with inactive regions , as previously shown in Hi-C [6 , 15 , 16] . While Hi-C is limited in its ability to detect short-range interactions at low resolution , 4C-ker can identify both short and long-range interactions with higher resolution at lower sequencing depth . One important consideration in 4C-Seq is to unravel how the profile of interactions generated in a population of cells relates to the physical constraints of chromosomes within the nucleus . For example , we need to better understand the implications of the differences in 4C-Seq profiles when an active or an inactive bait is used . Reduced interactions from an inactive bait likely reflect a less mobile compacted chromatin structure that could be embedded within the chromosome territory . To explore these relationships we need improved pipelines for integrating other genome wide techniques such as RNA-Seq , ATAC-Seq , and ChIP-Seq with 3C-based data sets . Only then can we learn whether inactive regions of the genome interact with regions that share epigenetic modifications and are bound by common regulatory factors as has been shown for active regions that are co-regulated [26 , 27] . Although 4C-Seq only provides information on interactions from a single viewpoint , it can help to identify intricate loop structures at a finer resolution than Hi-C , and this in turn will provide a basis for understanding regulatory interactions . Furthermore , it can identify long-range interactions in cis and in trans that likely reflect inter-TAD interactions on the same or different chromosomes . These interactions need to be validated by FISH analysis , which in contrast to chromosome conformation capture , faithfully reflects the appropriate chromatin compaction state and recapitulates the findings from individual live cells ( as opposed to averaging over populations ) [28] . Furthermore , FISH analysis can provide information about whether a particular region is embedded within a chromosome territory or looped away , which can be reflective of gene activity or association with repressive pericentromeric heterochromatin [29] . The 4C-ker pipeline can be adapted for analysis of data from new 3C-based techniques such as Capture-C [30] , T2C [31] and CHi-C [32] , that use oligonucleotides to enrich interacting fragments from multiple baits in a single experiment . Furthermore , the high resolution of 4C-Seq data can be used for determining the finer structure of domains identified with Hi-C . Finally , it should be pointed out that there is a great deal of variability between 4C-Seq experiments generated by different labs , and it is clear that the field would benefit from standardized protocols and quality control of datasets that lend themselves to comparisons between experiments from different sources . Going forward 4C-ker will provide a much-needed tool for comprehensive analysis of 4C datasets derived from different experimental approaches .
Animal care was approved by Institutional Animal Care and Use Committee . Protocols number is 150606–01 ( NYU School of Medicine ) . The sequence reads generated from a 4C-Seq experiment typically contain the primer sequence ending in the primary restriction enzyme followed by the interacting fragment captured by the bait . The portion of the read following the restriction enzyme sequence is mapped to a reduced genome—a set of unique sequences ( with the same length as the interacting fragment sequenced ) that are directly adjacent to all sites in the genome of the primary restriction enzyme used . We define these unique sequences as ‘potential fragments . ’ We used oligoMatch ( from UCSC command line tools ) to find all the primary restriction enzyme recognition sequences in the genome and a custom shell script ( provided ) was used to create the reduced genome . Reads were mapped to the reduced genome using Bowtie2 [32] ( command-line options: -N = 0 , in addition -5 was used to trim the barcode and primer sequence ) . We define the fragments in the reduced genome that have at least 1 read mapped to it as an ‘observed fragment’ . The read count at each observed fragment is extracted from the Bowtie output ( SAM file ) and transformed to a bedGraph file ( 4 columns with chr , start , end , count at each observed fragment ) that can be uploaded to IGV for visualization . A custom shell script is provided to generate these bedGraph files . For paired-end sequencing experiments the read containing the bait and the primary restriction enzyme was mapped as single-end data . Adaptive window sizes were determined using the k-th nearest neighbor approach to account for the change in 4C-Seq coverage in different regions . The value of k determines the number of observed fragments to be analyzed within each window . The window size is determined for each observed fragment as the linear distance to the k-th nearest observed fragment , which will result in a larger window size in regions where few fragments are observed and vice versa . On the bait chromosome , the adaptive windows are determined left and the right of the bait starting from the bait coordinate . For example , the first window on the left of the bait will be based on the k observed fragments on the left of the bait and the first window on the right will be based on the k observed fragments on the right of the bait . In this manner , the windows are built until the end of the chromosome . For trans interactions , this process is performed independently for each chromosome starting with the first fragment identified at the beginning of the chromosome . Window sizes are determined for each sample in a given dataset . Then a smooth spline ( smooth . spline function in R with a smoothing parameter of 0 . 75 ) is fitted to the window sizes separately for each chromosome in order to get a window size at each position along the chromosome that can be used for the entire dataset ( S1A Fig ) . To build the final windows we use overlapping windows to more accurately identify the borders of interacting domains . Cis: Using the bait coordinate , the size of the first window is predicted from the fitted spline and this is used to build a window to the left and right of the bait . Adjacent windows start at the mid-point of the bait window and the size is again determined by the fitted spline . In this manner , overlapping windows are generated for the region near the bait or the entire chromosome and will be used to analyze interactions for all samples in the given experiment . For the analysis near the bait we used k = 5 and , when analyzing the entire bait chromosome , k = 10 . Trans: Starting from the beginning of each trans chromosome , we predict the window size from the fitted spline . The next window starts from the mid-point of the first window and this process continues to the end of the chromosome . We used k = 15 for all trans analysis using 6bp cutters and k = 100 for 4bp cutters . To reduce the effect of PCR artifacts , fragments with counts greater than the 75th quantile within a given window are trimmed to this value . In this manner , we reduce the value of single fragments that have an extremely high count that are not supported by neighboring fragments with a similar signal . The counts at observed fragments within each window are normalized across all samples in the dataset using the method described in the DESeq2 [17] R package where each window is considered as a feature ( or gene ) . For windows in cis , the distance from the bait to the mid-point of each window ( in bp ) is also calculated . A pseudo-count of 1 is added to the normalized window counts and the distance value followed by a log10 transformation of the values . The log-transformation of the data results in an approximately linear function that describes the decrease in counts as the distance from the bait increases . In order for 4C-ker to take into account conditional dependencies among neighboring genomic elements we propose to use a three-state Hidden Markov Model ( HMM ) where the hidden states represent genomic regions that show high frequency of interactions in the population ( high interaction region-HI ) , low frequency ( low interaction region-LI ) , and no significant frequency of interactions ( no interaction-NI ) with the bait . A separate model was learned for cis and trans chromosomes . We used the depmixS4 R package [33] to specify and train the described HMM . For multiple conditions with the same bait , a merged set of domains is generated that contains those called as highly interacting in at least one of the conditions—Dataset-specific interacting domains ( DIDs ) . We can then obtain the raw window counts with each domain and use DESeq2 to perform a quantitative differential analysis . DESeq2 has been developed primarily to analyze RNA-Seq data but can also be applied to any count dataset that follows a negative binomial distribution . Therefore , we decided to use the method to look for quantitative differences between conditions in 4C-Seq . Similarity index was calculated based on a previously described method for dealing with more than 2 replicates [34] . CSm=mm−1 ( ∑i<jaij−∑i<j<ka_ijk+∑i<j<k<laijkl… . ∑iai ) , where m is the number of replicates in the dataset and aij is the sum of the size of overlapping HI domains between replicate i and j and ∑iai is the sum of the size of the merges HI domains from all replicates . Domains from both replicates were retained when 50% of the domains overlapped with the other replicates . When comparing with different number of replicates , we divide by m to get a score between 0–1 . The interactions defined for each replicate by the four methods were used to calculate the similarity index . Details of publically available datasets downloaded for this study can be found in S1 Table . We used datasets that had more than one replicate available in GEO and processed the FASTQ files using our pipeline . Datasets were further excluded if less than one million reads were available after removal of undigested and self-ligated 4C fragments . Samples were also required to have at least 40% of the reads on the cis chromosome and 40% coverage in the 2Mb region around the bait for 6bp cutters and 200kb for 4bp cutters as this is considered a standard quality control for a good 4C experiment [35] . Basic statistics for the datasets used can be found in S1 Table . The following datasets were generated from mouse cells for this study . Cd83 , Igh-Cγ1 baits on activated mature B cells , Igκ MiEκ , Tcrb Eβ bait in double negative ( DN ) T cells and immature B cells . See S1 Table for details of primers and enzymes used for these experiments . All datasets generated for this study can be found GEO ( GSE77645 ) . The 4C-Seq protocol was performed as described previously [14] and libraries were sequenced using the HiSeq2500 Illumina platform . Splenic mature B cells were isolated and induced to undergo class switch recombination as previously described [14] . Cells were collected on day 2 of activation . DN T cells , and immature B cells were isolated as described before [29 , 36] and pooled to obtain 10 million cells for each replicate at each developmental stage . Activated mature B cells for FISH analysis were isolated as described above . 3D-FISH was performed as described previously [37] . Interphase cells were analyzed by confocal microscopy on a Leica SP5 AOBS system ( Acousto-Optical Beam Splitter ) . Optical sections separated by 0 . 3μm were collected using Leica software and only cells with signals from both alleles ( >95% of cells ) were analyzed . Separation of alleles was measured in 3D from the center of mass of each signal using Image J software . DN T cells as well as immature B cell were isolated as described above . ATAC-Seq was performed in duplicate as described previously [38] with the following modifications: libraries were amplified with KAPA HiFi polymerase . Libraries were sequenced with HiSeq using 50 cycles paired-end mode . 50bp-paired-end reads were mapped to mm9 using Bowtie2 with the following parameters:—maxins 2000 , —very-sensitive Reads with MAPQ score < 30 were filtered out with Samtools , and duplicate reads were discarded using Picard tools . For each sample condition , biological replicates were merged with Samtools , and peaks were called using Peakdeck [39] with the following parameters: -bin 75 , -STEP 25 , -back 10000 , -npBack 100000 . Peaks were further filtered to a raw p-value cutoff of 1E-4 ( liftOver was used to convert the results to the mm10 genome ) . A custom script was used to determine peak maxima , and maxima were extended by 50bp on either side to yield peaks of ~100bp . DN and immature B cell were isolated as described above . RNA-seq libraries were prepared as previously described using the Ribo-Zero kit for depletion of ribosomal RNA [40] . Reads were mapped using Tophat version 2 . 0 . 6 [41]with the following parameters:—no-coverage-search -p 12—no-discordant—no-mixed -N 1 —b2-very-sensitive . Number of reads per gene ( RefSeq annotation ) was calculated using HTSeq-count [42] . Normalization of counts per gene was done using DESeq2 .
|
Circularized chromosome conformation capture , or 4C-Seq is a technique developed to identify regions of the genome that are in close spatial proximity to a single locus of interest ( ‘bait’ ) . This technique is used to detect regulatory interactions between promoters and enhancers and to characterize the nuclear environment of different regions within and across different cell types . So far , existing methods for 4C-Seq data analysis do not comprehensively identify interactions across the entire genome due to biases in the technique that are related to the decrease in 4C signal that results from increased 3D distance from the bait . To compensate for these weaknesses in existing methods we developed 4C-ker , a method that explicitly models these biases to improve the analysis of 4C-Seq to better understand the genome wide interaction profile of an individual locus .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"blood",
"cells",
"medicine",
"and",
"health",
"sciences",
"immune",
"cells",
"markov",
"models",
"immunology",
"mathematics",
"genome",
"analysis",
"genomic",
"signal",
"processing",
"genomic",
"libraries",
"white",
"blood",
"cells",
"animal",
"cells",
"chromosome",
"biology",
"hidden",
"markov",
"models",
"t",
"cells",
"probability",
"theory",
"antibody-producing",
"cells",
"signal",
"transduction",
"cell",
"biology",
"b",
"cells",
"genetics",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"physical",
"sciences",
"genomics",
"cell",
"signaling",
"computational",
"biology",
"chromosomes"
] |
2016
|
4C-ker: A Method to Reproducibly Identify Genome-Wide Interactions Captured by 4C-Seq Experiments
|
While there has been much recent focus on the ecological causes of adaptive diversification , we know less about the genetic nature of the trade-offs in resource use that create and maintain stable , diversified ecotypes . Here we show how a regulatory genetic change can contribute to sympatric diversification caused by differential resource use and maintained by negative frequency-dependent selection in Escherichia coli . During adaptation to sequential use of glucose and acetate , these bacteria differentiate into two ecotypes that differ in their growth profiles . The “slow-switcher” exhibits a long lag when switching to growth on acetate after depletion of glucose , whereas the “fast-switcher” exhibits a short switching lag . We show that the short switching time in the fast-switcher is associated with a failure to down-regulate potentially costly acetate metabolism during growth on glucose . While growing on glucose , the fast-switcher expresses malate synthase A ( aceB ) , a critical gene for acetate metabolism that fails to be properly down-regulated because of a transposon insertion in one of its regulators . Swapping the mutant regulatory allele with the ancestral allele indicated that the transposon is in part responsible for the observed differentiation between ecological types . Our results provide a rare example of a mechanistic integration of diversifying processes at the genetic , physiological , and ecological levels .
Adaptive diversification describes the splitting of an ancestral lineage into two derived groups due to frequency-dependent ecological interactions [1] . During this process , disruptive selection on a common ancestral type drives the creation of diversified ecotypes through a series of adaptive genetic changes [1 , 2] . Even though this process is of central importance in evolutionary biology [3] , examples are rare where the genetic changes that differentiate adaptively diversified species or ecotypes are known [4–8] . Even when genetic differentiation can be identified , it is often hard to establish a link to phenotypic differentiation , and even harder to show that the associated phenotypes played a causal role in the ecological mechanisms driving diversification [4–7] . Microbial experimental model systems greatly facilitate our ability to connect genotype to phenotype [9] . For instance , in static microcosms Pseudomonas fluorescens can readily diversify into three morphological types from a common ancestor [10–13] . One of these , the wrinkly spreader , dominates the liquid-air interface by forming a biofilm that it creates by overexpression of a polymer-forming operon [7] . While static microcosms provide different spatial niches and therefore an obvious mechanism for niche differentiation , it is less clear how diversification can occur in homogeneous liquid culture . We have used the bacterium Escherichia coli to investigate the ecological and genetic mechanisms of adaptive diversification in a homogeneous , well-mixed environment . When E . coli evolve in shaken serial batch culture with daily depletion of a mixture of glucose and acetate , in each batch they first use up all the available glucose and then undergo a diauxic switch to acetate consumption before entering stationary phase , after which they are transferred to a new batch of resources . After 1 , 000 generations in this homogeneous two-resource environment , E . coli readily diversify into two ecotypes that show different patterns of diauxic resource use . These types were previously dubbed Large and Small to reflect their colony morphology when cultured in a nutrient-rich environment [14–16] . Relative to the Small colonies , Large colonies exhibit high growth rates on glucose , slow growth rates on acetate , and a long lag between growth on glucose and growth on acetate . Thus , the diauxic growth profile of Larges is markedly different from the diauxic growth pattern of Smalls ( Figure 1A ) . The two ecotypes have repeatedly evolved from a single common ancestor , and their coexistence is maintained by negative frequency dependence generated by the daily , sequential depletion of resources [14–16] . This frequency dependence is likely to be generated by a trade-off between the metabolism of glucose and acetate [14]: a short switching time from glucose to acetate consumption is possible if acetate metabolism is active even during growth on glucose , but this in turn reduces the efficiency of glucose metabolism [17] . Because growth rate differences better reflect the selection pressures that caused the bacteria to diversify , we will refer to “slow-switchers” ( SS ) , which correspond to Large colonies , and “fast-switchers” ( FS ) , which refers to Small colonies . A fundamental advantage of the E . coli model over evolution in natural systems is the ability of E . coli to survive cryogenic preservation . Ecological and phenotypic change in divergent E . coli strains that evolved from a common clone can be compared to the cryogenically preserved and revived common ancestor , and the ancestral strain can be manipulated genetically to contain alleles from the derived ecotypes and vice versa . These advantages have allowed us to gain an integrative understanding of genetic , phenotypic , and ecological mechanisms underlying sympatric adaptive diversification due to competition for resources in E . coli . Here , we analyse the changes in resource consumption and the concomitant genetic changes of a lab-based , well-documented evolutionary diversification .
We isolated FS and SS strains from a diversified bacterial population based on their growth profiles and confirmed that the FS switching lag , i . e . , the time elapsed between the end of growth on glucose and the maximum growth rate on acetate , was shorter than that of the ancestral and the SS strains ( F2 , 12 = 2841 . 6 , p < 0 . 0001 , Figure 1A ) . To test the hypothesis that only the FS type has an active acetate metabolism during growth on glucose , we measured the dynamics of acetate concentrations for the different strains over a full day's growth . In wild-type E . coli , acetate metabolism is repressed during growth on glucose [18] , which incidentally generates acetate as a by-product . In wild-type strains , the net concentration of acetate in the medium should therefore increase during growth on glucose . For randomly selected strains of the FS and SS ecotypes and of their common ancestor , we monitored how the glucose and acetate concentrations changed during log growth on glucose ( hours 0 to 4 in Figure 1A ) . As was expected if acetate metabolism is inactivated by the presence of glucose , both the ancestor and the SS generated acetate as a by-product during the glucose consumption phase of growth , so that acetate concentration in the medium increased ( Figure 1B and 1C ) . In a striking deviation from this pattern , the FS strain did not accumulate acetate in its medium as it consumed glucose ( Figure 1B and 1C ) . This indicated probable failure of a genetic mechanism to repress acetate usage during growth on glucose . It is likely that failed repression of acetate metabolism in turn allows for a fast diauxic shift to acetate consumption at the end of the glucose phase . In an attempt to test this , we investigated the genetic basis of the de-repression of acetate metabolism in the FS strain . Likely acetate usage candidate genes include acetyl-CoA synthetase ( acs ) , which converts acetate to acetyl-CoA , and the three-locus acetate operon , aceBAK , which converts acetyl-CoA derivatives to malate ( Figure 2 ) [20] . The operon contains genes encoding malate synthetase A ( aceB ) , isocitrate lyase ( aceA ) , and isocitrate dehydrogenase kinase/phosphatase ( aceK ) , all of which are coexpressed; we selected aceB as a proxy to represent expression of the operon . We measured expression levels using quantitative PCR of aceB for the ancestral and derived strains in medium containing acetate as the sole carbon source , and contrasted this with expression in media containing glucose . When grown on acetate alone , aceB levels were high and equivalent for the ancestral , SS , and FS strains ( Figure 3 ) . This confirms the expectation that the acetate metabolism is active in all three types when growing on acetate . However , when growing on glucose , expression levels of the aceB gene dropped dramatically in the ancestor and the SS strain ( Figure 3 ) . This occurred independently of whether acetate was present in the medium or not and confirmed that , in these strains , acetate metabolism is repressed during growth on glucose . In contrast , the FS strain continued to express high levels of aceB , and by inference the entire acetate operon , even when growing on glucose ( Figure 3 ) . In the glucose-containing media , aceB expression in the FS was reduced relative to the expression levels in acetate-only medium , but remained high relative to the SS and ancestral expression . The high aceB expression levels of the FS during growth on glucose strongly indicate that the FS ecotype has evolved a genetic mechanism by which the acetate operon remains expressed in the presence of glucose . The enhanced aceB expression in the FS ecotype has two potential genetic causes: a change in the regulatory sequence of the acetate operon or a mutation in one of the operon's regulators . Through sequencing we confirmed that the regulatory region of aceB is identical in the FS , ancestor , and the SS . Next , we looked for mutations in the negative regulators of aceBAK because a decrease in negative regulation would cause constitutive expression of the aceBAK operon , congruent with the observed expression level changes . For the FS strain , we discovered that the isocitrate lyase repressor ( iclR ) gene , a negative regulator of aceBAK [21 , 22] , contains a transposable IS1 genetic element that terminates the iclR transcript when it is two-thirds complete . To determine whether this mutation was prevalent in the FS population in addition to our focal FS strain , we screened nine subsequent isolated FS and SS strains for this iclRIS1 allele: eight of nine FS strains carried the mutant allele , but neither of the SS strains , nor the ancestor , carried this insertion . We additionally PCR screened FS genotypes from two similarly evolved populations and recovered only alleles of ancestral size at this locus . Indeed , we would not expect the exact same mutation ( i . e . , insertion of a transposon ) to occur at the same site in two independently evolved populations , as it is likely that there are many different genetic mechanisms by which regulation of acetate metabolism can be altered . To determine how the IS1 insertion in the iclR gene affected acetate use in the FS , we substituted the ancestral iclRAnc allele into the FS genetic background and then estimated growth profile characteristics of the genetically modified strain . Inserting the ancestral iclR allele resulted in FS strains that had a significantly longer lag when switching from glucose to acetate use ( Figure 4A and 4C ) , however this altered lag was still significantly shorter than that of the ancestral strain ( Figure 4C ) . We concluded from this that the mutant iclRIS1 allele in the FS ecotype decreases the amount of time the FS requires to switch from consuming glucose to metabolising acetate , presumably because this allele deregulates the acetate operon and thereby enhances acetate metabolic activity during growth on glucose . We also inserted the FS mutant iclRIS1 allele into the ancestral genetic background , although recombinants were much less common in this direction , a fact that provides some evidence for genetic interactions between this locus and genes in the ancestral genome . In these strains with the ancestral genetic background , the mutant iclRIS1 allele did not affect switching times , which remained long ( Figure 4B and 4C ) . We speculate that the rarity of allelic recombinants during the allelic replacement procedure and the lack of change in switching lag indicate the presence of epistatic effects . In particular , the iclRIS1 insertion only seems to be effective in the genetic background of derived strains , but not in the genetic background of ancestral strains . This would not really be surprising , as the ancestor has no known evolutionary history in the glucose-acetate resource environment and does not carry the set of adaptively beneficial mutations that FS and SS must carry at other loci , based on differences in their growth curves from the ancestor ( Figure 1A ) . We hypothesize that one or more derived alleles interact with the iclR locus to compound the effect of the iclRIS1 insertion in the derived strains . Although the IS1 element disrupts the down-regulation of the acetate operon sufficiently to alter the switching time between resources , it does not exert sole control over the switching lag in the FS . This is evident because the FS with the iclRAnc allele switched to acetate earlier than either of the genetically modified ancestral strains ( Figure 4C ) . Furthermore , in the genetically modified strains , the iclRIS1 ( FS ) allele did not significantly affect colony morphology or growth rates on glucose or acetate ( unpublished data ) . This clearly indicates that this single mutation is not sufficient to cause all of the resource use changes between FS and the ancestor , or between FS and SS . In particular , iclR does not act alone to cause the critical trade-off in performance that enables coexistence between the FS and SS strains . ( Note that the FS strain and the ancestor never coexisted in the same population , and hence we would not expect to see evidence of a trade-off between these strains . ) The observed differences in the growth rates and the colony morphology between the various strains must therefore result from additional modifications to metabolism . Nevertheless , our data show that a genetic change in the regulation of genes controlling carbohydrate metabolism has contributed substantially to the differentiation of coexisting ecotypes in E . coli populations . Our results not only confirm that regulatory changes can provide a mechanism for rapid evolutionary change [4 , 6–8 , 23] , but they show that such regulatory changes may play a crucial role in processes of sympatric diversification . The results also show that such regulatory changes can act upon phylogenetically ancient central metabolic pathways such as the acetate switch found in microorganisms as diverse as gram negative E . coli , gram-positive Bacillus subtilis , and halophilic archea Haloferax volcanii [24] . The importance of acetate utilization on niche adaptation in nature is evident within the gammaproteobacteria , the taxonomic class containing E . coli , as evolutionary changes in the acetate utilization are correlated with pathogenicity in both closely related species such as Shigella [25] and less closely related species such as Yersinia spp [26] . Our current results demonstrate that similar evolutionary changes can also be observed in the laboratory during experimental evolution , and that such regulatory changes are important in niche specialization and differentiation [8] . These results establish a link between different levels of biological organization by showing how a genetic modification of gene regulation affects the expression of genes that are important for metabolic pathways , and how this gene expression in turn affects a trade-off in resource use that causes disruptive selection and competitive diversification .
Previously , we conducted an evolution experiment in which Populations 29 , 31 , and 33 were derived after 100 generations of evolution from an E . coli B strain ( REL606 [27] , hereafter called Anc for ancestor ) under a regimen of daily batch culture in Davis minimal medium supplemented with glucose and acetate [14 , 15] . Strains 33A ( SS ) and 33K ( FS ) were isolated from generation 1 , 000 ( day 150 ) of population 33 and have the characteristic growth rate parameters ( see growth curve assay below ) and colony morphologies for their respective ecotypes . SS colonies grow within 24 h on tryptone agar plates containing 0 . 005% tetrazolium dye , whereas FS colonies require >24 h to grow and are visibly smaller than SS colonies after 48 h . All assays requiring liquid cultures were conducted in Davis minimal ( DM ) medium and supplemented with 0 . 0002% thiamine HCl , 0 . 1% MgSO4 [28] , and carbon sources as follows: 410 mg/L D-glucose monohydrate ( DM-glucose ) , 410 mg/L sodium acetate trihydrate ( DM-acetate ) , 205 mg/L D-glucose monohydrate , and 205 mg/L sodium acetate trihydrate ( DM50:50 ) , unless otherwise noted . All bacteria were cultured at 37 °C . All liquid cultures were aerated by shaking at 250 rpm unless otherwise indicated . Three independent cultures of each of FS , SS , and Anc were sampled throughout log growth in DM50:50 to determine glucose and acetate concentrations . For each sample , we removed approximately 1 ml of culture , centrifuged to pellet the cells , and stored the supernatant at −80 °C . After all samples were obtained , we estimated the glucose concentrations in the supernatant using a Glucose ( HK ) Assay kit ( Sigma , http://www . sigmaaldrich . com ) , and acetate concentrations were determined using a UV-method Acetic Acid kit ( Boehringer Mannheim/R-biopharm , http://www . r-biopharm . com ) . Manufacturer-supplied standards and non-inoculated media samples were used as controls . Colonies were PCR assayed for the presence of the IclRIS1 allele . The sample included nine FS and ten SS from population 33 and three FS and three SS from each of two other populations ( 29 and 31 ) evolved under identical conditions . PCR was conducted in 25 μl volumes ( 400 nM each primer , 400 μM dNTPs , 1X reaction buffer , and 5 U Taq [Roche , www . roche-diagnostics . com] ) . Primers 5′-TCGAAAATACACGCTGCAAG and 5′-TTCCACTTTGCTGCTCACAC amplified from 217 bp upstream of the IclR start site through 83% of the gene . The template for each reaction was sampled from a single colony grown for 24 h ( for SS and Anc ) or 48 h ( for FS ) . Reaction conditions were as follows: 95 °C for 10 min , 30 cycles amplification ( 95 °C for 30 s , 64 °C for 30 s , and 72 °C for 1 min ) , and final extension period ( 72 °C for 5 min ) . PCR products were size separated in 2 . 0% agarose to assay for the slower migration of the iclRIS1 allele relative to the wild type . For the regulatory regions of acs , aceBAK , and iclR and for the iclR gene , we sequenced strains 33A , 33K , and Anc . Additional sequence data were collected for acs from one additional FS and SS from population 33 as well as two FS and two SS from two additional populations . Sequencing was performed by the University of British Columbia's Nucleic Acid and Protein Service ( NAPS ) , using the ABI PRISM Big Dye V . 3 . 1 sequencing kit ( Applied Biosystems , http://www . appliedbiosystems . com ) . Primers were from Treves et al . [29] ( acs regulatory region ) , 5′-GCTGGCGTAGTCACGGTAAT and 5′-ATCGGTTGTTGTTGCCTGTT ( aceB regulatory region ) , or as for PCR and qPCR . Three independent samples each of 33K , 33A , and Anc were prepared from early log growth in DM-glucose , DM-acetate , and DM50:50 . Cultures were collected on ice , pelleted , and resuspended in RNAlater ( Ambion , http://www . ambion . com ) . RNAs were prepared using RNeasy ( Qiagen , http://www . qiagen . com ) or Ribopure ( Ambion ) kits , following the manufacturers' protocols . Remaining genomic DNA was removed using DNAfree ( Ambion ) . cDNAs were generated using the TaqMan kit ( Applied Biosystems ) , primed with random hexamers . Reactions were carried out in duplicate in a 7000 SDS ( ABI ) using SYBR green master mix ( ABI ) according to the manufacturers protocol and primers designed with Primer Express 2 . 0 ( ABI ) . Primers used were: 5′-TGGCGTGGTGAGGCAAT and 5′-GGAAGAAATAGAGCGCAAAATCA ( aceB ) and 5′-GCTGATACCGCCCAAGAGTTC and 5′-CAGGATGTGATGAGCCGAC ( 23SrRNA-4 ) . Controls lacking reverse transcriptase were included for each sample to detect genomic DNA contamination and primer dimerization . As well , dissociation curves were used to confirm a single product . Standard curves were prepared for each primer set using five serial 5-fold dilutions of Anc genomic DNA template . RNA abundance was quantified by normalizing the quantity of cDNA aceB template to the quantity of cDNA 23SrRNA-4 template , chosen as a control because of its unchanging , high expression in both glucose and acetate growth conditions [20] . We transferred iclRIS1 to Anc and iclRAnc into FS using a suicide plasmid mediated technique [30] but substituting the suicide plasmid pRE112 [31] . Control transfers to return each allele into its original genetic background were also performed . Growth curve data were generated using a Bioscreen C plate reader ( MTX Lab Systems Inc , http://www . mtxlsi . com ) . Bacteria were inoculated into assay wells containing 250 μl aliquots of DM supplemented with 0 . 0002% thiamine HCl , 0 . 1% MgSO4 , and one of three resource combinations: 410 mg/L glucose , 410 mg/L acetate , or 41 mg/L glucose and 369 mg/L acetate ( DM10:90 ) . Cultures were grown for 24 h at 37 °C with continuous shaking and measured for wide band optical density every 10 min . Five technical replicates of each sample were averaged after anomalous growth curves were removed . We used C++ code provided by J . Tyerman to estimate three growth curve parameters: the maximum growth rates on glucose and acetate and switching time between resources , defined as the time elapsed between the end of growth on glucose and the maximum growth rate on acetate . Briefly , the code linearizes the data over a sliding window of nine time units to establish the changes in slope of the optical density over a 24-h growth period . The first critical point where the slope changes from positive to zero ( or negative ) defines the switching point . We calculate the glucose growth rate as the maximum slope between time 0 and the switching point , the acetate growth rate as the maximum slope between the switching point and the time of maximum optical density , and the switching lag as the difference in time units between the switching point and the maximum growth rate on acetate . Analyses of variance ( ANOVAs ) were calculated for these three parameter estimates for the FS , SS , and Anc strains and for the four genetically modified strains , Anc ( iclRAnc ) , Anc ( iclRIS1 ) , FS ( iclRAnc ) , and FS ( iclRIS1 ) , described above . ANOVAs were conducted in JMP [32] using Tukey Honestly Significant Difference ( HSD ) multiple comparisons . All Tukey HSD tests were significant for the FS , SS , and Anc multiple comparisons . Results from the genetically modified strains are given in Figure 4 .
The EchoBASE ( http://www . biolws1 . york . ac . uk/echobase ) accession numbers for the E . coli strains discussed in this paper are acetyl-CoA synthetase ( EB1417 ) , malate synthetase A ( EB0022 ) , isocitrate lyase ( EB0021 ) , isocitrate dehydrogenase kinase/phosphatase ( EB0025 ) , and isocitrate lyase repressor ( EB0486 ) .
|
Understanding the origin of diversity is a fundamental problem in evolutionary biology . The past decade has seen a shift in our understanding of speciation , away from considering geographical isolation as the main cause and towards elucidating how ecological interactions can drive diversification in populations that occupy a single and contiguous spatial area , a process called sympatric diversification . By culturing bacteria over many generations it is possible to observe processes of diversification in real time . This paper characterizes diversification caused by ecological interactions in bacteria at the physiological and genetic level . Propagating a single ancestral E . coli strain on a mixture of two resources , we found sympatric diversification into two descendant strains . This diversification occurs in a shared , well-mixed environment and is caused by competition for resources . We show that 1 ) the diversified strains use physiological pathways differently to consume the resources , 2 ) this physiological difference is caused by differences in the expression levels of genes controlling metabolism , and 3 ) this difference in gene expression is influenced by genetic differences in regulatory genes . Our paper thus contributes to an integrative understanding of sympatric diversification in E . coli at the genetic , physiological , and ecological levels .
|
[
"Abstract",
"Introduction",
"Results/Discussion",
"Materials",
"and",
"Methods",
"Supporting",
"Information"
] |
[
"evolutionary",
"biology",
"genetics",
"and",
"genomics",
"eubacteria"
] |
2007
|
Adaptive Diversification in Genes That Regulate Resource Use in Escherichia coli
|
We have developed an easy-to-use and memory-efficient method called PhenotypeSeeker that ( a ) identifies phenotype-specific k-mers , ( b ) generates a k-mer-based statistical model for predicting a given phenotype and ( c ) predicts the phenotype from the sequencing data of a given bacterial isolate . The method was validated on 167 Klebsiella pneumoniae isolates ( virulence ) , 200 Pseudomonas aeruginosa isolates ( ciprofloxacin resistance ) and 459 Clostridium difficile isolates ( azithromycin resistance ) . The phenotype prediction models trained from these datasets obtained the F1-measure of 0 . 88 on the K . pneumoniae test set , 0 . 88 on the P . aeruginosa test set and 0 . 97 on the C . difficile test set . The F1-measures were the same for assembled sequences and raw sequencing data; however , building the model from assembled genomes is significantly faster . On these datasets , the model building on a mid-range Linux server takes approximately 3 to 5 hours per phenotype if assembled genomes are used and 10 hours per phenotype if raw sequencing data are used . The phenotype prediction from assembled genomes takes less than one second per isolate . Thus , PhenotypeSeeker should be well-suited for predicting phenotypes from large sequencing datasets . PhenotypeSeeker is implemented in Python programming language , is open-source software and is available at GitHub ( https://github . com/bioinfo-ut/PhenotypeSeeker/ ) .
The falling cost of sequencing has made genome sequencing affordable to a large number of labs , and therefore , there has been a dramatic increase in the number of genome sequences available for comparison in the public domain [1] . These developments have facilitated the genomic analysis of bacterial isolates . An increasing amount of bacterial whole genome sequencing ( WGS ) data has led to more and more genome-wide studies of DNA variation related to different phenotypes . Among these studies , antibiotic resistance phenotypes are the most concerning and have garnered high public interest , especially since several multidrug-resistant strains have emerged worldwide . The detection of known resistance-causing mutations as well as the search for new candidate biomarkers leading to resistance phenotypes requires reasonably rapid and easily applicable tools for processing and comparing the sequencing data of hundreds of isolated strains . However , there is still a lack of user-friendly software tools for the identification of genomic biomarkers from large sequencing datasets of bacterial isolates [2 , 3] . While microbial genome-wide association studies ( GWAS ) can be successfully used in case of previously known genotype-phenotype associations caused by a single gene or only a set of few and specific mutations , more complex and novel associations would remain undetected . In addition , many bacterial species have extensive intra-species variation from small sequence-based differences to the absence or presence of whole genes or gene clusters . Choosing only one genome as a reference for searching for the variable components would be highly limiting . Alternative approaches use previously detected genomic features , either single nucleotide variations or longer sequences , behind the phenotype to create and train models using those features as the predictors . Not only antibiotic resistance but wide range of other phenotypes can be predicted , e . g host adaptation in invasive serovars [4] , needed minimum inhibitory concentrations of antibiotics [5] or virulence of the strains [6] . Using longer sequence regions , such as full genes in those models , requires assembled genomes as an input which adds data preprocessing step . The solution to avoid this is using k-mers , which are short DNA oligomers with length k , that enable us to simultaneously discover a large set of single nucleotide variations , insertions and deletions associated with the phenotypes under study . The advantage of using k-mer-based methods in genomic biomarker discovery is that they do not require sequence alignments and can even be applied to raw sequencing data . In recent years several publications using different machine learning algorithms and k-mers for detecting the biomarkers behind different bacterial phenotypes have been published . Among the latest , short k-mers and machine learning ( ML ) has been used to create minimum inhibitory concentration prediction models in assembled Klebsiella pneumoniae genomes for several antibiotics [7] . PATRIC and RAST annotation services include prediction of antimicrobial resistance with the species- and antibiotic-specific classifier k-mers which are selected using publicly available and collected metadata and the adaptive boosting ML algorithms [8] . Though providing a framework or predictive models for a specific species with a certain phenotype , those studies have not been concentrating on the creation of a software easily applicable by a wider public without an access to extensive computing resources but still having the need for analyzing large scale bacterial genome sequencing data with a reasonable amount of computing time . Only few papers describe software which we were able to compare with PhenotypeSeeker . The SEER program takes either a discrete or continuous phenotype as an input , counts variable-length k-mers and corrects for the clonal population structure [6] . SEER is a complex pipeline requiring several separate steps for the user to execute and currently has many system-level dependencies for successful compilation and installation . Another similar tool , Kover , handles only discrete phenotypes , counts user-defined size k-mers and does not use any correction for population structure [9] . The Neptune software targets so-called 'signatures' differentiating two groups of sequences but cannot locate smaller mutations , such as single isolated nucleotide variations , being the reason , it was not used in the comparison in current paper . The 'signatures' that Neptune detects are relatively large genomic loci , which may include genomic islands , phage regions or operons [10] . We created PhenotypeSeeker as we observed the need for a tool that could combine all the benefits of the programs available but at the same time would be easily executable and would take a reasonable amount of computing resources without the need for dedicated high-performance computer hardware .
PhenotypeSeeker consist of two subprograms: 'PhenotypeSeeker modeling' and 'PhenotypeSeeker prediction' . 'PhenotypeSeeker modeling' takes either assembled contigs or raw-read data as an input and builds a statistical model for phenotype prediction . The method starts with counting all possible k-mers from the input genomes , using the GenomeTester4 software package [11] , followed by k-mer filtering by their frequency in strains . Subsequently , the k-mer selection for regression analysis is performed . In this step , to test the k-mers’ association with the phenotype , the method applies Welch’s two-sample t-test if the phenotype is continuous and a chi-squared test if it is binary . Finally , the logistic regression or linear regression model is built . The PhenotypeSeeker output gives the regression model in a binary format and three text files , which include the following: ( 1 ) the results of association tests for identifying the k-mers most strongly associated with the given phenotype , ( 2 ) the coefficients of k-mers in the regression model for identifying the k-mers that have the greatest effects on the outcomes of the machine learning model , ( 3 ) a FASTA file with phenotype-specific k-mers , assembled to longer contigs when possible , to facilitate an user to perform annotation process , and ( 4 ) a summary of the regression analysis performed ( Fig 1 ) . Optionally , it is possible to use weighting for the strains to take into account the clonal population structure . The weights are based on a distance matrix of strains made with an alignment-free k-mer-based method called Mash [12] . The weights of each genome are calculated using the Gerstein , Sonnhammer and Cothia method [13] . 'PhenotypeSeeker prediction' uses the regression model generated by 'PhenotypeSeeker modeling' to conduct fast phenotype predictions on input samples ( Fig 1 ) . Using gmer_counter from the FastGT package [14] , the tool searches the samples only for the k-mers used as parameters in the regression model . Predictions are then made based on the presence or absence of these k-mers . PhenotypeSeeker uses fixed-length k-mers in all analyses . Thus , the k-mer length is an important factor influencing the overall software performance . The effects of k-mer length on speed , memory usage and accuracy were tested on a P . aeruginosa ciprofloxacin dataset . A general observation from that analysis is that the CPU time and the PhenotypeSeeker memory usage increase when the k-mer length increases ( Fig 2 ) . Previously described mutations in the P . aeruginosa parC and gyrA genes were always detected if the k-mer length was at least 13 nucleotides . We assume that in most cases , a k-mer length of 13 is sufficient to detect biologically relevant mutations , although in certain cases , longer k-mers might provide additional sensitivity . The k-mer length in PhenotypeSeeker is a user-selectable parameter . Although most of the phenotype detection can be performed with the default k-mer value , we suggest experimenting with longer k-mers in the model building phase . All subsequent analyses in this article are performed with a k-mer length of 13 , unless specified otherwise . PhenotypeSeeker was applied to the dataset composed of P . aeruginosa genomes and corresponding ciprofloxacin resistance values measured in terms of minimum inhibitory concentration ( MIC ) ( μg/ml ) , which is defined as the lowest concentration of antibiotic that will inhibit the visible growth of the isolate under investigation after an appropriate period of incubation [15] . We built two separate models using a continuous phenotype for one and binary phenotype for another . Binary phenotype values were created based on EUCAST ciprofloxacin breakpoints [16] . Both models detected k-mers associated with mutations in quinolone resistance determining regions ( QRDR ) of the parC ( c . 260C>T , p . Ser87Leu ) and gyrA ( c . 248C>T , p . Thr83Ile ) genes ( Fig 3 , S2 Table ) . These genes encode DNA topoisomerase IV subunit A and DNA gyrase subunit A , the target proteins of ciprofloxacin [17] . Mutations in the QRDR regions of these genes are well-known causes of decreased sensitivity to quinolone antibiotics , such as ciprofloxacin [18] . The classification model built using a binary phenotype had a F1-measure of 0 . 88 , prediction accuracy of 0 . 88 , sensitivity of 0 . 90 and specificity of 0 . 87 on the test subset ( Table A in S3 Table ) . The MIC prediction model built using the continuous phenotype had the coefficient of determination ( R2 ) of 0 . 42 , the Pearson correlation coefficient of 0 . 68 and the Spearman correlation coefficient of 0 . 84 ( Table M in S3 Table ) . In addition to the P . aeruginosa dataset , we tested a C . difficile azithromycin resistance dataset ( S2 Table ) studied using Kover in Drouin et al . , 2016 [9] . ermB and Tn6110 transposon were the sequences known and predicted to be important in an azithromycin resistance model by Kover [9] . ermB was not located on the transposon Tn6110 . PhenotypeSeeker found k-mers for both sequences while using k-mers of length 13 or 16 . Tn6110 is a transposon that is over 58 kbp long and contains several protein coding sequences , including 23S rRNA methyltransferase , which is associated with macrolide resistance [19] . The predictive models with all tested k-mer lengths ( 13 , 16 and 18 ) contained k-mers covering the entire Tn6110 transposon sequence , both in protein coding and non-coding regions . In addition to the 23S rRNA methyltransferase gene , k-mers in all three models were mapped to the recombinase family protein , sensor histidine kinase , ABC transporter permease , TlpA family protein disulfide reductase , endonuclease , helicase and conjugal transfer protein coding regions . The model built for the C . difficile azithromycin resistance phenotype had a F1-measure of 0 . 97 , prediction accuracy of 0 . 97 , sensitivity of 0 . 96 and specificity of 0 . 97 on the test subset ( Table A in S3 Table ) . In addition to antibiotic resistance phenotypes in P . aeruginosa and C . difficile , we used K . pneumoniae human infection-causing strains as a different kind of phenotype example . K . pneumoniae strains contain several genetic loci that are related to virulence . These loci include aerobactin , yersiniabactin , colibactin , salmochelin and microcin siderophore system gene clusters [20–24] , the allantoinase gene cluster [25] , rmpA and rmpA2 regulators [26 , 27] , the ferric uptake operon kfuABC [28] and the two-component regulator kvgAS [29] . The model predicted by PhenotypeSeeker for invasive/infectious phenotypes included 13-mers representing several of these genes . Genes in colibactin ( clbQ and clbO ) , aerobactin ( iucB and iucC ) and yersiniabactin ( irp1 , irp2 , fyuA , ybtQ , ybtX , and ybtP ) clusters showed the most differentiating pattern between carrier and invasive/infectious strains ( Fig 4; S2 Table ) . A 13-mer mapping to a gene-coding capsule assembly protein Wzi was also represented in the model . The model built for K . pneumoniae invasive/infectious phenotypes had a F1-measure of 0 . 88 , prediction accuracy of 0 . 88 , sensitivity of 0 . 91 and specificity of 0 . 78 on the test subset ( Table A in S3 Table ) . To measure the average classification accuracies of logistic regression models , all three datasets were divided into a training and test set of approximately 75% and 25% of strains respectively . A K-mer length of 13 was used , and a weighted approach was tested on binary phenotypes ( Table 1 ) . To reduce the influence of sequencing errors when using sequencing reads instead of assembled contigs as the input , we only counted 13-mers as being present in one of the input lists if they occurred at least 5 times in that input list . The PhenotypeSeeker prediction accuracy is not lower when using raw sequencing reads instead of assembled genomes , and therefore , assembly building is not required before model building . Our results with K . pneumoniae show that PhenotypeSeeker can be successfully applied to other kinds of phenotypes in addition to antibiotic resistance . In our trials , the model building on a given dataset took 3 to 5 hours per phenotype , and prediction of the phenotype took less than a second on assembled genomes ( Table 1 ) . The CPU time of model building by PhenotypeSeeker depends mainly on the number of different k-mers in genomes of the training set . The analysis performed on our 200 P . aeruginosa genomes showed that the CPU time of the model building grows linearly with the number of genomes given as input ( S1 Fig ) . The memory requirement of PhenotypeSeeker did not exceed 2 GB if default parameter settings are used , allowing us to run analyses on laptop computers ( S2 Fig ) if necessary . The p-value cut-offs during the k-mer filtering step influence the number of k-mers included in the model and have a potentially strong impact on model performance . Tables A-E in the S1 Table show the effects of different p-value cut-offs on model performances . We ran SEER and Kover on the same P . aeruginosa ciprofloxacin dataset and C . difficile azithromycin resistance dataset to compare the efficiency and CPU time usage with PhenotypeSeeker . In the P . aeruginosa dataset , SEER was able to detect gyrA and parC mutations only when resistance was defined as a binary phenotype . In cases with a continuous phenotype , those k-mers did not pass the p-value filtering step . Since Kover's aim is to create a resistance predicting model , not an exhaustive list of significant k-mers , it was expected that not all the mutations would be described in the output . gyrA variation already sufficiently characterized the resistant strains set , and therefore , parC mutations were not included in the model . The same applies to the PhenotypeSeeker results with 16- and 18-mers . parC-specific 16- or 18-mers were included among the 1000 k-mers in the prediction model ( based on statistically significant p-values ) but with the regression coefficient equal to zero because they were present in the same strains as gyrA specific predictive k-mers . In the C . difficile dataset , our model included the known resistance gene ermB and transposon Tn6110 . We were able to find ermB with both SEER and Kover . We also detected Tn6110-specific k-mers with SEER while running Kover with 16-mers instead of 31-mers as in the default settings . Regarding the CPU time , PhenotypeSeeker with 13-mers was faster than other tested software programs ( 3 . 5 hrs vs 14–15 hrs ) without losing the relevant markers in the output ( Table 2 ) . Using 16- or 18-mers , the PhenotypeSeeker’s running time increases but is still lower than with SEER and Kover .
PhenotypeSeeker works as an easy-to-use application to list the candidate biomarkers behind a studied bacterial phenotype and to create a predictive model . Based on k-mers , PhenotypeSeeker does not require a reference genome and is therefore also usable for species with very high intraspecific variation where the selection of one genome as a reference can be complicated . PhenotypeSeeker supports both discrete and continuous phenotypes as inputs . In addition , this model takes into account the population structure to highlight only the possible causal variations and not the mutations arising from the clonal nature of bacterial populations . Unlike Kover , the PhenotypeSeeker output is not merely a trained model for predicting resistance in a separate set of isolates , but the complete list of statistically significant candidate variations separating antibiotic resistant and susceptible isolates for further biological interpretation is also provided . Unlike SEER , PhenotypeSeeker is easier to install and can be run with only a single command for building a model and another single command to use it for prediction . Our tests using PhenotypeSeeker to detect antibiotic resistance markers in P . aeruginosa and C . difficile showed that it is capable of detecting all previously known mutations in a reasonable amount of time and with a relatively short k-mer length . Users can choose the k-mer length as well as decide whether to use the population structure correction step . Due to the clonal nature of bacterial populations , this step is highly advised for detecting genuine causal variations instead of strain-level differences . In addition to a trained predictive model , the list of k-mers covering possible variations related to the phenotype are produced for further interpretation by the user . The effectiveness of the model can vary because of the nature of different phenotypes in different bacterial species . Simple forms of antibiotic resistance that are unambiguously determined by one or two specific mutations or the insertion of a gene are likely to be successfully detected by our method , and effective predictive models for subsequent phenotype predictions can be created . This is supported by our prediction accuracy over 96% in the C . difficile dataset . On the other hand , P . aeruginosa antibiotic resistance is one of the most complicated phenotypes among clinically relevant pathogens since it is not often easily described by certain single nucleotide mutations in one gene but rather through a complex system involving several genes and their regulators leading to multi-resistant strains . In cases such as this , the prediction is less accurate ( 88% in our dataset ) , but nevertheless , a complete list of k-mers covering differentiating markers between resistant and sensitive strains can provide more insight into the actual resistance mechanisms and provide candidates for further experimental testing . Tests with K . pneumoniae virulence phenotypes showed that PhenotypeSeeker is not limited to antibiotic resistance phenotypes but is potentially applicable to other measurable phenotypes as well and is therefore usable in a wider range of studies . Since PhenotypeSeeker input is not restricted to assembled genomes , one can skip the assembly step and calculate models based on raw read data . In this case , it should be taken into account that sequencing errors may randomly generate phenotype-specific k-mers; thus , we suggest using the built-in option to remove low frequency k-mers . The k-mer frequency cut-off threshold depends on the sequencing coverage of the genomes and is therefore implemented as user-selectable . One can also build the model based on high-quality assembled genomes and then use the model for corresponding phenotype prediction on raw sequencing data .
PhenotypeSeeker was tested on the following three bacterial species: Pseudomonas aeruginosa , Clostridium difficile and Klebsiella pneumoniae . The P . aeruginosa dataset was composed of 200 assembled genomes and the minimal inhibitory concentration measurements ( MICs ) for ciprofloxacin . The P . aeruginosa strains were isolated during the project Transfer routes of antibiotic resistance ( ABRESIST ) performed as part of the Estonian Health Promotion Research Programme ( TerVE ) implemented by the Estonian Research Council , the Ministry of Agriculture ( now the Ministry of Rural Affairs ) , and the National Institute for Health Development . Isolated strains originated from humans , animals and the environment within the same geographical location in Estonia and belonged to 103 different MLST sequence types ( Laht et al . , Pseudomonas aeruginosa distribution among humans , animals and the environment ( submitted ) ; Telling et al . , Multidrug resistant Pseudomonas aeruginosa in Estonian hospitals ( submitted ) ) . Full genomes were sequenced by Illumina HiSeq2500 ( Illumina , San Diego , USA ) with paired-end , 150 bp reads ( Nextera XT libraries ) and de novo assembled with the program SPAdes ( ver 3 . 5 . 0 ) [30] . MICs were determined by using the epsilometer test ( E-test , bioMérieux , Marcy l'Etoile , France ) according to the manufacturer instructions . Binary phenotypes were achieved by converting the MIC values into 0 ( sensitive ) and 1 ( resistant ) phenotypes according to the European Committee on Antimicrobial Susceptibility Testing ( EUCAST ) breakpoints [16] . The resulted dataset consisted of 124 ciprofloxacin sensitive P . aeruginosa isolates ( 62% ) and 76 ciprofloxacin resistant P . aeruginosa isolates ( 38% ) and is deposited in the NCBI’s BioProject database under the accession number PRJNA244279 ( https://www . ncbi . nlm . nih . gov/bioproject/ ? term=PRJNA244279 ) . The C . difficile dataset was composed of assembled genomes of 459 isolates and the binary phenotypes of azithromycin resistance ( sensitive = 0 vs resistant = 1 ) , adapted from Drouin et al . , 2016 [9] . The isolates originated from patients from different hospitals in the province of Quebec , Canada and the genomes were received from the European Nucleotide Archive [EMBL:PRJEB11776 ( ( http://www . ebi . ac . uk/ena/data/view/PRJEB11776 ) ] . The dataset consisted of 246 azithromycin sensitive isolates ( 54% ) and 213 azithromycin resistant isolates ( 46% ) . The K . pneumoniae dataset was composed of reads of 167 isolates , originating from six countries and sampled to maximize diversity , and the binary clinical phenotype of human carriage status vs human infection ( including invasive infections ) status ( carriage = 0 vs infectious = 1 ) , adapted from Holt et al . , 2015 [31] . The reads were received from the European Nucleotide Archive [EMBL:PRJEB2111 ( https://www . ebi . ac . uk/ena/data/view/PRJEB2111 ) ] and de novo assembled with SPAdes ( ver 3 . 10 . 1 ) [30] . The dataset consisted of 36 isolates with human carriage status as phenotype ( 22% ) and 131 K . pneumonia isolates with human infection status as phenotype ( 78% ) . Abstractly , each test dataset was composed of pairs ( x , y ) , where x is the bacterial genome x∈{A , T , G , C}* , and y denotes phenotype values specific to a given dataset y ∈ {0 . 008 , … , 1024} ( continuous phenotype ) or y ∈ {0 , 1} ( binary phenotype ) . All operations with k-mers are performed using the GenomeTester4 software package containing the glistmaker , glistquery and glistcompare programs [11] . At first , all k-mers from all samples are counted with glistmaker , which takes either FASTA or FASTQ files as an input and enables us to set the k-mer length up to 32 nucleotides . Subsequently , the k-mers are filtered based on their frequency in strains of the training set . By default , the k-mers that are present in or missing from less than two samples are filtered out and not used in building the model . The remaining k-mers are used in statistical testing for detection of association with the phenotype . By default , PhenotypeSeeker conducts the clonal population structure correction step by using a sequence weighting approach that reduces the weight of isolates with closely related genomes . For weighting , pairwise distances between genomes of the training set are calculated using the free alignment software Mash with default parameters ( k-mer size of 21 nucleotides and sketch size of 1000 min-hasehes ) [12] . Distances estimated by Mash are subsequently used to calculate weights for each genome according to the algorithm proposed by Gerstein , Sonnhammer and Chothia [13] . The calculation of GSC weights is conducted using the PyCogent python package [32] . The GSC weights are taken into account while calculating Welch two-sample t-tests or chi-squared tests to test the k-mers’ associations with the phenotype . Additionally , the GSC weights can be used in the final logistic regression or linear regression ( if Ridge regularization is used ) model generation . In the case of binary phenotype input , the chi-squared test is applied to every k-mer that passes the frequency filtration to determine the k-mer association with phenotype . The null hypothesis assumes that there is no association between k-mer presence and phenotype . The alternative hypothesis assumes that the k-mer is associated with phenotype . The chi-squared test is conducted on these observed and expected values with degrees of freedom = 1 , using the scipy . stats Python package [33] . If the user selects to use the population structure correction step , then the weighted chi-squared tests are conducted according to the previously published method [34] . In the case of continuous phenotype input , the Welch two-sample t-test is applied to every k-mer that passes the frequency filtration to determine if the mean phenotype values of strains having the k-mer are different from the mean phenotype values of strains that do not have the k-mer . The null hypothesis assumes that the strains with a k-mer have different mean phenotype values from the strains without the k-mer . The alternative hypothesis assumes that the means of the strains with and without the k-mer are the same . The t-test is conducted with these values using the scipy . stats Python package [33] , assuming that the samples are independent and have different variance . If the user selects the population structure correction step , then the weighted t-tests are conducted [34] . In that case , the p-value is calculated with the function scipy . stats . t . sf , which takes the absolute value of the t-statistic and the value of degrees of freedom as the input . To perform the regression analysis , first , the matrix of samples times features is created . The samples in this matrix are strains given as the input and the features represent the k-mers that are selected for the regression analysis . The values ( 0 or 1 ) in this matrix represent the presence or absence of a specific k-mer in the specific strain . The target variables of this regression analysis are the resistance values of the strains . Thereupon , input data are divided into training and test sets whose sizes are by default 75% and 25% of the strains , respectively . The proportion of class labels in the training and test sets are kept the same as in the original undivided dataset . In the case of a continuous phenotype , a linear regression model is built , and in the case of a binary phenotype , a logistic regression model is built . The logistic regression was selected for binary classification task as it showed better performance on our datasets than other tested machine learning classifiers like support vector machine ( with no kernel and with Gaussian kernel ) and random forest . The performance of logistic regression models on our tested datasets in comparison to performance of other machine learning classifiers are shown in S3 Fig and in Tables A-L in S3 Table . The performance of linear regression model on P . aeruginosa dataset is shown in Table M in S3 Table . For both the linear and logistic regression , the Lasso , Ridge or Elastic Net regularization can be selected . The Lasso and Elastic Net regularizations shrink the coefficients of non-relevant features to zero , which simplifies the identification of k-mers that have the strongest association with the phenotype . To enable the evaluation of the output regression model , PhenotypeSeeker provides model-evaluation metrics . For the logistic regression model quality , PhenotypeSeeker provides the mean accuracy as the percentage of correctly classified instances across both classes ( 0 and 1 ) . Additionally , PhenotypeSeeker provides F1-score , precision , recall , sensitivity , specificity , AUC-ROC , average precision ( area under the precision-recall curve ) , Matthews correlation coefficient ( MCC ) , Cohen’s kappa , very major error rate and major error rate as metrics to assess model performance . For the linear regression model , PhenotypeSeeker provides the mean squared error , the coefficient of determination ( R2 ) , the Pearson and the Spearman correlation coefficients and the within ±1 two-fold dilution factor accuracy ( useful for evaluating the MIC predictions ) as metrics to assess model performance . To select for the best regularization parameter alpha , a k-fold cross-validation on the training data is performed . By default , 25 alpha values spaced evenly on a log scale from 1E-6 to 1E6 are tested with 10-fold cross-validation and the model with the best mean accuracy ( logistic regression ) or with the best coefficient of determination ( linear regression ) is saved to the output file . Regression analysis is conducted using the sklearn . linear_model Python package [35] . Our models were created using mainly k-mer length 13 ( “-l 13”; default ) . We counted the k-mers that occurred at least once per sample ( “-c 1”; default ) when the analysis was performed on contigs or at least five times per sample ( “-c 5” ) when the analysis was performed on raw reads . In the first filtering step , we filtered out the k-mers that were present in or missing from less than two samples ( “—min 2—max 2”; default ) when the analysis was performed on a binary phenotype or fewer than ten samples ( “—min 10—max N-10”; N–total number of samples ) when the analysis was performed on a continuous phenotype . In the next filtering step , we filtered out the k-mers with a statistical test p-value larger than 0 . 05 ( “—p_value 0 . 05”; default ) . The regression analysis was performed with a maximum of 1000 lowest p-valued k-mers ( “—n_kmers; 1000”; default ) when the analysis was done with binary phenotype and with a maximum of 10 , 000 lowest p-valued k-mers ( “—n_kmers 10000”; default ) when the analysis was performed with a continuous phenotype . For regression analyses , we split our datasets into training ( 75% ) and test ( 25% ) sets ( “-s 0 . 25”; default ) . The regression analyses were conducted using Lasso regularization ( “-r L1”; default ) , and the best regularization parameter was picked from the 25 regularization parameters spaced evenly on a log scale from 1E-6 to 1E6 ( “—n_alphas 25—alpha_min 1E-6—alpha_max 1E6”; default ) . The model performances with each regularization parameter were evaluated by cross-validation with 10-folds ( “—n_splits 10”; default ) . The correction for clonal population structure ( “—weights +”; default ) and assembly of k-mers used in the regression model ( “—assembly +”; default ) were conducted in all our analyses . SEER was installed and run on a local server with 32 CPU cores and 512 GB RAM , except the final step , which we were not able to finish without segmentation fault . This last SEER step was launched via VirtualBox in ftp://ftp . sanger . ac . uk/pub/pathogens/pathogens-vm/pathogens-vm . latest . ova . Both binary and continuous phenotypes were tested for P . aeruginosa and the binary phenotype in C . difficile cases . Default settings were used . Kover was installed on a local server and used with the settings suggested by the authors in the program tutorial .
|
Predicting phenotypic properties of bacterial isolates from their genomic sequences has numerous potential applications . A good example would be prediction of antimicrobial resistance and virulence phenotypes for use in medical diagnostics . We have developed a method that is able to predict phenotypes of interest from the genomic sequence of the isolate within seconds . The method uses a statistical model that can be trained automatically on isolates with known phenotype . The method is implemented in Python programming language and can be run on low-end Linux server and/or on laptop computers .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[
"sequencing",
"techniques",
"medicine",
"and",
"health",
"sciences",
"gut",
"bacteria",
"pathology",
"and",
"laboratory",
"medicine",
"statistics",
"pathogens",
"microbiology",
"pseudomonas",
"aeruginosa",
"genome",
"sequencing",
"antibiotic",
"resistance",
"mathematics",
"forecasting",
"genome",
"analysis",
"pharmacology",
"molecular",
"biology",
"techniques",
"klebsiella",
"bacteria",
"bacterial",
"pathogens",
"research",
"and",
"analysis",
"methods",
"clostridium",
"difficile",
"pseudomonas",
"antimicrobial",
"resistance",
"genomics",
"medical",
"microbiology",
"mathematical",
"and",
"statistical",
"techniques",
"microbial",
"pathogens",
"molecular",
"biology",
"gene",
"identification",
"and",
"analysis",
"genetics",
"microbial",
"control",
"biology",
"and",
"life",
"sciences",
"klebsiella",
"pneumoniae",
"physical",
"sciences",
"computational",
"biology",
"mutation",
"detection",
"statistical",
"methods",
"organisms"
] |
2018
|
A k-mer-based method for the identification of phenotype-associated genomic biomarkers and predicting phenotypes of sequenced bacteria
|
Temperate phages have the ability to maintain their genome in their host , a process called lysogeny . For most , passive replication of the phage genome relies on integration into the host's chromosome and becoming a prophage . Prophages remain silent in the absence of stress and replicate passively within their host genome . However , when stressful conditions occur , a prophage excises itself and resumes the viral cycle . Integration and excision of phage genomes are mediated by regulated site-specific recombination catalyzed by tyrosine and serine recombinases . In the KplE1 prophage , site-specific recombination is mediated by the IntS integrase and the TorI recombination directionality factor ( RDF ) . We previously described a sub-family of temperate phages that is characterized by an unusual organization of the recombination module . Consequently , the attL recombination region overlaps with the integrase promoter , and the integrase and RDF genes do not share a common activated promoter upon lytic induction as in the lambda prophage . In this study , we show that the intS gene is tightly regulated by its own product as well as by the TorI RDF protein . In silico analysis revealed that overlap of the attL region with the integrase promoter is widely encountered in prophages present in prokaryotic genomes , suggesting a general occurrence of negatively autoregulated integrase genes . The prediction that these integrase genes are negatively autoregulated was biologically assessed by studying the regulation of several integrase genes from two different Escherichia coli strains . Our results suggest that the majority of tRNA-associated integrase genes in prokaryotic genomes could be autoregulated and that this might be correlated with the recombination efficiency as in KplE1 . The consequences of this unprecedented regulation for excisive recombination are discussed .
Temperate bacteriophages are characterized by their ability to maintain their genome into the host , a process called lysogeny . Most temperate phages integrate their genome into the host's chromosome , becoming prophages . Alternatively , circularized phage genomes are maintained as episomes . Once integrated , the now so-called prophage is stable and replicates passively with its host genome . This situation can continue as long as outside conditions do not become threatening for the host , and therefore for the virus . Prophages are indeed able to detect many stressful signals , such as DNA damage , excessive heat or pressure [1]–[3] . By “listening” and hijacking the host's response to various stresses , prophages behave like perfect stress biosensors . Once the prophage is induced , the process of lysogeny escape is engaged , and the phage enters a lytic mode of development [1] . A crucial event in this process is the excision of the prophage from the host's chromosome . Replication of the viral genome follows , as well as the synthesis and the assembly of the virion proteins . Thus , excisive recombination is a highly regulated process that relies on two different levels of regulation: ( i ) protein activity , through the control of directionality by a recombination directionality factor ( RDF ) , and ( ii ) protein synthesis via the coordinated expression of the integrase and RDF genes . Temperate bacteriophages use site-specific recombination to integrate into and excise their genomes out of the host genomes . Integration consists of a strand exchange between the recombination region attP on the phage genome and attB on the bacterial chromosome leading to the formation of the recombined halves attL and attR at the junctions between the bacterial chromosome and the integrated phage genome ( Figure 1 ) . Lambda phage integrase has been extensively studied for its role in site-specific recombination and is essential for lysogeny establishment as well as for the transition to productive lytic development ( reviewed in [4] , [5] ) . The Int tyrosine recombinase catalyzes integrative and excisive recombination [6] , [7] . Xis acts as a recombination directionality factor ( RDF ) as it bears no catalytic activity but rather directs the Int-driven reaction toward excision [8] . Xis plays an architectural role in the formation of the excisive intasome by binding and bending DNA , and prevents reintegration of the excised phage genome [9]–[11] . Precise stoechiometry of Int and Xis proteins is required for the correct assembly of the intasome nucleoprotein complex [12] . Since the organization of the protein binding sites of the att regions is not conserved , this suggests that the intasome architecture may vary according to the number and orientation of the recombination protein binding sites [13] . The phage-encoded integrase is a hetero-bivalent DNA binding protein in which the N- and C-terminal domains bind to different DNA substrates . The C-terminal domain , where the catalytic activity takes place , binds to and recombines the identical core-type sequences present in attP and attB , or in attL and attR , depending on the direction of the reaction considered [14]–[16] . The N-terminal domain binds to arm-type sequences [17] , and this binding allows the assembly of the intasome , the nucleoprotein complex for site-specific recombination . Host-encoded proteins are also involved in this process , including IHF and Fis that bind and bend DNA in order to assist intasome formation [9] , [18]–[20] . Recombination occurs through pair-wise exchange of four DNA strands between two att substrates . A four-way Holliday junction is formed upon the exchange of one pair of strands and then resolved after the DNA cleavage activity is switched from one pair of strands to another [21]–[24] . In all temperate phages , site-specific recombination events are believed to be identical; however , the organization of the att regions varies from one family of phages to another according to the number and orientation of the recombination protein binding sites . This suggests that the assembly and final composition of the intasome might follow different paths to eventually end with the same recombination reaction . The KplE1 prophage ( also named CPS-53 ) is a defective prophage integrated into the argW tRNA gene in E . coli K12 ( Figure 1 ) . The prophage's remaining genome ( 10 . 2 kb ) contains 16 open reading frames ( ORF ) bordered by a duplicated core sequence of 16 nucleotides ( CTGCAGGGGACACCAT ) . None of these ORFs seems to encode a repressor consistent with the finding that KplE1 is not SOS-inducible ( M . Ansaldi , unpublished observation ) . Despite the small remnant genome , the KplE1 prophage can be excised in vivo [13] , [25] . The KplE1 recombination module has been analyzed , and indeed it contains all the elements required for site-specific recombination to occur , including RDF and integrase genes as well as the attL and attR recombination regions [26] . This recombination module is highly conserved in several enterobacteria phage genomes such as CUS-3 and HK620 that infect E . coli strains K1 RS218 and TD2158 , respectively , and Sf6 , which infects Shigella flexneri , as well as in prophages present in E . coli strains APEC-O1 and UTI89 [27]–[32] . One advantage of studying the KplE1 prophage is that we can dissect the excisive recombination and its regulation in vivo independently of prophage induction since the CI regulator module is missing in KplE1 . Directionality of the site-specific recombination has been studied using KplE1 DNA substrates as well as HK620 substrates and requires the RDF protein TorI to direct the recombination reaction towards excision [26] . One prominent feature of the KplE1 recombination is the orientation of the intS gene relative to the attL region ( Figure 1 ) . Indeed , the intS gene is transcribed from a dedicated promoter that overlaps with the attL region . In λ , int gene expression depends on the activity of two promoters PI and PL [1] , [5] . While lysogeny is established , int expression relies on the PI promoter located in the xis gene and allows transcription of int independently of xis . Therefore , this promoter is used to establish lysogeny and ensures that more Int than Xis is being made [33] . During the escape from lysogeny , xis and int are co-transcribed as a consequence of PL promoter activation and N antitermination ( Figure 1 ) . The differential expression of Int by these two promoters depends upon a site ( sib ) located distal to the int gene . Thus , lower amounts of Int are made , and Xis production is not affected by this element [34] . Based on the localization and orientation of the intS promoter that overlaps the attL recombination region ( Figure 1 ) , we performed preliminary experiments that led us to conclude that the intS gene is negatively autoregulated and poorly expressed during the exponential growth phase [26] . In this study , we further investigate the regulation of the intS gene in relation to the recombination efficiency . We provide in silico evidence that a majority of integrase genes associated with tRNA inserted prophages are predicted to negatively autoregulate . This prediction was subsequently confirmed in vivo with several examples . As a consequence , the integrase gene appears constantly expressed at a low level in KplE1 , and the control of excisive recombination seems to rely only on the RDF expression rather than on a coordinate expression of the integrase and RDF genes .
Previous work described the PintS promoter based on sequence analysis of the region upstream from the ATG starting codon [26] . This allowed the identification of putative −10 and −35 sequences close to the consensus sequences recognized by the σ70-RNA polymerase holoenzyme ( TAaAAa and TTGACA , respectively ) ( Figure 2C ) . To show that the RNA polymerase actually utilizes this promoter to start intS transcription , we experimentally determined the intS transcription start site . Primer extension analysis was performed using total RNAs extracted from a wild-type as well as an intS strain , annealed with a labeled primer hybridizing downstream from the intS ATG ( see Materials and Methods for details ) . In the presence of IntS ( Figure 2A , lane 1 ) , extension products were scarcely apparent . However , in the intS background ( Figure 2A , lane 2 ) we observed two main extension products , indicating that transcription started at T and A residues at positions 2464536 and 2464537 on the E . coli chromosome , respectively . These transcription start sites are correctly located relative to the σ70-RNA polymerase holoenzyme binding sites , and the A at position 2464537 is perfectly positioned relative to the −10 box [35] . This latter transcription start site was also detected in a genome-scale analysis of transcription in E . coli [36] . Altogether , these experiments confirmed the previous localization of the intS promoter and the downregulation of the intS gene by its own product . The intS promoter , due to its location , obviously overlaps with the attL recombination region , and thus overlaps with IntS and TorI binding sites as previously characterized [26] ( Figure 2C ) . In that study , we showed that the intS transcript originating at the chromosomal Pints promoter was five-fold more abundant in an intS background than in a wild-type strain . To study the influence of each protein binding site on PintS regulation in vivo , an accurate method was needed to quantify gene expression that would allow easy mutagenesis of the protein binding sites . We chose to use a gfp fusion-based vector ( pUA66 ) that contains a sc101 replication origin , which leads to a low copy number ( 3 to 4 copies in the logarithmic growth phase ) of the plasmid in vivo to avoid titration of the regulators [37] . The experiment was calibrated by cloning the entire attL region ( positions 2464344 to 2464630 on the E . coli chromosome ) in the pUA66 vector in order to measure pattL-gfp expression in various genetic backgrounds . Primer extension was used to control that transcription initiation occurred at the same site in this construct rather than in the chromosome ( Figure 2A , lane 3 ) . Indeed , the transcription start sites proved to be identical to those characterized on the chromosome when expressing the PintS promoter from a plasmid . Using this construct , we observed an increased transcription level of the PintS promoter compared to the chromosomal expression . This was likely due to a combination of two effects: the plasmid copy number and the fact that total RNAs were extracted from the LCB1019 strain that lacks the entire KplE1 prophage , and therefore the intS gene . Another explanation could be that this increase in transcription is linked to an increase in translation of the fusion . However , this is probably not the case because although integrase genes often contain rare codons that may slow down translation , a particular rare codon ( AGA ) is also present in the gfp gene . We measured the fusion expression with two different methods: direct fluorescence measurement ( Figure 3 ) , which gave a whole population measurement , and microscopic counting ( Figure S1 ) , which estimated the homogeneity of the fluorescent population . As indicated in Figure 3B , the attL-gfp wild-type fusion was expressed at a high level in the absence of IntS ( 6368±914 Units ) and was repressed in the presence of IntS ( 1270±208 Units ) , leading to a repression ratio of ∼5 when the control ratio of placZ-gfp expression was close to 1 in the same conditions . This ratio of ∼5 is in complete agreement with the values we obtained by measuring intS expression from the chromosomal gene with quantitative RT-PCR [26] , indicating that the fusion expression from several copies did not modify the regulatory ratio . Expression of the fusion was homogenous under all conditions , and the most resolved peaks were observed for cells producing TorI or IntS and therefore emitting little fluorescence ( Figure S1 ) . Thus , the results measured in the whole population ( Figure 3 ) reflect homogenous expression of the fusion . Looking at the recombination protein binding sites identified on attL ( Figure 2C ) , it was obvious that some of the TorI RDF binding sites were also near the −35 sequence . We thus looked at a possible effect of TorI on intS expression . As indicated in Figure 3B , overexpression of TorI in an intS background led to a strong decrease in expression of the pattL-gfp ( wt ) fusion ( compare 6368±914 with 1201±370 units ) . Taken together , these results show that intS expression is under the negative control of both TorI and IntS . The attL region contains five TorI binding sites organized in two blocks ( Figure 2C , red symbols ) . The first one ( I1;2 ) , encompasses sites I1 and I2 that are separated by 12 nucleotides ( positions 2464409 to 2464436 ) . The second block is composed of three binding sites ( I3;4;5 ) separated by 2 nucleotides ( positions 2464472 to 2464499 ) . We mutated each site by changing the sequences GTTCG , GATCG , GTCCG into CAAGC . When both sites of block I1;2 were mutated we did not observe any effect on the TorI mediated downregulation of intS ( pattL-gfp ( I1*;2* ) ) with a repression ratio of 5 . 4 ( Figure 3 ) . In contrast when the sites of the second block were changed , pattL-gfp ( I3*;4*;5* ) , TorI was no longer able to repress the expression of the fusion , meaning that at least one of these sites was important for repression . We thus measured the effect of each site independently . If the mutation of site I3 had little effect on the repression ratio ( 4 . 7 ) , the mutations of sites I4 and I5 led to expression of the fusion independent of the presence of TorI ( repression ratios of 1 . 0 and 1 . 7 , respectively ) . These two sites are the closest to the −35 sequence and are therefore appropriate candidates for mediating TorI repressor activity . We observed increased basal expression of the fusion ( from 6 , 400 to 15 , 000–19 , 000 units ) when the I5 site was mutated . This effect is probably due to the change in the nucleotides adjacent to the −35 sequence that results in a promoter-up phenotype . We then studied the implication of the arm-type binding sites ( P' sites , blue symbols in Figure 2B and 2C ) of the integrase . For the P' sites the conserved motif TAAA present in all P' sites was changed into its complement ATTT . Interestingly , none of the individual mutations led to derepression of the fusion; indeed , in all cases ( except for the P'1* , see below ) , the repression ratio ranged from 4 . 0 to 5 . 2 ( Figure 3B and 3C ) . The P'1 site's influence was more difficult to study since it overlapped with the −10 sequence ( Figure 2C ) . Thus , any mutation of the conserved motif led to an inactive promoter whose measured fluorescence did not exceed that of the promoter-less fusion ( Figure 3 , compare pUA66-gfp with pattL-gfp ( P'1* ) ) . Additional constructs were made to avoid this effect on the promoter activity; however , any change we made that altered IntS binding also affected promoter activity ( Figure 3 , constructs pattL-gfp ( P'1-A and B ) ) , and in the case the latter was not affected ( pattL-gfp ( P'1-C ) ) , neither was the down regulation of intS . In a control experiment , we mutated the core site , which is the binding site for the catalytic domain of the integrase , and this construct showed an unaltered repression phenotype ( Figure 3 ) as well as IntS binding similar to that observed with the wild-type sequence ( data not shown and [13] ) . Altogether , these data demonstrate that both TorI and IntS negatively regulates the intS gene in vivo and point to the TorI and IntS sites located near the −35 and −10 sequences as being responsible for the downregulation of intS gene expression . These results also show that the intS gene is tightly regulated and is thus expressed at a low level under all tested growth conditions . One could ask about the “raison d'être” of this atypical integrase gene regulation compared with the lambda int gene . For that purpose , we measured the efficiency of the excisive recombination reaction in vitro as a function of the integrase concentration . Briefly , 32 nM of attL and attR linear substrates were incubated at 37°C for 1 h in the presence of constant concentrations of TorI and IHF ( 1 . 6 µM and 0 . 25 µM , respectively ) and increasing concentrations of IntS ( 0 . 02 to 6 . 7 µM ) . The attP product was quantified by Q-PCR and the efficiency of the reaction was calculated as the percentage of substrates transformed into products . As the concentration of IntS increased , the efficiency of the reaction increased until a maximum level of ∼80% was achieved for an IntS concentration around 1 µM ( Figure 4 ) . However , when the IntS concentration exceeded 1 . 2 µM , we rapidly observed an inhibitory effect of IntS on the excisive reaction . Subsequently , the concentration range for which the efficiency of the reaction reached more than 50% was very narrow ( 0 . 8 µM up to 1 . 2 µM ) . These results show that to obtain the maximum efficiency in excisive recombination a precise integrase concentration is required . The effect of IntS overloading was then analyzed in vivo . Strain LCB6005 contains a Km resistance cassette in the tail fiber encoding gene ( tfaS ) of the KplE1 prophage , thus allowing an in vivo excision assay to be performed without any effect on the site-specific recombination process . This strain was transformed with the pJFi plasmid that contains the torI gene under the control of an IPTG inducible promoter as well as with the pBAD33 vector containing or not the intS gene under the control of an arabinose inducible promoter ( pBADintS ) . Colonies were counted after the different cultures induced with IPTG were plated on LB medium containing ampicillin or kanamycin ( see the Material and Methods section ) . ApR colonies are representative of the total number of cells since all contain the pJFi plasmid ( ApR ) , whereas KmR colonies originate from cells that have kept the tfaS::kan marker , and thus the KplE1 prophage . As shown before [25] , expressing torI from a multicopy plasmid ( pJFi ) is sufficient to promote excisive recombination . Indeed , in the strain containing the low copy vector alone ( pBAD33 ) , the maximal level of excision was achieved in the presence of TorI as revealed by a high ApR/KmR ratio ( Figure 5A ) , and the addition of the arabinose inducer did not impede the reaction's efficiency . However , in the presence of the pBintS plasmid , even without adding the arabinose inducer , we observed dramatically decreased recombination activity ( Figure 5A , compare bars 1 and 3 ) . It is striking that , even at a concentration of integrase that could not be detected on a Western blot ( Figure 5B , lane 3 ) , i . e . , in the absence of an inducer , the efficiency of the reaction underwent a 50-fold decrease . We explain this effect by the leakage of the pBAD promoter in the absence of glucose . Indeed , this promoter is induced in the presence of arabinose and repressed in the presence of glucose [38] . Since we do not use glucose in the medium , the pBAD promoter is not repressed , and some integrase is being made , although not sufficiently to immunodetect it . We therefore consider the empty vector as the actual negative control . Adding arabinose to the medium , which led to overproduction of IntS ( Figure 5B , lanes 4 to 8 ) , amplified this negative effect on the in vivo excision reaction . As a result , the in vivo recombination efficiency was negatively correlated with the increasing integrase concentration , thus confirming the results we obtained in vitro . To address the general relevance of the negative autoregulation of the intS gene , a large-scale in silico analysis of tRNA-associated integrase genes was performed on the complete prokaryotic genomes available at that time . The in silico outline is described in the “Materials and Methods” section . Experimentally well-characterized integrases such as λInt and KplE1IntS contain at least one of the three functional domains , Phage_integrase , Phage_integ_N , and Phage_integr_N , referred to as PF00589 , PF09003 and PF02899 in the Pfam database , respectively . By using these functional domains as queries , we detected 8368 protein homologs within 1014 complete prokaryotic genomes , and 1273 of the corresponding integrase genes ( 15% of the total ) are adjacent to a tRNA gene . These couples of tRNA-integrase genes ( called InTr shape ) constitute the primary data set used in this study ( Table S1 ) . InTr shapes were classified according to their gene coding orientation , leading to four different types of InTr shapes ( Figure 6A ) : STI ( Same orientation and T precedes I ) , SIT ( Same orientation and I precedes T ) , OC ( Opposite and Convergent orientation ) and OD ( Opposite and Divergent orientation ) . We then analyzed the distribution of the InTr copy-number per organism ( Figure 6B ) as well as the distribution of InTr shapes over the prokaryotic phylum ( Figure 6C ) . A detailed analysis of these data is available in the Text S1 . Overall analysis shows that the majority of the InTr shapes exhibits STI and OC shapes with 736 and 438 representatives , respectively . The other two classes ( SIT and OD ) occur relatively rarely ( less than 8% in total ) in the analyzed genomes . Therefore , the high occurrence of STI and OC shapes within the prokaryotes may highlight the functional importance of these shapes in microbial organisms . To study a possible correlation between the prevalence of InTr shapes and the autoregulation of the integrase genes as demonstrated for the intS gene , the number of putative autoregulated integrase genes was determined . Based on our experimental model , we proposed that STI and OD shapes should be subjected to autoregulation , since in these cases the integrase gene promoter overlaps with the recombination region , whereas SIT and OC shapes should show integrase gene expression independent of the integrase protein . Our in silico results indicated that InTr shapes containing Asn , Cys , Gln , Gly , Leu , Phe , SelC , and Ser tRNA genes were mainly predicted to autoregulated ( Table S2 ) . In contrast , the opposite conclusion can be drawn for InTr shapes containing Ile , Lys and Tyr tRNA genes , which is consistent with the observation that prophages are preferentially inserted in poorly expressed tRNA genes , probably to avoid a deleterious effect on cell fitness ( [39]–[41] . A detailed analysis of the distribution of InTr shapes with respect to tRNAs in prokaryotic genomes is available in Figure S2 and Table S3 . Out of the 1273 InTr shapes analyzed , 61 . 5% were detected as potentially autoregulated , most encoded within the Proteobacteria , Cyanobacteria , Bacteroidetes and Crenarcheota genomes ( Figure 6C and Table S2 ) . Thus , a situation that has rarely been described and studied in the literature is actually predominant in the sequenced prokaryotic genomes . We next addressed whether a relationship exists between the length of the intergenic region ( IR , Figure 6A ) and the fact that an integrase gene is predicted to be autoregulated . Therefore , the IR length was determined for each InTr shape , and the distribution of the obtained values was analyzed as a function of autoregulated and non-autoregulated InTr shapes ( Figure 7 ) . The lower values of the IR lengths are statistically associated with predicted non-autoregulated InTr shapes as the 95% confidence intervals of the mean IR length values are [157 . 5–158 . 4] for non-autoregulated InTr and [208 . 3–227 . 9] for predicted autoregulated InTr . These results clearly indicate that autoregulated InTr shapes are linked to large IRs . Our prediction is that autoregulation of the integrase mostly correlates with STI and OD shapes , and therefore the IR should be large enough to contain an entire attL region . As mentioned above , the average distance observed for predicted autoregulated InTr shapes [208 . 3–227 . 9] is perfectly compatible with the presence of an average attL region of 80–170 nucleotides . Biological validation of the autoregulation of integrase genes involved in STI and OD InTr shapes . To validate the in silico predictions , we chose to study the expression of several integrase promoters from E . coli strains K12 MG1655 and O157:H7 EDL933 . The promoters of the integrase genes were cloned into the pUA66 vector upstream of the gfp gene and the cognate integrase coding sequences were cloned into the pJF119EH vector ( see plasmid list in Table 1 ) . Regarding the InTr shapes , in addition to the well-characterized STI argW-intS , we studied 2 STI shape argW-intC ( the argW-intS homologous shape in EDL933 ) and selC-intL , 2 OC shapes argU-intD ( MG1655 ) and thrW-intH ( EDL933 ) and 1 OD ptwF-intF ( MG1655 ) . Of these integrase genes , 3 are predicted to be autoregulated ( intC , intF , and intL ) , and 2 should not exhibit autoregulation ( intD and intH ) . To avoid the influence of the chromosomal copies of MG1655 integrase genes , we transformed both kinds of plasmids ( the empty vector and the integrase encoding vector ) in the appropriate deletion mutant , and when applicable , we used the MG1655 mutant for the EDL933 equivalent . As indicated in Figure 8 , none of the OC shape associated integrase genes showed self-regulation , and the STI and OD shape integrase genes were negatively autoregulated . However , different regulation ratios were observed depending on the integrase gene considered . Interestingly , the pIntS-gfp fusion was repressed almost 15 times during the exponential growth phase ( time point ∼2 h ) whereas a repression ratio of 6 was measured during the stationary phase ( ∼4 . 5 h ) , which is consistent with the data shown in Figure 2 and Figure 3 . A similar expression pattern was obtained with the pPintC-gfp fusion for which the repression ratios were higher than for pPintS ( 28 in the exponential phase and 10 in the stationary phase ) . A high regulatory ratio was observed with the pPintF-gfp fusion whose expression was decreased around 23 times in the presence of the pIntF plasmid in exponential as well as in stationary growth phases , without any induction of the ptac promoter , indicating that the leak of the promoter allowed sufficient IntF production to produce a negative effect on the fusion expression . In contrast , the pPintL-gfp fusion was only down-regulated by a factor of 4 in the exponential growth phase , and this occurred in the presence of 0 . 1 mM of IPTG . Thus , the level of IntL required to lead to a negative effect on the fusion expression is probably higher than that necessary for the IntF integrase . One possible explanation is that integrase genes from E . coli MG1655 interfere with the downregulation of EDL933 genes . This hypothesis is strengthened by the fact that the regulatory ratios measured with the pPintC fusion were higher in an intS background than in a WT MG1655 background ( data not shown ) . Together , these results supported the in silico prediction that STI and OD shape associated integrase genes should be negatively autoregulated . However , this prediction could be associated with promoter and recombination region sequence analysis to ensure that the two overlap .
The regulatory switch leading to the controlled expression of the integrase and RDF proteins that allows the excision of the lambda prophage and therefore permits productive growth to resume has long been the paradigm for all temperate phages [1] , [5] . In this study , we show that the particular organization we identified for the KplE1 attL recombination region and related ( pro ) phages is widespread among the tRNA inserted prophages . The fact that the attL region overlaps the integrase promoter has several consequences: ( i ) the integrase gene is likely down-regulated by itself and the RDF , as long as the recombination protein and the RNA polymerase binding sites overlap sufficiently , ( ii ) the transcription of the integrase and RDF genes are uncoupled , and ( iii ) the regulatory switch that permits prophage excision relies on RDF gene expression . Full understanding of prophage excision control will require focusing on the expression of the RDF genes that are uncoupled to the integrase gene transcription .
Bacterial strains and plasmids are listed in Table 1 . Strains were grown in LB medium and , when necessary , ampicillin ( 50 µg mL−1 ) , chloramphenicol ( 25 µg mL−1 ) , kanamycin ( 25 µg mL−1 ) or IPTG ( 0 . 1–1 mM ) were added . Strains LCB6005 , LCB6006 , LCB6035 and LCB6036 are derivatives of ENZ1734 ( MG1655 ΔlacIZ ) [55] obtained by P1 transduction of the tfaS::KanR ( JW5383 ) , intS::KanR ( JW2345 ) , intD::KanR ( JW0525 ) and intF::KanR ( JW0275 ) markers [56] , respectively , into ENZ1734 . The kan gene was then removed from strains LCB6006 , LCB6035 and LCB6036 by using the pCP20 plasmid [57] to generate strains LCB6007 ( intS , KanS ) , LCB6037 ( intD , KanS ) , LCB6038 ( intF , KanS ) . Strains are described in Table 1 . To construct plasmid pBintS , the intS coding sequence was PCR-amplified using MG1655 chromosomal DNA as a template with appropriate primers . After enzymatic hydrolysis , the PCR product was cloned into the KpnI/HindIII sites of the pBAD33 vector [38] . Plasmid pattL-gfp was constructed by the insertion of the attL region ( 220 bp , Figure 2C ) into the XhoI and BamHI sites of the pUA66 vector [37] . A similar procedure was used to clone the promoter regions of intD , intH , intL intF into the pUA66 vector . Positions of the cloned sequences are indicated in Table 1 , and primer sequences are available upon request from the authors . The sequence accuracy of the cloned inserts was checked by sequencing . Total RNAs extracted from strains MC4100 and LCB1024 ( ΔintS ) , and strain LCB1019 ( ΔKplE1 ) containing pattL-gfp were hybridized with a primer complementary to the positions +40 to +64 relative to the ATG of intS ( attL-ter ) . attL-ter was 32P labeled by using [γ32P]ATP and T4 polynucleotide kinase ( Biolabs ) . A total of 12 µg of ARNs and 4 ng of labeled primer were incubated together with 200 units of Superscript III reverse transcriptase ( Invitrogen ) for 50 minutes at 50°C , followed by 10 minutes at 70°C to inactivate the enzyme . The sequencing ladder was PCR amplified with the same labeled primer and 5′ primer hybridizing to positions −196 to −173 relative to the ATG of intS ( attL-Kpn ) . The sequencing reaction was performed using the Thermo Sequenase Cycle Sequencing Kit ( USB Corporation ) . Extension and sequencing products were separated onto a 6 M urea 8% acrylamide ( 19∶1 ) gel . Mutations in the recombination protein binding sites were generated by an overlapping PCR procedure [58] . Mutated primers were used to amplify the protein binding sites whereas the wild-type primers attL-pro-XhoI and attL-ter-BamHI delimit the attL region . After enzymatic hydrolysis , mutated attL were cloned into pUA66 . Mutations in the IntS and TorI binding site are summarized in Figure 2C . All primer sequences used for mutagenesis are available upon request . IntS , TorI and IHF proteins were overproduced and purified near homogeneity as described [26] , [59] , [60] . All proteins were dialyzed in 40 mM Tris-HCl buffer ( pH 7 . 6 ) containing 50 mM KCl and 10% glycerol . Denaturing polyacrylamide gel electrophoresis ( SDS-PAGE ) was used to estimate the protein purity , and the Lowry method was used to estimate protein concentrations . Strain LCB6005 ( KanR gene inserted in the tfaS gene of KplE1 ) carrying plasmids pJFi and pBAD33 ( control ) or pJFi and pBintS were grown in LB medium supplemented by increasing amounts of arabinose as indicated in Figure 5 legend . When the A600 reached 0 . 5 units ( 0 . 5×109 cells mL−1 ) , IPTG ( 1 mM ) was added and the culture resumed for 2 h at 37°C under agitation . Culture dilutions were prepared and plated onto rich medium containing either ampicillin ( pJFi ) or kanamycin ( tfaS::kan ) . Numeration of the colonies plated on both antibiotics was performed and the ratio of ampicillin-resistant/kanamycin-resistant ( ApR/KnR ) colonies was calculated . This value is close to one when the excision rate is low and the tfaS::kan marker is present , and increases when excision efficiency increases and the cells no longer contain the KplE1 prophage . The values represent the average of at least three independent determinations . The IntS relative amount in crude extracts was analyzed after 12% SDS-PAGE with Western blot using a polyclonal IntS antiserum . Purified IHF , IntS and TorI were used in all experiments . All reaction mixtures ( 25 µl ) included 32 nM of linear attL ( attL-SpeI/attL-KpnI primers ) and attR ( attR-XbaI/attR-IHF2 primers ) in buffer containing 30 mM Tris-HCl ( pH 7 . 6 ) , 10 mM spermidine , 5 mM EDTA , 1 mg . mL−1 bovine serum albumin , 34 mM KCl and 5% glycerol . IHF ( 0 . 25 µM ) and TorI ( 1 . 6 µM ) were added in all samples in the presence of a range of IntS concentrations ( 0 . 02 to 8 µM ) . The reactions were carried out in optimized conditions: 37°C for 1 h . The best efficiency was obtained for IntS concentrations ranging from 0 . 8 to 1 . 2 µM , leading to an IHF:IntS:TorI protein ratio of 1∶4∶6 . The abundance of attP formed during in vitro excision assays was quantified by real-time PCR , using a known concentration of PCR amplified attP as a reference standard . The real-time PCR quantifications were performed with an Eppendorf Mastercycler ep realplex instrument and the SYBR Premix Ex Taq ( TaKaRa ) according to the manufacturer specifications . Serial dilutions of each in vitro reaction were mixed with 1 . 5 µM of primers and 6 µL of master mix in a 14 µL final volume . The primer pair used to quantify attP was attR-IHF2/attL-SpeI . PCR parameters were as follows: one cycle at 95°C for 2 min followed by 40 cycles at 95°C for 5 s , 55°C for 15 s and 72°C for 10 s . Excision efficiency was calculated as the percentage of the initial substrate ( 32 nM ) transformed into product . GFP fluorescence was measured on whole cells after an overnight aerobic growth at 37°C in LB medium supplemented by IPTG ( 0 . 1–1 mM ) for TorI and/or integrase induction ( Figure 3 ) . The pJF119EH empty vector was used as a negative control and to ensure that the growth conditions ( presence of ampicillin ) were identical for all strains . After centrifugation , bacteria were washed , resuspended and diluted in 0 . 25X M9 medium . Cells ( 150 µL ) were loaded on an Optilux black/clear Bottom Microtest 96-well assay plate ( Falcon ) . Alternatively , fluorescence intensity was measured on bacterial cultures over time . Precultures of the various strains were diluted in fresh LB medium containing the appropriate antibiotics and IPTG ( 0 . 1 mM ) when indicated . Each strain was assayed in quadruplet . The incubation protocol included an initial 5-min shake ( double orbital , 1 . 5 mm diameter , normal speed ) , followed by 85 cycles consisting of the following actions: a 1-sec measurement ( see below ) , a 6-min shake and a 1-min standing . The time course was performed at 37°C for approximately 10 h . A600 and fluorescence measurements were performed using the Infinit M200 instrument ( Tecan ) and the Tecan i-control 1 . 3 application ( 488 nm excitation wavelength , 521 nm emission wavelength , 160 gain , 20 µs integration time and 25 reads per sample ) . The value of blank ( 0 . 25X M9 or LB ) was withdrawn and normalized fluorescence intensities ( emission at 521 nm/A600 ) were calculated . The values represent the averages of at least four independent measurements . Microscopic analysis was performed using an automated and inverted epifluorescence microscope TE2000-E-PFS ( Nikon , France ) and adequate filters ( excitation 480±15 nm , emission 535±20 nm ) . Images were recorded with a CoolSNAP HQ 2 ( Roper Scientific , Roper Scientific SARL , France ) and a 40x/0 . 75 DLL “Plan-Apochromat” or a 100x/1 . 4 DLL objective; image analysis was conducted with MetaMorph 7 . 5 software ( Molecular Devices ) . For each cell preparation , a total of 25 images were taken randomly on different optical fields , and the average intensity of each cell was calculated ( Figure S1 ) . The complete genomes of 1014 prokaryotic ( 946 bacterial and 68 archaeal ) organisms available in December 2009 were downloaded from the NCBI ftp site ( ftp:/ftp . ncbi . nih . gov/genomes/Bacteria/ ) and constitute the primary data source . To identify integrase promoters overlapping the integration site , the analysis was restricted to prophage insertion targeted to tRNA sites . The HMMER-3 package [61] and self-written Perl scripts were then used to search for protein integrase homologs ( with phage λ int and E . coli intS as reference seed proteins ) in the complete genomes . The presence of one of these functional domains ( from Pfam 24 . 0 [62] ) , Phage_integrase ( PF00589 ) , Phage_integ_N ( PF09003 ) or Phage_integr_N ( PF02899 ) , was a requisite . Alignments with a score higher than the Pfam gathering thresholds were considered significant . Note that homologs with protein sizes lower than 140 amino acids ( corresponding to 80% of the Phage_integrase profile length ) were removed from the data . The obtained sequences were subsequently analyzed with the same software in order to locate additional known functional domains . In-house Perl scripts were used to define the domain organization . The search for tRNA genes , located in the region between the integrase gene and the downstream/upstream neighboring gene was performed by using the tRNAscan-SE program [63] . Finally , protein integrase homologs were filtered by the presence of an adjacent tRNA gene ( downstream or upstream of the integrase gene ) , leading to the final set of integrase homologs used in this study . We then computed the IR length as the distance in nucleotides between a given integrase gene and the immediately adjacent tRNA gene .
|
Temperate bacteriophages are widespread bacterial viruses that have the ability to replicate passively in their hosts as long as no stressful conditions are encountered , a process called lysogeny . Prophage-encoded genes may benefit the host in several ways such as providing resistance to antibiotics , increased pathogenicity , or increased fitness . Most temperate phages insert their genome into the host's chromosome by site-specific recombination . After prophage induction , usually under stressful conditions , the excisive recombination constitutes a key step toward productive phage development . In this paper , we study the regulation of integrase genes that encode the enzyme required for integrative as well as excisive recombination . We noticed that for prophages inserted in or near tRNA genes the orientation of the integrase gene relative to the tRNA is crucial for its regulation .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"microbiology/microbial",
"growth",
"and",
"development",
"microbiology/microbial",
"evolution",
"and",
"genomics",
"microbiology"
] |
2010
|
Tight Regulation of the intS Gene of the KplE1 Prophage: A New Paradigm for Integrase Gene Regulation
|
Fascioliasis is a pathogenic disease transmitted by lymnaeid snails and recently emerging in humans , in part due to effects of climate changes , anthropogenic environment modifications , import/export and movements of livestock . South America is the continent presenting more human fascioliasis hyperendemic areas and the highest prevalences and intensities known . These scenarios appear mainly linked to altitude areas in Andean countries , whereas lowland areas of non-Andean countries , such as Uruguay , only show sporadic human cases or outbreaks . A study including DNA marker sequencing of fasciolids and lymnaeids , an experimental study of the life cycle in Uruguay , and a review of human fascioliasis in Uruguay , are performed . The characterization of Fasciola hepatica from cattle and horses of Uruguay included the complete sequences of the ribosomal DNA ITS-2 and ITS-1 and mitochondrial DNA cox1 and nad1 . ITS-2 , ITS-1 , partial cox1 and rDNA 16S gene of mtDNA were used for lymnaeids . Results indicated that vectors belong to Lymnaea neotropica instead of to Lymnaea viator , as always reported from Uruguay . The life cycle and transmission features of F . hepatica by L . neotropica of Uruguay were studied under standardized experimental conditions to enable a comparison with the transmission capacity of F . hepatica by Galba truncatula at very high altitude in Bolivia . On this baseline , we reviewed the 95 human fascioliasis cases reported in Uruguay and analyzed the risk of human infection in front of future climate change estimations . The correlation of fasciolid and lymnaeid haplotypes with historical data on the introduction and spread of livestock into Uruguay allowed to understand the molecular diversity detected . Although Uruguayan L . neotropica is a highly efficient vector , its transmission capacity is markedly lower than that of Bolivian G . truncatula . This allows to understand the transmission and epidemiological differences between Andean highlands and non-Andean lowlands in South America . Despite rainfall increase predictions for Uruguay , nothing suggests a trend towards a worrying human infection scenario as in Andean areas .
The impact of climate change and global change is putting trematodiases in one of the main focuses of infectious disease actuality [1–4] . Among the food-borne trematodiases emphasized in the recent WHO Roadmap for neglected tropical diseases 2020 [5] , fascioliasis depicts a specific place due to its worldwide distribution , emergence , and estimated 17 million people infected throughout [6] . The climate change impact on fascioliasis is linked to the high dependence of both fasciolid larval stages and their freshwater lymnaeid snail vectors on climatic and environmental characteristics [2 , 7 , 8] . Additionally , fascioliasis emergence appears also related to global change aspects , such as import/export and management of livestock [6] , anthropogenic modifications of the environment [9] , travelling [10] and changing human diet traditions [11] . Fascioliasis morbidity in humans has been highlighted by the World Health Organization [12] . The acute and long-term chronic phases of this disease show high pathogenicity and immunosuppressive capacity [13–17] . Aspects adding concern are the clinical complexity and severity of symptoms and syndromes , important sequelae and even death [18] , pronounced diagnosis difficulties [19] and treatment problems [20] . South America stands out due to the high human prevalences and intensities reported in Andean countries as Bolivia [21–25] , Peru [26–28] and Chile [29 , 30] , and the cases from Ecuador [31] , Colombia [32] , and Venezuela [33] . However , in the non-Andean , lowland countries , human reports only concern sporadic and isolated cases , such as in Brazil [34] and Uruguay [35] . Uruguay has a wide farming and agricultural sector , with 70% of the export trade corresponding to livestock products and subproducts . Fasciola hepatica , locally known as “saguaypé” [36] , is distributed throughout the large flat lowlands of the whole country ( Fig 1 ) . Cattle and sheep are the most affected , which is related to mixed grazing [37] , with high prevalences [37–41] and great impact and economic losses [42] . Horses , sharing the same pastures with cattle and sheep , are the third affected species [43 , 44] . The liver fluke also infects wild rodents , including the capybara Hydrochoerus hydrochaeris ( Caviidae ) [45] and the nutria or river rat Myocastor coypus ( Myocastoridae ) [46] . Fasciolid eggs have also been found in the wild Pampas deer Ozotoceros bezoarticus ( Cervidae ) [47] . There is a direct effect of altitude on fascioliasis transmission [28] . High altitude lymnaeid vectors produce a higher number of metacercariae throughout a longer cercarial shedding period [48] . This higher transmission capacity is related to the longer life span and post-infection survival of the vector in such altitudes [48 , 49] . The high fascioliasis transmission in human fascioliasis endemic areas in Andean highlands appears linked to the Galba/Fossaria group species Galba truncatula , a very efficient vector of European origin introduced 500 years ago with the livestock transported by the Spanish "conquistadores" [6 , 48] . Recent reports indicate that other Galba/Fossaria species may also be involved in human fascioliasis endemic areas , such as Lymnaea neotropica in Peruvian highlands [50] , and in extreme aridity-dryness habitats in Argentina [51] together with L . viator ( = L . viatrix—for nomenclature see [30] ) . The later species is also involved in other places of Argentina [52] , whereas it was confused with G . truncatula in Chile [30] due to the difficulties in phenotypically differentiating between species of Galba/Fossaria [53] . In Uruguay , only two lymnaeid species have been reported [54: L . viator [55 , 56] and Pseudosuccinea columella [57] . Pseudosuccinea columella has been detected in 14 of the 19 departments of the country [36] , even naturally infected [58] . Although known as an important lymnaeid for the disease transmission to livestock [32] , veterinary responsibles have assigned only secondary importance to this species in Uruguay . This is because of its sporadic natural infection linked to its different ecology and pronouncedly lower cercarial shedding when compared to L . viator ( a maximum of 10 cercariae/snail in P . columella; between 100 and 200 cercariae/snail in L . viator ) . Therefore , all efforts have always been conducted to ascertain the epidemiological role of L . viator , both in nature and experimentally in the laboratory [36 , 59–66] . Three recent findings suggest the need to review the situation in Uruguay: ( i ) the involvement of Galba/Fossaria species as L . viator and L . neotropica in human fascioliasis endemic areas [50 , 51]; ( ii ) their experimentally verified high transmission capacity [67]; and ( iii ) the potential impact of climate change on fascioliasis and Galba/Fossaria species [1 , 2 , 9] , given the predictions on the climate change impact on Uruguay , including a rainfall increase [68] . Galba/Fossaria species , including both L . viator and L . neotropica , are amphibious snails which markedly depend on environmental and climatic factors such as temperature , water availability and evapotranspiration [2 , 7 , 9] . Thus , the predicted increase of rainfall [68] may a priori offer more possibilities for lymnaeid population growth and consequently the higher number and spread of vectors allow for an increased fascioliasis transmission . The main aim of the present study is to assess the fascioliasis situation in Uruguay by means of the following aspects: The aforementioned results furnish the baseline needed for the understanding of the reasons underlying the difference between the high human prevalences and intensities in Andean highlands and the only sporadic human cases or small outbreaks in non-Andean lowlands . This purpose is achieved by means of the appropriate analysis of the following objectives: The latter objective becomes crucial for a country as Uruguay where cattle raising is the most important activity of the primary sector; cattle are kept on more than 83% of farms; on more than half of them beef cattle are the main source of income . The most important beef breed is Hereford , with 76 . 0% of the herd . There are more than 6 , 500 farms specialising in dairying , with more than 750 , 000 animals , more than 90% of which are Holsteins [69] . Moreover , Hereford and Holstein breeds appear to be the most affected by fascioliasis in Uruguay , with prevalences of 56% ( 95% confidence interval: 51–61% ) and 68% ( 95% ID: 64–73% ) , respectively , pronouncedly higher than the prevalences in other breeds [41] .
Fasciolid materials were obtained from naturally infected animal hosts from Uruguay . A total of 46 fasciolid adult worms were obtained from livers of two Hereford , two Holando and two Aberdeen Angus breed cattle from three zones in the Salto department . Additionally , fasciolid eggs were obtained from a biliary filtrate of three horses from a slaughterhouse in Montevideo department ( Fig 1 ) . According to available facilities in obtaining materials , the collecting strategy was to sample materials from the western part of the country through which border the first fasciolids should have been introduced into Uruguay for the first time in the past , taking into account the original spread of livestock with the early Spanish conquerors . In Uruguay , fasciolids infect both cattle and sheep in the same places , because these two species are kept mixed in the grazing pastures and areas . Taking into account that F . hepatica infects cattle and sheep similarly ( i . e . , these two species do not select different fasciolid strains ) , the fasciolid material was obtained from cattle because this is the livestock regularly killed in the slaughterhouses , given that cattle raising is the most important activity of the primary sector in Uruguay . The material from horses was , however , not obtained in the same western areas , because horses also share the same pastures and should logically be infected by the same fasciolid haplotypes . Therefore , horses were selected from the Montevideo department , where these animals are managed in a somehow different way because of the neighbourhood of the big city . Fasciolids from the aforementioned cattle and horses were used for DNA marker sequencing and a F . hepatica isolate obtained in a 6-year-old Hereford cattle female from Salto was used for the experimental infections of lymnaeid snails . Uruguayan fasciolid materials have been deposited in the Museu Valencià d’Història Natural , Alginet , Spain , under the code MVHN-241016MD01 . Lymnaeid snail materials originated from three different populations in Uruguay , corresponding to the departments of Montevideo ( 6 specimens ) , Paysandú ( 4 specimens ) and Canelones ( 10 specimens ) ( Fig 1 ) and were used for their molecular characterization by DNA marker sequencing . Given the very low fasciolid larval stage prevalences in lymnaeid vectors in nature , a broader snail survey was not within the goals of this assessment of F . hepatica in Uruguay . Moreover , lymnaeids collected in Canelones and shortly maintained in the laboratory of the DILAVE "Miguel C . Rubino" , were finally transported and cultured in the Laboratory of the Valencia centre for a standardized experimental study . A total of 50 laboratory-borne specimens were used for the infection experiments with the aforementioned Uruguayan F . hepatica isolate . All lymnaeid specimens collected and used were preliminarily classified by shell morphology as Lymnaea viator , following the literature on lymnaeids in Uruguay ( Table 1 ) . Uruguayan lymnaeid materials have been deposited in the Museu Valencià d’Història Natural , Alginet , Spain , under the code MVHN-241016MD02 . The living standard of Uruguay is closely related to earnings from pastoral and agricultural exports of beef and wool . Extensive cattle and sheep rearing is the main activity of Uruguay , where half the grassland is in estancias ( usually large , simple buildings with thick walls , of a typical Spanish colonial style , with a lot of wrought iron ) exceeding 2 , 000 acres . More than 13 , 500 , 000 ha are under permanent pasture , almost 83% of the agricultural area [69] . Millions of sheep and cattle are raised in the country . The preference for pasture over cropland is due to the excellence of the grasslands and the variable rainfall that makes grain production unreliable . The ratio between sheep and cattle production shifts with demand [70] . The predominant sheep breeds in Uruguay are Corriedale , Merino and Polwarth , which represent 60% , 20% and 10% of the national sheep flock , respectively . These breeds generate income from the sale of wool and sheep meat ( surplus offspring and cast for age animals ) . Traditionally , wool has been the main product of the system . However , in recent years , the importance of sheep meat ( lambs and mutton ) has increased significantly [71] . Living Galba/Fossaria lymnaeids preliminarily classified as L . viator , collected in Canelones and shortly maintained in the DILAVE laboratory , were transported under isothermal conditions to the laboratory of Valencia . Transfer to Valencia was needed to allow for a standardized laboratory adaptation and subsequent experimental follow-up of the life cycle and transmission of Uruguayan flukes by Uruguayan lymnaeids under abiotic conditions enabling significant comparisons with endemic areas of other countries . The possible natural infection by fasciolids was always individually verified prior to the launch of laboratory cultures . This was performed by keeping each lymnaeid specimen isolated in a Petri dish containing a small amount of natural water . After 24 h , the presence or absence of motionless metacercarial cysts or moving cercariae was verified in each Petri dish . Afterwards , non-infected lymnaeids were arranged in standard breeding containers containing 2000 ml fresh water , to assure pure specific cultures . Finally , snails were adapted to and maintained under experimentally controlled conditions of 20°C , 90% relative humidity and a 12 h/12 h light/darkness photoperiod in precision climatic chambers ( Heraeus-Vötsch VB-0714 and HPS-500 ) . The water was changed weekly and lettuce added ad libitum . Eggs of F . hepatica from a 6-year-old bovine female from Salto were maintained in fresh water under complete darkness at 4°C until starting the embryogenic process . Embryogenesis was followed at 20°C at intervals of four days by counting eggs presenting an incipient morula , eggs including an advanced morula , eggs with outlined miracidium , and fully embryonated eggs containing a developed miracidium . Developed miracidia were forced to hatch by putting fully embryonated eggs under light and used for the experimental infection of snails [48] . Snails collected in the Uruguayan department of Canelones , shortly maintained in the laboratory of the DILAVE "Miguel C . Rubino" , and finally transported and kept in the Laboratory of the Valencia centre were used for the experiments . Only laboratory-borne specimens were used . Snails of different size within the length range of 4 . 7–7 . 6 mm ( mean 5 . 74 mm ) were used to assess infection susceptibility . A total of 50 lymnaeid specimens were infected monomiracidially by exposing each snail to 1 miracidium for 4 hours in a small Petri dish containing 2 ml of fresh water . The disappearance of the miracidium was taken as verification of its successful penetration into the snail . Snails were afterwards returned to the same standard conditions in the climatic chamber ( 2000 ml containers , 20°C , 90% relative humidity ( r . h . ) , 12 h/12 h light/darkness , dry lettuce ad libitum ) until day 30 post-infection , in which they were again isolated in Petri dishes to allow daily monitoring of cercarial shedding by individual snails . Lettuce was provided ad libitum to each snail in a Petri dish during both shedding and post-shedding periods until death of the snail . The chronobiology of the cercarial shedding was followed by daily counting of metacercariae in each Petri dish [48] . Furthermore , the strains of both F . hepatica and the lymnaeid used for the experimental assay were characterized by the sequencing of the aforementioned DNA markers . For that purpose , fasciolid metacercariae experimentally obtained and snails fixed after verification of the end of the cercarial shedding period were used ( Table 1 ) . Life cycle aspects analyzed and respective methods used are in agreement with the standards applied for such studies in Fasciola . Following this standardized way allows for significant comparisons of the transmission characteristics in different endemic areas [48] . Ethical approval for the animal work was provided by the Ethics Committee for Animal Experimentation and Welfare of the University of Valencia , Spain ( A1263915389140 ) . Additionally , the División de Laboratorios Veterinarios ( DILAVE ) , Montevideo , belongs to the corresponding national ministry ( Ministerio de Ganadería , Agricultura y Pesca—MGAP ) counting on its own ethics committee ( Comité de Etica para Uso de Animales de Experimentación—CEUA ) and its animal work is authorized by the Comisión Nacional de Experimentación Animal of Uruguay . Animal ethic guidelines regarding animal care strictly followed the institution’s guidelines based on Directive 2010/63/EU . Informed written consent was received from all animal owners ( farm: El Solar , Salto; owners: Sucesores de Alfredo Sanchis; official registry No . at the MGAP Ministry: 150 606 365 ) .
A total of 9 different marker sequences were obtained from the fasciolids . Nucleotide length of the sequences , their GC/AT content and reference codes are noted in Table 1 . Two haplotypes of the complete intergenic rDNA ITS1-5 . 8S-ITS2 region were detected in the fasciolids infecting cattle and also in horses in Uruguay . ITS-1 proved to have the same sequence in all specimens studied , corresponding to the haplotype A of this spacer ( Table 1 ) . ITS-2 showed two sequences in the Uruguayan fasciolids: haplotypes 1 and 2 GC ( Table 1 ) . Only one mutation in position 287 of the ITS-2 alignment allows the differentiation between both haplotypes: “C” in haplotype FhITS-2 H1 and “T” in FhITS-2 H2 . The mtDNA cox1 provided three different sequences with the same length . Their alignment showed 6 variable positions ( all of them singleton sites ) . These sequences proved to enter in the intraspecific variability of the 69 cox1 haplotypes so far known in F . hepatica , corresponding to the haplotypes Fhcox1-5 , Fhcox1-16 and Fhcox1-42 ( Fig 2 ) . The three haplotypes were found in cattle , whereas only Fhcox1-42 was found in horses ( Table 1 ) . The COX1 protein was 510 aa long , with start/stop codons of ATG/TAG , identical in all specimens analyzed , and corresponding to the haplotype FhCOX1-1 ( Fig 2 ) . The mtDNA nad1 sequence provided three different haplotypes with the same length . Their alignment showed 3 variable positions ( all of them singleton sites ) . These sequences also proved to enter in the intraspecific variability of the 51 nad1 haplotypes so far known in F . hepatica , corresponding to the haplotypes Fhnad1-2 , Fhnad1-12 and Fhnad1-14 ( Fig 3 ) . The three haplotypes were found in cattle , whereas only Fhnad1-12 was found in horses ( Table 1 ) . The NAD1 protein showed only one 300-aa-long haplotype with start/stop codons of GTG/TAG in all specimens analyzed , corresponding to the haplotype FhNAD1-1 ( Fig 3 ) . A total of 5 different marker sequences were obtained from the lymnaeids . Nucleotide length of the sequences , their GC/AT content and reference codes are noted in Table 1 . The ITS-2 sequences of lymnaeids from the three localities in Uruguay were identical . This unique sequence showed no one nucleotide difference with the ITS-2 haplotype H1 of the Galba/Fossaria species L . neotropica ( Table 1 ) . Similarly , the ITS-1 sequences were also identical in the three localities and without any difference when compared to the ITS-1 haplotype HA of L . neotropica ( Table 1 ) , which differs by two insertions in the “poli-A” region at the 3' end from haplotype HB ( positions 512 to 529 including 16 or 18 consecutive “A” in L . neo-HA and L . neo-HB , respectively ) . The 16S rRNA gene of the mtDNA provided only one haplotype in the three localities . This partial sequence presented a biased AT content , and proved to be identical to the provisional haplotype L . neo-16S HA ( Table 1 ) . The partial sequence of the mtDNA cox1 gene of lymnaeids from the three localities in Uruguay showed two haplotypes . The one found in Canelones proved to be identical to the haplotype L . neo-cox1 Ha from the type locality of L . neotropica . The second haplotype , present in the localities of Montevideo and Paysandu , proved to be identical to the provisional haplotype L . neo-cox1 He ( Table 1 ) . When comparing these two cox1 sequences from Uruguay with the five cox1 haplotypes of L . neotropica known so far , the resulting 672 bp-long alignment showed 76 variable positions , including two parsimony informative and 74 singleton sites . Nucleotide and amino acid differences are listed in Fig 4 . Experimental life cycle studies were undertaken with the Uruguayan F . hepatica combined haplotype FhITS2-1 , FhITS1-A , Fhcox1-5 , Fhnad1-2 found in Hereford cattle from Salto , and the L . neotropica combined haplotype L . neo-ITS2-1 , L . neo-ITS1-A , L . neo-16S-A , L . neo-cox1-a collected in the Canelones department ( Fig 5 ) . Results of embryogenesis inside the egg , lymnaeid snail infection , intramolluscan parasite larval development and influences of the latter on snail survival are noted in Table 2 . The use of identical experimental procedures and standardized abiotic factors allow for a significant comparison with the same data experimentally obtained with F . hepatica and G . truncatula from the high altitude pattern of the life cycle and disease transmission of cattle and sheep isolates of the liver fluke in the Northern Bolivian Altiplano , the human hyperendemic area with the highest human prevalences and intensities known ( Table 2 ) . The Uruguayan liver fluke isolate proved to follow a pronouncedly faster embryogenesis . The first developed miracidium appears already in day 15 , with the maximum percentage of eggs including fully developed miracidia in day 18 , whereas 46 and 58 days were needed by cattle and sheep isolates from the Bolivian Altiplano , respectively . Such a development speed is three times faster in the lowlands of Uruguay , even in a surprisingly high percentage of eggs of 88 . 2% ( whereas only in 24 . 9% and 16 . 4% for the two Bolivian isolates , respectively ) ( Table 2 ) . The high snail infectivity rate ( 74 . 5% ) of the Uruguayan isolate is worth mentioning , although similar to the Bolivian sheep isolate . The prepatent period ( days elapsed from infection day up to the first day of cercarial shedding ) in the Uruguayan isolate is markedly similar to that of the two Bolivian isolates . However , pronounced differences appear in the shedding period ( total number of days in which the snail was shedding cercariae ) , as well as in the total number of cercariae ( and subsequent metacercariae ) produced by each infected lymnaeid . The shedding period in the Uruguayan isolate proved to be very short , of only 1–19 days ( mean 9 . 6 days ) , despite of which the number of cercariae per snail was relatively high , of 4–1186 cercariae/snail ( mean 269 . 2 ) . These features are , however , far from the ones characterizing the Altiplanic liver fluke isolates , in which the shedding period is very long ( averages higher than 70 days in both isolates ) and the number of cercariae per snail is very high ( averages higher than 445 cercariae/snail in both isolates ) ( Table 2 ) . The geographical isolate did not seem to influence lymnaeid survival during the prepatent period , results obtained with the Uruguayan fluke being very similar to those in the two Bolivian isolates . Nevertheless , ( i ) the snail survival after the end of the shedding period , ( ii ) the postinfection longevity in shedding snails , and ( iii ) the longevity in non-infected snails , were all three pronouncedly shorter when dealing with the Uruguayan isolate than with the two Bolivian isolates ( Table 2 ) . Despite the fast and short intramolluscan larval development , the Uruguayan liver fluke isolate proved to be able to reach a marked extent of redial infection and massive presence of rediae in the local Uruguayan lymnaeids ( Fig 6A and 6B ) . The chronobiological pattern of cercarial emergence in the cattle isolate of F . hepatica is shown in ( Fig 7A and 7C ) . When the shedding period is analyzed from the day of the emergence of the first cercaria by each snail ( Fig 7A ) , the shedding process appears as an irregular succession of waves . After four days of an initial shedding of a reduced daily number of cercariae , a slow decrease of that number is observed . The higher acrophases take place at the end of the first week and during the second week . When the shedding period is analyzed from the day of the miracidial infection ( Fig 7C ) , most of the cercariae are shed between days 52 and 64 post-infection ( p . i . ) . The days 64 and 73 p . i . , in which all snails failed to shed any cercaria , suggest an intramolluscan larval development including up to a maximum of three redial generations . It is mainly for the first generation to produce and shed most of the cercariae . When comparing this chronobiological pattern of cercarial emergence in F . hepatica/lymnaeid snail from Uruguay ( Fig 7A and 7C ) with the chronobiological pattern in F . hepatica/G . truncatula from the Northern Bolivian Altiplano , performed under identical procedures and standardized experimental abiotic factors ( Fig 7B and 7D ) , four main differences should be highlighted:
Fasciolids from Uruguay molecularly prove to belong to widespread strains of F . hepatica , fitting within the intraspecific variability of this fasciolid species in Europe and Latin America . Indeed , F . hepatica was introduced into South America throughout a process which started 500 years ago at the time of the first Spanish colonizers , who were transporting livestock in their ships [6] . From the evolutionary point of view , such a period is very short for a parasite . Manter's parasitophylogenetic rule , about the slower evolution of parasites when compared to that of the hosts , should be considered here [80] . Moreover , the livestock host species in Latin America at present are the same than in its original spreading area in Europe 500 years ago , which means that the microhabitat of F . hepatica has not changed despite its hosts having been moved from one continent to another . Additionally , the evolutionary rates of the four DNA markers used are too low [81] as to expect mutations appearing by isolation in the Americas and differentiating American fasciolids from those of the Old World [6] . Short information may be inferred from the only one ITS-1 and two ITS-2 haplotypes of these evolutionarily conserved rDNA spacers . In fact , the worldwide spread of F . hepatica occurred only during the last 12 , 000–10 , 000 years , from the moment of the domestication of livestock in the Fertile Crescent in the Near East and in Old Egypt , when humans began to expand livestock species throughout . Similarly occurred with F . gigantica , although restricted to Africa and Asia where its specific Radix lymnaeid species are present . The absence of Radix in the New World ( only isolated populations of R . auricularia introduced into North America from Europe ) explain why only F . hepatica is present in the Americas [6] . A period of 12 , 000–10 , 000 years is too short for ITSs to give rise to mutations , given their evolutionary rate [81] . Thus , in regions where only F . hepatica is present , and consequently there is no possibility for hybridization with F . gigantica , the same only FhITS-1 HA is known [6] . Regarding ITS-2 , FhITS-2 H1 and FhITS-2 H2 found in Uruguay also correspond to the two haplotypes known in areas presenting genetically pure F . hepatica [6] . FhITS-2 H1 is the most widely distributed , whereas FhITS-2 H2 has interestingly been described from Spain and Andorra [6] and was already previously reported from Uruguay [82] . More information can be inferred from the three cox1 and three nad1 haplotypes of the faster evolving mtDNA [81] . Among cox1 , Fhcox1-5 and Fhcox1-16 are widely dispersed in South America , but Fhcox1-42 has only been found in Bolivia [6] . A similar picture is provided by nad1 . Fhnad1-2 and Fhnad1-14 are distributed throughout North and South America , whereas Fhnad1-12 has only been detected in Peru , Bolivia and Argentina [6] . The only two other complete mtDNA gene sequences of F . hepatica from Salt Lake City , Utah , USA [83] and the Geelon strain in Australia [84] are different from the F . hepatica haplotype group of the Iberian Peninsula and Latin America , at the level of both cox1 ( Fig 2 ) and nad1 ( Fig 3 ) . The single COX1 and NAD1 protein haplotypes found in Uruguay are the most abundant , both widely distributed in different countries and different host species [6] . The detection of Fhcox1-42 and Fhnad1-12 in both cattle and horses indicate infection from same sources , e . g . in Uruguay horses may become infected when grazing in pastures used for cattle and sheep [44] . The sequences of the four DNA markers used unambiguously demonstrate that the Galba/Fossaria lymnaeids collected in the three Uruguayan localities belong to the species L . neotropica . The ITS-2 and ITS-1 found in Uruguay are identical to L . neo-ITS-2 H1 and L . neo-ITS-1 HA in Perú and Argentina ( 74 , 50 , 51 , 75 ) , the latter differing by only two "A" insertions from L . neo-ITS-1 HB from Argentina [51] . Similarly , the 16S haplotype L . neo-16S HA found in Uruguay has also been reported from Perú and Argentina [50 , 51] . The first cox1 haplotype L . neo-cox1 Ha found in Canelones was already known from the type locality of L . neotropica and another area in Peru [50 , 74] . The second L . neo-cox1 He found in Montevideo and Paysandu was known only from Argentina [51] . So far , the only Galba/Fossaria species reported in Uruguay was L . viator [36 , 55 , 56 , 59–66] . Consequently , L . neotropica becomes a new species for the Uruguayan fauna . Its finding in areas located far away one another , indicate that this species should be widely distributed throughout the country . The question posed now is whether both L . neotropica and L . viator coexist in Uruguay , or there is simply only L . neotropica which has always been confused with L . viator . Indeed , DNA sequencing of lymnaeids started at the end of the last century , including markers of mtDNA [85] and nuclear rDNA [86] . Their progressive use highlighted the problems in specimen classification and species differentiation by traditional malacological approaches [48 , 72 , 73] . Interspecific differentiation in Galba/Fossaria became the main focus , due to their importance in the transmission of F . hepatica . The description of the new species L . neotropica and its molecular differentiation from L . viator and G . truncatula was an important step forward [74] . The molecular demonstration that the hitherto overlooked species L . schirazensis , without fascioliasis transmission capacity , had always been confused with G . truncatula and other Galba/Fossaria vector species , illustrated up to which level there was a chaotic systematic situation [53] , as verified in Venezuela [33] . Similarly , DNA sequencing proved the presence of G . truncatula , L . neotropica and L . schirazensis in the human fascioliasis hyperendemic area of Cajamarca , in Peru , where only L . viator had been involved [50] . All these major advances were posterior to all studies on lymnaeids in Uruguay . This suggests that we are only dealing with a classification confusion between L . neotropica on one side and L . viator and G . truncatula on the other side , similarly as happened in Peru , Venezuela , Argentina , and also Chile [30] . So , L . neotropica should probably be the only Galba/Fossaria species distributed throughout Uruguay . However , it should not be overlooked that L . neotropica and L . viator may coexist in the same area , as in Mendoza [87 , 88] and Catamarca [51] , both in Argentina . Studies in other areas of Uruguay are needed . For this purpose , it shall be considered that ITS-1 and 16S showed the highest and lowest resolution for interspecific differentiation , respectively , whereas cox1 was the best marker and ITS-1 the worst for intraspecific analyses [51] . Regarding genus ascription of L . neotropica and L . viator , their last molecular comparison , both one another and inside the Galba/Fossaria group , and maximum support values obtained for the internal branching nodes in the phylogenetic analysis of the species of Galba/Fossaria , demonstrated that these Neotropical species do not belong to the genus Galba defined by its Palaearctic type species G . truncatula [51] . Until sequence data from the very large number of Galba/Fossaria species known from the Nearctic region [89 , 90] is obtained , caution recommends to taxonomically keep L . neotropica and L . viator in the genus Lymnaea sensu lato . The sequences of L . neotropica from Uruguay and those from Peru [50 , 74] , Venezuela [33] and Argentina [51 , 75 , 88] , suggest that the spread of this lymnaeid throughout the Neotropical region should have occurred very recently , passively transported with livestock exchange , in a similar way as other Galba/Fossaria species spread throughout even different continents , as G . truncatula [6] and L . schirazensis [53] . Finally , it should be highlighted that all the haplotypes of the four DNA markers found in L . neotropica from Uruguay have also been found in two human fascioliasis endemic areas , such as Cajamarca in Peru [50] and Catamarca in Argentina [51] . Uruguayan F . hepatica and L . neotropica used for the experiments were selected to assess the disease transmission capacity of the most common and widely dispersed strains of both fluke and vector species . The embryonation time found in the Uruguayan couple proves to be very short . It fits within the fastest inside the range known when tested at 20°C [111–113] . In F . hepatica , the development of the miracidium inside egg is arrested below 9°C and above 37°C and has a duration between 9 and 161 days depending upon the temperature , the range 20–25°C offering the optimum for the hatching of a higher number of miracidia [114–117] . The detected infection percentages and prepatent period in monomiracidial infections may be considered as normal at 20°C when compared to similar studies carried out with F . hepatica isolates and G . truncatula specimens from European areas [111 , 114 , 118–122] . The prepatent period found in the Uruguayan couple also fits in the known range ( 43 . 1 ± 58 . 2 days ) for European F . hepatica/G . truncatula in the nature [123] . The shedding period in Uruguayan F . hepatica/L . neotropica is very short . In European F . hepatica experimentally infecting G . truncatula snails of the same size as ours under the same constant conditions of 20°C and 12 h/12 h photoperiod , the patent period lasted only 46 ± 27 . 6 days [124] . Similarly , results obtained in nature show that the patent period in Europe ranges between 5 . 0 and 9 . 3 days in the winter generation and 18 . 3–40 . 3 days in the summer generation [123] . The pronounced differences of the very short shedding period in the couple from the Uruguayan lowlands when compared to the very long one in the F . hepatica/G . truncatula couple from the highlands of the Bolivian Altiplano ( Table 2 ) [48] should be highlighted . The mean number of cercariae shed per individual lymnaeid in the Uruguayan couple is close to the mean of 238 . 5 cercariae/snail found in the European F . hepatica/G . truncatula model under the same experimental conditions [124] . A lower number of 114 . 9 ± 80 . 3 cercariae per monomiracidially infected snail were obtained with the same European couple . These experimental assays showed that the duration of the shedding and the number of cercariae were independent of the number of miracidia used for the infection of each individual lymnaeid . However , single-miracidium infections were most effective because of the much higher snail survival rate , despite the mean number of cercariae shed being the same as in multimiracidial infections [125] . However , the most important is the marked difference when compared to the pronouncedly higher mean cercarial numbers in the F . hepatica/G . truncatula couple from the Bolivian Altiplano ( Table 2 ) [48] . Differences in survival of different geographical strains of the same Galba/Fossaria species to F . hepatica infection have already been described [126] . The postinfection longevity in shedding L . neotropica from Uruguay is only slightly shorter than the 70 days p . i . usually observed in European G . truncatula [127–131] , with a maximum of 16 weeks described once [128 , 129] , and far from that known in other American lymnaeids such as 119 days p . i . for L . viator [132] and 113 . 4 days p . i . for L . bulimoides [133] . A fast development and extensive massive infection of the larval stages in L . neotropica ( Fig 6 ) may be the responsible for a quick snail mortality , similarly as in European G . truncatula . It was found that of 102 snails shedding on the first day , the number drastically reduced to only 56 on the second day and subsequently decreased on day 76 to four snails [124] . Regarding this aspect , the pronounced difference when compared with the capacity of G . truncatula from high altitude endemic areas to survive up to more than 4 months after the end of the shedding period ( Table 2 ) should be highlighted [48] . The cercarial shedding pattern detected in the Uruguayan couple does not disagree with the patterns observed by other authors on the F . hepatica/G . truncatula model [124 , 134–136] . When considering the shorter shedding period in the Uruguayan couple , the three shedding acrophases it shows ( Fig 7A and 7C ) fit well in the 4–5 shedding waves showed by the majority of G . truncatula in an experiment under constant conditions [124] . The delayed acrophases in F . hepatica/L . neotropica from Uruguay also agree with the pattern found in the European model . Here again , the pronounced differences of the couple from the Uruguayan lowlands with the F . hepatica/G . truncatula couple from the highlands of Bolivia [48] should be highlighted: regarding ( i ) the daily number of cercariae/snail , ( ii ) length of the shedding period , ( iii ) daily number of cercariae/snail , and ( iv ) cercarial production by the different redial generations . The longer post-infection longevity of G . truncatula under high altitude conditions [48 , 49] and the higher pathogenicity induced by the fast development and massive infection by F . hepatica in L . neotropica from Uruguayan lowlands underlie the aforementioned differences . Summing up , our experimental results demonstrate that the F . hepatica/L . neotropica couple from Uruguayan lowlands is markedly less efficient for the disease transmission than the F . hepatica/G . truncatula couple from the Andean highlands , although somewhat more efficient than the F . hepatica/G . truncatula couple from European lowlands . The latter result agrees with other experimental data indicating that L . neotropica and L . viator from Argentina are better hosts than European ( French ) G . truncatula in both allopatric and sympatric infections by Argentinian and French isolates of F . hepatica [67] . The efficiency results of both , our present study of the Uruguayan couple and the one of the Argentinian and French mixed couples [67] , may be interpreted taking into account that allopatric combinations of F . hepatica and lymnaeid species were proved to be more efficient than sympatric ones [137] . This capacity may be considered a useful strategy of the liver fluke for the colonization of new areas [6] . Indeed , the aforementioned historical analysis suggests that L . neotropica did not colonize Uruguay until at a maximum the first part of the 17th century , and consequently , around only 400 years ago . This is a very short period for the parasite from the evolutionary point of view . The aforementioned ( i ) Manter's rule [80] , ( ii ) similarity of ancestral European livestock and present Uruguayan hosts , and ( iii ) low evolutionary rates of DNA markers used [81] , should again be considered in this assumption . So , the term "allopatric" should be applied with caution here . Only a total of 95 human fascioliasis cases have been reported in Uruguay . The first report of a human subject infected by F . hepatica in Uruguay was in 1909 and concerned a 49-year-old women suffering from right hypocondrium pain and in whom a fluke was unexpectedly found near the main biliary duct during surgery [138] . Twenty years later , three liver fluke specimens were detected during a gall bladder surgical intervention of another patient [139] . Between 1935 and 1950 , isolated human cases were reported after egg detection in stool samples and/or duodenal exploration [140–142] . A familiar outbreak involving 11 members , probably linked to contaminated watercress consumption , was one year later diagnosed in the hospital of Paysandu . An exhaustive clinical and epidemiological study was performed [143] . The flooding events during the 1954/55 and 1958/59 periods were suggested to have spread fascioliasis and therefore related to three subsequently reported human epidemics [35] . Thus , a total of 31 cases were reported from the Florida department in 1960 [144] , 20 additional cases were compiled from seven different departments from the country inland , the majority from Florida , Canelones and San José in 1978 [145] , and finally another 16 cases diagnosed in an hospital during the 1953–1977 period were anatomo-pathologically described in 1979 [146] . Interestingly , among a review of patients infected by the HIV virus during the 1983–1988 period , F . hepatica eggs were found in the stools of a patient affected by AIDS and dying after 35 days hospitalization [147] . Only two additional cases were detected among a total of 951 samples during a wide serological and coprological survey performed in several localities of the departments of Artigas , Rivera , Florida and Salto throughout 1991 and 1993 ( Lopez Lemes et al . , 1992 and 1993 , in [35] ) . Another infected patient with fever and eosinophilia was noted to be diagnosed both coprologically and immunologically in 1990 by E . Zanetta , in the same article [35] . Another familiar outbreak involving only three subjects and linked to the consumption of wild watercress , was reported ( Lopez Lemes et al . , 1994 , in [35] ) . Finally , the last report was in 2003 about two female and one male clinical cases presenting with right hypocondrium pain , eosinophilia and history of watercress consumption [148] . The aforementioned review suggest a sporadic and isolated human infection risk throughout a wide hyperendemic animal fascioliasis zone in Uruguay , according to the epidemiological classification of WHO [149] . Several aspects merit , however , to be considered . Despite the distribution of the liver fluke covering the whole country [37] , the human infection risk does not appear to be homogeneous , i . e . it seems to be higher in given departments , as in Florida department [144 , 145] . Moreover , the unexpected finding of human infection in surgical interventions [138 , 139] , HIV-infected patient survey [147] , and wide surveys [35] , suggest underestimation of the real occurrence of human infection , similarly as in Argentina [150] . The familiar outbreaks also remember the human fascioliasis situation in the lowlands of Argentina . However , whereas in the physiographically highly heterogeneous Argentina available data suggested human endemic local areas which have been finally described [51] , the physiographic uniformity of Uruguay does not indicate such a scenario . Nevertheless , the three increases of patient numbers after the flooding events of the 1954/55 and 1958/59 periods [144–146] should be considered by the public health responsibles , given IPCC ( Intergovernmental Panel on Climate Change ) predictions of a rainfall increase within the future climate change impact affecting Uruguay [68] . Anyway , there is nothing indicating a trend towards a worrying human infection scenario such as in Andean areas . Neither the ecological characteristics and preferences of the main vector L . neotropica [64] , nor its transmission capacity verified in the present study , suggest such a future possibility . This does not mean , however , that the very fast larval development of F . hepatica and short shedding of high numbers of cercariae furnished by L . neotropica may take advantage of occasional , more or less prolonged flooding events to increase offspring , population densities and subsequently spread , thus enabling for an increase of familiar outbreaks or short transient epidemic situations . The comparison of the transmission characteristics and capacities in the F . hepatica/G . truncatula couple from Bolivia with the F . hepatica/L . neotropica couple from Uruguay allow for the understanding of the high transmission patterns and endemicity characteristics of human fascioliasis in Andean highlands , opposite to the rare/sporadic/low human infection in animal endemic areas in Uruguayan lowlands . Consequently , it may be concluded that L . neotropica may be responsible for a human endemic area only under special circumstances , as in isolated foci in aridity/dryness areas described in Argentina [51] . The transmission characteristics and capacities of the Uruguayan F . hepatica/L . neotropica couple are a priori better for a seasonal transmission of the disease , depending on local climatic features . Uruguay has a subtropical to temperate climate with very marked seasonal fluctuations [69] . The climate is sub-humid , because potential evapotranspiration in summer is greater than precipitation . Although rainfall is distributed throughout the year , great variations occur between years . The highest precipitation occurs , in general , in summer and autumn . In the first season , precipitation is very irregular , with summers lacking precipitation and others with more than 600 mm of rain . In the second season , precipitation shows minor variability . Although precipitation has a somewhat smaller volume in winter than in other seasons , there is no marked rainy season . The great rainfall variation , both in regularity and intensity , should be highlighted because it leads to droughts and floods in different seasons of the year . Mean temperatures of the coldest month ( July ) are 10 . 8°C and 13 . 0°C , and the warmest month averages ( January ) are 22 . 6°C and 25 . 1°C for the Southern and Northern regions , respectively [69] . In Uruguay , field studies indicated that the fluke life cycle is maintained throughout the whole year , although it considerably slows down in winter [35] . Lymnaeids naturally infected in autumn-winter , with mean maximum temperatures lower than 20°C and mean minimum ones below 10°C , showed a 4–8 month cercarial shedding , whereas this was reduced to only 37 days in summer [63] . In spring , shedding periods gradually shorten , which together with an increase of lymnaeid population densities at the end of spring , gives rise to an increase of the number of infected animals at the end of spring and summer [38] . In summer , temperatures are ideal for F . hepatica development , but the insufficient rainfall and high evapotranspiration resulting in a humidity shortage become important limiting factors for the lymnaeids [40] . The long survival and infectivity of metacercariae [151] also add to understand the human infection risk the year long , despite it being higher in spring .
|
Fascioliasis is a highly pathogenic zoonotic disease emerging in recent decades , in part due to the effects of climate and global changes . South America is the continent presenting more numerous human fascioliasis endemic areas and the highest Fasciola hepatica infection prevalences and intensities known in humans . These serious public health scenarios appear mainly linked to altitude areas in Andean countries , whereas lowland areas of non-Andean countries , such as Uruguay , only show sporadic human cases or outbreaks . To understand this difference , we characterized F . hepatica from cattle and horses and lymnaeids of Uruguay by sequencing of ribosomal DNA ITS-2 and ITS-1 spacers and mitochondrial DNA cox1 , nad1 and 16S genes . Results indicate that vectors belong to Lymnaea neotropica instead of to Lymnaea viator , as always reported from Uruguay . Our correlation of fasciolid and lymnaeid haplotypes with historical data on the introduction and spread of livestock species into Uruguay allow to understand the molecular diversity detected . We study the life cycle and transmission features of F . hepatica by L . neotropica of Uruguay under standardized experimental conditions to enable a comparison with the transmission capacity of F . hepatica by Galba truncatula at very high altitude in Bolivia . Results demonstrate that although L . neotropica is a highly efficient vector in the lowlands , its transmission capacity is markedly lower than that of G . truncatula in the highlands . On this baseline , we review the human fascioliasis cases reported in Uruguay and analyze the present and future risk of human infection in front of future climate change estimations .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] |
[
"invertebrates",
"livestock",
"medicine",
"and",
"health",
"sciences",
"ruminants",
"geographical",
"locations",
"tropical",
"diseases",
"vertebrates",
"fascioliasis",
"parasitic",
"diseases",
"animals",
"mammals",
"genetic",
"mapping",
"physiological",
"processes",
"gastropods",
"neglected",
"tropical",
"diseases",
"snails",
"molting",
"sheep",
"south",
"america",
"uruguay",
"molluscs",
"agriculture",
"people",
"and",
"places",
"helminth",
"infections",
"haplotypes",
"heredity",
"physiology",
"genetics",
"biology",
"and",
"life",
"sciences",
"cattle",
"amniotes",
"bovines",
"organisms"
] |
2017
|
DNA multigene characterization of Fasciola hepatica and Lymnaea neotropica and its fascioliasis transmission capacity in Uruguay, with historical correlation, human report review and infection risk analysis
|
Hookworms ( Necator americanus and Ancylostoma duodenale ) remain a major public health problem worldwide . Infections with hookworms ( e . g . , A . caninum , A . ceylanicum and A . braziliense ) are also prevalent in dogs , but the role of dogs as a reservoir for zoonotic hookworm infections in humans needs to be further explored . As part of an open-label community based cluster-randomized trial in a tribal area in Tamil Nadu ( India; 2013–2015 ) , a total of 143 isolates of hookworm eggs from human stool were speciated based on a previously described PCR-RFLP methodology . The presence of hookworm DNA was confirmed in 119 of 143 human samples . N . americanus ( 100% ) was the most prevalent species , followed by A . caninum ( 16 . 8% ) and A . duodenale ( 8 . 4% ) . Because of the high prevalence of A . caninum in humans , dog samples were also collected to assess the prevalence of A . caninum in dogs . In 68 out of 77 canine stool samples the presence of hookworms was confirmed using PCR-RFLP . In dogs , both A . caninum ( 76 . 4% ) and A . ceylanicum ( 27 . 9% ) were identified . Additionally , to determine the contamination of soil with zoonotic hookworm larvae , topsoil was collected from defecating areas . Hookworm DNA was detected in 72 out of 78 soil samples that revealed presence of hookworm-like nematode larvae . In soil , different hookworm species were identified , with animal hookworms being more prevalent ( A . ceylanicum: 60 . 2% , A . caninum: 29 . 4% , A . duodenale: 16 . 6% , N . americanus: 1 . 4% , A . braziliense: 1 . 4% ) . In our study we regularly detected the presence of A . caninum DNA in the stool of humans . Whether this is the result of infection is currently unknown but it does warrant a closer look at dogs as a potential reservoir .
Infections with hookworms ( Necator americanus and Ancylostoma duodenale ) remain a major public health problem in several low and middle-income countries [1 , 2] . It is estimated that ~439 million people are infected , resulting in a global disease burden of ~3 . 5 million disability adjusted life years ( DALYs; 62 . 3% of the DALYs attributable to soil-transmitted helminths; ~3% of all DALYs attributable to Neglected Tropical diseases ( NTDs ) ) [3] . The major morbidity associated with hookworm infections is caused by intestinal blood loss , iron deficiency anaemia , and protein malnutrition [4] , most of which occurs in children and pregnant women [5] . The current strategy to control the morbidity caused by these intestinal worms are embedded in large-scale school-based deworming programs , in which benzimidazole drugs ( albendazole and mebendazole ) are administered to schoolchildren regardless of their infection status [6 , 7] . However , it remains unclear whether these school-based deworming programs are the most efficient approach [8] . First , both prevalence and the intensity of hookworm infections increase as a function of age . Although most of the deworming programs target school-aged children , the major contributors of hookworm infection both in terms of prevalence and total egg excretion are adults , who are often not included in deworming programs [6 , 8] . Second , the eggs excreted in stool are non-infectious , and need to develop and hatch on the soil before larvae can transcutaneously enter the human host and cause disease [9] . Therefore it will be important to supplement deworming programs with improved water , sanitation and hygiene ( WASH ) to prevent re-infection [10] . Moreover , benzimidazole drugs have a moderate efficacy against hookworm and never reach 100% efficacy [11 , 12] . Third , it is traditionally assumed that hookworm infections are caused by the human hookworms N . americanus and A . duodenale , and hence hookworm infections in humans are solely due to the contamination of soil with human stool [13] . Infections with the hookworms ( e . g . A . caninum , A . ceylanicum , A . braziliense and Uncinaria stenocephala ) in dogs are also highly prevalent , and depending on the species these hookworms may also cause a variety of clinical symptoms in humans [14] . A . ceylanicum is the only known species to cause patent infection in humans [15] with symptoms ranging from gastrointestinal discomfort , epigastric pain , flatulence and diarrhoea , whereas the rest are mainly limited to lesions in the skin caused by migrating larvae ( cutaneuos larva migrans ) [16] . Migration to the intestine has been reported for A . caninum which may cause severe eosinophilic enteritis [14 , 17] . Recent studies also indicate that the role of animals as a source of hookworm infections in humans should not be ignored . For example , in a study done in a rural region of Cambodia [18] 64 out of 124 ( 51 . 6% ) individuals were found to be infected with the animal hookworm A . ceylanicum , of which the majority were mono-infections ( 89% ) [16] . Similarly , a study done in a tribal region in India [19] , found that human hookworm infections ( N . americanus 39/41; 95% and A . duodenale 6/41; 15% ) accounted for majority of the infections , whereas the animal hookworm A . ceylanicum only accounted for a minority of the infections ( 2/41; 5% ) , and hence these findings suggest that the rate of zoonotic transmission might vary across different geographical areas . Despite these studies , the role of animals as a reservoir for hookworm infections in human remains poorly explored . This lack of understanding of disease transmission among both humans and animals , is largely due to the fact that diagnosis of hookworm infections are based on the microscopic demonstration of eggs in stool , but it is impossible to differentiate animal and human hookworm eggs based on morphology . For this , molecular tools are more appropriate [20] . Second , various studies have identified hookworm species separately across humans [19 , 21 , 22] , dogs [23 , 24 , 25] or soil [26] , but to our knowledge there are no studies which have identified hookworm species using molecular techniques within both hosts and environment in the same geographical region . The present study aims at molecular identification of hookworms isolates from humans , dogs and soil from a tribal area in Tamil Nadu , India . The selection of this study area was based on ( i ) a high prevalence of hookworm infection in humans ( 38% ) [27] and ( ii ) the presence of factors that facilitates zoonotic hookworm transmission in humans .
This study was part of an open-label , community-based cluster randomized trial that was approved by the Institutional Review Board of Christian Medical College , Vellore , India . This trial is registered in the Clinical Trials Registry of India ( CTRI/2013/05/003676 ) . A description of the ethical considerations has been described in detail elsewhere ( Sarkar et al . , under review ) . In short , a written informed consent was obtained from parents/legal guardian for the collection of stool samples from children aged less than 18 years of age and an assent was obtained from 8–17 year old children . Participants older than 18 years of age signed their own informed consent form . Jawadhu hills are situated in Vellore and Thiruvannamalai district of Tamil Nadu ( southern India ) . It covers an area of 150 km2 and a population of approximately 80 , 000 of which the majority is tribal . The population is organized in 11 ‘panchayat’ ( a group of villages under one local administrative council ) and 229 villages [28] . The area is known to have red loamy soil [28] . The temperatures of the region ranges between 12°C and 33°C [28] . There is excessive rainfall ( >1000 mm ) [28] , with relative humidity varying from 40 to 85% [28] . The majority of the population is employed in the agricultural sector and lives in close proximity with animals , including dogs and cats . It is important to note that these animals are not confined , and although they belong to one household are found freely roaming through the village . Across the entire area there is a common practice of open-field defecation [27] . The study was part of an open-label , community-based cluster randomized trial conducted between 2013–2015 . The aim of the trial was to compare the hookworm re-infection rates for one year in a population that was subjected to varying cycles of deworming using albendazole . Therefore , 15 clusters ( villages ) were randomized into one of three different treatment arms: ( i ) single cycle , ( ii ) two cycles and ( iii ) four cycles . The timing of deworming in each of the three groups have been described in detail elsewhere ( Sarkar et al . , 2016; under review ) . In short , in the single cycle , individuals received a single oral dose of albendazole once in the beginning of the study and stool samples were collected 3 , 6 , 9 and 12 months post-treatment . In the treatment arm of two cycles , individuals received two single oral doses of albendazole . The first dose was given at the start of the trial and second dose after one month . The stool samples were collected 3 , 6 , 9 and 12 month after the administration of the second dose . In the treatment arm of four cycles , the first two doses of albendazole were given at the start of the trial with one-month interval , and an additional two doses of albendazole were given at 6 months after the 2nd dose . Stool samples were also collected after 3 and 6 months post 2nd dose and 3 , 6 , 9 and 12 months post 4th dose of albendazole . In the present study , stool ( humans and dogs ) and soil were collected from nine clusters included in the trial ( 3 per treatment arm ) , including Seramarthur , Jambudee , Alanjanur , Sinthalur , Koothatur , Villichanur , Keel Nadanur , Thimirimarathur and Pudhupattu . In 2013 the total population of the 9 villages was 2 , 906 habitants ( 1 , 492 males and 1 , 414 females ) belonging to 683 families . Human stool samples were collected as per trial protocol described above . Based on the treatment arm , stool samples were collected at an interval of three months until the end of one year after the last treatment . Field workers visited the house of the study a day before collection was scheduled and handed over a plastic stool container and wooden spatula . Containers were appropriately labelled . The stool samples were collected the following day and stored at the study area at 4°C before being transported to the laboratory using cold containers . In total 2 , 152 stool samples were collected from 711 individuals from the 9 selected villages . All samples were screened microscopically applying a saline wet mount . Stool in which hookworm eggs were found were subsequently screened using the McMaster egg counting method to estimate the intensity of infection ( faecal egg counts ( FECs ) expressed in number of eggs per gram of stool ( EPG ) ) . Finally , one stool sample per infected subject was withheld for molecular analysis , and stored at -70°C . If hookworm eggs were found in multiple samples from the same individual , the sample with the highest FEC was selected . A total of 146 out of the 711 individuals were found to be excreting eggs in at least one time point . The median ( 25th quantile ( Q25 ) - 75th quantile ( Q75 ) ) FEC among the infected individuals was 550 EPG ( 200–1 , 000 ) . The number of infected humans and the corresponding median FEC across the different villages are summarized in Table 1 . Field workers identified 10 houses per village based on structured questionnaire that had previously claimed dog ownership , and volunteered to help collect stool samples . As the dogs in these villages were not usually confined , the owners chained their dogs for a day and stool sample was collected into the plastic stool container using a spatula after the dog defecated . In each village , a single stool sample from 10 dogs was collected ( n = 90 ) . The stool samples were collected and stored at 4°C at the site of collection . The samples were transported to the laboratory in cold containers . As with human stool samples , dog stool were first screened with saline wet mount , after which stool of infected animals were re-examined using the McMaster egg counting method . All samples containing hookworm eggs were withheld for molecular analysis , and stored at -70°C . In total , 77 out of 90 dogs excreted hookworm eggs in stool . The median ( Q25—Q75 ) FEC across infected dogs was 350 EPG ( 100–650 ) . The number of infected dogs and the corresponding median FEC across the different villages are summarized in Table 2 . Soil samples were collected from common open defecation areas for each of these villages . Field workers opportunistically collected soil samples from hot spots of the defecation site chosen for the study . The soil samples that were collected were found to be loamy and wet . The collection of soil samples was in conjunction with human stool samples . Hookworm larvae are known to be present in the top soil early in the morning [29 , 30] and therefore sample collection was done between 8 a . m . and 10 a . m . Depending on the number of open defecation areas in a village and the number of cycles of deworming , 20 to 40 soil samples were collected per village . Approximately 250–300 grams of topsoil was collected and transported in plastic bags at room temperature to the laboratory on the same day . All samples were screened for the presence of hookworm-like nematode larvae applying a modified saline wet mount . In total , 271 samples were collected from 22 open defecation sites . In 78 samples out of 271 samples hookworm-like nematode larvae were identified . The total number of soil samples collected and the number of soil samples containing hookworm-like nematode larvae across the different villages are summarized in Table 3 . All samples were stored at 4°C until the molecular identification of the larvae .
From the 711 individuals that participated in the study , 146 individuals were found to be infected with hookworm using saline wet mount microscopy . Out of the 146 infected individuals , 143 individuals provided adequate quantity of stool samples to perform molecular characterization , and hookworm-DNA was detected in 119 individuals ( 83 . 2% ) . All had N . americanus , while A . caninum was found in 20 individuals and A . duodenale in 10 individuals . The distribution of the different hookworm species in human stool across the 9 villages is summarized in Table 1 . Based on a structured questionnaire , the odds of being infected with A . caninum when claiming dog ownership ( 11/299 ) was 1 . 71 ( 95% confidence interval = 0 . 69–4 . 18 ) times higher when no dog ownership was claimed ( 9/412 ) , but this was not statistically significant . On account of high prevalence of A . caninum DNA in human stool samples , dog stool samples were collected from the 9 villages . A total of 90 dog stool samples were collected , of which hookworm was detected in 77 samples using saline wet mount microscopy . Hookworm was detected in 68 of the 77 dog samples ( 88 . 3% ) selected for molecular characterization . A . caninum was the predominant species , being found in 52 dogs . A . ceylanicum was found in 19 ( 27 . 9% ) dogs . Among them three dogs had mixed A . caninum and A . ceylanicum infections . The distribution of the different hookworm species across the 9 villages is summarized in Table 2 . To assess the role of soil as a source of zoonotic hookworm infection , a total of 271 soil samples were collected from defecating areas across 9 villages . Hookworm-like nematode larvae were found in 78 out of the 271 samples that were collected . Of the 78 soil samples identified positive for hookworm-like nematode larvae , 72 ( 92 . 3% ) were found positive by PCR for hookworms . Molecular characterization of these 78 soil samples revealed the presence of a variety of hookworm species , including A . caninum , A . ceylanicum , A . duodenale , A . brazilense and N . americanus . The majority of these soil samples were contaminated with A . ceylanicum ( n = 41; 56 . 9% ) , followed by A . caninum ( n = 20; 27 . 7% ) and A . braziliense ( n = 1; 1 . 4% ) . The human hookworms were only found in the minority of the samples ( A . duodenale: n = 8; 11 . 1%; N . americanus: n = 1; 1 . 4% ) . The distribution of the hookworm species across the different villages is summarized in Table 3 . In total 69 hookworm isolates from different sources ( 29 human , 26 Dog and 14 soil ) were sequenced in the present study . The phylogenetic tree is provided in Fig 1 , and highlights that each species identified by the PCR-RFLP cluster nicely together with their respective reference sequences . The human-derived A . caninum sequences ( Study ID nos . Human20493 , Human16162 , Human11351 , Human11017 , Human9949 , Human9931 , Human9851 , Human9843 , Human9661 , Human9659 , Human6365 , Human6303 , Human6253 , Human1882 , Human1649 and Human1596 ) clustered with the A . caninum sequences from dogs and soil , forming a cluster with A . caninum reference sequences ( GenBank accession no . DQ438070 ) . The human-derived sequences of A . duodenale ( Study ID nos . Human2545 , Human2880 , Human2552 and Human2544 ) centred within a clade with the reference A . duodenale sequence ( GenBank accession no . EU344797 ) . The N . americanus sequences from humans clustered as a separate larger cluster . All 69 sequences were submitted to GenBank and assigned the accession numbers from KU996361 to KU996390 , and from KX155777 to KX155815 Due to low detection of N . americanus larvae in the soil samples but the high prevalence of hookworm ( N . americanus ) in humans , an experiment was carried out to assess the analytic sensitivity of detecting N . americanus larvae in soil samples using the standardized assay as described in material and methods section: . For the various concentrations ( 70 , 140 , 280 and 700 larvae ) of the larvae in the stock that was used to spike , there was a large variation in mean recovery rate ( % ) of the N . americanus L3-larvae , ranging from 51 . 1% when 70 larvae were added to 92 . 9% when 700 larvae were added . The results of the larvae recovery rate are presented in Table 4 . The larvae that were isolated from these experiments were characterized using PCR and were confirmed to be N . americanus .
There has been a worldwide upscale of drug donations to control the morbidity caused by hookworms , and to even attempts to eliminate these worms in confined geographical areas [6] . It is traditionally assumed that infections in humans are solely due to the human hookworms ( N . americanus and A . duodenale ) [6 , 34] , hence ignoring the possible role of animals as a reservoir for hookworm infections . This study determined the hookworm species in humans , dogs and soil from a tribal area in Tamil Nadu , India . The findings from our present work confirm that N . americanus are responsible for the majority of the hookworm infections in humans in these tribal communities [19] . In addition , we also found an unexpectedly high prevalence of animal hookworm DNA in humans , while our previous study [19] in Jawadhu hills had a low prevalence of animal hookworms ( 5% ) . These differences in occurrence might be explained by variation in prevalence across villages . Both studies covered different villages , and as illustrated in Table 1 , there was a large variation in animal hookworm infections in humans across villages ( A . caninum was found in 6 out 6 cases in Alanjanur , but was absent in Thimirimarathur and Villichanur ) . There was no significant evidence of an increased risk of hookworm infections with dog ownership , which is in contrast with the studies reported by Traub et al . , and Ugbomoiko US et al . , [24 , 35] , who did observe a significant increased risk . The lack of this evidence maybe due to the fact that animals are not confined , but are found freely roaming through the village . As a consequence of this they are able to randomly defecate within the village , which subsequently will increase the likelihood of infecting other habitants beyond their owner . In the present study a large proportion of human stools were found to contain A . caninum DNA . These observations can be explained by ( i ) A . caninum infections , ( ii ) passive passage of A . caninum eggs or larvae that are accidently ingested , but do not result in any infection and ( iii ) contamination of stool during sample collection with environmental A . caninum eggs or larvae . Although it is unlikely that passive passage explains the high proportion of stool samples containing A . caninum DNA , we have no conclusive evidence for any of the remaining potential causes either . Traditionally it is assumed that parasite DNA in stool is due the presence of eggs shed by adult worms . However , up to today there is no evidence yet that egg-laying adult A . caninum worms can develop in humans [14 , 36 , 37] , and hence one would not expect any amplification of DNA from A . caninum extracted eggs in human stool . In our study we were not able to provide evidence for the presence of A . caninum eggs in stool , as the human hookworm N . americanus was also detected in all cases of A . caninum . As a consequence of this , it is possible that the eggs in stool were shed by adult N . americanus worms only . A single-egg based speciation would have been ideal . Another potential source of parasite DNA is DNA that is directly released by immature or mature non-egg producing worms . To demonstrate the presence of both immature and non-egg producing mature A . caninum worms expulsion studies are required [36] . In these studies stool is collected over consecutive days following treatment to recover worms , which are then individually speciated . Both a single-egg based speciation and an expulsion study were out of scope of the present study . Another aspect that needs to be considered during the interpretation of our findings is the way the human samples were collected . Although all the study participants were informed about the importance of the study and need to collect fecal samples devoid of any soil , stool samples could have been contaminated with soil during collection because people in the study area defecate in the open and samples might have been scooped from the ground which could result in the presence of A . caninum DNA in human stool . In either case , these results emphasize the need for additional epidemiological surveys across various geographical settings to further explore the role of animals as a reservoir for zoonotic transmission . It is important to note that this does not only apply for hookworms , but is also of concern for other soil-transmitted helminths ( Trichuris trichiura and Ascaris lumbricoides ) . This is because dogs are known to harbor Trichuris spp . ( T . vulpis ) , which , similarly to canine hookworm species , is known to infect humans causing symptoms ranging from an asymptomatic infection to diarrhea or even dysentery [38] . T . vulpis has also been reported as a causative agent of visceral larva migrans [39 , 40 , 41] . In the dog stool samples , both known zoonotic hookworm ( A . caninum and A . ceylanicum ) species were found . Although A . ceylanicum is the only canine hookworm species that is known to cause patent infections in humans [42] , our present study did not identify any human A . ceylanicum infections , in spite of detecting it in both soil and canine stool samples . In contrast , a study from a rural village in Cambodia reported more than half of the hookworm infected individuals to be positive for A . ceylanicum ( 51 . 6% ) [18] A possible explanation for the presence of A . ceylanicum in dogs and soil , but not in humans , is the existence of two sub species ( haplotypes ) of A . ceylanicum , one with an animal origin and one with a human origin [42] . To differentiate these subspecies the cytochrome c oxidase ( COX ) subunit 1 gene of the hookworm is recommended [42] . In addition , the applied PCR-RFLP method may lack some sensitivity; this is in particular for mixed infections . As previously illustrated for other gastro-intestinal parasites ( e . g . Giardia; Geurden et al . , 2008 and Levecke et al . , 2009 ) , genus specific PCRs will preferentially amplify the most abundant species , and hence presence of the least abundant species may be underestimated [43 , 44] . This could be one possible explanation for missing out zoonotic A . ceylanicum infections in humans . For the soil samples that were collected from defecating areas , it was interesting to observe variety of different species of hookworm larvae ( A . braziliense , A . caninum , A . ceylanicum , A . duodenale and N . americanus ) . This can be attributed to open defecation practised in the study area and indiscriminate defecation by freely roaming stray dogs . There are a few limitations to our present study . First , N . americanus was rarely found in soil samples , whereas this hookworm species was found in all human stool samples . For this reason , we determined the analytical sensitivity of the isolation procedure using N . americanus L3-larvae to confirm efficiency of the assay to isolate and identify the species . The results of this seeding experiment suggest that the procedure was able to detect 3 . 5 larvae per gram of soil , and hence ruling out false negative test results due to loss of larvae . Another potential cause of absence of N . americanus could be inappropriate storage of the soil samples prior to the molecular analysis . In this study , the soil samples were first processed for hookworm-like larvae using modified saline wet mount microscopy , subsequently they were stored at 4°C ( up to 14 months ) until further processed for molecular identification . Unlike Ancylostoma spp . , N . americanus stored in cold temperature do not survive long [29] . It is therefore important to mention that the only case of N . americanus was observed in one out of four samples containing hookworm-like larvae that were processed almost immediately after collection for molecular speciation . Second , for the collection of soil samples , the samples were collected from area around the site of defecation/presence of stool ( human and dog ) , which makes the selection biased , and hence it increased the probability of finding hookworm larvae . In conclusion , in our study we regularly detected the presence of A . caninum DNA in the stool of humans . Whether this is the result of an infection is currently unknown but it does warrant a closer look at dogs as a potential reservoir . Nevertheless , there is a need for additional epidemiological surveys across different geographical settings to further unravel the role of animals as a reservoir for zoonotic transmission , and ultimately inform the health policy makers to adapt or improve measures to control soil-transmitted helminths as a public health problem .
|
Hookworm infections remain a major public health problem in both tropical and subtropical parts of the world . To control the disease burden attributable to hookworms , large-scale deworming programs , in which drugs are administered to schoolchildren regardless of their infection status , are currently being implemented in endemic regions . However , these programs face some difficulties . One of them is the uncertainty about the role of animals in the transmission of hookworm infections . It is commonly believed that human-specific hookworms cause these infections , but there is growing evidence that the role of some animal-specific hookworms as cause of infection in humans should not be underestimated . We determined the different hookworms in humans , dogs and soil ( eggs excreted by adult hookworms in stool are non-infectious , and need to develop and hatch on the soil before larvae can transmit disease by penetrating the skin ) in a tribal area in India . In this area , the transmission of hookworms between humans and dogs is possible . Our results highlight the presence of DNA from animal-specific hookworms in both soil and human stool . Although these findings suggest that these animals could act as reservoir for zoonotic hookworm infections in humans , they should be interpreted with caution . This is because we lack the evidence to confirm A . caninum infections in our study population . Other potential reasons for the presence of DNA in stool are contamination of stool with environmental eggs or larvae during sample collection and passive passage in which eggs or larvae are ingested but did result in any infection .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"sequencing",
"techniques",
"invertebrates",
"medicine",
"and",
"health",
"sciences",
"geographical",
"locations",
"vertebrates",
"india",
"parasitic",
"diseases",
"dogs",
"animals",
"mammals",
"necator",
"americanus",
"physiological",
"processes",
"developmental",
"biology",
"molecular",
"biology",
"techniques",
"research",
"and",
"analysis",
"methods",
"sequence",
"analysis",
"infectious",
"diseases",
"zoonoses",
"sequence",
"alignment",
"necator",
"molecular",
"biology",
"defecation",
"people",
"and",
"places",
"helminth",
"infections",
"asia",
"physiology",
"nematoda",
"biology",
"and",
"life",
"sciences",
"metamorphosis",
"amniotes",
"larvae",
"organisms"
] |
2016
|
Molecular Identification of Hookworm Isolates in Humans, Dogs and Soil in a Tribal Area in Tamil Nadu, India
|
Chagas disease is a neglected tropical disease that continues to affect populations living in extreme poverty in Latin America . After successful vector control programs , congenital transmission remains as a challenge to disease elimination . We used the PRECEDE-PROCEED planning model to develop strategies for neonatal screening of congenital Chagas disease in rural communities of Guatemala . These communities have persistent high triatomine infestations and low access to healthcare . We used mixed methods with multiple stakeholders to identify and address maternal-infant health behaviors through semi-structured interviews , participatory group meetings , archival reviews and a cross-sectional survey in high risk communities . From December 2015 to April 2016 , we jointly developed a strategy to illustratively advertise newborn screening at the Health Center . The strategy included socioculturally appropriate promotional and educational material , in collaboration with midwives , nurses and nongovernmental organizations . By March 2016 , eight of 228 ( 3 . 9% ) pregnant women had been diagnosed with T . cruzi at the Health Center . Up to this date , no neonatal screening had been performed . By August 2016 , seven of eight newborns born to Chagas seropositive women had been parasitologically screened at the Health Center , according to international standards . Thus , we implemented a successful community-based neonatal screening strategy to promote congenital Chagas disease healthcare in a rural setting . The success of the health promotion strategies developed will depend on local access to maternal-infant services , integration with detection of other congenital diseases and reliance on community participation in problem and solution definition .
Chagas disease is a vector-borne illness that can also be transmitted congenitally , via blood transfusion , organ donation , lab accidents or ingestion [1] . With the implementation of vector control programs , insect transmission of Trypanosoma cruzi has become less common [2] and vertical transmission has increased in importance [3–6] . In Argentina , a prospective study showed that 67 . 3% of 107 patients enrolled were infected congenitally , while only 4 . 7% via vector transmission [7] . In 2005 , the Guatemalan Ministry of Health ( MoH ) proposed to include congenital Chagas disease screening and treatment of children . [8] . However , program implementation has been limited by lack of evidence on congenital incidence rates . We are implementing a strategy to screen congenital transmission in populations at highest risk . Congenital Chagas disease is an acute infection [9] with 27–57% asymptomatic cases in children [10 , 11] . The consequences of infection in utero can be seen prior to birth , with spontaneous abortion and stillbirth and , upon birth , neonates have a higher mortality within the first two days [10] . Diagnosis of congenital Chagas can be achieved with varying degrees of sensitivity by screening the newborn´s blood within the first month after birth by microscopy , hemoculture or by polymerase chain reaction [5 , 12 , 13] . Infants may be screened serologically 10 months after birth , when maternal transplacental antibodies have waned [14] . Some potential risk factors for vertical transmission of Chagas disease include the degree of parasitemia [15–17] , the presence of acute infection in the mother [18] , and co-infection with HIV [17 , 18] . Treatment should be implemented immediately after diagnosis to improve prognosis [19] . The oral treatment must be monitored by trained health personnel due to potential adverse effects [5 , 15] . Thus , screening and treatment programs require access to maternal-infant care within an institutional platform . Over the past five years , we have worked at the municipality of Comapa in the Department of Jutiapa . This is a region of eastern Guatemala that , prior to the launching of the vector control program in 2000 , had some of the highest triatomine infestations ( >40% infested households ) [20] and seroprevalence in school-age children ( 13 . 75% ) [10] in the country . We extended our previous multidisciplinary study of Chagas disease vector control [21] , working in collaboration with the health personnel , communities and non-governmental organizations to establish a congenital Chagas disease healthcare program . This study aimed to improve congenital Chagas disease detection and treatment in this rural area of Guatemala through a multi-stakeholder driven strategy , based on the PRECEDE ( Predisposing , Reinforcing , and Enabling Causes in Educational Diagnosis and Evaluation ) PROCEED ( Policy , Regulatory and Organizational Constructs in Educational and Environmental Development ) model [22] for community interventions . After program implementation , newborns are being screened for Chagas disease at the Health Center ( HC ) .
The study obtained ethical approval from both the Universidad del Valle de Guatemala ( #108-10-2014 , #100-04-2014 ) and the Ministry of Health of Guatemala ( 01–2014 ) Institutional Review Boards . Individual written consents were obtained from participants before interviews and health access surveys . Comapa is a municipality located in the department of Jutiapa , in the southeastern region of Guatemala bordering El Salvador at -89°54′46 . 8″ and 14°6′38 . 6748″ ( Fig 1 ) . Comapa was selected to develop the congenital Chagas disease surveillance protocol due to the presence of a newly built maternity ward ( 2012 ) , the relevance of the disease to local health authorities , an ongoing Chagas diagnosis and treatment program for children and adults , and an incipient Chagas disease prenatal screening program . Prenatal screening includes a rapid diagnostic test ( if available ) at the HC in Comapa , with the rapid test provided by the non-governmental organization ( NGO ) , Médicos con Iberoamérica ( IBERMED ) , followed by a single ELISA test performed at the Area Laboratory in Jutiapa . Quantitative and qualitative research methods were used to understand the local socio-ecological system driving health behaviors . Ultimately , we aimed to develop and implement a sustainable community process for the surveillance of congenital Chagas disease . For this , we conducted the situational assessments of all phases of the PRECEDE component of the model: phase 1 ( social ) , phase 2 ( epidemiological ) , phase 3 ( educational and ecological ) and phase 4 ( administrative and policy ) . We also conducted one phase of the PROCEED component of the model , phase 5 ( design and implementation ) of the health promotion strategy [22] . We partnered with the MoH and identified stakeholders ( midwives , NGOs , Municipal offices , maternal HC and laboratory personnel ) throughout the study to ensure the joint identification of the problems and solutions . Table 1 shows the project phases , timeline , activities and stakeholders in chronological order .
Maternal-infant care in Comapa , Jutiapa , involves public health services , midwives and NGOs . We implemented a multi-stakeholder strategy for neonatal screening to offer timely diagnostics and treatment of congenital Chagas disease . The strategy was generated at the local level through a process including participatory activities with midwives and HC personnel , followed by community-based health communication and educational programs regarding Chagas disease management . To allow newborn screening and early treatment , the strategy requires ( 1 ) reaching the population at highest risk for infection through a community-based health communication program , ( 2 ) inclusion of midwives , clinic personnel and NGOs in the implementation of promotional materials for early diagnosis at the HC and ( 3 ) HC personnel trained to ( a ) take newborn blood samples , ( b ) perform a simple microscopic method to detect parasites in the blood sample , and ( c ) provide treatment and follow-up for infected neonates . The strategy takes into consideration current maternal-infant care policies and practices at the HC in Comapa , with inclusion of regional NGOs . It also takes advantage of the role played by midwives in informal maternal-infant care , as well as the current national policy requiring their training . Before our study , infants born to positive women were not screened . To promote newborn detection and treatment , education of midwives and women 15–45 years of age must be developed in a culturally appropriate way . The participatory meetings allowed the development of a socioculturally appropriate strategy for the promotion of congenital Chagas disease screening and treatment in the region . The newborn screening procedure was designed to have a low cost , requiring only microscopic evaluation of the newborn´s blood . The PCR was proposed to confirm microscopic results during method implementation . Once optimized , the parasitological method alone could be implemented in other endemic areas with high seroprevalence . Despite the limitations in maintaining trained personnel in rural areas , we propose that the microscopic method has a potential for sustainability due to its low cost , and could become a standard of care for newborns in these regions . However , rapid test based on the detection of T . cruzi IgM antigen would be better and should be considered as an alternative once available . On the other hand , the training procedures with midwives can become part of the ongoing program to improve maternal-infant health in the country . Cost estimations have not been included given that all procedures can be implemented without additional expenditure to ongoing activities at the Health Center . Limitations of the proposed strategy will likely include the sustainability of the community-level education programs to promote maternal-infant follow-up visits , the inclusion of the program in current prenatal screening programs such as HIV and syphilis , and the ability to maintain HC competency in parasitological diagnosis and record keeping [29] . As observed in South America [30] , social and technical constraints in Chagas disease management in Guatemala include lack of knowledge on the disease , loss to follow up , side effects that lead to treatment non-adherence , lack of communication between decentralized health system levels and lack of training on diagnostics and treatment . We propose that the inclusion of midwives as empowered stakeholders has resulted in referral of newborns to the health center . Future studies will evaluate the strengths and limitations of this strategy , and recommended improvement . The scaling up of the strategy will require a train-the-trainer program targeting reproductive health and nurse coordinators at the department level for prioritized areas . In addition , evidence of local transmission and education campaigns are needed to empower stakeholders at all levels . Targeted communication campaigns should be developed based on in-depth knowledge of the sociological and cultural behavior of the communities regarding maternal and neonatal care , and how they interact with the health authorities . Forms for recording screening and treatment of mothers and neonates must be developed or modified , and methods of reporting to epidemiological , vector control and policy authorities strengthened . A supply of treatment medication must also be ensured . The treatment of T . cruzi-infected women after delivery to reduce the risk of congenital transmission remains a challenge because there are no guidelines regarding treatment during the lactation period . Women with Chagas disease can breast feed , unless they are in the acute phase with high parasitemia , reactivated disease or have bleeding nipples [31] . In rural areas where women have multiple pregnancies , treating infected women during lactation would allow completion of the two-month course before another pregnancy [32] . Research in this area is needed to provide evidence . Finally , treatment before the first pregnancy reduces the risk of congenital transmission [33] and should be considered in future prevention strategies . A 3 . 9% seroprevalence in pregnant women attending the HC indicates that early congenital detection and treatment should be a priority in areas with similar historically high triatomine infestations [34] and seroprevalence [14] in Guatemala . To achieve elimination , more studies are needed to understand the prevalence of congenital disease in such populations . In similar areas with persistent triatomine infestation , the MoHs must ensure that Chagas disease control and prevention programs integrate innovative vector control strategies and attention to treatable congenital disease . Future assessment of the strategy is needed to ensure its long term effectiveness and sustainability . The strategy could be expanded to other congenital diseases by strengthening the network of midwives and maternity ward personnel through training in symptom detection at the community level and case referral to health facilities in areas with low access to health services . Given the recent emergence of Zika as a new vector-borne congenital disease , we propose that this stakeholder driven strategy could be implemented in areas with limited access to maternal-infant health services .
|
Chagas disease is caused by a parasite transmitted by insects that infest households living in extreme poverty conditions . The parasite can also be transmitted from mother to child during pregnancy . If detected at birth , the infection can be treated effectively with available drugs . However , access to professional neonatal healthcare is limited in rural communities such as those affected by Chagas disease . We developed a strategy to promote access to a simple neonatal diagnostic test in a rural region of Guatemala considered at risk for Chagas disease . The strategy included collaboration between Health Center personnel , midwives and non-governmental organizations that play a local role in maternal-infant care . During the implementation of a health promotion campaign , screening revealed previous infection in almost one of every 25 pregnant women . Most babies born to positive women were tested at the Health Center for parasites in blood . The implementation of similar strategies to prevent congenital Chagas in other rural areas should consider local maternal-infant care practices . This strategy of collaboration between Ministry of Health , community health workers , non-government organizations , academia and external governmental support could be expanded to screen for other diseases , such as Zika , that require early detection to improve overall infant health .
|
[
"Abstract",
"Introduction",
"Methods",
"Discussion"
] |
[
"medicine",
"and",
"health",
"sciences",
"midwives",
"medical",
"personnel",
"tropical",
"diseases",
"parasitic",
"diseases",
"developmental",
"biology",
"neglected",
"tropical",
"diseases",
"infectious",
"disease",
"control",
"public",
"and",
"occupational",
"health",
"infectious",
"diseases",
"protozoan",
"infections",
"congenital",
"disorders",
"people",
"and",
"places",
"health",
"screening",
"professions",
"chagas",
"disease",
"diagnostic",
"medicine",
"population",
"groupings",
"biology",
"and",
"life",
"sciences",
"neonates"
] |
2017
|
Towards Chagas disease elimination: Neonatal screening for congenital transmission in rural communities
|
Recently , Wu and colleagues [1] proposed two novel statistics for genome-wide interaction analysis using case/control or case-only data . In computer simulations , their proposed case/control statistic outperformed competing approaches , including the fast-epistasis option in PLINK and logistic regression analysis under the correct model; however , reasons for its superior performance were not fully explored . Here we investigate the theoretical properties and performance of Wu et al . 's proposed statistics and explain why , in some circumstances , they outperform competing approaches . Unfortunately , we find minor errors in the formulae for their statistics , resulting in tests that have higher than nominal type 1 error . We also find minor errors in PLINK's fast-epistasis and case-only statistics , although theory and simulations suggest that these errors have only negligible effect on type 1 error . We propose adjusted versions of all four statistics that , both theoretically and in computer simulations , maintain correct type 1 error rates under the null hypothesis . We also investigate statistics based on correlation coefficients that maintain similar control of type 1 error . Although designed to test specifically for interaction , we show that some of these previously-proposed statistics can , in fact , be sensitive to main effects at one or both loci , particularly in the presence of linkage disequilibrium . We propose two new “joint effects” statistics that , provided the disease is rare , are sensitive only to genuine interaction effects . In computer simulations we find , in most situations considered , that highest power is achieved by analysis under the correct genetic model . Such an analysis is unachievable in practice , as we do not know this model . However , generally high power over a wide range of scenarios is exhibited by our joint effects and adjusted Wu statistics . We recommend use of these alternative or adjusted statistics and urge caution when using Wu et al . 's originally-proposed statistics , on account of the inflated error rate that can result .
Genome-wide association studies ( GWAS ) have been remarkably successful at identifying the genomic locations of variants involved in a variety of complex diseases [2]–[7] . In spite of this success , some researchers have expressed disquiet at the issue of the ‘missing heritability’ [8] , namely the fact that the disease-associated single nucleotide polymorphisms ( SNPs ) identified through GWAS often account for only a small proportion of the the observed correlations in phenotype between relatives . This suggests that additional genetic factors remain to be found . Several explanations for this phenomenon have been suggested . Firstly , the SNPs identified through GWAS are likely to be surrogates in ( imperfect ) linkage disequilibrium ( LD ) with the true causal variants , and thus cannot be expected to fully account for their effects , particularly if the true causal variants are rare . Secondly , the low power of GWAS to detect loci of small effect means that many specific true loci remain undiscovered , even though the fact of their ( combined ) existence may be detectable from the observed genetic data [9] , [10] . Finally ( and the main focus of this communication ) is the fact that the single-locus ( SNP by SNP ) testing strategy generally undertaken as the primary analysis tool in a GWAS may be underpowered to detect loci that interact with other genetic or enviromental factors , since effects at such loci may not be visible unless the contributing interacting factors are also taken into account . The relationship between biological and statistical interaction has been hotly debated over many years [11]–[19] . It is now generally accepted that the lack of direct correspondence between statistical and biologial interaction makes it difficult to make strong inferences concerning biological mechanism from the existence of interaction terms in a statistical model . Nevertheless , the existence of such terms does imply that the interacting factors should at least both be ‘involved’ in disease in some way . Detection of statistical interaction thus provides a good starting point for a more focussed investigation of the joint involvement of the relevant factors , which can perhaps be better addressed through other types of experimental data . In addition , the increased detection power provided by statistical models that include interaction terms , when such terms do in fact operate [20] , motivates the development of improved methods for detecting and modelling statistical interaction , particularly in the context of GWAS . The hope is that such methods will be useful for detecting effects that may be missed in standard single-locus analysis , thus providing a complementary strategy to standard GWAS analysis approaches for detecting loci involved in disease . In case/control studies , statistical interaction is generally modelled as departure from a simple linear model describing the individual ( main ) effects of predictor variables on the predicted log odds of disease [17] . Consider two binary variables , and , whose presence/absence ( coded 0/1 ) is believed be associated with a disease outcome . Logistic regression models the main effects ( and ) and interaction term ( ) between the variables via the linear model ( 1 ) where represents the probability that an individual in the study is a case rather than a control . Applying this idea to genetic predictor variables ( such as SNP genotypes ) is complicated by the fact that genetic predictors are not binary , but rather take 3 levels according to the number of copies ( 0 , 1 , 2 ) of the susceptibility allele possessed . However , we can easily convert to a binary coding by assuming a recessive or dominant model for each of the factors considered ( thus collapsing two genotype categories to one at each locus ) . Alternatively , we can fit the above regression model using predictor variables coded ( 0 , 1 , 2 ) , according to the number of susceptibility allele possessed , thus imposing an additive model ( on the log odds scale ) within each locus for the effect of each susceptibility allele . Yet another approach would be to fit a more general nine parameter ( saturated ) genotype model [17] , that includes effects due to one or two copies of the susceptibility allele at locus 1 ( , ) , at locus 2 ( , ) , and four interaction parameters ( , , , ) representing the additional contribution to risk from combinations of these effects , resulting in the following model: ( 2 ) ( where here represents an indicator variable for the occurence of event ) . Given the simpler logistic regression model ( 1 ) , a variety of tests can be performed to assess the effects of the two contributing factors . ( Similar tests can be derived for logistic regression model ( 2 ) ) . A 3df test of tests for association at both loci , allowing for their possible interaction . A 2df test of tests for association at locus 2 , allowing for possible interaction with locus 1 . Such a test has been shown to be a powerful approach when interactions exist , while losing very little power when no interactions exist [20] . In the current communication , we will focus on the 1df test of i . e . a test of the interaction term alone . This test has the disadvantage of being generally underpowered compared to tests of main effects [21] . However , we might hope that loci with reasonably large main effects will be potentially detectable via a single-locus scan . We are interested in detecting loci that will be missed via single-locus analysis , i . e . those for which the interaction term is likely to be particularly important . Moreover , assuming we can construct a good test of , this test can potentially be combined with tests of the main effects [22] , allowing the construction of joint tests of association while allowing for interaction , if desired .
Recently , Wu and colleagues [1] proposed two novel statistics for genome-wide ( pairwise ) interaction analysis using case/control or case-only data . The statistics proposed by [1] were based on considering ‘haplotypes’ at two diallelic loci , and , with locus having alleles and and locus having alleles and . For linked loci , the concept of ‘haplotype’ corresponded to its usual interpretation in terms of the physical coupling of alleles on the DNA strand inherited from a single parent . For unlinked loci , the concept of ‘haplotype’ referred simply to the fact the alleles involved were inherited from the same parent ( a concept sometimes referred to as gametic phase disequilibrium ) , without necessarily implying any physical coupling of the alleles . Wu et al . [1] propose to detect interaction via consideration of the log odds ratio ( 3 ) where is the haplotype frequency of haplotype − ( i . e . the probability of this haplotype ) in some sample under consideration . We define a parameter vector , chosen to reparameterise the 4 haplotype frequencies in terms of the allele frequencies , , , , and , and a ‘linkage disequilibrium’ ( LD ) ( or more generally , for unlinked loci , allelic association ) parameter , , such thatNote that the odds ratio in ( 3 ) relates to the odds of a allele appearing on a ‘haplotype’ in coupling with a allele ( and a allele with ) i . e . it acts a measure of correlation between alleles at the two loci , rather than relating to the odds of disease . No correlation ( ) corresponds to the situation where the allelic association parameter . Wu et al . ( 2010 ) propose that , under the null hypothesis of no interaction , , where and refer to calculating within the sample of cases and controls respectively . If , in addition , there is no population-level allelic association between alleles at and , then . Wu et al . [1] give a complicated description motivating their use of , however this quantity can perhaps more easily be motivated by analogy with classical ‘case-only’ analysis [23] , [24] , [25] . Case-only analysis stems from the observation that , for binary predictor variables , a test of the interaction term in the logistic regresssion model ( 1 ) can be obtained by noticing that it equals the ‘ratio of odds ratios’:where and are the joint probabilities that binary variables and take values and , i . e . , calculated within the sample of cases and controls , respectively . If variables and are uncorrelated in the controls ( or , equivalently , in the general population under a rare disease assumption ) then the denominatorand a test of interaction can be constructed by testing whetherThis test has the advantage [23] , [24] , [25] of being substantially more powerful than the usual logistic regression test of . If we are not willing to assume that variables and are uncorrelated in the controls , then a natural test of interaction can instead be constructed by testing whetheror , equivalently , whetherConsidered in this light , the log odds ratio considered by Wu et al . [1] can be seen as analagous to the quantity used in case-only analysis , if the unit of analysis is defined to be a ‘haplotype’ ( rather than an individual ) and if binary variables and are defined as indicator variables for the two possible alleles at each locus on the haplotype . To test for interaction , Wu et al . [1] propose two test statistics , one for case-only and one for case/control analysis , which we denote as and ( 4 ) ( 5 ) Here is the log OR ( for − and − alleles being in coupling , as opposed to − and − ) , is its estimated variance ( calculated using the delta method ) , and and refer to quantities calculated within the sample of cases and controls respectively . The case-only test should be suitable provided there is no correlation ( e . g . due to LD ) between alleles at the two loci . The case/control test is more suitable if we expect correlation between alleles at the two loci due to the fact they are linked , or induced by other influences such population stratification [26] ) . In order to actually calculate and , we need to know ( or estimate ) the ‘haplotype’ frequencies in cases and controls . Even for linked loci , haplotypes are not generally observed , but luckily many programs exist to estimate haplotype frequencies ( often via an EM algorithm ) given unphased genotype data . Most if not all such programs assume Hardy-Weinberg equilibrium ( HWE ) in order to perform the calculation . We expect HWE to hold in the general population ( and thus in controls , under a rare disease assumption ) . Under the global null hyothesis of no association between disease status and the loci in question ( via either main effects or interactions ) , haplotype frequencies in cases should be identical to those in controls , and HWE should also hold in the cases . However , under the alternative hypothesis of association and/or interaction , HWE will not necessarily hold in the cases [27] ( unless the disease model is assumed to result from multiplicative haplotype effects [28] ) , meaning that haplotypes in cases cannot be considered to come together independently . We return to this point later . Wu et al . [1] provide the following formulae for their proposed statistics: ( 6 ) ( 7 ) where and are the number of sampled case and control individuals , and and are estimators of the haplotype frequencies in cases and controls , respectively . However , the denominators in these formulae ( based on calculating the asymptotic variances of and ) are only correct if haplotypes are actually observed i . e . there is no phase uncertainty . Consequently , we expect these variance estimates to be too small if haplotype frequencies are estimated from unphased genotype data , resulting in test statistics that are too large . In Text S1 we use results from Brown [29] and application of the delta method to calculate the correct asymptotic variances of and . We refer to our corresponding resulting test statistics as ‘adjusted’ Wu statistics: ( 8 ) ( 9 ) ( where now relates to the correct asymptotic variance of as given in Text S1 ) . Interestingly , if one calculates this variance under the null hypothesis that ( as might be reasonable when performing case-only analysis , where this assumption is in any case required ) , it turns out that the resulting variance is exactly double that derived by Wu et al . [1] . In these circumstances , our case-only statistic would be exactly half of the original Wu case-only statistic . This suggests that another way to construct a valid version of Wu's case-only statistic would be to simply divide the original statistic by two . In computer simulations ( data not shown ) , we found negligible differences between between our ‘adjusted’ statistic and , and thus , in our Results section , we only report results for . Two fast approaches for testing interaction ( in addition to a slower logistic regression based approach ) are implemented in the computer program PLINK [30] . For a set of individuals ( either cases or controls ) , PLINK takes unphased genotype data as shown in Table 1 and expands it out to the 22 allelic table shown in Table 2 . The log odds ratio in this table can be calculated as with estimated variance . PLINK's fast-epistasis tests test whether correlation between alleles at the two loci exists ( case-only test ) or is different between cases and controls ( case/control test ) via the following test statistics: ( 10 ) ( 11 ) Here and again refer to quantities calculated within the sample of cases and controls respectively . These statistics are seen to have exactly the same form as the Wu and adjusted Wu statistics , but with the log odds ratio and its estimated variance relating to slightly different quantities , namely those quantities shown in Table 2 . Apart from the difference in , the main difference between PLINK's statistics and those proposed by Wu et al . is that fact that , in PLINK , no estimation of phased haplotype frequencies is performed . Nevertheless , the log odds ratio can be shown to be exactly that which would be obtained if one did estimate haplotype counts , assuming that the middle cell ( ) in Table 1 resolves into phased genotypes −/− or −/− with equal frequencies . The haplotype counts implicitly utilized by PLINK are therefore similar to what would be obtained from an EM algorithm , except that in PLINK the middle cell is resolved assuming no correlation between alleles at the two loci , resulting ( presumably ) in a set of estimated haplotype frequencies that will be biased towards showing lower levels of allelic association . We hypothesise that this bias towards lower levels of allelic association might partly account for the inferior performance of PLINK observed by Wu et al . [1] . Although the log odds ratio in PLINK corresponds to what would be obtained from attempting to resolve phase while assuming no correlation between alleles at the two loci , the variance estimate is based on counting independent alleles rather than haplotypes ( where is the total number of individuals in Table 1 ) . The formula for the variance estimate assumes that there are 3 independent cell probabilities in Table 2 . However , since the data in Table 2 was originally derived from Table 1 , considering these data as realisations from a multinomial distribution , we can see that in fact there should be 8 parameters corresponding to 8 independent cell probabilities . In Text S1 , we use the delta method to calculate the correct asymptotic variances of and , based on the multinomial data in Table 1 . We refer to the corresponding resulting test statistics as ‘adjusted’ fast-epistasis statistics: ( 12 ) ( 13 ) where now relates to the correct asymptotic variance of as given in Text S1 , and and again refer to quantities calculated within the sample of cases and controls respectively . Both the methods of Wu et al . [1] and the fast-epistasis tests implemented in PLINK operate by turning a question about statistical interaction into a question about allelic association ( or correlation ) , namely , whether association between alleles two loci exists ( case-only test ) or is different between cases and controls ( case/control test ) ) . However , many different measures of allelic association ( usually calculated for linked loci , and thus assumed to reflect LD ) have been proposed . Arguably the most popular are Lewontin's [31] and Pearson's product-moment correlation coefficient ( or the square of it , ) [32] . In most current genetic applications , these measures are calculated based on known or estimated haplotype frequencies . Wellek and Ziegler [33] pointed out that one advantage of is that it may be calculated without estimating phase , simply by applying it to two variables , and , coded ( 0 , 1 , 2 ) according to the number of susceptibility alleles possessed at each locus . Wellek and Ziegler [33] examined the performance of as a measure of LD using either estimated ( phased ) haplotype frequencies or using unresolved genotype data and showed that , if HWE holds , the loss of precision for estimating was negligible when using unphased genotypes rather than ( phased ) haplotypes . If HWE does not hold , Wellek and Ziegler found the genotype-based estimator of to be unbiased but the haplotype-based estimator to be strongly biased , i . e . to not reflect the ‘true’ value of based on the true haplotype frequencies . This would seem an unappealing property of the haplotype-based estimator , if the goal is to accurately estimate the true level of allelic association ( or LD ) between two loci . However , if the purpose is rather to test for interaction ( via testing whether correlation between alleles two loci exists ( case-only test ) or is different between cases and controls ( case/control test ) ) , it is possible that such a bias could be advantageous in terms of improving power . Since the method of Wu et al . [1] relies on estimating ‘haplotypes’ within the sample of cases ( under a potentially incorrect HWE assumption ) , we hypothesise that the bias pointed out by [33] might also contribute to the superior performance observed by Wu et al . [1] for their approach compared to PLINK . Given a genotype-based estimator of , Wellek and Ziegler [33] propose using Fisher's transformation to calculate a quantityand its estimated variance . A natural pair of statistics for testing interaction based on might therefore be:In computer simulations ( data not shown ) , we found the performance of these statistics to be virtually identical to statistics based on the correlation coefficient itself . We therefore instead define our Wellek and Ziegler inspired statistics based on the correlation coefficient as: ( 14 ) ( 15 ) where again and refer to quantities calculated within the sample of cases and controls respectively . Formulae for the correlation coefficient and its estimated variance are given by Wellek and Ziegler [33] . Note that the test based on the difference in the correlation coefficients between cases and controls , , was also proposed by Kam-Thong et al . [34] and implemented in a program called EPIBLASTER . In EPIBLASTER , is used as a screening step , prior to performing a full logistic regression analysis on the subset of pairs of loci showing some loose level of significance with . Although designed to test specifically for ( statistical ) interaction , several of the test statistics proposed above can be shown to be sensitive to the situation where there is , in fact , no interaction , but one or both of the loci display main effects ( see details in Text S2 ) . This is rather unsatisfactory as , even if one of the loci does have a genuine main effect , this phenomenon could lead to potentially increased false positive rates with respect to detection of the other locus ( through its apparent – but false – interaction with the locus that has genuine main effects ) . Ideally , one would hope that detection of a significant interaction effect would indicate genuine interaction , but , even if this is not the case , one would at least hope that both loci identified have some involvement in disease ( with their precise joint effects - interactive or otherwise - being determinable through further , more focussed , statistical or biological investigation ) . In order to address this issue , we propose two new ‘joint effects’ tests that are sensitive only to either a ) a genuine interaction effect or b ) ( if the disease is not sufficiently rare ) , main effects present at both loci . Our tests are motivated by a desire to test the same interaction parameter as tested by Wu et al . [1] . However , unlike some previously-proposed tests , our new tests can be shown to have the advantage of not being sensitive to main effects at a single locus . Moreover , under a rare disease assumption , our new tests can also be shown to be insensitive to main effects at both loci , thus reflecting genuine interaction . Thus , application of our joint effects tests will not result in an inflated type 1 error rate with respect to the detection of loci that are not involved in the disease ( even though , for a common disease , our tests could potentially result in an inflated type 1 error with respect to whether the pair of loci actually interact , in the usual statistical sense ) . Our new tests are based on the counts in Table 1 , calculated separately within the sample of cases and controls . Consider using each of the four top left cells in Table 1 in turn , to estimate four odds ratios relative to the baseline ( bottom right ) cell:In Text S3 we show that , under a rare disease assumption , these estimated odds ratios can be considered as estimates of the following functions of , where refers to to the log odds ratio estimated in the method of Wu et al . [1]: ( 16 ) To construct our proposed tests , we therefore propose to use the four relationships in ( 16 ) as four estimating equations for , and test the hypothesis that ( case/only test ) or that is equal for cases and controls ( case/control test ) . Further motivation for our tests is provided in Text S3 . Note that corresponds to the situation where all four of the ‘interaction’ odds ratios ( , , , ) equal 1 . We construct two separate estimates of , using the data in Table 1 as tabulated for either cases or controls . Equation ( 16 ) implies that we can estimate via a weighted average:where relates to the estimate of obtained from Table 1 , and the weights are chosen to sum to 1 and make the variance of minimum ( see Text S1 for details ) . Having estimated and its variance ( see Text S1 ) separately using data from either cases or controls , we can then construct ‘joint effects’ tests: ( 17 ) ( 18 ) where again and refer to the quantities calculated within cases and controls respectively . A difficulty with estimation arises when . If this occurs , we replace the objective quantity bywhich reduces to zero if ( i . e . possesses the same desirable property under the null hypothesis ) . Writing in terms of , we obtain four estimating equations for instead of , and we estimate as:with optimal weights chosen to make the variance minimum as before . Estimating the variance of as , this results in alternate versions of our joint effects tests: ( 19 ) ( 20 ) where again and refer to quantities calculated within cases and controls respectively . Text S3 motivates our ‘joint effects’ tests through consideration of the relationship between the original Wu et al . [1] method and standard logistic regression . A natural question of interest is the relationship between the other two methods described here ( FE , WZ ) and standard regression approaches – and , in particular , to what extent the different odds ratios ( ) estimated by these methods correspond to the usual interaction parameters ( and , , , ) in Equations ( 1 ) and ( 2 ) . In Text S4 we show , for each of these methods , the relationship between the parameters estimated in that method and those estimated in standard logistic or linear regression . In addition , in Text S5 , we show that the WZ case-only statistic can be viewed equivalently as a score test with respect to the interaction parameter . It would be of interest to determine whether a similar relationship holds for the other statistics considered here . However , providing this derivation for the remaining statistics is beyond the scope of the current manuscript , and we defer it to future work . We performed computer simulations to evaluate the performance ( type 1 error and power ) of the various test statistics described above . For the Wu and adjusted Wu methods , haplotype frequencies in cases and controls were calculated from unphased genotype data using an EM algorithm as implemented in either PLINK or the R library ‘Genetics’ . The general structure of the disease models we considered is shown in Table 3 , assuming two loci G and H , each having two alleles and . We simulated 1000 cases and 1000 controls from a general population assumed to be in HWE . Writing the haplotype frequencies in the general population as ( − ) for , we considered the same two sets of haplotype frequencies considered by [1]: When the two SNPs were not in LD , we examined the performance of both case/control and case-only statistics . When the SNPs were simulated to be in LD , we examined only the performance of case/control statistics ( since we know that case-only statistics will show inflated type 1 error in this situation ) . To investigate type 1 error we considered 8 scenarios , each using 10 , 000 data replications . To investigate power we considered a further 4 scenarios , each using 1 , 000 data replications . The structure of the simulated models and the parameter values assumed are given in Tables 3 and 4 . Note that in Tables 3 and 4 we denote the baseline , main effect and interaction parameter values ( , , , in Equation ( 1 ) ) as ( , , , ) respectively . In each scenario apart from 5c and 5d , the baseline regression coefficient was chosen to equal , corresponding to a baseline penetrance of 2% . For Scenarios 5c and 5d we assumed a rarer disease , with baseline penetrance 0 . 0001 . For each power scenario , we increased from 0 ( no interaction ) to a value at which the power to detect an effect ( at significance level 0 . 01 ) was close to 100% for the best-performing statistics . In addition to the test statistics described above , when comparing power ( Scenarios 6–9 ) we also calculated several additional statistics . Firstly , as an ‘optimal’ test we considered analysing the data assuming the ‘correct’ model ( i . e . imposing the correct structure in terms of whether a model was assumed to be additive , dominant or recessive at each locus , see Table 4 ) . For case/control data this was achieved by using logistic regression with the correct coding of predictor variables at each locus , and then comparing models in which an interaction term was or was not included via a likelihood ratio test . For case-only data , the ‘optimal’ analysis was implemented by using the Wellek and Ziegler statistic ( 14 ) with the correct coding of predictor variables ( corresponding to an additive , dominant or recessive model ) at each locus . For comparison , we also considered ‘sub-optimal’ tests where an incorrect coding for the simulation model was used . Secondly , we considered an ‘ideal’ version of the Wu et al . statistics ( Equations ( 6 ) and ( 7 ) ) , in which we assumed haplotypes could be inferred without error . In this case , the formulae proposed by Wu et al . [1] should be correct , as there is no increase in the asymptotic variances used in the denominator due to phase uncertainty . Although not achievable in practice , for theoretical interest we investigated the performance of the Wu et al . statistics ( with respect to both type 1 error and power ) in this ‘ideal’ situation . To gain additional insight into the properties of the methods considered , for Scenario 7 we noted the ‘haplotype’ frequencies and resulting LD measures , and obtained from the EM algorithm applied ( separately ) to cases and controls ( as used in the Wu and adjusted Wu approaches ) . These were compared to the true haplotype frequencies and correlation measures ( as implied by the generating model ) , the genotype-based correlation coefficient ( as used in the Wellek and Ziegler inspired approaches ) , and the haplotype frequencies and correlation measures calculated from Table 2 ( which are , effectively , those used by PLINK ) . As an illustration of the methods described , we also applied them to real data from a publicly available genome-wide data set consisting of 1748 cases of Crohn's disease and 2938 population-based controls obtained from the Wellcome Trust Case Control Consortium ( WTCCC ) [2] . Since this exercise was purely for illustrative purposes , in the interests of time we limited our analysis to that of a single chromosome , chromosome 22 . We used the same quality control procedures as the WTCCC [2] to remove poor-quality SNPs and samples prior to analysis . This generated 5750 SNPs across chromosome 22 , resulting in 16 , 528 , 375 pairwise combinations to be tested for interaction .
Figure 1 shows quantile-quantile ( QQ ) plots of the distribution of the different test statististics calculated in Scenario 1 ( so under the global null of no effects at either locus ) . For a test that is performing correctly ( i . e . with well-calibrated type 1 error ) , we would expect to see all points lying on the line with slope equal to 1 . We find this to be true for all methods except the original Wu et al . [1] statistics ( Equations ( 6 ) and ( 7 ) ) , which show strong departure from the line , indicating a severe inflation in type 1 error . Figure 2 shows QQ plots for Scenario 2 in which locus G has a recessive main effect . Again the original Wu et al . [1] statistics show a severe inflation in type 1 error . A severe inflation is also seen for the Wellek and Ziegler inspired statistics and PLINK's fast-epistasis tests ( both the original and our adjusted version ) in case/control analysis , when the two SNPs considered are in LD in the general population . ( Some theoretical explanation for these results can be found in Text S2 ) . This inflation in the presence of LD is not seen for the ideal Wu statistics or for our new joint effects statistics . For case-only analysis , we see a small inflation in type 1 error for PLINK's fast-epistasis test , which is corrected through use of our adjusted version of this test . We also see a slight deflation in type 1 error ( indicating the method is conservative ) for our adjusted Wu statistic . A similar pattern is seen for Scenario 3 ( in which locus G has a dominant main effect , see Figure S1 ) except that , in this case , the Wellek and Ziegler inspired case/control statistic does not appear to show inflated type 1 error in the presence of LD , and , for case-only analysis , PLINK's fast-epistasis test shows a slight deflation ( rather than inflation ) in type 1 error , while our adjusted Wu statistic shows a slight inflation . Correct type 1 errors are achieved by the ideal Wu statistics and by our new joint effects statistics . Results from Scenario 4 ( in which locus G has an additive main effect ) are shown in Figure S2 . In this case , all methods appear to have correct type 1 error except the original Wu et al . [1] statistics and the Wellek and Ziegler inspired case/control statistic in the presence of LD . Figures S3 , Figure S4 , Figure 3 , and Figure 4 show the results from Scenarios 5a , 5b , 5c , 5d , in which both loci have main effects . Provided the disease is rare ( Figure 3 and Figure 4 ) , our joint effects statistics show correct type 1 error , while the adjusted fast-epistasis and Wellek and Ziegler methods can show inflated type 1 errors , particularly in the presence of LD . ( Some theoretical explanation for these results can be found in Text S2 ) . The Adjusted Wu method has generally correct type 1 error although it appears to be slightly conservative for case/only analysis in Figure 4 . When the disease is more common ( Figures S3 and S4 ) , the presence of main effects appears to have an impact on the type 1 error of virtually all methods , indicating that none are completely immune from detecting pairs of loci that are both involved in disease , but which do not , in fact , require any statistical interaction term to describe their action . The only method that appears immune to this problem is the ideal Wu statistic applied to case/control ( but not to case-only ) data . Figure 5 shows power curves for Scenario 6 ( RecessiveRecesssive model ) for all methods considered , including methods that assume ‘correct’ or ‘incorrect’ knowledge of the true structure of the underlying generating model . The left hand panels show results when there are no main effects , while the right hand panels show results in the presence of a main effect at locus G . We use solid lines to represent methods that have been shown ( Figure 1 , Figure 2 , Figure 3 , Figure 4; Figures S1 , S2 , S3 , S4 ) or would be expected on theoretical grounds to have correct type 1 error . We use dashed lines to represent methods that have been shown ( Figure 1 , Figure 2 , Figure 3 , Figure 4; Figures S1 , S2 , S3 , S4 ) to have incorrect type 1 error under the relevant generating model ( and whose ‘power’ should therefore be interpreted cautiously in the light of that fact ) . In all cases , we find that the highest power among methods that correctly control the type 1 error is seen for ‘optimal’ tests that impose the correct structure , while the lowest power is seen for ‘sub-optimal’ tests that impose the incorrect structure , as might be expected from standard statistical theory . Amongst the other tests , no method consistently outperforms the others; in some cases our joint effects test has highest power , in other cases the adjusted Wu or adjusted or original fast-epistasis tests perform best . The ideal Wu test ( in which we assume haplotypes can be estimated without uncertainty ) shows generally lower power than the other tests considered , in this scenario . Figure 6 shows power curves for Scenario 7 ( DominantDominant model ) . The original Wu statistic shows apparent highest power , but this observation is tempered by the fact that we know it has inflated type 1 error . Again , highest power among methods that correctly control the type 1 error is generally obtained for ‘optimal’ tests that impose the correct structure , although in some cases this power is closely matched by the adjusted Wu or joint effects tests . The original and adjusted fast-epistasis tests show low power when applied to case/control data . The ideal Wu test also shows generally low power when applied to either case/control or case-only data . Figure S5 shows power curves for Scenario 8 ( AdditiveAdditive model ) . Most methods perform fairly similarly , except for analysis under an incorrect model and the ideal Wu test , which both show lower power . For case/control data , in this scenario , the Wellek and Ziegler test slightly outperforms most other tests . Figure S6 shows power curves for Scenario 9 ( DominantAdditive model ) . Again we find that the highest power among methods that correctly control the type 1 error is seen for ‘optimal’ tests that impose the correct structure , while the lowest power is seen for either for the ideal Wu statistic , or for ‘sub-optimal’ tests that impose the incorrect structure . Amongst the other tests , no method consistently outperforms the others; in some cases the Wellek and Ziegler test shows highest power , whereas in other cases the joint effects or adjusted Wu statistics show highest power . Table 5 shows the true and estimated haplotype frequencies and correlation measures , as used by several different methods , under one particular setting for simulation Scenario 7 . When data is simulated without LD between the loci , we see that , in controls , both the EM algorithm ( as used in the Wu et al . and adjusted Wu methods ) and the allele counting algorithm ( used in PLINK's fast-epistasis method ) give very similar results with respect to estimated haplotype frequencies and resulting correlation measures . The correlation measures ( along with the Wellek and Ziegler genotype-based correlation coefficient ) are correctly estimated as being close to 0 . The slight departure from 0 results from the fact that the disease is not particularly rare , and so the presence of an interaction effect will cause unaffected controls , as well as cases , to show some slight correlation between alleles at the two loci . In cases ( with no LD ) however , the story is very different . All methods show correlation between alleles at the two loci , however the haplotype frequencies and resulting correlation measures estimated using PLINK's allele counting algorithm seem to be much closer to the true generating values . The EM algorithm ( as used in the Wu et al . and adjusted Wu methods ) produces upwardly biased estimates , presumably because of the incorrect ( within cases ) HWE assumption made . This results in much higher apparent correlation , which could plausibly increase power when testing whether correlation between alleles two loci exists ( case-only test ) or is different between cases and controls ( case/control test ) ) . However , the power of any given test will depend not just on the level of apparent correlation , but also on the estimated variance of the correlation measure used , and our results overall suggest that the bias induced by the incorrect HWE asssumption does not necessarily always translate to a substantially improved power . In the presence of LD , for controls the EM algorithm ( as used in the Wu et al . and adjusted Wu methods ) appears to better capture the true haplotype frequencies and resulting correlation measures , while the PLINK's allele counting algorithm produces results that are biased downwards ( i . e . towards showing lower levels of correlation ) . For cases , PLINK's allele counting algorithm produces correlation measures that are biased downwards from the true values , while the EM algorithm produces correlation measures that biased upwards . Given that any analysis in the presence of LD needs to be based on the difference in correlations between cases and controls , it is unclear to what extent these biases will operate to improve power for one method over another , although the results shown in Figure 6 suggest that these bias may partly account for the high power of the adjusted Wu methods in that scenario . Figure S7 shows the results from applying the different methods to 5750 SNPs across chromosome 22 genotyped in the WTCCC Crohn's disease dataset . Since SNPs on the same chromosome are likely to be in LD , we limited our analysis to the case/control version of all statistics considered . Given the large number of potential tests performed ( 16 , 528 , 375 pairwise combinations ) , for the joint effects , fast-epistasis and Wellek and Ziegler inspired methods , we only output results passing a value threshold of 0 . 001 ( although note that , for the fast-epistasis statistic , PLINK in fact only performed a total of 13 , 818 , 410 tests that passed its validity criteria ) . The QQ plots ( Figure S7 ) show that the joint effects , fast-epistasis and Wellek and Ziegler inspired statistics all follow the expected distribution under the null hypothesis , even in this tail ( ) of the distribution . We also noted that , for these three methods , the proportion of tests falling into this tail was , as expected ( data not shown ) . The most computationally efficient implementation was PLINK , which took approximately 20 minutes to perform 13 , 818 , 410 tests . The Wellek and Ziegler and joint effects methods were considerably slower , each taking 20 hours ( on the same computer system ) to perform 16 , 528 , 375 tests . We implemented the Wellek and Ziegler and joint effects statistics through code written by ourselves in R , and so these times could be considerably reduced by re-writing the code ( e . g . in C++ ) and making use of mechanisms for efficient binary data storage . The original and adjusted Wu methods were prohibitively slow to calculate for all 16 , 528 , 375 pairwise combinations , most likely because of the requirement of these methods to estimate haplotype frequencies from unphased genotype data ( e . g . via an EM algorithm ) . ( We implemented these methods through code written by ourselves in R; calculation might be achievable in reasonable time through use of more efficient programming in C++ , binary data storage and parallel execution on a computer cluster ) . Figure S7 therefore shows the results for the original and adjusted Wu methods for a subset of 10813 SNP pairs consisting of the first and the thousandth SNP , each paired with all others . Even in this reduced data set , we can see that the adjusted Wu statistic follows the expected distribution under the null hypothesis while the original Wu statistic shows an inflated distribution , in line with the results we found in our computer simulations . The results in Figure S7 do not provide any strong evidence for the existence of interactions between SNPs on chromosome 22 in the WTCCC Crohn's data . However , it is of interest to see to what extent the different methods implicate the same ‘top SNP pairs’ . Figure S8 plots the observed test statistics for the joint effects , fast-epistasis and Wellek and Ziegler inspired statistics against one another . The results from these three methods are seen to be broadly correlated , with the same SNP pairs tending to fall at the extreme of the distribution , regardless of which method is used . Since we were unable to calculate the Wu and adjusted Wu statistics for all pairs of SNPs , at the suggestion of a reviewer , we used another approach for calculating these statistics , which we hoped would be computationally quicker . We used a phasing algorithm to infer haplotypes across chromosome 22 , for each individual . We carried out this step using the program SHAPEIT [35] , which has the advantage of outputting for each individual not just a single “most likely” haplotype configuration , but additionally allows one to store the uncertainty and sample a set of haplotype configurations . We sampled 100 replicate haplotype configurations for each individual . Since the idea of the Wu method is to compare ‘apparent LD’ within cases to that within controls , we initially carried out the phasing in case and control groups separately , although we later compared our results to those obtained when phasing the cases and controls together . Having generated 100 replicates of phased haplotypes , we then calculated , for each pair of SNPs , the mean ( over the 100 replicates ) haplotype frequencies in cases and controls . ( The haplotype frequencies within each replicate were calculated simply by counting resolved case and control haplotypes ) . We used these mean haplotype frequencies in the formulae for the Wu and adjusted Wu statistics ( Equations 7 and 9 respectively ) . Note that these formulae were derived on the basis of sampling theory under the assumption of a certain number of observed haplotypes , and it is unclear whether the same theoretical arguments should apply to haplotype frequencies that have been estimated in a different way . In particular , SHAPEIT uses a hidden Markov model that is motivated by population genetics principles , resulting in a greater borrowing of information across SNPs and individuals than is used in the other approaches . This fact , together with the fact we averaged ( over 100 replicates ) , suggests that the haplotype frequencies ( and thus and ) estimated from SHAPEIT may be more accurate and less variable than those estimated in the other approaches , thus requiring a smaller variance in the denominator of the test statistic . To address this issue , we used an additional strategy of calculating the variance directly from the 100 replicates . Within each replicate , we calculated the haplotype frequencies and log odds ratios and . We then calculated the sample mean and variance of and over the 100 replicates and constructed a ‘SHAPEIT variance-based Wu ( SVBW ) test statistic’: Figure S9 shows QQ plots for the Wu and adjusted Wu test statistics ( Equations 7 and 9 ) applied to the mean estimated haplotype frequencies from SHAPEIT , for the subset of 10813 SNP pairs consisting of the first and the thousandth SNP , each paired with all others . The test statistics ( shown in red and black ) are seen to be considerably deflated in comparison to the expected distribution , suggesting that the variance of the SHAPEIT-derived haplotype frequencies is indeed considerably lower than that implied by Equations 7 and 9 . We noticed , however , that the test statistics appeared to be approximately half their expected value . We therefore constructed an alternative ‘SHAPEIT mean-based Wu ( SMBW ) test statistic’: ( 21 ) which can be seen to be equivalent to the original Wu case/control statistic , but under the assumption of double the number of haplotypes . The SMBW test statistics ( shown in green ) are seen to closely follow the expected distribution , suggesting that variance of the SHAPEIT-derived haplotype frequencies is indeed equivalent to what would be obtained from observing twice the number of haplotypes . We consulted the description of the algorithm used by SHAPEIT [35] and noticed that it involves an iterative procedure of updating an individual's current haplotype configuration by sampling haplotypes from a set of currently resolved haplotypes ( for the other individuals in the data set ) , in such a way that recombination and mutation events are allowed for . This means that , for SNPs close together , the sampling procedure would effectively be sampling alleles from currently-resolved haplotypes ( where is the number of individuals in the data set ) while for SNPs that are far apart , a recombination event is virtually guarranteed and so the sampling procedure is effectively sampling from haplotypes constructed by sampling the alleles at each SNP independently . Since the majority of our pairwise tests involve SNPs that are far apart , the majority of the tests will indeed closely correspond to effectively observing haplotypes . Note that this argument is quite similar to the argument that could be used to justify the construction of PLINK's fast-epistasis statistic on the basis of alleles . Logically , one would expect that the variance could be estimated even better by allowing for the actual recombination distance between each pair of SNPs , so that SNPs that are closer together are considered to have a probability of undergoing a recombination and thus being sampled from haplotypes , and probability of not undergoing a recombination and thus being sampled from haplotypes . ( For definition of , see [35] ) . However , we found implementation of this approach resulted in test statistics that did not follow the expected ( on 1 df ) distribution quite as well as simply assuming or haplotypes ( data not shown ) . One possible explanation is that the iterative nature of the SHAPEIT algorithm means that even SNPs that lie close together are likely to be subject to a recombination event at some point during the procedure , generating closer to effective haplotypes . Further work , beyond the scope of this paper , would be required to follow up the explanation for these observations in more detail . Figure S10 ( Panel ( a ) ) shows the QQ plot for the SHAPEIT variance-based Wu ( SVBW ) test statistic , for the subset of 10813 SNP pairs consisting of the first and the thousandth SNP , each paired with all others . Although the majority of the points do lie on the expected line , there are a number of outliers . We noticed that the most severe outliers corresponded to pairs of SNPs that lie within 1 cM of one another ( shown in red ) , suggesting that the variance of the haplotype frequencies within short regions may perhaps be under-estimated by the SHAPEIT algorithm . ( Another explanation is that these are true interactions and/or haplotype effects , however this seems a little unlikely given that they are not identified by any other method ) . We removed all pairs of SNPs that lie within 1 cM of one another from both the SVBW and SMBW results , which resulted in test statistics that followed the expected distribution more closely ( Panels ( b ) and ( c ) ) . Panel ( d ) shows a comparison between the resulting SMBW and SVBW test statistics , showing how extremely similar they are . Figure S11 Panels ( a ) and ( b ) show a comparison between the SMBW and SVBW statistics and the AWu statistic , while Panels ( c ) and ( d ) show a comparison between the SMBW and SVBW statistics and the JE statistic . Although these different test statistics are by no means identical , they are seen to be broadly correlated , as expected . Figure S12 shows a comparison of the SMBW ( left hand panels ) and SVBW ( right hand panels ) results from haplotypes estimated by applying SHAPEIT to cases and controls separately ( y axes ) or together ( x axes ) . Points marked in red on the top panels correspond to SNP pairs where the SNPs are less than 1 cM apart; these pairs are seen to generate outliers for the SVBW test regardless of whether SHAPEIT is applied to cases and controls separately or together . These outliers do not occur with the SMBW test when SHAPEIT is applied to cases and controls together . The bottom panels repeat these plots , but with SNP pairs where the SNPs are less than 1 cM apart removed . Overall the results from applying SHAPEIT to cases and controls separately ( y axes ) or together ( x axes ) are seen to be highly correlated , particularly for the SMBW test . We investigated the outliers ( where the results were very different according to whether cases and controls were phased separately or together ) and noticed that the vast majority of these corresponded to SNPs whose minor allele frequency is close to 0 . 5 , and for which there had been a swap with respect to which allele was designated as the minor allele between the case and control groups , when phased separately . This resulted in an incorrect matching of haplotypes between case and control groups , resulting in an incorrect test statistic . ( Interestingly , the 7 outliers for which the test statistic is close to 0 when the cases and controls were phased separately are also seen as outliers when compared to the AWu and JE tests ( Figure S11 ) , indicating that the results from SHAPEIT applied to cases and controls together are concordant with the AWu and JE results ) . We found that the allele swap problem had occurred in 46 out of the 5750 SNPs considered i . e . just under 1% of the results presented from applying SHAPEIT to cases and controls separately were incorrect . This might suggest that the strategy of phasing cases and controls together is more reliable , although in practice one could avoid this problem when phasing cases and controls separately by performing a more careful check at the analysis stage . Intuitively , one might expect that the strategy of phasing cases and controls separately might be more powerful when constructing tests that are based on haplotype differences between cases and controls , but a detailed comparison of the relative power of these two approaches would required further investigation . Although the SHAPEIT approaches appear to result in more accurate haplotype estimation than the EM algorithm-based Wu and AWu approaches , generating haplotype frequency estimates that can ( with care ) be translated into Wu-like interaction tests , in our hands , implementation of these approaches was not computationally faster than the original Wu and AWu methods . Although generation of 100 replicates of phased chromosome 22 haplotypes in SHAPEIT was relatively fast ( taking around 28 hours on our system ) , our program for generating the resulting SMBW and SVBW test statistics ended up taking about 3 seconds per SNP pair . ( For each SNP pair we needed to read in – or store in memory – 100 replicates of phased haplotypes for each individual , in order to pick out the required alleles at the two SNPs , and then calculate haplotype frequencies , and , within each replicate , followed by the mean and variance of these quantities across replicates ) . No doubt more efficient programming , binary data storage and implementation on a computer cluster could considerably speed up this procedure . Given the close correspondence between the SMBW and SVBW tests , together with the better performance of SMBW for SNPs that lie close together , a natural first step might be to initially focus on SMBW alone , for which and within each replicate , and all variances across replicates , would not need to be calculated .
Here we have investigated , through theoretical derivation , computer simulations and a real data example , the properties of several previously-proposed statistics for performing genome-wide interaction analysis using case/control or case-only data [1] , [30] , together with a number of alternative statistics proposed by ourselves and others [33] , [34] . Our main finding is that the statistics proposed by Wu and colleagues [1] show substantially increased type 1 error due to the incorrect variance estimates used ( Equations ( 6 ) and ( 7 ) ) which do not account for the uncertainty induced when estimating phased haplotype frequencies from unphased genotype data . This inflation in type 1 error can be corrected by using a variance estimate that accounts for this uncertainty , as in our adjusted Wu statistics . All other methods investigated appear to show adequate control of type 1 error under the null hypothesis of no genetic effects ( main effects or interactions ) , although several methods ( including the fast-epistasis method implemented in PLINK [30] and the Wellek and Ziegler method [33] , [34] ) can show increased type 1 error when there is a main effect at one or both loci , particularly if there is also LD . Only the ideal Wu method and our new joint effects statistics achieve consistent control of type 1 error in the presence of a main effect at just one of the loci . In terms of power , comparison of the different methods is somewhat complicated by the fact that several of the methods show increased type 1 error in different circumstances . However , even when comparing methods that control the type 1 error rate in a given situation , no method consistently outperforms all others . Generally high power over a range of scenarios is exhibited by the Wellek and Ziegler statistics [33] , [34] and by our new joint effects statistics and adjusted Wu statistics . Given that , out of these options , only the joint effects statistics achieve adequate control of type 1 error in the presence of a single main effect , this might suggest that the joint effects tests would be the overall preferred option . Although the ideal Wu method also shows adequate control of type 1 error in the presence of a single main effect , observation of known haplotypes , as required by this method , is unachievable in practice . Even if it were achievable , e . g . through experimental assays that allow determination of haplotypes , or through the use of larger numbers of markers to help infer phase between the two SNPs in question , Figure 5 and Figure 6 and Figures S5 and S6 show that the power achieved by the ideal Wu approach is generally lower than for other approaches . This slightly counter-intuitive result might be due to the fact that the ideal Wu method is not affected by the bias that results from incorrectly assuming HWE when estimating haplotype frequencies in cases , a bias that can potentially increase power . Somewhat surprisingly , many of our results appear to contradict results presented by Wu and colleagues [1] who found in simulations ( using similar generating models to those considered here ) and application to real data that their method gave adequate control of type 1 error and higher power than competing methods ( including logistic regression analysis under the correct model ) . We have been unable to fully determine the reason for these discrepencies , even after discussion with the authors of [1] , although our discussions have highlighted some possible explanations . With respect to the simulation results , our current understanding is that the simulations performed by Wu et al . did not , in fact , include any consideration of haplotype uncertainty ( their simulations simply assumed haplotypes could be observed without error – as , in a simulated data set , they can ) . This explains the apparently correct type 1 error observed by Wu et al . but it means that all their simulations ( of both type 1 error and power ) are highly misleading with respect to illustrating how their method might perform in practice ( where haplotype uncertainty will invariably exist , particularly at loci that are not in strong LD ) . It also does not explain the difference in power we see compared to Wu et al . when we also assume haplotypes can be observed without error ( our ‘ideal’ Wu statistics ) . We speculate that one possible explanation for this difference might be that Wu et al . assumed in their simulations that haplotypes come together independently in cases ( which is true under a multiplicative haplotype model [28] , but not under recessive or dominant models ) . It is unclear what effect such an erroneous assumption would have on the power of the different methods , but it might possbly explain why Wu et al . found their method to give consistently higher power than logistic regression analysis under a correct model , whereas we find ( as might be expected from statistical theory ) that logistic regression analysis under a correct model gives generally higher power than the adjusted or ideal Wu statistics . The explanation of these simulation discrepencies also does not explain why Wu et al . found correct ( or possibly slightly deflated ) type 1 error in analysis of real data ( see QQ plot shown in Figure 1 of Wu et al . ( 2010 ) [1] ) , whereas in our own application of the original Wu et al . ( 2010 ) method to real data ( Figure S7 ) , we found the same general inflation of test statistics as we observe in computer simulations . One possibility is that Wu et al . inadvertedly divided by a factor of two when using their formulae ( our Equations ( 6 ) and ( 7 ) ) to calculate the desired test statistics . This would result in a test that would approximately correspond to our adjusted Wu statistic . In any case , unless or until these issues can be resolved , we recommend use of our new joint effects or adjusted statistics , and urge caution when using Wu et al . 's [1] originally-proposed statistics , on account of the inflated error rate that can result . We have focussed in this communication on methods that test for interaction per se i . e . that test ( or attempt to test ) the interaction term in a linear model ( such as Equation 1 ) . As mentioned previously , if one prefers to test combinations of terms ( e . g . in order to implement tests of association allowing for interaction [20] , [17] ) one may do so by combining a test of the interaction term with some test of the other terms [22] . It is well-known ( and indeed can be seen from Figure 5 and Figure 6 and Figures S5 and S6 ) that case-only tests are more powerful than case/control tests for testing interaction , provided there is no population-level correlation between the two variables being tested . Although such an assumption should in principal be reasonable when testing genetic variants that are located sufficiently far apart as to be expected not to show LD , in practice GWAS data often does display long-range allelic association [17] , possibly due to population structure [26] or other confounding influences . This suggests that , in application to GWAS data , the case/control versions of the statistics described here might be preferred over the case-only versions , in spite of their lower power . Alternatively , construction of weighted combinations of the case-only and case/control statistics [36] ) might prove a more powerful approach . Several authors have recently proposed the use of retrospective likelihoods [37] , [26] that can increase power by exploiting an assumption of gene-gene independence in the underlying population ( or in controls , if the disease is rare or controls unselected ) . These methods have been used , for example , in a conditional search exercise exploiting known loci for prostate cancer in a multi-stage GWAS [38] . The advantage of these frameworks is that they allow the incorporation of covariates ( such as principal components scores ) to account for population stratification , as well as allowing a wider class of tests . Since the methods described here can all be formulated in terms of ( prospective ) linear or logistic regression models ( see Text S3 and S4 ) , in theory such approaches could be applied to the tests described here . However , an advantage of the current formulations is that closed-form expressions for the tests are available , which makes them attractive when carrying out all pairwise interaction scans in GWAS , on account of the fact that the tests are rapidly computed . R code for implementing the joint effects , Wellek and Ziegler and adjusted Wu statistics described in this manuscript is available on request from the authors .
|
Gene–gene interactions are a topic of great interest to geneticists carrying out studies of how genetic factors influence the development of common , complex diseases . Genes that interact may not only make important biological contributions to underlying disease processes , but also be more difficult to detect when using standard statistical methods in which we examine the effects of genetic factors one at a time . Recently a method was proposed by Wu and colleagues [1] for detecting pairwise interactions when carrying out genome-wide association studies ( in which a large number of genetic variants across the genome are examined ) . Wu and colleagues carried out theoretical work and computer simulations that suggested their method outperformed other previously proposed approaches for detecting interactions . Here we show that , in fact , the method proposed by Wu and colleagues can result in an over-preponderence of false postive findings . We propose an adjusted version of their method that reduces the false positive rate while maintaining high power . We also propose a new method for detecting pairs of genetic effects that shows similarly high power but has some conceptual advantages over both Wu's method and also other previously proposed approaches .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"public",
"health",
"and",
"epidemiology",
"computer",
"science",
"mathematical",
"computing",
"mathematics",
"epidemiology",
"statistics",
"genetics",
"biology",
"computational",
"biology",
"computerized",
"simulations",
"genetics",
"and",
"genomics",
"numerical",
"analysis"
] |
2012
|
Improved Statistics for Genome-Wide Interaction Analysis
|
Salmonella enterica serovar Weltevreden ( S . Weltevreden ) is an emerging cause of diarrheal and invasive disease in humans residing in tropical regions . Despite the regional and international emergence of this Salmonella serovar , relatively little is known about its genetic diversity , genomics or virulence potential in model systems . Here we used whole genome sequencing and bioinformatics analyses to define the phylogenetic structure of a diverse global selection of S . Weltevreden . Phylogenetic analysis of more than 100 isolates demonstrated that the population of S . Weltevreden can be segregated into two main phylogenetic clusters , one associated predominantly with continental Southeast Asia and the other more internationally dispersed . Subcluster analysis suggested the local evolution of S . Weltevreden within specific geographical regions . Four of the isolates were sequenced using long read sequencing to produce high quality reference genomes . Phenotypic analysis in Hep-2 cells and in a murine infection model indicated that S . Weltevreden were significantly attenuated in these models compared to the classical S . Typhimurium reference strain SL1344 . Our work outlines novel insights into this important emerging pathogen and provides a baseline understanding for future research studies .
Salmonella enterica is a globally distributed Gram-negative enteric bacterial species that is responsible for significant levels of morbidity and mortality in both humans and animals [1] . S . enterica is currently classified into six subspecies and >2 , 500 serovars on the basis of the White- Kauffmann-Le Minor scheme that exploits specific typing sera against O ( lipopolysaccharide ) and H ( flagella ) antigens [2] . S . enterica organisms can also be loosely assigned into the so called typhoidal or non-typhoidal Salmonella ( NTS ) serovars . Typhoidal Salmonella are typically adapted to cause systemic disease in humans e . g . Salmonella enterica serovar Typhi ( S . Typhi ) the cause of the typhoid fever [3 , 4] . In contrast , NTS are more frequently , but not exclusively , associated with localized gastroenteritis and are more promiscuous and zoonotic than typhoidal serovars , and can infect multiple hosts e . g . S . Typhimurium [5 , 6] . Genetically , S . enterica is regarded as a broad and ancient species; different serovars/isolates can vary by 100 , 000s of single nucleotide polymorphisms ( SNPs ) and contain a variable array of genomic islands and prophages [7 , 8] . Further , many Salmonella serovars are monomorphic or clonal [6 , 9] , however , there can be significant genetic diversity within a specific serovar e . g . S . Typhimurium [10] and generally studies on genetic diversity within a particular serovar are scarce , particularly using newer high-throughput approaches such as whole genome sequencing . Most human NTS infections are commonly regarded as having a zoonotic source , at least in developed countries [11] . Some serovars , such as S . Cholerasuis , are predominantly associated with a particular mammalian host ( pigs in the case of S . Cholerasuis ) whereas others are more promiscuous . Indeed , the sources of many outbreaks of human salmonellosis are difficult to trace and often remain unidentified . In cases where the source is known , these have commonly been associated with raw or undercooked meat , eggs , vegetables or contamination has occurring during processing [12 , 13] . Notably , the frequency of outbreaks associated with particular serovars can vary in terms of source , incidence and geographical distribution . Salmonella enterica serovar Weltevreden ( S . Weltevreden ) is being increasingly reported as a cause of diarrheal disease in humans [14 , 15] . This little described serovar is emerging as a significant food-borne pathogen in Asia where it is reported to be associated with fish or aquatic food production systems [14] . S . Weltevreden has also been reported in Europe; an outbreak of diarrhea in Scandinavia was linked to alfalfa sprouts contaminated with S . Weltevreden [16] . Whole genome sequencing ( WGS ) has been used to define structure within various populations of bacterial pathogens , including some of the key serovars of S . enterica [17 , 18] . Genome sequences of four S . Weltevreden isolates have been previously decoded [19 , 20] , but information regarding their phylogenic structure is limited . Here , we report a phylogenetic analysis of a global collection of S . Weltevreden; additionally providing complete reference genomes to facilitate future analysis . Furthermore , aiming to assess their pathogenic potential in comparison to other Salmonella we have phenotyped selected S . Weltevreden isolates in order to establish the value of particular experimental models .
The 115 strains contributing to this study , their origins and their resulting accession numbers are described in S1 Table . All DNA samples were extracted using the wizard genomic DNA extraction kit ( Promega , USA ) and sequenced using the Illumina HiSeq 2500 platform . A reference genome of S . Weltevreden 10259 was generated using both Illumina and PacBio sequencing technologies . The assembly was built initially using the PacBio sequence reads then manually assembled . Raw sequencing reads of each sample were run through a Kraken database [21] ( 0 . 10 . 6 ) for taxonomic identification . All of the Illumina sequences were aligned to the complete reference genome generated from S . Weltevreden 10259 . The reads , in FASTQ format , were first split into groups containing 1 , 000 , 000 reads . Each group of reads was individually aligned using SMALT ( https://www . sanger . ac . uk/resources/software/smalt/ ) ( 0 . 7 . 4 ) . Aligned reads were merged using SAMtools [22] ( version 0 . 1 . 19 ) , coordinate sorted , and outputted as a BAM file . Optical duplicates were identified using Picard ( http://broadinstitute . github . io/picard/ ) ( 1 . 9 . 2 ) . Statistics regarding each mapping were generated using BamCheck [22] including read coverage of the reference genome , reads aligned , perfect pairs , unmapped reads and actual insert size . The resulting data was evaluated manually to identify poor quality sequencing data . Illumina-generated sequences were assembled using a pipeline ( https://github . com/sanger-pathogens/vr-codebase ) developed at the Wellcome Trust Sanger Institute . For each genome , the de novo short-read assembler Velvet [23] ( 1 . 2 . 09 ) was used to generate multiple assemblies by varying the k-mer size between 66% and 90% of the read length using Velvet Optimiser ( https://github . com/tseemann/VelvetOptimiser ) . From these , the assembly with the highest N50 was chosen . Contigs were excluded from the assembly if they were shorter than the target fragment size ( 400 bases ) . A scaffold assembly of the contigs was built by iteratively running SSPACE [24] ( version 2 . 0 ) beginning with the contigs which were predicted to map next to each other . The reads were then mapped again to the scaffold assembly and perfect pairs were excluded . Next , gaps identified as one or more N’s , were targeted for closure by running 120 iterations of GapFiller [25] ( version 1 . 11 ) , using a decreasing read evidence threshold . Finally , the reads were aligned back to the improved assembly using SMALT ( https://www . sanger . ac . uk/resources/software/smalt/ ) and a set of statistics was produced for assessing the quality of the assembly . PacBio raw read data for each sample was manually assembled using the PacBio SMRT analysis pipeline ( https://github . com/PacificBiosciences/SMRT-Analysis/ ) ( 2 . 2 ) . The raw unfinished assemblies all produced a single non-circular chromosome plus some other small contigs , some of which were plasmids or unresolved assembly variants . If the ends of a contig overlapped , they were identified as candidates for circularization using a protocol recommended by PacBio ( https://github . com/PacificBiosciences/Bioinformatics-Training/wiki/Circularizing-and-trimming ) . A virtual break was manually introduced into the chromosome sequence at the thrA gene , to match the starting point of other published S . enterica references . Plasmids were also artificially broken at the replication gene . The sequences were then circularized using the genome assembler , Minimus [26] ( version 2 part of AMOS version 3 . 1 ) , which removed the overlapping sequence . Quiver was then used by the circularized sequence and the raw reads to correct errors in the circularized region . As high quality short read data from Illumina were available , ICORN2 ( version 0 . 97 ) was used to correct minor errors in the assembly , providing a very high quality reference sequence , as assessed by REAPR [27] assembly was subsequently annotated with Prokka [28] . All sequences and assemblies are freely available; accession numbers are provided in S1 Table . SNPs were called on each set of aligned reads using mpileup with the parameters ‘samtools mpileup -d 1000 -DSugBf ref bam’ . The raw SNPs were then passed into BCFtools and were filtered into a higher quality set . A virtual pseudo-genome was then constructed by substituting the base call at each site ( variant and non-variant ) into the reference genome . For a SNP to be called the depth had to be greater than 4 reads , and be present on both strands , with at least 75% of reads containing the SNP at that position . The mapping quality had to be greater than 30 ( less than 1 in 1000 probability that the mapping was incorrect ) . If a SNP failed to meet these criteria it is substituted with an ‘N’ . Insertions with respect to the reference genome were ignored . Deletions with respect to the reference genome were filled up with ‘N’ characters in the pseudo-genome in order to keep it aligned and at the same length relative to the reference genome . Heterozygous sites were turned into homozygous alleles by selecting the first allele in the BCF file . However , if the first allele was an insertion or deletion ( indel ) , the second allele in the BCF file was taken . If the second allele was also an indel , a single ‘N’ character was used . All of the pseudo-genomes were then merged into a single multi-FASTA alignment file , including the reference sequence . Since the median fragment size for the isolates sequenced on Illumina was 400 bases , the reference genome was blasted against itself ( 2 . 2 . 31 ) in order to identify repeats larger than 400 bases . All bases falling within these regions were replaced with ‘N’ in the multi-FASTA alignment file and therefore not included in the analysis . The filtered multi-FASTA alignment was then checked for recombination using Gubbins [29] ( 1 . 3 . 4 ) . Five iterations of Gubbins were run and in each iteration a phylogenetic tree was constructed with RAxML [30] with the GAMMA GTR model , and internal ancestral sequences were inferred using FastML [31] ( version 3 . 1 ) . Recombinant sequences were detected and a multi-FASTA alignment with the recombinant regions was masked out . This data was then used as the input to the next iteration . RAxML with 100 bootstraps was then run over the final multi-FASTA alignment to provide a high quality phylogenetic tree in newick format . The population structure of the phylogenetic tree was validated using a Bayesian statistical approach . Hierarchical BAPS [32] ( version 6 . 0 ) was used to perform a hierarchical clustering of the multi-FASTA alignment to reveal a nested genetic population structure . Once the clusters were identified , SNPs , which uniquely defined each of the clusters ( in 100% of isolates in a cluster ) were extracted using BioPericles ( https://github . com/sanger-pathogens/BioPericles ) ( version 0 . 1 . 0 ) . Exploiting the multi-FASTA alignment with recombination removed , a consensus sequence was generated for each cluster and any bases which varied or contained missing data were replaced by ‘N’ . The consensus sequences were merged into a single multi-FASTA alignment file and SNP locations were identified using snp sites ( https://github . com/sanger-pathogens/snp_sites ) ( version 2 . 0 . 1 ) . Each SNP was then annotated using the reference annotation ( 10259 ) GFF3 file . An annotated VCF file was produced with VEP syntax [33] listing the type of change ( intergenic/ synonymous/ nonsynonymous ) , the amino acid ( before and after ) and the amino acid position in the gene , along with the coordinates of each SNP relative to the reference genome , the reference base , the allele base and the presence and absence of the variant in each cluster . These cluster defining SNPs were then further annotated with the functional annotation of the gene they occurred in . Antimicrobial resistance was predicted from each sample’s raw sequencing reads using ARIBA [34] ( version 0 . 4 . 1 ) , which performs antibiotic resistance identification by assembly and alignment . A manually curated input database of known resistance genes in FASTA format was used as input along with the paired end sequencing reads in FASTQ format . The resistance gene sequences were first clustered using CD-hit [35] ( 4 . 6 ) . The raw reads were then aligned to a representative sequence for each resistance cluster . Reads which mapped and their complimentary strand equivalents were extracted . A local assembly was performed on the reads for each cluster ( version 3 . 5 ) , where the resistance genes for the cluster were used as ‘untrusted contigs’ . This generates a candidate gene along with sequence on either side if the gene is present in the reads . MUMmer [36] ( 3 . 23 ) was then used to identify differences between the assembled contig and the known resistance gene and the results were reported along with any variation found and quality flags . These were manually inspected and samples with 100% matches to resistance genes and with a complete open reading frame were flagged as being potentially candidates for visual inspection . A pan genome was constructed using Roary [37] ( version 3 . 2 . 5 ) from the annotated assemblies of the sample set with a percentage protein identity of 95% . The protein sequences were first extracted and iteratively pre-clustered with cd-hit ( version 4 . 6 ) down to 98% identity . An all against all blast ( version 2 . 2 . 31 ) was performed on the remaining non-clustered sequences and a single representative sequence from each cd-hit cluster was selected . The data were used by MCL [38] ( version 11–294 ) to cluster the sequences . The preclusters and the MCL clusters were merged and paralogs were split by inspecting the conserved gene neighborhood around each sequence ( 5 genes on either side ) . Each sequence for each cluster was independently aligned using PRANK [39] ( version 0 . 140603 ) and combined to form a multi-FASTA alignment of the core genes . To facilitate the analysis of invasion assays , S . Typhimurium SL1344 and S . Weltevreden C2346 , 10259 , 98_11262 and 99_3134 were transformed with the plasmid pSsaG that directs the expression of GFP from the ssaG promoter [40] . Hep-2 cells were cultured in Glasgow’s minimal essential medium ( GMEM , Sigma ) supplemented with 2 mM L-Glutamate and 10% ( volume/volume ) heat- inactivated fetal bovine serum ( FBS ) . Cells were seeded into 24-well plates ( 105 cells per well ) and cultured overnight . Salmonella were initially cultured at 37°C with agitation ( 250 rpm ) in 5ml LB broth for 4 . 5 hours . An aliquot was then diluted 1:50 in LB broth and grown at 37°C overnight as a static culture to optimize Salmonella pathogenicity island 1 ( SPI1 ) gene expression . For infection , the bacterial cultures were re-suspended in fresh GMEM media supplemented with 2 mM L-Glutamate and 10% ( volume/volume ) heat-inactivated FBS , in order to obtain a multiplicity of infection ( MOI ) 50 . The MOI was confirmed by plating 10μl spots of 10-fold serial dilutions of the bacterial solution onto agar plates . After 30 minutes of incubation ( to allow Salmonella invasion ) , cells were washed with phosphate-buffered saline before adding GMEM supplemented with gentamicin ( 50μg/ml ) . Cells were incubated for the appropriate length of time and then washed and lysed with 0 . 1% Triton X-100 . Dilutions of the cell lysates were plated onto agar plates to determine the number of intracellular bacteria . Alternatively , cells were washed and fixed onto 13mm coverslips with 4% formaldehyde then stored in PBS for confocal or electron microscopy . Salmonella infected cells were washed twice with the wash buffer from the Cytotoxicity 3 kit after fixation and permeabilized with the permeability buffer from the same kit for 10 minutes . The cells were then blocked with the block buffer for 20 min at room temperature and stained with goat anti-Salmonella CSA-1 antibody followed by tagged secondary antibody . Glass coverslips were mounted onto a microscopic slide along with ProLong Gold antifade reagent DAPI ( Invitrogen ) . The preparations were observed with an LSM510 META confocal microscope ( Zeiss ) . Six groups of five C57BL/6 mice were challenged intravenously with 2 x 103 CFU of S . Typhimurium SL1344 , S . Weltevreden C2346 , 10259 , 98_11262 , 99_3134 or PBS as a control . The mice were followed for four days . They were all subsequently culled at day 4 , or earlier if they were critically moribund . For the streptomycin infection model , six groups of five C57BL/6 mice each were pre-treated with 10 mg of streptomycin ( 200μl of a stock solution of 50mg/ml of streptomycin ) 24 hours before challenge . The first group ( naïve ) was orally inoculated with PBS; the second , the third and the fourth groups were orally challenged with approximately 5 . 5x 105 CFU of respectively S . Typhimurium SL1344 , S . Weltevreden C2346 , S . Weltevreden 10259 , S . Weltevreden 98_11262 and S . Weltevreden 99_3134 . The mice were sacrificed four days post challenge and caecum was removed from all mice for further analysis . Part of the caecum was used for histology to look for inflammation and the remaining part was plated on LB agar in order to check for bacterial colonization of the colon .
One hundred and fifteen S . Weltevreden isolates were collected from 18 countries encompassing South Asia , Southeast Asia and Oceania ( S1 Table ) . These isolates were collected from a diverse range of sources including the environment , food , animal waste , human feces and blood . The collection of organisms spanned over 50 years ( isolated between 1940 and 2013 ) . DNA from all 115 S . Weltevreden isolates were sequenced using an Illumina platform and these sequences , together with four other previously published S . Weltevreden sequences [19 , 20] , were assessed using the reference Multi-Locus Sequence Type database ( http://mlst . warwick . ac . uk/mlst/dbs/Senterica ) and confirmed to be Sequence Type ( ST ) 365 . Table 1 summarizes the souce and the location of the isolates in this study . In order to generate a complete reference genome for S . Weltevreden , DNA from the isolate 10259 obtained from a stool of a Vietnamese child with diarrheal disease [41] , was sequenced using both the Illumina HiSeq and PacBio RSII long read sequence platforms . After manual adjustment , a single contig representative of the main S . Weltevreden chromosome was assembled , along with an additional contig representing a large plasmid . The chromosome of S . Weltevreden 10259 ( Accession number LN890518 ) was a single circular molecule of 5 , 062 , 936 bps harboring 4 , 723 predicted coding DNA sequences ( CDSs ) with an average G+C content of 52 . 1% ( Fig 1 ) . The single plasmid , named pCM101 ( Accession number LN890519 ) , was 98 , 756 bps with 98 predicted CDSs . High quality ( PacBio ) reference genomes were generated for three additional isolates: C2346 ( Accession numbers LN890520 and LN890521 ) , 98_11232 ( Accession numbers LN890522 and LN890523 ) and 99_3134 ( Accession numbers LN890524 to LN890526 ) . The contiguous S . Weltevreden 10259 reference genome was compared with 14 other isolates representative of the breadth across different S . enterica serovars . A single high quality assembly was chosen for each serovar from a publically available dataset and a pan genome was constructed [42] . A core Salmonella genome of 3 , 269 genes was identified , representing 3 , 161 , 517 bps , with SNPs at 134 , 485 positions . These data were used to create a phylogenetic tree to in which to position S . Weltevreden within the context of the broad species , S . enterica [43] ( Fig 2 ) . The nearest phylogenetic neighbor to S . Weltevreden was S . Elisabethville with a difference of 16 , 693 bps in the core genes of the representative isolates . Notably , this observation was also supported by the similarities in their serological composition ( S . Weltevreden; O:3 , O:10 or 15 and r , z6 positive , S . Elisabethville; O:3 , O:10 and r , 1 , 7 positive ) . S . Agona also mapped close to S . Weltevreden; this finding was in agreement with previously published data [20 , 43] . Nothing is currently known about the phylogenetic organization with the single serovar of S . Weltevreden . Consequently , a phylogenetic tree was constructed using data from 115 sequenced S . Weltevreden . SNPs were present at 22 , 569 positions; these data were then filtered using Gubbins to refine the final phylogeny by removing fragments of the genome with a recombinant signal . A total of 218 recombination blocks were identified in the sample set ( S1 Fig ) , which reduced the number of core SNPs to 2 , 601 . The subsequent phylogenetic analysis identified two primary clusters , designated here as the ‘Island Cluster’ and the ‘Continental Cluster’ correlating broadly with where the organisms were predominantly isolated ( Fig 3 ) . In total 112 SNPs discriminated between the two main clusters ( S2 Table ) ; these SNPs were found to distributed uniformly across the genome with no obvious high-density clusters . None of the identified mutations were found to be associated with premature stop codons . Indeed , our data suggest that in general pseudogene formation is not a dominant feature of the S . Weltevreden genome . We were able to broadly subdivide S . Weltevreden into five subclusters , again correlating largely to where the organisms were isolated , or their hypothetical origin . The Island Cluster contains two subclusters , one drawn primarily from islands in the Indian Ocean ( Indian Ocean subcluster ) and the other from islands in Oceania or from nearby Southeast Asian countries ( Oceania subcluster ) . The phylogenetic structured suggested several independent introductions onto the various sampled islands and that these organisms subsequently evolved independently . The Continental cluster contained multiple isolates from Vietnam ( a primary sampling site ) , with three distinct subclusters , which captured the circulating lineages , Vietnam 1 , Vietnam 2 and Vietnam Antimicrobial resistant ( AMR ) . The Vietnam AMR subcluster was distinct from the other S . Weltevreden because 3/13 ( 23% ) isolates had coding sequences that were associated with AMR ( S1 Table ) , which overall we found to be rare within S . Weltevreden . Notably , with the overall phylogenetic structure of S . Weltevreden no particularly evident clustering was observed between the sources ( environmental/human ) of the bacterial isolates . Indeed , isolates from animal , food , the environment and humans were arbitrarily distributed throughout the tree . Furthermore , no evidence of significant temporal clustering was observed within the structure . Plasmid DNA was found to be common , with equivalents of pCM101 ( the plasmid described in S . Weltevreden 10259 ) detectable in 90% of the sequenced isolates . A phylogenetic tree of the pCM101 structure mirrored that of the main chromosome , indicating that this plasmid has co-evolved with the chromosome , indicating a long-term likely synergistic relationship . This phylogenetic signal was determined by relatively little variation , with only 970 SNPs discriminating the plasmids on the phylogenetic tree . Further , when a single region of recombination was excluded from seven organisms that were isolated on La Réunion Island , the number of SNPs was reduced to only 48 . Many isolates scattered across the tree harbored more than one plasmid . Known AMR genes were detected in seven isolates . Isolate 2013_2776 , originating from a food source in France ( potentially imported from Southeast Asia ) was found to harbor six AMR genes ( aph ( 3’ ) -Ia; aminoglycoisides , oqxA/B; general efflux , strA/B; streptomycin and tetB; tetracycline , which were located on a plasmid similar to plasmid pSH111_227 ( accession JN983042 ) found in S . Heidelberg . Indeed , of the isolates with laboratory confirmed resistance against antimicrobials ( S1 Table ) , seven harbored additional predicted plasmids , equivalents of which have been previously associated with AMR ( Table 2 ) [44–46] . A predicted pan genome of the sequenced S . Weltevreden isolates was created using annotated de novo assemblies . We identified a core of 4 , 046 CDSs present in each of the sequenced isolates using this approach , compared to 2 , 572 core CDSs identified using the broader set of Salmonella serovars . The total S . Weltevreden accessory genome was comprised of an additional 7 , 923 CDSs . An mean of 15 new predicted CDSs was added to the pan genome with every sequenced isolate and there were underlying collection of unique genes that were only found in single isolates ( S2 Fig ) ; the majority of these appeared to be attributable to mobile genetic elements and/or gene islands . Therefore , given the rate of novel gene acquisition and the increasing size of the pan genome , S . Weltevreden seems to have a submissive genome structure with the ability to routinely acquire and lose additional genetic material . Prophage-like regions are commonly associated with rapid evolution in S . enterica [47 , 48] . In order to align the phylogeny with phage-like signatures the genome sequences of 10259 , C2346 , 98_11262 and 99_3134 , representing a cross section of diversity within the tree , were investigated using “PHAST” ( http://phast . wishartlab . com/ ) . Several apparently complete prophage sequences were identified within each bacterial isolate , with a mean of 12 prophage elements per isolate . Most prophages identified using PHAST were shared by all isolates; these included the classical Salmonella phages Gifsy 1 , Gifsy 2 , Fels 1 and entero PsP3 [49] . The Salmonellae have the ability to both adhere to and invade cultured cells [50] . Consequently , cultured Hep-2 cells were exposed independently to S . Typhimurium SL1344 ( pSsaG ) , S . Weltevreden C2346 ( pSsaG ) ( Human asymptomatic stool isolate ) , S . Weltevreden 10259 ( pSsaG ) Human symptomatic stool isolate ) , S . Weltevreden 98_11262 ( pSsaG ) ( Human bloodstream isolate ) and S . Weltevreden 99_3134 ( pSsaG ) ( Human bloodstream isolate ) at a multiplicity of infection ( MOI ) of ~50 bacteria per human cell ( Fig 4 ) . All S . Typhimurium and S . Weltevreden tested were able to invade Hep-2 epithelial cells . S . Typhimurium SL1344 ( pSsaG ) exhibited a consistently stronger fluorescent signals at both two and six hours post infection , when compared to all S . Weltevreden . No significant difference in intracellular bacterial burden was observed between individual S . Weltevreden isolates . Importantly , there were consistently lower levels of GFP-positive S . Weltevreden using microscopic imaging than with S . Typhimurium , indicating that they are much less invasive in this assay than the archetypal S . Typhimurium SL1344 . To assess the bacterial burden a gentamicin-killing assay was performed with infected Hep-2 epithelial cells . For S . Typhimurium SL1344 , there was a consistent increase in the number of viable internalized bacteria between two and six hours post infection . At six hours post infection , there was a significant difference in the number of viable bacteria recovered compared to the two hour post infection time point for S . Typhimurium SL1344 ( p = 0 . 0001 , two sided t-test ) ( Fig 5 ) . Conversely , no significant difference in recovered numbers was observed between the two and six hour time point for the S . Weltevreden isolates . Furthermore , there was a consistently lower level of invasion by all S . Weltevreden in comparison to S . Typhimurium SL1344 . There are several murine models of Salmonella infection , which include the classical systemic typhoid model [51] , and the streptomycin pre-treatment model [52] , the latter of which is more relevant for gastroenteritis . Consequently , S . Weltevreden isolates were evaluated in both of these murine infection models in comparison again to S . Typhimurium SL1344 . To determine the systemic virulence of S . Weltevreden , C57bl/6 ( Salmonella susceptible , Nramp-1 negative ) mice were infected intravenously with S . Typhimurium SL1344 , S . Weltevreden C2346 and S . Weltevreden SW 10259 with 2 , 000 CFU . Mice were followed for four days to monitor disease severity . Mice infected with S . Weltevreden survived the four days post infection and remained well thereafter until being killed . In contrast , the majority of the C57bl/6 mice infected with S . Typhimurium SL1344 reached the disease severity endpoint two days post infection; others reached this state by day four and were killed ( Fig 6 ) . In order to study the potential of S . Weltevreden to cause gastroenteritis , and to disaggregate the potential mechanism of infection , streptomycin pre-treated C57bl/6 mice were orally challenged with S . Weltevreden isolates C2346 and 10259 and S . Typhimurium SL1344; the inflammatory response and evidence of infection in the caecum were compared using histopathological analysis . Four days post infection , S . Weltevreden C2346 and 10259 induced pronounced inflammation on the caecum , characterized by edema in the submucosa , with distinct cellular infiltrate in the submucosa , the lamina propria , and the epithelial layer , as well as the presence of immune cells in the intestinal lumen . Crypt elongation and erosive changes in the surface epithelium were also observed . This inflammatory response was indistinguishable from S . Typhimurium SL1344 . Thus , in contrast to the attenuated phenotype displayed by S . Weltevreden in the systemic murine model , a similar pattern in intestinal pathology were observed between both these serovars in the caecum after challenge of streptomycin-treated mice . The degree of colonization by the different S . enterica isolates was also measured by weighing sections of the caecum and liver and enumerating surviving Salmonella ( Fig 6 ) . S . Typhimurium SL1344 exhibited a significantly higher level of cecal and liver colonization in comparison to the S . Weltevreden isolates . In contrast , there was no significant difference in cecal and liver colonization between the S . Weltevreden isolates .
In this study , a combination of WGS , phylogenetic and in-vitro/in-vivo phenotyping were used to characterize the emerging Salmonella serovar S . Weltevreden . Additionally , four reference genomes ( and corresponding reference strains ) that will be of value for further genetic and genomic work on this increasing important serovar were generated . Our analysis revealed that the S . Weltevreden genome is larger than those of many other S . enterica serovars with a mean size over five million basepairs . Much of this additional genetic material was attributed to the accessory genome , where complete prophage and additional prophage-related elements were found to be common . Another little known serovar—S . Elisabethville , was found to be phylogenetically closest S . enterica serovar to S . Weltevreden . Remarkably , S . Elisabethville shares core serological properties with S . Weltevreden , but S . Elisabethville is not a common pathogen in humans . It will be interesting to see if this related serovar emerges in humans in the future , as has been the case for S . Weltevreden . Our analysis additional found that S . Weltevreden is monophyletic serovar with two major phylogenetic clusters consisting of a largely “Continental” isolates and “Island” isolates . Thus , there is evidence of a significant degree of geographical structuring within the S . Weltevreden population . Some geographical clustering is also detectable within the subphylogeny , suggesting that S . Weltevreden continues to evolve within a specific geographical region , as opposed to frequently spreading from one location to another . Geographical subclustering has been detected in other serovars , including S . Typhimurium ST313 clades within sub-Saharan Africa [53] . These data suggest that after introduction Salmonella clades can become established in a specific environment or a human population where they can start to persist and evolve in an isolated manner . We found that 112 SNPs were cluster-specific; these could be of value for future epidemiological tracking studies . For example , it may be possible to determine if these SNPs can be used map potential transmission routes within and between different human and animal populations . Previous links have been described between seafood and S . Weltevreden causing human disease [54] . Here , some of the cluster-associated ( or private SNPs ) could be exploitable in SNP-based assays for the rapid identification of S . Weltevreden isolates in the field . Such approaches have been developed for other Salmonella serovars , including S . Typhi , for the urban and international tracking of typhoid fever [55 , 56] . It is meaningful that the phylogenetic structure of the S . Weltevreden sequenced here did not correlate with date of isolation , disease type or source ( environment , animal , human ) . Therefore , it was not possible to link particular genotypes to disease syndromes . The inability to link genotype to human disease is suggestive , implying that factors such as infectious dose , host susceptibility and immune status or the local environment may be influencing the patterns of disease . More thorough epidemiological studies of S . Weltevreden are required to identify environment-to-human , animal-to-human , or human-to-human transmission routes . Notably , AMR was found not to play a big influence in shaping the molecular epidemiology of with S . Weltevreden , as AMR genes were found in very few isolates . The relatively rare AMR isolates harbored several plasmids comprised of structures that have been previously described in other Gram-negative bacteria . Thus , S . Weltevreden clearly has the capacity to acquire AMR genes , however they were rare . We speculate that the lack of AMR genes in this relatively submissive genome is related to the ecology of the organism , the it will be important to maintain surveillance on the serovar in order to actively detect any increasing AMR trends . Phenotypic characterization of S . Weltevreden showed an overall attenuated virulence potential in different models of disease when compared to S . Typhimurium SL1344 . S . Weltevreden isolates were significantly less invasive in terms of their ability to enter and replicate in Hep-2 cells . Similar to in vitro observations in Hep-2 cells , S . Weltevreden isolates were moderately attenuated in both the mouse in both intravenous and oral streptomycin treated infection models , than S . Typhimurium SL1344 . In fact , mice intravenously infected with S . Weltevreden were able to survive four days post infection . However , despite this attenuation in Hep-2 cells and mice S . Weltevreden can clearly cause significant human disease and other approaches will be required to define virulence mechanisms associated with this emerging pathogen . In conclusion , for the first time we have studied the phylogenetic structure and aimed to define the virulence potential of S . Weltevreden , an emerging cause of Salmonella induced infections in tropical regions . Our data show that S . Weltevreden is complex serovar and whilst we did observe geographical clustering this pathogen appears to have a more permeable genome that many other Salmonella . The acquisition of horizontally acquired DNA appears to be the main evolutionary driving force within the serovar , however AMR genes are rare . We additionally can show that S . Weltevreden has a distinct virulence-associated phenotype in conventional laboratory Salmonella virulence assays , which clearly suggests attenuation in comparison to S . Typhimurium SL1344 . Our study provides new insights into this organism and will serve as a platform for future research on this emerging Salmonella serovar .
|
Organisms belonging to the species Salmonella enterica are a major cause of infection globally . Such infections can be zoonotic in origin or transmitted between humans . One of the most notable features of the genus Salmonella is that the dominant serovars that cause human infections change over time , with new threats periodically emerging . These trends often go unnoticed and are underreported in low-income locations . There is good evidence that Salmonella Weltevreden is emerging ( particularly in low-income countries in the tropics ) as a significant cause of diarrhea and sometimes invasive bacterial disease in humans . However , little is known about the phylogenetic structure or virulence potential of this unstudied serovar . Here , we provide a detailed phylogenetic analysis of S . Weltevreden through whole genome sequencing and bioinformatics tools . We identify that specific phylogenetic clusters are associated with geographical regions , providing novel data regarding the global distribution of different clades and high quality reference genomes to facilitate future work . We additionally show that S . Weltevreden has a distinct virulence-associated phenotype in conventional laboratory Salmonella pathogenicity assays that will guide future investigations on this serovar .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"sequencing",
"techniques",
"taxonomy",
"medicine",
"and",
"health",
"sciences",
"pathology",
"and",
"laboratory",
"medicine",
"pathogens",
"microbiology",
"salmonellosis",
"sequence",
"assembly",
"tools",
"bacterial",
"diseases",
"phylogenetics",
"data",
"management",
"enterobacteriaceae",
"phylogenetic",
"analysis",
"genome",
"analysis",
"molecular",
"biology",
"techniques",
"bacteria",
"bacterial",
"pathogens",
"salmonella",
"typhimurium",
"research",
"and",
"analysis",
"methods",
"sequence",
"analysis",
"infectious",
"diseases",
"computer",
"and",
"information",
"sciences",
"genomics",
"sequence",
"alignment",
"medical",
"microbiology",
"microbial",
"pathogens",
"salmonella",
"enterica",
"evolutionary",
"systematics",
"molecular",
"biology",
"salmonella",
"molecular",
"biology",
"assays",
"and",
"analysis",
"techniques",
"genetics",
"biology",
"and",
"life",
"sciences",
"computational",
"biology",
"evolutionary",
"biology",
"organisms"
] |
2016
|
A Phylogenetic and Phenotypic Analysis of Salmonella enterica Serovar Weltevreden, an Emerging Agent of Diarrheal Disease in Tropical Regions
|
Infection with the human liver fluke Opisthorchis viverrini induces cancer of the bile ducts , cholangiocarcinoma ( CCA ) . Injury from feeding activities of this parasite within the human biliary tree causes extensive lesions , wounds that undergo protracted cycles of healing , and re-injury over years of chronic infection . We show that O . viverrini secreted proteins accelerated wound resolution in human cholangiocytes , an outcome that was compromised following silencing of expression of the fluke-derived gene encoding the granulin-like growth factor , Ov-GRN-1 . Recombinant Ov-GRN-1 induced angiogenesis and accelerated mouse wound healing . Ov-GRN-1 was internalized by human cholangiocytes and induced gene and protein expression changes associated with wound healing and cancer pathways . Given the notable but seemingly paradoxical properties of liver fluke granulin in promoting not only wound healing but also a carcinogenic microenvironment , Ov-GRN-1 likely holds marked potential as a therapeutic wound-healing agent and as a vaccine against an infection-induced cancer of major public health significance in the developing world .
Approximately 10 million people in Thailand and Laos are infected with the South East Asian liver fluke Opisthorchis viverrini [1 , 2] . Infection with O . viverrini , a one-centimeter long flatworm that inhabits the bile ducts , is strongly associated with the induction of cholangiocarcinoma ( CCA ) , cancer of the bile ducts [3] . The World Health Organization’s International Agency for Research on Cancer classifies infection with O . viverrini as a ‘group 1 carcinogen [1 , 3 , 4 , 5] . In Thailand and neighboring countries , cyprinid fish that are intermediate hosts for O . viverrini are eaten raw as a staple of the diet [1 , 2] . Infected individuals in endemic areas suffer the world’s highest incidence of CCA , 65 times that experienced in non-endemic regions , and accounting for up to 81% of liver cancers in this region [3 , 4] . CCA is a primary cancer originating in cholangiocytes , the epithelial cells that line the biliary tree . It has long latency , is invasive , metastasizes , is relatively non-responsive to anti-tumor agents and has a dismal prognosis . How opisthorchiasis induces cholangiocarcinogenesis is likely multi-factorial , involving immunopathogenesis , increased consumption of dietary carcinogens , and the secretion of parasite proteins mitogenic for cholangiocytes [2] . We described a liver fluke-derived homologue of the human growth factor granulin , termed Ov-GRN-1 , from the excretory/secretory ( ES ) products of O . viverrini [2 , 6 , 7] . Ov-GRN-1 binds to cholangiocytes in experimentally infected hamsters and stimulates proliferation of fibroblasts and CCA cell lines . Here we sought to determine whether Ov-GRN-1 possesses wound healing capacity and might therefore function to repair the chronic damage it causes in the bile ducts during feeding activity and the ensuing chronic inflammation . Moreover , given the physiologic and genetic similarities between chronically healing wounds and cancer [8] , we sought to address whether Ov-GRN-1 promotes cellular changes that are conducive to the establishment of a tumorigenic environment .
Using fluorescence microscopy we report that recombinant Ov-GRN-1 ( rOv-GRN-1 ) labeled with Alexa Fluor 488 ( AF ) was putatively internalized by ~75% of cells from an immortalized human cholangiocyte cell line , H69 ( Fig 1A and 1B , S1A–S1D Fig ) . Cholangiocytes co-cultured with rOv-GRN-1-AF exhibited significantly higher ( P < 0 . 001 ) per cell fluorescence intensity ( 6 . 4-fold , or 15 . 3-fold RFU/mole ) than cholangiocytes co-cultured with a control recombinant protein ( thioredoxin-AF , rTRX-AF ) that had been expressed and purified under identical conditions ( S1E Fig ) . Using 3D-structured illumination microscopy , rOv-GRN-1-AF was detected between the apical and basal actin filaments of cells in monolayer , confirming internalization in cholangiocytes of the liver fluke granulin ( Fig 1C and 1D , S1 Movie ) . The precursor of human granulin is expressed as a seven-domain granulin unit , known as progranulin ( PGRN ) , and initiates context-dependent autocrine and paracrine signaling cascades [9 , 10 , 11 , 12] . PGRN is internalized by cells and targeted to a specific organelle , commonly lysosomes , when bound to co-factors such as sortilin or CpG nucleic acid motifs [9 , 10 , 11 , 13] . Attempts to identify the sub-cellular location of rOv-GRN-1 after internalization by cholangiocytes using a range of organelle-specific markers suggested a cytosolic location , as specific co-localization to organelles was not apparent ( S2 Fig ) . The lack of involvement of an organelle suggested direct cell entry followed by interactions with signaling cascades , rather than the more conventional growth factor receptor-based signal initiation . While unusual , direct cell entry and interaction with signaling molecules is known for small growth factors with alkaline tails , such as basic FGF [14 , 15]; the C-terminus of Ov-GRN-1 is highly basic [7] with a predicted pI of 12 , characteristics that also support this mode of cell entry . Previously , we silenced expression of the Ov-grn-1 gene using RNA interference ( RNAi ) that reduced cell proliferation of cholangiocytes co-cultured with the liver flukes [16] . To address the role of Ov-GRN-1 in wound repair we silenced expression of Ov-grn-1 using RNAi and assessed the ability of ES products from dsRNA-treated flukes to accelerate cell proliferation and wound repair . Levels of mRNA encoding Ov-GRN-1 were depleted by 97% in worms transduced with dsRNA specific for Ov-grn-1 but not affected by control dsRNA specific for luciferease ( luc ) ( S3 Fig ) . ES products were collected from culture supernatants of dsRNA-treated flukes and effects of the ES on proliferation of cholangiocytes assessed . ES products collected on days 1 , 5 and 7 from Ov-grn-1dsRNA-treated flukes reduced cell proliferation by ~48% ( P < 0 . 01; F ( DFn , DFd ) = 24 . 27 ( 3 , 7 ) ) compared to ES from luc-treated flukes ( Fig 2A and S3 Fig ) . To ensure that Ov-grn-1-dsRNA treatment did not have a major impact on the ES composition of the flukes , we compared ES profiles from Ov-grn-1- and luc-dsRNA treated flukes by SDS-PAGE , and did not detect obvious differences in protein yield or composition ( S4 Fig ) . At the outset , we assessed the role of Ov-GRN-1 in wound repair using in vitro scratch assays given that the procedure is a facile surrogate of cell migration and wound closure [17] . dsRNA-treated flukes were co-cultured in Transwell plates such that they were separated from the underlying cells by a porous membrane , but ES products could traverse the inner membrane of the chamber . Firstly , we showed that ES products from luc dsRNA-treated flukes substantially accelerated wound healing compared to both cholangiocytes and CCA cell lines that were not co-cultured with flukes ( Fig 2C and 2D ) . Secondly , and pivotal to this study , significantly less wound healing/closure was induced by Ov-grn-1 dsRNA-treated flukes in both cholangiocytes over 36 hours ( P < 0 . 01–0 . 0001; Fig 2B and 2C ) and CCA cells over 18 hours ( P < 0 . 001–0 . 0001; Fig 2C ) than with control luc dsRNA-treated flukes . Fewer cells crossed the margin of the wound of the scratched monolayers cultured with ES products from Ov-grn-1 dsRNA-treated flukes at the early time points ( 6–12 h , Fig 2B–2D ) , suggesting the involvement of cell migration in scratch closure rather than closure due simply to cell proliferation [17] . To confirm the role of Ov-GRN-1 in in vitro wound healing 20 nM rOv-GRN-1 was shown to be sufficient to significantly accelerate healing of a cholangiocyte monolayer compared to cells exposed to control protein ( rTRX ) ( F ( DFn , DFd ) = 16 . 32 ( 2 , 33 ) ; P < 0 . 01 ) ( Fig 2E ) . To determine whether rOv-GRN-1 could accelerate wound repair in vivo , sub-cutaneous deep lesions were surgically inflicted between the ears on laboratory mice , treatment applied and the injury covered with spray plaster , after which the rate of wound healing was quantified at intervals of 24 hours for four days [18] ( Fig 3A ) . This method is considered to be superior to the conventional abdomen wound protocol when investigating growth factors , since it quantifies healing primarily from epithelial re-growth rather than skin contraction [18 , 19] . Daily application of 56 pMoles of rOv-GRN-1 significantly accelerated wound healing within 2–4 days compared wound closure in response to application of a control protein ( rTRX ) ( F ( DFn , DFd ) = 32 . 08 ( 2 , 16 ) ; P < 0 . 01–0 . 001 ) or PBS ( Fig 3B ) . Angiogenesis is an integral aspect of wound healing , is essential for the vascularization of new tissue , and is a cardinal hallmark of carcinogenesis . The chorioallantoic membrane ( CAM ) assay is a commonly accepted in vivo model of vertebrate angiogenesis [7 , 20 , 21]; moreover , the ancestral lineage of the granulin family of growth factors [22] made us conclude that the CAM assay was a suitable mean by which to assess angiogenic properties of Ov-GRN-1 . Quail eggs were implanted with rOv-GRN-1- or PBS-soaked membranes . Membranes with two picomoles ( P < 0 . 05 ) or 20 picomoles ( P < 0 . 0001 ) of rOv-GRN-1 induced angiogenesis ( F ( DFn , DFd ) = 108 . 4 ( 2 , 9 ) ) ( Fig 3C ) in the embryo developing within the egg . We employed isobaric tags for relative and absolute quantitation ( iTRAQ ) of changes in expression of cholangiocyte proteins induced by rOv-GRN-1 . Using the Scaffold program , we reliably validated 215 proteins in cholangiocytes identified by Mascot compared to cells at baseline and at subsequent intervals ( S1 Table ) . rOv-GRN-1 induced >50% change in detectable expression levels ( P < 0 . 05 ) of 70 cholangiocyte proteins at ≥1 time point compared to control cells ( Fig 4A and S2 Table ) . During co-culture of up to eight hours there was substantial up-regulation of protein expression , after which moderation or down regulation of the proteins ensued beyond 16 hours from the start of the analysis ( Fig 4A ) . Three KEGG pathways with 12 significantly regulated proteins each—the spliceosome , endoplasmic reticulum protein processing and metabolic pathways ( Fig 4B ) were revealed by protein ontology analysis in the cholangiocytes cultured with the parasite granulin . Euclidean distance clustering revealed the internal patterning of temporal translational changes ( Fig 4A ) , where group X proteins underwent a short-term up-regulation ( 0 . 5–8 h ) followed by a lessening of expression . Group Y proteins also underwent a short-term up-regulation followed by a substantial down-regulation . Group Z proteins were distinct due to their high and rapid short-term up-regulation . Notably , six of the 13 group Z proteins are associated with the spliceosome ( Fig 4A and 4B ) . The dysregulated proteins were subjected to a network analysis ( Fig 4C ) . When the top-25 most highly up-regulated proteins were considered , proteins involved in the spliceosome pathway were most highly represented ( Fig 4D ) , and included the top three ( HNRNPA3 , THOC4 and NONO ) and nine of the top 25 most highly up-regulated proteins . Mass spectrometry is constrained in its ability to characterize changes in low abundance proteins such as growth factors and cytokines . We therefore assessed the changes in cholangiocyte gene expression after one and 24 hours of co-culture with rOv-GRN-1 using gene arrays targeting epithelial to mesenchymal transition ( EMT ) , oncogenesis , wound healing and Toll-like receptor signaling ( S3 Table ) . Thirty genes underwent an Ov-GRN-1-induced change ( P < 0 . 05 ) in regulation ( Fig 4E and S4 Table ) , including four which exhibited >50% change in expression levels . Three of the four upregulated genes encoded proteins from the C-X-C ligand chemokine family of cytokines: cxcl1 , cxcl2 and cxcl8 ( also known asinterleukin-8 ) ; the fourth gene encoded for serine/threonine kinase 11 ( stk11 ) , also known as liver kinase B1 . Another member of the cxcl family , cxcl5 , was significantly upregulated , but fell below the 50% cutoff ( 43% ) .
We report for the first time the secretion of a growth factor from a metazoan pathogen that promotes wound healing of mammalian host tissue in vivo . The implications of the findings are multi-fold and significant . Firstly , the instrumental role described here for Ov-GRN-1 in orchestrating wound repair implies that this protein represents an attractive target for the development of a vaccine that thwarts regulation of the microenvironment within the biliary tract parasitized by the liver fluke . Indeed we previously showed that antibodies to rOv-GRN-1 block proliferation of cholangiocytes [7] , which further bolsters the proposition of a vaccine with both anti-infection and anti-cancer properties . One potential caveat of a vaccine that blocks wound repair however is the consequences of an aggressive inflammatory response in the absence of wound resolution , including uncontrolled sepsis or other complications , so appropriate consideration is warranted . Second , the findings highlight the potential therapeutic application of Ov-GRN-1 as a novel biologic for treating both acute and chronic wounds , such as recalcitrant ulcers on the extremities of diabetic patients [23] . Mammalian granulins play diverse roles continuously during development from the embryo into adult life , including key roles in tissue remodeling and inflammation [22] . Mutations in the human granulin gene result in a spectrum of conditions , including neurodegenerative disorders [24] and malignant growth and metastasis [25] . Indeed , granulin has a central role in carcinogenesis of a range of malignancies [22]; pertinent to our findings , granulin is over-expressed in hepatocellular carcinoma ( HCC ) [26] and renders HCC cells resistant to Natural Killer cell-mediated cytotoxicity by modulating expression of MHC-associated genes [27] . By contrast , granulins of pathogens have received little attention . We detected O . viverrini granulin ( Ov-GRN-1 ) in the ES products of adult flukes and provided the first evidence of a parasite growth factor that drove proliferation of host cells [6 , 7] . The recent report of the O . viverrini genome revealed additional members of the granulin family–a single granulin domain protein ( Ov-GRN-2 ) and a pro-granulin ( PGRN ) containing eight granulin subunits [28] . Products of either of these genes were not evident within the ES proteome [6] and their role in the host-parasite relationship is unclear . The mechanisms by which vertebrate or liver fluke granulins drive cell proliferation and wound repair are poorly understood . Vertebrate PGRN contains seven individual granulin subunits that are post-translationally processed . Mouse PGRN but not the individual subunits of granulin binds to TNF receptors ( TNFR ) , and antagonizes TNF signaling [29] . Dissimilar to PGRN , Ov-GRN-1 consists of a signal peptide and a single granulin motif [7] . Although the ability of rOv-GRN-1 to bind to TNFR has not been investigated , probing a microarray of the human proteome microarray [30] with labeled rOv-GRN-1 failed to reveal binding to any isoforms of TNFR , or indeed to any other obvious cell surface receptors , on the array . Cholangiocyte proteins involved in the spliceosome pathway were significantly regulated after exposure to rOv-GRN-1 in vitro . The majority of intron removal from pre-RNA molecules is catalysed by the spliceosome , a large ribonucleo-protein complex that consists of five small nuclear ribonucleo-protein particles ( snRNP , U1-6 ) and >150 other proteins [31] . One critical component of the wound healing process that is heavily regulated by RNA binding and splicing is the epithelial to mesenchymal transition ( EMT ) , which increases the migratory and invasive properties of cells and thereby promotes wound closure [32 , 33 , 34] . However , cancerization also is an untoward consequence of EMT , and aggressive tumours often display dysregulated expression of spliceosome proteins [31 , 35] . Liver fluke granulin stimulated expression of genes encoding the chemokines CXCL1 , 2 , 5 and 8 ( also known as IL-8 ) . These chemokines signal through the receptor CXCR2 [36 , 37] by internal transactivation of the epidermal growth receptor ( EGFR ) and EGFR signaling through the mitogen activated protein kinase ( MAPK ) pathways [38] . Chemokines play central roles in wound repair , angiogenesis and recruitment of immune cells [36 , 39 , 40] . Inhibitors of MAPK signaling block rOv-GRN-1-induced cell proliferation [7] , and the increased expression of cxcl genes induced in cholangiocytes by rOv-GRN-1 may underlie this observation . In addition , expression levels of transcripts encoding several kinases including stk11 and irak1 were markedly stimulated by the parasite granulin . Both STK11 ( liver kinase B1 ) and IRAK1 ( Interleukin-1 receptor associated kinase 1 ) control signaling in inflammatory pathways and regulate chemotaxis in diverse processes including wound healing [41 , 42] . Moreover , somatic mutations in stk11 [43] and irak1 [44] associated with malignancy . Upregulation of these kinases during proliferation of cholangiocytes within the liver fluke-parasitized biliary tree may , therefore , increase the likelihood of these mutations . Topical application of picomoles of rOv-GRN-1 significantly accelerated repair of wounds in the skin of mice . Although liver fluke granulin triggers changes in the cellular proteome that establish a pre-tumorigenic environment , short-term therapy would reduce the likelihood of inducing cancer in patients . Whereas advances in understanding the impaired angiogenesis in non-healing wounds have been reported , few effective agents that promote or expedite wound healing and closure are yet available [45] . The ability of rOv-GRN-1 to accelerate wound healing in mice and promote angiogenesis in vivo revealed that this growth factor holds noteworthy promise for a new category of medicines for non-healing wounds and related indications . Other growth factors are of interest for their therapeutic properties , notably human PGRN due to its ability to bind to TNFR . Indeed , recombinant human PGRN inhibits TNF-activated signaling and protected against inflammation in rodent models of arthritis [29] . PGRN further exerts its anti-inflammatory influence by inducing naïve T cells to transform into FOXP3-expressing regulatory T cells ( Tregs ) [46] , a lymphocyte type that is underrepresented in inflammatory diseases but the presence of which is a hallmark of helminth infections [47 , 48] . Indeed we speculate now that Ov-GRN-1 may be the major inducer of Tregs during opisthorchiasis , but this hypothesis clearly warrants testing . In conclusion , we have shown using gene silencing and recombinant protein technologies that the most carcinogenic of parasitic helminths , the liver fluke O . viverrini , secretes a growth factor which in isolation is sufficient to repair wounds both in monolayers of cultured human cholangiocytes and in the skin of mice . While our mouse cutaneous wound healing studies are informative and shed light on the potential therapeutic application of Ov-GRN-1 for chronic wounds , they do not directly address the role of the protein in host-fluke interactions in the biliary tree . With recent advances in genome editing using CRISPR-Cas9 , we will soon be well placed to knock out the Ov-grn-1 gene , facilitating in vivo studies that will specifically address the role of the protein in healing parasite-induced wounds in the bile ducts . Ov-GRN-1 therefore is a worthy candidate at which to target novel interventions—drugs and/or vaccines with both anti-helminth and anti-cancer activity . Moreover , Ov-GRN-1 offers potential as a novel biologic for treating acute and chronic wounds where normal tissue repair mechanisms are insufficient . Now more than ever , there is acute need for new therapeutics to combat the epidemic of inflammatory diseases , particularly diabetes and associated chronic ulceration . The therapeutic efficacy of parasitic helminths and their secreted products in treating inflammatory diseases is clear-cut [49] . The present findings indicate that parasite growth factors , which by their very nature have evolved to repair damaged tissues within their hosts , offer great promise as a novel therapeutic modality informed by millennia of host-parasite coevolution .
Excretory/secretory ( ES ) and somatic proteins were harvested from adult O . viverrini grown in laboratory hamsters as described [7 , 50] . Briefly , O . viverrini metacercariae harvested from naturally infected cyprinoid fish were used to infect hamsters ( Mesocricetus auratus ) by stomach intubation . Hamsters were euthanized three months after infection , when adult O . viverrini flukes were removed from the biliary tract . The flukes were washed and cultured in modified RPMI-1640 ( Life Technologies ) containing penicillin and streptomycin at 37°C/5% CO2 for three days . Culture supernatant was retained as ES products of the parasites , and stored at -80°C [50] . Ov-grn-1 pET41a or thioredoxin ( trx ) cDNAs contained within the pET32a ( Novagen ) plasmid were transfected into BL21 Escherichia coli cells ( Life Technologies ) and used to create recombinant protein with auto-induction as previously described [7 , 51] . Briefly , ZYM-5052 culture media was supplemented with 100 μM Fe ( III ) Cl3 and 100 μg L-1 kanamycin to produce recombinant protein ( rOv-GRN-1 ) or 50 μg L-1 ampicillin to produce rTRX [51] . Two hundred ml of inoculated media in a 1L baffled Erlenmeyer flask was incubated overnight at 37°C with 300 rpm rotation to induce expression with auto-induction . Purification of rOv-GRN-1 was achieved using an AKTA10 purification system at 4°C ( GE Healthcare ) [52] . The BL21 E . coli pellet was lysed with 3 freeze/thaw cycles followed by sonication ( Q4000 sonicator , Qsonix ) on ice . Twenty g of the resulting pellet was solubilized in 400 ml urea-containing nickel binding buffer ( 8 M urea/300 mM NaCl/50 mM imidazole/50 mM sodium phosphate pH 8 [Sigma] ) at 4°C for 24 h with slow agitation . After filtration through 0 . 22 μM membranes , supernatants were incubated in nickel chelate resin on 2× 5 ml Histrap IMAC columns ( GE Healthcare ) . The columns were washed in increasing concentration of imidazole ( two column volumes [CV] at 50 mM/5 CV at 100 mM ) after which bound material was eluted in 500 mM imidazole in binding buffer . The control rTRX protein was expressed and affinity purified similarly , but under native conditions ( without chaotropes ) , as described [52] . Refolding of urea-denatured rOv-GRN-1 was performed with 28 mL of G10 Sephadex ( GE ) resin on a XK16/20 column ( GE ) [52] . A 120 ml Superdex 30 XK16/60 column ( GE ) was used to fractionate three ml of refolded rOv-GRN-1 into 150 mM NaCl , 50 mM sodium phosphate , pH 6 , at a flow rate of 1 ml min-1 . Fractions containing rOv-GRN-1 monomer eluting at a size equivalent of ~1 kDa were pooled . Protein concentration was established using a combination of the Bradford assay ( Bio-Rad ) and absorbance at 280 nm . The cholangiocyte cell line H69 is a SV40-transformed bile duct epithelial cell line derived from a non-cancerous human liver [53] and was obtained in 2010 from Dr . Gregory J . Gores , Mayo Clinic , Rochester , Minnesota . H69 cells and cells of the human cholangiocarcinoma ( CCA ) cell line KKU-M214 were maintained in T75 cm2 vented flasks ( Corning ) as monolayers as described [52 , 53 , 54 , 55] with minor modifications . KKU-M214 cells were maintained with regular splits using 0 . 25% trypsin ( Life Technologies ) every 2–5 days in complete media ( RPMI with 10% fetal calf serum [FCS] and 1× antibiotic/antimycotic ) at 37°C under 5% CO2 . Cell proliferation assays were performed with low nutrient media containing 0 . 5% FCS . H69 cells were maintained under similar conditions with growth factor supplemented media [54] ( DMEM/F12 with high glucose , 10% FCS , 1×antibiotic/antimycotic , 25 μg ml-1 adenine , 5 μg ml-1 insulin , 1 μg ml-1 epinephrine , 8 . 3 μg ml-1 holo-transferrin , 0 . 62 μg ml-1 hydrocortisone , 13 . 6 ng ml-1 T3 and 10 ng ml-1 EGF–Life Technologies ) . Low nutrient media for H69 cells was 5% complete media , i . e . 0 . 5% FCS and 5% of the growth factor concentrations for complete media . The identities as human-derived of both cell lines were confirmed with single tandem repeat ( STR ) analysis ( 15/15 positive loci across two alleles ) and mycoplasma free at the DNA diagnostics centre ( U . S . A . ) , accredited/certified by CAP , ISO/IEC 17025:2005 through ACLASS . Cells were seeded at 1500 cells per well in 200 μl of complete media as described above in E-plates ( ACEA Biosciences ) and grown overnight while monitored with an xCELLigence SP system ( ACEA Biosciences ) which monitors cellular events in real time by measuring electrical impedance across interdigitated gold micro-electrodes integrated on the bottom of tissue culture plates [56] . Cells were washed three times with PBS and replaced with 180 μl of low nutrient media as described above and incubated for a minimum of 6 h before further treatments . Treatments were prepared at 10× concentrations and added to each well in a total volume of 20 μl . The xCELLigence system recorded cell index readings hourly for 5–6 days after treatment . Cell index readings were normalized before treatment and cell proliferation ratios were determined from biological triplicates and represent the relative numbers of cells compared to control cells . H69 cells in complete media ( see above ) that were grown to confluence in 6 well plates ( Falcon ) were wounded by scratching the cell monolayer with a disposable 200 μl pipette tip , as described [17] . The wound in the monolayer was photographed regularly and closure was assessed using ImageJ software ( National Institute of Health , U . S . A . ) . Wound widths over time were plotted and compared to controls with matched 2-way ANOVA and Dunnett’s correction for multiple comparisons . For cell scratch assays performed in co-culture with liver flukes , wounded monolayers of cells in 6 well plates were co-cultured with 10 adult liver flukes that had been subjected to RNA interference to silence expression of Ov-grn-1 ( below ) in the upper chamber of Transwell ( 4 μm pore size ) inserts ( Corning , USA ) . Recombinant rOv-GRN-1 and rTRX ( control ) proteins were amine labeled with Alexa Fluor 488 ( AF488—Life Technologies ) [57] . H69 cholangiocytes were grown to 50% confluence on optical quality glass bottomed culture dishes containing a 0 . 17 mm thick cover glass ( World Precision Instruments ) . AF488-labeled proteins were added to cells at a final concentration of 3 μM and incubated for 18 h at 37°C under 5% CO2 . Cells were fixed in 4% paraformaldehyde/PBS for 20 min at room temperature . Cells were permeabilized in 0 . 1% Triton X-100/PBS and stained with 10 μM DAPI and 165 nm Alexa Fluor 568 Phalloidin ( Life Technologies ) . Specimens were mounted in 5% N-propyl-Gallate ( Sigma ) in 80% glycerol/PBS . For localization studies , cells were fixed in 4% paraformaldehyde in PBS for 20 min at room temperature and then permeabilized in 0 . 1% Triton X-100/PBS . Fixed and permeabilized cells were probed with either LAMP1 ( lysosomes ) , Rab5 , EEA1 ( early endosome ) , Rab7 ( late endosomes ) , GRP78 BiP ( endoplasmic reticulum ) or anti-golgin97 ( Golgi ) antibodies at a 1:200 dilution , followed by incubation with Alexa Fluor 568 goat anti-mouse or Alexa Fluor 568 goat anti-rabbit antibodies at a 1:1000 dilution . Conventional fluorescence imaging was performed with a 60× ( NA1 . 4 ) objective using an A1 confocal research microscope ( Nikon ) or a DeltaVision personal research microscope ( Applied Precision , GE Healthcare ) . Super Resolution imaging was performed using a DeltaVision OMX 3D-Structured Illumination Imaging system ( Applied Precision , GE Healthcare ) as previously described [58] and images were processed as described elsewhere [59] . The chorioallantoic membrane assay ( CAM ) assay was established based on previous studies using quail eggs [21 , 60] . Briefly , fertilized eggs of the quail Cortunix cortunix were incubated at 37°C in a humidified incubator for five days . The surface of the eggshell was sanitized by wiping with 70% ethanol . Subsequently , a 0 . 5-cm square window of shell was surgically resected . The CAM with visible blood vessels was gently pulled down after which the window was sealed with clear tape . Eggs were incubated at 37°C for 18 h . Subsequently , filter paper presoaked in 20 μl of 2 or 20 pMoles of rOv-GRN-1 was implanted . The surgical window was resealed , and the eggs incubated at 37°C for 18 h . Eggs were chilled and the surgical window was fixed with 25% glutaraldehyde . Implanted filter papers were trimmed and washed with PBS prior to counting the blood vessels using an Olympus SZX12 dissecting microscope with a light box using 32× magnification . A head biopsy model was employed , as recommended for assessment of growth factors in wound healing [18 , 19] . Briefly , five female BALB/c mice per group ( rTRX , PBS and rOv-GRN-1 ) were anesthetized ( intraperitoneal xylazine 16 mg kg-1; ketamine 80 mg kg-1 ) , after which the crown of the head was shaved with an electric razor . Mice were anesthetized three days later and the surgical site was sterilized with 70% ethanol wipes . A skin-deep wound of 5 mm in diameter was inflicted on the crown of the head using biopsy punch ( Zivic instruments ) . The lesion was rinsed with antiseptic ( Betadine , Sanofi ) , after which 56 pMoles of rOv-GRN-1 , rTRX or PBS suspended in 1 . 5% methyl cellulose ( Sigma ) in 50 μl was applied . Thereafter , the lesion was covered with Elastoplast Spray Plaster ( Beiersdorf ) . Progress of the wound , and wound closure , was documented with photographs taken at cumulative 1 . 6× magnification using a dissection microscope ( Olympus ) fitted with a Nikon D200 camera , each day for five days . Wound closure was ascertained in an unblinded fashion by comparison of the surface area of the lesion with the size as documented immediately after the wound was inflicted , with the assistance of ImageJ software . H69 cholangiocytes were cultured in complete medium until ~50% confluence was reached in T75cm2 flasks . Cells were washed three times in PBS , 13 . 5 ml of low nutrient medium was added and cells were grown overnight at 37°C in 5% CO2 . rOv-GRN-1 or rTRX ( 500 nM ) were prepared in pre-warmed low nutrient media and 1 . 5 ml was added to each flask for a final concentration of 50 nM recombinant protein in media . Cells were grown for 0 . 5 , 1 , 4 , 8 , 16 , 24 and 48 h , washed 3× in PBS and snap frozen then stored at -80°C . Cells were lysed in three ml of 0 . 2% SDS with 3× freeze/thaw cycles and centrifuged at 4000 g to remove cell debris . The protein in the supernatant was precipitated with methanol [61] . Precipitated protein was prepared as per manufacturer’s instructions from the 8-plex iTRAQ [62] kit ( AB SCIEX ) as previously described [63] . Briefly , 100 μg of protein samples for each time-point were digested with 2 μg of trypsin ( Sigma-Aldrich ) at 37°C for 16 h . Each sample was labeled with different iTRAQ labels and was subsequently combined into one tube for OFFGEL fractionation and LC-MS/MS analysis . A 3100 OFFGEL Fractionator ( Agilent Technologies ) with a 24 well setup was used for peptide separation based on isoelectric point ( pI ) , as described [64] . Sample clean up and desalting were performed using a HiTrap SP HP column ( GE Healthcare ) and a Sep-Pak C18 cartridge ( Waters ) . Samples were separated with the OFFGEL Fractionator and collected fractions were desalted using ZipTip ( Millipore ) followed by evaporation by centrifugation under vacuum . The sample was reconstituted , desalted and separated with an analytical nano-HPLC column ( 150 mm x 75 μm 300SBC18 , 3 . 5 μm , Agilent Technologies ) before being applied to a Triple TOF 5600 mass spectrometer ( Applied Biosystems ) ; the results were analyzed as described [64] . Database searches were performed on the SwissProt database ( version September 2013 ) using MASCOT search engine v4 . 0 ( Matrix- Science ) with parameters as previously described [64] . Findings from Mascot searches were validated using the program Scaffold v . 4 . 2 . 1 ( Proteome Software Inc . ) [65] . Peptides and proteins were identified using the Peptide Prophet algorithm [66] , using a probability cut-off of 95% ( peptides ) or 99% probability ( proteins ) , and contained at least two identified peptides ( proteins ) [67] . Proteins containing similar peptides that could not be differentiated based on tandem mass spectrometry ( MS/MS ) analysis were grouped to satisfy the principles of parsimony . A false discovery rate ( FDR ) of <0 . 1% was calculated using protein identifications validated using Scaffold v . 4 . 2 . 1 . Furthermore , a FDR of <0 . 4% for the proteins identified was calculated using protein identifications validated by Scaffold . Proteins sharing significant peptide evidence were grouped into clusters . Channels were corrected in all samples according to the algorithm described in i-Tracker [68] . Acquired intensities in the experiment were globally normalized across all acquisition runs . Individual quantitative samples were normalized within each acquisition run , and intensities for each peptide identification normalized within the assigned protein . The reference channels were normalized to produce a 1:1 fold change . Normalization calculations were performed using medians to multiplicatively normalize data . A protein-protein interaction analysis was performed using the String software ( http://string-db . org/ ) based on compiled available experimental evidence [69] . Adult flukes from hamsters were transformed with Ov-grn-1 targeted dsRNA ( residues 49–333 of the 444 nucleotide transcript [7] ) by square wave electroporation [70] . Briefly , 20 flukes in 100 μl of RPMI 1640 medium were dispensed into a 4 mm gap electroporation cuvette containing 5 μg dsRNA followed by a square wave pulse of 125 volts of 20 milliseconds duration . Transformed parasites were cultured for 1 , 2 , 3 , 5 and 7 days after treatment . Total RNA was isolated from parasites and Ov-grn-1 expression measures using qRT-PCR with SYBR Green ( TAKARA Perfect Real-time kit , Japan ) and O . viverrini actin ( GenBank EL620294 . 1 ) as a reference transcript [70] . The mRNA levels of Ov-grn-1 were normalized to actin mRNA and are presented as the unit value of 2-ΔΔCt where ΔΔCt = ΔCt ( treated worms ) – ΔCt ( control , luciferase dsRNA-treated worms ) [70 , 71] . ES products from treated and control worms were collected and tested for cell proliferation activity ( above ) . The time point at which maximum cell proliferation was attained with ES products from Ov-grn-1 ds-RNA-treated flukes was used to calculate the percent reduction in cell proliferation relative to ES products from luc dsRNA-treated flukes . ES products from dsRNA-treated ( Ov-grn-1 and luc ) worms were assessed by SDS-PAGE with silver staining to ensure that the protein profiles were consistent between treatments . Specific gene pathways in H69 cholangiocytes exposed to rOv-GRN-1 as described above were investigated by qRT-PCR . Cells from 6-well plates were harvested employing a cell scraper after 1 h ( “early” time point ) or 24 h ( “late” time point ) after the addition of recombinant proteins , and total RNA was isolated using the miRNeasy Mini Kit ( Qiagen ) . The concentration , purity and integrity of the RNA were evaluated using spectrophotometry ( Nanodrop 1000 ) and an Agilent 2100 Bioanalyzer . The RNAs were stored at -80°C until processed for cDNA synthesis and qPCR following the RT2 Profiler PCR Array protocol ( Qiagen ) . Four RT2 Profiler PCR Arrays ( Qiagen ) were screened—Wound healing ( PAHS-121Z ) ; Oncogenes and Tumor Suppressor genes ( PAHS-502Z ) ; Epithelial-Mesenchymal Transition ( EMT ) ( PAHS-090Z ) ; Toll-like Receptors ( TLR ) ( PAHS-018Z ) . Ct values were exported and analyzed for significance using RT2 Profiler PCR Array Data Analysis software version 3 . 5 ( http://pcrdataanalysis . sabiosciences . com/pcr/arrayanalysis . php ) . The relative quantitation , included in the software , was performed using the 2-ΔΔCt method employing a panel of 5 house keeping genes as follows: beta actin ( NM_001101 ) , beta-2-microglobulin ( NM_004048 ) , glyceraldehyde-3-phosphate dehydrogenase ( NM_002046 ) , hypoxanthine phosphoribosyltransferase 1 ( NM_000194 ) , and ribosomal protein , large , P0 ( NM_001002 ) . Control groups ( cells exposed to media alone ) were used as calibrator samples . Three biological replicates were assessed and included in the analysis . The qPCR experiments were performed using a Bio-Rad iCycler iQ5 with an initial activation step of 95°C for 10 min followed by 40 cycles of 95°C for 10 sec and 60°C for 1 min . A melting curve analysis from 55°C to 95°C and 0 . 5°C temperature increment every 30 sec was included at the end of the run . Statistical analyses were conducted using GraphPad Prism 6 . 02 software . For cell proliferation studies , two-way ANOVA with Sidak’s multiple comparison tests were used to compare the changes in proliferation induction of ES products from Ov-grn-1- compared to luc-dsRNA treated flukes . Degrees of freedom for the F-test output of the ANOVA were calculated with DFn and DFd representing the degrees of freedom of the numerator and denominator , respectively . For CAM studies , statistical analysis compared treatment ( rOv-GRN-1 ) and media alone controls using one-way ANOVA with Dunnett’s correction for multiple comparisons . For wound healing studies , closure rate of wounds was compared by 2-way ANOVA with Dunnett’s correction for multiple comparisons . For proteomics studies with cell lines , differentially expressed proteins were determined using Kruskal-Wallis Test and results were expressed in log2 ratios . Proteins with a P-value < 0 . 05 and a significant log2 fold-change >0 . 6 or <-0 . 6 ( for upregulated and downregulated proteins respectively ) were considered in subsequent analyses . For gene expression studies , the fold change values of the genes from the four analyzed gene arrays were exported to GraphPad Prism 6 . 02 , pooled and plotted in a volcano plot and the significantly dysregulated genes ( P ≤ 0 . 05 ) plotted as a gene expression heatmap using Microsoft Excel .
|
The oriental liver fluke Opisthorchis viverrini infects millions of people in SE-Asia and kills 26 , 000 people each year due to parasite-induced liver cancer . The mechanisms by which the parasite causes cancer are complex , but a role for excessive wound healing in response to feeding parasites in the bile ducts has been proposed . We show that a growth factor ( granulin ) secreted by the worm gets into bile duct cells and drives wound healing and blood vessel growth . We delve into this “supercharged” wound healing process and uncover a range of signaling molecules that initiate healing , but when dysregulated , can result in a deadly liver cancer . On the upside , this liver fluke growth factor is now a candidate drug for the development of novel wound healing therapeutics to treat chronic wounds , such as diabetic ulcers . Understanding this process is another step on the road to developing a vaccine to reduce both parasite burdens and the incidence of the most prevalent and fatal cancer in Thailand and surrounding countries .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2015
|
Carcinogenic Parasite Secretes Growth Factor That Accelerates Wound Healing and Potentially Promotes Neoplasia
|
Oxidative stress serves as an important host/environmental signal that triggers a wide range of responses in microorganisms . Here , we identified an oxidative stress sensor and response regulator in the important multidrug-resistant nosocomial pathogen Enterococcus faecium belonging to the MarR family and called AsrR ( antibiotic and stress response regulator ) . The AsrR regulator used cysteine oxidation to sense the hydrogen peroxide which results in its dissociation to promoter DNA . Transcriptome analysis showed that the AsrR regulon was composed of 181 genes , including representing functionally diverse groups involved in pathogenesis , antibiotic and antimicrobial peptide resistance , oxidative stress , and adaptive responses . Consistent with the upregulated expression of the pbp5 gene , encoding a low-affinity penicillin-binding protein , the asrR null mutant was found to be more resistant to β-lactam antibiotics . Deletion of asrR markedly decreased the bactericidal activity of ampicillin and vancomycin , which are both commonly used to treat infections due to enterococci , and also led to over-expression of two major adhesins , acm and ecbA , which resulted in enhanced in vitro adhesion to human intestinal cells . Additional pathogenic traits were also reinforced in the asrR null mutant including greater capacity than the parental strain to form biofilm in vitro and greater persistance in Galleria mellonella colonization and mouse systemic infection models . Despite overexpression of oxidative stress-response genes , deletion of asrR was associated with a decreased oxidative stress resistance in vitro , which correlated with a reduced resistance to phagocytic killing by murine macrophages . Interestingly , both strains showed similar amounts of intracellular reactive oxygen species . Finally , we observed a mutator phenotype and enhanced DNA transfer frequencies in the asrR deleted strain . These data indicate that AsrR plays a major role in antimicrobial resistance and adaptation for survival within the host , thereby contributes importantly to the opportunistic traits of E . faecium .
Enterococci are commensal Gram-positive cocci of intestinal origin . First reported as a cause of infective endocarditis in 1899 , enterococci have also become , over the past 20 years , the 2nd–3rd most common organisms isolated from healthcare-associated infections [1] , [2] . In the USA , the emergence of enterococci as nosocomial pathogens was associated with a gradual replacement of Enterococcus faecalis by Enterococcus faecium and an epidemic spread of vancomycin-resistant E . faecium [3] . Acquisition of resistance to ampicillin and then to vancomycin , impacting the antibiotic treatments of choice , has been assumed to be the major factor responsible for transforming this organism from its docile , commensal nature into a significant nosocomial pathogen [3] . Reports on the transfer of vancomycin resistance from enterococci to methicillin-resistant Staphylococcus aureus stress the need to better understand the molecular epidemiology , as well as the transmissibility and virulence of enterococci , to control further spread and develop treatment and eradication strategies [4] , [5] . Mortality associated with vancomycin-resistant E . faecium infections is high but is more related to severe underlying diseases in infected patients than to production of bacterial virulence factors [6] . One of the most remarkable features of E . faecium isolates is their striking capacity to colonize both healthy carriers and patients , to survive to the host defences and to spread in the hospital environment , leading to major outbreaks [7] . The factors underlying its colonization capacities , including host-persistence , environmental stress response and adaptation , are only poorly understood . In addition to antibiotic resistance genes , several virulence genes have been identified in E . faecium of which espEfm and acm ( encoding a surface protein and a collagen adhesin , respectively ) have been experimentally proven to be important for infection in animal models [8] , [9] . In numerous organisms , virulence genes are controlled by environmental stresses and involve alternative σ factors of RNA polymerase and specific transcriptional regulators . Enterococci lack a σB-like general stress σ factor , but approximately 10 transcriptionnal regulators have been shown to be involved in virulence and stress response in the related bacterium E . faecalis [10]–[13] . Deciphering the regulatory pathways that lead to virulence and antibiotic resistance is crucial to understand the mechanisms by which E . faecium can colonize and infect critically ill patients . MarR family transcriptional regulators play key roles in several bacterial species , including SarA , MgrA , and their homologs in Staphylococcus aureus [14]–[18] . These regulators utilize cysteine oxidation to sense oxidative stress and regulate bacterial responses . The MarR sub-family of OhrR ( organic hydroperoxide resistance regulator ) regulators , found in Bacillus subtilis and in numerous other Gram-positives , regulate bacterial resistance to organic hydroperoxides using similar redox-sensing mechanisms [19]–[23] . Interestingly , in pathogenic bacteria such as S . aureus and Pseudomonas aeruginosa , MarR regulators seem to play broad regulatory roles that have profound effects on global properties of the pathogen . MgrA ( multiple gene regulator A ) is the first example of utilizing this mechanism to regulate antibiotic resistance and expression of virulence factors in S . aureus [24] , [25] . In recent work , the MarR family transcriptional regulator OspR ( oxidative stress response and pigment production regulator ) , homologous to MgrA , was found to play key roles in antibiotic resistance and virulence regulation in P . aeruginosa [26] . These discoveries raise the possibility that the opportunistic microorganism E . faecium may also harbor a MgrA/OspR homologue that could assume global roles in pathogenesis through sensing oxidative stress . We report the finding of a MarR family oxidative sensing regulator , AsrR ( antibiotic and stress response regulator ) , in E . faecium . A search for MgrA/OspR homologues in E . faecium identified AsrR that shares significant sequence identities with OspR and OhrR proteins . AsrR was found to possess the winged-helix DNA binding motif and the two cysteine residues present in the MarR family members and to exert a global regulatory role on adaptive responses , antimicrobial resistance , oxidative stress response , autolysis , and pathogenicity in E . faecium . These results should help shed light on the understanding of the multifaceted adaptative response in E . faecium and its remarkable colonizing capacities .
The global regulators MgrA of S . aureus and OspR of P . aeruginosa play key roles in virulence regulation [15] , [16] , [26] . Using BLASTP analysis , we identified a MgrA/OspR homologue in the genome sequence of the E . faecium E1162 clinical isolate [7] . A single significant hit was obtained with the deduced protein of EfmE1162_0374 showing 34% and 44% amino acid identity with MgrA and OspR , respectively . After further study , we renamed EfmE1162_0374 as asrR ( for antibiotic and stress response regulator ) based on the observed phenotypes presented below . Pfam analysis showed that the deduced AsrR protein possessed the MarR-type helix–turn–helix motif placing it in the MarR protein family ( Figure 1A ) . Similarly to OspR , AsrR harbors two cysteine residues , found at positions 11 and 61 ( Figure 1A ) . These residues have been shown to play a major role in oxidative stress sensing in OspR [26] . Sequence comparison showed that asrR was conserved among all E . faecium isolates and that asrR putative homologs were present in Enterococcus gallinarum and Enterococcus casseliflavus but not in E . faecalis ( data not shown ) . Fifty-six base pairs upstream of asrR , the EfmE1162_0373 locus , subsequently renamed ohr , encoded a putative protein highly similar to Ohr proteins described in numerous Gram-positive bacteria ( Figure 1B ) . Usually , ohr is part of a two-gene operon and is co-transcribed with an upstream adjacent gene ohrR encoding a transcriptional regulator [20] , [23] , [27] . This organization was not found in E . faecium since no ohrR homologue was found upstream of the EfmE1162_0373 locus . However , the homology of asrR with ohrR ( 42% nucleotide identity ) suggested that AsrR may control the expression of ohr . RACE-PCR experiments in E . faecium HM1070 identified one promoter upstream of both asrR and ohr genes , and we showed that cotranscription of ohr and asrR may occur from the ohr promoter ( Figure 1B , Figure S1A , Figure S1B ) . We also determined experimentally the AsrR binding site upstream of the ohr gene ( Figure 1B , Figure S1C ) . A putative AT-rich inverted repeat sequence was found that overlapped the AsrR binding box of ohr ( Figure 1B ) , which is consistent with the fact that proteins of the MarR family specifically bind palindromic or pseudopalindromic sites using a conserved winged helix fold [22] , [28] . The direct interaction of AsrR with the ohr and asrR promoters was tested by electromobility shift assay ( EMSA ) ( Figure 1C ) . Purified His6-tagged AsrR bound specifically to the ohr and asrR promoter sequences , while failing to shift a non-promoter DNA fragment used as a control ( Figure 1C ) . In addition , the binding was lost in the presence of unlabelled competitor and restored in the presence of non-competitor DNA ( Figure 1C ) . Finally , 69 bp downstream of asrR , EfmE1162_0375 encoded a putative permease of unknown function conserved among enterococci ( Figure 1B ) . Suspecting that asrR expression was modulated by oxidative stress , we used quantitative real-time PCR ( qRT-PCR ) to analyze the expression of asrR in E . faecium HM1070 after 10 or 20 min of a 2 mM hydrogen peroxide ( H2O2 ) challenge ( Figure S2 ) . In addition , we also analyzed the expression of ohr since AsrR interacts directly with the promoter of this gene ( see above ) . We observed a strong induction of expression of both genes after 10 min of H2O2 treatment . Induction was higher for the ohr gene and decreased similarly for both genes after 20 min of H2O2 challenge ( Figure S2 ) . By contrast , no ohr upregulation was found in response to H2O2 oxidative stress in the ΔasrR mutant strain ( data not shown ) . As suspected , we showed that AsrR was a functionnal sensor of the oxidative stress . Indeed , we showed by using EMSA that after a treatment with 10 mM of H202 the oxidized His6-tagged AsrR was no longer able to bind to the ohr promoter ( Figure 2A ) . In addition , this effect was reversible since the addition of a reducing agent ( i . e . DTT ) restored the binding ability of the AsrR protein ( Figure 2A ) , and it was also dependent of the H2O2 concentration ( Figure 2B ) . To identify the set of AsrR-regulated genes in E . faecium , the transcriptome of the ΔasrR mutant was compared to those of the E . faecium HM1070 parental strain and of the knock-in ΔasrR::asrR complemented strain . Since the E . faecium microarray was custom-made based on the E1162 genome , an in silico comparative genomic hybridization was performed between HM1070 ( entirely sequenced , unpublished ) and E1162 E . faecium genomes . We found that 73 . 5% of probes ( 3924 of a total of 5337 ) designed for E1162 had 100% identical targets in HM1070 DNA . In addition , 6 . 8% of probes ( 364 of a total of 5337 ) had only one or two mismatches . Therefore , the E1162 E . faecium microarray appeared to be suitable for HM1070 transcriptome analysis since around 80% of the probes were conserved in both genomes . Both the parental strain and knock-in complemented derivatives were used for comparative transcriptome analysis to minimize the influence of unexpected random mutations that could have occurred during the construction of the asrR null mutant . We observed 87 genes significantly upregulated and 94 genes downregulated in the ΔasrR mutant strain in comparison to both the parental and complemented strains ( Figure S3A ) . Nine and 33 genes showed modified transcriptionin the mutant when compared to the parent only or to the derivative only , respectively , and were not considered further ( Figure S3A ) . To validate these results , we compared expression ratios obtained by microarrays and by qRT-PCR for seven selected genes and obtained an excellent correlation ( r2 = 0 . 99 ) ( Figure S3B ) . Expression ratios of key AsrR-regulated genes are shown in Figure 3 . Most genes shown in Figure 3 were up-regulated in the mutant and are classified in functional groups . A first functional group was composed of four genes homolog to those of the dlt operons involved in the resistance to cationic antimicrobial peptides ( CAMPs ) in E . faecalis , Bacillus subtilis , and S . aureus [29]–[31] as well as a pbp5 gene , encoding a penicillin-binding protein responsible for β-lactam resistance in E . faecium ( Figure 3A ) [32] . Noticeably , genes involved in the adhesion to extracellular matrix ( ECM ) including the well-characterized acm gene [9] , [33] and two ecbA paralogous genes [34] encoding microbial surface components recognizing adhesive matrix molecules ( MSCRAMM ) adhesins were also strongly up-regulated in the ΔasrR mutant ( Figure 3B ) . Several genes putatively involved in oxidative stress response were up-regulated , including the kat and gpx genes encoding a putative manganese-containing catalase and a putative glutathione peroxidase , respectively , and the aforementioned ohr gene that was the most upregulated gene in the ΔasrR strain ( Figure 3C ) . Numerous genes encoding putative transposon conjugative transfer proteins that could enhance DNA exchange and horizontal transfer were upregulated in the absence of AsrR ( Figure 3D ) . In addition , homologs of genes known in other bacteria to be involved in the adaptation to environmental changes , uvrA and mutS2 encoding an UV resistance determinant and a putative anti-recombination endonuclease , respectively , were downregulated in the ΔasrR strain ( Figure 3E ) . EfmE1162_0375 encoding a putative permease and located directly downstream of asrR was upregulated in the mutant as well as the gspA paralogous genes encoding general stress proteins ( Figure 3E ) . Two glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) homologues of GapA and GapB , reported as S . aureus virulence factors [35] , were upregulated in the ΔasrR strain ( Figure 3E ) . The gls24 and glsB genes , involved in bile salts stress response and virulence of E . faecium , were repressed by AsrR ( Figure 3E ) [36] . Finally , as previously described for several transcriptional regulators , asrR deletion also modulate expression of other transcriptional regulators , in particular that of SigV that was previously characterized in E . faecalis ( Figure 3E ) [30] , [37] . Taken together , these results indicated that AsrR acts as a global regulator in E . faecium , functioning mainly as a repressor of numerous genes involved in antibiotic and CAMP resistance , adhesion to ECM , oxidative stress response and adaptative response . Using upstream regions of several of those genes upregulated in the asrR deleted mutant , we computationally identified a 15-bp putative DNA binding box ( Figure S1D ) . In the following experiments , we tested the phenotypic effects of the modulation of expression of the various functional groups of genes . E . faecium can survive a wide range of stresses during its life cycle . The role of AsrR in the response to H2O2 and organic oxidative stresses was tested by survival analysis and growth on plates containing oxidants ( Figure 4A , Figure 4B ) . We performed survival experiments with a 2 mM H2O2 challenge for 30 min on cells in exponential or stationnary growth phases , and the ΔasrR strain was found to be around one order of magnitude more susceptible to hydrogen peroxide stress than the parental and complemented strains in both conditions ( Figure 4A ) . Note that resistance to H2O2 dramatically decreased on growing cells ( Figure 4A ) . We then performed a 2 mM H2O2 challenge for 30 min on cells in exponential growth phase in the presence of deferoxamine ( DFX ) , an iron chelator , or tiron , a superoxide anion scavenger ( Figure 4A ) . Interestingly , if the addition of DFX or tiron significantly increased the survival of both strains , as expected , the ΔasrR mutant was still significantly impaired as compared to the parental strain ( Figure 4A ) . In addition , the growth on BHI plates of the ΔasrR derivative was also impaired by the addition of 0 . 5 mM menadione , an organic peroxide ( Figure 4B ) . No significant differences were observed between the ΔasrR mutant and the parental strain when grown on plates containing other organic peroxides , such as tertiary-buthylhydroperoxide and cumene hydroperoxide ( data not shown ) . Taken together , these results confirm that AsrR plays a role in the E . faecium oxidative stress response . The role of AsrR in the oxidative stress response , was further tested in vivo . Survival of the parental , complemented and ΔasrR strains was monitored by counting of viable bacteria inside murine macrophages over a 3-day period ( Figure 4C ) . Clearance of the ΔasrR mutant was slightly faster than that of the parent and of the complemented mutant , and correlated with its increased in vitro oxidative stress sensitivity . Finally , we estimated the intracellular concentration of hydroxyl radicals by FACS ( fluorescence-activated cell sorting ) experiments in both parental and ΔasrR mutant strains by measuring the fluorescence intensity of a probe specific for reactive oxygen species ( ROS ) ( Figure 5 ) . The basal hydroxyl radical level was similar in both strains ( Figure 5 ) . Interestingly , exogenous H2O2 treatment ( 0 . 5 mM or 2 mM , 10 min ) increased the intracellular amount of hydroxyl radicals in both strains but no significant difference was found between the E . faecium HM1070 and ΔasrR strains ( Figure 5 ) . The effect of asrR deletion on the activity of various antimicrobials against E . faecium HM1070 was tested . The ΔasrR strain was more resistant to penicillin G and ampicillin ( MIC of 1 and 0 . 5 µg/ml , respectively ) than the parental ( MIC of 0 . 125 µg/ml for both antibiotics ) and the complemented strain ( MIC of 0 . 25 and 0 . 125 µg/ml , respectively ) ( Figure 6A ) . These results are consistent with the pbp5 gene upregulation in the ΔasrR mutant ( Figure 3A ) . No differences were observed for vancomycin ( MIC of 1 µg/ml for the three strains ) ( Figure 6A ) and 24 other antibiotics tested ( data not shown ) . Because glycopeptides and β-lactams are bactericidal against E . faecium , we tested if asrR deletion could promote survival to these drugs . Time-kill analysis was carried out in the presence of vancomycin , penicillin G , and ampicillin ( 4× MIC ) . The bactericidal activity of penicillins and vancomycin against the ΔasrR strain was markedly reduced ( by approximately one order of magnitude after 6 h and 24 h , respectively ) as compared to the parental and complemented strains ( Figure 6B ) . Finally , in agreement with tolerance to β-lactams [38] , tests with Triton X-100 showed that autolysis was twice more rapid for the parental and the complemented strains than for the ΔasrR strain ( Figure S4 ) . Bacterial cells have to cope with the CAMPs produced by other prokaryotic microorganisms and eukaryotic cells . The ΔasrR mutant exhibited noticeable growth on BHI plates supplemented with nisin ( a bacterial CAMP ) as compared to the parental and complemented strains ( Figure 6C ) . No significant differences were observed between the ΔasrR mutant and the parental strain when grown on BHI plates containing colistin methanesulfate ( data not shown ) . Previous studies have identified the dlt operon as crucial for response to CAMPs in numerous Gram-positive bacteria [29]–[31] , [39] which is consistent with upregulation of dlt in the absence of AsrR . Like other Gram-positive microorganisms , enterococci are able to produce biofilms on abiotic surfaces . The ability of ΔasrR , parental and complemented strains to form a biofilm on polystyrene microtiter plates was evaluated ( Figure 7A ) . To quantify biofilm production , the OD600 in wells where bacteria have been cultured was determined after crystal-violet staining ( Figure 7B ) . The parental and complemented strains did not produce biofilm after 24 h of incubation at 37°C whereas the asrR mutant adhered to the surface and formed significant amounts of biofilm ( Figure 7B ) . Adhesion to host-cells is a crucial step in the infection process and for host-colonization . Upregulation of the acm and ecbA genes , encoding major MSCRAMM adhesins , in the absence of AsrR prompted us to evaluate the contribution of AsrR to E . faecium adherence to HT-29 intestinal epithelial cells ( Figure 7C ) . A high percentage of ΔasrR bacteria attached to the HT-29 cells ( median value 44% ) , while the parental and complemented strains showed significantly lower levels of attachment ( median values 18 and 25% , respectively ) ( P<0 . 01 ) ( Figure 7C ) . Inactivation of the postreplicative DNA repair pathways has been shown in a wide variety of microorganisms to result in a mutator phenotype [40] . Since the uvrA gene , encoding a putative excision repair protein , was downregulated in the absence of AsrR , we determined the mutation frequency to spectinomycin resistance in the parent and the constructs . The ΔasrR strain displayed five-fold increase ( P = 0 . 024 ) in mutation frequencies ( 4 . 4×10−8±1 . 8×10−8 ) as compared to the parental and complemented strains ( respectively 8 . 5×10−9±1 . 1×10−9 and 9 . 1×10−9±1 . 3×10−9 ) . Considering both the strongly upregulated expression of genes involved in conjugation of transposons and the downregulated expression of mutS2 in the absence of AsrR ( Figure 3D and 3E ) , we studied the involvement of AsrR in DNA transfer . We conjugated the integrative conjugative transposon Tn916 ( which confers tetracycline resistance ) from strain Streptococcus agalactiae UCN78 to E . faecium HM1070 , ΔasrR , and ΔasrR::asrR , and subsequently from this set of strains to E . faecalis BM4110 . Note that integration site of the Tn916 in HM1070 , ΔasrR , and ΔasrR::asrR strains did not influence the transfer frequency of three transconjugants tested for each constructed donor strains ( data not shown ) . However , in three independent experiments , the ΔasrR/Tn916 strains displayed a four-fold mean increase in Tn916 transfer frequency ( 5 . 2×10−6±1 . 3×10−6 ) as compared to parental ( 1 . 5×10−6±2 . 2×10−6 ) and complemented ( 1 . 7×10−6±1 . 9×10−6 ) strains ( P = 0 . 039 ) . Increased mammalian cell adhesion and biofilm formation of the ΔasrR mutant lead us to test the impact of AsrR on colonization of the host . To assay pathogenicity , we used larvae of the moth Galleria mellonella of which the innate immune system shares a high degree of structural and functional homology with that of mammals [41] . As described previously , only weak lethality for the larvae was observed with the parental strain [42] and no significant differences were found with the mutant ( data not shown ) . Then , Galleria larvae were infected with ΔasrR , parental , or complemented strains , sacrificed at 0 h , 24 h , 48 h , and 72 h and bacterial counts were monitored in host homogenates ( Figure 8 ) . The parental and complemented loads markedly decreased following infection ( from 1×106 to 3×104 and 5×104 CFU/larva 72 h post-infection , respectively ) whereas the ΔasrR load decreased only slightly , after stabilizing ( 2 . 8×105 CFU/larva 72 h post-infection ) . In correlation with the insect model , the ΔasrR mutant strain showed statistically significant increase of bacterial burdens in kidney and liver tissues 168 h post-infection ( Figure 9 ) . The ΔasrR mutant exhibited an increase of 1 . 17 log unit in the kidneys ( P = 0 . 002 ) ( Figure 9A ) and 0 . 70 log unit in the livers ( P = 0 . 011 ) ( Figure 9B ) compared to the burdens of the HM1070 parent strain . The ΔasrR::asr complemented strain loads were restored to the wild-type level in both tissues confirming the involvement of AsrR in the E . faecium pathogenicity .
Reactive oxygen species were originally considered to be exclusively detrimental to bacterial cells . However , redox regulation involving ROS is now recognized as a vital component to bacterial signaling and regulation [43]–[45] . Some members of the MarR family modulate the transcription of virulence and/or stress genes using an oxidative sensing mechanism . In particular , studies on OspR of P . aeruginosa and MgrA of S . aureus have shown that the activity of these regulators that sense oxidative stress is not limited to oxidative stress response but has pleiotropic effects [15] , [26] . The sensing mechanism of OspR has been recently described [26] . A cysteine residue , Cys-24 , is used by OspR to sense a potential oxidative stress and to regulate bacterial response . Cys-24 is first likely oxidized and the resulting sulphenic intermediate is trapped by a second cysteine , Cys-134 , to form an intermonomer disulphide bond . The inactive form of OspR dissociates from promoter DNA resulting in modulation of gene expression [26] . We identified a gene , asrR , which encodes a functional homologue of the OspR/MgrA proteins and is present in all sequenced E . faecium strains and absent in E . faecalis . Two cysteine residues are present in the protein sequence encoded by asrR indicating that AsrR belongs to the 2-Cys protein family , which senses peroxides [26] , [46] , [47] . Similarly to OspR , our data show that oxidative stress leads to inactivation of AsrR , resulting in loss of binding to promoter DNA , which leads to prompt modulation of gene expression . Investigation of the AsrR regulon identified numerous targets consistent with the pleiotropic phenotype resulting from its inactivation . Oxidative stress acts as a signal modulating AsrR activity , but it remains a challenge to which bacteria have to cope with during infection . Indeed , our results show that ArsR played a role in the survival against H2O2 challenge as well as into phagocytic cells since three important genes from the oxidative stress regulon ( i . e . kat , gpx , and ohr ) are overexpressed in the absence of AsrR . However , this appears a priori in contradiction with the higher susceptiblity of the null-mutant strain to both in vitro oxidative stress and oxidative burst in mouse macrophages . Despite the fact that enterococci possess a kat gene , catalase enzyme can only be formed when heme or manganese is present and these organisms are considered as catalase-negative bacteria [48] , [49] . While little is known about mechanisms of oxidative stress response in E . faecium , it has been shown in E . faecalis that peroxidases important for the survival under oxidative stress and into macrophages are Tpx ( thiol peroxidase ) , Npr ( NADH peroxidase ) , and Ahp ( alkyl hydroperoxide reductase ) [50] , so that E . faecium homologs are not upregulated in the ΔasrR mutant . In E . faecalis , gpx encodes a glutathione peroxidase of which activity is regenerated by a glutathione reductase [51] . Since the gene encoding this reductase does not appear to be a member of the AsrR regulon , the impact of gpx overproduction alone on the oxidative damages restoration should be reduced in the mutant strain . Also , the data obtained with fluorescent ROS-specific probe confirmed that ΔasrR mutant strain did not better detoxify hydroxyl radicals than the wild-type strain . Interestingly , addition of ROS scavengers during the H2O2 challenge reduced the sensitivity of both wild-type and ΔasrR mutant strains , the mutant being still more sensitive . Under our conditions , it seems that hydrogen peroxide is capable of damaging the bacterial cell independently of the formation of hydroxyl radicals formed via the Fenton's reaction [52] . Therefore , although oxidative stress leads , through AsrR derepression , to overproduction of detoxification proteins , the absence of difference in intracellular ROS accumulation suggests that asrR deletion may lead to an increased oxidative susceptibility , independently of ROS detoxification pathways . One hypothesis is that a modification of the bacterial cell wall , for which evidence is also provided by our transcriptomic analysis and autolysis assay , may lead to increased oxidative susceptiblity [53] . AsrR regulation was not restricted to oxidative stress response but extended to modulation of expression of multiple targets . First , AsrR modulated resistance and tolerance to cell-wall active antimicrobial agents . In E . faecium , resistance to penicillins is due to production of the low-affinity penicillin-binding protein PBP5 [54] and overproduction of PBP5 increases the level of ampicillin resistance [32] , [55] . Accordingly , the increase in MICs of penicillins after deletion of asrR in E . faecium HM1070 may be explained by upregulation of the pbp5 gene . Interestingly , recent reports on the role of MgrA and OspR in antibiotic resistance reinforce the implication of these MarR regulators in β-lactam resistance [26] , [56] , [57] . In addition , the activity of the CAMP nisin , which damage the bacterial membrane , was reduced against the asrR null mutant . The dlt operon encodes proteins that alanylate teichoic acids , the major components of the cell wall of Gram-positive bacteria . This generates a net positive charge on bacterial cell walls , that repulses positively charged molecules and confers resistance to CAMPs [29] , [31] , [39] . Therefore , AsrR could contribute to modulate resistance to nisin in E . faecium through regulation of the dlt operon . These data suggest that the E . faecium Dlt resistance system is effective against CAMPs , as previously shown for E . faecalis [30] . Besides resistance , tolerance to antibiotics is an efficient pathway for bacteria to escape antimicrobial-induced killing . Bactericidal activity of both penicillins and vancomycin was significantly decreased in the asrR null mutant . Importantly , these antibiotics remain primary therapeutic choices for the treatment of enterococcal infections [58] . The molecular basis for tolerance remains poorly understood , and processes involved are much more complex than previously thought [59] . As suggested by transcriptomic data as well as biofilm and autolysis phenotypes , the higher tolerance to β-lactams and glycopeptides of the asrR mutant might be due to a modification of cell wall composition ( peptidoglycan , lipoteichoic acids ) and/or of intrinsic control of lysis ( murein hydrolase activity ) [59] . Similarly to other pathogens , adherence of E . faecium to exposed host ECM is likely the first step in the infection process . MSCRAMMs are proteins that adhere to components of the ECM [34] . To date , three MSCRAMMs , Acm , Scm and EcbA , have been characterized in E . faecium adhesion [34] , [60] . Acm has been previously shown to interact with collagen type I and to a lesser extent with collagen type IV [33] , [61] , whereas Scm and EcbA bind to collagen type V and fibrinogen [60] , [62] . The in vivo function of Acm has been thoroughly investigated highlighting its role in the pathogenesis of experimental E . faecium endocarditis [9] . Consistent with the literature , marked up-regulation of both acm and ecbA genes expression promotes ability of the null-mutant strain to adhere to epithelial cells . In addition , as previously described for MgrA in S . aureus [63] , we report that AsrR represses biofilm formation in E . faecium . Following primary adhesion , biofilm establishes a protected environment for growth that enables bacteria to proliferate by restricting antibiotic access and shielding the bacterial pathogen from host immune defences [64] . Modulation by AsrR of biofilm formation and expression of MSCRAMM proteins indicates that this regulator may contribute to the host-colonization by E . faecium , a hypothesis confirmed in a Galleria persistence model and in a murine systemic infection model . Galleria insect model has been widely used to evaluate virulence of numerous pathogens [65]–[67] but has also recently been shown to be suitable for the study of E . faecium host-persistence [42] . Interestingly , the increased persistence of the null-mutant E . faecium strain inside larvae is correlated with its persistence following infection in mouse kidneys and liver . Good correlation between mouse and insect models has already been reported in the literature [68] . Beyond immediate cellular adaptation to stress , E . faecium organisms adapt their genome to hostile environmental conditions through acquisition of beneficial genes from external sources or by de novo mutations . The UV resistance genes ( uvr ) that are part of the SOS systems , have been analyzed in detail in E . coli and E . faecalis [40] , [69] , [70] , and UvrA is the initial DNA damage-sensing protein in nucleotide excision repair [71] . AsrR deletion in E . faecium causes the downregulation of uvrA and , interestingly , the null mutant strain showed a mutator phenotype . We thus speculate that , under oxidative stress , E . faecium cells will promote mutations through AsrR-mediated deregulation of uvrA which would be globally profitable in hostile environments even though some may be deleterious to individual cells . Long-term adaptation may also benefit from genetic changes due to acquisition of pre-evolved functions via horizontal transfer [72] . Interestingly , we found that , in the absence of AsrR , E . faecium increases the transfer frequency of conjugative transposon Tn916 , which might be linked to both the strong up-regulation of conjugative transposon protein and the down-regulation of the mutS2 gene . Also , the DNA damage response role in the regulation of transfer of mobile genetic elements was previously described in Bacillus subtilis [73] . Considering the adaptive role of uvr and mutS2 genes and DNA exchange , AsrR may contribute to the E . faecium long-term adaptation by modulating its mutability and DNA transfer capacities . Lastly , the locus encoding the transcriptional regulator SigV , previously described as involved in E . faecalis virulence and CAMP response [30] , was overexpressed in the mutant strain . It is likely that some members of the AsrR regulon are under several transcriptional controls . Currently , except for ohr , the possibility that AsrR modulates gene expression in E . faecium through interactions with other regulators cannot be excluded . Such a regulatory cascade has been shown for E . faecalis , S . aureus , and Salmonella enterica [13] , [74] , [75] . The present investigation provides evidence that AsrR plays a key role in E . faecium adaptation , antimicrobial resistance and pathogenicity , which is summarized in a model ( Figure 10 ) . AsrR , which is inactivated in the presence of oxidative stress , is a global repressor ( direct or indirect ) of expression of the genes involved in all the important steps during the early infection process and allows host-colonization ( Figure 10 ) . It has been shown that nitric oxide-mediated activation of bacterial defence is important for the in vivo virulence of Bacillus anthracis [76] . Similarly , it can be speculated that the oxidative stress acts as a signal to promote the transition from the commensal to the opportunistic state thus rendering E . faecium more pathogenic .
The strains and plasmids used in this study are listed ( Table S1 ) [7] , [77]–[80] . E . faecium was grown at 37°C in Brain Heart Infusion ( BHI ) , Mueller-Hinton ( MH ) or Trypticase-Soy ( TS ) broth , or on BHI agar ( Difco Laboratories ) . Escherichia coli were grown in Luria-Bertani ( LB ) broth or on LB agar ( Difco Laboratories ) . When appropriate , antibiotics were added at the following concentrations: ampicillin 100 µg/ml; erythromycin 150 µg/ml for E . coli and 50 µg/ml for E . faecium; fusidic acid 40 µg/ml; kanamycin 30 µg/ml; lincomycin 30 µg/ml; rifampin 50 µg/ml; spectinomycin 150 µg/ml; streptomycin 150 µg/ml and tetracycline 100 µg/ml . Plasmid pG ( + ) host9 is a temperature-sensitive E . coli–Gram-positive shuttle vector used for allelic replacement in Gram-positive bacteria [77]; pORI23 is an E . coli-Gram-positive shuttle vector used for complementation studies in Lactococcus lactis [78] . Vector pCR2 . 1-TOPO ( Invitrogen ) was used as recommended by the manufacturer for TA sub-cloning and cloning steps . Chromosomal DNA isolation , restriction endonuclease digestion , DNA ligation , and transformation of electrocompetent cells were performed using standard protocols or manufacturers instructions . For RACE-PCR and qRT-PCR , total E . faecium RNA was isolated using the RNeasy midi kit ( Qiagen ) as recommended by the manufacturer . For microarray experiments , total RNA was isolated as follows . Strains were cultured overnight in 3 ml of BHI broth within 10 ml tubes at 37°C with aeration by rotary shaking at 150 rpm and pre-warmed BHI broth ( 25 ml in 50-ml Falcon tube ) was inoculated in duplicate with the overnight culture to a starting absorbance at 600 nm ( OD600 ) of 0 . 025 and then incubated at 37°C as described above . One culture was used to monitor growth by measuring OD600 . When OD600 reached 0 . 5 ( mid-exponential growth phase ) , RNA was isolated from the second culture and the OD600 was measured to confirm equal growth in the duplicate cultures . For RNA isolation , 2 ml of each culture were transferred into an Eppendorf tube and spun down at 13 , 000×g for 20 s and the cell pellets were snap-frozen in liquid nitrogen . The time between removal from the incubator and freezing of the cell pellets was approximately 60 s . Within 20 min after freezing , 1 ml of TRI reagent ( Ambion ) was added to the frozen pellets and the suspension was transferred into a 2-ml tube filled with 0 . 5 g of 0 . 1 mm zirconia/silica beads ( Biospec ) . Cells were disrupted by beadbeating three times for 1 min with intermittent cooling on ice . RNA was then extracted following Ambion's TRI reagent protocol . Residual chromosomal DNA was removed by treating samples with the TURBO DNA-free kit ( Ambion ) . Extracted RNA samples were quantified using a Nanodrop 1000 spectrophotometer ( Isogen Life Science ) and stored in 70% ethanol-83 mM sodium acetate buffer ( pH 5 . 2 ) at −80°C . PCR amplification was carried out in a final volume of 50 µl containing 40 pmol of each oligonucleotide primer , ca . 100 ng of template DNA , using the GoTaq Flexi DNA polymerase kit ( Promega ) as recommended by the manufacturer . Primers used were designed based on the E . faecium HM1070 asrR cluster sequence ( Accession number JQ390466 ) ( Table S2 ) [81] . The transcriptional start sites of asrR and ohr were determined using the 5′ RACE system kit ( Invitrogen ) according to the manufacturer's instructions . The specific RACE-PCR primers were designed using the Primer3 software ( http://frodo . wi . mit . edu/primer3 ) ( Table S2 ) . For RT-PCR , cDNA was synthesized from total RNA ( ∼1 . 5 µg ) by using the Superscript III First-Strand Synthesis System ( Invitrogen , Breda , The Netherlands ) according to the manufacturer's instructions . Using synthesized cDNAs , qRT-PCR was performed using Maxima SYBR Green/ROX qPCR Master Mix ( Fermentas , St . Leon-Rot , Germany ) and a StepOnePlus instrument ( Applied Biosystems , Nieuwekerk a/d IJssel , The Netherlands ) with the following program: 95°C for 10 min , and subsequently 40 cycles of 95°C for 15 sec , 55°C for 1 min . Relative transcript levels were calculated using REST 2009 Software ( Qiagen ) . Expression of tufA was used as a housekeeping control gene . The asrR deletion mutant ( ΔasrR ) was derived from E . faecium HM1070 by allelic exchange with a truncated copy of asrR as described [77] . Approximately 500 bp fragments upstream and downstream of asrR were amplified by PCR using HM1070 chromosome as template and primer pairs asrR-DC1-F/asrR-DC2-R and asrR-DC3-F/asrR-DC4-R ( Table S2 ) [81] . The forward primer binding to the 3′-end of asrR ( asrR-DC3-F ) and the reverse primer to the 5′-end ( asrR-DC2-R ) were modified to carry the same restriction site ( Table S2 ) [81] . Following restriction , ligation and amplification using asrR-DC1-F/asrR-DC4-R , the resulting fragment carrying the truncated asrR copy was cloned in the temperature-sensitive shuttle vector pG ( + ) host9 to create plasmid pG ( + ) host9ΩasrR-KO ( Table S1 ) [7] , [77]–[80] . The hybrid plasmid was introduced in the chromosome of HM1070 by electrotransformation and homologous recombination followed by excision of the wild-type copy as described [77] . In-frame deletion of the asrR gene was confirmed by PCR and sequencing . As described in Figure 1 , around 42% of the sequence containing the helix-turn-helix DNA binding domain and the second Cys residue were deleted in the ΔasrR strain . No significant difference was found when comparing the growth kinetics of the parental and mutant strains ( Figure S5 ) . An in trans complemented ΔasrR/pOri23ΩasrR strain was constructed ( Table S1 ) [7] , [77]–[80] , and complementation was confirmed in the in vitro physiological tests used in this study as well as in the murine macrophages experiments . Since only partial complementation was found with this construct ( data not shown ) , we decided to construct the knock-in complemented strain , ΔasrR::asrR ( Table S1 ) [7] , [77]–[80] . Subsequently , all the experiments were conducted using the ΔasrR::asrR strain except for those with the mouse macrophages . For asrR in trans complementation , the asrR coding sequence from E . faecium HM1070 including the predicted ribosomal binding site was amplified with the primer pair asrR-pOri23-F/asrR-pOri23-R ( Table S2 ) [81] and cloned in pOri23 ( Table S1 ) [7] , [77]–[80] . The resulting plasmid pOri23ΩasrR was introduced into the ΔasrR mutant strain by electrotransformation ( Table S1 ) [7] , [77]–[80] . For asrR knock-in complementation , the entire asrR sequence of HM1070 was amplified with the asrR-DC1-F/asrR-DC4-R primers ( Table S2 ) [81] and cloned in pG ( + ) host9 to create plasmid pG ( + ) host9ΩasrR-KI which was introduced into the ΔasrR strain ( Table S1 ) [7] , [77]–[80] , excised and allele replacement was obtained and verified as described above . A 460-bp fragment encoding AsrR was amplified by PCR from E . faecium HM1070 chromosome using primers AsrR-F and AsrR-R ( Table S2 ) [81] . The product was cloned in pQE30 ( Qiagen ) downstream of the RGS-His6 tag sequence ( Table S1 ) [7] , [77]–[80] . The pQE30ΩasrR plasmid was electroporated in E . coli M15[pREP4] ( Table S1 ) [7] , [77]–[80] , and expression of the his-tagged recombinant peptide was performed using IPTG induction ( 1 mM final concentration ) as described [82] . Briefly , purification from E . coli M15[pREP4]/pQE30ΩasrR lysates was achieved by Ni2+-affinity chromatography using Ni-NTA resin ( Qiagen ) under native conditions . Samples were desalted on PD-10 columns ( Amersham Biosciences ) and protein concentrations were determined using the Bio-Rad protein assay . DNA fragment from the ohr and asrR promoter regions was amplified , labelled by PCR with [γ-32P]dATP and incubated with purified His6-tagged AsrR ( 10 to 200 ng ) in interaction buffer [40 mM Tris HCl [pH 7 . 5] , bovine serum albumin 200 µg/ml , 2 mM CaCl2 , 2 mM dithiothreitol and poly ( dI-dC ) µg/ml] at room temperature for 30 min . Designated amounts of H2O2 and or DTT were used as previously described [26] . The DNA-AsrR mixtures were electrophoresed in 12 . 5% polyacrylamide gels in 0 . 5× Tris-borate-EDTA ( TBE ) at 180 V that were dried and analyzed by autoradiography . DNase I footprinting assays were performed as previously described [82] using a D-4 labelled DNA fragment of the ohr promoter . The capillary electrophoresis was performed using a CEQ8000 sequencing apparatus ( Beckman Coulter ) . The determination of the DNA sequence of the protected region was performed after co-migration of the footprinting assay and the corresponding sequence reaction . The MEME suite ( http://meme . sdsc . edu/meme/intro . html ) was used to generate a putative AsrR binding box logo on several DNA sequences of regulon members . The E . faecium E1162 sequence was used as a reference and gene tags or ORFs numbers are indicated according to its annotation ( Genbank accession ABQJ00000000 ) [7] . Transcriptome comparisons were performed between the ΔasrR mutant , the parental HM1070 and the ΔasrR::asrR complemented strains grown to mid-exponential ( OD600 = 0 . 5 ) phase . For each strain , bacterial RNA was extracted from four independent cultures as described above , and used for cDNA synthesis and labelling as detailed below . RNA samples were prepared and labelled with Cy3 or Cy5 as previsouly described [83] . Dyes were switched between samples to mimimize the effect of dye bias . E . faecium microarrays ( Agilent , Palo Alto , CA ) were hybridized with 300 ng labelled cDNA . The experiments for comparison of the transcriptomes of the ΔasrR mutant , parental HM1070 or complemented ΔasrR::asrR strains were performed with four independent biological replicates . Slides were then scanned using an Agilent Technologies Scanner G2505B . Data were extracted from the scanned microarrays with Agilent Feature Extraction software ( version 10 . 7 . 1 ) , which includes a Lowess normalization step for the raw data . The microarrays used in this study were custom-made E . faecium E1162 arrays using Agilent's 8×15K platform ( containing 8 microarrays/slide ) , as described previously [83] . After removal of the data for the different controls printed on the microarray slides , the normalized data for each spot were analyzed for statistical significance using the Web-based VAMPIRE microarray suite ( http://sasquatch . ucsd . edu/vampire/ ) [84] , [85] . A spot was found to be differentially expressed between two samples using the threshold of a false discovery rate smaller than 0 . 05 . An open reading frame was found to be differentially expressed when all four spots representing the open reading frame were significantly differentially expressed ( False Discovery Rate for each spot <0 . 05 ) between samples . The average expression ratio of each significantly regulated open reading frame was determined by calculating the log-averages of the expression ratios of each individual probe . Finally , changes of 2-fold for upregulated and 0 . 5-fold for downregulated genes in the mutant strain were also introduced as biological significance limits . Microarray data were submitted to the MIAMExpress database and are accessible under accession number no . E-MEXP-3528 . The nucleotide sequence of the ohr-asrR region in E . faecium HM1070 has been deposited in the GenBank database under accession no . JQ390466 . Minimum inhibitory concentrations ( MIC ) were determined by the broth microdilution technique as recommended by the Comité de l'Antibiogramme de la Société Française de Microbiologie ( http://www . sfm-microbiologie . org ) [86] . For the determination of mutation frequencies , ca . 1010 cells from an overnight broth culture were plated onto BHI agar plates supplemented with spectinomycin and the mutation frequency was determined relative to the count of viable organisms plated in four independent experiments . Time–kill curves were determined for exponentially growing enterococcal cultures and an antibiotic concentration equal to 4× the MIC as described [87] . Briefly , bacteria were inoculated 1∶20 in 10 ml of fresh MH broth containing antibiotic and incubated at 37°C . Bacterial survival was monitored by CFU counts after 0 , 3 , 6 , 24 , and 48 h of incubation in three independent experiments by plating the cultures on BHI agar plates . Transfer of Tn916 carrying tetracycline resistance from strains S . agalactiae UCN78 ( Table S1 ) [7] , [77]–[80] to E . faecium HM1070 , ΔasrR and ΔasrR::asrR was attempted by filter mating . Transconjugants were selected on BHI agar plates containing tetracycline , rifampicin and fusidic acid . For each strain , three transconjugants were selected and used in the following experiment to quantify the influence of the integration site on the transfer frequency . Transfer of Tn916 from strains HM1070/Tn916 , ΔasrR/Tn916 and ΔasrR::asrR/Tn916 to E . faecalis BM4110 was attempted as described above . Transconjugants were selected on BHI agar plates containing tetracycline , streptomycin and lincomycin whereas parental donor cells were selected on BHI agar plates containing rifampicin , fusidic acid and tetracycline . Transfer frequency data are of three independent experiments and statistical analysis was performed with the two-tailed Student's t test . E . faecium ΔasrR , parental HM1070 , and complemented strains grown in exponential growth phase in BHI were harvested , washed twice with cold phosphate-buffered saline ( PBS; Gibco ) , resuspended in the same buffer supplemented with 0 . 1% Triton X-100 ( Sigma ) , incubated at 37°C without shaking and autolysis was monitored by measuring the decrease in OD600 on a microplate reader system ( Multiskan Ascent , Thermo Electron Corporation ) . The initial OD600 value was fixed at 100% , and the results are the means ( ± standard deviation ) from three independent experiments . E faecium HM1070 , ΔasrR mutant and complemented strains suspensions were standardized to an OD600 of 1 in 0 . 9% NaCl and 10 µl aliquots of 10-fold dilutions were spotted on BHI agar plates supplemented with various amounts of colistin methanesulfate ( Sigma ) and nisin ( Sigma ) . Experiments were repeated at least three times and representative data are shown . E faecium HM1070 , ΔasrR mutant , and complemented strains suspensions were standardized to an OD600 of 1 in 0 . 9% NaCl and 10 µl aliquots of 10-fold dilutions were spotted on BHI agar plates supplemented with various amounts of menadione , tertiary-buthylhydroperoxide , and cumene hydroperoxide . Experiments were repeated at least three times and representative data are shown . Resistance of E . faecium to oxidative killing by H2O2 was tested as described with slight modifications [88] . Bacteria were grown 16 h in BHI broth and sub-cultured in 10 ml broth at a starting density of OD600 at 0 . 05 . Cultures were grown to mid-exponential phase ( OD600 = 0 . 5 ) or to stationary phase ( OD600 = 1 . 4 ) , harvested by centrifugation , resuspended in 0 . 9% NaCl with 2 mM H2O2 , placed into a 37°C water bath , and samples were enumerated on plates immediately and 30 min following H2O2 challenge . For H2O2 killing assays in the presence of iron or superoxide anion scavengers , cultures ( OD600 = 0 . 5 ) were resuspended in 2 mM H2O2-containing 0 . 9% NaCl supplemented with 100 µM of deferoxamine or 3 . 3 mM tiron , respectively , and processed as described [88] . The detection of intracellular hydroxyl radical was carried out as described [89] . All data were collected using a Epics XL Beckman Coulter flow cytometer with a 488 nm argon laser and a 505–545 nm emission filter ( FL1 ) at low flow rate . In all experiments , cells were grown as described above , stressed with 0 , 0 . 5 , or 2 mM of H2O2 during 10 min and washed with PBS buffer . At least 30 , 000 cells were collected for each sample . To detect hydroxyl radical formation , we used the fluorescent reporter dye 3′- ( p-hydroxyphenyl ) fluorescein ( HPF; Invitrogen ) at a concentration of 10 µM . Flow data were processed and analyzed with Kaluza V1 . 2 . Bacteria that had been grown overnight were inoculated 1∶100 in 10 ml of TS broth with 0 . 25% glucose and shared into 96-microwell polystyrene plates ( NUNC , Denmark ) . After 24 h of static incubation at 37°C , the plates were washed three times with PBS and stained with 1% crystal violet for 30 min . The wells were rinsed with distilled water and ethanol-acetone ( 80∶20 , vol/vol ) . After drying , OD600 was determined using a microplate reader ( Multiskan Ascent , Thermo Electron Corporation ) . Each assay was performed in triplicate in at least three independent experiments . For visualization , bacteria were grown as described above using 12-well polystyrene plates ( CytoOne , Starlab International , Germany ) and were similarly processed than above and directly examined . The in vivo colonization model was carried out as described [42] . Galleria larvae were infected ( 1 . 8×106±0 . 5×106 CFU/larva ) and batches of 10 alive larvae were sacrificed at 0 , 24 , 48 , and 72 h post-infection and homogenized as described previously [42] . The t0 time point was determined immediately following injection . Homogenates were plated onto BHI agar plates containing aztreonam ( 100 µg/ml ) and rifampicin ( 60 µg/ml ) , and CFU were counted after 24 h of incubation . Results represent means ( ± standard deviation ) of at least three independent experiments . Mouse experiments were performed with the approval of an institutional animal use committee ( see below ) . Mice were housed in filter-top cages and had free access to food and water . The in vivo-in vitro model of survival within murine macrophages was carried out as described [90] . Briefly , E . faecium strains were grown in BHI for 16 h , pelleted and resuspended in an adequate volume of PBS for injection . Male BALB/c mice ( 10 weeks old; Harlan Italy S . r . l . ) were infected with 107 to 108 cells by intraperitoneal injection and after 6 h infection macrophages were collected by peritoneal lavage , centrifuged , and suspended in DMEM containing 10 mM HEPES , 2 mM glutamine , 10% bovine fetal serum , and 1× nonessential amino acids supplemented with vancomycin ( 10 µg/ml ) and gentamicin ( 150 µg/ml ) . The cell suspension was dispensed into 24-well tissue culture plates and incubated at 37°C under 5% CO2 for 2 h . After exposure to antibiotics to kill extracellular bacteria ( i . e . , at 8 h postinfection ) , the infected macrophages were washed , and triplicate wells of macrophages were lysed with 0 . 1% sarkosyl . Note that , nor the HM1070 strain , neither the ΔasrR mutant were found to be sensitive to the lytic treatment ( data not shown ) . The lysates were diluted in BHI broth and plated on BHI agar to quantify the number of viable intracellular bacteria . The remaining wells were maintained in DMEM with the antibiotics for the duration of the experiment . The same procedure was performed at 24 , 48 , and 72 h post-infection . All experiments were performed at least three times . The intravenous systemic infection model was performed as described previously [91] . Briefly , overnight cultures of the strains grown in BHI broth supplemented with 40% heat-inactivated horse serum were centrifuged , and the resulting pellets were resuspended in sterile PBS to achieve final concentrations of 1×109 cells/ml . Aliquots of 100 µl from each strain suspension were used to inject the tail veins of groups of 10 mice each . The infection experiments were repeated three times . The mice were monitored with twice-daily inspections , and 1 , 3 , and 7 days after infection they were killed using CO2 inhalation . The kidneys and livers were then removed aseptically , weighed , and homogenized in 5 ml of PBS for 120 s at high speed in a Stomacher 80 apparatus ( Pbi International ) . Serial homogenate dilutions were plated onto Enterococcus Selective Agar ( ESA; Fluka Analytical ) to determinate the CFU numbers . All experiments were performed three times . The mouse experiments were performed under a protocol approved by the Institutional Animal Use and Care Committee at the Università Cattolica del Sacro Cuore , Rome , Italy ( Permit number: Z21 , 11/01/2010 ) and authorized by the Italian Ministry of Health , according to the Legislative Decree 116/92 , which implemented the European Directive 86/609/EEC on laboratory animal protection in Italy . Animal welfare was routinely checked by veterinarians of the Service for Animal Welfare . Comparisons between groups were performed with different statistical tests ( one-way analysis of variance with a Bonferroni correction post test or non-parametric Wilocoxon signed-rank sum test ) using GraphPad Prism software ( version 5 . 00 ) for Windows ( GraphPad Software , San Diego , CA ) . For all comparisons , a P value of less than 0 . 05 was considered as significant .
|
Multiple antibiotic‐resistant isolates of the opportunistic pathogen Enterococcus faecium have emerged and spread worldwide . However , studies aimed at identifying mechanisms that underlie the transformation of E . faecium from its commensal nature into a nosocomial pathogen are scarce . We report pleiotropic roles for a novel oxidative‐sensing regulator , called AsrR ( antibiotic and stress response regulator ) , in E . faecium . Based on transcriptomic analysis , phenotypic studies , and animal models , we demonstrate that asrR deletion is responsible for i ) diminished susceptibility to penicillins , vancomycin , and cationic antimicrobial peptides , ii ) increased adhesion to human cells and biofilm formation , iii ) a mutator phenotype and enhanced DNA transfer frequencies , iv ) decreased resistance to oxidative stress both in vitro and in murine macrophages , and v ) increased host‐persistence in both insect and mouse models . AsrR is a stress‐sensor and is promptly inactivated in the presence of hydrogen peroxide . Therefore , oxidative stress , which is a main challenge during infection , may be a significant signal used by E . faecium to promote opportunistic traits . This provides a significant resource combining , for the first time in E . faecium , a global transcriptomic approach and a thorough phenotypic study , which places AsrR as a key regulator modulating pathogenicity , antimicrobial resistance , and environmental adaptation .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"medicine",
"bacterial",
"diseases",
"pathogenesis",
"infectious",
"diseases",
"enterococcus",
"infection",
"medical",
"microbiology",
"microbial",
"pathogens",
"biology",
"microbiology",
"bacterial",
"pathogens",
"gram",
"positive"
] |
2012
|
AsrR Is an Oxidative Stress Sensing Regulator Modulating Enterococcus faecium Opportunistic Traits, Antimicrobial Resistance, and Pathogenicity
|
Pluripotent cells such as embryonic stem ( ES ) and induced pluripotent stem ( iPS ) cells are the starting point from which to generate organ specific cell types . For example , converting pluripotent cells to retinal cells could provide an opportunity to treat retinal injuries and degenerations . In this study , we used an in vivo strategy to determine if functional retinas could be generated from a defined population of pluripotent Xenopus laevis cells . Animal pole cells isolated from blastula stage embryos are pluripotent . Untreated , these cells formed only epidermis , when transplanted to either the flank or eye field . In contrast , misexpression of seven transcription factors induced the formation of retinal cell types . Induced retinal cells were committed to a retinal lineage as they formed eyes when transplanted to the flanks of developing embryos . When the endogenous eye field was replaced with induced retinal cells , they formed eyes that were molecularly , anatomically , and electrophysiologically similar to normal eyes . Importantly , induced eyes could guide a vision-based behavior . These results suggest the fate of pluripotent cells may be purposely altered to generate multipotent retinal progenitor cells , which differentiate into functional retinal cell classes and form a neural circuitry sufficient for vision .
Every body organ consists of tissue groups with multiple cell types . Therefore , recovery from organ damage requires the replacement of a variety of distinct cell types . The retina is the light detecting tissue of the eye , and consists of seven major cell classes . All retinal cell classes are generated from a common multipotent retinal progenitor . The conversion of pluripotent cells to retinal progenitors would , in theory , provide a source of all the cell classes necessary for retinal repair . Pluripotent cells treated with extrinsic factors express retinal markers in vitro and in vivo [1]–[5] . When transplanted to the subretinal space of mice lacking functional photoreceptors , human embryonic stem cells directed toward a retinal lineage integrate into the outer nuclear layer , express photoreceptor markers , and restore a light response as determined by the electroretinogram ( ERG ) [5] . However , it is not known if any other functional retinal cell classes can be derived from pluripotent cells , or if these cells can form the complex neural network necessary for vision . Vertebrate retina formation begins in the anterior neural plate in a region called the eye field . A group of transcription factors collectively called the eye field transcription factors or EFTFs ( pax6 , rx1 , tbx3 [or ET] , nr2e1 [or tailless] , six3 , lhx2 , and six6 [or optx2] ) and the neural patterning gene otx2 are essential for normal eye formation [6] . When overexpressed together in developing Xenopus embryos by RNA microinjection , these genes can induce eye-like structures as defined by the expression of markers for some retinal cell classes [7] . The overall aim of this study was to determine if pluripotent cells overexpressing these transcription factors could be intentionally driven toward retinal progenitors that differentiate into multiple retinal cell classes and form a functional retina . Primitive ectoderm cells isolated from the animal pole of blastula stage embryos are pluripotent . If treated with the appropriate inducer , they can form endodermal , mesodermal , or ectodermal cell types [8] . Here we provide evidence that these pluripotent cells misexpressing the EFTFs differentiate into all retinal cell classes . Using both the ERG and a vision-based behavioral assay , we found that eyes generated from EFTF-expressing pluripotent cells are molecularly and functionally similar to the normal eye . These results suggest that by using the correct combination of gene products , it may be possible to reprogram more readily available pluripotent cell types to a multipotent retinal cell lineage useful in healing damaged or diseased retinas .
Coordinated misexpression of the EFTFs ( pax6 , tbx3 , rx1 , nr2e1 , six3 , and six6 ) and otx2 in developing embryos results in formation of ectopic eye tissue [7] . One interpretation of this result is that EFTFs induce an eye field-like fate in pluripotent cells . This interpretation predicts that transcripts expressed in the eye field should also be induced in EFTF-expressing pluripotent cells . Therefore , we used microarray analysis to perform pairwise comparisons of the transcriptional profile of EFTF-expressing pluripotent cells to three regions of the Xenopus embryo: the eye field ( EF ) including underlying mesendoderm , posterior neural plate ( PNP ) and the non-neural , flank , lateral endoderm ( LE , Figure 1A ) . Of approximately 14 , 400 transcripts evaluated by the Affymetrix GeneChip X . laevis Genome Array , 365 were induced greater than 2-fold in EFTF-expressing pluripotent cells . Of the 365 EFTF-induced transcripts , 31 . 5% , 8% , and 0 . 5% were uniquely shared with EF , PNP , or LE , respectively ( Figure 1B ) . We used an unsupervised hierarchical clustering algorithm to compare the transcriptional profiles of all samples . This analysis compares the relative change of all probe sets on the array and unbiasedly groups the four tissue samples based on their similarity [9] . These results indicate the eye field and EFTF-expressing pluripotent cells are more closely related to each other than to either the PNP or LE tissues ( Figure S1 ) . Importantly , 25 of the 115 transcripts , shared by EFTF-expressing pluripotent cells and the EF , encode for 15 genes that are both expressed in retinal stem/progenitor cells and required for normal eye formation in frogs , fish , mice , or humans ( Figure 1C; Table S1 ) . The expression levels for 13 of these transcripts were not significantly different when comparing eye field and EFTF-expressing pluripotent cells ( p<0 . 05 considered significant ) . By comparison , only two of the 15 genes were not significantly different when comparing the eye field to PNP or LE data sets ( Table S1 ) . These results suggest EFTF-expressing pluripotent cells share a common transcriptional profile with the eye field , and are consistent with the hypothesis that the EFTFs direct nonretinal pluripotent cells to an eye field-like retinal cell lineage . The EFTFs may transiently alter the pattern of gene expression in pluripotent cells , yet fail to stably direct cells toward a retinal cell fate . To test this possibility , we transplanted EFTF-expressing pluripotent cells to the flank of X . laevis embryos . To ensure that donor and host tissues could be easily distinguished , we generated EFTF-expressing cells from transgenic embryos constitutively expressing a variant of yellow fluorescent protein ( Venus YFP ) [10] . Transplanted control pluripotent cells isolated from Venus YFP embryos formed only sheets of epidermis ( number of animals or eyes [N] = 108; unpublished data ) . In contrast , EFTF-expressing pluripotent cells formed YFP-expressing , pigmented spheroids in 23% of transplants ( N = 566 ) . We compared the morphology of the induced tissues to the normal eye ( Figure 2A and 2B ) . Thirty-one transplants ( 24% of pigmented spheroids ) had an eye-like morphology consisting of retinal pigment epithelium ( RPE ) and the trilayered cup structure of a normal retina ( Figure 2C and 2D ) . We used immunofluorescence and in situ hybridization to detect specific retinal cell classes . All 31 expressed YFP throughout and molecular markers for two or more retinal cell classes . Markers used included Islet-1 and hermes for retinal ganglion cells , Tyrosine Hydroxylase for amacrine cells , R5 for Müller glia , XAP2 for rod photoreceptors , and Calbindin for cone photoreceptors ( Figure 2D , 2E , 2G , and 2H; Tables S3 and S4 ) . The inner nuclear layers of flank retinas were also strongly labeled for gamma-aminobutyric acid ( GABA ) , which stains horizontal , and a subset of amacrine cells , and Calretinin , which predominantly labels bipolar cells but also subsets of amacrine and retinal ganglion cells ( Figure 2F; Tables S3 and S4 ) . These results suggest that EFTF-expressing cells , like bona fide eye field cells , are determined , forming retinal cells and even eye-like structures , when transplanted to another region of the embryo . In addition to differentiated retinal cells , the amphibian eye contains a population of self-renewing retinal stem cells located in the retinal periphery in a region called the ciliary marginal zone or CMZ . Retinal stem cells are derived from the embryonic eye field and are the source of new retinal cells throughout the life of the animal [11] , [12] . To determine if mitotic cells were present in the periphery of flank eyes , tadpoles were briefly placed in a solution containing 5-bromo-2-deoxyuridine ( BrdU ) . BrdU immunoreactivity was detected in all retinas tested ( Figure 2H; Table S4 ) . To determine if mitotic cells were enriched in the peripheral retina , we determined the relative proportion of BrdU-positive cells in the peripheral and central retina . We found that 12 . 5% of peripheral cells yet only 0 . 6% of central cells were BrdU-labeled in flank retinas ( 2 , 063 cells from ten flank retinas were counted , p<0 . 001 ) . We observed similar percentages in wild-type retinas , which contain 18 . 6% and 0 . 12% BrdU-labeled peripheral and central cells , respectively ( 2 , 672 cells counted in ten control retinas ) . Based on a Student's paired t-test , there was no statistically significant difference in the percentage of BrdU-labeled cells when control and flank eyes were compared ( p>0 . 1 ) . Despite the remarkable ability of EFTF-expressing pluripotent cells to form eye-like structures on the tadpole flank , it was not possible to record ERGs because of the small size of the induced flank eyes ( flank eye volume: 0 . 007±0 . 003 mm3; control eye volume: 0 . 021±0 . 002 mm3 , at stage 41 , N = 5 ) . In addition , retinal ganglion cell axons exiting the flank eyes failed to reach their normal tectal targets . When DiI was applied to the RGC layer of induced eyes , labeled processes were observed projecting dorsocaudally along the spinal cord away from the tectum ( unpublished data ) . Similar paths are taken by RGC axons exiting ectopic eyes generated by transplantation of eye primordia [13]–[15] . Misrouting has been attributed to the absence of the directional cues required for normal RGC axon guidance . These observations prompted us to replace the eye field with EFTF-expressing pluripotent cells to ask if the induced retinal cells could generate a normal eye and if induced retinal ganglion cells would extend axons to and synapse with their normal tectal targets . Xenopus embryos from which an eye field has been removed survive , develop normally , but lack an eye on the operated side . We grafted donor ( YFP-only or EFTF/YFP-expressing ) pluripotent cells to host embryos from which one of the two eye fields had been removed ( Figure 3A and 3B ) . When cultured in isolation , pluripotent cells isolated from blastula stage embryos autonomously form epidermis , loosing competence to form other structures . We reasoned , however , if cells were successfully reprogrammed to multipotent retinal cells , and grafted to the anterior neural plate , an eye should form . YFP-expressing pluripotent cells grafted in place of the eye field never formed eyes ( Figure 4A; N = 79 ) . As expected , these cells generated only epidermis ( Figure 4A–4C ) . In contrast , EFTF-expressing pluripotent cells generated YFP-labeled eyes in 63% of transplants ( N = 74 ) . The majority of transplants created mosaic eyes ( 55% ) , likely a consequence of the difficulty in surgically removing all traces of the eye field . Nevertheless , 8% of the transplants formed retinas expressing YFP throughout and were therefore completely donor derived . We focused our attention on eyes derived entirely from the EFTF-expressing pluripotent cells ( Figure 4D–4H ) . We observed YFP-labeled processes exiting the induced eye at the optic nerve head that followed the expected trajectory of RGC axons to their tectal targets ( Figure 4E ) . We also observed some YFP-labeled cells in the epidermis , forebrain , and occasionally olfactory bulb ( Figure 4E and 4F and unpublished data ) . Induced eyes contained a lens , RPE , and the trilayered structure of a normal retina ( Figure 4G ) . Interestingly , while the retina and RPE of induced eyes were YFP-labeled , lens cells were most often not , indicating that the lens formed from host tissue ( Figure 4G ) . This result is consistent with previous experiments in Xenopus demonstrating the inductive cues and origin of the retinal and lens primordia are distinct [16] , [17] . These observations suggest the EFTFs can direct pluripotent cells to an embryonic eye field-like cell fate . If this hypothesis is correct , transplantation of eye fields should yield similar results . Therefore , we transplanted embryonic eye fields from YFP transgenic embryos to wild-type hosts . Embryonic eye fields formed mosaic and complete eyes at frequencies similar to those observed with EFTF-expressing pluripotent cells ( mosaic eyes 68% , complete eyes 16%; N = 38; Figure 4I and 4J ) . These results suggest EFTF-expressing pluripotent cells are directed to an embryonic eye field-like cell fate . Interestingly , in addition to the tissues formed by EFTF-expressing pluripotent cells , transplanted eye fields also formed muscle and head mesenchyme ( Figure 4J and unpublished data ) . If the EFTFs induce multipotent retinal progenitor cells , every retinal cell class , including retinal stem cells , should form in induced eyes . To better discern cell classes and determine how well donor cells integrated into the retina we deliberately generated mosaic eyes . EFTF/YFP expressing pluripotent cells were transplanted to embryos from which only one-half of one eye field had been removed ( Figure 3C ) . This allowed us to directly compare the patterns of marker expression in normal and induced retina , and identify cell classes on the basis of their distinctive morphology and location within the eye . Control , YFP-only expressing pluripotent cells never generated mosaic retinas ( Figures 5A and S2A; N = 57 ) . This demonstrates that even though the anterior neural plate is essential for proper eye morphogenesis , cultured pluripotent cells were not directed to a retinal fate—even when grafted directly into the eye field . In striking contrast , cultured EFTF-expressing pluripotent cells generated the seven classes of differentiated retinal cells observed in the normal retina . Retinal cell classes were identified based on their morphology , location within the retina , and using cell class specific markers ( Figures 5 and S2 , S3 , S4 , S5 , S6 ) . RGCs lie on the vitreal surface of the retina adjacent to the lens . RGC axons are the only neural processes that leave the retina . Consistent with the presence of donor-derived RGCs , YFP-labeled processes were detected exiting the eye at the optic nerve head ( Figures 4E , 5B , and S2B ) . All other cell classes were similarly identified on the basis of their morphology and retinal location ( Figures 5C–5J , S2 , S3 , S4 , S5 , S6 ) . Molecular markers for retinal ganglion , amacrine , bipolar , horizontal , Müller glia , and rod and cone photoreceptor cells ( Table S3 ) identified these cell types ( Figures 5B , 5D–5N , and S2 , S3 , S4 , S5 , S6 ) . The expression patterns of cell class-specific markers in mosaic retinas appeared the same in host and donor derived regions suggesting that donor cells generated all retinal cell classes in approximately normal ratios ( Figures 5K–5N and S2 ) . Consistent with this idea , we observed no significant difference in retinal ganglion , inner nuclear layer , or rod photoreceptor cell density when endogenous and induced regions of mosaic retinas were compared ( Figure 5O ) . We did not stain every mosaic retina for all seven retinal cell classes . However , we typically stained each mosaic retina for three or four cell class-specific markers and always detected cells expressing every marker tested ( three of three or four of four ) . In addition , every mosaic retina contained columns of YFP-positive cells that spanned the entire width of the retina—from RPE to RGCs ( see Figures 5B , 5K–5N , and S2B–S2J for examples ) . The layering within these columns was indistinguishable from the adjacent , control retina and contained cells with the appropriate morphology for their nuclear layer ( rods and cones in the ONL , bipolar , amacrine , horizontal , and Müller glia in the INL , and RGCs in the GCL ) . These results suggest that EFTF-expressing cells are multipotent , as they differentiate into the seven retinal cell classes of the mature retina . We also cultured animals in BrdU and used immunocytochemistry to detect mitotically active cells in mature retinas generated from EFTF-expressing cells . BrdU immunoreactivity was detected in the dorsal and ventral , donor-derived CMZ of every animal tested ( Figure 5P and 5Q; N = 43 ) . BrdU immunoreactivity is not conclusive evidence for the presence of retinal stem cells . However , when coupled with the observation that induced eyes continued to grow throughout the life of the animals ( Figure 4D and 4H ) , these results suggest EFTF-induced eyes contain a self-renewing population of retinal cells in the retinal stem cell niche of the CMZ . Normal anatomical structure and the expression of cell class-specific markers does not demonstrate normal cellular function and connectivity . Therefore , we next measured the ERG to determine if induced cells could form the intricate neural network necessary to detect and process light stimuli . ERGs recorded from eyes wholly derived from induced retinal cells were similar to those recorded from eyes on the unoperated side of the same animals and eyes of stage-matched wild-type controls ( Figure 6A and 6B ) . ERG responses of control and induced eyes were similar in sensitivity and time course ( Figure 6A ) . Rod and cone photoreceptors in the outer retina transduce light into electrical signals . Bipolar cells relay the visual information from photoreceptors to the inner retina for further processing . Activation of bipolar cells elicits the characteristic b-wave of the ERG ( Figure 6A; arrows ) . In response to green ( 520 nm ) , dim flashes ( 0 . 4 photons/µm2 ) the b-waves are small and exhibit slow kinetics . In response to brighter flashes , b-waves were progressively larger and faster . Figure 6B shows the amplitude of the b-waves plotted as a function of flash intensity . b-waves increased progressively with intensity , saturating in response to the brightest flashes ( >30 photons/µm2 ) . The b-waves of the induced eyes were slightly desensitized ( approximately 30% ) relative to controls , as reflected by the slight shift of the Michaelis-Menten functions fit to the data ( Figure 6B ) . Immunostaining for YFP confirmed the eyes from which ERGs were recorded were completely derived from donor tissue ( unpublished data ) . ERGs not only indicate the presence of mature retinal cells , but also demonstrate that light enters the eye appropriately , phototransduction takes place , synapses form between retinal neurons , and synaptic transmission is functional within the eye . The ERGs and the presence of optic nerves projecting to the brain ( Figure 4E ) prompted us to ask if the animals could respond to a visual stimulus using the induced eye . Xenopus show a stage-dependent phototropic behavior; premetamorphic tadpoles raised in a 12-h light cycle prefer a white color background [18] . When placed in a half-white/half-black test tank , sighted animals swim to and remain on the white side . In contrast , the same animals blinded by axotomy of the retinal ganglion cell axons from both eyes ( double axotomy ) show no color preference . We generated eyes in wild-type hosts using EFTF-expressing pluripotent cells from YFP transgenic donors , grew animals to premetamporphic stages , and tested for phototropic behavior using the Background Color Preference Assay [18] . The majority of transplants ( 12 of 13 ) generated mosaic eyes ( 24%–52% of the retina was YFP-labeled ) . However , a retina of one animal was entirely derived from EFTF-expressing pluripotent cells . An age-matched control and the animal with one induced eye both showed strong white background preference , spending 89%±6% and 88%±4% of swimming time on the white side of the test tank , respectively ( Figure 6C and 6D ) . We then severed the right optic nerve , leaving the optic nerve from the induced eye intact ( single axotomy ) . Single axotomy was also used to blind the right eye of the control animal . No statistical difference in background color preference was detected following single axotomy , suggesting the induced eye could guide behavior . To confirm this conclusion , we next severed the optic nerve from the induced eye . Double axotomy resulted in a significant reduction in the phototropic behavior of both the control and experimental animal . White background preference was reduced with axotomy of the induced eye ( 79%±6% to 57%±5% , p = 0 . 009 ) and the age-matched wild-type control ( 83%±6% to 50%±2% , p = 6 . 1×10−5; Figure 6C and 6D ) . The left eye did not contain transplanted cells ( Figure 6E ) , while the induced retina expressed YFP throughout , confirming that it formed entirely from the transplanted donor cells ( Figure 6F ) . These results suggest that EFTF-expressing pluripotent cells can form eyes with the functional cell classes and circuitry necessary for phototropic behavior . Noggin strongly induces the expression of the EFTFs in frog primitive ectoderm and human ES cells [3] , [7] . In Xenopus , Noggin can direct the progeny of early blastomeres , not normally destined to contribute to the eyes , to a retinal fate [19] , [20] . Human ES cells treated with Noggin , Dickkopf-1 , and Insulin-like Growth Factor-1 proteins differentiate into functional photoreceptors [5] . Given our ability to test retinal function in the Xenopus system , we wondered if mammalian Noggin could mimic the effect of the EFTFs and generate functional eyes . We first asked if Noggin-treated Xenopus pluripotent cells would differentiate into retinal cells in vitro . Pluripotent cells were cultured in mouse Noggin protein for 5 d and immunocytochemistry was used to detect retinal specific markers . While untreated pluripotent cells never expressed retinal markers ( 25 explants , npublished data ) , Noggin-treated explants expressed markers for rod and cone photoreceptors , as well as inner nuclear layer cells ( Figure 7A–7C; Table S5 ) . Rods and cones formed both rosettes and pseudo outer nuclear layers similar to those observe in the normal retina , with interspersed rods and cones . Labeled cell classes could also be identified based on their morphology . Oil droplets , outer , and inner segments were observed in cells expressing photoreceptor specific markers ( Figure 7B and 7C ) . Calretinin-expressing ( likely bipolar ) cells were also observed extending processes toward the peduncles of nearby layered photoreceptors ( Figure 7C ) . We next transplanted pluripotent cells cultured in Noggin to the neural plate of stage 15 embryos . Every embryo receiving a transplant of Noggin-treated pluripotent cells formed a YFP expressing eye ( Figure 7D , N = 38 ) . YFP-only pluripotent cells never formed eyes ( unpublished data , N = 31 ) . Embryos were cultured to stage 51 , and tested for phototropic behavior . An animal with one Noggin-induced eye and an age-matched wild-type animal spent 98%±0 . 8% and 96%±3% of swimming time on the white side of the test tank , respectively ( Figure 7E and unpublished data ) . Following single axotomy to blind the right ( control ) eye , a white color preference was still evident in both animals ( Figure 7E and unpublished data ) . Visual guided behavior was significantly reduced however , when the optic nerve from the second ( induced ) eye was severed ( Noggin-induced eye , 95%±4% to 57%±7% , p = 0 . 002; control , 78%±5% to 52%±5% , p = 0 . 001 ) . We confirmed the induced eye was completely donor derived . YFP-labeled projections were observed and sectioning of the animal tested and confirmed that all cells in every section of the induced eye were YFP-positive ( Figure 7F–7H and unpublished data ) . These results suggest that Noggin , similar to the EFTFs it induces , directs pluripotent cells to a multipotent retinal progenitor lineage . Induced retinal cells differentiate into functional retinal cell classes and form a neural network sufficient for vision .
The induced retinal cells generated here are multipotent , as they differentiate into all retinal cell classes and even express markers for some retinal cell subtypes . We observed stereotypical retinal cell morphology in eyes completely derived from EFTF-expressing pluripotent cells and in mosaic retinas . Lamba and colleagues recently demonstrated that human embryonic stem cells could be directed to differentiate into photoreceptors and respond to a light stimulus as measured by ERG in crx−/− mice ( a model for Leber's congenital amaurosis ) [5] . This important finding strongly suggests that in vitro generated cells can form functional retinal cells in a nonfunctioning retina . We show that pluripotent cells can generate all retinal cell classes . When transplanted to the anterior neural plate , these cells form eyes and a neural network capable of driving a vision-guided behavior . Not all donor cells were reprogrammed to a retinal lineage by the EFTFs . Why this might be is unclear . One hypothesis is that a subset of the pluripotent cells was refractory to the effects of the EFTFs . Alternatively , some cells may not have received a sufficient dose or the appropriate relative level of each EFTF . This possibility is intriguing , since subsets of the EFTFs are coordinately expressed in specific regions of the forebrain and developing olfactory placodes [7] . It will be interesting to determine if specific EFTF subsets can fate pluripotent cells to other ( nonretinal ) neural lineages . Eye fields and early optic vesicles form eye-like structures with lenses when transplanted ectopically or when cultured in vitro [21]–[26] . Similarly , EFTF-expressing pluripotent cells formed ectopic eye-like structures when transplanted to the embryonic flank . Although induced cells expressed retinal markers and exhibited retinal cell morphology in culture , morphologically normal eyes never formed . These results demonstrate fundamental differences between the induced retinal cells and the embryonic eye field . In isolation , EFTF-expressing pluripotent cells may be unable to recapitulate the complex inductive and morphological changes necessary for complete eye formation in culture . EFTFs may redirect pluripotent cells to a more restricted lineage than isolated eye fields , which may include additional tissues necessary for in vitro eye formation . Our transplantation experiments are consistent with this interpretation . While both eye fields and EFTF-expressing pluripotent cells generated epidermis , forebrain , and olfactory tissue when grafted to the anterior neural plate , transplanted eye fields also formed muscle and head mesenchyme ( Figure 4 and unpublished data ) . A requirement for mesoderm is supported by the work of Asashima , showing that tissue recombinants including lateral marginal zone or the mesoderm inducer Activin A generates muscle as well as pax6 expressing eye-like structures with lens in vivo as well as in vitro [27] , [28] . The presence of lens in these tissue recombinants may also be telling as EFTF-expressing pluripotent cells very infrequently formed lens . The reduced ability of the EFTFs to induce lens in the embryonic flank may be the cause of the abnormal layering we observed , since the lens and retina are dependent on each other for their normal formation . It is interesting that YFP-expressing pluripotent cells were rarely detected in the lens , yet often present in the cornea since the surface ectoderm forms both the lens and the cornea . All surgeries were performed on stage 15 embryos . At this developmental stage , the presumptive lens ectoderm lies lateral to , and outside the neural plate [29] , [30] . Therefore , successful removal of part or the entire eye field would spare the host presumptive lens ectoderm in the majority of the animals . Following neurulation , the externally localized ( YFP-negative ) lens ectoderm comes to lie over the evaginating optic vesicle , they make contact , the lens placode forms , and together they invaginate to form the optic cup and lens vesicle ( reviewed in [31] ) . We speculate , that once the YFP-negative lens vesicle separates from the surface ectoderm , the cornea forms from the remaining surface ectoderm ( YFP-positive ) and neural crest-derived mesenchyme [32] . YFP-expressing pluripotent cells did not form neural tissue ( retina or brain ) , but were always observed in the epidermis of tadpoles . This suggests that the transplanted , control cells move from the eye field to the surface ectoderm outside the neural plate . Conversely , EFTF-expressing pluripotent cells ( and transplanted eye fields ) remained in the anterior neural plate . In Xenopus , expression of the EFTFs pax6 , rx1 , as well as otx2 redirect ventral progenitors normally destined to form epidermis into the eye field where they form retina [20] . It will be important to identify the transcriptional targets of the EFTFs and the mechanism ( s ) responsible for these morphological movements . Eyes generated from pluripotent cells continued to grow throughout the life of the animals and contained a population of mitotically active cells in the CMZ suggesting the presence of retinal stem cells . During early embryonic development , cells throughout the forming retina are proliferative . Progenitors of the central retina are the first to leave the cell cycle and differentiate into the six neuronal and one glial cell class of the mature retina . In some species ( including Xenopus ) , the most peripheral cells of the retina never differentiate but form the retinal stem cells of the CMZ [33] , [34] . The Wnt , Shh , BMP , Insulin/IGF , and FGF signaling pathways have all been implicated in establishing and maintaining the adult retinal stem cell niche ( reviewed in [34]–[37] ) . Determining the effect ( s ) of these extrinsic factors on induced retinal cells and the formation of the CMZ in induced eyes will help define the molecular mechanisms regulating the formation and maintenance of retinal stem cells and the niche in which they reside . EFTF-expressing pluripotent cells form functional retinal cells and eyes when transplanted to the developing embryo , but can these induced retinal cells also differentiate into all the functional retinal cell classes in a mature normal or damaged retinal environment ? Cultured mouse , monkey , and human pluripotent cells can be driven down a retinal lineage as determined by the expression of marker genes , including the EFTFs [2]–[4] . When transplanted to the mouse retina , human ES cells directed to a retinal lineage partially restore light-elicited ERG responses [5] . Noggin is a key component in the cocktail used to bias human ES cells to a retinal lineage , and we found it was also sufficient to direct Xenopus pluripotent cells to retinal cell classes and eventually functional eyes . The assays presented here demonstrate a rapid and simple system for testing individual and combinations of intrinsic and extrinsic factors for their ability to direct pluripotent cells to multipotent retinal cells , with the option of testing them functionally . Future studies can now address how to maintain induced retinal cell cultures in a proliferative , multipotent state and drive them to all the retinal cell classes necessary to repair a damaged or degenerating mature retina .
Eye field ( EF ) and PNP isolated from regions AB1 , 2+B2 and HI3 , 4 , respectively of stage 15 embryos ( coordinate system described in [38] ) . LE was isolated from stage 15 lateral flank at A–P coordinates midway between eye field and PNP . Samples were pipetted into microfuge tubes cooled on dry ice to quickly freeze the isolated tissues . RNA from 40 EF , PNP , LE dissected tissues ( approximately 250 µm2 each ) or ten YFP-expressing control or YFP- and EFTF-expressing animal caps ( EFTF-expressing pluripotent cells ) were used for each chip analysis . Analyses were performed in triplicate on tissue from embryos on three different days from three different females . RNA extraction , probe synthesis and purification , chip hybridization , and scanning were carried out at SUNY-MAC ( Syracuse , New York ) . GeneSpring GX , version 7 . 3 . 1 ( Agilent Technologies ) was used for analysis . Cel file data was preprocessed using RMA method with the average signal intensity normalized to pluripotent cells isolated from Venus YFP RNA injected embryos or stage 15 whole embryos . MIAME-compliant supplementary information is included online at Gene Expression Omnibus ( http://www . ncbi . nlm . nih . gov/geo/index . cgi; GSE9173 and GSE9175 ) . Capped RNAs coding for pax6 , tbx3 , rx1 , nr2e1 , six3 , six6 , otx2 , and/or venus YFP were generated as previously described and injected into both cells of 2-cell staged transgenic venus YFP embryos . Expression of YFP in transgenic embryos is weak until approximately stage 15 . Experimental and control embryos were injected with venus YFP cRNA , which allowed selection of embryos successfully injected . We collected animal caps at stage 9 [7] , [10] , also called pluripotent cells here . Tissue was cultured in 0 . 7× MMR containing 50 µg/ml gentamicin sulphate to the equivalent of stage 15 . For Noggin protein-treated caps , stage 9 animal cap explants were cultured with 500 nM mouse Noggin protein ( Sigma-Aldrich , catalogue number N6784 ) in 0 . 7× MMR and 50 µg/ml gentamicin sulphate and grown to sibling embryo stage 15 for transplantation . A 250-µm2 region of LE or the left eye field was removed from stage 15 wild-type host embryos [38] using a Gastromaster . One half of the cultured pluripotent cells was grafted to the wound . For mosaic analysis , 125 µm2 ( ∼1/2 ) of the left eye field was removed and a size-matched fragment of animal cap was grafted . For Noggin cultured explants , Noggin-treated caps were grown in the same Noggin-containing solution above until sibling tadpoles reached stage 40–41 . Embryos were cultured to the indicated developmental stage and 12-µm , cryostat sections were immunostained as previously described [39] . The following antibodies were used: mouse XAP2 ( 1∶10; Developmental Studies Hybridoma Bank ( DSHB ) , Iowa City , Iowa , clone 5B9 ) ; mouse R5 ( 1∶5; kindly provided by W . A . Harris , Cambridge University , Cambridge , United Kingdom ) ; rabbit anti-Calbindin ( 1∶500; VWR , catalog number 80001-624 ) ; rabbit anti-GABA ( GABA:1∶500; ImmunoStar , catalog number 20094 ) ; mouse anti-Islet-1 ( 1∶100; DSHB clone 39 . 4D5 ) ; rabbit anti-Calretinin ( 1∶100; Novus Biologicals , catalog number NB200-618 ) ; anti-Tyrosine Hydroxylase ( TH: 1∶500; ImmunoStar , catalog number 22941 ) ; rabbit anti-GFP ( 1∶750; Invitrogen , catalog number A-11122 ) , mouse anti-GFP ( 1∶500; Sigma , catalog number G6539 ) . The slides were mounted in solution of FluorSave reagent ( VWR ) , 2%1 , 4-Diazabicyclo[2 . 2 . 2]octane ( DABCO: SIGMA ) and 10 µg/ml 4′ , 6-Diamidino-2-phenyindole , dilactate ( DAPI: SIGMA ) . Anti-GABA staining was performed as published [40] . In situ hybridization for hermes was performed as previously described [7] , [39] . Whole embryos images were captured using a Leica MZ16A fluorescence stereomicroscope with a MicroPublisher 3 . 3 RTV digital camera ( Q-Imaging ) and Q-Capture software ( V 2 . 7 . 1 ) . Sections stained for markers described above were visualized using a Leica DM6000 B upright fluorescence light microscope with motorized Z-focusing . A low-light , high-speed Retiga-SRV camera ( Q-Imaging ) captured images and sent them to the Volocity Imaging ( Improvision Inc . a PerkinElmer Company; V 5 . 0 . 3 ) software . The Volocity software package allowed us to acquire , visualize , deconvolve , and quantitatively measure tissue sections , as well as , retinal cells . Movies were made using this software and exported as QuickTime movies . Eye volumes were calculated by summing the area of every eye section and multiplying by the section thickness ( 12 µM ) as previously described using Volocity Software [19] . Tadpoles were allowed to swim in a solution of 5 mM BrdU ( Roche Inc . ) in 0 . 1× MMR for 1–4 h at room temperature and sections stained as previously described [39] . Percent of BrdU-labeled cells were determined by dividing their number by the total number of nuclei ( DAPI ) . 50 µm of central retinal sections were compared to the two peripheral regions of the same sections . Peripheral regions were defined as those regions containing cells with an elongated neuroepithelial morphology . In Figure 5O , the number of rod photoreceptor , inner nuclear layer , and retinal ganglion cells were determined by counting the nuclei of cells expressing XAP2 , Calretinin , or in the RGC layer , respectively . ERGs were performed as previously described [41] . Vision-based behavioral assay was modified from Moriya et al . [18] . A ½ Gallon Flex-Tank ( NASCO ) was colored half black and half white ( outer tank ) . Animals were placed in a second clear tank then inside the colored outer tank . The inner tank was rotated 180° following each 2 min trial . Behavior was recorded using a digital camcorder . Ten 2-min trials were run on two consecutive days for a total of 20 trials . A within-subject design , statistical analysis using a Student's t-test , paired two-tailed distribution was used . p≤0 . 05 was considered significant . Results shown as mean±standard error of the mean . Sham and unoperated animals behaved similarly ( unpublished data ) . Animals were anesthetized in 0 . 01% tricaine . The optic nerve was located and a 26-gauge needle and forceps were used to sever and displace the optic nerve from the optic tract . Tadpoles recovered in water containing 50 µg/ml gentamicin . The Committee for the Humane Use of Animals at SUNY Upstate Medical University approved all procedures .
|
The goal of regenerative medicine is to replace dead or dying cells . Successful cell replacement depends on the ability of donor cells to differentiate into all functional cell types lost in the target organ . Blindness resulting from retinal disease or damage , for example , would require the replacement of as many as seven specialized cell types found in the retina . The most celebrated characteristic of pluripotent cells is their ability to differentiate into any adult cell type . This defining feature , however , presents the challenge of identifying the conditions for their conversion to the cell types needed for tissue repair . We asked if pluripotent cells could be directed to generate all the retinal cell types necessary to form a functional eye in the frog , Xenopus laevis . If left untreated , transplanted pluripotent cells only form the epidermal layer of the skin . However , when forced to express the eye field transcription factor ( EFTF ) genes , the cells differentiate into all seven retinal cell classes and eventually organize themselves into a functioning eye that can detect light and guide tadpoles in a vision-based behavior . Our results demonstrate that pluripotent cells can be purposely altered to generate all the functional retinal cell classes necessary for sight .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"developmental",
"biology/cell",
"differentiation",
"neuroscience/neurodevelopment",
"developmental",
"biology/neurodevelopment",
"developmental",
"biology/molecular",
"development",
"developmental",
"biology/organogenesis"
] |
2009
|
Generation of Functional Eyes from Pluripotent Cells
|
Paramyosin is a thick myofibrillar protein found exclusively in invertebrates . Evidence suggested that paramyosin from helminths serves not only as a structural protein but also as an immunomodulatory agent . We previously reported that recombinant Trichinella spiralis paramyosin ( Ts-Pmy ) elicited a partial protective immunity in mice . In this study , the ability of Ts-Pmy to bind host complement components and protect against host complement attack was investigated . In this study , the transcriptional and protein expression levels of Ts-Pmy were determined in T . spiralis newborn larva ( NBL ) , muscle larva ( ML ) and adult worm developmental stages by RT-PCR and western blot analysis . Expression of Ts-Pmy at the outer membrane was observed in NBL and adult worms using immunogold electron microscopy and immunofluorescence staining . Functional analysis revealed that recombinant Ts-Pmy ( rTs-Pmy ) strongly bound to complement components C8 and C9 and inhibited the polymerization of C9 during the formation of the membrane attack complex ( MAC ) . rTs-Pmy also inhibited the lysis of rabbit erythrocytes ( ER ) elicited by an alternative pathway-activated complement from guinea pig serum . Inhibition of native Ts-Pmy on the surface of NBL with a specific antiserum reduced larvae viability when under the attack of complement in vitro . In vivo passive transfer of anti-Ts-Pmy antiserum and complement-treated larvae into mice also significantly reduced the number of larvae that developed to ML . These studies suggest that the outer membrane form of T . spiralis paramyosin plays an important role in the evasion of the host complement attack .
Trichinellosis is one of the common parasitic zoonoses and is a serious public threat in both developing and developed countries [1]-[6] . Trichinella spiralis ( T . spiralis ) infection is initiated by the consumption of meat contaminated with infective muscle larvae ( ML ) . With the aid of host gastric juice , ML are released from cysts and migrate to the small intestine where they develop into adult worms in 2–3 days . Five days post-infection , gravid females begin to produce newborn larvae ( NBL ) , which penetrate the intestinal mucosa and enter the lymphatic vessels and bloodstream [7] . NBL travel through capillaries to various organs and finally invade the muscles , where they form cysts . During the life cycle of T . spiralis in the host , all developmental stages are exposed to host complement , which is the first line of defense against pathogenic organisms and is a functional bridge between the innate and adaptive immune responses [8] . The ability to evade complement attack is essential for the survival of parasites within their respective hosts [9] . As early as 1911 , the presence of complement-fixing antigens from larvae of T . spiralis was reported in antiformin extracts of pepsin-digested rat muscle [10] . Complement -fixing antigens have since been used to diagnosis of trichinosis of trichinellosis [11] , [12] . Subsequent studies have reported that the complement elements C3 , C5 [13] , C1q , C8 and C9 [14] , [15] directly bind the ML of T . spiralis . All three stages of T . spiralis are capable of activating complement via the classical or alternative pathways [14] , or the lectin pathway [16] . However , it is still unknown whether the activation of the complement is detrimental or beneficial to the parasite . NBL might be the most potent activators [13] . Molecules or structures on the outermost cuticle/epicuticle of the parasite directly bind complement and appear to protect the parasite from an attack by inhibiting the formation of the membrane attack complex ( MAC ) [14] , [15] . Rats with normal levels of C6 or those with a C6-deficiency have similar susceptibilities to infection by T . spiralis . However , C3 , C8 and C9 were found to bind worms , suggesting that T . spiralis has efficient mechanisms for protecting against complement attack [15] . However , the precise molecular basis for this resistance is still unknown . Paramyosin is a thick myofibrillar protein found exclusively in invertebrates [17] . Experimental evidence has shown that paramyosin from helminths serves not only as a structural protein but also as an immunomodulatory agent [18]–[22] . It has been reported that paramyosin from Taenia solium inhibits C1 function [18] . Paramyosin from Schistosoma mansoni acts as an immunological defense molecule by binding C1q [18] , the Fc fragment of IgG [19] , C8 and C9 [20]–[21] . Recently , paramyosin from Clonorchis sinensis was shown to bind both human collagen and C9 [22] . In our previous study , a full-length cDNA encoding T . spiralis paramyosin ( Ts-Pmy ) was cloned by immunoscreening an adult T . spiralis cDNA library with infected immune sera [23] , Recombinant Ts-Pmy ( rTs-Pmy ) elicited partial protective immunity against a T . spiralis larval challenge in BALB/c mice [24] . In the present study , we investigated ability of rTs-Pmy to bind to host complement components and to protect against host complement attack . Our data show that rTs-Pmy binds complement components C8 and C9 and inhibits the complement-mediated killing of NBL , providing more evidence that Ts-Pmy plays an important role in the evasion of the host immune response to facilitate the survival of T . spiralis in its host .
All experimental animals were purchased from Laboratory Animal Services Center of Capital Medical University ( Beijing , China ) . All experimental procedures were reviewed and approved by the Capital Medical University Animal Care and Use Committee and were consistent with the NIH Guidelines for the Care and Use of Laboratory Animals . T . spiralis ( ISS 533 strain ) was maintained in female ICR mice . ML were recovered from the muscles of infected mice by a standard pepsin/hydrochloric acid digestion method as described previously [14] . Adult worms were obtained from the intestine of a rat infected orally with 800 T . spiralis ML [25] . NBL were obtained from fertile female adult worms cultured overnight in RPMI 1640 at 37°C . Crude somatic extracts of the different stages of T . spiralis were prepared by conventional methods [26] , and the protein concentration was determined by the BCA assay ( Pierce , USA ) . Total RNA was extracted from T . spiralis ML , adult worms and NBL with an RNAeasy mini kit ( Qiagen , Germany ) according to the manufacturer's instructions . Total first-strand cDNAs were reverse transcribed from the total mRNAs using a Sensiscript Reverse Transcription kit ( Qiagen , Germany ) . The specific forward primer ( 5′- ACC AAC TGA GGG CTT TGC A-3′ ) and reverse primer ( 5-′ AAT ATT CAT GTC CTT CTT CCA TCA C-3′ ) , based on Ts-Pmy coding sequence of 1830–2730 bp , were used to amplify Ts-pmy cDNA fragments ( 900 bp ) from reverse transcribed total cDNA from different developmental stages of T . spiralis using a PCR kit ( TaKaRa , China ) . Reactions without the addition of reverse transcriptase were used as negative controls . The amplified products were analyzed in 2% agarose gels and stained with DNA Green ( Tandz , USA ) . rTs-Pmy was expressed in E . coli BL-21 ( DE3 ) using the pET-28a expression system ( Novagen , USA ) and purified with Ni-affinity chromatography ( Qiagen , USA ) , as described previously [23] . The antiserum against rTs-pmy was raised in rabbits immunized three times with 150 µg rTs-Pmy . The monoclonal antibody against rTs-Pmy ( mAb 7E2 ) was obtained using a conventional hybridoma technique , and IgG was purified with HiTrap rProtein A affinity columns ( Amersham Biosciences , USA; data not shown ) . Crude somatic extracts of T . spiralis ML , adult worms and NBL were subjected to SDS-PAGE on a 12% acrylamide gel and transferred onto a PVDF membrane ( Millipore , USA ) . After being blocked with a 5% ( w/v ) skim milk solution in Tris-buffered saline ( TBS ) containing 0 . 05% Tween-20 ( Sigma , USA ) ( TBS-T ) for 1 hour at room temperature , the membrane was incubated with mAb 7E2 at a concentration of 50–100 ng/mL in TBS-T . Peroxidase-conjugated goat anti-mouse IgG ( 1∶5 , 000; Sigma , USA ) was used as secondary antibody . The reaction was visualized with enhanced chemiluminescence reagent ( Pierce , USA ) and exposed to a BioMax film ( Kodak , USA ) . In some experiments , the Odyssey two color infrared imaging system was used according to manufacturer's instructions . A total of 150 NBL released by adult female worms in culture were pretreated with heat-inactivated ( 56° for 30 min ) rabbit anti-rTs-Pmy serum , the same serum pre-absorbed with rTs-Pmy or heat-inactivated normal rabbit serum ( NRS ) ( 2 , 20 or 40 µL ) in a total volume of 50 µL in a 96-well plate for 1 hour at room temperature . Heat-inactivated rabbit antiserum against Ts87 was used as a non-relevant antibody control ( 40 µL ) . Then , 100 µL of freshly pooled normal guinea pig serum ( NGS ) was added as a complement source into each well , and the incubation was continued for 24 hours in a 5% CO2 incubator at 37°C . Heat-inactivated NGS ( INGS ) was added as a control . NBL mortality was monitored under an inverted microscope based on motility ( The worms without any movement during 30 seconds of observation and total stretch-out were scored as dead ) and the fluorescent staining of the DNA-binding dye SYTOX Green ( Invitrogen , USA ) [27] . Experiments were run in triplicate . Percent mortality was calculated as described elsewhere [21] , [28] . To determine the viability of NBL treated with anti-rTs-Pmy serum and NGS , treated NBL were passively transferred into BALB/c mice intravenously . Briefly , 2000 NBL per group were pretreated with 100 µL of heat-inactivated anti-rTs-Pmy rabbit serum , the same serum pre-absorbed with rTs-Pmy or heat-inactivated NRS , and incubated with 400 µL of fresh NGS for 12 hours in a 5% CO2 incubator at 37°C . INGS were used as a control . After being incubated with NGS or INGS , NBL were washed with serum-free RPMI-1640 , re-suspended in 0 . 25 ml of PBS and injected into the lateral tail vein of BALB/c mice ( 8 mice for each group ) as described elsewhere [29] . Muscle larvae burdens were collected and counted on the 26th day after injection [14] . Complement-mediated lysis of rabbit erythrocytes ( ER ) was performed via the alternative complement pathway as described by Hong et al . [14] . To test whether Ts-Pmy acts as an inhibitor or neutralizer of the complement activated by the alternative pathway , fresh NGS ( 6% ) were pre-incubated with various amounts of rTs-Pmy ( 0 , 10 , 20 or 40 µg ) in Mg-EGTA solution ( 5 mM MgCl2 , 10 mM EGTA ) for 30 min before adding the mixture to fresh , washed ER ( 3×108 ) in 0 . 1 mL of GVB ( Veronal-buffered saline , pH 7 . 4 , containing 0 . 1% gelatin and 0 . 02% NaN3 ) for 30 min at 37°C . Lysis was stopped by adding 1 mL of cold GVB containing 10 mM EDTA . After being centrifuged at 4 , 400×g for 10 min at 4°C , the amount of hemoglobin released into the supernatant was measured at 412 nm , and the percent lysis relative to the number of cells lysed completely by water was calculated . To determine whether rTs-Pmy binds to C8 and C9 , purified human C8 , C9 ( Merck , Germany ) and non-relevant control BSA ( Sigma , USA ) ( 1 µg each ) were subjected to SDS-PAGE under reducing conditions and transferred to a PVDF membrane . After blocking with 5% milk in TBS-T , the membrane was incubated with 10 mL of rTs-Pmy ( 5 µg/mL in TBS-T ) for 3 hours at 37°C and then with mAb 7E2 ( 50 ng/mL ) for 1 hour at room temperature . Peroxidase-conjugated goat anti-mouse IgG ( 1∶4 , 000; Sigma , USA ) was used as the secondary antibody . Bands were visualized with ECL ( Pierce , USA ) . To determine the specific binding of Ts-Pmy to C8 and C9 , rTs-87 was used as a control to react with C8 and C9 that had been transferred to a PVDF membrane and probed with anti-Ts87 antiserum . For the reciprocal experiment , the same amount of rTs-Pmy and BSA ( 1 µg ) was transferred to a PVDF membrane and incubated with C9 ( 0 . 5 µg/mL ) for 2 hours at 37°C . After being washed with TBS-T , the membrane was incubated with monoclonal anti-C9 antibody ( 1∶4 , 000; Abcam , USA ) . To determine the competition between the soluble C9 and blotted C9 for binding to Ts-Pmy , rTs-Pmy was pre-incubated with C9 1∶6 ( w/w , 5 µg/mL of rTs-Pmy and 30 µg/mL of C9 ) for 1 hour at 37°C before being incubated with the membrane . The membrane was then reacted with mAb 7E2 ( 50 ng/mL ) and IRDye-800CW-conjugated goat anti-mouse IgG ( 1∶10 , 000; LI-COR , Germany ) . To determine the effect of rTs-Pmy on Zn2+-activated C9 polymerization [30] , 3 µg of C9 was pre-incubated with various amounts of rTs-Pmy at 37°C for 40 min and incubated with 50 µM ZnCl2 in 20 mM Tris buffer , pH 7 . 2 for 2 hours at 37°C . Inhibition of C9 polymerization was shown by SDS-PAGE on a 2 . 5 to 25% acrylamide gradient gel under reducing conditions; the gel was visualized by staining with Coomassie blue [31] or analyzed by western blot with an anti-C9 antibody . A non-relevant T . spiralis antigen Ts87 was used as a control . rTs-Pmy inhibition of C9 polymerization on ER was performed as reported by Tschopp et al . [32] . One hundred microliters of normal human serum ( NHS ) supplemented with 5 µg of C9 was pre-incubated with 30 µg of rTs-Pmy for 40 min at 37°C prior to the addition of 3×106 ER and continued incubation for 1 hour at 37°C . Lysed cells were washed three times with 3 mL of TBS containing 5 mM EDTA ( pH 8 . 0 ) and centrifuged for 20 min at 4 , 800×g at 4°C . Sediments were washed three times with 3 mL of 0 . 5 mM PBS ( pH 8 . 0 ) . Lysed cell pellets were analyzed by SDS-PAGE under reducing conditions on a 2 . 5 to 15% gradient acrylamide gel followed by either Coomassie blue staining or Western blot with an anti-C9 antibody . Results were expressed as the mean ± SD . Differences between groups were assessed by SPSS 10 . 0 ( SPSS Inc . , USA ) using One-Way ANOVA; p<0 . 05 was considered to be statistically significant .
The transcription of Ts-Pmy mRNA at different developmental stages in T . spriralis was analyzed by RT-PCR with Ts-Pmy specific primers . Ts-Pmy mRNA was transcribed in all developmental stages ( ML , NBL and adult worm; Figure 1A ) . The size of the amplified cDNA fragments was 900 bp , which is the same size as predicted by the DNA sequence . Western blot analysis showed that an approximately 100 kDa band was recognized by mAb 7E2 in the somatic extracts the three developmental stages of T . spiralis ( ML , adult and NBL ) ( Figure 1B ) . The results show that Ts-Pmy is expressed in all three developmental stages ( ML , adult worm and NBL ) of T . spiralis at the level of both mRNA transcription and protein expression . Previously , immunofluorescent staining of worm sections demonstrated that Ts-Pmy was expressed on the surface of T . spiralis larvae [23] . Here , immunoelectron microscopy confirmed that Ts-Pmy was expressed on the outer membrane of the cuticle of the NBL ( Figure 1Ca ) and adult worm ( Figure 1Cb ) . No significant staining was observed using normal mouse serum at the same dilution ( c and d ) . Using immunofluorescence staining , Ts-Pmy was also observed on the surface of intact adult worms ( Figure 1Da ) and NBL ( Figure 1Db ) . To determine whether Ts-Pmy expressed on the parasite surface protects NBL from being killed by the host complement , heat-inactivated rabbit anti-rTs-Pmy serum was used to block Ts-Pmy on the NBL . The antiserum- or normal rabbit serum-treated NBL were challenged with the complement from fresh NGS . As shown in Figure 2 , after being blocked with anti-rTs-Pmy serum , NBL viability was significantly decreased following incubation with fresh NGS compared to those of the normal rabbit serum group . The increase in complement-mediated killing of NBL blocked with anti-rTs-Pmy serum was observed in an antiserum dose-dependent manner . After being absorbed with rTs-Pmy , the anti-rTs-Pmy serum had a minimal or non-existent effect on the complement-mediated NBL killing . Incubating with rabbit antiserum against recombinant Ts87 , a specific T . spiralis secreted protein [33] , that is also located on the surface of worm [34] , did not significantly increase complement-mediated NBL killing ( Figure 2 ) . The number of ML recovered from mouse muscle was reduced by 95 . 1% for those NBL treated with anti-rTs-Pmy rabbit serum and NGS complement compared with those treated with normal rabbit serum/NGS 26 days after being transferred intravenously into mice ( Figure 3; p = 0 . 006 ) . No significant reduction was observed for ML developed from NBL treated with rTs-Pmy pre-absorbed anti-rTs-Pmy serum . Heat-inactivated NGS ( INGS ) had a smaller killing effect on NBL treated with anti-rTs-Pmy rabbit serum ( 72 . 0% ML reduction ) or with the same serum pre-absorbed with rTs-Pmy ( 48 . 5% ML reduction ) compared with the normal rabbit serum control ( Figure 3; p = 0 . 048 ) . These studies further indicated that Ts-Pmy on the surface of NBL may protect larvae from complement-mediated killing . Complement-mediated ER lysis via the alternative pathway was significantly inhibited by adding rTs-Pmy in a dose-dependent manner ( p<0 . 001 ) . ER lysis was reduced by 55 . 7% when 10 µg of rTs-Pmy was added and by 91 . 9% when 40 µg of rTs-Pmy was added ( Figure 4 ) . After being transferred to a membrane , the human C8 and C9 were probed with rTs-Pmy and detected with anti-rTs-Pmy mAb . Western blot analysis demonstrated that rTs-Pmy strongly bound to human C8 ( α , β chains ) and C9 ( Figure 5B ) . The amount of rTs-Pmy bound to the immobilized C9 was dramatically decreased ( Figure 5E ) after being absorbed with soluble C9 . Reciprocally , the binding of C9 to rTs-Pmy immobilized on a membrane was also demonstrated by incubating the membrane with soluble C9 and detecting with an anti-C9 mAb ( Figure 6 ) . In the control experiment , the same amount of recombinant Ts87 did not bind to C8 and C9 ( Figure 5D ) . C9 polymerization leads to the creation of transmembrane channels that are critical for complement-mediated cytolysis [35] . To determine whether Ts-Pmy inhibits the Zn2+-induced C9 polymerization , C9 was mixed with various amounts of rTs-Pmy and then incubated with 50 µM ZnCl2 . As shown in Figure 7 , rTs-Pmy inhibited Zn2+-induced C9 polymerization in a dose-dependent manner . When the amount of rTs-Pmy was increased to 10 µg , it completely inhibited the polymerization of 3 µg of C9 ( Figure 7 , lane 6 ) . The same amount of recombinant Ts87 did not inhibit C9 polymerization . In addition , rTs-Pmy inhibited the formation of the complement complex ( poly-C9 ) on the surface of the ER ( Figure 8 , lane 3 ) .
Recent studies have shown that the complement system plays a role in granulocyte recruitment and parasite impairment in nematode infection prior to antibody production [36] . Complement components have been implicated in the killing of Strongyloides stercoralis larvae [37] and Schistosoma mansoni schistosomula [21] . The ability to evade complement attack is essential for the survival of tissue-dwelling nematodes within hosts . Many parasites , especially those living in or in contact with blood , seem to have developed parallel routes to escape complement attack [38] . T . spiralis infective larvae ( ML ) , adults and NBL are able to bind to the complement components [14] and evade complement-mediated killing [15] . However , the mechanism by which T . spiralis evades complement-mediated killing is still not completely understood . Paramyosin is an essential muscle protein in invertebrates , forming the core of thick myofilaments that determine the length and stability of muscles [17] . It has been suggested that the surface expression of paramyosin by Schistosoma mansoni [20] and Taenia solium [39] may inhibit the complement cascade of the immune system . Thus , in the present study , we examined the expression and localization of Ts-Pmy in the three developmental stages of T . spiralis and showed that Ts-Pmy is present on the outer membrane of the cuticle of the adults and NBL of T . spiralis . Our results were consistent with other observations that identified paramyosin in the tegument and on the surface of Schistosoma mansoni , Schistosoma japonicum [19] , Echinococcus granulosus [40] , Taenia solium [39] and Fasciola hepatica [41] . The existence of a surface-exposed form of paramyosin suggests a role as a potential modulator of the host immune system . It was previously observed that paramyosin on the surface of helminth parasites bound to the Fc of IgG [19] and IgA [42] , collagen and at least three complement components , C1q [18] , C8 and C9 [20] . The fact that Ts-Pmy is expressed on the outermost layer of the parasite indicates a possible role in the first line of defense against the host immune response . Our data confirmed that surface-exposed Ts-Pmy binds to complement C8 and C9 , which are important components of the complement activation cascade and comprise the membrane attack complex ( MAC ) [43] . Polymerization of C9 induced by Zn2+ was highly inhibited by rTs-Pmy , indicating that the assembly of the MAC was impaired . The alternative complement pathway that activates the complement complex or poly-C9 on the rabbit erythrocytes ( ER ) was also greatly inhibited by rTs-Pmy . These studies suggest that Ts-Pmy binding to C8 and C9 inhibits the assembly and formation of the MAC , creating an effective strategy for the parasite to evade complement attack [28] . Similar strategies are also adopted by Schistosoma mansoni [28] , Entamoeba hisolytica [44] , Herpesvirus saimiri [45] and Trypanosoma cruzi [46] . Functional analysis in this study revealed that rTs-Pmy inhibited the lysis of ER triggered by guinea pig serum , indicating that the binding of Ts-Pmy to C8/C9 or other complement elements inhibits the complement activation pathways . Whole T . spiralis worms also inhibited of complement-mediated hemolysis [14] , suggesting that native Ts-Pmy or other modulating molecules on the surface of the worm might have similar functions as complement inhibitors . Such observations are consistent with the findings of other studies of paramyosin in Schistosoma mansoni [20] , [21] . Additional evidences from this study show that native Ts-Pmy on the surface of T . spiralis effectively protected NBL from attack by host complements . Blocking Ts-Pmy on the surface of T . spiralis with anti-rTs-Pmy antiserum reduce of NBL viability when under attack by complement in vitro . In vivo passive transfer experiments with NBL treated with anti-Ts-Pmy antiserum and NGS also showed a significant reduction in the number of NBL that developed into ML in mouse muscle , indicating that the infectivity of these larvae after Ts-Pmy blocked by antibodies was seriously impaired . Our studies suggest that the outer membrane form of paramyosin expressed by T . spiralis has a role in host immunomodulation , presumably by inhibiting the formation of the MAC and thereby protecting the parasite from being damaged by activated complement . This modulation is an effective survival strategy for T . spiralis to live within its host . Disruptions of the immunomodulatory function of Ts-Pmy could be explored as an alternative strategy to control T . spiralis infection . As a result , T . spiralis paramyosin is under further evaluation as a potential laboratory reagent to study host complement function and as a potential vaccine antigen . The specific epitope in Ts-Pmy that reacts with the complement or elicits protective immunity is currently being defined .
|
Trichinellosis is a serious food borne parasitic disease caused by the consumption of meat contaminated with the infective larvae of Trichinella spiralis . The ability of the tissue-dwelling parasite to evade the host complement attack is essential for its survival and for establishing infection in the host . This study describes the expression of paramyosin , a muscular protein in invertebrates , on the surface of Trichinella spiralis and its role in the defense against the host complement attack as a survival strategy . Using a specific antiserum , expression of Trichinella spiralis paramyosin was detected on the outer membrane of the adult worms and newborn larvae . Functional analysis revealed that recombinant Trichinella spiralis paramyosin protein strongly bound human complement components C8 and C9 and inhibited the formation of the complement membrane attack complex . Neutralization with a specific antiserum greatly impaired the protective effect of paramyosin on the viability and infectivity of Trichinella spiralis newborn larva when under attack by complement . These studies suggest that the outer membrane form of Trichinella spiralis paramyosin plays an important role in the evasion of the host complement attack and is therefore a good target for vaccine and pharmaceutical development .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"medicine",
"infectious",
"diseases",
"trichinellosis",
"parasitic",
"diseases"
] |
2011
|
Trichinella spiralis Paramyosin Binds to C8 and C9 and Protects the Tissue-Dwelling Nematode from Being Attacked by Host Complement
|
Clostridium difficile is the primary cause of nosocomial diarrhea and pseudomembranous colitis . It produces dormant spores , which serve as an infectious vehicle responsible for transmission of the disease and persistence of the organism in the environment . In Bacillus subtilis , the sin locus coding SinR ( 113 aa ) and SinI ( 57 aa ) is responsible for sporulation inhibition . In B . subtilis , SinR mainly acts as a repressor of its target genes to control sporulation , biofilm formation , and autolysis . SinI is an inhibitor of SinR , so their interaction determines whether SinR can inhibit its target gene expression . The C . difficile genome carries two sinR homologs in the operon that we named sinR and sinR’ , coding for SinR ( 112 aa ) and SinR’ ( 105 aa ) , respectively . In this study , we constructed and characterized sin locus mutants in two different C . difficile strains R20291 and JIR8094 , to decipher the locus’s role in C . difficile physiology . Transcriptome analysis of the sinRR’ mutants revealed their pleiotropic roles in controlling several pathways including sporulation , toxin production , and motility in C . difficile . Through various genetic and biochemical experiments , we have shown that SinR can regulate transcription of key regulators in these pathways , which includes sigD , spo0A , and codY . We have found that SinR’ acts as an antagonist to SinR by blocking its repressor activity . Using a hamster model , we have also demonstrated that the sin locus is needed for successful C . difficile infection . This study reveals the sin locus as a central link that connects the gene regulatory networks of sporulation , toxin production , and motility; three key pathways that are important for C . difficile pathogenesis .
Clostridium difficile , a major nosocomial pathogen , is the causative agent of antibiotic-associated diarrhea and pseudomembranous colitis [1 , 2] . Every year , nearly half a million cases of C . difficile infections ( CDI ) occur in the United States and result in approximately 14 , 000 deaths [3] . C . difficile toxins damage the colonic epithelium , which results in moderate to severe diarrhea [4] . Recent studies have shown that these toxins are essential for C . difficile pathogenesis [4–7] . Due to the strictly anaerobic nature of the vegetative cell , C . difficile survives outside the host in the form of dormant spores , which are highly resilient and resistant to most disinfectants . Thus , C . difficile spores are critical for its host to host transmission and persistence in the hospital environment [8] . C . difficile Toxins A and B are encoded by the tcdA and tcdB genes respectively , and their expression is dependent on TcdR , an alternative RNA polymerase sigma factor [9–11] . Environmental stresses , such as alteration of the redox potential , high temperature , or limitation of nutrients like glucose , and biotin , modulate toxin production by influencing the expression of tcdR [9–12] . Similar to toxin production , the sporulation pathway in C . difficile is also known to be influenced by nutrient availability and uptake [13 , 14] . The regulators involved in controlling toxin synthesis in response to nutrients are the global regulatory proteins CcpA and CodY [14–18] . Among them , CcpA mediates glucose-dependent toxin gene repression [15 , 16] , and CodY blocks the transcription of toxin genes during the exponential growth phase of the bacterial culture [17 , 18] . Other than affecting toxin production , mutations in codY and ccpA were also found to affect sporulation [13 , 16] . Other genes that are known to influence both toxin production and sporulation include spo0A , sigH , and rstA [19–22] . New evidence suggests that the toxin , motility , and sporulation regulatory networks are linked together in C . difficile [19 , 23 , 24] . The sigma factor SigD needed for transcription of the flagellar operon was identified to regulate tcdR transcription to influence toxin production [25 , 26] positively . Mutations in spo0A , rstA , and sigH also influenced motility along with toxin production and sporulation [19–22] . This study identified that mutation of the sin locus in C . difficile could affect toxin production and sporulation along with motility and thus reports a new regulatory element of this network . In Bacillus subtilis , the sin ( sporulation inhibitor ) locus codes for two proteins SinR and SinI and regulates several genes involved in sporulation , motility , competency , proteolysis , and biofilm formation [27–31] . In this study , we have created C . difficile sin locus mutants in two different strains . Using RNA-Seq analysis , we compared the transcriptome of the mutants with respective parent strains to identify and assess the transcriptional regulation of sin locus coded regulators . Follow up phenotypic analyses and complementation experiments showed that the Sin regulators in C . difficile are also pleiotropic as in B . subtilis . Here , their regulatory roles in toxin production , sporulation , and motility were further investigated and discussed .
In B . subtilis , the sin locus carries two small ORFs , sinI and sinR [32 , 33] ( Fig 1A ) . B . subtilis SinR ( BsSinR ) is a DNA-binding protein that binds to a conserved DNA sequence upstream of the translational start site of target genes to negatively control their transcription . SinI , encoded by a gene adjacent to sinR , has an antagonistic relationship with SinR and binds directly to the SinR protein to inhibit its activity . This causes the pathways that were repressed by SinR to switch on . In B . subtilis , SinR contains 113 aa , and the DNA binding domain is located at the N-terminus part , which spans from residues 5–61 [32 , 33] ( Fig 1A ) . The C-terminal part of SinR forms alpha-helices and is responsible for multimerization and SinI interaction . The SinI protein , on the other hand , resembles a truncated SinR without the DNA binding region and carries only the alpha-helical structure to drive the hetero-dimerization of SinR-SinI complex [32–34] . In C . difficile the sin locus contains two ORFs CDR20291_2121 and CDR20291 _2122 ( in C . difficile R20291 reference genome ) , which codes for proteins that are 43% and 35% identical to B . subtilis SinR , respectively ( Fig 1B ) . Both these proteins are predicted to be DNA-binding since they carry HTH ( Helix-Turn-Helix ) domains in their N-terminal regions . Hence we named CDR20291_2121 as sinR and CDR20291_ CD2122 as sinR’ . The C . difficile SinR ( CdSinR ) contains 112 amino acids , and its predicted HTH domain spans residues 11 to 66 . The SinR’ ( CdSinR’ ) protein carries 105 aa , and its predicted HTH domain spans from residues 7 to 62 ( Fig 1B ) . Both CdSinR and CdSinR’ shows the highest homology to BsSinR in this DNA-binding domain , where within the 50 residues of HTH domain , 13 of them are identical and 19 of them represent conservative substitutions ( Fig 1C ) . CdSinR and CdSinR’ shows similarity with each other ( 33% identity ) only in their N terminal DNA binding domain . The C terminus multimerization domains of these proteins show variations , and there is less similarity of CdSinR and CdSinR’ to BsSinR and each other in this region . In various Bacillus sp . SinR homologs are known to control the expression of the genes adjacent to the sin loci . Thus , identifying genes adjacent to the sin loci were helpful in predicting at least a few functions of the Sin regulators in these bacterial species . For example , in B . subtilis , the sin locus is adjacent to the tapA-sipW-tasA operon , and SinR represses the expression of this operon whose products are involved in the production of the biofilm matrix [31] . In Bacillus anthracis , the sin locus is next to calY that codes for camelysin , a cell surface associated protease , and SinR in this species is known to repress the calY expression [35] . In C . difficile , the sin locus is located in between cynT ( codes for carbonic anhydrase ) and CDR20291_2123 ( unknown function ) ( Fig 1B ) and is not close to any other genes that are known to be essential for virulence in this pathogen . Thus , the location of the sin locus in C . difficile chromosome did not provide us any clues about its possible functions . To get more information about the locus and its role in C . difficile physiology we decided to construct and characterize mutants in sin locus . An erythromycin resistant marker was introduced in the sinR at nucleotide 141 using Clostron , a TargeTron-based group II intron in C . difficile JIR8094 [36] and R20291 strains [37] . The presence of the retargeted intron in the correct gene in both mutant strains was confirmed by PCR ( S1 Fig ) . In B . subtilis , three different promoters drive the transcription of the sin genes [33] . In B . subtilis , the polycistronic sinIR transcript is produced from two different promoters , and the sinR transcript is driven from an independent promoter immediately downstream of sinI ( Fig 1A ) [33] . In C . difficile , the operon upstream of sin locus transcribes in the opposite direction , and no read-through transcription of sin locus is possible from its promoter ( Fig 1B ) . Using cDNA prepared from the JIR8094 and the mutant strain , we performed RT-PCR analysis and checked for the presence of sinR , sinR’ and sinRR’ transcripts ( S2 Fig ) . We could detect sinR , sinR’ and also the read through sinRR’ transcripts , which confirmed that the sinR and sinR’ are transcribed as a single transcript ( S2 Fig ) . When the same analysis was performed using the mutant strain cDNA both the sinR , sinR’ and sinRR’ transcripts were absent ( S2 Fig ) . The QRT-PCR analysis of the sinR mutant showed significant reduction of both sinR and sinR’ transcript levels ( S2 Fig ) . It also revealed that similar to B . subtilis , the C . difficile sin locus is expressed between late-exponential and early-stationary growth phase ( 10 to 12 h ) ( S2 Fig ) . Similar results were obtained in RT-PCR analyses of cDNA from the R20291 strain ( S2 Fig ) . When we performed the western blot analysis using the SinR and SinR’ specific antibodies ( see M&M ) , both SinR and SinR’ were found to be absent in the mutant ( S3 Fig ) . Our western blot and the RT-PCR results together suggest that sinR and sinR’ are part of an operon . However , there is a possibility that sinR’ could have an independent promoter coded within the sinR coding region , which was not expressed in the growth conditions tested . Since the insertion of the intron in sinR ( first gene in the operon ) disrupted both sinR , sinR’ transcripts , and SinR , SinR’ production in the growth conditions tested , we named the mutant strains with the disrupted sinR gene as JIR8094::sinRR’ and R20291::sinRR’ . We first analyzed the impact of sin locus inactivation on the growth of C . difficile in TY medium . During the exponential phase of the growth , both parents and mutants grew at a similar rate . However , when they entered the stationary phase , we observed a decrease in the turbidity of the mutant cultures as measured as OD@600 nm ( S3 Fig ) . We performed the Triton X-100 autolysis assay to check the influence of SinRR’ on global autolysis of C . difficile [25] . We used the 16h old stationary phase culture to perform this assay , where the R20291::sinRR’ lysed at a faster rate compared to the parent ( S4 Fig ) . These results suggested that inactivation of sinRR’ induced autolysis in C . difficile . In B . subtilis , SinR along with another regulatory protein SlrR represses the expression of lytA-lytB-lytC and lytF autolysins [38] . Our initial observation of lysis phenotype in the sinRR’ mutants suggested that like B . subtilis SinR , C . difficile SinR might also be controlling the autolysin genes . In B . subtilis the SinR is a pleiotropic regulator and controls various pathways including autolysis [29–31 , 33 , 38 , 39] . We suspected that SinR and SinR’ in C . difficile might also regulate several targets to control multiple functions . Hence , to identify the sinRR’ regulated pathways in C . difficile , we performed the transcriptome analysis of the sinRR’ mutants in comparison with their respective parents . Based on the growth pattern of the sinRR’ mutants ( S3 Fig ) and the expression kinetics of sinRR’ in the parent strains ( S2 Fig ) , we decided to compare the transcriptomes of mutant strains with their respective parent strains during the early stationary phase ( i . e . , 12 h of growth ) in TY medium . We used three biological replicates and genes were considered differentially expressed if the fold change was ≥ log2 1 . 5 and their adjusted p-value was ≤0 . 05 . In the RNA seq analysis , it was observed that 437 and 425 genes were over-expressed in R20291::sinRR’ and in JIR8094::sinRR’ mutant strains , respectively , while 668 and 208 genes were under-expressed in R20291::sinRR’ and JIR8094::sinRR’ mutant strains , respectively . Results from the transcriptome analysis confirm that as in B . subtilis , SinRR’ in C . difficile also regulates a wide range of genes involved in several pathways including sporulation , motility , metabolism , membrane transport , stress response and toxin synthesis ( Fig 2A ) . A list of genes identified to be differentially regulated in mutants R20291::sinRR’ and JIR8094::sinRR’ compared to their parent strains are listed in S4 , S5 , S6 and S7 Tables respectively . To test and validate the transcriptome profiles , we performed relevant phenotypic assays and functional analysis with parent and mutant strains for major pathways ( sporulation , motility , toxin production and autolysis ) that were suggested to be regulated by SinR and SinR’ . We have included following strains in the phenotypic analysis: parent strain , sinRR’ mutant , sinRR’ mutant with pRGL311 ( plasmid with sinRR’ under its native promoter ) , and sinRR’ mutant with pRG334 ( plasmid with sinRR’ under the inducible promoter ) . To determine the independent role of SinR and SinR’ in the phenotypes , the sinRR’ mutant with plasmids: pRG300 ( sinR gene alone with its promoter region ) ; pRG310 ( sinR under the inducible promoter ) ; and pRG306 ( sinR’ alone under the inducible promoter ) were used . Western blot analysis with SinR and SinR’ specific antibodies were performed to confirm their expressions from the constructs , and the sinRR’ mutant with vector alone was used as negative controls ( Fig 2B ) . Growth curve analysis showed when sinRR’ was expressed from its promoter or the inducible promoter in the sinRR’ mutant , no autolysis was observed , and they grew similar as the wild type ( Fig 2C and S4 Fig ) . In the Triton X-100 autolysis assay , a partial recovery from autolysis was observed when either SinR or SinR’ alone was expressed in the mutant ( S4 Fig ) . To determine the role of sinRR’ on sporulation , we grew the test strains on 70:30 sporulation agar for 30h . Initial analysis through phase contrast microscopy detected no spores in R20291::sinRR’ ( Fig 3A ) . Transmission electron microscopy ( TEM ) further confirmed this observation ( Fig 3B ) . Fully mature spores could be detected in R20291 , whereas the sinRR’ mutant cells were devoid of any spores . Similar results were obtained for JIR8094::sinRR’ mutant as well ( S5 Fig ) . We performed ethanol treatment based sporulation efficiency assay where the ability of the bacteria to produce viable spores were analyzed by counting the total number of CFU ( Colony Forming Units ) following ethanol treatment . The mean sporulation efficiency of the parental strain R20291 was 18 . 7% ( Fig 3C ) . The sinRR’ mutant strain did not produce any spores , and the percentage of sporulation was near zero . We were surprised by the observation that expression of either sinRR’ or sinR/sinR’ alone also did not revive the sporulation in the sinRR’ mutants ( Fig 3C ) . Sporulation in C . difficile is initiated with the activation of Spo0A , which in turn triggers early sporulation gene transcription [22 , 40] . Transcripts of spo0A were 3 . 5-fold and 2 . 9-fold under-expressed in JIR8094::sinRR’ and in R20291::sinRR’ strains respectively , when compared to parent strains . We performed western blot analysis with the Spo0A specific antibodies [41] . We detected GDH ( glutamate dehydrogenase ) for loading control since its production was found to be unaffected in the sinRR’ mutants . Western blot analysis showed that in R20291::sinRR’ the Spo0A was absent or below the detectable level ( Fig 3C , S5 Fig ) . Lower production of Spo0A can result in down-regulation of all sporulation genes under its control . Our transcriptomic data indeed found many sporulation-associated genes to be affected ( Tables 1 , S4 and S6 ) in the sinRR’ mutant . The QRT-PCR analysis performed on selected sporulation genes confirmed their down-regulation in the sinRR’ mutants ( S8 Table ) . Since our transcriptome analysis and western blot analysis revealed a lower Spo0A in R20291::sinRR’ , we decided to test whether the asporogenic phenotype of the sinRR’ mutants is due to the lower production of Spo0A . We expressed spo0A from its native promoter ( pRGL312 ) in the R20291::sinRR’ and production of Spo0A in sinRR’ mutants was verified through the western blot analysis using Spo0A specific antibodies ( Fig 3C ) [41] . To our surprise , production of Spo0A in the sinRR’ mutants did not induce the sporulation in the R20291::sinRR’ strain ( Fig 3C ) . For sporulation to proceed normally , the Spo0A protein should get activated by phosphorylation [42] . Spo0A~P then acts as a transcriptional activator for many downstream genes in the sporulation pathway that includes sigma factors , the forespore specific sigF , and the mother cell-specific sigE [22 , 40 , 42] . We performed QRT-PCR to detect the transcripts of Spo0A~P activated sigF and sigE genes . We did not observe increases in sigF and sigE transcript levels in the spo0A expressing sinRR’ mutant when compared to the sinRR’ mutant with vector alone control . This result suggests that activation of Spo0A to Spo0A~P is affected in the sinRR’ mutant . In Bacillus sp . , the pathway that controls Spo0A phosphorylation is well characterized [43–47] . In Clostridia , the components of this phosphorelay are absent , and it has been hypothesized that sporulation-associated sensor kinases may directly phosphorylate the Spo0A for its activation . In C . difficile , four orphan kinases ( CD630_01352 , CD630_2492 , CD630_01579 , and CD630_1949 ) are present , among which , the CD630_1579 kinase was shown to phosphorylate Spo0A in vitro , and the CD630_2492 mutant was found to be less efficient in sporulation [48] . In the transcriptome data , the CD630_1579 and the CD630_ 2492 kinases were to be under-expressed ~1 . 5-fold and ~3-fold , respectively , in the JIR8094::sinRR’ mutant . However , their homologs CDR20291_1476 and CDR20291_2385 in the R20291::sinRR’ were not affected suggesting that these kinases might not be the main reason for Spo0A inactivation in the sinRR’ mutants . Since the regulatory network of Spo0A activation is largely unknown , there is a possibility that unknown kinases could have been affected in sinRR’ mutants . The JIR8094 strain was intrinsically non-motile due to mutations within the flagellar operon [49] . Hence , we choose only R20291 and R20291::sinRR’ to perform motility-related experiments . The R20291::sigD mutant and the R20291::sinRR’ strains with vector alone ( pRPF185 ) were used as the controls . Exponentially growing bacterial cultures were spotted on BHI with 0 . 3% agar and was incubated at 37°C for 36h to monitor motility . The bacterial cultures expressing sinRR’ , or sinR or sinR’ from the tet-inducible promoters were spotted on BHI with 50 ng/ml of ATc and 0 . 3% agar . In the motility assays , the R20291::sinRR’ strain was defective in motility ( Fig 4C and S6 Fig ) . The transcriptome analysis supported our observation , where sigD , the sigma factor needed for the transcription of the flagellar operons , was found to be 14-fold under-expressed in the R20291::sinRR’ ( Fig 4A , S4 Table ) along with other motility-related genes . Electron microscopic analysis followed by negative staining failed to detect flagellar structures in the R20291::sinRR’ ( Fig 4B ) . A dot blot analysis with FliC ( the flagellar structural protein ) specific antibodies also confirmed the absence of flagella in the R20291::sinRR’ strain ( S6 Fig ) . Expression of sinRR’ from its promoter or the inducible promoter revived the motility ( Fig 4C ) . Interestingly , expression of SinR alone was sufficient to revive the motility in the R20291::sinRR’ strain , whereas the SinR’ expression alone did not have any effect ( Fig 4C ) . SigD is needed for the transcription of the flagellar operon in C . difficile [25 , 26] . To determine whether the non-motile phenotype of sinRR’ mutant is due to the reduced levels of sigD in the sinRR’ mutants , we expressed sigD from the tetracycline-inducible promoter by introducing the construct pRGL291 into the R20291::sinRR’ strain ( S1 ) . We observed motility was partially restored in the R20291::sinRR’ when the sigD expression was induced ( Fig 4C ) , suggesting that sinRR’ controls motility by controlling the expression of sigD in C . difficile . The transcriptome analysis and the follow-up QRT-PCR ( Fig 5A , Table 2 , S4 , S6 and S8 Tables ) result suggested sin locus’s role in toxin gene regulation . Toxin ELISA was performed with the cytosolic protein extracts of sinRR’ mutants and their respective parent strains . Bacterial cultures expressing either sinRR’ or sinR/sinR’ alone from the tetracycline-inducible promoter were grown for 6h in TY medium and were induced with 50ng/ml of ATc for 5 hours . Cytosolic proteins harvested from these induced cultures were used for toxin ELISA . We observed a six-fold reduction in toxin production ( Fig 5A ) in the R20291::sinRR’ when compared to the R20291 strain . In JIR8094::sinRR’ however , a moderate two-fold reduction in toxin level was recorded when compared to the parent strain ( S7 Fig ) . Expression of sinRR’ in the mutants brought the toxin production back to the level comparable to the parent strains . As we observed in the motility assay , expression of sinR alone was sufficient to bring back the toxin production in the sinRR’ mutant , while expression of sinR’ did not show any effect . In C . difficile , SigD positively regulates tcdR , the sigma factor needed for toxin gene transcription [25 , 26] . Interestingly , the expression of sigD from an inducible promoter revived the toxin production in sinRR’ mutants , suggesting that sinRR’ controls both toxin production and motility by regulating sigD in C . difficile . We observed that SigD expression in the sinRR’ mutants partially recovered both the motility and the toxin production in that strain ( Fig 4C and Fig 5A ) . The main question that arises from this observation is how SinR controls sigD expression . The sigD gene is part of the flagellar operon , whose transcription is directly controlled by the intracellular cyclic di-GMP ( c-di-GMP ) concentration [26 , 50] . Within the cells , the c-di-GMP is synthesized from two molecules of GTP by diguanylate cyclases ( DGCs ) and is hydrolyzed by phosphodiesterases ( PDEs ) [50 , 51] . The functionality of several of these C . difficile DGCs and PDEs has been confirmed by expressing them heterologously in Vibrio cholerae , where they resulted in phenotypes ( biofilm formation and motility ) that correspond to elevated or lowered levels of intracellular c-di-GMP [51] . In C . difficile when CD630_1420 ( dccA ) was expressed from an inducible promoter , it resulted in elevated levels of intracellular c-di-GMP and reduced bacterial motility [50] . In R20291::sinRR’ , ten-fold more ( -3 . 3 Log2 fold ) dccA ( CDR2029_1267 ) transcript was observed ( S5 Table ) compared to parent . We measured the intracellular concentration of c-di-GMP ( S8 Fig ) and observed a nearly three-fold increase in the c-di-GMP concentration in the sinRR’ mutant compared to the parent R20291 strain ( Fig 5B ) . This elevated intracellular level of c-di-GMP in sinRR’ mutants can block the sigD expression , which in turn will result in reduced motility and toxin production ( Figs 4C and 5B ) . Hence , when sigD was expressed from the tetracycline-inducible promoter ( which is not affected by c-di-GMP concentration ) , motility and toxin production in the sinRR’ mutant could be revived . These two findings corroborate our conclusion that elevated levels of c-di-GMP in sinRR’ mutant plays a major role in controlling its toxin production and motility . We are currently performing experiments to test whether SinR can directly regulate dccA in C . difficile . Results from the sinR and sinR’ complementation experiments showed that expression of SinR alone could revive the toxin production and the motility in the R20291::sinRR’ strain , whereas SinR’ expression alone did not have any effect on the toxin production or the motility ( Figs 4C and 5A ) . These results suggested that among SinR and SinR’ , only SinR can directly influence the toxin production and the motility , which raised the question on the role of SinR’ in these pathways . To find the answer , we created a sinR’ mutant which expressed SinR in the absence of SinR’ ( S9 Fig ) . Our repeated attempts to create a sinR’ mutant using the similar technique in the JIR8094 background failed for unknown reasons . Mutation in sinR’ was confirmed by PCR ( S9 Fig ) and western blot analysis using SinR’ specific antibodies . As expected the SinR’ mutant produced SinR protein , but not the SinR’ ( S9 Fig ) . The R20291::sinR’ grew almost similar to the parent strain and did not show any profound autolysis phenotype as the R20291::sinRR’ ( S6 Fig ) . We performed the assays to measure sporulation , motility and toxin production in the R20291::sinR’ . In the sporulation assay , it was found that R20291::sinR’ produced nearly three-fold more spores than the parent R20291 strain ( Fig 6A ) . The R20291::sinR’ was more motile than the R20291 strain ( Fig 6B ) . Similarly , a 2 . 5-fold increase in the toxin production was observed in the R20291::sinR’ when compared to the parent strain ( Fig 6C ) . These initial results revealed that SinR’ can negatively influence sporulation , toxin production , and motility . In our complementation of R20291::sinRR’ we showed that presence of SinR’ alone in the C . difficile cells in the absence of SinR could not influence either toxin production or the motility ( Fig 4C and Fig 5A ) . Hence , SinR’ must be influencing these pathways through its action on SinR . For example , if SinR’ is an inhibitor of SinR then the absence of SinR’ in the R20291::sinR’ would result in increased SinR activity , which in turn may result in increased sporulation , toxin production and motility in this strain . To test this hypothesis , we performed two experiments . First , tested the effect of over-expressed SinR in the wild-type strain; Second , we checked for physical interaction of SinR with SinR’ proteins by performing pull-down experiments . The plasmid construct with either sinR ( pRG300 ) or sinR’ ( pRG306 ) under tetracycline-inducible promoter were introduced into R20291 parent strain and were tested for their toxin production , sporulation , and motility upon induction with ATc . The R20291 strain with the vector alone was used as the control in these assays . To perform the sporulation assay , we used bacterial cultures grown in 70:30 medium supplemented with 50 ng/ml of ATc for 36 hours . Sporulation efficiency was enumerated as described in the method section . Overexpression of sinR in R20291 strain increased its sporulation efficiency 2 . 5-fold ( 45% ) when compared to the control strain , where the average sporulation efficiency was 18% . Overproduction of SinR’ in R20291 , however , reduced the sporulation efficiency to 5% ( Fig 7A ) . Overproduction of SinR in R20291 resulted in increased motility as well ( Fig 7B ) . In C . difficile , toxin production is minimal during exponential phase ( ~4 to 8h ) of the bacterial culture and reaches its maximum during the stationary phase ( 12h -16h ) [9] . To detect any positive influence of both SinR and SinR’ on toxin production in the parent strain , we chose to use the 8h time point . The bacterial cultures were grown for 6h in TY medium and were induced with 50 ng/ml of ATc for two hours before harvesting their cytosolic protein for Toxin ELISA . Results from these experiments showed that overexpression of sinR resulted in a nearly 2 . 5-fold increase in the toxin production in the R20291 strain when compared to the R20291 with vector alone control ( Fig 7C ) . No significant effect on toxin production was observed when sinR’ was overexpressed in R20291 ( Fig 7C ) . This could be because sin locus is expressed only during the early stationary phase ( 10-12h ) in C . difficile ( S2 Fig ) . We performed toxin ELISA at 8h time-point when SinR is predicted to be lower in the bacterial cells . If SinR’ acts on toxin production primarily by repressing SinR , then overexpression of SinR’ at this time-point will not have any effect on toxin production . Nevertheless , results from this overexpression studies demonstrated that increased SinR content in C . difficile could result in increased toxin production , motility , and sporulation . In B . subtilis , SinR monomers bind with each other to form a homotetramer , which would then bind to upstream sequences of the target genes to repress their expression [34 , 52] . SinI in B . subtilis binds with SinR and prevents the SinR homotetramer formation and thus blocks its activity [52] . To test the protein-protein interaction of C . difficile SinR with SinR’ , we performed GST pull-down experiments using SinR-6His and SinR’-GST . Purified SinR-6His protein was mixed with crude lysates from E . coli expressing SinR’-GST . When we passed this mixture through the Ni++ affinity chromatography column , we pulled out SinR-6His along with SinR’-GST , suggesting the tight association of SinR with SinR’ ( Fig 8A , lanes 5 , 7 ) . In control , the GST alone did not interact with the SinR-6His ( Fig 8A , lanes 6 , 8 ) , confirming protein specific interaction between SinR with SinR’ . These results provided compelling evidence that SinR’ affects toxin production and sporulation indirectly by binding with SinR to inhibit its activity on its target genes . Transcriptome analysis of the R20291::sinRR’ showed up-regulation of codY , an important global regulator by ~3 to 30 fold compared to parent strains ( S5 Table , S8 Table ) . CodY is highly conserved in many Gram-positive bacteria [53–55] . In B . subtilis it regulates several metabolic genes and controls competence , sporulation , and motility [56–58] . In C . difficile , the codY mutant produced more toxins and spores than the parent strains and thus it is a repressor of these pathways [14 , 17 , 18] . We hypothesized that many phenotypes and transcriptional changes we observe in the sinRR’ mutant could be related to the up-regulation of codY in these mutant strains . To investigate whether SinR and SinR’ or both controls codY expression by binding to the promoter region of codY , we carried out electrophoretic mobility shift assays ( EMSAs ) . We used radiolabeled DNA probe that contained the putative promoter region of the codY gene and performed binding reactions using purified SinR-6His or SinR’-6His proteins . First , we tested SinR alone at increasing concentrations and found that it can shift the probe when used above 100 nM concentration ( Fig 7B ) . When SinR’ was used similarly , it was unable to cause the mobility shift of the probe , even at the highest concentration ( Fig 7B ) . We then tested whether SinR’ would prevent SinR from binding to the codY promoter region . To do this , we used increasing amounts of SinR’ , in the presence of a fixed amount of SinR ( Fig 7B ) . The results show that the presence of SinR’ in the reaction mix could prevent SinR from binding to the DNA . As a negative control , we used a DNA probe that contained the promoter region of gluD , which codes for glutamate dehydrogenase ( GDH ) . Neither SinR nor SinR’ was able to shift the control DNA even at the highest concentrations tested ( S10 Fig ) . Based on these results , we conclude that SinR binds specifically to codY promoter region to control its transcription . This result also provided evidence that the SinR’ interaction with SinR prevents its regulatory activity on its target gene . In a recent study , CodY was found to negatively regulate sinRR’ expression in the C . difficile 630Δerm strain [14] . A CodY putative binding site was identified in the sin locus upstream sequence , and reporter fusions with the sin locus promoter revealed the CodY could negatively regulate sin locus expression in this strain . However , in the UK1 strain ( belongs to the ribotype 027 as R20291 ) , the promoter fusion revealed a positive regulation of sin locus by CodY . Because of these contradictory observations , one could not conclude whether CodY regulates sin locus . To examine the role of CodY on sin locus expression , we performed EMSA with purified CodY-6His and the putative CodY binding region upstream of sin locus . An oligonucleotide with putative CodY binding sequence upstream of sinR was synthesized ( ORG 721 ) ( S2 Table ) and was radioactively labeled with [γ- 32 P] dATP . A double-stranded DNA probe was generated after annealing with the complementary oligonucleotide ( ORG722 ) . It is worth noting no sequence difference was found within this putative sin promoter regions of the UK1 , R20291 , JIR8094 and 630Δerm genomes . We also generated probes with a known CodY binding sequence upstream of the tcdR gene ( using ORG719 and ORG720 ) and with non-specific sequence ( ORG702 and ORG723 ) as positive and negative controls respectively . EMSA was performed by incubating the radioactively labeled probes with varying concentrations of purified CodY-6His . We found that CodY could bind to the sequence upstream of sin locus at the concentration of 400 nM ( Fig 9A ) . As expected the shift was observed with the positive control probe , while no shift could be observed with the non-specific DNA probe even with high protein concentrations ( Fig 9A ) . Binding of CodY to its targets most of the time results in repression of their transcription [17 , 18 , 58] . However , there are few targets where CodY was found to promote transcription [58] . To check whether CodY has any positive influence on sin locus expression in UK1 strain as reported [14] , we performed western blot analysis and looked for SinR and SinR’ in UK1 strain and its codY mutant ( UK1::codY ) . Results showed that SinR and SinR’ protein content in the UK1::codY mutant was higher than in the UK1 parent strain ( Fig 9B ) . Our data demonstrate that CodY has a negative impact on SinR and SinR’ production in this strain . Since our repeated attempts to create codY mutants in R20291 and JIR8094 strains failed , we could not include them in this analysis . Nonetheless , our results from the EMSA and the western blot analyses corroborate the negative regulation of the sin locus by CodY . Since the C . difficile sin locus was found to be important for the regulation of many important pathways under in vitro growth conditions , we wanted to determine its significance in C . difficile pathogenesis . We used the hamster model in which C . difficile infection is known to cause severe disease signs [59 , 60] . Syrian hamsters were gavaged with 2 , 000 vegetative cells of C . difficile strain R20291 or with R20291::sinRR’ and monitored for C . difficile infection . Fecal pellets were collected daily until animals developed diarrheal symptoms . All ten animals infected with parental strain R20291 succumbed to the disease within five days after bacterial challenge . Two of the ten animals infected with R20291::sinRR’ exhibited disease symptoms within two days after challenge ( Fig 10A ) . Diseased hamsters were sacrificed ( see M&M ) , and their cecal contents were collected for toxin ELISA and CFU count . All surviving sinRR’ mutant infected hamsters ( 8 in total ) and uninfected control hamsters were also sacrificed fifteen days post-infection , and their cecal contents were also tested for toxins and C . difficile cells . Toxins could be detected ( S11 Fig ) in the cecal contents of all the diseased hamsters ( 10 from R20291 group and two from R20291::sinRR’ group ) , which confirmed the occurrence of CDI in them . However , toxins could not be detected in the eight hamsters that survived the R20291::sinRR’ challenge . The cecal contents of R20291 infected hamsters contained nearly 107 colony-forming units per gram . No C . difficile could be recovered from the cecal contents of any of the R20291::sinRR’ challenged animals , including of the two hamsters that came down with CDI in this group ( Fig 10B ) . If the sinRR’ mutant lyses in vivo , as we observed in in vitro growth conditions , it could explain why we could not recover any C . difficile cells but could detect toxins in the cecal contents of that two hamsters that came down with the disease after sinRR’ mutant challenge . Since nearly 80% of the animals survived the R20291::sinRR’ challenge , we conclude that members of the SinRR’ regulon are needed for C . difficile successful pathogenesis .
This study aims to decipher the role of the SinRR’ regulators in C . difficile physiology . In C . difficile , there has been no data explaining their function , except for a few expression analyses , where mutations in sigH , tcdR , codY , spo0A , opp , app were found to affect the expression of the sin locus [13 , 14 , 21 , 22 , 60] . Initial clues about the role of SinR and SinR’ in sporulation came from the work performed by Saujet et al . where they showed increased expression of sinR in the asporogenous sigH mutant , suggesting it to be a negative regulator of sporulation as in the case of B . subtilis [22] . However , sinR was found to be up-regulated in the hyper-sporulating oligopeptide transporter opp-app mutant and was down-regulated in the hypo-sporulation tcdR mutant [13 , 60] . These later studies suggested the positive influence of SinR on sporulation . In this work , we mutated the sin locus in two different C . difficile strains and conclusively showed that unlike B . subtilis SinR , which inhibits sporulation , C . difficile SinR has a positive effect on sporulation . Transcriptome analysis of sinRR’ mutants revealed that in addition to sporulation , genes involved in motility , transport , stress response , cell wall biogenesis , and various metabolic pathways were also affected . It is worth noting that cynT , the gene adjacent to sin locus ( Fig 1B ) , is one among the many metabolic genes that were found to be down-regulated in the sinRR’ mutants ( S4 and S6 Tables ) . The analysis also revealed that the sin locus mutations could affect the transcription of many important regulators , including codY , sigD , spo0A , and tcdR . This observation compelled us to hypothesize that SinRR’ might be indirectly influencing transcription of many of these genes by controlling their regulators . For example , changing in the transcription of codY , a global regulator can affect the gene regulatory circuits of various pathways . CodY is known to be a sensor of the metabolic state of the cell . During the exponential growth phase , when the nutrients are abundant , CodY binds to branched-chain amino acids ( BCAAs ) , and GTP and acts primarily as a repressor of various alternative metabolic pathways [17 , 18] . When nutrients become limited in the cell , CodY is no longer bound by the cofactors and the transcriptional repression by CodY is alleviated on its targets . In C . difficile , CodY controls toxin production and sporulation in addition to metabolic pathways [14 , 17 , 18] . The transcription of codY was found to be up-regulated in the R20291::sinRR’ ( S5 Table ) , ( Fig 11 ) . This observation of increased codY transcription in the asporogenic sinRR’ C . difficile mutant is consistent with the recent findings that a C . difficile codY mutant hyper- sporulates [14] . To test whether increased CodY activity in the mutant is the reason for its lower toxin production and sporulation , we tried to isolate a sinRR’-codY double mutant and were unsuccessful even after several attempts . However , our EMSA experiments with purified SinR and codY upstream DNA showed that the SinR could specifically bind to this region , possibly to repress its transcription . We have also shown that purified CodY , in turn , can bind with the upstream region of sin locus upstream region to control its expression . Since the sin locus codes for both sinR and its antagonist sinR’ , SinR repression on codY would be moderate when compared to CodY’s repression on the sin locus . Also , when the cells enter the stationary phase , CodY repression on the sin locus may be alleviated in the absence of its co-substrates and will result in the sin locus expression , which we found to be essential for sporulation initiation . We performed dot blot analysis with cytosolic proteins of R20291 and R20291::sinRR’ and determined that CodY in R20291::sinRR’ was only moderately higher than R20291 ( S12 Fig ) . This could be due to the cell to cell variation in gene expression within the test population . For example , only 18% of the R20291 population enters sporulation in the growth conditions we tested . In C . difficile , only cells with low or inactive CodY enter sporulation . If we consider sporulation as an indirect measure for inactive CodY in a bacterial cell , we can say that the CodY production or activity was affected only in a fraction of cells in the parent population . To overcome this issue , we compared the CodY content in R20291::sinR’ cells ( which produce more SinR ) with R20291::sinRR’ . It is worth to note that nearly 50% of R20291::sinR’ culture enters sporulation . Nearly two-fold more CodY could be detected in R20291::sinRR’ cells when compared to R20291::sinR’ cells . Other than modulating CodY content in C . difficile , SinR could also affect the CodY activity indirectly by affecting the concentrations of CodY substrates ( BCAA and GTP ) . The transcriptome analysis indeed showed numerous metabolic genes to be affected in the sinRR’ mutant . In the JIR8094::sinRR’ mutant , codY was not among the differentially regulated genes . However , in this strain ccpA was up-regulated nearly 13 . 5-fold ( S7 Table ) . Similar to CodY , CcpA also represses toxin gene expression in C . difficile [15 , 16] . Thus lower toxin production in JIR8094::sinRR’ could be due to the higher CcpA activity in this mutant ( Fig 11 ) . We are currently testing whether ccpA is directly regulated by SinRR’ . We are also setting up experiments to check whether increased CcpA has any role in controlling codY expression in the JIR8094::sinRR’ strain . SigD is one other regulator whose expression was found to be affected in the sinRR’ mutants . In C . difficile , the sigD expression is repressed by elevated levels of c-di-GMP [50] . The enzyme , diguanylate cyclase coded by dccA synthesize c-di-GMP from GTP . In this study , we have shown the expression of dccA is up-regulated in sinRR’ mutant ( S5 Table and Table 2 ) and the observation of three-fold higher intracellular concentration of c-di-GMP in the sinRR’ mutant , corroborated the transcriptome data . These results suggest that SinR and SinR’ regulates motility and toxin production indirectly by regulating the c-di-GMP production . Another scenario that can result in higher intracellular c-di-GMP concentration is when c-di-GMP degrading phosphodiesterases are reduced within the cell . In C . difficile , pdcA codes for a c-di-GMP phosphodiesterases , and it was recently identified to be repressed by CodY [61] . RNA-Seq analysis did not identify pdcA as one among the differentially regulated genes in R20291::sinRR’ strain . However , it was under-expressed nearly 4-fold in JIR8094::sinRR’ ( S6 Table ) mutant . Increased CodY activity in the sinRR’ mutant could indirectly result in increased c-di-GMP concentration , which in turn can suppress toxin production and motility . In B . subtilis , the SinR’s repressor’s activity on its target genes is inhibited by SinI , which is coded in the same operon ( Fig 1A ) . In B . subtilis the polycistronic sinRI transcripts are produced from two upstream promoters . The monocistronic sinR transcripts are driven from a promoter located within the coding region of sinI . Regulating the transcription rate of sinRI and sinR helps B . subtilis to control its SinR and SinI content . Our RT-PCR and QRT-PCR analysis detected sinRR’ transcripts in C . difficile . We have also shown that disrupting sinR by insertion mutagenesis affects both sinR and sinR’ transcription . These results suggest that sinRR’ is transcribe as a bicistronic message . However , there is a possibility that sinR’ may have an independent promoter within sinR coding sequence as in B . subtilis . Our QRT-PCR analysis repeatedly detected lower levels of sinR’ , sinRR’ transcripts than the sinR transcripts . Western blot analysis also revealed lower levels of SinR’ than the SinR in growth conditions tested ( Fig 2B-lane 2and S9C Fig ) . There is a possibility that mRNA degradation from the 3’ end can result in lower levels of sinR’ transcripts , which in turn can result in lower levels of SinR’ than SinR . We did not detect any secondary structures upstream of sinR’ that can influence its translation rate or translation initiation . We , however , noted that the RBS of sinR’ are just two nucleotides away from the sinR stop codon . Ribosome complex occupying the sinR stop codon can prevent the assembly of new ribosome complex at the sinR’ RBS to initiate translation . Since SinR’ has a DNA binding domain , it is also possible that SinR’ may work as direct regulator independently from SinR and may have its own targets for regulation . In such case , SinR’ may not always be available to inhibit SinR function . A transcriptome analysis of sinR’ mutant and its comparison with sinRR’ transcriptome may help us to identify direct targets of SinR’ . In B . subtilis , other than SinI , SinR also interacts with SlrR and SlrA to regulate genes involved in matrix formation ( the eps and tap-sipW-tas operon ) , autolysis ( lytABC ) and motility ( hag , encoding flagellin ) [29 , 31] . In B . subtilis , the SlrR is a DNA binding protein , and it is homologous to SinR . Conversely , SlrA is a small protein devoid of any DNA binding domains and is homologous to SinI [38 , 39] . While SlrR can form heterodimers with SinR to repress lytABC and hag expression , it can also inhibit SinR’s repression activity on eps and tap-sipW-tas operons [38 , 39 , 62] which are needed for biofilm formation . The C . difficile sin locus codes for two DNA binding proteins SinR and SinR’ and their interactions resembles the interaction between B . subtilis SinR-SlrR . Similar to B . subtilis SlrR , C . difficile SinR’ carries a DNA binding domain and it will be interesting to analyze whether SinR-SinR’ complexes together are needed for the repression of any genes . It is important to note that the autolysis phenotype of sinRR’ mutant was complemented only when both SinR and SinR’ were expressed ( Fig 2C ) . This suggests that like B . subtilis SinR-SlrR complex , SinR-SinR’ complex together repress autolysis in C . difficile . No lytABC homologs could be identified in C . difficile genome , and the precise reason for autolysis in sinRR’ mutant is not clear yet . However , the RNA-Seq data revealed that nearly 6% of the differentially expressed genes in sinRR’ mutant plays a role in cell wall synthesis or assembly . This highlights that SinR and SinR’ play an important regulatory role in this pathway . Among the phenotypes tested , asporogenesis of the sinRR’ mutant was the only one we could not complement . Even the expression of spo0A failed to initiate sporulation in this mutant . Transcripts of Spo0A~P activated sigE and sigF did not show any increase when spo0A was expressed in the sinRR’ mutant , suggesting the Spo0A remain unphosphorylated and inactive . We are currently performing additional experiments to test this hypothesis . Another regulatory checkpoint for sporulation initiation is chromosomal DNA replication and segregation . This is achieved through the action of Soj and Spo0J in B . subtilis , where they repress sporulation until chromosomal segregation has occurred . They block the spo0A dependent transcription in B . subtilis [63] . The spo0J and soj homologs in C . difficile are CD630_3671 and CD630_3672 , respectively in an operon , which also carries CD630_3673 , an additional Spo0J-like orthologue . In both JIR8094::sinRR’ and R20291::sinRR’ , all three genes were up-regulated ~3 fold . Hence , the inactivation of Spo0A could result partly because of the up-regulation of the soj operon in the sinRR’ mutants ( Fig 11 ) . But the function of soj and spo0J in C . difficile should be determined before we can speculate their roles in asporogenesis of sinRR’ mutants . BLAST search revealed that SinRR’ to be unique to C . difficile and its close relative Clostridium sordellii . The sin locus is absent in other Clostridia . Even though sporulation-specific sigma factors appear to be conserved among Clostridia , recent studies have suggested that sporulation initiation and regulation of C . difficile to be distinct [64 , 65] . Since the sin locus appears to play a significant role in sporulation initiation and regulation , it is reasonable to speculate its presence could be one of the reasons why the regulation of sporulation initiation is distinct in C . difficile . In summary , our study supports earlier reports that in C . difficile , virulence , sporulation , metabolism and motility pathways are inter-connected [13–24] . While many regulators in this network are yet to be identified , here we present the evidence that SinRR’ play a central role in this regulatory network . SinR regulates multiple pathways by controlling other global regulators . Finding genes that are directly under SinR regulation may lead to the identification of new regulatory genes and gene products that are important for C . difficile pathogenesis .
All animal procedures were performed with prior approval from the KSU Institutional Animal Care and Use Committee ( protocol #3657 ) . Animals showing signs of disease were euthanized by CO2 asphyxia followed by thoracotomy as a secondary means of death , in accordance with Panel on Euthanasia of the American Veterinary Medical Association . Kansas State University is accredited by AAALAC International ( Unit #000667 ) and files an Assurance Statement with the NIH Office of Laboratory Animal Welfare ( OLAW ) . KSU Animal Welfare Assurance Number is D16-00369 ( A3609-01 ) , and USDA Certificate Number is 48-R-0001 . Kansas State University utilizes the United States Government Principles for the utilization and care of vertebrate animals used in testing , research and training guidelines for appropriate animal use in a research and teaching setting . Bacterial strains and plasmids used in this study are listed in S1 Table and cloning strategies used are listed in S1 Text . Clostridium difficile strains were grown anaerobically ( 10% H2 , 10% C02 and 80% N2 ) in TY ( Tryptose and Yeast extract ) agar or broth as described previously [60 , 66] . Erythromycin ( Erm; 2 . 5 μg ml-1 ) , Lincomycin ( Linc 20ug/ml ) , Cefoxitin ( Cef; 25 μg/ml ) , thiamphenicol ( Thio; 15 μg ml-1 ) were added to culture medium whenever necessary . Sporulation was induced in respective C . difficile strains by growing them in 70:30 sporulation medium ( 63 g Bacto-Peptone , 3 . 5 g Protease-Peptone , 11 . 1 g BHI , 1 . 5 g Yeast-Extract , 1 . 06 g Tris base , 0 . 7 g NH4SO4 , 15 g agar per liter ) [67] . Escherichia coli strain S17-1 [68] was used for conjugation and cultured aerobically in Luria-Bertani ( LB ) broth and supplemented with chloramphenicol ( 25μg ml-1 ) or ampicillin ( 100μg ml-1 ) as indicated . ClosTron gene knockout system [69] was used to construct sinRR’ and sinR’ mutants . For sinRR’ disruption , the group II intron insertion site between nucleotides 141 and 142 in sinR gene in the antisense orientation was selected using the Perutka algorithm , a Web-based design tool available at http://www . Clostron . com . For sinR’ mutant construction , the group II intron insertion site between nucleotides 129 and 130 in the sense direction was selected . The designed retargeted intron was cloned into pMTL007-CE5 as described previously [59 , 70] . The resulting plasmids pMTL007-CE5::Cdi-sinR-141s or pMTL007-CE5::Cdi-sinR’-129s was transferred into C . difficile cells by conjugation as described earlier [59 , 70] . The potential Ll . ltrB insertions within the target genes in the C . difficile chromosome was conferred by the selection of erythromycin or lincomycin resistant transconjugants in 5 μg ml-1erythromycin or 20 μg ml-1 lincomycin plates . PCR using gene-specific primers ( S2 Table ) in combination with the EBS-U universal and ERM primers was performed to identify putative C . difficile mutants . DNeasy Blood and Tissue Kit ( Qiagen ) was used to extract chromosomal DNA from the C . difficile cultures . Primers used throughout the study are listed in S2 Table and S3 Table . Geneclean Kit ( mpbio ) was used to gel extract the PCR products , and QIAprep Spin Miniprep Kit ( Qiagen ) was used to extract plasmid DNA . Standard procedures were used to perform routine cloning . Sporulation assays were performed in 70:30 sporulation medium as described previously [60] . C . difficile strains were grown on 70:30 sporulation agar . After 30 h of growth , cells were scraped from the plates and suspended in 70:30 sporulation liquid medium to an OD600 of 1 . 0 . Cells were immediately serially diluted and plated onto TY agar with 0 . 1% taurocholate to enumerate viable vegetative cells and spores . To determine the number of spores present , 500μl of the samples from each culture were mixed 1:1 with 95% ethanol and incubated for 1hour to kill all the vegetative cells . The ethanol-treated samples were then serially diluted , plated on TY agar with 0 . 1% taurocholate and incubated at 37°C for 24 to 48 hours to enumerate the number of spores . Dividing the number of spores by the total number of CFU and multiplying the value by 100 determined the percentage of ethanol-resistant spores . The results were based on a minimum of three biological replicates . C . difficile strains were grown in 70:30 medium as described above . At indicated time points , 1 ml of culture was removed from the anaerobic chamber , centrifuged at 17 , 000g for 1min and suspended in 30μl of sterile PBS . A thin layer of 0 . 7% agarose was applied to the surface of slide and 2μl of concentrated culture was placed on it . Phase contrast microscopy was performed using 100x oil immersion objective on OLYMPUS BX41 microscope . The PixeLINK camera was used to acquire the view of at least three fields for each strain . All steps in sample preparation were performed at room temperature and solutions were prepared in 1X PBS ( phosphate-buffered saline ) unless indicated otherwise . For transmission electron microscopy , cells ( 1010 ) were fixed overnight in a solution of 2% glutaraldehyde and 2% paraformaldehyde . The cells were thoroughly rinsed with 1X PBS ( 5 minutes each ) and post-fixed with 1% osmium tetroxide with constant rotation for 1–2 hours . The samples were then washed thrice with 1X PBS ( 5 minutes each ) , enblock stained with 2% Uranyl acetate in water for 1hr with light protection , and finally washed three times ( 5 min each ) with distilled water . The cells were further dehydrated in a graded 50% -100% acetone series ( vol/vol ) for 5 minutes and infiltrated in graded EMBED 812/Araldite resin ( Electron Microscopy Sciences ) at RT with constant rotation . Thin sections of polymerized resin were placed on copper grids and stained with 2% alcoholic uranyl acetate and Reynolds' lead citrate respectively . Sections were examined with a transmission electron microscope ( Philips CM100 ) and regions containing the cross-section of the cells were photographed at 80 kV for image analysis . To visualize the flagella , whole bacterial cells harvested from overnight cultures were processed as above and were negatively stained with 2% uranyl acetate before transmission electron microscopy analysis . We isolated total RNA from three biological replicates of each strain belonging to early-stationary phase ( 12 hours after inoculation ) and quality was checked using Agilent 2100 Bioanalyzer . The RNA-Seq was performed as previously described [60] . Briefly , we depleted the rRNA content in the selected samples using Epicenter Bacterial Ribo-Zero kit . Strand-specific single end cDNA libraries were prepared using Truseq Small Stranded Total RNA sample prep kit Illumina as per the manufacturers’ instructions . Illumina HiSeq2000 sequencer ( multiplexing three samples per lane ) was used to sequence libraries . Sequences were cleaned with AlienTrimmer [71] of adapter sequences . Only high-quality sequences with a minimum of 30 nucleotides in length were considered for further analysis . Cleaned genes were aligned to reference genomes ( FN545816 . 1 and AM180355 . 1 ) using Bowtie ( version 1 . 0 . 1 ) [25 , 60 , 72] . DESeq2 version 1 . 8 . 3 was used to perform normalization and differential analysis . Genes were considered differentially expressed if the fold change was ≥ log2 1 . 5 and their adjusted p-value was ≤0 . 05 . SinR , SinR’ and CodY proteins were overexpressed in Rosetta E . coli DE3 cells using pET16B expression system . The ORFs for cloning were PCR amplified from JIR8094 chromosome using gene-specific primers ( listed in S2 Table ) , and the amplified gene fragments were then digested with Xho1 and BamH1 to clone into pET16B digested with the same enzymes . The resulting plasmids were then transformed into E . coli Rosetta DE3 ( Novagen ) competent cells to obtain recombinant strains . To overexpress SinR-6His , and SinR’-6His , the E . coli recombinant strains were grown at 37°C in LB medium containing chloramphenicol ( 25μg ml-1 ) and ampicillin ( 100ug ml-1 ) . Protein expression was achieved by inducing with 1mM IPTG at 17°C overnight . Cells were harvested by centrifugation , and the 6His-tagged proteins were purified by affinity chromatography on Ni++ agarose ( Sigma-Aldrich ) beads following the manufacturer’s recommendations . The anti-SinR used in this study was raised against SinR-His6 in rabbits by Lampire Biologicals ( Everett , PA ) . The anti-SinR’ was raised against SinR’-His6 in mice by Lampire Biologicals ( Everett , PA ) . C . difficile cells for western blot analysis were harvested and washed in 1x PBS solution before suspending in sample buffer ( Tris 80mM; SDS 2%; and Glycerol 10% ) for sonication . Whole cell extracts were then heated at 100°C for 7 min and centrifuged at 17 , 000 g for 1 min , and the proteins were separated by SDS-PAGE and electro-blotted onto PVDF membrane . Immobilized proteins in the membranes were then probed with specific antibodies at a dilution of 1:10 , 000 . The blot was subsequently probed with HRP-conjugated secondary antibodies at a dilution of 1:10000 . Immuno-detection of proteins was performed with ECL Kit ( Thermo Scientific ) following the manufacturer’s recommendations and were developed using Typhoon 9100 scanner . Cytosolic toxins from 16h old C . difficile cultures grown in TY medium were measured as described previously [70 , 73] . In brief , one ml of C . difficile cultures were harvested and suspended in 200 μl of sterile PBS , sonicated and centrifuged to harvest the cytosolic protein . One hundred μg of cytosolic proteins was used to measure the relative toxin levels using C . difficile premier Toxin A &B ELISA kit from Meridian Diagnostics Inc . ( Cincinnati , OH ) . C . difficile cultures were grown until mid-exponential phase at 37°C . After adjusting their OD@600 to 0 . 5 , 3μl of each strain was inoculated by stabbing or spotting into BHI medium with 0 . 3% w/v agar in tubes and plates respectively . After incubation at 37°C , the motility was quantified by measuring the radius of the cultures at different time points . Motility assay was performed in 4 replicates and independently repeated at least three times . To express SinR’-GST protein we cloned the sinR’ gene in the pGST-parallel2 expression system [74] . First , the sinR’ gene was PCR amplified using primers ORG619 and ORG620 ( S2 Table ) and R20291 chromosomal DNA as a template . The PCR fragments were then cloned in between NcoI and SalI sites of the pGST-parallel2 vector . The resulting plasmid was then transformed into E . coli Rosetta DE3 competent cells to obtain recombinant strain . To overexpress SinR’-GST , E . coli recombinant strains were grown at 37°C in LB medium containing chloramphenicol ( 25μg ml-1 ) . Protein expression was achieved by inducing with 1mM IPTG at 17°C overnight with mild agitation . To perform the pull-down experiment , 200 μgs of whole cell lysate proteins from the E . coli cells expressing SinR’-GST was mixed with ~20 μgs of purified SinR-6His protein and incubated at 4°C for 1hr . The mixture was then passed through the Ni++ affinity column ( Sigma-Aldrich ) to trap and elute SinR-6His protein . Whole lysates from E . coli cells expressing GST alone was also mixed with purified SinR-6His protein , and this control mixture was processed in the same way as the test sample . The elutes from Ni++ columns were then separated by SDS-PAGE and were electro blotted onto PVDF membrane . Membranes with immobilized proteins were then probed with either Anti-6His antibodies at 1:10 , 000 dilution or with anti-GST antibodies at the dilution of 1:5000 . Immunodetection of proteins was performed with Pierce ECL 2 Western blotting Substrate Kit ( Thermo Scientific ) and the Typhoon 9100 scanner . SinR and SinR’ binding was performed with radioactively labeled DNA probes . The codY upstream and the gluD upstream regions were amplified using primer pairs ORG629- ORG630 and ORG72-ORG73 , respectively and the products were cloned into a pGEMT cloning vector . The region was then excised from the plasmid construct using EcoRI and was radiolabeled using Klenow fragment of DNA polymerase I ( NEB . labs ) and [α- 32 P]dATP-6000 Ci/mmol ( PerkinElmer Life Sciences ) . Binding experiments with radioactively labelled codY upstream DNA with SinR-6His or SinR’-6His was performed using reaction buffer containing 10 mM Tris–HCl ( pH 8 . 0 ) , 0 . 1 mM DTT , 150 mM KCl , 0 . 5mM EDTA , 0 . 1% Triton X-100 and 12 . 5% glycerol . For binding experiments containing both SinR and SinR’ , proteins were mixed in the reaction buffer at a specified concentration and were incubated at room temperature for 30 minutes before adding the DNA probe . Reactions were loaded onto a 6% native polyacrylamide gel in 1XTBE ( Tris/Borate/EDTA ) and subjected to electrophoresis at 100 V for 45 minutes . Gels were then dried , and the autoradiography was performed with Molecular Dynamics Phosphor-Imager technology . For the CodY binding experiments , the upstream region of the sin locus with the predicted CodY binding sequence ( shown as underlined ) 5’ TAGAAA ATTTTTTTAATTTTCAAAATATATTCTACATATCTAA was synthesized and was labeled with [γ- 32 P]dATP-6000 Ci/mmol ( PerkinElmer Life Sciences ) using T4 polynucleotide kinase . It was then annealed with the complementary oligo to generate double-stranded DNA probe . Known CodY binding sequence upstream of the tcdR gene was similarly synthesized ( S2 Table ) and used as a positive control . A non-specific double-stranded DNA was used as negative control ( S2 Table ) . The DNA-protein binding reactions were carried out at room temperature for 30 min in 10μl volume containing 1x binding buffer [10mM Tris pH 7 . 5 , 50mM KCl , 50μg BSA , 0 . 05% NP40 , 10% Glycerol , 10 mM GTP and 2mM ILV ( Isoleucine , Leucine and Valine ) , 100 μg/ml poly dI-dC and 800nM of DNA probe with varying concentration of purified CodY protein . DNA probe in reaction buffer was incubated for 10 min at RT before adding purified CodY-6His protein . The reaction was stopped by adding 5ul of gel loading buffer and electrophoresed at 100V for 1 . 5 h using 6% 1XTBE gel in 0 . 5X TBE buffer containing 10 mM ILV . Gels were then dried , and the autoradiography was performed with Molecular Dynamics Phosphor-Imager technology . Syrian golden hamsters ( 100–120 g ) were used for C . difficile infection . Upon their arrival , fecal pellets were collected from all hamsters , homogenized in 1 ml saline , and examined for C . difficile by plating on CCFA-TA ( Cycloserine Cefoxitin Fructose Agar- 0 . 1% Taurocholate ) to ensure that the animals did not harbor indigenous C . difficile . After this initial screen , they were housed individually in sterile cages with ad libitum access to food and water for the duration of the study . Hamsters were first gavaged with 30 mg/kg clindamycin [59 , 75] . C . difficile infection was initiated five days after clindamycin administration by gavage with vegetative cells . We used vegetative C . difficile cells because of the test strain R20291::sinRR’ is asporogenic and do not produce any spores . Bacterial inoculums were standardized and prepared immediately before challenge as described in our earlier study [59] . They were transported in independent 1 . 5 ml Eppendorf tubes to the vivarium using the Remel AnaeroPack system ( one box for each strain ) to maintain viability . Immediately before and after infecting the animal , a 10 μL sample of the inoculum was plated onto TY agar with cefoxitin to confirm the bacterial count and viability . There were five groups of animals , including the uninfected control group . Ten animals per group were used for the infection . Approximately , 2000 C . difficile vegetative cells of R20291 strain and R20291::sinRR’ were used for the animal challenge . In the uninfected control ( group 5 ) only five animals were used , and they received only antibiotics and sterile PBS . Animals were monitored for signs of disease ( lethargy , poor fur coat , sunken eyes , hunched posture , and wet tail ) every four hours ( six times per day ) throughout the study period . Hamsters were scored from 1 to 5 for the signs mentioned above ( 1-normal and 5-severe ) . Fresh fecal pellets were collected daily from every animal to monitor C . difficile colonization until they began developing diarrheal symptoms . Hamsters showing signs of severe disease ( a cumulative score of 12 or above ) were euthanized by CO2 asphyxiation . Surviving hamsters were euthanized 15 days after C . difficile infection . Thoracotomy was performed as a secondary mean of death . The cecal contents from these hamsters were collected in 15ml Nalgene tubes , secured air tight and were transported to the lab using Remel AnaeroPack system . They were then immediately subjected to CFU enumeration . For CFU enumeration , the daily fecal samples or the cecal contents collected post-mortem were resuspended in 1X PBS , serially diluted and plated onto CCFA agar with 0 . 1% Taurocholate ( CCFA-TA ) . The CFU were counted after 48 h of incubation . The survival data of the challenged animals were graphed as Kaplan-Meier survival analyses and compared for statistical significance using the log-rank test using GraphPad Prism 6 software ( GraphPad Software , San Diego , CA ) .
|
In Bacillus subtilis , sporulation , competence and biofilm formation are regulated by a pleiotropic regulator called SinR . Two sinR homologs are present in C . difficile genome as an operon and henceforth labeled as sinR and sinR’ . Our detailed investigation revealed that in C . difficile , the SinR and SinR’ are key master regulators needed for the regulation of several pathways including sporulation , toxin production , and motility .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"bacteriology",
"medicine",
"and",
"health",
"sciences",
"gut",
"bacteria",
"toxins",
"pathology",
"and",
"laboratory",
"medicine",
"gene",
"regulation",
"pathogens",
"bacillus",
"microbiology",
"vertebrates",
"animals",
"mammals",
"toxicology",
"toxic",
"agents",
"bacterial",
"sporulation",
"prokaryotic",
"models",
"experimental",
"organism",
"systems",
"bacteria",
"bacterial",
"pathogens",
"research",
"and",
"analysis",
"methods",
"clostridium",
"difficile",
"hamsters",
"microbial",
"physiology",
"medical",
"microbiology",
"gene",
"expression",
"microbial",
"pathogens",
"genetic",
"loci",
"pathogen",
"motility",
"bacterial",
"physiology",
"rodents",
"eukaryota",
"virulence",
"factors",
"bacillus",
"subtilis",
"genetics",
"biology",
"and",
"life",
"sciences",
"amniotes",
"organisms"
] |
2018
|
Pleiotropic roles of Clostridium difficile sin locus
|
The accurate prediction of the structure and dynamics of DNA remains a major challenge in computational biology due to the dearth of precise experimental information on DNA free in solution and limitations in the DNA force-fields underpinning the simulations . A new generation of force-fields has been developed to better represent the sequence-dependent B-DNA intrinsic mechanics , in particular with respect to the BI ↔ BII backbone equilibrium , which is essential to understand the B-DNA properties . Here , the performance of MD simulations with the newly updated force-fields Parmbsc0εζOLI and CHARMM36 was tested against a large ensemble of recent NMR data collected on four DNA dodecamers involved in nucleosome positioning . We find impressive progress towards a coherent , realistic representation of B-DNA in solution , despite residual shortcomings . This improved representation allows new and deeper interpretation of the experimental observables , including regarding the behavior of facing phosphate groups in complementary dinucleotides , and their modulation by the sequence . It also provides the opportunity to extensively revisit and refine the coupling between backbone states and inter base pair parameters , which emerges as a common theme across all the complementary dinucleotides . In sum , the global agreement between simulations and experiment reveals new aspects of intrinsic DNA mechanics , a key component of DNA-protein recognition .
Binding of DNA to proteins or small molecules is modulated by subtle sequence-dependent variations inherent to the structure and dynamics of free DNA , which facilitate or disfavor the structural fit with cognate partners [1–4] . Given the many DNA targets , a purely experimental characterization of their structure and dynamics is an enormous task . The structural biology of DNA would be greatly helped if one could describe and predict the sequence-dependent intrinsic mechanical and structural preferences of the double helix . That would pave the way to a fuller understanding of DNA malleability in direct and indirect readout . Molecular Dynamics ( MD ) simulations in explicit solvent can potentially explore the properties of any B-DNA sequence of moderate length , considering the extensive sampling afforded by modern computational resources [5 , 6] . However , MD simulations are only as reliable as the underlying energy model , typically treated with a classical force-field . Development of force-fields is complex , requires extensive efforts , and needs precise reference experimental data [7] . This latter requirement has been a complicating factor for DNA , given the paucity of reliable experimental data reflecting the fine structural details of DNA in solution [8–10] . The situation has improved in recent years , especially with respect to the DNA backbone , for which additional experimental information have been gathered from X-ray crystallography and NMR ( see below ) . In response , force-field shortcomings regarding the DNA backbone were addressed , including via QM studies on model compounds [11–13] , motivated by the realization that the backbone is an essential component of the intrinsic mechanical couplings in DNA . Statistical analyses of X-ray structures of free B-DNA have unveiled that , among the five dihedral angles along the phosphate linkage , ε and ζ present bi-modal distributions [14–19] , referred to as BI with ε/ζ:trans/g- and BII with ε/ζ:g-/trans [20 , 21] . In contrast , α , β and γ appear to prefer overwhelmingly one conformation ( α/β/γ:g-/t/g+ ) [14 , 15 , 17 , 19] . Importantly , the BI and BII conformers are associated to distinctive values of the helicoidal parameters , X-disp ( base displacement ) , slide , roll and twist [15 , 16] . In addition , the density of BI or BII phosphate groups in a window of 4 consecutive base pairs is coupled to the groove dimensions [22] . Hence , the modulation of B-DNA shape observed in X-ray structures is associated with the conformation of ε/ζ backbone dihedrals . NMR solution studies echoed these findings and provided additional information about the sequence dependent behavior of the BI and BII populations . In NMR , this equilibrium is reflected by the 31P chemical shifts ( δPs ) [21 , 23] , which can be translated in terms of BII percentages [24 , 25] . Correlations between NMR-measured δPs and internucleotide distances [24 , 26] reflect the coupling between the backbone states and inter base pair parameters observed in X-ray structures . Consistency between NMR and X-ray results extends to the relation between the BII densities and the width of the minor groove , which can also be observed by NMR [27] . In addition , the compilation of a sizeable set of δPs documented the effect of B-DNA sequence on BII propensities [28] , initially inferred from X-ray structures [15 , 16] . Of the 10 unique complementary dinucleotides ( NpN•NpN ) , CpG•CpG , CpA•TpG , GpC•GpC , GpG•CpC are characterized by BII percentages markedly higher than the average ( 21% ) ; ApN•NpT ( N: any base ) and TpA•TpA can be globally considered as BI-rich; GpA•TpC is an intermediate case , with BII percentage only slightly lower than average . These intrinsic sequence-specific BII propensities in solution were summarized on a scale called TRX [28] , by reference to the couplings with Twist , Roll and X-disp . TRX was recently validated further by a large set of δPs collected on new DNA oligomers [27] . This detailed information on DNA in solution offers a precious framework for testing and refining DNA force fields . Thus , the AMBER force fields Parm98 [29] and Parm99 [30] stabilized artefactual α/γ conformations that caused severe distortions in DNA [19 , 31] . Such undesirable α/γ transitions were corrected in the subsequent potential Parmbsc0 [12] . First observed by NMR [32 , 33] and then generalized and quantified in the TRX scale [34] , the modulation of CpG BII propensity by the 3'- and 5'-neighbors was qualitatively retrieved by Parmbsc0 [35] . Also , the sensitivity of the BI ↔ BII equilibrium to the type of monovalent cation ( K+ , Na+ ) was demonstrated by NMR [26] . Parmbsc0 simulations do not seem to reproduce this dependence , yet they suggest a mechanism that could explain how K+ and Na+ affect the backbone motions [35] . Concerning the CHARMM family of force-fields , the early thorough systematic calibration of the DNA backbone torsional energetics for CHARMM27 [17 , 36] prevented artefactual α/γ transitions and resulted in a force-field which treats B-DNA robustly [6 , 37 , 38] . Importantly , CHARMM27 , like Parmbsc0 , correctly represent the mechanical coupling between the backbone states and the helical parameters [9 , 38 , 39] . Nevertheless , the remaining shortcomings in Parmbsc0 MDs [5 , 35 , 40 , 41] cannot be ignored , in particular regarding CpG , CpA and TpG that show a systematic deficit in BII with respect to the NMR data [27 , 28 , 32–34 , 42–45] . The CHARMM27 force-field also did not reproduce the experimentally documented BII percentages [9 , 39] . A simulation of the Drew-Dickerson dodecamer with Parmbsc0 [40] and a NMR/modeling study with CHARMM27 [46] also raised the issue of unrealistic BII propensities . In response , two force-fields were recently conceived to improve the DNA backbone representation: Parmbsc0εζOLI [13] , derived from Parmbsc0 - , and CHARMM36 [11] , built on CHARMM27 . Parmbsc0εζOLI and CHARMM36 were developed guided by DNA X-ray structures and a small set of BII percentages extracted from NMR . In initial tests with B-DNA , both force fields notably increased the sampling of the BII form compared to prior potentials [11 , 13] . Since twist and groove shape are coupled to the BI ↔ BII equilibrium , the structural outcome obtained with Parmbsc0εζOLI significantly differs from that yielded by Parmbsc0 [13] . These initial tests are encouraging and call for a more systematic examination of the performance of these potentials , especially in the light of experimental data not used to train the force-fields . The present work exploits a wealth of recent 31P NMR chemical shifts on the DNA backbone motions , to thoroughly evaluate the performance of the Parmbsc0εζOLI and CHARMM36 potentials . These data were collected on four DNA dodecamers [27] , independent of those used to develop those force-fields . Together , the dodecamers cover a 39 bp segment in the 5’ half of sequence 601 , the best artificial sequence at forming nucleosome core particle [47] , which is therefore important to understand how DNA is packaged . The TRX approach [28] combined with the analysis of the 31P chemical shifts of the four dodecamers [27] provides evidence that the intrinsic structural characteristics of the free sequence 601 largely account for its strong affinity for the histone core . In addition to their biological relevance , the 72 dinucleotides ( NpN ) of the four dodecamers behave as expected from the TRX scale regarding the effect of sequence on the BII propensities [27] . These dodecamers and the attending experimental information are therefore ideally suited to evaluate Parmbsc0εζOLI and CHARMM36 , with emphasis on the representation of the BI ↔ BII equilibrium and the coupled helicoidal parameters . Importantly , we show how the improvements brought by these new potentials lead to new insights into DNA structure and dynamics , which are essentially consistent across the two force-fields . The first step was to compare the BII percentages inferred from δP measurements to those generated by MDs . The simulated fine modulation by the sequence is not yet fully satisfactory . For instance the simulated BII populations of some steps , as those of GpC with both force fields , tend to be underestimated . However , Parmbsc0εζOLI and CHARMM36 represent the backbone behavior much more realistically than previous force fields . CHARMM36 in particular shows a very good ability to obtain BII-rich steps . This advance enabled to examine for the first time the conformational combinations corresponding to the states of facing phosphate groups , i . e . BI•BI , BI•BII , BII•BI and BII•BII . We find that the conformational states of the two facing phosphate groups of any complementary dinucleotide are not correlated in either Parmbsc0εζOLI or CHARMM36 simulations . An important practical consequence is that the populations of the combinations of facing phosphates can be easily deduced from the overall individual BII populations inferred experimentally for every phosphate . This approach reveals that the four dodecamers contain a sizable number of steps where BII-containing states , BI•BII , BII•BI and BII•BII , dominate . Such quantification is critical for accurately describing the conformational landscape explored by the complementary dinucleotides , because the backbone combinations are tightly associated with helical parameters , as documented here . Overall , our results deepen our understanding of the intrinsic B-DNA mechanics , which is a key player in the indirect readout of DNA sequences by proteins .
Each of the four dodecamers ( Table 1 ) was simulated with the Parmbsc0εζOLI [13] ( P-MDs ) or CHARMM36 [11] ( C-MDs ) force-field , resulting in a total of 8 MDs . The MDs of Oligo 1 , 2 and 3 were 450ns each , while for Oligo 4 the trajectories were extended to one microsecond . Additional sampling was performed on Oligo 4 since its alternation of BI and BII-rich dinucleotides is especially relevant to test the convergence of backbone dynamics . During the present simulations , the base pairs N2→N11•N14→N23 were stable , with ~99% of Watson-Crick pairing . The root mean square deviations ( RMSDs ) between a regular canonical B-DNA and the simulated snapshots fluctuated around 2 . 6±0 . 6 Å in P-MDs and 2 . 1±0 . 5 Å in C-MDs ( S1 Fig ) . The slightly larger RMSDs for P-MDs versus C-MDs gave the first indication of subtle differences between the force-fields , but one should refrain from interpreting these differences in terms of relative validity of the two force-fields , since canonical B-DNA is a somewhat artificial construct . Then , we examined the five dihedral angles of the phosphodiester backbone , α , β , γ , ε and ζ . In both P- and C-MDs , α/β/γ conform to the canonical g-/trans/g+ pattern observed in free DNA [14 , 15 , 17–19] . The torsions ε and ζ , which undergo correlated motions , define the BI and BII states ( Fig 1 ) . The convergence of the BII populations is of evident relevance , especially to compare simulated BII percentages to their experimental counterpart . Previous analyses of very long trajectories ( up to ~45 μs ) with Parmbsc0 and CHARMM36 showed reasonable convergence of the fast motions ( timescale < 100ns ) on internal parts of DNA oligomers after only ~50ns [5] . A similar conclusion was drawn from μs simulations with Parmbsc0 , using as convergence criteria the average helical and backbone parameters [41] . These previous studies indicate that the timescale of the present MDs should be amply sufficient to investigate the backbone motions . Indeed , our results confirm this expectation , keeping in mind that constraints were applied to the terminal base-pairs to maintain their Watson-Crick base-pairing . A detailed justification of the protocol is given in Materials and Methods . Clearly , the BII population of some phosphates did not converge over the first 50 ns , which may be considered as a reasonable equilibration time . Thus , for each phosphate , convergence was monitored by plotting its BII percentage over increasing trajectory lengths , from 150ns upwards ( S2 Fig ) . For each Oligo treated with Parmbsc0εζOLI or CHARMM36 , convergence of the BII populations was reached within the simulation times , including for phosphates with high BII populations ( S2 Fig ) . An additional test was performed with the MDs of Oligo 4 extended to 1 μs , by comparing the beginning -from 50 to 150ns- and the end -from 900 to 1000ns- of trajectories ( S3 Fig ) . This analysis produced very similar BII percentages on each step of Oligo 4 , with only slight differences ( 8% for the worst case ) on some BII-rich steps . Thus , the first 100ns of production ( from 50 to 150ns ) , while not sufficient to ensure a complete convergence , surprisingly offer a rather good estimation of the backbone behavior , supporting the expectation that the simulations are essentially converged for practical purposes with respect to the BI/BII balance on the 450ns time scale . We recall that the MDs were performed restraining the Watson-Crick hydrogen bonds in the first and last base pairs . We chose this protocol since , with unrestrained MDs , convergence issues were observed in conjunction with fraying events involving larger than expected motions of the terminal regions , consonant with previous reports [5 , 13 , 48] . Since the structural signature of these long-lived fraying events in the unrestrained MDs is not supported by the NMR measurements ( see “Restrained base pairing on the first and last base pairs” in Materials and Methods ) , Watson-Crick pairing restraints of the first and last base pairs were applied in the analyzed MDs . This remedied the convergence concerns otherwise observed in the unrestrained MDs ( see the example of restrained and unrestrained 1 μs P-MDs of Oligo 4 in S4 Fig ) . Importantly , however , the BII percentages calculated on the central part ( N3→N10•N15→N22 ) of the dodecamers are quasi identical ( correlation coefficient of 0 . 98 ) in both restrained and unrestrained MDs . This indicates that the behavior of the internal steps was reproducible in different conditions , another reassuring element concerning the convergence of the simulations presented here . Overall , we observe that the DNA backbone dynamics is essentially converged with MDs of several hundredth ns . This convergence timeframe is realistic considering that the phosphate groups undergo rapid ( nano-picosecond timescale ) conformational exchange according to NMR [49 , 50] . Convergence in terms of sampling of the backbone states allows us to concentrate the following analysis on the influence of the force-fields . In free DNA X-ray structures the distribution of the pseudo-angle ( ε-ζ ) is characterized by two major peaks centered around ( ε-ζ ) = -90° ( BI , ε in trans and ζ in g- ) and ( ε-ζ ) = 90° , ( BII , ε in g- and ζ in trans ) ( S5 Fig ) . Between these two maxima , a region covering ( ε-ζ ) values from -60 to +70° contains phosphate linkages with ε:trans typical of BI and ζ:trans typical of BII . Therefore this region may be considered ambiguous in terms of BI/BII categorization . The separation between BI and BII is commonly set at the minimum of the ( ε-ζ ) distribution , which is close to ( ε-ζ ) = 0 in the X-ray distribution ( S5 Fig ) . BI and BII are thus usually characterized by negative and positive ( ε-ζ ) values , respectively . The pattern observed in the X-ray structures for the ( ε-ζ ) distribution is globally preserved in P-MDs and C-MDs , while influenced by the force-fields ( Fig 2 ) . Thus , the operational definition of the BI and BII states in MDs must be carefully scrutinized , and possibly adapted . In addition to the ( ε-ζ ) histograms , the sugar populations in the south , east and north puckers ( e . g . Oligo 4 sugars in S6 Fig ) were considered , since this criterion is relevant to the definition of BI and BII . Indeed , crystallographic and NMR investigations established that BI is tolerant in terms of surrounding 5’ and 3’ sugar puckers , while BII is restricted to south puckers , especially with respect to 5' sugars [15 , 18 , 50] . The ( ε-ζ ) histogram of C-MDs has a minimum at ( ε-ζ ) = 30° , located at a tail of the ε/ζ:trans/trans region ( Fig 2 ) . 5’ south sugars are observable in both BI and BII regions; 5' east sugars fall in ( ε-ζ ) from -110 to -50° , inside the conventional BI region; 5' north sugars are associated to a larger range of ( ε-ζ ) values , but are suppressed above ( ε-ζ ) = 30° ( Fig 2 ) . This sugar behavior and the minimum of ( ε-ζ ) at 30° offer an analogy with the X-ray observations , such that ( ε-ζ ) = 30° was deemed suitable to separate BI from BII with CHARMM36 . Using instead the conventional cutoff ( ε-ζ = 0 ) to separate BI from BII would not result in a dramatically different description , since the average BII population inferred with ( ε-ζ ) > 0° would only be 4% higher than that based on ( ε-ζ ) > 30° . However , in view of the distribution of 5’ north sugars , the criterion ( ε-ζ ) > 30° was preferred to define BII with CHARMM36 . Incidentally , the present C-MD data confirm the strong association between BII and 5’ south sugars . With Parmbsc0εζOLI , no clear minimum is observed in the ( ε-ζ ) distribution between the BI and BII peaks , which are separated by a flat ( ε-ζ ) distribution ( Fig 2 ) . This intermediate region , centered around ( ε-ζ ) = 0 , contains ε/ζ:trans/trans conformers ( Fig 2 ) , which represent 10% of the snapshots . In absence of a clear minimum in the ( ε-ζ ) distribution , histograms of ε and ζ were considered . This approach was adopted in previous Parmbsc0 trajectories [15 , 18 , 50 , 51] , where the BII linkages were defined relative to the minimum of the distribution . Here , with the minimum of the ζ distribution at ζ = 230° , this approach would designate as BII the range above ( ε-ζ ) = -50° , a strongly negative value . Conversely , the transition from BI to BII would be at ( ε-ζ ) = 40° if chosen to be at the minimum of the ε histogram ( ε = 240° ) . So , the ε and ζ histograms do not provide a coherent definition of BI and BII ranges for P-MDs here . In addition , the sugar dynamical regime is of little help since Parmbsc0εζOLI generates only a few north sugars ( Fig 2 and S6 Fig ) . In absence of any convincing more specific rationale to assign the ε/ζ:trans/trans snapshots to either BI or BII with Parmbsc0εζOLI , we adopted the common BII definition , ( ε-ζ ) > 0° . Such a decision is somewhat arbitrary but the uncertainty it introduces is limited . Indeed , shifting the ( ε-ζ ) dividing value by 20° ( ( ε-ζ ) = -20° or +20° ) only changed the BII population by +/-2% . In sum , BII percentages were extracted using ( ε-ζ ) > 0° for P-MDs and ( ε-ζ ) > +30° for C-MDs throughout this work . These considerations illustrate the difficulty in defining BI and BII unambiguously , in a manner which would be meaningful and transferable across different structural models . It also draws the attention to some differences between Parmbsc0εζOLI and CHARMM36 regarding their representations of sugars and the ( ε-ζ ) distribution . Yet , the following sections show that a consistent overall picture emerges from CHARMM36 and Parmbsc0εζOLI . The four dodecamers studied here correspond to 72 dinucleotide steps , excluding the terminal steps . The corresponding 72 31P chemical shifts ( δPs ) were measured and converted to BII percentages , BII%from NMR , using an empirical procedure based on a calibration involving a comparison of NMR and X-ray structural data [24] ( see also Materials and Methods ) . In this procedure , δPs of pure BI and pure BII states are assumed to be sequence-independent , even if they could be modulated by the dinucleotide sequence , as suggested by a computational study [52] . However , previous studies showed that neglecting subtle sequence effect on δPs of pure BI and pure BII produced reasonable estimations [27 , 53] , for instance with points where BI and BII are expected to be equally populated [27] . Another indication of the protocol reliability is the consistency between the average BII percentage either derived from the average δPs of the 72 steps considered here ( 19% of BII , from δPav = -4 . 20 ppm at 30° ) or inferred from statistics of X-ray structures ( 20% of BII ) [15] . A first test of the force fields is to compare the NMR-inferred and simulated BII populations , averaged on the 72 dinucleotides . The simulated overall average BII percentages are 11% in P-MDS and 18% in C-MDs . Thus , Parmbsc0εζOLI somewhat underestimated the BII populations , as noted before [13] . The excellent agreement of the CHARMM36 value with experiment is an obvious improvement compared to CHARMM27 , which severely underrepresented BII [9 , 11] . That the force fields , in particular CHARMM36 , produce overall BII population commensurate with experimental data is very encouraging , considering that the treatment of the backbone by previous force-fields fell outside the experimental range . Since the dinucleotides have markedly different propensities to populate BII [15 , 27 , 28 , 51 , 53] , reproducing the sequence effects is a more stringent test of the force-fields , examined in the following . A previous dataset of 323 measured δPs has established that the 16 dinucleotides ( NpN steps for a single strand in a duplex context ) composing B-DNA are associated with specific δP values [28] . Since δP translates into a BII propensity it implies that the BI/BII populations are primarily controlled by the dinucleotide sequence [28] . The additional 72 δPs considered here conform to this sequence pattern , validating the notion of dinucleotide-specific BII propensity [27] . Thus , the sequence-dependent BII populations derived from δPs provide a rare opportunity to test the sequence-dependent behavior of DNA force-fields in solution . One notes that adjustments made to Parmbsc0εζOLI and CHARMM36 to increase the BII populations were not tailored depending on the base sequence , in contrast with , for instance , the CMAP approach [54] . In other words , the same backbone force-field parameters are applied to any sequence . Therefore , differences in the backbone behavior during simulations can only be ascribed to intrinsic sequence-dependent properties . To examine whether Parmbsc0εζOLI and CHARMM36 reproduce the effect of sequence on BII populations , the simulated BII percentages were compared to their experimental counterparts , considering the individual phosphates ( BII%from MD versus BII%from NMR , given in S1 Table ) . BII%from MD and BII%from NMR are overall moderately correlated ( Table 2 and S7 Fig ) . The simulated BII percentages of half of the 72 steps ( 53% for both P-MDs and C-MDs ) are within BII%from NMR ±10% , where the 10% interval corresponds to the tolerance allowed around the NMR-based BII percentages ( see Materials and Methods ) . The comparison between BII%from NMR and BII%from MD is shown in Fig 3 for each non-terminal phosphate of the four dodecamers . A more detailed analysis , in particular on the dinucleotides present in several occurrences in the dodecamers , shows that Parmbsc0εζOLI and CHARMM36 correctly reproduce the low or moderate BII%from NMR ( < 20% ) of CpT , GpT , ApC , ApG , ApA and TpT ( Fig 3 and Table 3 for the steps present in several occurrences in the dodecamers ) . However , this overall agreement suffers some shortcomings . The BII population of TpA tends to be either underestimated or overestimated with Parmbsc0εζOLI and CHARMM36 , respectively . GpC steps in BII are quasi systematically underestimated by both force fields , as well as one of the three CpA•TpG in Oligo 4 . Parmbsc0εζOLI generated too low BII percentages on CpC and GpG in Oligo 2 and most CpG ( seven on a total of ten ) . In C-MDs , the representation of CpC , GpG and CpG is reasonable whereas inversions of BII% occurs at two complementary CpG•CpG steps . Indeed , the NMR gives asymmetrical BII percentages for C2pG3•C22pG23 in Oligo 1 , with 79% of BII for C2pG3 and 42% for C22pG23 . C-MD gave the reverse , with 30% and 75% of BII for C2pG3 and C22pG23 , respectively . A similar situation arises for C10pG11•C14pG15 in Oligo 3 . Our results confirm that adjustments specifically aimed at enhancing access to the BII state produce convincing , positive effects , especially perceptible in CHARMM36 . That the increase in simulated BII populations is not distributed uniformly along the sequences ( Fig 3 ) is not trivial since , as noted above , the computational models were not parametrized to reproduce the BII% for specific dinucleotides , but were only adjusted to be generically more permissive to BII . Admittedly , discrepancies still exist between the experimental sequence effect on the BI↔BII equilibrium and Parmbsc0εζOLI or CHARMM36 . However , an essential point is that the simulations are now sufficiently BII-rich to extend the analysis to aspects of the backbone dynamics that eludes experimental approaches . The phosphate groups facing each other across the strands can adopt homogeneous combinations , BI•BI or BII•BII , or hybrid combinations , BI•BII or BII•BI ( denoted here BI•BII|BII•BI , where the vertical bar means logical “or” ) . The populations of these combinations are especially meaningful from the point of view of B-DNA mechanics , because inter base pair parameter values are associated to the conformational states of two facing phosphate linkages [15 , 16 , 24 , 40 , 55–58] , as also addressed below . The behavior of facing phosphate linkages cannot be deduced from δP measurements , which report time and ensemble-averaged BII percentages for individual phosphates . The present simulations offer the opportunity to inspect the dynamic behavior of phosphate linkages in complementary dinucleotides and to estimate possible correlation . Indeed , several steps in C-MDs , in particular CpG•CpG , CpC•GpG and TpA•TpA , adopt the three combinations , BI•BI , BI•BII|BII•BI or BII•BII ( Table 4 and Fig 4 ) . In P-MDs , BI•BI and BI•BII|BII•BI are also frequently observed , but the BII•BII populations are almost inexistent ( Table 4 ) , consistent with Parmbsc0εζOLI generating fewer BII conformers than CHARMM36 . Fig 5 illustrates the statistics of the transitions between the facing phosphate combinations for steps that adopt BII•BII in both P-MDs and C-MDs . The same result holds for any other complementary step investigated here in which the facing phosphates undergo BI ↔ BII transitions . With both force fields , the large majority of the transitions between BI•BI , BI•BII|BII•BI and BII•BII involves only one of the two facing phosphates ( BI•BI ↔ BI•BII or BII•BI; BI•BII or BII•BI ↔ BII•BII ) . BI•BII|BII•BI ↔ BII•BII are infrequent in P-MDs , the BII•BII state being poorly populated . In both P-MDs and C-MDs , the simultaneous transitions of two phosphate states ( BI•BII ↔ BII•BI; BI•BI ↔ BII•BII ) are very rare , representing at most 5% of the total number of transitions ( Fig 5 ) . The populations of BI•BI , BI•BII|BII•BI and BII•BII can be addressed with simple elements from probability theory , summarized here before comparison to the MD data . Pi ( BII ) is the probability that phosphate i is in BII , the complementary event has probability Pi ( BI ) = 1 –Pi ( BII ) . The states of facing phosphate pairs i , j are characterized by pair probability distributions , Pi , j ( BI•BI ) , Pi , j ( BI•BII ) , Pi , j ( BII•BI ) and Pi , j ( BII•BII ) . Because the facing phosphate groups are either BI or BII , the probabilities satisfy the relations: Pi , j ( BI•BI ) + Pi , j ( BI•BII ) = Pi ( BI ) Pi , j ( BI•BI ) + Pi , j ( BII•BI ) = Pj ( BI ) Pi , j ( BII•BI ) + Pi , j ( BII•BII ) = Pi ( BII ) Pi , j ( BI•BII ) + Pi , j ( BII•BII ) = Pj ( BII ) Summing the last two equations gives: [Pi , j ( BII•BI ) + Pi , j ( BI•BII ) ] + 2 Pi , j ( BII•BII ) = [Pi ( BII ) + Pj ( BII ) ] ( 1 ) The first term on the left in ( 1 ) is the probability of BI•BII|BII•BI , Pi , j ( BII•BI|BI•BII ) . Note that here Pi , j ( BII•BI|BI•BII ) does not denote a conditional probability , but simply the probability of states BII•BI or BI•BII . So Eq 1 is equivalent to Pi , j ( BII•BI|BI•BII ) + 2 Pi , j ( BII•BII ) = Pi ( BII ) + Pj ( BII ) . ( 2 ) Eq 2 is general , as it follows directly from the definitions and it does not rely on any assumption about the independence ( correlation ) of facing phosphate groups . One now examines the case when the states of the two facing phosphate groups are independent of each other . Then , the pair probabilities factorize: Pi , j ( b•b' ) = Pi ( b ) Pj ( b' ) , where b and b’ stand for any of the phosphate states . In particular , we have Pi , j ( BII•BII ) = Pi ( BII ) Pj ( BII ) ( 3 ) Pi , j ( BII•BI|BI•BII ) = Pi ( BII ) + Pj ( BII ) − 2Pi ( BII ) Pj ( BII ) . ( 4 ) Eq 4 follows from Eqs ( 2 ) and ( 3 ) and from the relations: Pi ( BI ) = 1- Pi ( BII ) and Pj ( BI ) = 1- Pj ( BII ) . Eqs 3 and 4 mean that , under the assumption of statistical independence of the two individual facing phosphates , the knowledge of the single phosphate probabilities Pi ( BII ) and Pj ( BII ) is sufficient to find the probabilities of BII•BI|BI•BII , BII•BII , and then also of BI•BI by using equation: Pi , j ( BI•BI ) = 1– [Pi , j ( BII•BI|BI•BII ) + Pi , j ( BII•BII ) ] . The next step was to test the possibility of uncorrelated facing phosphates against data collected from the MDs . Thus , Pi , j ( BII•BI|BI•BII ) , Pi , j ( BII•BII ) , Pi ( BII ) and Pj ( BII ) were evaluated as the proportions of these states in the MD trajectories; Pi , j ( BII•BI|BI•BII ) was compared to [Pi ( BII ) + Pj ( BII ) −2Pi ( BII ) Pj ( BII ) ] in P-MDs and C-MDs; Pi , j ( BII•BII ) was compared to [Pi ( BII ) Pj ( BII ) ] in C-MDs only , since they generate sizable BII•BII populations in contrast with P-MDs . The agreement between the compared quantities is clearly visible in Fig 6 , with correlation coefficients of 0 . 99 . That is , the distribution of BII steps between the BI•BII|BII•BI and BII•BII combinations in complementary dinucleotide matches Eqs 3 and 4 very well . Thus , the conformational states of the facing phosphates are statistically independent of each other . In sum , the ability of both Parmbsc0εζOLI and CHARMM36 to generate phosphates visiting BII enabled to gain new insights into their dynamics and populations . Thus , simultaneous transitions of two facing phosphate groups are very rare . The two force-fields unambiguously support the notion of statistical independence of the conformational states of individual , facing phosphates . This means that the populations of the three combinations of facing phosphates can be simply expressed from the BII propensities of individual phosphate groups , in particular from experimental data , as developed in the next section . Considering that the notion of uncorrelated facing phosphate is convincing , the probabilities of states BI•BII|BII•BI and BII•BII were calculated using Eqs 3 and 4 , respectively , and equating Pi ( BII ) and Pj ( BII ) to Pi from NMR ( BII ) and Pj from NMR ( BII ) ( equivalent to BII%from NMR given in S1 Table ) . This has the advantage to use the experimental data directly ( δP-derived BII percentages ) to quantify the phosphate states , bypassing the limitations in the simulated estimates of the phosphate state populations . The resulting experimentally inferred BI•BII|BII•BI and BII•BII populations along the four dodecamers are shown in Fig 7 and the values are given in S2 Table . According to this approach , most CpG•CpG and GpC•GpC , as well as the only CpC•GpG , are characterized by high percentages ( 45% and more ) of BI•BII|BII•BI ( Fig 7 ) . CpG•CpG in Oligos 1 and 3 , CpC•GpG in Oligo2 and CpA•TpG in Oligo 4 are in addition more than 20% in BII•BII ( Fig 7 ) . Overall , BI•BI is not the most frequent state in 12 steps , out of a total of 36 , in the four dodecamers . As seen above ( Fig 3 ) , the individual BII percentages extracted from MDs differ from those inferred from NMR; accordingly , the corresponding respective populations of BI•BII|BII•BI and BII•BII are not identical . However , the match between C-MD and experimentally inferred data is reasonable ( S8 Fig ) , with correlation coefficients of 0 . 62 for BI•BII|BII•BI and 0 . 57 for BII•BII . So , CHARMM36 appears to represent the sequence-dependent behavior of the pairs of facing phosphates better than that of individual phosphates ( Table 2 ) . This improvement reflects in part compensatory effects between the two strands of CpG•CpG steps , in which the asymmetric individual BII percentages are inversed in C-MDs and NMR ( see the above section “Sequence-dependent BII propensities from simulations versus NMR” ) . Overall , the realization that the states of facing phosphates are independent enables to derive their populations from δP-based BII percentages . Applying this approach reveals that all the complementary dinucleotides in the four dodecamers populate both BI•BI and BI•BII|BII•BI , some of them also display significant percentages of BII•BII ( Fig 7 and S2 Table ) . This prevalence of BII-containing steps is of real importance for the DNA intrinsic mechanics , as examined next . BI•BI , BI•BII|BII•BI or BII•BII are associated to different values of slide , roll and twist in X-ray structures [15 , 16 , 39] . However , the requirement to select only very high resolution X-ray structures to ensure the accuracy of backbone dihedral angles [59] drastically limits the data for analysis . A previous study [28] underlined that BII conformers in such X-ray structures occur almost exclusively in CpG , CpA , TpG , GpG , and GpC; furthermore , the BII•BII combination was only observed in CpA•TpG . The improved representation of the DNA backbone with Parmbsc0εζOLI and CHARMM36 offers the opportunity to broaden the analysis of the helicoidal parameters associated to the facing phosphate combinations for a larger variety of complementary dinucleotides than in X-ray datasets . Consistent results between both force fields would of course strengthen the conclusions . The mean values of the six inter base pair parameters ( shift , slide , rise , slide , roll and twist ) were calculated for each conformational combination of the facing phosphate groups , after merging all equivalent conformational combinations across complementary steps . Slide , roll and twist are found very sensitive to the facing backbone conformational combinations ( Table 5 ) contrary to invariant shift and tilt ( S3 Table ) . As in a previous study using Parmbsc0 [40] , rise variations are observed , but the change between BI•BI and BII•BII does not exceed 0 . 2 Å in both P-MDs and C-MDs ( S3 Table ) . It is striking that there is almost quantitative agreement on the changes of slide , roll and twist across BI•BI , BI•BII|BII•BI and BII•BII with both force-fields ( Table 5 ) . Yet , CHARMM36 does not increase the twist as much as Parmbsc0εζOLI from BI•BI to BII•BII . Considering the 36 individual complementary dinucleotides of the four oligomers ( Fig 8 ) confirms this concordance . Only isolated departures between Parmbsc0εζOLI and CHARMM36 appear when the variation of the helical parameters is examined in individual complementary steps ( examples in S9 Fig ) . In P-MDs , the rolls of TpA•TpA are systematically 5±2 . 5° larger than in C-MDs , for all backbone combinations; the twist of CpG•CpG and CpA•TpG in BII•BII is 6 . 5±1 . 5° higher in P-MDs than in C-MDs . The MD results not only systematically documented the couplings , but they enabled comparison of the variability ( standard deviations ) of the helicoidal parameters for BI•BI , BI•BII|BII•BI and BII•BII . In both P-MDs and C-MDs , the slide and twist variabilities are greater in BI•BI compared to BI•BII|BII•BI or BII•BII ( Fig 9 ) . A similar , but attenuated , trend is observed for the roll in C-MDs ( where the roll standard deviations are higher than in P-MDs ) . Thus , the simulations suggest that BII containing combinations are stiffer than BI•BI , at least for slide and twist . This , combined with the suppression of north sugars in 5’ of the BII linkages , might entropically disfavor the BII conformers . However , other contributions will affect the net balance of the BI ↔ BII equilibrium . Indeed , the above quantitative analysis makes clear that BII is frequently populated , and is the dominant conformer at some base steps . The concordant results from P-MDs and C-MDs considerably strengthen and extend the view of the couplings between the facing phosphate states and inter base pair parameters gleaned from X-ray structures . Here , MDs inform about the behavior of a large range of dinucleotides , comprising those that are moderately or even barely propitious to BII . They reveal a general mechanical property of free DNA . Thus , BII containing complementary dinucleotides are characterized by more positive slide , more negative roll and higher twist than those in BI . As most steps have access to the BI•BII|BII•BI states ( see the preceding section ) , the DNA deformation cost upon protein binding could be less important than expected when BII-like features are required for recognition . This can be illustrated by the TTAAA sequence in Oligo 3 . This segment is considered as one of the strongest anchoring points in the nucleosome assembly [60–64] , by forming multiple interactions with histones H3 and H4 . In the X-ray structures of nucleosome containing the sequence 601 ( PDB entries 3ZL0 , 3ZL1 [63] and 3MVD [61] ) , TTAAA•TTTAA displays rather variable but globally negative rolls ( -7 ±6° ) . According to both NMR and MDs , in their free state , these steps are mainly in BI•BI , associated to rolls of 4 . 4±3° . However , they also explore the BII•BI|BI•BII states , with rolls of -2 . 5±1° . So , the free TTAAA sequence spontaneously visits conformations closer than expected to its bound counterpart .
Assessing the extent to which MD simulations correctly represent B-DNA structural features in solution , their sequence dependency and populations remains an ongoing challenge and a necessary step to gain confidence in the role that DNA simulations may play in biophysics and structural biology . Part of the difficulty is to obtain experimental data in solution , suitable for comparison with simulations . Here , Parmbsc0εζOLI [13] and CHARMM36 [11] , specifically developed to improve the representation of the DNA backbone , were tested with respect to the sequence-specific BI and BII populations in four dodecamers , derived from 31P chemical shifts ( δPs ) [24] . The results show that the Parmbsc0εζOLI and CHARMM36 potentials produce substantial BII populations , closer to their experimentally inferred counterpart than those obtained with preceding force fields [5 , 9 , 40 , 41] . Many simulated BII propensities of the four dodecamers compare satisfactorily to experiment , a very encouraging achievement . This provides the foundation to understand the factors underpinning the differentiated BI ↔ BII equilibrium behavior across base steps . In particular , this context may be better adapted to investigate the quantum-mechanical origin of the phosphate chemical shifts [51] . However , the experimental sequence effect on BII propensities is still imperfectly reproduced by simulations , each force field displaying its own weaknesses . Parmbsc0εζOLI , as reported by its developers [13] , globally underestimates the BII propensities . The CHARMM36 biases include generating too high BII percentages on TpA or , conversely , suppressing the BII character of GpC and some CpA and TpG . The procedure translating experimental δP to BII% is not devoid of uncertainties [51] , but they would not account for the most severe discrepancies . For instance , the simulated TpA being BII-richer than GpC is clearly inconsistent with both NMR and X-ray data [15 , 24 , 28] . There was no evidence that the residual discrepancies in the sequence effect on the BI and BII populations resulted from insufficient sampling . The BII percentages were found converged well before the half microsecond timescale under monitored MD length increase . Since the BI↔BII exchange occurs in the pico to nanosecond time range in RMN experiments [21 , 49] , which is short compared to current simulation times , one does not expect that the BI↔BII equilibrium distributions would be significantly affected by increasing the sampling time . However , one cannot exclude the existence of hypothetical and currently unknown slower motions , on a longer time scale not probed by the present simulations , which might influence the BI ↔ BII equilibrium . Instead , progress is likely to require further refinements of the potentials . Considering the charged nature of the phosphate groups , it is possible that polarisable DNA force-fields will be required to reach a more satisfactory treatment of the sequence-dependent DNA properties [7 , 65] . Despite some limitations , the present MDs are very helpful to scrutinize the DNA backbone dynamics , especially the behavior of facing phosphate groups within complementary dinucleotides . In that respect , Parmbsc0εζOLI or CHARMM36 yielded similar features despite strong differences in their conception and parametrizations . This convergence strengthens the results . First the simulations indicate that concomitant BI ↔ BII transitions on two facing phosphates are much rarer than transitions involving only one phosphate . Second , statistical analysis of the simulations established that the conformational states of the two individual phosphates within a complementary dinucleotide were independent of each other . As a consequence , the BI•BI , BI•BII|BII•BI and BII•BII populations can be assessed from the individual BII percentages inferred from δPs , using straightforward equations . Importantly , this approach reveals that there is a sizable number of steps where BII-containing states dominate . Thus , more than one fourth of the 36 complementary dinucleotides spend more time in BI•BII|BII•BI and BII•BII than in BI•BI; all the complementary dinucleotides explore BI•BII|BII•BI in addition to BI•BI , with various populations of BI•BII|BII•BI; however , BII•BII is more restricted , apparently only significantly populated in a few types of BII-rich steps . Since the behavior of facing phosphates was uncorrelated in all the 36 complementary steps studied here , one can reasonably infer that this is a general property of any B-DNA . Thus , according to the general and predictable sequence effect on experimental BII propensities [27 , 28] , the BII-containing combinations ( BI•BII|BII•BI and BII•BII ) are expected to be largely represented or even statistically dominant in CpG•CpG , CpA•TpG , GpC•GpC and GpG•CpC . The steps less propitious to BII , GpA•TpC , ApN•NpT ( N: any base ) and TpA•TpA , favor BI•BI but they also present modest fractions of BI•BII|BII•BI . Such findings are of fundamental importance because of the strong couplings between these fine-grained backbone states and the inter base pair parameters of slide , roll and twist , consistent with initial observations on X-ray structures [15 , 16] . Such couplings are not only confirmed here , but further characterized in solution for a broader range of steps , offering a unifying theme underpinning the intrinsic mechanics of B-DNA . Given the recurrent occurrence of BII-containing combinations , it follows that the accessible conformational landscape of most complementary dinucleotides extends into a region characterized by positive slide , negative roll and high twist ( “BII profile” ) . This enhanced intrinsic malleability is relevant to the reading of DNA by proteins , since it increases the repertoire of states which may be critical to initiate selective recognition by facilitating local , structural DNA adjustments upon protein binding . The implication of BII-rich steps in indirect readout mechanisms , via their ability to modulate the DNA shape , has been previously highlighted [9 , 34 , 43 , 66] . In addition , the present work touched upon the counterintuitive example of the BI-rich ( positive rolls ) TTAAA segment in Oligo 3 , which nevertheless also accesses negative rolls ( BII•BI|BI•BII ) in solution , reminiscent of the pattern of negative rolls observed in its nucleosome-bound form . So , the energetic penalty induced by the DNA deformation upon protein binding could be less than expected in many cases , especially when BII-like features are involved for the structural fit between the partners . Thus , the present characterization of free DNA is conceptually relevant to a deeper understanding of the selective recognition of DNA . The investigated force-fields Parmbsc0εζOLI or CHARMM36 may also prove advantageous when simulating such events .
Four oligodeoxyribonucleotides of 12 base pairs ( bp ) ( sequences in Table 1 ) were studied by NMR and MD simulations . These sequences , placed end to end after discarding the terminal base pairs , recompose a continuous 39 bp segment corresponding to the 5’ part of the non-palindromic sequence 601 , selected from SELEX experiments for its very high-affinity for association with the histone octamers [67] . Sample preparation and NMR spectroscopy protocols were reported in a previous study [27] . All the NMR data are available in the Biological Magnetic Resonance Bank , entry 19222 . BII percentages ( BII% ) of the phosphate linkages along the four dodecamers were inferred from the phosphate chemical shifts ( δPs , referenced to trimethyl phosphate ) collected at 30° , using the equation BII ( % ) = 143 δP + 621 [24] . This equation is based on an empirical procedure that assumes the same δPs for purely BI or BII states of every dinucleotide , which is unlikely to be strictly correct [51] . Although previous studies showed that it is a reasonable approximation [27 , 53] , we decided to allow a large tolerance of ±10% on the BII percentages inferred from the experimental δPs to take into account uncertainty on the translation procedure . MD simulations were performed with the Parmbsc0εζOLI force-field [13] using the AMBER 14 program [68] , or the CHARMM36 force-field [11] with program NAMD [69] . Parmbsc0εζOLI and CHARMM36 simulations were carried out following protocols as comparable as possible . Yet , with Parmbsc0εζOLI and CHARMM36 , we used the counterion parameters classically associated to the Amber [70] and CHARMM [71] force-fields , respectively . Parmbsc0εζOLI and CHARMM36 simulations were performed at constant temperature ( 300K ) and pressure ( 1bar ) using the Berendsen algorithm [72] . The integration time-step was 2fs and covalent bonds involving hydrogen were constrained using SHAKE [73] . The non-bonded pair-list was updated heuristically . Long-range electrostatic interactions were treated using the particle mesh Ewald ( PME ) approach [74] . Non-bonded interactions were treated with a 9Å direct space cut-off in AMBER and with a force-shift function from 10 to 12 Å [75] with CHARMM36 . In AMBER , the centre-of-mass motion was removed every 10ps . With both Parmbsc0εζOLI and CHARMM36 , each dodecamer in initial standard B-DNA conformation was neutralized with 22 Na+ ions ( minimal salt condition , ~50 mM Na+ ) , in explicit TIP3P water molecules [76]; the primary boxes were truncated octahedrons with solvent extending 15Å around the DNA . The water molecules and counterions were energy-minimized and equilibrated at 100K around the constrained DNA for 100ps in the NVT ensemble; the entire system was then heated from 100 to 300K in 10ps by 5K increments with harmonic positional restraints of 5 . 0 kcal/mol/Å2 on the DNA atoms . The molecular dynamics simulations were continued in NPT , without notable change in volume . The positional restraints were gradually removed over 250ps and followed by the production phase . During the simulations , distance restraints were applied between base atoms of the first and last base pairs of each dodecamers , to prevent their opening . No restraint was applied on any of the internal nucleotides . The application of restraints on the terminal base pairs is justified in the next section , which highlights the benefits of conducting DNA simulations work alongside experimental characterization . MD snapshots were saved every 1 ps . During the simulations with Parmbsc0εζOLI and CHARMM36 , distance restraints were applied to maintain the Watson-Crick base-pairing in the first and last base pairs of each dodecamers , to prevent their opening . These restraints were applied on the terminal base pairs between base atoms involved in Watson-Crick hydrogen-bonding ( Distancedonor/acceptor = 2 . 9±0 . 2Å ) via a parabolic potential with a force-constant of 10 kcal/mol/Å2 . The application of these restraints was motivated by the behavior of terminal base-pairs in unrestrained simulations , which are not presented in details here . We only give a summary of the unrestrained terminal base-pairs simulations compared to relevant NMR data , to justify the application of restraints in the presented MDs . In the unrestrained simulations with Parmbsc0εζOLI , the first ( N1:N24 , N for any base type ) and last ( N12:N13 ) base pairs were generally open . These terminal bases , once extruded , got involved in various structural patterns that persisted during several hundreds of nanoseconds . In the most prevalent conformations , these bases interact with the penultimate phosphate group , insert into the minor groove or mispair with an antepenultimate base . These conformations impact some χ angles , are associated with unusual backbone dihedrals in N1pN2 , N11pN12 , N13pN14 or N23pN24 , and break the stacking with the 3' or 5' neighbors ( N2 , N11 , N14 or N23 ) . With CHARMM36 , the first two base pairs opened after a few nanoseconds and , as in Parmbsc0εζOLI MDs , adopted multiple non-canonical structures . In these unrestrained MDs , base pair opening only occurred at the termini of the dodecamers and did not propagate further . Such behavior is not specific to our simulations since it was previously described for MDs with Parmbsc0 and Parmbsc0OLI [5 , 13 , 48] or CHARMM36 [5] , which used DNA sequences and simulation protocols different from the ones used here . The re-orientation of the terminal bases towards the internal double stranded part of the DNA are not supported by the NMR data collected at 20 and 30°C on the four dodecamers . In one-dimensional 1H spectra at 30° ( 303K ) , the imino proton resonances are lost in the terminal base pairs while they are clearly observable in all the internal base pairs ( from N2:N23 to N11:N14 ) . This excludes long-lived disruption of the Watson-Crick hydrogen bonds in the penultimate base pairs ( MDs with CHARMM36 ) or mispairing between a terminal base and an antepenultimate base ( MDs with Parmbsc0εζOLI ) . The glycosidic bonds of the terminal nucleotides , probed by the intranucleotide distances H1'-H6/8 , adopt the anti conformation . Furthermore , the numerous sequential NOEs between the penultimate and antepenultimate residues ( N2pN3 , N10pN11 , N14pN15 or N22pN23 ) do not support extensive break of their stacking or abnormal structural features . NMR measurements also give information about the terminal steps , N1pN2 , N11pN12 , N13pN14 and N23pN24 . The corresponding 31P chemical shifts are in the range of the internal phosphates . Intense , well defined 31P-1H4' couplings testify that the 3’ terminal phosphate groups conform to usual backbone conformation , since these couplings are observable only when α/β/γ are in g-/trans/g+ [21 , 77] , the typical conformation of B-DNA . Finally , sequential NOE connectivities , clearly visible in all the terminal steps , imply that the fraying events do not generate large distance between open terminal bases and the penultimate residues . In agreement with a detailed study of this topic [48] , our NMR data indicate that current force fields do not yet provide a satisfactory description of the fraying of the terminal base pairs . The convergence issues induced by the behavior of the terminal regions in our unrestrained MDs are discussed in the Result section . The phosphate group linkages were characterized by torsion angles ε , ζ , α , β and γ following the conventional threefold staggered torsional pattern: gauche plus ( 60±40° ) , trans ( 180±40° ) and gauche minus ( 300±40° ) . The sugar ring conformations were categorized according to their pseudorotation phase angle: north ( 300 to 50° ) , east ( 50 to 120° ) and south ( 120 to 220° ) . DNA structures were analyzed with Curves5 [78] and 3DNA [79] . Both programs produced almost identical helical parameter values . The inter base-pair parameters presented here for complementary dinucleotides NpN•NpN are those from Curves5 . Only the 10 central base-pairs of each dodecamer were analyzed .
|
The ability to simulate computationally the structure and dynamics of biomolecules is a major goal of structural biology . Such simulations require the calculation of the forces and energy of the system , typically with extensively parametrized functions called “force-fields” . Developing reliable force-fields has been very challenging for DNA , mainly because the simulations have to reproduce subtle , complex , sequence-dependent differences that also remain difficult to capture experimentally in solution . Here , we take advantage of an extensive set of recent experimental ( NMR ) data gathered on selected DNA oligomers to test the performance of a new generation of force-fields for DNA simulations . Our results demonstrate impressive progress towards more realistic simulations of DNA . The agreement between experiment and simulations is now good enough to incite further interpretation of the experimental observables and yield new original insights into the intrinsic DNA mechanics . In sum , this work shows that reliable DNA simulations provide a much finer understanding of the structural and dynamical B-DNA behavior , which will be essential to account for DNA recognition by proteins .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[] |
2015
|
Simulations Meet Experiment to Reveal New Insights into DNA Intrinsic Mechanics
|
For the past four decades the compositional organization of the mammalian genome posed a formidable challenge to molecular evolutionists attempting to explain it from an evolutionary perspective . Unfortunately , most of the explanations adhered to the “isochore theory , ” which has long been rebutted . Recently , an alternative compositional domain model was proposed depicting the human and cow genomes as composed mostly of short compositionally homogeneous and nonhomogeneous domains and a few long ones . We test the validity of this model through a rigorous sequence-based analysis of eleven completely sequenced mammalian and avian genomes . Seven attributes of compositional domains are used in the analyses: ( 1 ) the number of compositional domains , ( 2 ) compositional domain-length distribution , ( 3 ) density of compositional domains , ( 4 ) genome coverage by the different domain types , ( 5 ) degree of fit to a power-law distribution , ( 6 ) compositional domain GC content , and ( 7 ) the joint distribution of GC content and length of the different domain types . We discuss the evolution of these attributes in light of two competing phylogenetic hypotheses that differ from each other in the validity of clade Euarchontoglires . If valid , the murid genome compositional organization would be a derived state and exhibit a high similarity to that of other mammals . If invalid , the murid genome compositional organization would be closer to an ancestral state . We demonstrate that the compositional organization of the murid genome differs from those of primates and laurasiatherians , a phenomenon previously termed the “murid shift , ” and in many ways resembles the genome of opossum . We find no support to the “isochore theory . ” Instead , our findings depict the mammalian genome as a tapestry of mostly short homogeneous and nonhomogeneous domains and few long ones thus providing strong evidence in favor of the compositional domain model and seem to invalidate clade Euarchontoglires .
Human and cow genomes have been shown to possess a complex architecture , in which compositionally homogeneous and nonhomogeneous domains of varying lengths and nucleotide composition are interspersed with one another [1] , [2] . These empirically derived compositional architectures are mostly incompatible with the “isochore theory” [3]–[6] , according to which the genomes of warm-blooded vertebrates are depicted as mosaics of fairly long isochores —typically 300 kb or more—each possessing a characteristic GC content that differs significantly from that of its neighbors , and each classifiable by GC content into six or less isochore families [7]–[14] . Numerous methods for segmenting DNA sequences into contiguous compositionally-coherent domains have been proposed in the literature . These methods differ from one another in the number and types of parameters used in the segmentation process , as well as in the levels of user intervention . Unfortunately , even methods that limit user input to a few parameters yield incongruent results with one another [15] , whereas methods that rely on subjective user intervention [e . g . , 16] preclude independent replication of the results and are , thus , unscientific . Through comparison of performances against benchmark simulations , Elhaik , Graur , and Josić [2] identified a segmentation method , DJS [17] , that outperformed all others . However , DJS failed to partition sequences with low compositional dispersion and had difficulties in identifying short homogeneous domains . To rectify these inadequacies , Elhaik et al . [15] devised IsoPlotter—a recursive segmentation algorithm that employs a dynamic threshold , which takes into account the composition and length of each segment . Most importantly , IsoPlotter is an unsupervised algorithm , i . e . , it requires no subjective user intervention , and through benchmark validation , it was shown to yield unbiased results [15] . The compositional domains identified by IsoPlotter are contiguous genomic segments , each with a characteristic GC content that differs significantly from the GC contents of its adjacent upstream and downstream compositional domains . By comparing the GC content variance of compositional domains with that of the chromosomes on which they reside , compositional domains can be further classified into two types: “compositionally homogeneous domains , ” or simply “homogeneous domains , ” and “compositionally nonhomogeneous domains” or “nonhomogeneous domains . ” A subset of long homogeneous domains , where “long” is arbitrarily defined as ≥300 kb , are termed “isochoric” domains ( sensu [12] ) . By segmenting the human genome with IsoPlotter , we found that one-third of the genome is composed of compositionally nonhomogeneous domains and the remaining is a mixture of many short compositionally homogeneous domains and relatively few long ones [15] . “Isochoric” domains cover less than a third of the human genome . Similar results were obtained for the cow genome [1] . Here , we characterize the compositional architecture of ten completely sequenced mammalian genomes and an avian outgroup , and attempt to identify quantitative trends in the evolution of homogeneous and nonhomogeneous domains . Seven attributes of compositional domains are used , many of which were previously used to characterize compositional architectures [1] , [18]–[28] . Each genome is defined by: ( 1 ) the number of compositional domains , ( 2 ) compositional domain-length distribution , ( 3 ) density of compositional domains , ( 4 ) genome coverage by the different domain types , ( 5 ) degree of fit to a power-law distribution , ( 6 ) compositional domain GC content , and ( 7 ) the joint distribution of GC content and length of the different domain types . Our results are interpreted in light of two currently competing phylogenetic hypotheses depicting the evolution of eutherian mammals for which traditional phylogenetic tools provided ambiguous answers [e . g . ] , [ 29 , 30] ( Figure 1 ) . Further , our results support the so-called “murid shift” hypothesis , and suggest that homogeneous and nonhomogeneous domains are biologically different . This evolutionary study represents a dramatic departure from earlier studies that either extrapolated from a few genes to the entire genome [e . g . ] , [ 10] , [ 31 , 32] , used unreliable proxies to infer the composition of domains [e . g . ] , [ 31 , 33] , or used irreproducible methodologies [e . g . ] , [ 16 , 34] . Our results will be compared with claims made by proponents of the “isochore theory . ” Sadly , we are forced yet again to confront the “isochore theory , ” because despite its being refuted numerous times [e . g . ] , [ 18] , [ 35] , [ 36 , 37] , proponents of the theory and those invested in it continue to pursue the notion of isochores aggressively , relentlessly , and vociferously [e . g . ] , [ 31] , [ 38] , [ 39]–[46] . It seems that T . H . Huxley's dictum on “the great tragedy of science” being “the slaying of a beautiful theory by an ugly fact” does not easily apply to the concept of “isochores . ”
Genome statistics for compositional , homogeneous , nonhomogeneous , and “isochoric” domains are shown in Table 1 . In Table S1 we present the same data partitioned by individual chromosomes . The mean number of compositional domains in a mammalian genome in our sample is approximately 96 , 000 , with opossum having the largest number of domains ( 107 , 000 ) , and rat having the smallest ( ∼63 , 000 ) . On average , over two thirds of all mammalian domains are homogeneous , but this proportion varies with taxon ( Table 1 ) . Opossum has the smallest fraction of homogeneous domains ( 59% ) followed by murids ( 62% ) . By contrast , pig ( 71% ) and horse ( 74% ) genomes are the most enriched for homogeneous domains . Isochoric domains constitute only a tiny fraction of the compositional domains , from 0 . 7% in horse and dog to 2 . 1% in rat . The mean compositional-domain length varies from ∼25 , 700 bp in primates to ∼38 , 500 bp in murids ( Table 1 ) . The median length is much smaller in all taxa , indicating an extreme skewed distribution towards very short domains . For example , half of the compositional domains in rat are shorter than 9 , 216 bp . The mean and median lengths of homogeneous and nonhomogeneous domains within a taxon are practically indistinguishable . The largest homogeneous domain among all species is one 10 . 5-megabase ( Mb ) long ( GC content of 36% ) found in the opossum genome . In the human genome , the largest homogenous domain is about half that length ( 5 . 2 Mb ) . Almost all the distributions of homogeneous domain lengths in all studied species ( Figure 2 ) are significantly different from each other ( Kolmogorov-Smirnov goodness-of-fit test , p<0 . 01 ) , however , this is due to the large sample sizes . The magnitude of the differences between homogeneous and nonhomogeneous domain lengths is very small in all species ( area overlap>98% , Cohen's d<0 . 05 ) with the chicken genome exhibiting borderline similarity ( area overlap 97% , Cohen's d<0 . 05 ) . A comparison of the cumulative distributions of domain lengths indicates that the top percentile in murids consists of domains larger than 511 kb , whereas the top percentile in the laurasiatherian genomes consists of domains larger than 281 kb ( Figure 3 ) . In mammalian genomes , the proportion of long homogeneous domains ( ≥300 kb ) , i . e . , “isochoric” domains , out of all domains is 1% and twice that in murids ( 2 . 02% ) . Similar cumulative distributions were observed for compositional and nonhomogeneous domains ( Figure S1 ) . Domain density measures the average number of compositional domains per Mb . When divided into GC-poor and GC-rich compositional domains it ranges from 0 to 90 domains/Mb for GC-poor domains and up to 121 domains/Mb for GC-rich domains ( Figure 4 ) . Homogeneous domains are more dense for both GC-poor ( 0–57 domains/Mb ) and GC-rich ( 0–98 domains/Mb ) domains compared to nonhomogeneous GC-poor ( 0–43 domains/Mb ) and GC-rich ( 0–64 domains/Mb ) domains , respectively . In regions of low domain densities , the density of GC-rich domains is higher than the density of GC-poor domains . That is , genomic regions with fewer domains are more likely to be GC-rich , whereas denser genomic regions are more likely to harbor GC-poor domains ( Figure S2a ) . On average , murid chromosomes are the least dense ( 26 domains/Mb ) , whereas the horse genome is the most dense ( 49 domains/Mb ) . The chromosomal domain densities of opossum are as low as murids for homogeneous domains ( 21 and 16 domains/Mb , respectively ) and as high as primates for nonhomogeneous domains ( 13 domains/Mb ) . The overall chromosomal domain densities position opossum ( 34 domains/Mb ) between murids ( 26 domains/Mb ) and other mammals ( 43 domains/Mb ) . Similar patterns were observed when comparing the compositional domain densities of GC- rich and GC-poor domains ( Figure S3 ) ; the opossum and primate genomes have the highest density for GC-rich domains ( 21 and 18 domains/Mb , respectively ) . By contrast , the opossum's genome low density for GC-poor domains ( 10 domains/Mb ) is lower even than that of murids ( 16 domains/Mb ) . The overall domain density in opossum ( 31 domains/Mb ) is between that of murids ( 25 . 5 domains/Mb ) and primates ( ∼38 . 5 domains/Mb ) . Domain density largely varies among chromosomes and chromosome types . Density differences between chromosomes can reach 100% ( Figures 4 , S2 ) with sex chromosomes having a lower density than the average autosome ( Table S1 ) . These results indicate that the processes that shaped the inter-chromosomal domain organization acted non-uniformly on all chromosomes and their effect on domain lengths was highly variable in different lineages implying the existence of a compositional constraint on chromosomal heterogeneity . In Figure 5 , we show the relative genomic coverage of compositional domains as a function of domain homogeneity and length . The genomic coverage by homogeneous domains ranges from ∼79% in primates and murids to ∼85% in horse . By defining “isochoric” domains as compositionally homogeneous domains longer than 300 kb , we find that the genomic coverage by “isochores” in mammals is a trifling 27% , compared to 16% in the chicken . Murids and opossum have the largest genomic coverage by “isochoric” domains ( 34% and 37% , respectively ) . Relaxing the “isochore” definition to include homogeneous domains larger than 100 kb , as proposed by Nekrutenko and Li [47] , slightly increases the “isochoric” portion of the genome to 38% . These results , in themselves , are sufficient to invalidate the “isochore theory” or at least diminish its applicability . The distribution of domain lengths in the human genome is commonly depicted as a power-law distribution over a large range of length scales [e . g . ] , [ 18] , [ 48 , 49] . A distribution is said to follow a power-law if its histogram is a straight line when plotted on a log-log scale [50] , [51] . To gauge the power-law model , we used two approaches: first , we compared the cumulative distributions of homogeneous domain lengths to the maximum likelihood power-law fits . In all cases , the complementary cumulative distribution function P ( x ) and their maximum likelihood power-law fits deviate from a straight line , and the p-value is sufficiently small ( Kolmogorov-Smirnov , p<0 . 01 ) that the power-law model can be ruled out ( Figure 6 ) . In other words , there is a very small probability that the data can be modeled by a power-law . An even weaker fit was obtained using compositional domains and nonhomogeneous domains ( Figure S4 ) . Next , we tested the power-law behavior of domain lengths using the random group formation model . We found that the same deviations from a power-law-like behavior were also predicted by the random group formation model [52] ( Figure S5 ) . The deviations of the data from power-law behavior are caused by the excess of short domains and low frequency of long domains . These findings are at odds with earlier contentions that the mammalian genome is a mosaic of long homogeneous domains with very few short domains [e . g . ] , [ 12] , [ 49 , 53] . However , we note that earlier results are not based on the length distribution of actual domains as some authors chose to avoid the excess of short domains – that cause the deviation from power-law – by concatenating them to form artificially long domains [e . g . ] , [ 54 , 55] . We believe that the decision as to whether or not neighboring domains should be concatenated should rely solely on their homogeneity rather than on attempts to make the data fit a preconceived model . Moreover , if domain lengths are truly drawn from a power-law distribution , the power-law model should fit the data over more than three orders of magnitude [50] . In reality , the power-law fit is quite poor and should thus be rejected ( Figures 6 , S4 , S5 ) . Our findings are in agreement with previous studies that rejected the power-law behavior of compositional domains , although they relied on a small dataset and incomplete genomic sequences [56]–[61] . We reported similar findings in three ant genomes [19]–[21] . The GC contents of the homogeneous and nonhomogeneous domains in eutherians exhibit a non-normal distribution ( Lilliefors goodness-of-fit test , p<0 . 05 ) with a mean of 42–44% and a standard deviation of 5 . 7–8 . 5% . The GC distributions of compositional domains of the same type are significantly different from one another , particularly between related taxa ( Kolmogorov-Smirnov goodness-of-fit test , p<0 . 01 ) ; however , this is due to the large sample sizes . Similar to the patterns observed in compositional domain lengths , the small differences in the GC contents of homogeneous and nonhomogeneous domains allow grouping the species into five taxonomic groups: Primates , Laurasiatheria , Muridae , opossum , and chicken ( Figure 7 ) . Of these groups , only the Primate and Laurasiatheria exhibit a high degree of similarity in compositional domain length distribution . Murids and opossum have the most variable GC distribution ( 38% area nonoverlap ) ( Figure 7 ) . With the exception of the murid genomes ( γ≈0 . 29 ) , the low frequency of short GC-poor domains and the abundance of medium GC-rich domains causes mammalian GC distributions to be highly right-skewed ( 0 . 56<γ<0 . 77 ) ( Figure 7 , Table S2 ) . Opossum ( γ≈1 . 12 ) and chicken ( γ≈0 . 86 ) are the most right-skewed of all species , due to the high abundance of short GC-rich and medium-short GC-rich domains , respectively . To further study the GC content fluctuations within compositional domains , we looked at their compositional variability . Compositional variability is measured from the standard deviation ( GCσ ) of the GC content of each domain calculated over short nonoverlapping windows within the domain ( see Materials and Methods ) . Figure 8 presents two-dimensional joint distribution of homogeneous domain GC content and GCσ . Interestingly , the GCσ values of most mammalian domains are narrowly distributed around 11% GCσ and , with the exception of opossum that exhibits a smaller variation . In other words , GC-rich domains are more erratic in their composition ( high GCσ ) than GC-poor domains ( low GCσ ) . The high compositional variability of horse and dog is also reflected in the wide range of GCσ values compared with those of the Cetartiodactyla species . The opossum is exceptional in exhibiting a GCσ gradient toward smaller GCσ . The opossum compositional makeup characterized by its low GC content and narrow GCσ distribution appears to be an intermediate between mammals and murids . The narrow GCσ distribution in the murid genomes is also confounding . The murid joint distributions are largely symmetric about the x-axis ( Figure 8 ) , suggesting that the evolutionary processes that shaped the compositional organization of the genome were symmetrical . Similar trends were obtained for the nonhomogeneous domains ( Figure S6 ) . The two-dimensional joint distributions of homogeneous domain GC content and length are shown in Figure 9 . These measures are not correlated ( r = ∼0 ) . As shown before , the majority of domains in all genomes are short ( 6–8 kb ) , and their GC content distributes close to the mammalian genome mean GC content . With the exception of murids , homogeneous domains are significantly more AT-rich than nonhomogeneous domains ( Kolmogorov-Smirnov goodness-of-fit test , p<0 . 01 ) . The genomic landscape topologies of primates , laurasiatherian , and murids are remarkably similar with short ( 103–104 bp ) GC-rich domains 1 . 3–1 . 7 times more frequent than GC-poor domains and medium-large ( 105–106 bp ) GC-rich domains 1–2 times more frequent than GC-poor domains ( Table S2 ) . This ratio is opposite for both domain size groups ( 0 . 7 and 0 . 32 , respectively ) in opossum , which implies a major domain fusion process that affected the tetrapod genome . Domains in the murid genome have a distinct length distribution compared to other mammals . The murid genome has an abundance of over 2 , 500 medium-long ( 105–106 bp ) and long ( >106 bp ) GC-rich domains compared to all other genomes ( ∼500–1 , 591 ) ( Table S2 ) . By comparison , in the AT-rich opossum genome , GC-poor domains are twice more frequent than GC-rich domains . The opossum genome is particularly enriched in over 3 , 500 medium-long and long GC-poor domains compared with only 486 GC-rich domains . Similar results were observed for nonhomogeneous domains ( Figure S7 ) . Table 2 summarizes the supporting evidence for the two phylogenetic hypotheses contrasting the validity of Euarchontoglires clade based on the defined genetic attributes . Although the attributes are not independent , qualitatively they provide a strong support for the second hypothesis that places Primates with Laurasiatheria to the exclusion of Muridae , thereby invalidating clade Euarchontoglires ( Figure 1 ) .
One of the most fascinating features of mammalian genomes is the uniformity of GC content over hundreds and hundreds of thousands base-pairs termed short- and long-range correlations , respectively . Although these structures have been known for over three decades [3] , only few explanations were proposed in an evolutionary framework . Most of the explanations for the long-range correlations were related to the “isochore theory . ” The “isochore theory” posits the mammalian genome is composed of a mosaic of isochores , long homogeneous domains ( typically ≥300 kb ) that cover the majority of the genome of “warm-blooded” vertebrates; whereas only a small portion of the genome consists of non-“isochoric” regions . The “cold-blooded” vertebrate genome was described as less compositionally heterogeneous and devoid of GC-rich isochores [5] , [12] . Although the theory failed to explain the compositional patterns later found in fish and reptiles [e . g . ] , [ 43 , 62] , its importance has been in stimulating follow-up studies that attempted to correlate various biological phenomena with compositional and organizational features . Eventually , following conflicting findings [e . g . ] , [ 15] , [ 36] , [ 37] , [ 62 , 63] , ambiguity as to the interpretation of the theory predictions [18] and contradictory revisions of the theory's main principles [e . g . , 55] ( Table S3 ) , the original theory was de facto abandoned by most scientists ( with the exception of its proponents ) , leaving open the basic questions: how , when , and why in the course of evolution , did mammalian genomes acquire their current composition and organization ? The most effective approach to understanding the compositional organization of human and mammalian genomes is by comparative analysis – preferably a large-scale one . In a previous analysis of the human genome , Elhaik et al . [15] proposed a compositional domain model to explain its genomic architecture . The compositional domain model portrays the human genome as a mixture of mostly short and very few long homogeneous and nonhomogeneous domains in a ratio of 2∶1 . Under this model , “isochoric” domains consist of only a small fraction of all compositional domains [15] . Here , we extended the analysis to ten mammalian genomes and tested whether the outcomes fit within the isochoric or the compositional-domain models using seven genomic attributes . Our findings are discussed under two different phylogenetic hypotheses , for which traditional phylogenetic analyses provided ambiguous answers ( Figure 1 ) . Table 2 summarizes the evidence in support of either hypothesis . The mammalian genome is covered by a complex medley of nonhomogeneous domains of various lengths ( 32% ) , short ( 103–104 bp ) homogeneous domains ( 36% ) , medium-short ( 104–105 bp ) homogeneous domains ( 26% ) , medium-long ( 105–106 bp ) homogeneous domains ( 4% ) , and only a miniscule fraction of 0 . 16% long ( 106–107 bp ) homogeneous domains ( Table S2 ) . On average , homogeneous domains longer than 300 kb , i . e . , isochores , constitute less than 2% of all domains and cover less than 28% of the mammalian genome ( Table 1 ) . Short homogeneous domains have wide GC content distributions and the GC content of long homogeneous domains is distributed slightly below the mammalian genome mean GC content ( Figure 9 ) m whereas the GC content of long nonhomogeneous domains is distributed slightly above it . Under the “isochore model” where the vast portion of the genome was considered to be composed of long homogeneous domains , their length distribution was thought to display a power-law distribution [18] , [49] , [53] , [64] . We demonstrated that the power-law model is inconsistent with the data due to the high abundance of short domains and the scarcity of long domains ( Figures 6 , S4 , S5 ) . Short domains are major components of the mammalian genome and cannot be dismissed as “false positives“ [55] . Overall , our results support the compositional domain model and limit the applicability of the isochore model to less than 30% of the average mammalian genome . Homogeneous or “relatively homogeneous” [9] domains were speculated to be biologically different from nonhomogeneous domains [7] , [18] , [55] , yet we found only minor differences between and within chromosomes , most of which stemmed from the differences in the proportions of the two domain types ( Tables 1 , S2 ) . Interestingly , with the exception of murid genomes , we found that homogeneous domains are significantly more AT-rich than nonhomogeneous domains ( Figures 9 , S7 ) , which may suggest biological importance . To support such hypothesis , additional biological properties should be used to test whether or not this distinction is biologically meaningful . Most genome characteristics within higher taxa follow phylogenetic relatedness . For example , the genomes of the three primates are very similar to each other , as are the genomes of the two murids . The genome characteristics of the Pegasoferae ( horse and dog ) differ slightly from those of cetartiodactyls ( cow and pig ) , possibly adding support for the validity of clade Pegasoferae ( Figure 1 ) . However , the possibility that the similarity between horse and dog is due to the poor quality of their genomic sequences cannot be excluded . We have evidence obtained by comparing a draft of the cow genome ( build 3 . 1 ) with the finished version ( build 4 . 0 ) [1] that draft genomes contain an abundance ( ∼90% ) of short compositional domains ( <10 kb ) , thus rendering drafts genomes artificially similar to one another . Overall , the laurasiatherian genomes are more similar to the primate genomes than the murid genomes , which , in turn , are more similar to the opossum genome than to any other genome ( Table 2 ) . The murid genome is distinguished from the primate and laurasiatherian genomes mainly by its narrow GC content distribution ( Figure 7 ) , larger GC-rich domains ( Figures 2 , 3 ) , smaller GC content standard deviation for both GC-poor and -rich domains ( Figure 8 ) , and the unique shape of its joint distribution of compositional domain GC content and length ( Figure 9 ) . Differences in the compositional patterns between murids and other mammals were previously termed the “murid pattern” [65] or “murid shift” [66] . The “shift” was attributed to a smaller variation in the composition of isochoric domains compared to other mammals [66]; however , we found that the differences between the murid lineage to other mammals are found in the entire murid genomes and are not unique to “isochoric” domains . A possible explanation to the “shift” may be in the different evolutionary origin of murids ( Figure 1b ) . Moreover , the similarity between the murid and opossum genomes suggests the effect was not unique to murids and may have originated in the eutherian ancestor . The two phylogenetic hypotheses tested here differ in the validity of clade Euarchontoglires . According to the first hypothesis ( Figure 1a ) , murids arose relatively late in mammalian evolution and are grouped with Primates under Euarchontoglires . Considering the relatively fast mutation rate of the murids [67] , the most parsimonious explanation would be that their genomic organization is a derived state , possibly as a result of a “shift” or a genomic transition that affected the entire linage . Under this hypothesis , the genomic transition resulted in the fusion of nearly half of the short domains of extreme GC content together with other domains . Elongated domains were created due to the decrease in GC content variability and the fusion of neighboring domains . Subsequently , domain density was reduced and the compositional fluctuations were “flattened” resulting in higher homogeneity between domains . The process dramatically decreased the proportion of short domains ( 52% ) that are highly frequent in other mammalian genomes ( 60% ) . Conversely , these fusions increased the proportion of longer domains ( medium-short = 40% , medium-long = 7 . 5% , long = 0 . 28% ) compared to all other mammalian domains ( medium-short = 36% , medium-long = 4% , long = 0 . 15% ) . The proportion of long GC-poor domains increased as well but in smaller proportion than GC-rich domains . Further evidence for this transition can be found in the frequency distribution of GC content standard deviation that is relatively devoid of heterogeneous domains compared to other mammalian genomes ( Figure 8 ) . Moreover , Muridae have genomes that are markedly homogeneous in both poor- and GC-rich domains , as opposed to mammalians genomes that are highly heterogeneous in their GC-rich domains and homogeneous in their GC-poor domains ( Table S2 ) . We note that genome elongation could also result from segmental duplication; however , we do not know of a segmental duplication that acts selectively on segments with certain composition . According to the second hypothesis ( Figure 1b ) , murids arose early in the mammalian evolution and their genomic architecture reflects an ancestral state . The “typical” mammalian genome thus evolved from this ancestral pattern leading to a wider compositional distribution and shorter domains . This view is supported by the similar genomic structure ( Tables 1 , S2 ) and genome homogeneity shared between the murid and opossum genomes . A similar hypothesis was tested by Mouchiroud , Gautier , and Bernardi [68]; however , because they assumed the existence of isochores that cover the mammalian genome , their conclusions are limited to few “isochoric” domains . Unfortunately , the representation of marsupial mammal as outgroup yielded more questions than answers as opossum reflected either unique genomic characteristics or oscillated between murid and non-murid characteristics ( Tables 1–2 ) . Thus , although the results showed a high resemblance between murids and opossum in support of the second hypothesis ( Table 2 ) , additional evidence would be necessary before ruling out the first hypothesis ( Figure 1 ) . It is possible that with the accumulation of additional genomic sequences of intermediate species this question would be answered . In light of these findings , it will be intriguing to identify which evolutionary mechanisms shaped the transitions that affected the murid and opossum genomes . Understanding these biological mechanisms and their evolutionary implications is a key factor in reconstructing the evolutionary history of mammalian genome evolution .
Nine eutherian genomes that are either fully sequenced or have reliable genomic drafts were included in this study: human ( Homo sapiens build 37 . 1 ) , chimpanzee ( Pan troglodytes build 2 . 1 ) , orangutan ( Pongo abelii build 1 . 2 ) , mouse ( Mus musculus build 37 . 1 ) , rat ( Rattus norvegicus build 4 . 1 ) , horse ( Equus caballus build 2 . 1 ) , dog ( Canis familiaris build 2 . 1 ) , pig ( Sus scrofa build 2 . 1 ) , and cow ( Bos taurus build 4 . 1 ) . The gray short-tailed opossum ( Monodelphis domestica build 2 . 1 ) was used as an outgroup to the eutherians , and chicken ( Gallus gallus build 2 . 1 ) was used as an outgroup to the mammals . Genomes were downloaded from http://www . ncbi . nlm . nih . gov/Genomes/ . Nulls , i . e . , unknown , undetermined , or ambiguous characters in the genomic sequences , were discarded . There are two phylogenetic hypotheses in the literature for the taxa under study ( Figure 1 ) . The two hypotheses are supported by molecular data though differ in their outcome . The difference between the two phylogenetic trees concerns the relative kinship of murids ( mouse and rat ) and laurasiatherians ( horse , dog , cow , and pig ) to primates ( human , chimpanzee , and orangutan ) . In the first scheme [e . g . ] , [ 29] , [ 69] , [ 70]–[72] , primates cluster with the murids within clade Euarchontoglires ( Figure 1a ) . In the second scheme [e . g . ] , [ 30 , 73] , primates cluster with the laurasiatherians to the exclusion of murids ( Figure 1b ) . The clustering of Perissodactyla ( horse ) and Carnivora ( dog ) into Pegasoferae to the exclusion of Cetartiodactyla ( cow and pig ) is accepted by both alternative phylogenies [69] . Version 2 of IsoPlotter [15] of the IsoPlotter+ pipeline [28] was obtained from https://github . com/sean-dougherty/isoplotter/and used to partition each of the genomes into compositionally distinct domains . IsoPlotter recursively maximizes the difference in GC content between adjacent segments , as measured by the Jensen-Shannon divergence statistic [17] . The halting criterion was obtained via a dynamic threshold calculated in real-time according to the length of each segment and the standard deviation of its GC content . The compositional domains inferred by the segmentation procedure were classified into homogeneous and nonhomogeneous as in Elhaik et al . [15] . For convenience , domains are sometimes divided by order of magnitude of their length into: short ( 103–104 bp ) , medium-short ( 104–105 bp ) , medium-long ( 105–106 bp ) , and long ( 106–107 bp ) domains . The mean GC content of all mammalian genomes in this study ( 40 . 9% ) was used as a critical value . A compositional domain was defined as GC-rich or GC-poor if its GC content was higher or lower , respectively , than the critical value . For each species and for each domain category , log domain-lengths were sorted and smoothed . Smoothing was carried by dividing the log domain-lengths into 1 , 000 groups of equal size and then using the mean domain length of each group to calculate a histogram with 38 bins ranging from 8 to 16 . To test whether or not two distributions are different , we used the Kolmogorov-Smirnov goodness-of-fit test and the False Discovery Rate ( FDR ) correction for multiple tests [74] . Because the differences between the distributions were highly significant due to the huge sample sizes , we also calculated the effect size , first by using the nonoverlapping percentage of the two distributions , and then by using Hedges' g estimator of Cohen's d [75] . If the area overlap was larger than 98% and Cohen's d was smaller than 0 . 05 , we considered the magnitude of the difference between the two distributions to be too small to be biologically meaningful . The distributions of domain GC contents were calculated in a similar manner . To smooth the GC content distributions , domain GC contents were divided into 1 , 000 groups of equal size , and the mean domain GC content of each group was used to calculate a histogram with 38 bins ranging from 0 to 1 . The remaining calculations were carried as described above . To test whether the GC-content distributions of homogeneous and nonhomogeneous domains fit a normal distribution , we used the Lilliefors ( 1967 ) test . This test is a two-sided goodness-of-fit test suitable when a fully-specified null distribution is unknown and its parameters must be estimated . It tests the null hypothesis that domain GC contents come from a distribution in the normal family , against the alternative that they do not come from a normal distribution . We also estimated the standardized skewness ( γ ) of the GC content distributions using the “skewness” function in Matlab , which first centralizes the distribution by subtracting it from its mean , calculates its third ( k3 ) and second ( k2 ) moments , and then computes the skewness , so that: GC0 = GC – μ ( GC ) , k3 = μ ( GC03 ) , k2 = μ ( GC02 ) , and γ = k3/k21 . 5 . We used two approaches to test the fit of the domain-length distributions to power-laws . First , the minimum domain length and the power-law exponent were estimated for the domain lengths of each genome according to the goodness-of-fit based method described in Clauset , Shalizi , and Newman [51] . The observed domain lengths were then compared to the domain lengths generated from the parameters previously estimated , and the similarity between the two distributions was calculated using the Kolmogorov-Smirnov statistic [76] . Based on the observed goodness-of-fit , we calculated a p-value that quantifies the probability that the data were drawn from the hypothesized distribution . We used the Matlab scripts plfit . m ( version 1 . 0 . 5 ) , plpva . m ( version 1 . 0 . 6 ) , and plplot . m ( version 1 . 0 ) in www . santafe . edu/~aaronc/powerlaws/ ( Clauset , Shalizi , and Newman [51] . Second , Baek and et al . [52] showed that the random group formation ( RGF ) model is a form of general distribution , free from system-specific assumptions , of which pure power-laws are a special case . We used this model to test the data fitting into the power-law model using the online application http://www . tp . umu . se/~garuda/Comp . html .
|
The non-uniformity of DNA composition in mammalian genomes has been known for over four decades . Many attempts have been made to provide a concise description of this heterogeneity and to identify the evolutionary driving forces behind this compositional phenomenology . The first concise description of the genome suggested an isochoric structure according to which the mammalian genome consists of a mosaic of long , compositionally homogenous DNA sequences . With the advent of genome sequencing , this description was found to be inappropriate . We have recently proposed an alternative “compositional domains” model that depicts the human and cow genomes as composed of mixture of compositionally homogeneous and nonhomogeneous domains . Most of these domains are very short . Since its proposal , this model has been validated in plethora of invertebrate genomes . Here , we test the validity of this model on eleven mammalian and avian genomes using seven attributes of compositional domains and discuss their evolution . We also use these attributes to decide between two competing phylogenetic hypotheses . Our findings provide strong supporting evidence for the “compositional domains” model and indicate that rodents are not as close to primates as envisioned by the Euarchontoglires hypothesis .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"genome",
"evolution",
"evolutionary",
"modeling",
"biology",
"and",
"life",
"sciences",
"molecular",
"evolution",
"computational",
"biology",
"evolutionary",
"biology",
"genomics",
"statistics",
"evolutionary",
"theory"
] |
2014
|
A Comparative Study and a Phylogenetic Exploration of the Compositional Architectures of Mammalian Nuclear Genomes
|
The Schistosomiasis Consortium for Operational Research and Evaluation ( SCORE ) has launched several large-scale trials to determine the best strategies for gaining and sustaining control of schistosomiasis and transitioning toward elimination . In Côte d’Ivoire , a 5-year cluster-randomized trial is being implemented in 75 schools to sustain the control of schistosomiasis mansoni . We report Schistosoma mansoni infection levels in children one year after the initial school-based treatment ( SBT ) with praziquantel and compare with baseline results to determine the effect of the intervention . The baseline cross-sectional survey was conducted in late 2011/early 2012 and the first follow-up in May 2013 . Three consecutive stool samples were collected from 9- to 12-year-old children in 75 schools at baseline and 50 schools at follow-up . Stool samples were subjected to duplicate Kato-Katz thick smears . Directly observed treatment ( DOT ) coverage of the SBT was assessed and the prevalence and intensity of S . mansoni infection compared between baseline and follow-up . The S . mansoni prevalence in the 75 schools surveyed at baseline was 22 . 1% ( 95% confidence interval ( CI ) : 19 . 5–24 . 4% ) . The DOT coverage was 84 . 2% . In the 50 schools surveyed at baseline and one year after treatment , the overall prevalence of S . mansoni infection decreased significantly from 19 . 7% ( 95% CI: 18 . 5–20 . 8% ) to 12 . 8% ( 95% CI: 11 . 9–13 . 8% ) , while the arithmetic mean S . mansoni eggs per gram of stool ( EPG ) among infected children slightly increased from 92 . 2 EPG ( 95% CI: 79 . 2–105 . 3 EPG ) to 109 . 3 EPG ( 95% CI: 82 . 7–135 . 9 EPG ) . In two of the 50 schools , the prevalence increased significantly , despite a DOT coverage of >75% . One year after the initial SBT , the S . mansoni prevalence had decreased . Despite this positive trend , an increase was observed in some schools . Moreover , the infection intensity among S . mansoni-infected children was slightly higher at the 1-year follow-up compared to the baseline situation . Our results emphasize the heterogeneity of transmission dynamics and provide a benchmark for the future yearly follow-up surveys of this multi-year SCORE intervention study .
Schistosomiasis is a neglected tropical disease that exerts a considerable public health problem in 78 tropical and subtropical countries [1] . In 2013 , it was estimated that schistosomiasis affected more than 250 million people worldwide with 90% of the reported cases concentrated in sub-Saharan Africa [2] . Since the mid-1980s , the World Health Organization ( WHO ) emphasizes morbidity control using the drug praziquantel as the main pillar of the global strategy to fight schistosomiasis [3] . Praziquantel is the drug of choice because it is efficacious against the adult stages of all Schistosoma species parasitizing humans , is inexpensive ( the average cost to treat a school-aged child was US$ 0 . 2 per treatment in 2013 ) , and has a good safety profile [4–8] . For morbidity control , praziquantel is being administered to at-risk populations without prior diagnosis , a strategy commonly known as ‘preventive chemotherapy’ [9] . The recommended frequency of drug administration is based on the level of endemicity in a given study area . According to WHO , in areas with high schistosomiasis endemicity ( prevalence ≥50% ) , all school-aged children and adult people at risk of infection should be treated annually [10] . In areas with moderate endemicity ( prevalence 10–50% ) , all school-aged children should be treated once every two years . In low endemic areas ( prevalence <10% ) , school-aged children should be treated twice during their time in school; first at school entry and then again in their last year of schooling [11 , 12] . However , these prevalence thresholds are arbitrary . Hence , the Schistosomiasis Consortium for Operational Research and Evaluation ( SCORE ) launched a series of studies to strengthen the evidence-base how best to gain and sustain the control of schistosomiasis , including cost considerations [13] . Two 5-year cluster-randomized trials are being implemented in Côte d’Ivoire and Kenya [14 , 15] . These trials are school-based with three treatment arms ( 25 schools per arm ) and aim to assess whether annual school-based treatment ( SBT ) with praziquantel for four years ( arm A ) , annual SBT in years 1 and 2 , followed by “drug holidays” in years 3 and 4 ( arm B ) , or SBT in years 1 and 3 , spaced by“drug holidays” in years 2 and 4 ( arm C ) will substantially reduce the prevalence and intensity of Schistosoma infection and keep infection at low levels . Here , we present the effect of the first SBT with praziquantel on Schistosoma mansoni infection among school-aged children in western Côte d’Ivoire , as revealed by a detailed follow-up survey conducted in May 2013 , compared to baseline data collected from December 2011 to February 2012 . Specifically , we determined changes in the prevalence and intensity of S . mansoni infections among children in the 50 schools that belong to treatment arms A and B , and discuss consequences for the ongoing cluster-randomized trial and , more generally , for schistosomiasis control interventions in Côte d’Ivoire and elsewhere .
The study protocol was approved by the institutional research commissions of the Swiss Tropical and Public Health Institute ( Basel , Switzerland ) and the ‘Centre Suisse de Recherches Scientifiques en Côte d’Ivoire’ ( CSRS; Abidjan , Côte d’Ivoire ) . Ethical approval was obtained from the ethics committees in Basel ( reference no . EKBB 279/10 ) and the Ministry of Public Health in Côte d’Ivoire ( reference no . 1994 MSHP/CNER ) . At the onset of the study , regional directors of the education and health sectors , education inspectors , village authorities , local community members , and teachers were sensitized in detail about the objectives of the research project . Parents and guardians of study participants provided written informed consent for children to participate . After the baseline parasitologic survey , in the frame of the first SBT conducted in June 2012 , school-aged children living in the catchment area of participating schools were offered treatment with praziquantel at a single oral dose of 40 mg/kg of body weight [16] . The baseline survey was carried out from December 2011 to February 2012 , the SBT in June 2012 , and the first follow-up survey was conducted in May 2013 in eligible schools located in four regions of western Côte d’Ivoire: Cavally , Guemon , Haut-Sassandra , and Tonkpi . Details of the study area and population surveyed have been described elsewhere [15 , 17] . The Cavally and Sassandra rivers and their tributaries represent the major hydrographic features of the study area [18 , 19] . Buyo , a hydroelectric dam built across the Sassandra River in 1981 , formed a lake with an estimated surface area of 600 km² [20] . In western Côte d’Ivoire , the sources of water are traditional wells , rain water , rivers , water supply dams , ponds , creeks , fountains , natural spring water , and tap water [21] . The main reasons for human water contact that might lead to schistosomiasis transmission are washing dishes , washing children , fetching water , fishing , swimming , farming , and playing [22] . Despite the existence of latrines in numerous households , open defecation is commonly practiced [22–24] . The aim of the SCORE sustaining schistosomiasis control study implemented in western Côte d’Ivoire is to determine the best strategy of preventive chemotherapy with praziquantel to sustain schistosomiasis mansoni control in moderate endemicity settings [15 , 17] . For this purpose , the S . mansoni prevalence in n schools in three treatments arms is compared over a study period of four years . The prevalence of S . mansoni is determined by testing m children in those schools where there is subsequent treatment . The effect of the different treatment intervals on the S . mansoni prevalence will be estimated using the following logistic regression model: log ( pijt / ( 1 − pijt ) ) = μ +αi + βt + ɣik , where pijt denotes the prevalence of S . mansoni in school j receiving treatment i in year t , μ is an intercept term , αi is the effect of treatment i , βt is the effect of time t , and γik is the time by treatment interaction . Generalized estimating equations have been used to fit these longitudinal data [25] . To take into account variation in the S . mansoni prevalence among schools , an overdispersion parameter φ was introduced into the model . When φ = 1 , all schools under the same treatment have identical prevalences , whereas φ increases with increasing variation of prevalence levels between villages . The calculations revealed that studying 20 schools per arm and evaluating 100 individuals per school would result in minimum effect sizes of 5–12% with or without overdispersion . In order to increase the chance of detecting differences between the intervention arms , the number of intervention units was increased to 25 per arm . Consequently , a total of 75 schools with a S . mansoni prevalence of 10–24% according to results from an eligibility survey were randomized to one of the three treatment arms [15] . Treatment arm A receives SBT with praziquantel once every year for four years , arm B receives SBT in years 1 and 2 , followed by “drug holidays” in years 3 and 4 , and arm C receives SBT in years 1 and 3 , alternated by “drug holidays” in years 2 and 4 [15] . Before administration of the first round of treatment , a detailed baseline survey was conducted . Following the SCORE harmonization protocol , all 75 schools were included in the baseline parasitologic survey implemented in Côte d’Ivoire from December 2011 to February 2012 . Children were treated with praziquantel in June 2012 . Only the 50 schools belonging to treatment arms A and B were subjected to the first follow-up survey carried out in May 2013 , while the 25 schools belonging to treatment arm C were not subjected to a follow-up survey , as they were on “drug holidays” in year 2 . Baseline and follow-up surveys pursued cross-sectional designs . Study procedures have been detailed elsewhere [15 , 17] . In brief , in each of the selected schools , approximately 100 children were invited to participate in the study . Inclusion criteria were as follows: ( i ) age of children ranging between 9 and 12 years; ( ii ) presence of an informed consent sheet signed by parents/guardians; and ( iii ) children themselves assented orally . Over three consecutive days , children were invited to submit a portion of their own morning stool in a 125-ml plastic container . Every day , filled stool containers were collected by trained field enumerators and sent to the hospital laboratories in the towns of Biankouma , Danané , Douékoué , Guiglo , Kouibly , and Man for processing . Stool specimens were subjected to the Kato-Katz method [26] . In brief , duplicate Kato-Katz thick smears were prepared from a single stool sample , using 41 . 7 mg plastic templates . The thick smears were allowed to clear for at least 60 min and examined by experienced laboratory technicians under a light microscope at low magnification . Eggs from S . mansoni , and additionally from soil-transmitted helminth species , were counted and recorded for each species separately . For quality control , 10% of the slides were randomly selected and re-read by a senior microscopist . In case of discrepancies , the results were discussed with the concerned microscopists and the slides re-read until agreement was reached [27] . In June 2012 , children aged 5–15 years enrolled in the 75 study schools and non-enrolled school-aged children living in the school catchment areas were offered free-of-charge treatment with praziquantel ( 40 mg/kg ) using a dose pole according to WHO guidelines [16] . Praziquantel was administered by trained teachers to children , following a directly observed treatment ( DOT ) approach . Children remained under medical observation and adverse events were recorded within 4 hours post-treatment . Treatment was led by the ‘Programme National de Lutte contre la Schistosomiase , les Géohelminthiases et la Filariose Lymphatique’ ( PNL-SGF ) , and supported by staff from the ‘Programme National de Santé Scolaire et Universitaire’ ( PNSSU ) , the CSRS , and the ‘Université Félix Houphouët-Boigny’ . Praziquantel tablets were supplied by the Schistosomiasis Control Initiative ( SCI; London , United Kingdom ) . The overall number of school-aged children residing in each village was obtained by adding up the number of school-aged but non-school attending children as recorded by the community health workers and the number of children registered in school , as detailed by school teachers . Trained teachers administered praziquantel to children ( those attending school , and the non-enrolled children ) and recorded the number of treated children . Baseline survey data were entered into Microsoft Excel ( 2010 Microsoft Corporation ) , while data from the first follow-up survey were directly entered into smartphones and then uploaded to a database maintained on a central server ( Open Data Kit ) in Atlanta , United States of America . Statistical analyses were performed with STATA version IC13 . 1 ( Stata Corporation; College Station , United States of America ) . The final analysis included children aged 9–12 years who had at least four Kato-Katz thick smear readings at the parasitologic surveys done both at baseline and follow-up . To obtain individuals’ eggs per gram of feces ( EPG ) , we divided the total S . mansoni egg counts from the multiple Kato-Katz slides per child by the total number of Kato-Katz thick smears and multiplied by a factor of 24 . S . mansoni-positive individuals were stratified into three infection intensity categories: ( i ) light ( 1–99 EPG ) , ( ii ) moderate ( 100–399 EPG ) , and ( iii ) heavy ( ≥400 EPG ) [16] . Moreover , we calculated S . mansoni prevalence and arithmetic mean ( AM ) EPG for positive individuals per school and treatment arm . With regard to soil-transmitted helminth infections that were also identified with the Kato-Katz technique , a child was considered positive if at least one egg of Ascaris lumbricoides , hookworm , or Trichuris trichiura was detected in one of the slides . We employed a χ² test to assess a potential association between S . mansoni prevalence and age or sex . Reduction in the prevalence and intensity of S . mansoni infection per school was calculated using the following formulae [28]: prevalence reduction = [ ( prevalence at baseline—prevalence at first follow-up ) / ( prevalence at baseline ) ] X 100 . Reduction in the intensity of infection = [ ( AM EPG at baseline—AM EPG at first follow-up ) / ( AM EPG at baseline ) ] X 100 . The treatment coverage rate was assessed by using the following formula: coverage of the mass drug administration ( MDA ) = [ ( number of school-aged children with DOT recorded by teachers ) / ( overall number of school-aged children registered in school and recorded by health workers ) ] X 100 . Geographic coordinates of each school were recorded using a hand-held global positioning system ( GPS ) receiver ( Garmin Etrex 30; Olathe , United States of America ) . Arc Map 10 . 2 . 1 ( Environmental Systems Research Institute Inc . ; Redlands , United States of America ) was used to generate maps of the changes of S . mansoni prevalence and intensity of infection ( AM EPG ) from baseline to follow-up .
The baseline survey was conducted in the 75 schools meeting eligibility criteria from December 2011 to February 2012 , and 7 , 478 children were invited to participate ( Fig 1 ) . Among them , 168 pupils were excluded from further analyses , because their age was outside the 9–12 years range . Additionally , 299 children were excluded because they did not provide sufficient stool to prepare at least quadruplicate Kato-Katz thick smears . The final study population for analysis of the baseline survey consisted of 7 , 011 children . There were more boys ( n = 4 , 173 ) than girls ( n = 2 , 838 ) . The mean age was 10 . 5 years . The number of children in treatment arms A , B , and C was 2 , 410 ( 34 . 4% ) , 2 , 348 ( 33 . 5% ) , and 2 , 253 ( 32 . 1% ) , respectively . In May 2013 , 4 , 966 children from the 50 schools belonging to intervention arms A and B were invited to participate in the first follow-up survey . According to the SCORE harmonization protocol , children attending schools belonging to study arm C were not surveyed . Among the pupils attending schools included in arms A and B , who were invited to participate , 49 children had an age outside the 9–12 years range , and 250 children did not provide enough stool for at least quadruplicate Kato-Katz thick smears . Hence , results of 4 , 667 children were included for further statistical analyses . There were more boys ( n = 2 , 640 ) than girls ( n = 2 , 027 ) . The children’s mean age was 10 . 3 years . There were 2 , 379 children in treatment arm A and 2 , 288 in treatment arm B . At baseline , before the implementation of the first SBT with praziquantel , the examination of at least four Kato-Katz thick smears per child revealed an overall S . mansoni prevalence of 22 . 1% among the 75 schools surveyed . The prevalence at the unit of the school ranged from 1 . 0% to 54 . 0% . S . mansoni infection was significantly associated with age ( χ² = 25 . 2 , p <0 . 001 ) , higher prevalence was predominantly observed among older children . The prevalence of S . mansoni was significantly higher among boys than girls ( 24 . 3% versus 18 . 7%; χ² = 29 . 9 , p <0 . 001 ) . The overall S . mansoni prevalence in treatment arms A , B , and C was 18 . 8% ( 95% CI: 17 . 2–20 . 3% ) , 20 . 5% ( 95% CI: 18 . 9–22 . 2% ) , and 27 . 2% ( 95% CI: 25 . 3–29 . 0% ) , respectively . With regard to the AM infection intensity , the respective values were 93 . 5 EPG ( 95% CI: 62 . 6–124 . 4 EPG ) , 96 . 2 EPG ( 95% CI: 74 . 5–117 . 9 EPG ) , and 88 . 1 EPG ( 95% CI: 71 . 5–104 . 7 EPG ) ( Table 1 ) . As summarized in Table 1 , at the first follow-up survey , the overall S . mansoni prevalence in arms A and B showed a statistically significant decline from 19 . 7% ( 95% CI: 18 . 5–20 . 8% ) at baseline to 12 . 8% ( 95% CI: 11 . 9–13 . 8% ) at the 1-year follow-up . In arm A , a decrease from 18 . 8% ( 95% CI: 17 . 2–20 . 3% ) to 11 . 2% ( 95% CI: 9 . 9–12 . 4% ) was observed , corresponding to a reduction of 40 . 4% , while in arm B the prevalence declined from 20 . 5% ( 95% CI: 18 . 9–22 . 2% ) to 14 . 5% ( 95% CI: 13 . 1–16 . 0% ) , a reduction of 29 . 3% . Fig 2 indicates the dynamics of the S . mansoni prevalence from baseline to first follow-up survey on a school-by-school basis , stratified by treatment arm . Among the 25 schools belonging to treatment arm A , the S . mansoni prevalence dropped in 23 schools ( S1 Table ) . The most significant decreases occurred in Dio , Pona 2 , Siambly , and Gregbeu , where at the 1-year follow-up , no eggs of S . mansoni were found in the stool of the children examined . However , in Biélé , the S . mansoni prevalence increased significantly from 36 . 0% ( 95% CI: 26 . 4–45 . 6% ) to 79 . 0% ( 95% CI: 70 . 9–87 . 1% ) , while a non-significant increase from 12 . 0% ( 95% CI: 5 . 5–18 . 5% ) to 20 . 7% ( 95% CI: 12 . 0–29 . 4% ) was observed in Séohoun-Guiglo . In treatment arm B , the prevalence of S . mansoni decreased in 20 out of the 25 schools included ( S1 Table ) . In two schools , the prevalence dropped prominently to zero from 24 . 0% in Semien and from 25 . 6% in Diehiba . A significant increase in the S . mansoni prevalence was observed in Ziondrou from 31 . 6% ( 95% CI: 22 . 0–41 . 1% ) to 62 . 0% ( 95% CI: 52 . 3–71 . 7% ) . An increase in prevalence was also observed in Dah , Douandrou 1 , Koulouan , and Guessabo 2 , but without statistical significance . Taken together , as shown in Fig 3A , among the 50 schools surveyed at the first follow-up , a reduction of the S . mansoni prevalence of 25% and above was observed in 39 schools ( 78 . 0% ) . In six schools , the changes ranged from -25% to +25% . An increase of 25% and above was recorded in five schools ( 10 . 0% ) . The increase in prevalence was observed mainly in the central part of Guemon region , eastern Tonkpi region , and western part of Haut-Sassandra region . The overall S . mansoni AM EPG in arms A and B increased from 94 . 9 EPG ( 95% CI: 76 . 2–113 . 6 EPG ) at baseline to 109 . 3 EPG ( 95% CI: 82 . 7–135 . 9 EPG ) at the 1-year follow-up survey . However , this increase was not statistically significant . As shown in Table 2 , in arm A , an increase from 93 . 5 EPG ( 95% CI: 62 . 6–124 . 4 EPG ) to 123 . 7 EPG ( 95% CI: 70 . 7–176 . 7 EPG ) was observed , corresponding to an increase of 32 . 3% , while in arm B the AM EPG at baseline ( 96 . 2 EPG , 95% CI: 74 . 5–117 . 9 EPG ) and the 1-year follow-up ( 97 . 8 EPG , 95% CI: 75 . 5–120 . 0 EPG ) remained basically the same . The proportion of children with heavy infections ( ≥400 EPG ) increased from 4 . 9% to 6 . 3% . Fig 4 displays the changes of the S . mansoni AM EPG in all the schools of treatment arms A and B from baseline to the first follow-up . In arm A , the S . mansoni AM EPG decreased in 16 ( 64 . 0% ) out of the 25 surveyed schools ( S1 Table ) . However , a statistically significant decrease in AM EPG from 33 . 0 EPG ( 95% CI: 13 . 9–52 . 0 EPG ) to 5 . 5 EPG ( 95% CI: 3 . 8–7 . 2 EPG ) was observed in only one school; Tobly Bangolo . Increases in S . mansoni AM EPG were observed in nine schools . However , the increase lacked statistical significance in all schools . In treatment arm A , the proportion of children with moderate ( 100–399 EPG ) and heavy infections ( ≥400 EPG ) increased from 9 . 5% to 13 . 5% and from 4 . 6% to 7 . 2% , respectively . In arm B , a decrease of the S . mansoni infection intensity was observed in 13 ( 52 . 0% ) out of the 25 schools ( S1 Table ) . With the exception of one school , this decrease was not statistically significant . The AM EPG decreased significantly in Mangouin school from 178 . 0 EPG ( 95% CI: 77 . 7–278 . 3 EPG ) to 30 . 4 EPG ( 95% CI: 4 . 2–56 . 6 EPG ) . In the remaining 12 schools , the AM EPG increased , but these increases lacked statistical significance . The proportion of children with moderate and heavy infection intensities increased from 12 . 9% to 18 . 7% , while the proportion of heavy infections decreased slightly from 6 . 0% to 5 . 7% . Fig 3B shows the spatial distribution of S . mansoni AM EPG reduction after the intervention in the study area . The AM EPG decreased by at least 25% in 25 schools ( 50 . 0% ) . In eight schools ( 16 . 0% ) , the change varied from -25% to +25% . The AM EPG increased by 25% and more in 17 schools ( 34 . 0% ) . An increase of S . mansoni infection intensity by 25% and more was only focally observed; in Tonkpi region and central Guemon region . During the SBT carried out in June 2012 , the estimated number of the school-aged population in the study area was 31 , 832 children . Among them , 26 , 804 swallowed praziquantel tablets at the SBT , resulting in an overall DOT coverage of 84 . 2% . Stratified by treatment arm , we found a DOT coverage of 79 . 2% ( range: 31 . 9–97 . 9% ) for arm A , 84 . 8% ( range: 61 . 5–98 . 5% ) for arm B , and 88 . 4% ( range: 75 . 1–98 . 9% ) for arm C . The individual DOT coverage rates achieved in the 75 villages are shown in S2 Table . A coverage of 75% and above was achieved in 57 schools ( 76 . 0% ) , while a coverage of less than 75% was reported in the remaining 18 schools . Yaoudé ( in arm A ) reported a coverage below 50% . The DOT coverage was not significantly correlated with changing levels of S . mansoni prevalence ( Spearman ρ = -0 . 11; p = 0 . 43 ) , while it was significantly correlated with AM EPG ( Spearman ρ = 0 . 32; p = 0 . 02 ) ( Fig 5 ) .
Preventive chemotherapy with praziquantel is the backbone of the global strategy against schistosomiasis and other helminthiases [12 , 29] . Our findings show that one year after an initial treatment with praziquantel in 50 schools of western Côte d’Ivoire that met inclusion criteria of a SCORE harmonization protocol ( prevalence ranging between 10% and 24% ) [15] , the overall S . mansoni prevalence was reduced from 19 . 7% to 12 . 8% , while there was no significant change in the overall AM EPG . The overall DOT coverage in the study area was 84 . 2%; hence , above the 75% coverage recommended by WHO [16] . At school level , the picture on the impact of the SBT was less clear cut . Decreases in prevalence and infection intensity were observed in some schools and increases in others . Among the six schools that showed higher prevalences of S . mansoni at the 1-year follow-up compared with baseline , in only one school , the treatment coverage was <75% . The changes in the AM EPG level were significantly correlated with the coverage rate . The overall reduction of the S . mansoni prevalence in the first year of this SCORE project ( 35 . 0% ) is in line with studies assessing the S . mansoni prevalence 12 months post-MDA in central Sudan and Uganda , where reductions of S . mansoni prevalence of 36 . 7% and 39 . 5% were observed , respectively [30 , 31] . The treatment coverage in these two studies was reported to be 100% and 79 . 2% , respectively [31 , 32] . In the Sudan study , treatment of children with praziquantel was conducted by trained nurses and medical officers , while in Uganda , the treatment was carried out by trained teachers and community drug distributors [31 , 32] . A survey conducted 6 months after praziquantel treatment in Sierra Leone where the overall treatment coverage was 94 . 0% found a reduction of the S . mansoni prevalence of 44 . 6% [33] . Another study carried out in Sierra Leone reported an even higher reduction in the S . mansoni prevalence of 67 . 2% , as determined three years after three rounds of praziquantel administration [34] . In contrast , studies conducted in Zambia and Kenya showed that 2 years after the withdrawal of praziquantel treatment led to an increase of S . mansoni prevalence [35 , 36] . It is important to note that these studies showed that the impact of MDA on the S . mansoni prevalence varied depending on the infection status in a given area , and the frequency and number of treatment rounds . Repeated treatments over short time periods can lead to a high reduction in S . mansoni prevalence compared to longer treatment intervals . Similar baseline S . mansoni prevalences were observed in two preceding studies in Sierra Leone and Uganda ( 49% and 42% , respectively ) , but the decrease in S . mansoni prevalence was lower in Uganda , where the intensity of infection , and thus the level of transmission , was higher . A plausible explanation of this observation arises from rapid re-infection , which is related to the force of infection , and which is likely higher where S . mansoni transmission is intense . Indeed , the level of schistosomiasis transmission , which is governed by various factors , such as local environmental determinants , climate , water contact patterns , intermediate host snail distribution , and ecology , may affect the impact of MDA [37–40] . When interpreting these results , one has to bear in mind , however , that the prevalence of S . mansoni was determined by an insensitive diagnostic approach; single stool samples subjected by single ( Uganda ) or duplicate Kato-Katz thick smears ( Sierra Leone ) . Hence , the diagnostic approach was less rigorous than in the current study in Côte d’Ivoire , where only those children who had at least quadruplicate Kato-Katz thick smears examined before and after treatment were included in the final analysis . In our study , in the schools Biélé and Ziondrou , the S . mansoni prevalence had significantly increased one year after SBT with levels in excess of 60% . Since the DOT coverage in both schools was high ( 75 . 2% in Biélé and 91 . 9% in Ziondrou ) , we assume that there are major transmission hotspots in the area , where children become rapidly re-infected . Re-emergence of S . mansoni and S . haematobium after treatment in high-endemicity areas has previously been reported from other studies in Côte d’Ivoire and Niger [41 , 42] . One explanation might be migration of people , including those infected with S . mansoni or S . haematobium , into treated villages . A considerable population movement has , for example , been observed in Côte d’Ivoire due to socio-political unrest in 2011 [43] , hence at the start of our study . A lack of access to safe water , sanitation , and hygiene ( WASH ) might also be the reason for rapid reinfection . Noteworthy , when interviewing the local village leaders , they reported that people in the area frequently use well water for washing and bathing , while ponds and rivers serve as the main natural water contact sites . While some houses have latrines , many people still practice open defecation . Another explanation of the increase in S . mansoni prevalence might be the target population of the treatment strategy . The present study focused on school-aged children . Preschool-aged children and adults also harbor Schistosoma worms , and hence , they act as reservoir of transmission source of re-infections [36] . Yet , there are other local conditions that might foster S . mansoni transmission in Biélé and Ziondrou that warrant further investigation . For example , one might want to assess the frequency and duration of water contact in children and associated re-infection patterns , and the transmission force caused by intermediate host snails populating waterbodies located in close proximity to the surveyed schools . It will be important to assess in future surveys whether individuals had indeed received praziquantel in the past treatment round , or whether they were immigrating from other areas after the last survey , or had traveled to highly endemic areas over the past year . Ideally , the reinfection pattern would be determined by following a cohort of children , including immunological markers of the individuals that might favor or delay reinfection , and molecular markers of the infecting parasites . An increase of S . mansoni infection within the frame of ongoing treatment programs has also been observed elsewhere . In Senegal , for example , an elevated S . mansoni prevalence was found 10 months after praziquantel administration [44] . More recently , in Ségou district in Mali , the national control program had revealed an increase of the S . mansoni prevalence after four rounds of MDA in 7- to 14-year-old children [45] . It has been assumed that these increases of S . mansoni infections after praziquantel treatment might be explained by partial resistance to praziquantel , the acquisition of new infection , and high force of transmission [46–48] . Taken together , our data show that SBT resulted in marked decreases of S . mansoni prevalence , but the intensity of infection among infected children did not change significantly . Hence , with a single treatment round , the force of transmission in terms of egg excretion in the school-aged population has not been changed in most of our study schools . Monitoring the impact of multiple treatment rounds and “drug holidays” over the next years will provide stronger evidence of what multiple SBT rounds can achieve [13 , 15] . Clearly , sustainable control and eventual elimination of schistosomiasis requires multiple intervention packages , such as preventive chemotherapy ( perhaps extended to all age groups ) , intensified case management , control of intermediate host snails , provision of WASH , and setting-specific information , education , and communication ( IEC ) [49–51] . In Côte d’Ivoire , the control of schistosomiasis at a national scale is still at an early stage . Indeed , the PNL-SGF was only launched shortly before this SCORE project . For the success and sustainability of schistosomiasis control in Côte d’Ivoire–and elsewhere in sub-Saharan Africa–it will be important that , in addition to preventive chemotherapy , other control measures are considered and implemented [6 , 7 , 49 , 52] . The present study showed that one year after SBT with praziquantel , the overall prevalence of S . mansoni infection had decreased significantly . However , in certain hotspot schools , the S . mansoni prevalence had increased unexpectedly . The infection intensity among S . mansoni-infected children was similar at the 1-year follow-up . These results demonstrated that the dynamic of schistosomiasis in the study areas is heterogeneous and that a single round of treatment is insufficient to have a lasting effect . It will be important to monitor the dynamic of schistosomiasis over the course of this SCORE study , in order to deepen our understanding of the dynamics of schistosomiasis transmission in a moderately endemic setting .
|
Schistosomiasis is a parasitic worm disease that is widespread in sub-Saharan Africa . To better understand how to gain and sustain the control of schistosomiasis and how to eliminate this disease in different epidemiologic settings , the Schistosomiasis Consortium for Operational Research and Evaluation ( SCORE ) has launched several multi-year studies that are being implemented in East and West Africa . This article highlights how the Schistosoma mansoni infection levels changed one year after an initial treatment with the anti-worm drug praziquantel given to children aged 5–15 years in western Côte d’Ivoire . Infection and treatment data at school level were available from more than 4 , 600 children in 50 schools . One year after the treatment that had been received by more than 80% of the children , the overall S . mansoni prevalence decreased from 19 . 7% to 12 . 8% , while the intensity of infection among S . mansoni-positive children slightly increased . In several schools , the S . mansoni intensity and , particularly the prevalence , increased unexpectedly . Our findings show that the dynamics of schistosomiasis transmission varies from one village to another . It will be interesting to monitor changes over longer time periods as this SCORE study unfolds .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[] |
2016
|
Sustaining Control of Schistosomiasis Mansoni in Western Côte d’Ivoire: Results from a SCORE Study, One Year after Initial Praziquantel Administration
|
Recent genome-wide experiments in different eukaryotic genomes provide an unprecedented view of transcription factor ( TF ) binding locations and of nucleosome occupancy . These experiments revealed that a large fraction of TF binding events occur in regions where only a small number of specific TF binding sites ( TFBSs ) have been detected . Furthermore , in vitro protein-DNA binding measurements performed for hundreds of TFs indicate that TFs are bound with wide range of affinities to different DNA sequences that lack known consensus motifs . These observations have thus challenged the classical picture of specific protein-DNA binding and strongly suggest the existence of additional recognition mechanisms that affect protein-DNA binding preferences . We have previously demonstrated that repetitive DNA sequence elements characterized by certain symmetries statistically affect protein-DNA binding preferences . We call this binding mechanism nonconsensus protein-DNA binding in order to emphasize the point that specific consensus TFBSs do not contribute to this effect . In this paper , using the simple statistical mechanics model developed previously , we calculate the nonconsensus protein-DNA binding free energy for the entire C . elegans and D . melanogaster genomes . Using the available chromatin immunoprecipitation followed by sequencing ( ChIP-seq ) results on TF-DNA binding preferences for ~100 TFs , we show that DNA sequences characterized by low predicted free energy of nonconsensus binding have statistically higher experimental TF occupancy and lower nucleosome occupancy than sequences characterized by high free energy of nonconsensus binding . This is in agreement with our previous analysis performed for the yeast genome . We suggest therefore that nonconsensus protein-DNA binding assists the formation of nucleosome-free regions , as TFs outcompete nucleosomes at genomic locations with enhanced nonconsensus binding . In addition , here we perform a new , large-scale analysis using in vitro TF-DNA preferences obtained from the universal protein binding microarrays ( PBM ) for ~90 eukaryotic TFs belonging to 22 different DNA-binding domain types . As a result of this new analysis , we conclude that nonconsensus protein-DNA binding is a widespread phenomenon that significantly affects protein-DNA binding preferences and need not require the presence of consensus ( specific ) TFBSs in order to achieve genome-wide TF-DNA binding specificity .
Binding of TFs to their target sites on the DNA is a key step during gene activation and repression . An existing paradigm assumes that the main mechanism responsible for specific TF-DNA recognition is TF binding to short ( typically 6–20 bp long ) DNA sequences called specific consensus motifs , or specific TF binding sites ( TFBSs ) . It has been known for a long time , since the seminal studies of Iyer and Struhl [1] , that genomic context surrounding specific TFBSs significantly influences TF-DNA binding preferences . However , general rules describing the mechanisms responsible for such influences remain unknown . Recently , the model organism ENCODE ( modENCODE ) project has revealed genome-wide comprehensive maps of TF-DNA binding and nucleosome occupancy in C . elegans [2–7] and in D . melanogaster [8–10] . Remarkably , these studies have challenged the existing paradigm and revealed that a large fraction of TF-DNA binding events occurs in genomic regions depleted of specific consensus motifs . Such genomic regions with enhanced overall TF-DNA binding but depleted in consensus motifs are oftentimes of low sequence complexity , which means that they are enriched in repeated DNA sequences . We have recently proposed that repetitive DNA sequences characterized by certain symmetries and length scales of repetitive sequence patterns ( see below ) exert a statistical potential on DNA-binding proteins , affecting their binding preferences [11–15] . This effect of protein binding to repetitive DNA sequences in the absence of specific base-pair recognition is different from the concept of nonspecific protein-DNA binding introduced and explored in seminal studies of von Hippel , Berg , et al . [16–21] . In particular , von Hippel and Berg defined two related mechanisms for nonspecific protein-DNA binding [19] . The first mechanism is DNA sequence-independent , and it assumes that DNA exerts an electrostatic attraction upon DNA-binding proteins , modulated by the overall DNA geometry [19] . It has been proposed that DNA-binding proteins use different conformations in specific and nonspecific binding modes [16–20 , 22] . The second mechanism assumes that mutated specific DNA consensus motifs retain a reduced binding affinity for sequence-specific TFs [19] . Nonspecific protein-DNA binding might become significant since the statistical probability to find such imperfect motifs in many genomic locations by random chance is high for eukaryotic genomes [19 , 23] . The importance of nonspecific protein-DNA binding has been experimentally demonstrated for a number of systems both in vivo [24 , 25] and in vitro [26–31] . We demonstrated recently that repetitive DNA sequence patterns characterized by certain symmetries lead to nonconsensus protein-DNA binding that can be enhanced or reduced depending on the symmetry type [11] . We use the term nonconsensus protein-DNA binding in order to emphasize the point that the nonconsensus protein-DNA binding free energy is computed without using any experimental information on specific protein-DNA binding preferences ( see below ) . For example , we showed that repetitive homo-oligonucleotide sequence patterns , such as repeated poly ( A ) /poly ( T ) /poly ( C ) /poly ( G ) tracts lead to statistically enhanced nonconsensus protein-DNA binding affinity [11] . Our results indicated that such nonconsensus binding significantly influences nucleosome occupancy [12] , TF-DNA binding preferences [13] , and transcription pre-initiation complex binding preferences [14] in yeast . In addition , using the protein binding microarray ( PBM ) method , we have recently directly measured the nonconsensus protein-DNA binding free energy for several human TFs [15] . We have demonstrated that , remarkably , the magnitude of the identified nonconsensus effect reaches as much as 66% of consensus ( specific ) binding [15] . In this study we explore the extent and significance of the nonconsensus protein-DNA binding mechanism for a large number of proteins belonging to different structural families . First , we investigate the nonconsensus effect in more complex , multicellular organisms , using the available ChIP-seq data obtained for ~100 TFs in C . elegans [2 , 3] and D . melanogaster [10 , 32] . Next , we perform the analysis of high-resolution in vitro universal protein-DNA binding microarray ( PBM ) data obtained for ~90 eukaryotic TFs belonging to 22 different DNA-binding domain types [33–35] . In addition , we identify protein sequence features that statistically distinguish between proteins with stronger and weaker response to nonconsensus repetitive DNA sequence elements , respectively . We stress the point that in vitro analysis is free of confounding factors present in a cell , such as nucleosomes and indirect TF-DNA binding . Our previous experimental in vitro study of nonconsensus protein-DNA binding was performed for only 6 TFs [15] . The present analysis of the vast amount of in vitro TF-DNA binding data extends this number to more than an order of magnitude , suggesting that the nonconsensus mechanism most likely represents the statistical law rather than the exception . Therefore , the results reported here strongly support our conclusion that nonconsensus protein-DNA binding is a widespread phenomenon that significantly affects protein-DNA binding preferences in eukaryotic genomes , and need not require the presence of consensus ( specific ) TFBSs in order to achieve genome-wide TF-DNA binding specificity .
We compared the predicted landscape of nonconsensus protein-DNA binding free energy with the genomic binding profiles of 69 transcriptional regulators in C . elegans [2 , 3] and 30 transcriptional regulators in D . melanogaster [10 , 32] , as determined by ChIP-seq in the modENCODE project [2 , 3 , 8 , 10] . We computed the nonconsensus binding free energy landscape using a simple approach that we developed previously [11] . Briefly , we used a set of random protein-DNA binders as a proxy for nonspecific protein-DNA interactions in a crowded cellular environment ( Fig 1 ) . Next , to each location along the C . elegans and D . melanogaster genomes , we assigned an average free energy of nonconsensus protein-DNA binding , 〈F〉TF , where the averaging is performed over an ensemble of random binders ( see Methods for further details ) . The free energy value at each sequence location is entropy-dominated , and it is influenced exclusively by the presence of repetitive DNA sequence patterns [11] surrounding that location . We use the term DNA sequence correlations to describe the repetitive DNA patterns , and the term correlation scale to describe the length of the patterns ( Methods ) . The larger the correlation scale , the larger the number of repetitive sequence patterns , and thus the stronger the nonconsensus protein-DNA binding effect [11] . Importantly , the genomic DNA sequence constitutes the only input for the nonconsensus binding model , i . e . the model does not have any fitting parameters ( Methods ) . We found that the nonconsensus protein-DNA binding free energy correlates negatively with the combined TF occupancy in both the C . elegans and the D . melanogaster genomes , i . e . the lower the nonconsensus binding free energy , the higher the combined TF occupancy ( Fig 2 ) . Fig 2a and 2c illustrate this correlation for free energy profiles , 〈〈F〉TF〉seq , averaged over genomic sequences aligned with respect to the TSS . A statistically significant correlation at the single gene level is also observed , on average , without sequence alignment with respect to the TSS ( Fig 2b and 2d ) . In these analyses both genomes show statistically significant negative correlations , with the correlation being more pronounced in C . elegans . We verified that the predicted free energy landscape is qualitatively robust with respect to variations in the model parameters ( i . e . the sliding window width , L , and the TFBS size , M ) ( S1 Fig ) . In addition , we validated that the predicted free energy landscape is determined by the presence of repetitive sequence patterns , and not by the average genomic nucleotide content . To show this , we shuffled the DNA sequence in each sliding window along the genome to obtain random DNA sequences with a fixed nucleotide content , and we computed the normalized free energy , δF = F−Frand , where Frand is the free energy of the random , shuffled sequences , averaged over different random realizations ( Methods ) . As shown in S2 Fig , the normalized free energy δF is robust with respect to global variations in the genomic nucleotide content . The predicted reduction in the nonconsensus free energy upstream of TSSs ( Fig 2a and 2c ) stems from the enhanced level of homo-oligonucleotide sequence correlations ( i . e . repetitive homo-oligonucleotide sequence patterns , such as repeated poly ( dA:dT ) tracts ) . This effect can be intuitively understood in the following way . As shown in our previous work , the presence of enhanced homo-oligonucleotide sequence correlations within a DNA region generally leads to the widening of the protein-DNA binding energy spectrum in this region [11] . For example , in the statistical ensemble of random binders interacting with DNA sequence that contains long homo-oligonucleotide tracts with two alternating types of nucleotides ( such as alternating poly ( dA:dT ) and poly ( dT:dA ) tracts ) , the width ( i . e . the standard deviation ) of the binding energy spectrum , σUhomo , will be universally larger than the corresponding width for the case of entirely random DNA sequence , σUhomo≃2⋅σUrandom [11] . This result is independent of the microscopic details of the protein-DNA interaction potential , U , and it is simply the consequence of the central limit theorem [36 , 37] . The wider energy spectrum , σUhomo>σUrandom , universally leads to the statistically lower free energy , Fhomo < F random [38] , and therefore to a higher nonconsensus protein-DNA binding affinity . The computed probability distributions of the nonconsensus protein-DNA binding energy and the free energy in the C . elegans genome , further illustrates this mechanism ( S3 Fig ) . Thus , the nonconsensus protein-DNA binding mechanism can significantly influence TF-DNA binding preferences in the C . elegans and D . melanogaster genomes , complementing the conventional , specific protein-DNA recognition mode . We stress the fact that the minimum of the average nonconsensus protein-DNA binding free energy landscape does not align precisely with the maximum of the average TF occupancy profile in both C . Elegans and D . melanogaster genomes ( Fig 2a and 2c ) . Such mismatch is also observed between the average nonconsensus protein-DNA binding free energy landscape and the average nucleosome profile ( see below , Fig 3a and 3c ) , similar to the case as we previously observed for the yeast genome [12] . Combination of additional factors not taken into account in our model but present in vivo might explain a possible origin of such a mismatch . These factors include , first , steric constrains imposed by the presence of nucleosome particles [39]; second , steric constrains imposed by the transcription pre-initiation complex ( PIC ) [40]; and third , the presence of specific TFBSs [41] . We also assessed the effect of nonconsensus protein-DNA binding on nucleosome binding preferences in the C . elegans and D . melanogaster genomes . Genome-wide measurements of nucleosome occupancy show a typical nucleosome depleted region upstream of the TSSs , and a well-positioned +1 nucleosome [2 , 4 , 42] . In D . melanogaster , an oscillating nucleosome occupancy pattern was observed , similar to the one in yeast [43] , while the C . elegans genome-wide nucleosome occupancy profile does not demonstrate such strong oscillations [4 , 42] . The computed nonconsensus free energy landscapes show a statistically high , positive correlation with the nucleosome occupancy profile in both genomes ( Fig 3 ) . In particular , the average nonconsensus free energy shows a pronounced minimum in the upstream nucleosome depleted region ( Fig 3a and 3c ) , similar to the one observed in yeast [12] . In Fig 3b and 3d we also observed , at the single gene level , statistically significant correlation between the average nucleosome occupancy and the average free energy of nonconsensus binding ( Methods ) . Sequences with lower nonconsensus protein-DNA binding free energy have , on average , lower nucleosome occupancy . We suggest that the observed effect stems from the competition between TFs that experience enhanced nonspecific attraction towards upstream promoter regions ( i . e . , reduced level of the nonconsensus free energy ) and nucleosome-forming histones . It is important to stress that the presence of repetitive DNA sequence elements in promoter regions might also affect histone-DNA binding due to the nonconsensus mechanism , and as a result of it , the nucleosome formation . How exactly individual histones and histone complexes respond to different repetitive DNA sequence patterns remains an open question . This issue is further complicated by the fact that several additional mechanisms influence histone-DNA binding in promoter regions . Namely , genome-wide , in vitro nucleosome reconstruction experiments demonstrate that nucleosome-free regions ( NFR ) can be formed to some extend even in the mixture of purified genomic DNA with histones [44 , 45] . However , intrinsic DNA sequence preferences of nucleosomes still remain an open issue [46] . In particular , it has been recently demonstrated that AT-rich sequences present in many NFRs have little effect on the stability of nucleosomes [46] . Rather it appears that ATP-dependent chromatin modifiers constitute a major factor regulating nucleosome-binding preferences in vivo [43 , 46] . Here we provide an additional , highly significant validation for the proposed mechanism of nonconsensus protein-DNA binding by the analysis of the available in vitro TF-DNA binding data obtained using the protein-binding microarray ( PBM ) technology [35 , 47–49] . The PBM technology allows to simultaneously measure binding of a TF to tens of thousands of 36-bp long DNA sequences in a single experiment [35] . The PBM method is free from the confounding factors , such as the effect of competing TFs and nucleosomes on TF-DNA binding preferences . Here , we used the currently available ‘universal PBM’ data for 91 TFs ( belonging to 22 distinct DNA-binding domains ) from C . elegans , D . melanogaster , and mus musculus [33 , 34 , 50] ( Fig 4 and S1 Table ) . The DNA libraries used in these ‘universal PBM’ experiments were designed in such a way that they cover all possible 8-mer DNA sequences [35] , giving an unbiased view of TF-DNA binding specificity . Overall , there are ~45 , 000 distinct DNA sequences in this library , and thus the TF-DNA binding strength was measured for each TF to all these sequences [33 , 34 , 50] . We computed the nonconsensus TF-DNA binding free energy , 〈f〉TF , for each 36-bp long DNA sequence in the library using the procedure described above ( Fig 1 ) . Contrary to the case of genomic sequences , here we do not move the sliding window along the DNA sequence since each sequence is short , L = 36 bp , and therefore a single value of 〈f〉TF is assigned to each DNA sequence . Remarkably , for 69 out of 91 analyzed TFs ( i . e . 76% ) we detected a statistically significant , negative correlation between the nonconsensus protein-DNA binding free energy and the measured in vitro TF-DNA binding intensity . This is in agreement with the results obtained for the in vivo TF-DNA binding data ( Fig 2b and 2d ) . Twelve TFs ( i . e . 13% ) did not show a statistically significant correlation , and interestingly , ten TFs ( i . e . 11% ) showed an opposite , positive correlation ( S1 Table ) . The latter observation is remarkable , since it demonstrates that a non-negligible fraction of TFs can respond to DNA symmetries ( represented by our free energy model ) in an opposite way compared to the majority of other TFs . However , statistically , the average TF-DNA binding preferences show highly significant , negative correlation with the computed free energy of nonconsensus protein-DNA binding ( Fig 4 ) in agreement with the in vivo results ( Fig 2b and 2d ) . In order to identify what structural and sequence features are responsible for the anomalous behavior of these 11% of TFs , we classified all TFs according to the DNA-binding domain ( DBD ) families they belong to . However , we have not identified any particular DBD families that are unique to those 11% of TFs ( S1 Table and S4 Fig ) . We have also not identified any preference of these TFs with respect to any particular biological function , according to the gene ontology ( GO ) classification . Therefore , the question what sequence and structural features of proteins are responsible for the positive correlation between the free energy and the experimentally measured in vitro TF occupancy remains open . Next , in order to identify protein sequence features that might be responsible for enhanced nonconsensus TF-DNA binding , we separated TFs ( we used 82 mouse TFs for this analysis ) into two groups . The first group contained 41 TFs with the strongest negative correlation between the free energy and the measured TF occupancy . The second group contained the remaining 41 TFs . We have analyzed the amino acid correlation properties in these two groups of TFs . Our working hypothesis here is that enhanced amino acid sequence correlations in TF sequences are responsible for enhanced nonconsensus TF-DNA binding . We use the term “sequence correlations” in order to describe repetitive sequence patterns . We have previously used a similar analysis in order to investigate protein sequence features responsible for enhanced level of protein structural disorder and protein-protein interaction promiscuity [36] . In particular , we have analyzed the frequency of occurrence of the following repetitive amino acid sequence patterns in each TF group: [aa] , [aXa] , [aXXa] , and [aXXXa] , where a represents each amino acid type and X represents an arbitrary amino acid ( S2 Table ) . For example , when we compute the frequency of [Lys-X-Lys] pattern , we count the total number of the occurrence of this pattern in each protein sequence , irrespectively to the identity of X . As a result of this analysis , we have identified three patterns that demonstrated a statistically significant difference of frequencies between the two TF groups: [Lys-XX-Lys] ( enriched in the first TF group; Kolmogorov-Smirnov p-value , pks ≃ 0 . 01 ) , [Arg-Arg] ( enriched in the second TF group; pks ≃ 0 . 02 ) , and [Leu-X-Leu] ( enriched in the first TF group;pks ≃ 0 . 05 ) ( S2 Table ) . In addition the overall compositional fraction of Lys was enriched in the first TF group ( pks ≃ 0 . 01 ) ( S2 Table ) . The fact that the most statistically significant enrichment ( distinguishing the two TF groups ) is observed for the [Lys-XX-Lys] and [Arg-Arg] patterns is encouraging since positively charged Lys and Arg are obviously the key amino acids responsible for TF binding to the negatively charged DNA molecule . Two conclusions can be drawn from our results . First , that the intrinsic propensity for nonconsensus protein-DNA binding is imprinted both into the DNA and the protein . Since our simple nonconsensus binding model treats proteins as random binders , it captures general trends in the binding profiles of most , but not all , TFs . Second , nonconsensus and specific ( consensus ) protein-DNA binding mechanisms are tightly interlinked , and both of these mechanisms cooperate in determining the overall protein-DNA binding preferences in eukaryotic genomes . The fact that our simple random-binder model ( without any fitting parameters and without any protein-DNA binding specificity built in ) provides such a good statistical description of the measured DNA binding strength for the majority of TFs strongly suggests that the nonconsensus mechanism is quite general and it represents the statistical law rather than the exception . However , more accurate , atomistic models describing nonconsensus protein-DNA binding interactions are necessary in order to improve the accuracy of our predictions for different proteins .
Our analyses of the effect of nonconsensus protein-DNA binding demonstrate that the combined genome-wide binding preferences of 69 TFs in C . elegans and 30 TFs in D . melanogaster are significantly , negatively correlated with the predicted nonconsensus free energy landscape ( Fig 2 ) . Our analyses also show that the experimentally derived nucleosome occupancy in C . elegans and in D . melanogaster is significantly , positively correlated with the predicted nonconsensus protein-DNA binding free energy ( Fig 3 ) . This trend is qualitatively similar to the one that we previously observed in yeast [12] . The results shown in Figs 2 and 3 strongly suggest that TFs compete with nucleosomes for nonconsensus binding to DNA . Such a competition between TFs and nucleosomes could lead to the enhanced TF binding cooperativity previously predicted by Mirny [51] and Teif et al . [52] . We suggest that nonconsensus protein-DNA binding greatly enhances such nucleosome-induced cooperativity between TFs , and most importantly , in order to achieve this enhancement , promoters do not require the presence of specific , consensus TF binding sites . We stress the important point that the predicted effect of nonconsensus TF-DNA binding most likely affects many but not all TFs . We expect for example , that stress response TFs , such as for example Msn2 in yeast [53] , might be insignificantly influenced by the nonconsensus mechanism . Our model predicts that genomic loci enriched with repetitive sequences , such as in heterochromatin , should also be enriched with TF binding . However , the ChIP-seq analysis in such regions is impeded by the fact that multi-mapping reads from long repetitive region will be filtered out by most peak-calling algorithms , therefore identifying interactions in these regions remains a challenging problem [54] . Interestingly , there are evidences that regions of heterochromatin are not actually transcriptionally inert and non-coding RNA molecules are transcribed from repeated DNA sequences in pericentromeric heterochromatin in different eukaryotic genomes [55] . A recent study even demonstrated [56] that some TFs bind directly to the major satellite repeat DNA sequences that are present in pericentromeric heterochromatin regions and might play a significant role in the mouse heterochromatin formation . Further experiments and analysis of TF binding to the heterochromatin would reveal whether nonconsensus binding play an important role in these regions as well . Our analysis of available in vitro TF-DNA binding data from protein-binding microarray ( PBM ) experiments ( Fig 4 and S1 Table ) demonstrates that statistically , on average , in vitro TF-DNA binding preferences negatively correlate with the computed nonconsensus free energy landscape , and showed qualitatively similar behavior to the one observed in vivo ( compare Fig 2b and 2d with Fig 4 ) . This additional analysis is important for several reasons . First , the in vitro TF-DNA binding preferences are not affected by the presence of other proteins and histones , which can compete with the protein or cause an indirect binding to the DNA . Second , the TF binding intensity is measured in PBM experiments at significantly higher accuracy compared to ChIP-seq experiments . Third , the usage of non-genomic sequences that cover all possible 8-mer DNA sequences , eliminates possible sequence bias that might exist in the genomic sequences , and thus PBM measurements provide an entirely independent validation of the nonconsensus protein-DNA binding effect . Finally , the present analysis performed for ~90 TFs extends our previous analysis performed for only 6 TFs [15] by more than an order of magnitude , thus strongly suggesting the generality of the nonconsensus protein-DNA binding effect in eukaryotic genomes . Interestingly , ten TFs ( i . e . 11% ) showed an opposite , positive correlation between the free energy and the measured TF-DNA occupancy ( S1 Table ) . The latter observation is remarkable , since it demonstrates that a non-negligible fraction of TFs can respond to DNA symmetries ( represented by our free energy model ) in an opposite way compared to the majority of other TFs . However , we failed to identify any particular structural , sequence , or functional features unique to this set of TFs . This failure might stem from the small number of proteins that exhibited such behavior . Yet , we were able to identify repetitive amino acid sequence patterns that are responsible for enhanced nonconsensus TF-DNA binding ( S2 Table ) . In particular , for the group of TFs characterized by the strongest nonconsensus TF-DNA binding preferences , the most statistically significant enrichment is observed for the [Lys-XX-Lys] pattern , while the frequency of [Arg-Arg] pattern is reduced in this group ( S2 Table ) . The latter result is intuitively sound since both Lys and Arg are the key amino acids responsible for TF binding to the negatively charged DNA molecule . Importantly , in this study , our random-binder statistical mechanics model for protein-DNA interactions does not use any experimentally pre-determined information on either low-affinity or high-affinity TF-DNA binding sites . The genomic DNA sequence constitutes the only experimental parameter of the model . In addition , our model does not have any fitting parameters . Contrary to the case of specific protein-DNA binding that requires the presence of a 6 to 20-bp long specific DNA motif ( unique for each individual TF ) , the nonconsensus protein-DNA binding effect stems from multiple nonspecific interactions between the TF and a relatively long ( few tens of bp ) DNA fragments enriched with repetitive sequence patterns . The fact that different TFs are affected in a statistically similar way by entirely different DNA sequences containing similar repetitive patterns constitutes the key difference between the nonconsensus and specific protein-DNA recognition modes . What exactly is the interplay between nonconsensus DNA repetitive sequence elements and consensus ( specific ) sequences and how their combination influences the overall binding of proteins to the DNA and the expression levels of genes are important questions yet to be explored . We suggest that repetitive nonconsensus sequence elements might have similar influence on TF-DNA binding and on gene expression as repeats of consensus ( specific ) DNA sequence elements ( i . e . homotypic clusters ) [57] . However , an important difference between these two types of repeated sequence elements is that nonconsensus repeats can affect many different TFs in a similar way , while homotypic clusters are more specific to a limited set of TFs . Repetitive sequence elements located near the consensus ( specific ) motif , could increase the TF association rate , by inducing the one-dimension “sliding” of the TF , and improving its search for the specific binding site [20 , 58] . The presence of many weaker sites flanking a strong binding site could lead to a funnel effect [59–62] , where the molecules are directed to the strong binding site as depicted in S5a Fig . It could also stabilize binding sites that are not strong enough individually [63 , 64] and increase the ability of binding sites to “withstand mutations” [65] . We use the C . elegans Hlh-1 protein as an example demonstrating that nonconsensus DNA sequence elements might stabilize the binding to specific consensus elements in vivo ( S5b Fig ) . The analysis of Hlh-1 binding sites ( based on the genome-wide ChIP-seq measurements [2 , 3] in C . elegans ) demonstrates that only 5% of the total number of Hlh-1 specific motifs in the genome is bound by Hlh-1 ( S5b Fig ) . We sorted the genomic sequences containing the Hlh-1 motif ( consensus motifs were reported in [3] ) into two groups: the first group contains DNA sequences that were experimentally determined as being bound by Hlh-1 , while the second group contains unbound DNA sequences . S5c Fig represents the average nonconsensus protein-DNA binding free energy computed for each of these two sequence groups . We observed that the nonconsensus free energy is reduced for the group that contains bound sequences as compared with the group that contains unbound sequences . The computed p-values show that this result is statistically significant ( S5c Fig ) . This example supports the hypothesis that nonconsensus sequence elements might provide the funnel effect in vivo . Additional analysis and experimental measurements of the kinetics of TF-DNA binding to consensus ( specific ) sequence elements embedded in different nonconsensus DNA backgrounds , should shed more light on this hypothesis . Future in vitro measurements of binding preferences for additional TFs [66] , combined with high-resolution in vivo ChIP-seq and ChIP-exo analysis , will help to complete the molecular picture of design principles for nonconsensus protein-DNA binding and its functional significance .
We used the set of 23 , 287 C . elegans genes based on Wormbase annotation , WS228 [2 , 67] , and 12 , 188 D . melanogaster genes annotated in [10] . We used experimentally measured binding preferences of 69 C . elegans TFs ( S3 Table ) , as determined by the Gerstein and Snyder labs [2 , 3]; for computing the D . melanogaster TF occupancy we used binding preferences of 30 TFs ( S4 Table ) determined by the White lab [8] . TF-DNA binding preferences for both genomes were measured using ChIP-seq assays ( modENCODE project ) . We defined TF occupancy for each genomic location as the total number of bound TFs at each location along the genome . We used experimentally measured , genome-wide , normalized nucleosome occupancy determined by the paired-end Ilumina sequencing in C . elegans [4 , 5]; we also used the genome-wide map of H2A . Z nucleosome occupancy in D . melanogaster embryos ( 0–12 hr ) ( determined in [32] ) . We used experimentally measured in vitro binding intensity for the C . elegans , D . melanogaster , and mus musculus TFs ( S1 Table ) , determined using the protein-binding microarray ( PBM ) technology [33 , 35 , 47–49] . In order to compute the nonconsensus protein-DNA binding free energy landscape , we generate an ensemble of random DNA binders as a proxy for the phenomenon of nonconsensus protein-DNA binding in a crowded cellular environment [11] . Our model does not use any experimentally pre-determined protein-DNA binding preferences in order to model protein-DNA binding . The actual DNA sequences of the C . elegans and D . melanogaster genomes constitute the only input parameter for our model . In order to compute the free energy of nonconsensus protein-DNA binding at any given location along a DNA sequence , we position the center of the sliding window of width L = 50 bp at that location . The 50 bp length is a typical sliding event distance of a protein along the DNA under physiological conditions [68 , 69] ( Fig 1 ) . We assume that a model protein ( random binder ) makes M bp contacts with the DNA ( Fig 1b ) and that the model protein-DNA interaction energy at each genomic position i is simply a sum of M interaction energies: U ( i ) =−∑j=iM+i−1∑α={A , T , C , G}Kαsα ( j ) ( 1 ) where sα ( j ) represents the elements of a four-component vector of the type ( δαA , δαT , δαC , δαG ) , and δαβ = 1 if α = β , or δαβ = 0 if α ≠ β . For example , if the A nucleotide is positioned at the coordinate j along the DNA , then this vector takes the form: ( 1 , 0 , 0 , 0 ) . If , for example , the DNA sequence contains entirely poly ( A ) at a given genomic location , then a random binder makes all M contacts with the A nucleotide , and hence at this location the resulting energy , Eq ( 1 ) , will be simply , MKA . In order to generate each model protein , we draw the values of KA , KT , KC , and KG from Gaussian probability distributions , P ( Kα ) , with zero mean , and standard deviation σα = 2kBT , where T is the temperature and kB is the Boltzmann constant . We have shown previously that the resulting free energy is qualitatively robust with respect to the choice of model parameters [11] . The energy scale , 2kBT ≃ 1 . 2 kcal/mol , is chosen to represent a typical strength of a hydrogen bond , or an electrostatic bond that a protein makes with one DNA bp [16 , 19] . For each model random binder , we define the partition function of protein-DNA binding within the chosen sliding window of width L bp: Z=∑i=1Lexp ( −U ( i ) /kBT ) ( 2 ) and the corresponding free energy of nonconsensus protein-DNA binding in this sliding window: F=−kBTlnZ ( 3 ) We then assign the computed F to the sequence coordinate in the middle of the sliding window . Next , we move the sliding window along the DNA sequence and we compute F at each sequence location . This procedure allows us to assign the free energy of nonconsensus protein-DNA binding to each DNA bp within the genome . Next , we repeat the described procedure for an ensemble of 250 model random binders ( Fig 1 ) and compute the average free energy , 〈FTF〉 , over this ensemble , at each sequence location . We stress that the resulting free energy is qualitatively robust with respect to the choice of the sliding window size , L , within a wide range of values ( S1 Fig ) . In addition , the free energy profiles are statistically robust with respect to a moderate variation of the value of M , within a typical range of the TF binding site size ( S1 Fig ) . We verified that the predicted free energy landscape is dominated by DNA sequence correlations , and not by the average nucleotide composition ( S2 Fig ) . In particular , for each random binder , in each sliding window we computed the normalized free energy , δF = F−Frand , where Frand is the free energy computed for a randomized sequence ( in the same sliding window as F ) and averaged over 25 random realizations . In order to compute the p-value for S5c Fig , we first selected all the 800 bp-long sequences containing the exact binding motifs for each TF . For example , genome-wide , we have overall 9258 sequences containing the consensus Hlh-1motif . Among those 9258 sequences , 442 sequences were experimentally determined as bound by Hlh-1 , while the rest of 8816 sequences were unbound . In order to compute the p-value , we compiled 105 pairs of groups containing 442 and 8816 sequences , respectively , randomly chosen from the original 9258 sequences . These 105 pairs of groups represent randomized analogs for the original groups of bound and unbound Hlh-1motifs . Second , for each of these pairs of random groups we computed the average free energies , 〈f〉 , of nonconsensus binding separately for the randomized bound and unbound groups , as described above . Third , for each pair of randomized groups we computed the difference of the integrated free energy within the interval ( -400 , 400 ) between the two randomized groups . Finally , we computed the probability that this difference is equal or larger than the actual value of the difference . The latter probability was taken as the p-value .
|
Interactions between proteins and DNA trigger many important biological processes . Therefore , to fully understand how the information encoded on the DNA transcribes into RNA , which in turn translates into proteins in the cell , we need to unravel the molecular design principles of protein-DNA interactions . It is known that many interactions occur when a protein is attracted to a specific short segment on the DNA called a specific protein-DNA binding motif . Strikingly , recent experiments revealed that many regulatory proteins reproducibly bind to different regions on the DNA lacking such specific motifs . This suggests that fundamental molecular mechanisms responsible for protein-DNA recognition specificity are not fully understood . Here , using high-throughput protein-DNA binding data obtained by two entirely different methods for ~100 TFs in each case , we show that DNA regions possessing certain repetitive sequence elements exert the statistical attractive potential on DNA-binding proteins , and as a result , such DNA regions are enriched in bound proteins . This is in agreement with our previous analysis performed for the yeast genome . We use the term nonconsensus protein-DNA binding in order to describe protein-DNA interactions that occur in the absence of specific protein-DNA binding motifs . Here we demonstrate that the identified nonconsensus effect is highly significant for a variety of organismal genomes and it affects protein-DNA binding preferences and nucleosome occupancy at the genome-wide level .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods"
] |
[] |
2015
|
Nonconsensus Protein Binding to Repetitive DNA Sequence Elements Significantly Affects Eukaryotic Genomes
|
Negative examples – genes that are known not to carry out a given protein function – are rarely recorded in genome and proteome annotation databases , such as the Gene Ontology database . Negative examples are required , however , for several of the most powerful machine learning methods for integrative protein function prediction . Most protein function prediction efforts have relied on a variety of heuristics for the choice of negative examples . Determining the accuracy of methods for negative example prediction is itself a non-trivial task , given that the Open World Assumption as applied to gene annotations rules out many traditional validation metrics . We present a rigorous comparison of these heuristics , utilizing a temporal holdout , and a novel evaluation strategy for negative examples . We add to this comparison several algorithms adapted from Positive-Unlabeled learning scenarios in text-classification , which are the current state of the art methods for generating negative examples in low-density annotation contexts . Lastly , we present two novel algorithms of our own construction , one based on empirical conditional probability , and the other using topic modeling applied to genes and annotations . We demonstrate that our algorithms achieve significantly fewer incorrect negative example predictions than the current state of the art , using multiple benchmarks covering multiple organisms . Our methods may be applied to generate negative examples for any type of method that deals with protein function , and to this end we provide a database of negative examples in several well-studied organisms , for general use ( The NoGO database , available at: bonneaulab . bio . nyu . edu/nogo . html ) .
Despite the recent influx of machine learning algorithms applied to function prediction , there has been relatively little study devoted to the issue of class imbalance in function labels . This imbalance stems from the fact that the current standard set of labels for protein functions , the Gene Ontology ( GO ) database [1] , rarely stores which proteins do not possess a function . If no annotation is present for a given gene to a particular GO term , it does not mean that such a gene is a negative example for that term , but rather that it is either a negative example or a positive example that has yet to be annotated . This situation arises due to experimental constraints: function assays are typically applied to single proteins and that protein function can be context dependent , making negative statements/labels quite uncertain , and leading to very few ( or for most protein functions , not any ) verified negative examples . This imbalance presents an obvious problem for the vast majority of machine learning techniques , which require enough examples of both the positive and negative class in order to train an accurate predictor . Without these labeled negative examples , authors often resort to heuristics in order to define the non-positive class; but mistakes stemming from these heuristics can lead to false negatives in the training set , and are detrimental to classifier performance . The situation described above , in which the only known labels are of the positive class , is not unique to the protein function prediction ( PFP ) problem , but also occurs in several other domains . It has been given the name Positive-Unlabeled ( PU ) learning , and there has been a surge of interest lately in this particular subset of semi-supervised machine learning problems . One branch of PU algorithms attempts to learn in a one-class scenario , as has been applied to biology , specifically mRNA detection [2] . As the authors point out , however , 2-class machine algorithms often perform better when the negative class can be well defined . In another 1-step algorithm [3] , the authors demonstrate that if certain conditions hold , learning without explicitly knowing negative examples is possible and even more accurate than existing methods . Unfortunately , this assumption requires the probability of a true positive example being labeled to be independent of the example itself ( the set of observed positive labels should be selected at random from the total set of true positives ) . Since GO terms are often propagated via homology methods , there is a high degree of correlation between many of the labeled positive examples , and so this assumption does not hold in our domain . Thus we focus on the majority of PU algorithms , which proceed by first predicting a set of reliable negative examples before applying a traditional machine learning classifier to the enriched data as usual . These 2-step algorithms take many forms ( see [4] for review of these methods ) , but in this work we will refer to two main subcategories: passive 2-step PU algorithms , which learn the negative examples through a separate mechanism from the classifying algorithm , and active 2-step PU algorithms , which work in conjunction with the classifier to learn the negative examples . The main focus of PU-learning literature has been to improve text classification [4] , a problem in which labeling a document's topics is time-intensive , and it is not practical to label all the topics a document does not contain . Yet the analogies to protein function are clear: proteins are rarely labeled with the functions they do NOT possess , and proteins are nearly always multi-topic , in that the annotation of a protein to a particular GO-term does not exclude the potential for several other functional classifications ( we use the word “function” synonymously with “GO term” , regardless of which branch of GO that term occurs in ) . Therefore PU algorithms are applicable to the function prediction problem , and hold great potential for improvements in machine learning algorithms applied in this context . For example , we have previously shown that more-reliable negative examples boost the predictive power of protein function prediction algorithms [5] . We proceed by focusing directly on the first step of the PU learning task , namely generating a reliable set of negative examples for protein function and directly evaluating the quality of our negative examples , rather than their indirect effect on classifier performance . While PU learning has been applied to the biological domain before [2] , [6] , [7] , to the best of our knowledge no study has focused on evaluating the quality of negative examples for GO functions . We examine many of the heuristics used for protein function negative examples in the past , including: designating all genes that don't have a particular label as being negative for that label [8] , randomly sampling genes and assuming the probability of getting a false negative is low ( often done when predicting protein-protein interactions , as in [9] ) , and using genes with annotations in sibling categories of the category of interest as negative examples [10] , [11] . To these heuristics we add two common PU algorithms used in text classification but here adapted to PFP , the Rocchio algorithm [12] and the “1-DNF” algorithm [13] , as well as our ALBNeg algorithm [5] , and one of the few previously-published protein-negative-example-selection algorithms , the AGPS algorithm [7] . In addition , we present two new techniques: the first , Selection of Negatives through Observed Bias ( SNOB ) , is an extension of our ALBNeg algorithm ( which can itself be viewed as a generalization of the “1-DNF” PU algorithm ) , while the second , Negative Examples from Topic Likelihood ( NETL ) , is based on a Latent Dirichlet Topic model of GO data . Our algorithms , as well as competing algorithms borrowed from text classification , require only existing GO annotations in order to predict negative examples . As new annotations are continuously added to GO this allows testing via training on archived GO data , and examining the number of incorrectly predicted negative examples using current GO data to identify true positives that were predicted to be negative . The AGPS method utilizes additional feature data , such as Gene Expression , Protein-Protein-Interaction , etc . , but can still be evaluated in the same manner as the other algorithms . We provide a case study to show how these examples can benefit the performance of other algorithms , specifically a function prediction method tested in A . thaliana[14] . Additionally , we demonstrate increases in function prediction accuracy when our negative examples are used , testing on human , mouse , and yeast proteins , using our earlier-published function prediction algorithm [5] . Lastly , we provide a resource , NoGO , which contains lists of high-quality negative examples for GO categories in a variety of well-studied organisms ( Human , Mouse , Worm , Yeast , Rice , and Arabidopsis ) .
Function prediction results are biased negatively ( estimations of function prediction accuracy are typically lower limits ) by the fact that a positive prediction without a corresponding validation annotation might simply indicate lack of study of the gene rather than an incorrect prediction . It therefore follows that negative example validations are biased by the same effect , but positively ( estimated error rates are lower bounds ) . Just because a gene is not annotated with the function in the validation data doesn't guarantee that it was correctly identified as a negative example . In order to attempt to rigorously evaluate potential negative example selection algorithms , we utilize the average number of false negative predictions over categories in each of the three branches of GO . We determine false negatives through a temporal holdout in order to mitigate bias [15] , running all of our algorithms on data from the human genome obtained in Oct . 2010 , and then validating with data obtained in Oct . 2012 . This process involves restricting the training phase of all algorithms to data available in Oct . 2010 , removing the potential for test and training data correlation that can happen during cross-validation . Any gene that was predicted as a negative example from 2010 data , which received a positive annotation in the 2012 data , is considered an error in prediction ( a false negative example ) . For extra stringency , we consider an “Inferred by Electronic Annotation” ( IEA ) evidence code annotation as an indication of false negativity ( even though these types of annotations are traditionally considered less reliable ) . For completeness , we also include an evaluation without considering IEA annotations , presented in Figures S4 and S5 . Prediction errors are calculated separately for each GO term , and then averaged together within each branch of GO . Only categories that have between 3 and 300 annotations are evaluated , so as to consider only terms specific enough to be interesting but not so specific as to have little chance of being validated , since prediction errors can be observed only if new annotations appear for the category in question in the Oct 2012 data that were not present in the Oct 2010 data . Additionally we focus on a specific GO term in human ( RNA Binding ) , augmenting the temporal validation with annotations from a recent high throughput screen for RNA binding proteins [16] . Lastly , we evaluate using a gold-standard set for a single GO term in the yeast genome [17] . As the trivial solution ( predicting no negative examples ) would obviously have the lowest number of false negatives , we present results in two dimensions , where the vertical axis is average number of false negatives , and the horizontal axis is number of negative examples predicted ( in this setup , the origin represents the trivial solution , while the upper right corner of the plot represent choosing all non-positive genes as negatives ) . Algorithms that do not have the capability to vary the number of negative examples that they predict appear as points on the performance graph , instead of lines . Because prediction errors can be evaluated only if new annotations appear during the course of the temporal holdout time period , the error rate calculated is an observed error rate , rather than the true error rate . This observed rate will vary in magnitude from GO term to GO term , as it is bounded from above by the number of new annotations . Since the magnitude of the number of false negatives in each branch is dependent on the total number of new annotations added in that branch between 2010 and 2012 , the numbers cannot be compared across branches . In order to provide a reference point that is comparable across each branch , we treat the performance of random selection of negative examples as a baseline . Thus while the magnitude of the observed error rate cannot be compared across branches , the difference between an algorithm and the random baseline is comparable , both across branches and between GO-terms of differing specificity . Our first novel negative example selection algorithm , Selection of Negatives through Observed Bias ( SNOB ) , is an extension of our previously published ALBNeg algorithm [5] , which selected negative examples for a function based on whether or not a gene's most specific functional annotations had ever appeared alongside that function . ALBNeg in turn can be viewed as a generalization of a popular passive 2-step PU-learning algorithm known as “1-DNF” negative example selection . This algorithm works in the context of text classification by identifying words that are enriched among the positive class , and using as negatives all unlabeled documents that do not contain any of these positive “indicator” words [4] . We consider each GO term annotation as a “word” in the “document” of a protein , then apply the “1-DNF” technique to choose negative examples for a protein function by excluding proteins with GO terms that are enriched among proteins containing the function of interest . In ALBNeg , we generalized the idea of “enrichment” , by computing the empirical conditional probability of the GO function of interest , denoted g , given the presence of each other GO function in all three branches [5] . Proteins whose most specific annotations had non-zero conditional probabilities of appearing in a gene alongside g were ruled out from the potential negative set for g , effectively using the conditional probability as an indicator of potential positivity in the same way that the “1-DNF” algorithm uses enriched terms . In our new algorithm ( SNOB ) , presented here , we follow the same approach as ALBNeg , and for each GO term g , compute the pairwise empirical conditional probability of seeing g given the presence of each other GO term . We further develop ALBNeg , i ) by including IEA annotations in our calculations as well . We then obtain a score for each protein for each GO term g , by averaging the conditional probabilities of all GO terms ( including IEA ) annotated to that protein , ii ) by including all GO terms in the average , not just the most specific terms , and iii ) instead of choosing all proteins with a score of 0 as negatives for the function g , we allow the user to set a desired number n of negative examples , and choose the n proteins with the lowest scores as our negatives for g . See the Methods section for details of this calculation . Our second novel algorithm , Negative Examples from Topic Likelihood ( NETL ) , again treats proteins analogously to “documents” , with the GO terms annotated to each protein serving analogously to a document's “words” , but now we consider the proteins to have latent “topics” as well . These hidden topics represent the “true” function of the protein , both accounting for new functions ( functions not annotated because they have to be verified/tested ) as well as errors and missannotations ( having a GO annotation does not guarantee that a protein actually performs the function in question due to potential errors in annotation , especially with IEA annotations ) . We can then apply a multi-topic inference algorithm , specifically Latent Dirichlet Allocation [18] , to learn the distribution of these latent topics , or “true” functions , and also learn the conditional distribution of the “words” or annotated GO terms based on those topics . Once these distributions are known , NETL selects as negatives the proteins whose latent topic distributions are as dissimilar from the positive class as possible , allowing the user to specify how many negative examples are desired . Ideally each latent topic would represent a single GO term , but since the size of the vocabulary in our corpus is also equal to the number of GO terms , this is not feasible . Instead , we utilize the GO hierarchy to select fewer but more general topics , while ensuring coverage of the entire GO tree . Such a setup does not guarantee an intuitively interpretable relation between the latent topics and specific GO terms: topic x does not directly correlate to any one GO term , but rather is likely a combination of GO terms . Thus the calculation of the likelihood of a particular protein being a negative example for a particular GO term is infeasible , and must instead be inferred through a similarity metric ( see methods ) . In order to provide a reference for the quality of our algorithm's negative examples , we include past heuristics used for negative example selection , as well as the popular passive 2-step PU algorithms , “1-DNF” and “Rocchio” , which we have adapted to the PFP context through the GO term “word” and protein “document” mechanism described above . In the case of the Rocchio algorithm , we made an additional adjustment allowing the number of negative examples to be varied ( See Methods for details ) . We have chosen to focus on passive 2-step PU algorithms as the performance of active 2-step methods is intertwined with the performance of the underlying classification algorithm , as well as the input feature data . A stronger classifier will produce better negative examples , as will a classifier that can use more discriminative data . This increases the difficulty of judging the relative performance of active 2-step PU algorithms , as different classifiers utilize different mechanisms and datasets . These underlying differences make it difficult to correctly attribute relative performance of negative example selection to the 2-step algorithm itself , as opposed to the quality of the classifier or underlying data . Additionally , 2-step algorithms are self-reinforcing , in that the classifier identifies as negatives those proteins which are most different from the positive class by whatever mechanism that classifier is using , which only reinforces that particular kind of discrimination when the classifier is run again with the negative examples in the second step . In general , a classifier is better served with negative examples that are actually more similar under the classifying mechanism , in order to force the classifier to be more discriminative . Lastly , the passive 2-step algorithms presented here function solely with GO data input , allowing for very rapid calculations and avoiding the need to gather large amounts of feature data , which can quickly become difficult for less-studied organisms . The exception to our focus on passive 2-step algorithms is the AGPS algorithm , which is an active 2-step PU algorithm with which we make a comparison . We have included this algorithm , as it is one of the few explicit negative example selection algorithms in the protein function prediction ( PFP ) literature . Results for the methods tested on the human proteome are presented in Figure 1 . Among the methods tested , all algorithms performed better than the random baseline , with the exception of the sibling algorithm , whose weakness is also confirmed in [10] . The heuristic of choosing all non-positive genes as negative also does not perform better than the baseline , as it is itself a special case of the baseline where the number of negative examples is allowed to be the size of the genome ( minus the number of positive examples ) . The best performance was achieved by the SNOB algorithm , which achieved an equal or lower average number of false negatives than all other algorithms , heuristics , and the baseline , across all three branches . The NETL algorithm , as well as our adaptation of the Rocchio algorithm to PFP , also exhibited strong performance compared with other algorithms . Driving the performance of SNOB was its ability to achieve significantly fewer false negative predictions for more general GO categories ( categories with more annotations in the human genome ) . Figure S1 shows false negative rates broken down by the specificity of the function , demonstrating that while the Rocchio algorithm can compete with or even outperform our SNOB algorithm on the most specific categories , it is eclipsed by SNOB in the more general ones . This discrepancy among categories is most likely driven by the fact that the SNOB algorithm directly utilizes the co-occurrence of functions ( See the Methods section ) , and thus has less information to work with for the most specific categories . While not performing as well as SNOB , our previously published ALBNeg algorithm still achieves comparable or better performance than the AGPS algorithm . This comes as somewhat of a surprise , as AGPS has the benefit of access to a wealth of biological data beyond the GO information utilized by our algorithms , and much of that data post-dates the training GO annotations , providing unfair bias due to the correlation of many data types with GO annotations . However , with that additional data comes additional noise , and we recognize that the AGPS algorithm might be able to improve upon its performance with additional parameter tuning and feature selection among the data inputs . The results presented in Figure 1 represent the average of a large number of individual evaluations , each with an error rate whose magnitude can vary largely depending upon the specificity of the term . We encourage the reader to examine Figure S1 , which presents the same results but broken down by specificity , reducing the information lost by averaging . These results agree with those in Figure 1 . To further substantiate our evaluation , we focused on one particular molecular function term: GO:0003723 RNA Binding , presented in Figure 2 . We augmented the temporal holdout validation data with additional annotation not yet present in GO , but which have been experimentally verified in [16] via a large-scale genomics experiment designed to detect mRNA binding proteins genome-wide . These additional annotations significantly increase the number of potential false negative examples , allowing for greater discrimination between algorithms . Continuing in the same patterns as the entire human genome evaluation , the NETL , SNOB , and Rocchio algorithms perform similarly , and significantly better than the random baseline , with SNOB edging out the other two algorithms for larger numbers of negative predictions . Both NETL and Rocchio , however , maintain a zero false negative rate for a larger number of predicted negative examples than SNOB . AGPS and ALBNeg do well , but only provide a small number of negative examples , and both predict one false negative while NETL and Rocchio achieve zero errors at the same number of negative examples . The “1-DNF” algorithm performs very poorly on this category . In order to further explore the potential biases in the evaluation of negative example selection methods , we include evaluation on a gold-standard set of annotations in yeast , obtained from [17] . This golden set , for the biological process term GO:0007005 Mitochondrial Organization , represents an exhaustively verified set of annotations , such that all positive and negative occurrences of this GO term are known across the entire yeast genome . Because the number of true positives and negatives is known , this GO term in yeast allows us to utilize cross-validation on the data to calculate a Receiver-Operator Characteristic ( ROC ) curve or point for each algorithm . While cross-validation is problematic in the evaluation of function-prediction in general , due to the interconnectedness of GO and many types of feature data which introduces large positive bias into the evaluation , here we are examining and holding out only GO terms , and so such bias is mitigated . In the yeast golden set , we see similar results ( presented in Figure 3 ) as in our evaluation with human data: The SNOB algorithm is the strongest performer , followed closely by the Rocchio and NETL algorithms . The ALBNeg algorithm also performs well , achieving zero false assertions of negative functionality with a large number of predicted negative examples ( 473 . 2 on average ) . The 1-DNF algorithm also achieves zero false assertions of negative functionality , but with fewer predicted negative examples ( only 76 . 6 on average ) , and the AGPS method predicts fewer negative examples than ALBNeg , with a much higher number of false negatives ( 2 . 6 on average ) . It is also worth noting that 59 of the 4625 negative examples in the golden set had received positive annotations for GO:0007005 in the years since the golden set was formed ( the annotations set is updated accordingly here ) . In order to demonstrate the importance of high quality negative examples , we use our previously published algorithm [5] to predict functions across all three branches of GO , for human , mouse , and yeast proteins . We validate these predictions with a temporal holdout ( see methods ) , which enables us to compute the area under the curve ( AUC ) for the Receiver-Operator-Characteristic ( ROC ) plot . We repeat this process using negative examples selected by each of the best-performing negative-example-selection methods , as well as with random negative examples to serve as a baseline . Results are presented in Figure 4 . Comparing the average AUC_ROC values of function prediction with the negative examples selected by each method , we see relative performance very similar to our earlier evaluation of negative example quality . All three of the negative-example-selection algorithms yield much stronger function prediction performance than when negative examples are selected randomly from proteins lacking the positive example . Between the three algorithms , performance is fairly similar , with function prediction utilizing the SNOB negative examples slightly outperforming the other methods . We apply our SNOB algorithm to the work of Puelma et al . [14] , which employs discriminative local subspaces in gene expression networks to predict function in Arabidopsis thaliana . We choose this work as a case study because the authors specifically mention the importance of negative examples in their work , and devise an algorithmic approach for selecting high-confidence negative examples for the 101 biological process categories they used to test their PFP method . We use their provided data to select negative examples with SNOB , generating the same number of negative examples per category as the author's original algorithm ( a total of 313592 across all categories ) . Table 1 shows the results of our case study , demonstrating that even though our algorithm only had access to 1/3 of the data it usually requires ( here the authors provided only Biological Process data , and no data from the other two branches of GO ) , SNOB produces significantly fewer false negatives , negative examples with greater specificity , and performs better when evaluated by the metric chosen by the authors . It is also interesting to note that even though the rate of false negatives is very small ( originally only 0 . 6% ) , further reduction still produces performance gains in downstream function prediction . We have collected negative example predictions from the SNOB , NETL , and Rocchio algorithms in an online database for use by other researchers . While the NoGO database uses the most current annotations for its ranking of negative examples , we have also included false negative rates for each species in the database , obtained from temporal holdouts on older data , to allow researchers to have a reference for the quality of negative examples in that organism . We describe the quality by the area under the false negative curve , as a percentage of the area under the random baseline curve , allowing the number of negative examples to range up to 20% of the size of the genome of that organism . Results are presented in Figure 5 . SNOB and Rocchio achieve the lowest overall errors , with the performance gap between Rocchio and NETL larger than in our other evaluations ( see figure S2 for performance broken down by organism ) . The reduction of the performance gap between NETL and Rocchio in Figure 5b as compared to Figure 5a , indicates that while Rocchio performs better on more general categories , NETL's performance is on par with or better than Rocchio for the more specific GO terms ( and thus a greater number of GO terms ) . It is also interesting to note that across all organisms , SNOB and Rocchio perform similarly on cellular component terms , SNOB has stronger performance on molecular function terms , and Rocchio performs better on biological process terms , suggesting systematic differences in the way that GO annotations relate to each other within each of the three branches . Our Web interface to the NoGO database provides a plot for each GO function that shows the number of false negative predictions as a function of the number of negative examples chosen ( Figure 2 is an example of such a plot , for GO:0003723 ) . This allows researchers to make an informed decision about which algorithm to use for their specific organism , GO terms , and task . These plots also allow researchers to determine how many negative examples to use for each category ( see methods ) .
We have demonstrated ( using the human , yeast , and A . Thaliana proteomes ) that the SNOB algorithm achieves significantly lower prediction errors when predicting negative examples than several previously described alternative approaches ( including heuristics , techniques borrowed from PU-learning in text classification , and other negative-example prediction algorithms ) . These results , supported by additional literature that has explored the inter-relationships between GO categories [19] , [20] , indicate that despite lacking a significant number of negative annotations , the GO database encodes implicit information about likely negative examples via its positive annotations . Additionally , these pairwise term implications span all three branches of GO ( cellular component , biological processes and molecular function ) . Despite the success of our approach , there will inevitably be cases where the information from GO alone is not enough to predict a good set of negative examples . So-called “moonlighting” proteins , for example , can have unique combinations of functions that defy conventional annotation patterns . Additionally , approaches that rely on existing GO annotations are limited to proteins that have already been studied to some extent , which in many organisms can be a relatively small proportion of the genome . For these reasons , our group is considering active methods that can incorporate additional data types ( such as gene expression , protein-protein interaction , domain structure , etc . ) . The algorithms presented here represent a significant improvement over the active 2-step AGPS method that has access to data outside of GO . Our SNOB algorithm achieved a lower false negative rate than any other comparison algorithm tested , significantly lower than the “1-DNF” algorithm that served as its conceptual basis . Through our case study in Arabidopsis , SNOB also demonstrated its ability to improve existing function prediction algorithms . Youngs et al . 2013 [5] showed that even a moderate increase in the quality of negative examples has the power to improve function prediction in general , and those results are replicated here by our case study in human , mouse , and yeast . We have shown the ability of high quality negative examples to improve function prediction accuracy , again with the SNOB algorithm achieving the best results . Additionally , this case study represents a very basic use of these negative example methods , and we believe even further accuracy can be gained by more careful selection of the number of negative examples chosen for each prediction task . Further work includes the incorporation of additional data types , and potentially the use of active 2-step PU methods . Another potentially fruitful avenue is the explicit incorporation of the GO hierarchy in a negative example method . While GO annotations obey the “true path rule” , meaning that every protein with an annotation a also implicitly has all annotations which are ancestors of a , negative annotations follow the inverse of this rule: a protein p that is a negative for g is also implicitly a negative for all descendants of a . This rule holds for the molecular function branch of GO , but is more complex in the biological process and cellular component branches , as there is more than one type of ancestry ( terms may be direct descents , or connected by a “part-of” link , for example ) . These differences most likely account for some of the systematic performance differences of different algorithms on each branch of GO across all the organisms in the NoGO database . These systematic performance differences across branches , combined with the fact that our GO-term specificity effects algorithms' relative performance , suggest the potential utility of ensemble methods ( a combination of methods that use one of multiple algorithms depending on a GO term's specificity , placement in the tree , and desired size of the negative class ) . It is quite natural to think that the optimal algorithm will be quite different for predicting rare functions ( functions that with only a handful of examples of per genome ) and common functions ( like information processing proteins that have hundreds of paralogous examples per genome ) . Further exploring the differences between the performance of NETL and SNOB for rare and common functions separately is likely to result in improved performance via hybrid methodologies . In conclusion , we have presented a significant step forward in the calculation of negative examples for protein function prediction . Following the example set for negative protein-protein interactions by the Negetome database [21] , we have made our predictions readily available for a variety of organisms . Our NoGO database also includes useful statistics to allow researchers to choose the number of desired negative examples and the likely false negative rate of those examples when used in their own experiments and algorithms .
Data for the human genome was obtained from the GO database archive , with training annotations obtained from October 2010 and validation annotations from October 2012 . The set of genes was obtained from HUGO by selecting all protein-coding gene symbols , resulting 19060 genes . GO terms for these genes were gathered by querying all official symbols for all annotations that have at least one annotated protein in the human genome , resulting in 7432 biological process categories , 2681 molecular function categories , and 997 cellular component categories . GO terms are fully propagated according to the “True Path Rule” , meaning that an annotation of a protein to a particular term also implies annotations to all ancestral terms . For the RNA Binding term example , there were 686 positive annotations ( including IEA ) in our training data , and with an additional 157 annotations added in temporal holdout validation data . To these 157 new annotations , we added an additional 381 annotations , which were obtained from [16] , but are not yet present in GO . This raised the total of potential false negatives to 538 . For the case study in Arabidopsis Thaliana , all data was obtained from the supplementary materials provided by [14] . Annotation data for the GO:0007005 golden set in yeast was obtained from [17] , with training GO annotations obtained from the GO ontology in April 2013 . The yeast annotations were taken for the same set of genes as the original positive and negative classes defined in [17] , comprised of 4966 unique yeast gene symbols , with annotations in 4226 biological process categories , 2231 molecular function categories , and 820 cellular component categories . Data for the NoGO database was obtained from GO for each organism , with training data for the negative examples collected in April 2013 , and training data for the validation plots collected in October 2011 and validated with the April 2013 data . The gene sets for each organism were also obtained from GO , by extracting all unique official gene symbols within that organism which had at least one annotation in any branch of GO . Table 2 lists the number of genes and GO categories for each organism , as well as the NCBI Taxa ID for each specific species used . In order to generate the validation plots in Figure 1 and Figure S1 , we plot the average number of false negatives as a function of the number of negative examples . For algorithms that allow the specification of the size of the negative class , we sample the number of false negatives at 100 , 200 , 500 , 1000 , 2000 , and 3000 negative examples . The average number of false negatives is determined using the temporal holdout , by seeing how many proteins that were designated as negative received an annotation in the function in question ( including an IEA annotation ) . Functional categories that received no new annotations during the temporal holdout are not evaluated , nor are categories with fewer than 3 or more than 300 annotations . Plots are broken down by branch of the GO hierarchy , with each plot showing an average of the results for functions in that branch that meet the specified criteria . The plot for Figure 2 is identical in construction , but for one specific GO category , rather than an average over GO categories . The plots in Figure 5 and Figure S2 are three representations of algorithmic performance on all organisms in the NoGO database , and each organism , respectively . The leftmost graph was generated by sampling the number of false negatives at negative class sizes equal to 0 . 1% , 0 . 5% , 1% , 2 . 5% , 5% , 10% , 15% and 20% of the size of the genome of the organism in question . This value is then turned into a single number by computing the area under the sample curve for each algorithm , and for the random baseline . These numbers are summed over all categories in the organism ( or in the case of Figure 5 across all categories in all organisms ) , and then divided by the number obtained from the random baseline . The central graph is calculated identically , except here the area under the curve for each algorithm is divided by the random baseline area before being summed over all categories , meaning that each GO category contributes equally to the score , regardless of the number of annotations in that category . The rightmost graph represents the total false negative rate , over all GO categories in each branch , when predicting a number of negative examples equal to the number of positive annotations for that GO category . All false negative statics are obtained via a temporal holdout . Note that in the plots for performance in the NoGO database , it is possible for algorithms to appear worse than the random baseline . This is due to the fact that the random baseline chooses from all possible unlabeled proteins , whereas the algorithms are constrained to only those proteins with GO annotations . Since it can often be the case that new annotations in the temporal holdout set are concentrated among proteins that are already partially annotated , the GO-restricted algorithms are penalized over the random baseline . The Selection of Negatives through Observed Bias algorithm takes as its basis the pairwise conditional probability calculation of seeing annotation a given the presence of annotation m , which is specified for the ALBias algorithm in Youngs et al . , 2013: , where is the number of genes where m appears alongside g in the dataset , and is the total number of genes annotated with m in the dataset . As mentioned in the results , SNOB removes the restriction that the score is calculated from leaf annotations only , or that a protein must have an annotation in the same branch as the GO term in question in order to be chosen as a negative . In addition , all annotations are utilized , including IEA annotations . The score vector , which holds the scores for all genes as potential negative examples for a given GO function a , is calculated as the average of the conditional probabilities of all other annotations in each gene , which is efficiently calculable as: , where A is the annotation matrix of the dataset , with each row representing a gene and each column a GO category , W is the diagonal matrix with Wii equal to the total number of annotations for protein i , and P is the conditional probability matrix with . These scores are then ranked to produce a list of negative examples , with the lower scores indicating higher probability that a particular protein is a negative example for the GO term in question . For the Negative Example from Topic Likelihood algorithm , we again formulate a protein as a document , with GO annotations ( including IEA ) from all three branches as the words in that document . We then run Latent Dirichlet Allocation ( Code obtained from David Blei's “lda-c” package ) on the document corpus to identify the parameters of the Dirichlet topic distribution , and perform inference on each document to obtain the posterior topic distribution given the GO terms present in that protein ( See [18] for the details of LDA ) . Ideally , we would set the number of latent topics t equal to the number of GO categories m , but this choice yields infinite perplexity in the corpus , as the number of unique words w = m as well . In order to achieve w >> t , to increase the quality of the learned topics , yet also to preserve coverage of all GO categories , we set the number of topics for each organism equal to the total number of annotated direct descendants of the root ontology terms . For example , in our Human validation data , the biological process node has 27 direct descendants with annotations in the data , the molecular function node has 14 direct descendants , and the cellular component node has 10 , for a total of 51 latent topics . By invoking the inverse of the true path rule , whereby negative examples are propagated downwards through the GO graph , this approach guarantees coverage of all GO categories for the purposes of negative example selection . Since LDA discovers latent topics , which are not predefined before the algorithm is run , it is not immediately obvious which learned topic corresponds to which GO term . Indeed despite our efforts to ensure coverage of every GO category directly descended from a root node , it is not necessarily the case that the correspondence between the topics and the selected GO terms are 1–1 . Instead it is possible , even likely , that some combinations of topics/GO terms relate to each other , making exact inference of the probability that a given protein possesses a given GO term difficult under the LDA model . To overcome this problem , we chose to represent the positive class with the average of the Dirichlet posterior vectors for all proteins annotated to the function in question ( including IEA annotations ) . Then for each unlabeled protein u , we calculate a Distributional-Overlap Score ( DOS ) representing the similarity of topics distributions between u and the positive class average topic distribution . This score can be viewed as a symmetric simplification of the Kullback-Leibler Divergence metric , and is calculated simply as , where and are two Dirichlet posterior parameter vectors ( since each posterior vector sums to 1 , the DOS score is also bounded by [0 , 1] ) . The unlabeled proteins are then ranked according to this score , with the lowest DOS values indicating the most likely negative proteins , as these are proteins which are least likely to share topics with the positive class of proteins . In order to calculate the random baseline , we consider the positive class to be all proteins with an annotation in the function of interest ( including an IEA annotation ) , and all other proteins to be the unlabeled class . We sample uniformly at random without replacement from those unlabeled proteins in order to pick negative examples , allowing the user to specify the desired size of the negative class . In order to reduce noise from this stochastic operation , we calculate the baseline 100 times for each branch of GO , and then display the average of those 100 calculations . In order to adapt the Rocchio algorithm to protein function , we follow the pseudocode in [12] , treating the set of GO terms across all three branches as our lexicography , each protein as a document , and the annotations of that protein as a word . This formulation allows the computation of the tf-idf vectors required by the algorithm , and for each function we treat the positive class as all proteins with an annotation in that function ( including IEA ) , and the rest of the proteins as the unlabeled class . The algorithm then builds a representative vector for the positive and unlabeled class , and computes the cosine similarity of the tf-idf vector for each unlabeled protein with each of the representative vectors . Where the traditional algorithm would assign as negative examples all proteins whose similarity to the unlabeled class vector is greater than to the positive class vector , we assign a score to each protein , defined as: UnlabeledSimilarity – PositiveSimilarity . This allows us to rank the proteins in terms of confidence of their negativity , with the highest-scoring proteins as the most likely to be negative examples . For the 1-DNF algorithm , we again formulate proteins as documents and GO terms across all three branches as words . We proceed according to the pseudocode laid out in [4] , utilizing as the positive class all proteins with an annotation in the function of interest ( including IEA ) . Other GO terms that appear more frequently in the positive set than the unlabeled set are considered our “enriched” words , and negative examples are all proteins that are not in the positive class and do not contain any of these enriched words . As there is no immediately obvious way to translate this decision into a score , we only implemented this algorithm for one choice of the number of negative examples , rather than thresholding it to allow the user to specify the desired size of the negative class . Code for the AGPS algorithm was generously provided by the authors of [7] . AGPS requires features to operate , which we obtained through the similarity networks provided by the Genemania server [22] . Each of these networks ( 235 networks for human , 297 for yeast ) represents similarity between pairs of genes according to a particular datatype . For human data it was necessary to translate the networks from being specified by ENSEMBL ids to gene symbols by using the HUGO lookup for gene symbol and ENSEMBL pairs . For both yeast and human , we performed a simple linear combination of all of the networks , where each component network and the final network was normalized according to the scheme: , where D is the diagonal row sum matrix of W . Once the final network was obtained ( a 19060×19060 matrix for human , 4966×4966 for yeast ) , we applied Principal Component Analysis to reduce the feature size to a 19060×200 matrix and a 4966×200 matrix , which were the input feature sets for AGPS for each organism , respectively . We ran the algorithm provided by the authors using all of the default constants provided , but as described in the author's text , ran cross-validation for each category and only used negative examples that were chosen in the majority of the cross validation runs . We choose to segment data into 5 cross-validation segments . AGPS was only validated on functional categories with at least 85 annotations ( the reliance of the method on cross-validation increases the number of necessary positive examples for a meaningful result ) . The lengthy runtime of the algorithm also restricted our application of the method to function categories with more than 85 annotations . To allow for a fair comparison to other methods we utilized the inverse of the true path rule , and for GO functions with fewer than 85 annotations in the human genome , we set the negative examples as the union of all of the negative examples of all parent categories of that GO term . For the heuristic that chooses siblings as negatives for a function , we follow the specification laid out in [11] , whereby a protein is a negative for a function if it is annotated to the parent of that function , but not to the function itself . This includes proteins annotated to sibling categories , as well as those annotated to the parent but to none of the children of that parent . Because some function categories will have no proteins that satisfy these requirements , we revert in this case to the strategy of choosing all non-positive proteins as negative , where the positive class is all proteins with an annotation in the function in question ( not including IEA annotations ) . As Mostafavi 2009 points out , the sibling approach is problematic in that many sibling categories are not mutually exclusive , but we present the technique here for completeness . Since the heuristic will produce different numbers of negative examples for different function categories , the point on the validation plot corresponding to this algorithm represents an average over different sizes of the negative class . For function prediction , we used our previously published algorithm [5] . Training GO annotations were obtained from the GO archive in April 2013 , with validation annotations obtained in December 2013 . Input data included protein-protein interaction , Interpro database data [23] , gene expression data , sequence similarity , and phylogenetic profiles . Predictions were made for all terms in all three branches , regardless of specificity , but validations were calculated only for those terms that received new annotations during the temporal holdout period . For each term predicted , the number of negative examples was selected to be the maximum of the number of positive examples of that term , or 20% of the size of the genome . A further restriction capped the number of negative examples at 50% of the number of non-positive genes for the function in question . The area under the curve of the Receiver Operator Characteristic plot was calculated using the methodology presented in [5] . Negative examples are available in the NoGO database , located at: bonneaulab . bio . nyu . edu/nogo . html . Negative examples are currently available for the following species: Human , Mouse , Yeast , Rice , Arabidopsis and Worm . For each function in each organism , a ranked list of genes shows the most to least likely negative examples , available for the SNOB , NETL , and Rocchio algorithms described here . All negative examples were computed using GO data from April 2013 . Accompanying each list is a validation plot ( See Figure 2 for a sample , GO:0003723 in Homo Sapiens ) , which shows the performance of SNOB against a random baseline , trained on GO data obtained from October 2012 and validated with data from April 2013 . This plot gives a researcher an idea of the relative performance of the SNOB algorithm against the random reference , in order to give confidence as to the likelihood of false negatives , and also allows a researcher insight into how many negative examples to choose based on the false negative rate presented in the graph . MATLAB code for generating negative examples from custom data is also be available from the downloads section of the NoGO database , as well as directly from: http://markula . bio . nyu . edu:8080/downloads . The database will be updated with negative examples computed from new GO annotations in April 2014 , and then subsequently every three months .
|
Many machine learning methods have been applied to the task of predicting the biological function of proteins based on a variety of available data . The majority of these methods require negative examples: proteins that are known not to perform a function , in order to achieve meaningful predictions , but negative examples are often not available . In addition , past heuristic methods for negative example selection suffer from a high error rate . Here , we rigorously compare two novel algorithms against past heuristics , as well as some algorithms adapted from a similar task in text-classification . Through this comparison , performed on several different benchmarks , we demonstrate that our algorithms make significantly fewer mistakes when predicting negative examples . We also provide a database of negative examples for general use in machine learning for protein function prediction ( The NoGO database , available at: bonneaulab . bio . nyu . edu/nogo . html ) .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods",
"and",
"Materials"
] |
[
"algorithms",
"computer",
"and",
"information",
"sciences",
"mathematics",
"applied",
"mathematics",
"biology",
"and",
"life",
"sciences",
"physical",
"sciences",
"computational",
"biology"
] |
2014
|
Negative Example Selection for Protein Function Prediction: The NoGO Database
|
Almost 60 years ago , Severo Ochoa was awarded the Nobel Prize in Physiology or Medicine for his discovery of the enzymatic synthesis of RNA by polynucleotide phosphorylase ( PNPase ) . Although this discovery provided an important tool for deciphering the genetic code , subsequent work revealed that the predominant function of PNPase in bacteria and eukaryotes is catalyzing the reverse reaction , i . e . , the release of ribonucleotides from RNA . PNPase has a crucial role in RNA metabolism in bacteria and eukaryotes mainly through its roles in processing and degrading RNAs , but additional functions in RNA metabolism have recently been reported for this enzyme . Here , we discuss these established and noncanonical functions for PNPase and the possibility that the major impact of PNPase on cell physiology is through its unorthodox roles .
In bacteria and eukaryotes , PNPase has an important function in RNA decay . PNPase catalyzes the processive degradation of single-stranded RNA in the 3ʹ to 5ʹ direction using inorganic phosphate as the nucleophile to attack the 3ʹ phosphodiester bond releasing a ribonucleoside diphosphate ( NDP ) . A metal cofactor , Mg2+ or Mn2+ , is required to stabilize the transition state of the phosphate during the reaction [30] . To degrade RNA efficiently , PNPase must bind a single-stranded stretch of RNA at least six nucleotides in length at the 3ʹ terminus [31 , 32] . PNPase subsequently degrades RNA in a stepwise motion , rapidly removing discrete segments of 6 to 7 nucleotides between short pauses [33] . A stem loop structure in an RNA substrate can act as a roadblock that halts degradation by PNPase . In vitro studies using a set of GC-rich RNA hairpins of varying length followed by single-stranded sequences demonstrated that a hairpin with a stem as short as seven base pairs inhibited degradation by PNPase [32] . However , PNPase also rapidly degrades some natural hairpins such as the Rho-independent terminators that end many sRNAs [34] . Thus , while very stable hairpins can block the exoribonucleolytic activity of PNPase , merely resulting in 3ʹ end trimming [35] , this enzyme can degrade many natural double-stranded RNAs provided that a single-stranded region is present at the 3ʹ end to initiate degradation . This balance between degradation and inhibition is important for maturation of some tRNA transcripts , in which PNPase removes Rho-independent terminators but stops short of the CCA determinants [36–39] . Likewise , structural features of the 16S rRNA preribosomal particle are likely important for preventing excessive 3ʹ trimming of the 16S rRNA by PNPase and other exoribonucleases during its maturation [40 , 41] . Crystal structures for PNPase from Streptomyces antibioticus [42] , E . coli [30 , 43] , Caulobacter crescentus [44] , and Homo sapiens [45] have been solved , and much of the substrate specificity and activity of PNPase can be assigned to distinct structural and organizational features of this enzyme . Each PNPase monomer consists of two RNase PH-like domains , which are separated by an α-helical domain and are followed by a KH and S1 domain ( Fig 1A ) . As a functional enzyme , PNPase is assembled into a torus-shaped trimer in which alternating RNase PH-like subunits and α-helical domains form a central ring from which the KH/S1 domains extend outward ( Fig 1B and 1C ) [30 , 42–44] . The first RNase PH-like domain contributes to RNA and NDP binding , and the second domain additionally possesses enzymatic activity [43 , 46] . The active site is positioned in a shallow groove along the inner rim of the trimer ( Fig 1D , panel iii ) , and a constriction point formed by the FFRR loop at the entrance of the core ring creates a pore that only allows access by single-stranded RNA ( Fig 1D ) [43] . Catalytic activity of the central ring is facilitated by the other domains; the α-helical domain appears to regulate access of phosphate or NDP to the active site [30 , 47 , 48] , and the KH and S1 domains each contribute to capturing and binding RNA substrates [44 , 49] . Additionally , the KH domain imparts RNA directionality through the interactions of the conserved GSGG loop with the RNA backbone [44] . Finally , the processivity of PNPase is attained through its ring-like structure that retains RNA substrates via multiple RNA-binding interactions , including hydrogen binding between the GSGG loops of the KH domains and the RNA phosphate backbone ( Fig 1D , panel i ) and base stacking interactions between the aromatic phenylalanines in the conserved FFRR loops and a ribonucleotide base ( Fig 1D , panel ii ) [44] . While PNPase can function independently as a 3ʹ to 5ʹ exoribonuclease , in bacteria , PNPase also serves as a component of an organized RNA degradation machine ( Fig 2A ) . Termed the degradosome , this multiprotein complex is responsible for bulk mRNA decay [50 , 51] . At the core of this RNA-degrading machine in gram-negative bacteria is the endoribonuclease RNase E , which initiates RNA decay . The essential N-terminal domain of RNase E contains the active site and additional features including the S1 domain and 5ʹ sensor pocket important for binding many RNAs ( reviewed in [52] ) . The C-terminal domain is required for formation of the RNA degradosome and contains binding sites for other proteins; in the canonical E . coli RNA degradosome , these proteins include the glycolytic enzyme enolase , the DEAD-box RNA helicase RhlB and PNPase [53–55] . However , the RNase E–based degradosome can vary in composition between species or even within the same organism depending on cellular conditions [56] . In C . crescentus , aconitase is exchanged for enolase , and RNase D was validated as a legitimate degradosome component [57 , 58] . Gram-positive bacteria have a similar RNA degradation machine that is likewise organized around a core endoribonuclease , in this case , RNase Y . Like RNase E , RNase Y has an unstructured C-terminal region with specific binding sites for PNPase , enolase , and an RNA helicase [59] . However , despite these functional similarities , the two proteins are evolutionarily distinct , belonging to different protein superfamilies [60] . Unlike RNase E , RNase Y features an N-terminal region with an RNA-binding KH domain and an active site–containing HD domain , and its C-terminal domain additionally interacts with 5ʹ to 3ʹ exoribonucleases [60 , 61] . PNPase and the RNA degradosome of the gram-negative bacterium E . coli have been studied in much detail . As part of the degradosome , PNPase cooperates with RNase E to degrade a specific set of mRNAs , many of which encode proteins involved in macromolecule biosynthesis or modification [50] . Additionally , within the degradosome , binding of RhlB and PNPase to the RNase E scaffold is necessary for the degradation of highly structured RNA sequences , termed repeated extragenic palindrome ( REP ) elements that are found in some mRNAs [62–64] . Although PNPase participates in mRNA decay as a component of the RNA degradosome , the enzyme primarily functions independent of this machine . This is evident by the number and distribution of PNPase and RNase E molecules in E . coli . PNPase is approximately three to five times more abundant than RNase E and is mostly distributed throughout the cytoplasm ( 69% of PNPase; Fig 2A ) , whereas the majority of RNase E ( 91% ) is located near or on the cell membrane [65] . Moreover , only a minority of E . coli PNPase trimers are bound to RNase E at any one time [53] . The independent function of PNPase is also evident by global gene expression profiling , which demonstrated that many mRNAs stabilized by the absence of PNPase are not significantly impacted by loss of the degradosome [50] . Furthermore , PNPase functions independently of the RNA degradosome in tRNA processing [37 , 38 , 66] , rRNA degradation [40 , 67] , and sRNA-mediated gene regulation [68] . In bacteria , the fully assembled 70S ribosome is made up of the small 30S subunit containing the 16S rRNA and the large 50S subunit consisting of the 23S rRNA and 5S rRNA . In E . coli and presumably most gram-negative bacteria , PNPase and RNase R perform routine rRNA quality control by degrading fragments of 16S and 23S rRNAs that might otherwise compete with mature rRNAs for ribosomal proteins and impair proper ribosomal assembly [69 , 70] . These rRNA fragments are generated by RNase E [70] , which may cleave rRNAs that cannot be assembled into functional ribosomes due to improper processing , damage , or overabundance relative to ribosomal proteins . In E . coli , PNPase was not required for rRNA decay induced by nutrient starvation [71] . In some bacteria such as the radiation-resistant D . radiodurans , PNPase mediates rRNA degradation in response to nutrient starvation with the assistance of the Ro sixty-related protein , Rsr ( Fig 2A ) [72 , 73] . Ro was originally identified in human cells as an autoantigen recognized by antibodies from lupus erythematosus patients [74] . Ro and its homologs , which are present in many vertebrates and in roughly 5% of bacterial genome sequences [75] , bind to structured noncoding RNAs called Y-RNAs [13 , 72 , 74 , 76 , 77] and are involved in rRNA decay [73 , 78 , 79] . In D . radiodurans , Y-RNA tethers Rsr to PNPase resulting in the formation of the “RYPER” complex ( Fig 2A ) that degrades structured RNA including the 5S , 16S , and 23S rRNAs [72 , 73] . Y-RNAs have a highly conserved structure that consists of an extended stem generated by pairing between bases at the 3ʹ and 5ʹ ends and a large internal loop that in many cases is decorated with two hairpins [80 , 81] . A conserved region within the Y-RNA stem binds to Rsr , whereas a region containing two hairpins , one resembling a T-arm of a tRNA , binds to the KH and S1 domains of PNPase [72 , 82] . The 3ʹ end of the misfolded rRNA appears to thread through the central pore of the toroid-shaped Rsr protein and into the central channel of PNPase , where it is degraded [72] . Although for many years it was largely assumed that PNPase only contributed to RNA processing and degradation in bacteria , it has become increasingly clear over the last decade that PNPase also plays an important role in regulating sRNA function and stability [68] . In bacteria , sRNAs range in size from 50 to 250 nucleotides and alter gene expression by sequestering regulatory proteins or by base-pairing with target mRNAs to modulate translation and transcript stability ( reviewed in [83 , 84] ) . Many sRNAs interact with RNA chaperones , such as FinO [85] , ProQ [86 , 87] , or the host factor for phage Qβ ( Hfq ) [88 , 89] , to facilitate this process . In the latter case , Hfq protects sRNAs from degradation by occluding an endoribonuclease cleavage site [90 , 91] and facilitates sRNA–mRNA annealing [92 , 93] . In E . coli and its close relative Salmonella Typhimurium , binding by Hfq dictates whether an sRNA is degraded or stabilized by PNPase . For Hfq-independent sRNAs such as CopA , CsrB , and CsrC , PNPase degrades these RNAs following initial cleavage by RNase E [94 , 95] . Even some Hfq-binding sRNAs such as SraL , RybB , and MicA are destabilized by PNPase [94 , 96]; however , PNPase degrades only the pool of MicA that is not bound by Hfq [97 , 98] . Indeed , PNPase binds and rapidly degrades several Hfq-binding sRNAs in vitro but only in the absence of Hfq [34] . In the presence of Hfq , PNPase instead forms a stable ternary complex with Hfq and sRNAs , and experimental evidence supports the existence of this complex in E . coli [34] . In vivo , PNPase stabilizes many Hfq-dependent sRNAs , and deletion of the gene encoding this RNase paradoxically results in reduced sRNA stability [34 , 68 , 97 , 98] . What is the role of PNPase in facilitating sRNA-mediated gene regulation ? In our speculative model , PNPase forms a complex with Hfq that is mediated by sRNAs ( Fig 2A ) . Within this complex , PNPase is unable to degrade the sRNA because its 3ʹ end is bound to Hfq . After sRNA–mRNA pairing , Hfq is released from the complex and in most cases each RNA is first cleaved by an endoribonuclease ( RNase E ) , followed by rapid degradation of the resulting sRNA and mRNA fragments by PNPase . In the absence of PNPase , specific mRNA fragments accumulate and go on to pair with additional sRNAs resulting in their cleavage by RNase E . By this mechanism , these mRNA fragments act to deplete the pool of specific sRNAs , resulting in decreased regulation of their mRNA targets . Given that Hfq-binding sRNAs in E . coli and other gram-negative bacteria regulate many physiological processes—including DNA repair [99 , 100] , motility [101–104] , biofilm formation [102 , 105–111] , and antibiotic resistance [112–114]—and that PNPase regulates sRNA stability and function , we postulate that the majority of the phenotypes associated with the loss of functional PNPase are due to its role in degrading or stabilizing sRNAs . There is already some evidence supporting this hypothesis . For example , the reduced rate of spontaneous mutation observed for an E . coli pnp deletion strain [115] may originate from reduced ArcZ sRNA levels , as disruption of the negative regulation of mutS by ArcZ also reduces the spontaneous mutation rate in E . coli [100] . Similarly , the role of PNPase in promoting biofilm formation may be due to its function in stabilizing Hfq-dependent sRNAs; recent studies have collectively shown that deletion of pnp , hfq , or genes encoding the sRNAs DsrA or ArcZ from E . coli each resulted in defects in biofilm formation [108 , 116] . Studies of the mammalian PNPase have been fraught with controversy , and many functions have been reported for the human PNPase ( hPNPase ) , including mitochondrial RNA import , processing , and decay and miRNA and mRNA degradation ( Fig 2B ) . Careful studies mapping the cellular location of hPNPase indicate that it mainly resides in the mitochondrial inner membrane space ( IMS ) located between the outer and inner membranes [117 , 118] . hPNPase is guided to the IMS via a mitochondrial targeting sequence that is cleaved off when it is translocated into the IMS [118] . Upon overexpression , hPNPase accumulates in other cellular compartments such as the cytoplasm [119] , but conditions in which the natively expressed hPNPase is found in this space have not been identified until recently . A newly published study revealed that natively expressed hPNPase can also be released into the cytoplasm upon mitochondrial outer membrane permeabilization during programmed cell death , whereupon hPNPase contributes to global apoptotic RNA decay by degrading mRNAs and polyadenylated noncoding RNAs ( Fig 2B ) [120] . Within the IMS , PNPase is a peripheral membrane protein that reportedly binds the 5S rRNA and the RNase P and RNase MRP RNAs to facilitate importation of these RNAs into the mitochondrial central space , or matrix [25] . This import function of PNPase did not appear to require its catalytic activity , and intriguingly , a 20-nucleotide stem loop structure found in the RNase P and MRP RNAs was sufficient for PNPase-mediated mitochondrial import [25] . In addition , Wang and colleagues [25] found that processing of polycistronic tRNA transcripts in vivo required the RNase P RNA . These results appear to conflict with previous work showing that the MRP RNA was undetectable [121] or at infinitesimal levels [122] in HeLa cell mitochondria , that only a very small number of RNase P RNA molecules were associated with the mitochondria of HeLa cells [122 , 123] , and that a reconstituted mitochondrial RNase P lacking an RNA component was functional in processing mitochondrial precursor tRNAs [124] . As argued by Wang and colleagues [25] , it is possible that RNase P exists in mammalian mitochondria in both the protein-only and H1 RNA-containing forms and that the RNA-containing form of RNase P is much less abundant . Both RNases serve critical roles in mitochondria , in which RNase MRP cleaves the RNA primers used for mitochondrial DNA replication and RNase P processes the large mitochondrial polycistronic transcripts that give rise to 22 tRNAs , 12S and 16S rRNAs , as well as 13 mRNAs encoding electron transport chain ( ETC ) components involved in oxidative phosphorylation ( reviewed in [125] ) , i . e . , the synthesis of ATP that is powered by the transfer of electrons from NADH or FADH2 to O2 . Likewise , a stable PNPase knockout in mouse embryonic fibroblasts resulted in the loss of both mitochondrial DNA and cellular respiration , supporting a role for PNPase in mitochondrial DNA maintenance [126] . The vital function of hPNPase in facilitating proper expression of the ETC components is further evidenced by the fact that knockdown of PNPase in HEK293 cells leads to impairment of the ETC and disruption of oxidative phosphorylation [117] . Additionally , several recent clinical reports demonstrate that patients suffering from hereditary hearing loss , delayed myelination , axonal neuropathy , and Leigh syndrome have mutations in PNPT1 , the gene encoding hPNPase [26–29 , 127] . In several of these reports , the authors provided evidence that the hPNPase encoded in these patients’ genomes contributes to a defect in oxidative phosphorylation and mitochondrial RNA import [27–29] . hPNPase also catalyzes mitochondrial RNA decay with assistance from the suppressor of Var 1 , 3 ( SUV3 ) RNA helicase [128–130] . The involvement of hPNPase in this process requires that it associate with the SUV3 helicase in the mitochondrial matrix . Consistent with some hPNPase binding SUV3 in the mitochondrial matrix , PNPase coimmunoprecipitated with SUV3 from mitochondrial cell extracts and foci of exogenously produced hPNPase and SUV3 colocalized with mitochondrial DNA and RNA [128 , 130] . Furthermore , knockdown of hPNPase in HeLa or T-Rex 293 cells resulted in stabilization of mitochondrial mRNAs [128 , 129] , and depletion of hPNPase or SUV3 led to accumulation of mitochondrial double-stranded RNA [131] . hPNPase also facilitates degradation of the c-myc mRNA [132 , 133] and miRNAs including miR-221 , miR-222 , and miR-106b in vitro and in vivo upon overexpression [134] . miRNAs are a class of sRNAs in humans that regulate gene expression by base-pairing with target mRNAs ( reviewed in [135] ) . However , to degrade these RNAs , hPNPase must reside in the cytoplasm , but this has only been shown to occur during apoptosis or upon exogenous overexpression in human cells [119 , 120 , 136] . Thus , the role of PNPase in degrading these RNAs may not be relevant under most physiological conditions . Aside from this potential degradation role , PNPase also appears to facilitate the import of miR-378 into mitochondria , resulting in down-regulation of the mt-ATP6 transcript and a reduction in ATP synthase activity [137] .
Although PNPase has been studied for over 60 years , new functions for this old enzyme have been recently uncovered . In bacteria , a novel function for PNPase in degrading some sRNAs and protecting others has been discovered [68 , 94 , 138] . Because each sRNA can potentially regulate hundreds of distinct transcripts , PNPase impacts many , if not most , regulatory circuits in bacterial cells . Therefore , we postulate that the vast majority of phenotypes associated with loss of PNPase function in bacteria are due to its role in mediating sRNA stability . Equally exciting were the discoveries that hPNPase mediates the importation of RNA into the mitochondrial matrix [25] and degrades mRNAs and polyadenylated noncoding RNAs upon release into the cytoplasm following mitochondrial outer membrane permeabilization during apoptosis [120] . Given the recent discovery that PNPase is critical for mitochondrial DNA maintenance [126] , hPNPase is vital to the proper replication and function of mitochondria and to human life . Considering that these recent discoveries of additional functions for PNPase were made after more than a half century of study , we expect to see many more exciting findings reported on this ancient enzyme in the years to come .
|
Widely distributed among bacteria and eukaryotes , including humans , polynucleotide phosphorylase ( PNPase ) is a critical enzyme in RNA metabolism that functions in most organisms as a 3ʹ to 5ʹ exoribonuclease . In bacteria , inactivation of the gene encoding PNPase results in a wide range of consequences , including impaired growth , diminished stress responses , and loss of virulence . In mammals , PNPase has an essential role in mitochondrial function . Mutations in the gene encoding the human PNPase ( hPNPase ) that reduce its activity can lead to hereditary hearing loss , encephalomyopathy , severe axonal neuropathy , delayed myelination , and Leigh syndrome . In this review , we highlight both the canonical and unorthodox activities that have been reported for PNPase . Specifically , we examine its role in bacterial mRNA and rRNA decay , RNA processing , and small regulatory RNA ( sRNA ) degradation and stabilization . Furthermore , we explore the recently reported findings on the function of hPNPase in mitochondrial RNA import and degradation and cytoplasmic mRNA and noncoding RNA decay . Despite being discovered more than six decades ago , we are still only beginning to grasp the breadth of mechanisms by which the enzymatic activities of PNPase contribute to cellular and organismal physiology .
|
[
"Abstract",
"Introduction",
"Conclusions"
] |
[
"nucleic",
"acid",
"synthesis",
"non-coding",
"rna",
"sequences",
"nucleases",
"enzymes",
"dna-binding",
"proteins",
"enzymology",
"polynucleotides",
"nucleotides",
"review",
"mitochondria",
"bioenergetics",
"rna",
"synthesis",
"cellular",
"structures",
"and",
"organelles",
"chemical",
"synthesis",
"research",
"and",
"analysis",
"methods",
"genome",
"complexity",
"rna",
"structure",
"genomics",
"proteins",
"ribonucleases",
"molecular",
"biology",
"ribosomes",
"biosynthetic",
"techniques",
"cytoplasm",
"biochemistry",
"rna",
"hydrolases",
"ribosomal",
"rna",
"cell",
"biology",
"nucleic",
"acids",
"genetics",
"biology",
"and",
"life",
"sciences",
"energy-producing",
"organelles",
"non-coding",
"rna",
"computational",
"biology",
"macromolecular",
"structure",
"analysis"
] |
2018
|
Polynucleotide phosphorylase: Not merely an RNase but a pivotal post-transcriptional regulator
|
The fusion of the human immunodeficiency virus type 1 ( HIV-1 ) with its host cell is the target for new antiretroviral therapies . Viral particles interact with the flexible plasma membrane via viral surface protein gp120 which binds its primary cellular receptor CD4 and subsequently the coreceptor CCR5 . However , whether and how these receptors become organized at the adhesive junction between cell and virion are unknown . Here , stochastic modeling predicts that , regarding binding to gp120 , cellular receptors CD4 and CCR5 form an organized , ring-like , nanoscale structure beneath the virion , which locally deforms the plasma membrane . This organized adhesive junction between cell and virion , which we name the viral junction , is reminiscent of the well-characterized immunological synapse , albeit at much smaller length scales . The formation of an organized viral junction under multiple physiopathologically relevant conditions may represent a novel intermediate step in productive infection .
Strategies for antiretroviral therapy have recently focused on inhibiting human immunodeficiency virus ( HIV ) adhesion , fusion , and entry . The biochemical properties of the dynamic binding interactions between host cell and viral receptors have been well characterized [1] , [2] . However , how these interactions may work together for viral adhesion to progress toward an effective fusion event is not well understood . In particular , whether a single viral particle has the ability to spontaneously organize receptors at the cellular plasma membrane is unknown . Viral adhesion occurs on length and time scales that are difficult to monitor in real time because of the limited spatial and temporal resolution of current light and electron microscopes and the small size of virions ( 100nm in diameter , with entry into the cell taking place after only a few minutes [3] , [4] ) . Here we use stochastic modeling to test the fundamental hypothesis that the cell-virus interfacial area forms an organized ultrastructure during viral adhesion , similar to the well-characterized immunological synapse [5] but at much smaller length scales ( i . e . 0 . 1µm versus 10µm , respectively [3] , [6] ) . Virus-cell adhesion is primarily governed by bimolecular bonds formed between the viral surface protein gp120 , and its cellular receptor CD4 [7] . gp120 molecules are arranged on the viral surface in trimers [8] , [9] . For type-1 HIV ( HIV-1 ) , productive infection is also dependent on the subsequent binding of a cellular co-receptor , most commonly CCR5 or CXCR4 [10] . gp120-coreceptor binding induces a dynamic refolding of viral surface proteins which provides the driving force for fusion of the viral and cellular membranes [11] , [12] . The formation of lipid microdomains on length scales similar to the viral diameter have been shown to result in protein colocalization [13] . Here we hypothesize that viral protein adhesion to cellular receptors , coupled with plasma membrane rigidity can produce highly organized protein structures at the cell surface on the same length scales as the virion and lipid rafts [13] . We used stochastic modeling based on recent single-molecule force spectroscopy measurements [14] to assess the spatial and temporal organization of cellular receptor CD4 and co-receptor CCR5 at the plasma membrane , as they dynamically interacted with gp120 on the viral surface ( Fig . 1 ) . We will discuss how the energy of bond formation between gp120 on the virion and receptors on the cell surface acts as an organizing force amidst disordering thermal energy . Specifically , thermal energy drives stochastic movement of the laterally diffusing receptors on the plasma membrane , the formation and destruction of bonds between viral proteins and cellular receptors , and the deformation of the plasma membrane . The conditions that we explore here are designed to determine how each type of stochastic movement contributes to the organization of cellular receptors at the plasma membrane . In addition , we study whether viral particles with gp120 trimers that are capable of diffusing on the viral surface result in distinct receptor organization between the virus and the cell . The combined system of virion , plasma membrane , viral receptors and cellular receptors studied here , is modeled as a succession of discrete states . The transition to new states is assumed to be a Markov process [15] and computed using the local-steady state approximation to the Fokker-Planck equation [16] . The organizations of virus-cell bonds discussed here were produced using the 3-D location of receptors on the plasma membrane actively bound to gp120 molecules relative to the center of the virion itself .
We performed simulations considering the dynamic progress of the junction between the virion and the cell surface to be a stochastic , Markov process [15] . The system itself consists of a rigid sphere ( the 100nm-diameter virion ) interacting with a deformable surface ( the plasma membrane , 200×200nm in dimension ) . The virion is populated with gp120 trimers , which can bind receptors ( CD4 and CCR5 ) located on the plasma membrane ( Fig . 1 ) . All entities within the system including the plasma membrane , the configurations of viral proteins , the position of the virion itself , the dynamics of the bonds between the virus and the cell , as well as the positions of the cellular receptors ( which are either bound or unbound to gp120 ) , are specified by discrete states in Markov dynamics . The probabilities of generating a particular sequence of states , or a trajectory , are governed by the transition rates between these states . The transition rates between states that involve changes in the physical configuration of the system ( i . e . the movement of the virion , the positions of the proteins and the plasma membrane from their current position to each possible new position ) were calculated according to the local-steady state approximation of the Fokker-Planck equation [16] , Here , is the forward association rate for the possible state adjacent to the current state ( ) , is the diffusion coefficient of each physical entity of the system ( membrane proteins , viral proteins , etc . ) whose position has changed between the current and future state , is the change in position of that physical parameter , is the difference in system energy which results from the movement of that physical parameter between states and is where is Boltzmann constant and is the absolute temperature . This approximation calculates the rate of transition between the current state and subsequent adjacent state using the total change in energy that accompanies the progression from one discrete state to another . Specifically , a favorable change in energy ( e . g . the relaxation of a bond between viral proteins and cellular receptors ) results in an increased transition rate and an unfavorable change in energy ( e . g . the deformation of the plasma membrane ) results in a decreased transition rate from one state to the next . Each possible state differs from the current state by the discrete position of any physically real object within the system ( e . g . the x , y , z position of the virion , the position of a CD4 protein or a CCR5 protein , a discrete point along the plasma membrane , etc . ) or by the creation or destruction of a bond between a receptor and gp120 . For example , a single CD4 protein with no neighboring proteins on the plasma membrane would offer four possible new states based on its physical location because of the Cartesian coordinates used to define the flexible plasma membrane . In this example , is an experimentally determined diffusion coefficient for CD4 on a cellular membrane , is the three-dimensional distance between the current position of CD4 and that of each available discrete position within the plasma membrane . For an unbound CD4 protein , a new state dictating the movement of the protein would use . However , an unbound CD4 protein may also offer additional states which do not correspond to the physical movement of the CD4 if it is capable of binding an unbound gp120 protein . Those additional states will vary from the current state by the creation of a previously nonexistent bond; the forward rates toward these states are discussed below . However , should the CD4 protein in question already be participating in an existing bond , the forward rate corresponding to the movement of the bound CD4 protein would have a nonzero between states . This nonzero is the change in energy of the existing bond and determines the probability that this CD4 protein moves in an energetically favorable ( relaxing the existing bond ) or an energetically unfavorable ( applying tension or compression to the bond ) manner . For the case of receptors located on the plasma membrane , the forward rate constant is calculated using a determined by the distance between discrete points along the plasma membrane , which vary during the simulation according to the membrane deformation . For system parameters such as the virion position , is a fixed step size which dictates that the forward rates will vary only according to the change in energy of the existing bonds . Similarly , the transition rates involving a change in z-position of discrete plasma membrane points are calculated using a fixed . Here the z-position corresponds to the height of each plasma membrane point in the z-axis while the plasma membrane itself is oriented in the x-y plane . A fixed , dictates that the transition rates of plasma membrane points will vary only according to the local membrane free energy , as discussed below , and if it be the location of a bound cellular receptor , the change in energy of that particular bond . For the systems examined here , the elapsed time between states and the distance over which physical objects move are of such a small order of magnitude that two assumptions can be made . First , the local energy landscape is approximated to be linear . Second , the probability density is assumed to be at a local steady state . Therefore , at small length scales , the system of the virion and plasma membrane is well described by the high-friction limit of the Fokker-Plank equation . The on and off rates ( and ) for CD4 and CCR5 bond formation with gp120 were calculated using experimentally measured rates [14] . Initial values , , were computed using a model described by Hummer and Szabo [17] , Here , is the effective diffusion coefficient , is the molecular spring constant , is the distance along the free energy well from the minimum to bond rupture , and is the minimum bond potential energy . This value of was then used to calculate for all bonds using the energy relationFor already existing bonds between cell and virion , values were calculated using the relationwhere the bond potential energy , , was calculated using a parabolic approximation of the Lennard-Jones potential , Here , is the distance along the energy potential calculated by subtracting the length of the proteins involved in the bond from the shortest distance between the location of the proteins on the plasma membrane and the virion . The total probability of transitioning out of the current state was equal to the sum of all forward rates for each possible destination state:The specific destination state of the system was determined by a pseudo random number generator ( PRNG ) . Briefly , once all possible states available to the current state are determined and their forward rates are calculated , the probability ( or rate constant ) describing the likelihood that the system transitions away from the current state is the total sum of all forward rates , . To determine which of these possible states is the next destination state , the PRNG yields a random number , , uniformly distributed between 0 and 1 . is multiplied by resulting in a random position between 0 and , which corresponds to a particular adjacent state . The system is subsequently updated , newly available states are determined , and their new forward rate constants are calculated according to the imposed changes ( e . g . if a gp120-CD4 bond breaks , new rates are calculated for the newly free gp120 and CD4 molecules ) . This process was repeated , updating each new state sequentially . Energy changes that governed the evolution of the system included those of individual gp120-CD4 bonds , gp120-CCR5 bonds , and the deformation of the plasma membrane with a specified elastic modulus and surface tension . The fluctuations of the membrane , diffusion of the receptors and the diffusion of the virus are described by Fokker-Planck equations . The simulation methodology is described in Atilgan et al . [18] . The total free energy of the system , , is given bywhere is the free energy of the plasma membrane calculated using the Canham-Helfrich form [19] , [20] , Here , is the mean curvature of the membrane , is the local area , is the elastic modulus and is the surface tension of the membrane . It should be noted that electrostatic interactions between protein pairs not bound together and between the viral and cellular membranes were not included in the computation of the energy for a given state . In addition , we simplified our model by assuming that the concentration of the local actin filament network beneath the cellular membrane is sufficiently low so as to not dictate plasma membrane deformation [21] . The time elapsed as the system stepped from one state to another was also calculated and used to determine the total time elapsed during the simulation , starting at = 0s . The duration of each time step was calculated using the equationHere , is the time step from the to the +1 state and is the uniformly distributed random number between 0 and 1 provided by the PRNG . The plasma membrane was initially defined as a completely flat surface . To simulate a more realistic interaction between cell and virion , the plasma membrane was allowed to evolve during the initialization of the system without the ability to form productive bonds with the virion above it . After this brief initialization ( 2×106 sequential iterations ) , the simulation of receptor-mediated viral adhesion to the cell surface was allowed to begin . Surface proteins on the plasma membrane were randomly distributed during the initialization according to the PRNG and concentrations of diffusing , unbound proteins were kept constant throughout the simulation . Proteins on the viral surface were either evenly spaced over the entire particle as previously described [22] , or randomly distributed according to the PRNG using a random zenith , θ , between 0 and π according to the probability density distribution , and random azimuth , , uniformly distributed between 0 and 2π . Throughout the simulation , proteins had a finite volume , so that other proteins were not allowed to diffuse through one another on either the viral surface or the plasma membrane . In addition , the plasma membrane and virion could not occupy the same space . If a physical obstacle was encountered , the forward rate for that adjacent state was set equal to 0 . The actual lengths of gp120 , CD4 and CCR5 molecules were also used when calculating bond interaction distances and free energies ( Table 1 ) . Lastly , gp120 trimers located on the viral surface were capable of binding up to three CD4 molecules and three CCR5 molecules at a time . As stated earlier , CCR5 adhesion to gp120 in contingent on a previously existing gp120-CD4 bond . In our system , gp120 trimers were not allowed to bind CCR5 unless that trimer was already involved in a gp120-CD4 bond . However , the number of CCR5 bonds was not allowed to exceed the number of CD4 bonds formed with a single gp120 spike , i . e . no synergistic effect was imposed throughout a gp120 trimer that would allow a single CD4 adhesion to promote multiple CCR5 bonds . For a more detailed explanation of the modeling algorithm see Text S1 .
First , we studied the development of viral-cell adhesion with a viral gp120 organization in which protein trimers were evenly distributed over the viral surface and were not allowed to move . We observed that as the system progressed towards steady state , the gp120-CD4 bond probability distribution displayed three distinct phases of organization ( Fig . 2A ) . The system quickly transitioned from single gp120 trimers bound to the plasma membrane , which we call phase I , to a second state through rotation and translocation , which allowed multiple gp120 trimers to bind to cellular receptors and produced a single broad node of bond probability formation , which we call phase II ( Fig . 2B ) . Accordingly , upon initial physical contact between the cell and virion , the gp120-CD4 bond probability distribution displayed a single maximum near the viral center ( ∼0nm ) . This is a result of initial gp120-CD4 bonds occurring most preferably at the closest point on the virion to the plasma membrane ( Fig . 2B ) . The shift of the bond probability maximum from the viral center is the result of averaging the initial bond locations of multiple simulations ( = 8 ) where the viral particle is binding to a non-uniform plasma membrane surface . At the initial time of contact the closest point on the cellular membrane is not always presented to the virion directly at = 0nm . Phase II resulted in a broadening of the CD4 bond probability distribution , during which its maximum shifted to a distance = 37nm from the center ( radius of the virion , 50 nm ) , i . e . CD4 receptors participating in the viral junction organized into a ring-like structure or corona ( Fig . 2B ) . CCR5 bonds also organized into a corona , with a spatial distribution similar to that of CD4 but containing much fewer bonds ( Fig . 2B ) . Through continued rotation and translocation of the virion , the interfacial region between cell and virion further evolved to develop a central , “anchor” gp120 spike surrounded by trimers bound to adjacent cellular receptors . This configuration produced a bimodal bond probability distribution of bound cellular receptors , which we call phase III ( Fig . 2 , A–D ) . While the organization of the adhesive junction between the cell and virion developed , the flexible plasma membrane spontaneously deformed and engulfed the virion . The deformation of the plasma membrane increased the cell surface area which was close enough to the virion to allow further receptor binding to gp120 trimers , thereby increasing the total bond number ( Fig . 2E ) . The time of formation of an organized viral junction ( Fig . 2 , C and D ) remained fairly constant over multiple simulations with different ( random ) initial states indicating that the final organization and number of bonds in the viral junction were relatively independent of the initial positions of the cellular membrane , virion , and receptors . The virion formed a stable adhesion interface and progressively increased the number of bonds . Simultaneously , the position of the virion above the membrane decreased as the membrane deformed to engulf the virion with time . Here , the change in the vertical height of the virion from its position at initial cell contact ( i . e . the depth of virion engulfment ) is referred to by , and the evolution of as adhesion progresses from Phase I to Phase III is illustrated in Fig 2F . Interestingly , after CD4 and CCR5 bond probabilities had reached steady states , the depth of engulfment of the virion continued to increase until the plasma membrane reached an equilibrium deformation ( Fig . 2 , G and H ) . Again , there are two counteracting “forces” that dictate the direction in which the organization of the virion-cell interface progresses . The first force is the energetically favorable formation of bimolecular bonds between proteins on the viral and cell surfaces; the second force is the energetically unfavorable deformation of the plasma membrane . The plasma membrane may be maintained in an unfavorable , deformed position if it is permissive of an increase in bond number ( Fig . 2H ) . Fixed gp120 units on the viral surface forced the virion to maximize bond formation by rotating and laterally moving the virion so as to minimally deform the plasma membrane , while recruiting new receptors to bind . Eventually , the depth of engulfment of the virion and the radial profile of the plasma membrane stabilized at heights that no longer exposed new gp120 units to cellular receptors on the plasma membrane . The organization of spatially fixed gp120 molecules on the viral surface regulates the organization of bonds between cell and virion ( Fig . 3 ) . Together these results suggest two important findings: ( i ) CD4-gp120 bimolecular bonds can be highly organized in the interfacial region between the cell and virus; ( ii ) the receptor organization is dynamic: initially a peak forms at the center , followed by a corona or bull's eye pattern of virus-cell bonds develops , and finally a peak and a corona co-exist while the plasma membrane deforms and engulfs the virion ( Fig . 2 , C and D ) . In the following , we conduct simulations over a wide range of parameters to investigate the effects of viral protein organization , receptor concentration , plasma membrane rigidity , and overall bond stability between the cell and virion , on the organization of cellular receptors . Recent studies suggest that the increased density of gp120 on the surface of viral particles could increase infection [3] , [23] , [24] . Therefore , we studied the effect of gp120 density ( at fixed positions on the viral surface ) on the organization of the viral junction over the physiological range of 7–20 gp120 trimers per virion [3] . We found that , when viral particles contained few gp120 ( 7–9 trimers ) , organized viral junctions did not form and little plasma membrane deformation occurred ( Fig . 4 ) . For an increased number of gp120 trimers per virion ( 14–20 trimers ) , the distance between adjacent gp120 trimers was sufficiently reduced that spontaneous deformations in the plasma membrane could result in an increase in the number of bonds ( Fig . 4 ) . The number of bound CD4 and CCR5 bonds correlated directly with the number of gp120 trimers on the viral surface ( data not shown ) , indicating that reduced gp120 density result in fewer virus-cell bonds , a less dynamic protein organization within the viral junction , and greatly reduced plasma membrane deformation . Recent work suggests that gp120 units may diffuse on the viral surface [25] . To determine whether diffusing gp120 trimers on the viral surface could also produce an organized viral junction , we conducted simulations allowing gp120 trimers to diffuse freely on the viral membrane ( Fig . 5 ) . The presence of freely diffusing gp120 trimers allowed for the formation of three times as many CD4 bonds than in the fixed gp120 case ( Figs . 2E and 5E ) . These bonds also formed 10 times faster than for virions with fixed gp120 positions . While the number of CD4 bonds increased beyond the steady state value of the fixed case , the number of CCR5 bonds did not change significantly from the fixed case ( Figs . 2E and 5F ) . Initially the CD4 bond probability profile in the organized viral junction was similar to that of the case of spatially fixed gp120 units . Then , the maximum of the bond probability grew outward resulting in a local maximum at a stable distance ∼27nm from the center ( Fig . 5A ) . After a slight delay , the CCR5 bond probability grew in a similar manner , but was dwarfed by the probability of forming CD4 bonds ( Fig . 5A ) . The probability distribution of bound cellular receptors did not feature a single peak directly beneath the center of the virion because of the deformation of the plasma membrane . The resulting curvature of the plasma membrane while induced by the virion , did not exactly match the virion curvature . These mismatched curvatures resulted in only a fraction of the area directly under the virion to be close enough to mediate the formation of bonds between viral and cellular receptors ( Fig . 5 , G and H , Stable CD4 and Rigid PM ) . These mismatched curvatures did not result from specific random starting conditions , distributions shown here result from eight independent simulations . As in the fixed gp120 case , virions with freely diffusing gp120 induced membrane deformation and resulted in viron engulfment ( Fig . 5 , G and H ) . Ultimately , membrane deformation resulted in a depth of virion engulfment of = 22nm . However , while the number of bonds increased much faster than in the fixed gp120 case , plasma membrane deformation and virion engulfment occurred at rates similar to those in the fixed gp120 case ( Figs . 2F and 5G ) . The location of bound cellular receptors corresponds to the shortest distance between the virion and the plasma membrane . When the plasma membrane is completely flat ( κ = ∞ ) , the virion is brought in close proximity to the cell to maximize the adhesion competent viral surface area ( Fig . 6A ) , cellular receptors at = 0 nm cannot bind the virion due to lack of space , resulting in a corona ( Fig . 5 , B , G , and H , hyper rigid PM ) . When the rigidity of the plasma membrane is relatively low ( κ = 20 kbT/nm ) , the maximum in the probability of bound receptors away from = 0nm , i . e . a corona of bound receptors is formed . When the rigidity of the plasma membrane is increased ( κ = 100 kbT/nm; Fig . 5 , D and H , Rigid PM ) , the steady state deformation of the plasma membrane is reduced and the maximum probability of bound receptors is shifted toward the radial center . Simple geometric analogies for the soft and rigid plasma membrane case are a sphere sitting in the bottom of a cone compared to a sphere sitting at the bottom of a larger sphere , respectively ( Fig . 6 , B and C ) . The slight decrease in bond number over time for the case of diffusive gp120 ( Fig . 5E , Stable CD4 and Rigid PM ) also results from the membrane deformation . Initially , bond number increases rapidly because diffusive gp120 molecules can concentrate between the virion and the plasma membrane . Bond number decreases slowly when the plasma membrane deforms , limiting the adhesion-competent area for cellular receptors to occupy ( Fig . 6 ) . Taken together these results suggest that the potential ability of gp120 molecules to diffuse in the viral membrane has a qualitative effect on the type of organization of the virus-cell interface and a quantitative effect on the number of bound receptors in that interface . We note that a virion with diffusing gp120 trimers produces patterns of bound receptors similar to those recently reported using electron tomography [25] , especially those produced by CD4 receptors with biphasic stability ( Fig . 5C , Unstable CD4; see more details below and in the Discussion section ) . Previous work suggests that gp120 trimers are unevenly distributed on the viral surface [3] , [25] and that these trimers may form fixed clusters [3] . We examined whether the formation of an organized viral junction underneath the virion containing a corona of bonds at an intermediate radial distance required uniformly distributed gp120 trimers by examining viral adhesion governed by randomly distributed gp120 trimers on the viral surface . Frequently , interfaces developed by these virions contained randomly clustered gp120 units which resulted in complex bond distributions similar to that produced with evenly placed gp120 ( Fig . S1 ) . However , the distance between the concentric rings of maximum bond probability depended on the spacing of gp120 on the viral surface and varied from one simulation to the next . This double-corona viral junction was a direct result of the non-uniform gp120 spike distribution on the viral surface . The interaction between local viral gp120 organization and receptors on the plasma membrane resulted in the progression of bond organization through distinct phases similar to the case of evenly distributed gp120 ( Fig . 2B and Fig . S1 ) . The virion first made contact with the plasma membrane , rotated , and as the deformation of the plasma membrane continued , a bimodal bond probability was established as virion rotation presented gp120 trimers previously unattainable to the receptors . The distance between gp120 trimers ultimately determined the distance between the nodes of the bond probability distribution ( Fig . S1 ) . Indeed , other simulations resulted in distributions that resembled phase II organization , i . e . a single-maximum bond probabilities single ( Fig . 2A , Fig . S1 ) . Unbound gp120 trimers that were spaced too far away on the viral surface for the virion to successfully rotate and expose to the cellular receptors resulted in halting the progression of bond organization at a single , off center probability maxima ( Fig . S1 ) . Together these results suggest that the final organization of CD4 receptors bound to fixed gp120 at the virus-cell interface before viral entry depends on the spatial organization of gp120 molecules on the surface of virions . The depletion of cholesterol from cellular membranes may significantly inhibit HIV-1 infection [26] and patients treated with cholesterol lowering statins seem to present decreased viral loads [27] . Aside from affecting subcellular pathways activated by statins , the depletion of cholesterol can dramatically affect the mechanical stiffness of the plasma membrane [28] . Previous work has demonstrated that varying cholesterol levels have a direct effect on the deformability of lipid vesicles [28] . We found that the plasma membrane rigidity critically influenced the spatial organization within the viral junction ( Figs . 5D and 7 , A and B ) . Cells with a completely rigid plasma membrane ( κ = ∞ ) coupled with evenly distributed gp120 units never progressed beyond the second phase of adhesion/bond formation described above ( Fig . 7B ) . This completely rigid plasma membrane resulted in slightly fewer CD4 bonds than in viral junctions involving cells with a flexible membrane and virions with fixed gp120 trimers ( Fig . 7E ) . A fivefold increase in membrane rigidity ( κ = 100 kbT/nm vs . 20 kbT/nm ) did not significantly change the steady state number of bonds in the viral junction ( Fig . 7E , Rigid PM ) and viral junctions typically took approximately twice as long to progress beyond the second phase of adhesion . Moreover , a fivefold increase in membrane rigidity resulted in a little more than double the depth of virion engulfment ( Fig . 7G , Rigid PM ) . For systems containing diffusing gp120 trimers adhering to receptors on a more rigid membrane , the virion depth of engulfment quickly stabilized at almost half the depth observed with a more flexible plasma membrane ( Fig . 5G ) . This was accomplished without a significant change in the numbers CCR5 bonds ( Fig . 5H ) . However , a more rigid membrane in conjunction with diffusive gp120 resulted in increased CD4 bond numbers and a higher probability of bond towards the center , = 0 nm ( Fig . 5 , D and E , Rigid PM ) . Together these results suggest that the mechanical properties of the plasma membrane can affect the organization and distribution of bonds within the viral junction . If organized virion-cell bonds are indeed important for successful infection , then the effect that cholesterol concentration has on the mechanical properties of the plasma membrane may contribute to its effect on HIV-1 infection . The concentration of CD4+ T cells harvested from the blood and lymph node tissue of infected patients correlates with infection and the ratio of [CCR5]∶[CD4] [29]; infection will correlate more to CD4 expression when CCR5 is expressed in limiting amounts and vice versa [30] . Our results indicate that the steady state number of CCR5 bonds in the viral junction depended most critically on the concentration of CCR5 , and did not strongly depend on the mechanical stiffness of the plasma membrane or the number , mobility , and the organization of gp120 on the virions ( Fig . 5F and 7F ) . For fixed gp120 trimers on the virion , a tenfold increase in CCR5 concentration produced a twofold increase in steady state number of CCR5 bonds in the viral junction ( Figs . 2E and 7F , for 10∶1 and 1∶1 [CD4]∶[CCR5] molar ratios , respectively ) . When CCR5 concentration was increased to the same level as CD4 , the physical hindrance of additional bound proteins located in an area of comparable size became more of a determining factor for protein organization than when CCR5 was present at a lower concentration . For example , increasing the CCR5 concentration resulted in a CD4 bond probability distribution containing three local maxima ( Fig . 7C ) . The increased number of CCR5 molecules , coupled with the increased stiffness and decreased length of the CCR5 bond quickly resulted in a virion with comparably little rotational freedom . The virion was bound by so many more CCR5 proteins than in any other simulation described here , that it had a much greater resistance to rotation and translocation so to minimize bond free energy ( because of the higher stiffness of CCR5 bonds ) . Figures 7 , G and H , compare the depths of virion engulfment and the plasma membrane profiles for increased CCR5 concentration , [CD4]∶[CCR5] = 1∶1 . While the plasma membrane has a similar z-position for a more rigid membrane ( ∼10nm away from viral center ) , the virion itself is sitting almost 10 nm higher than with a rigid membrane . This increased resistance ultimately results in phase III organization of bonds , however the increased CCR5 bonds at larger radii produced a CD4 bond probability distribution with three local maxima ( Fig . 7C ) . The two outward CD4 bond probability maxima ( ∼45nm and 65nm ) could have actually formed a single maximum , were it not for the steric hindrance of the shorter CCR5 bond coupled with the local plasma membrane deformation . The shorter CCR5 bonds concentrate in a small area of the plasma membrane and surrounded by the longer CD4 bonds which are able to diffuse over a larger area . Unable to occupy the same space , two local CD4 probability maxima flanking the outward CCR5 bond probability maxima were formed . We recently demonstrated that the presence of CCR5 could result in a decrease of CD4-gp120 bond stability [14] . Therefore , we examined the effect of CD4-gp120 bond instability on viral junction formation and organization . For fixed gp120 , CD4-gp120 bond instability resulted in fewer CD4 bonds between the cell and virus , averaging 8 at long time scales , as well as a decrease in average CCR5 bond number to practically zero ( Fig . 7 , E and F ) . The CD4 bond probability distribution was noticeably bare at the center of the virion-cell interface region compared to the stable CD4 bond case , indicating a slight difference in the third phase of organization ( Fig . 7D ) . Initial CD4 bonds were formed and increased in number similarly to the previous cases . However , when CCR5 bonds began to form and the CD4 bonds became unstable . CD4-gp120 bonds broke and the CD4 molecules diffused away , leaving only the CCR5 bond between the cell and virus . Ultimately , the last CCR5 bond broke and could not reform as no CD4 bonds were sufficiently close to initiate the gp120 conformation change . This production of CCR5 bonds and destruction of local CD4 bonds continued while the viral-cell interface as a whole maintained a constant number of bonds ( Fig . 7 , E and G ) , forming a globally stable adhesion interface ( Fig . 7D ) . Interestingly , imposing a biphasic gp120-CD4 instability while also allowing gp120 trimers to diffuse on the viral surface most accurately recreated the bond organization previously observed using electron tomography ( Fig . 5C ) .
Our computational results suggest that a viral particle can induce the formation of a highly organized ring-like ultrastructure of cell receptors bound to viral proteins , which we termed the viral junction . Our model involves biochemical ( e . g . binding constants ) and biophysical parameters ( e . g . membrane stiffness ) that have previously been measured . The diffusion rate constants of the plasma membrane and gp120 on the viral surface were assumed . However , these two unknown constants only set the rate of formation of a viral junction , not its steady state organization . Results from the model suggest that the formation of an organized viral junction is robust against relatively large variations in receptor concentrations , virion properties , physical properties of the plasma membrane , and dynamic properties of virus-cell bimolecular bonds . The simulations revealed that several factors contribute to the organization of bonds on the flexible membrane . The ability of CD4 and CCR5 molecules to diffuse while bound to gp120 contributes to the organization of the viral junction by increasing bond formation while decreasing plasma membrane deformation . For fixed gp120 ( i . e . unable to diffuse on the viral surface ) , the organization of the viral junction primarily depends on the gp120 distribution on the viral surface . While for the diffusive gp120 case , the organization of the viral junction primarily depends on the deformability of the plasma membrane . The longer and more flexible CD4 bond compared to the CCR5 bond , coupled with the local deformation of the plasma membrane , often result in different bond organization for these two receptors . Our computational model suggests that the mechanical properties of the plasma membrane work in concert with viral gp120 organization to organize cellular receptors at the virion-cell interface . Changes in the stiffness of the plasma membrane , which could be mediated by changes in cholesterol content [26] , affect the properties of the viral junction , including the total number of bonds between cell and virion . The virion-cell bonds work with the plasma membrane to reduce the overall potential energy of the system by two mechanisms . First , an energetically unstable bond forms where the plasma membrane is not locally deformed and the membrane is subsequently deformed to stabilize the bond . Second , the membrane spontaneously deforms to an unfavorable configuration and before it can relax a receptor forms a bond that maintains the deformation of the plasma membrane . The finite deformability of the plasma membrane limits how much the membrane can spontaneously deform without forming new bonds . By spacing gp120 trimers too far away on the viral surface for the plasma membrane receptors to spontaneously encounter them , the distribution of gp120 dictates the extent of plasma membrane deformation . Recently it has been suggested that endocytosis plays an important role in HIV-1 infection [4] . The formation of an organized viral junction , which is computationally described here within a small 200×200nm area , could occur anywhere from the plasma membrane to within an endocytic vesicle . Previous theoretical work using thermodynamic steady states predicted that an increase in membrane rigidity would decrease the number of bonds between cell and virion [21] . However , this conclusion was reached assuming a uniformly binding viral surface . Here , discrete locations of adherent gp120 trimers reveal that the point at which the plasma membrane is unable to bind and continue deformation is dependent on the distance at which unbound gp120 units are spaced . A computational model of how CD4 receptor organization on a planer surface responds to binding gp120 trimers has previously been introduced [31] . This earlier work focused on the rate at which the gp120 molecules of a trimeric spike become bound to CD4 as a function of gp120 density and CD4 diffusion coefficient . While this work predicts an increase in local CD4 concentration under the virion due to bond formation , it does not appear to display receptor organizations similar to those predicted by our work . This difference may stem from this earlier works use of a rigid plane to simulate the cellular membrane and the absence of co-receptor adhesion . Here , we employ a flexible membrane , which we demonstrate plays a critical role in the organization of the viral junction . The addition of coreceptor adhesion also offered insight into the roles that CD4 and CCR5 bond micromechanics play in the formation of the viral junction . Lastly , we had the advantage of using experimentally measured kinetic and micromechanical values for gp120-CD4 bonds . Recent reports suggest that gp120 could be partially disorganized or diffuse on the viral surface [25] and that viral infection correlates with gp120 concentration [3] , [23] . Therefore , we studied the effects of diffusing vs . non-diffusing gp120 , as well as gp120 density on the formation , organization , and dynamics of the viral junction . Viral particles with diffusive gp120 organized cellular receptors into a distinct corona to maximize the number of bonds between the cell and virus ( Fig . 5 , A–D ) . Sougrat et al . suggest that a virion adhering to a cell induces a similar formation of proteins between the cell and virus , called the ‘entry claw’ , with comparatively little gp120 elsewhere on the virion . It was suggested that this concentration gradient in gp120 along the viral surface is best explained by the ability for gp120 to diffuse on the surface . The conditions that best reproduced the bond organization observed using electron tomography is a combination of diffusive gp120 trimers with an imposed biphasic stability on the gp120-CD4 bond , suggesting that gp120 are not permanently organized on the viral surface . However , the gp120 gradient observed by Sougrat et al . can also be explained by the possibility that virions with clustered gp120 trimers have preferential binding as well as the possibility that gp120 trimers could be cleaved from the virion during sample preparation for electron tomography . Therefore , in addition to the diffuse case , we also considered the effect of gp120 configurations on viral junction organization . An experimental test of our computational predictions is challenging given the small sizes of the virion and associated viral junction . However , the advent of super resolution microscopy approaches , such as photoactivated localization microscopy ( PALM ) [32] , could help determine the organization of the viral junction through co-labeling of gp120 and receptors CD4 and CCR5 . Combined immunolabeling and cryo-electron microscopy ( EM ) or tomography could also help assess the organization of the viral junction; however EM often creates artifacts especially when visualizing the plasma membrane . If we assume that certain bond configurations result in enhanced infection , then the subpopulation of infectious viral particles not only depends on gp120 density , but also gp120 organization on the viral surface . A viral population containing particles equipped with a functional gp120 organization ( i . e . with relatively clustered gp120 on the viral surface ) could lead to the formation of an organized viral junction , while those with dysfunctional organizations ( with relatively distant gp120 ) , may be unable to infect cells . Since the formation of a viral junction depends on receptor density and the mechanical properties of plasma membrane , viral junction formation could be cell type-specific . One might speculate from the spectrum of mechanical properties exhibited by different cell types as well as the distribution of viral particle size and gp120 concentration that a subset of viral particles might preferentially infect one cell type while another subset of particles could preferentially infect another cell type . Although the formation of the viral junction would occur at the much smaller length scales than that of a whole cell , the organization of the viral junction is reminiscent of the immunological synapse . It is therefore tempting to speculate on the possible signaling function of the viral junction . For instance , gp120 binding induces assembly of a local actin network and coreceptor binding induces disassembly of actin filaments [33] , [34] . By analogy to the better-characterized immunological synapse , we speculate that part of this signaling function could be related to the complex evolution of an organized viral junction by relying on one organizational phase between receptor and coreceptor for an initial signaling event and a subsequent organizational phase for a secondary signaling event .
|
The entry of human immunodeficiency virus ( HIV ) into cells is the target for new therapies preventing HIV infection . While intermediate steps of viral entry have been characterized , the progression between these steps and how they result in productive infection are not well understood . By using stochastic modeling , we examine the initial interaction of a single viral particle with a flexible plasma membrane populated with viral receptors . The model predicts the formation of an organized receptor ultrastructure beneath the viral particle , which we name viral junction and which may contribute to productive viral infection . The organization of the viral junction depends on receptor density , CD4 bond stability , membrane mechanical flexibility , as well as viral protein organization and density .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"computational",
"biology/molecular",
"dynamics",
"biophysics/biomacromolecule-ligand",
"interactions",
"cell",
"biology/cell",
"adhesion",
"infectious",
"diseases/hiv",
"infection",
"and",
"aids"
] |
2010
|
Organization of Cellular Receptors into a Nanoscale Junction during HIV-1 Adhesion
|
Among the Reduviidae family , triatomines are giant blood-sucking bugs . They are well known in Central and South America where they transmit Trypanosoma cruzi to mammals , including humans , through their feces . This parasitic protozoan is the causative agent of Chagas disease , a major public health issue in endemic areas . Because of the medical and economic impact of Chagas disease , the presence of other arthropod-borne pathogens in triatomines was rarely investigated . In this study , seven triatomines species involved in the transmission of T . cruzi were molecularly screened for the presence of known pathogens generally associated with arthropods , such as Rickettsia , Bartonella , Anaplasmataceae , Borrelia species and Coxiella burnetii . Of all included triatomine species , only Eratyrus mucronatus specimens tested positive for Bartonella species for 56% of tested samples . A new genotype of Bartonella spp . was detected in 13/23 Eratyrus mucronatus specimens , an important vector of T . cruzi to humans . This bacterium was further characterized by sequencing fragments of the ftsZ , gltA and rpoB genes . Depending on the targeted gene , this agent shares 84% to 91% of identity with B . bacilliformis , the agent of Carrion’s disease , a deadly sandfly-borne infectious disease endemic in South America . It is also closely related to animal pathogens such as B . bovis and B . chomelii . As E . mucronatus is an invasive species that occasionally feeds on humans , the presence of potentially pathogenic Bartonella-infected bugs could present another risk for human health , along with the T . cruzi issue .
Triatomine bugs ( order Hemiptera , family Reduviidae , subfamily Triatominae ) are blood-sucking arthropods ( “kissing bugs” ) , most of which can feed both on animals and humans . All stages from first instar to male and female adults are strictly hematophagous and responsible for a relatively large blood intake due to their large size . They are mainly sylvatic and feed on small wild mammals but can also feed on birds and bats [1] . Triatomines occupy diverse natural ecotopes , such as mammal and bird nests , hollow trees , caves and rock fissures [2] , but also rural environments , as they can prosper in crevices in houses [1] . These arthropods are distributed world-wide but the vast majority of the 140 recognized species is found in the Americas [3] . They are particularly well studied in South America , where they transmit an endemic flagellate pathogen , T . cruzi , the etiological agent of Chagas disease [1] . Also known as the American trypanosomiasis , Chagas disease is a neglected tropical disease , the first human parasitic disease in the endemic areas . T . cruzi is transmitted through the feces of infected kissing bugs , causing heart failure 10 to 30 years post-infection for almost 30% of individuals [4] . Because of the public health impact of Chagas disease , studies on kissing bugs are mainly focused on this theme . As a matter of fact , the presence of other human pathogens was never described in the hundred years that it has been known that kissing bugs could transmit pathogens . Only the presence of Wolbachia and Arsenophonus species was investigated based on the fact that these obligate intracellular bacteria are known to be endosymbionts of many arthropods [5 , 6] . The presence of zoopathogenic arthropod-borne viruses was also investigated . To our knowledge , there is no report of pathogen detection in dejections or in triatomines themselves , except for T . cruzi , although Arsenophonus nasoniae was once reported to be detected in an eschar of a human [7] . Regarding viruses , two have been described in these bugs . Triatoma virus is reported as strictly entomopathogenic , particularly for its principal host , Triatoma infestans [8] , while African swine fever virus was detected in Triatoma gerstaeckeri but not transmitted to pigs [9] . French Guiana is an 84 , 000 km2 overseas department and region of France bordered by Brazil and Suriname . Because of its many different ecosystems , particularly a dense rainforest , French Guiana is a biodiversity hotspot and one of the 21 areas where Chagas disease is endemic [10] . Among the 27 described species of triatomines in this area , many are invasive species . That is to say that many of them temporarily leave their sylvatic or peridomestic dwellings in order to invade houses . The main vector community of French Guiana comprises highly anthropophilic bugs belonging to the Panstrongylus , Rhodnius and Eratyrus genera [10] . They accidentally feed on humans [11] and also on potentially infected animals since they easily feed on domestic animals or wild mammals . Aiming to add to knowledge regarding bacteria and triatomine association , we screened seven species of triatomines bugs from French Guiana by molecular biology for the presence of arthropod-borne bacteria such as Rickettsia , Bartonella , Borrelia , Anaplasma , Wolbachia , Ehrlichia species and Coxiella burnetii .
Triatomine specimens were collected in French Guiana from 1991 to 2013 using light traps or interception traps by one of the authors ( JMB ) and by the Société Entomologique Antilles-Guyane ( SEAG ) as part of an inventory of French Guiana’s insects . Triatomines were caught in forests ( Horses Mountains , Kaw Mountains ) , peridomiciliary areas ( Degrad Kwata , Kaw Mountains , Nancibo ) or urban areas ( Sinnamary , Kourou savannah ) as displayed in Fig 1 . All triatomine specimens were morphological identified with the Bérenger et al . morphological key [12] and kept dried as insect collections . Seven T . cruzi vectors were included in this study: Rhodnius prolixus ( n = 10 ) , Rh . pictipes ( n = 10 ) , Rh . robustus ( n = 10 ) , Triatoma infestans ( n = 10 ) , Panstrongylus geniculatus ( n = 10 ) , P . rufotuberculatus ( n = 4 ) , Eratyrus mucronatus ( n = 23 ) . Dried triatomines were rinsed in sterile water and air-dried on filter paper before cutting lengthwise in two equal halves , using a sterile surgical blade for each specimen . One half and the legs were stored at -20°C as a backup sample and the other legless half was selected for molecular analyses . Each half triatomine was crushed with a sterile pestle in 400 μL of a G2 buffer solution containing 40 μM of proteinase K ( Qiagen ) and incubated at 56°C overnight . After 1 minute of centrifugation at 7000 x g , 200 μL of the supernatant was then collected prior to DNA extraction . Triatominae genomic DNA was individually extracted using the EZ1 DNA tissue extraction kit ( Qiagen , Hilden , Germany ) according to the manufacturer’s instructions . Triatominae DNAs were then eluted in 100 μL of Tris EDTA ( TE ) buffer using the DNA extracting EZ1 Advanced XL Robot ( Qiagen ) as previously described [13] . DNAs were either immediately used or stored at -20°C until used for molecular analysis . The DNA extracting EZI Advanced XL Robot was disinfected after each batch of extraction as per the manufacturer recommendations to avoid cross-contamination . DNA samples were individually tested by genus-specific PCR using primers and probes targeting specific sequences of Bartonella spp . , but also Rickettsia spp . , Coxiella burnetii , Borrelia spp . , and all Anaplasmataceae species [14] as previously described [15] ( Table 1 ) . Real-time quantitative PCR ( qPCR ) was carried out according to the manufacturer’s protocol using a CFX Connect Real-Time PCR Detection System ( Bio-rad , Hercules , CA , USA ) with the Eurogentec Takyon qPCR kit ( Eurogentec , Seraing , Belgium ) . Bartonella elizabethae , Rickettsia montanensis , Coxiella burnetii , Anaplasma phagocytophilum and Borrelia crocidurae DNAs were used as positive qPCR controls for the primers and probe targeting respectively all Bartonella , Rickettsia , Coxiella burnetii and Borrelia species . DNAs were tested at different concentrations to avoid PCR inhibition . For each run , a PCR mix without DNA was used as negative control . Standard PCR targeting a 710 bp region of the invertebrate cytochrome oxidase I ( COI ) gene was performed on PCR negative triatomines to control DNA extraction . DNA samples that were positive with Bartonella-qPCR were submitted to conventional PCR amplification using a Bio-Rad Thermocycler ( Bio-Rad Laboratories , Hercules , CA ) prior to sequencing . For Bartonella species identification , primers targeting Bartonella rpoB , gltA and ftsZ genes fragments were used as previously described [16] . DNA from Bartonella elizabethae served as PCR positive control and mixture without DNA as negative control . The cycling protocol consisted of 15 min at 95°C followed by 35 cycles of denaturing at 95°C for 30 s , annealing at 50°C for 30 s ( 58°C for rpoB gene ) , extension 1 min at 72°C , followed by a final cycle of 1 min at 72°C and sampling held at 4°C . Amplification products were separated by electrophoresis through a 1 . 5% agarose-tris-borate-EDTA gel containing ethidium bromide . PCR products were sequenced using a Big Dye Terminator kit and an ABI PRISM 3130 Genetic Analyser ( Applied BioSystems , Courtabeauf , France ) . The sequences were analyzed using the ABI PRISM DNA Sequencing Analysis software version 3 . 0 ( Applied BioSystems ) and compared to sequences available in the GenBank database using the BLAST algorithm ( http://blast . ncbi . nlm . nih . gov/Blast . cgi ) . The partial sequences of ftsZ and rpoB genes of Bartonella amplified from the sample EmG01 are available in GenBank at #KX377404 and #KX377405 . Phylogeny of the detected Bartonella with other members of the Bartonella genus was established with TOPALi 2 . 5 software ( Biomathematics and Statistics Scotland , Edinburgh , UK ) . Available sequences of ftsZ , gltA and rpoB genes of validated Bartonella species were retrieved from the National Center for Biotechnology Information ( NCBI ) based on the results of the BLAST program . Multiple sequence alignment was performed with the ClustalW multiple sequence alignment program , which is included in the BioEdit software .
Triatomines were collected in eight different localities in French Guiana with no selection regarding species and sex ( convenient sampling ) . Among the triatomines collected , Eratyrus mucronatus ( Fig 2 ) accounted for 20% of catches and 29 . 8% of the specimens analyzed . Details related to collection , such as sampling area and triatomines’ sex , are indicated in Table 2 . Of 23 E . mucronatus samples , 22 ( 95 . 6% ) were male . Further details regarding other collected species have been listed elsewhere [12] . DNAs extracted from all the triatomines were included to assess the presence of Bartonella species . Of 23 Eratyrus mucronatus samples , 13 ( 56 . 5% ) were positive by Bartonella spp . -specific qPCR , with cycle threshold ( Ct ) values ranging from 13 . 23 to 25 . 91 ( mean: 21 . 44 ) ( Table 2 ) . These specimens were from six distinct regions and collected between 1993 and 2003 . All positive specimens were male , and a majority of them were collected in sylvatic and peridomestic areas: the Kaw Mountains ( 38 . 4% ) and Horses Mountains ( 38 . 4% ) . All samples tested negative for the presence of Rickettsia spp . , Borrelia spp . Anaplasma spp . , Ehrlichia spp . , Wolbachia spp . and Coxiella burnetii . Bartonella spp . was only detected in Eratyrus mucronatus specimens . A 787 bp fragment of the Bartonella spp . rpoB gene was amplified using conventional PCR primers prior to sequencing . Only three ITS2-qPCR positive samples were also positive for rpoB by standard PCR . Sequencing failed for two of them . Comparison of the one rpoB resulting sequence against the NCBI database using the BLASTN program indicated that it possessed 90% identity with the ATCC Bartonella bacilliformis 35685D-5 strain ( #CP014012 . 1 ) and with the B . bacilliformis KC583 strain ( #CP000524 . 1 ) . The next closest cultivated strains were a B . bovis strain [17] and a B . chomelii strain [18] , both with 89% identity . Our genotype also possesses 87% identity with a Bartonella ancashensis strain [19 , 20] . Available sequences of Bartonella rpoB gene were retrieved from NCBI and compared to the Bartonella sequence described hereby . This Bartonella genotype formed a distinct clade , with a strain of Bartonella bacilliformis as the closest clade based on rpoB gene analysis ( Fig 3 ) . All samples allowed amplification of a single 333 bp fragment of the ftsZ gene by standard PCR . Blast analysis revealed 91% identity with the aforementioned 35685D-5 and KC583 B . bacilliformis strains ( Fig 4 ) . A total of 12 out of 13 samples were successfully amplified by standard PCR targeting a fragment of the gltA gene . BLAST analysis showed 88% identity with uncultured Bartonella species detected in bank voles [21] , deer [22] and bats from Africa [23] , but also with B . bovis and B . chomelii strains ( Fig 5 ) . Based on the gltA gene , our genotype is 84% similar to B . bacilliformis strains . Only a single rpoB sequence was obtained but all ftsZ and gltA sequences obtained were identical for all E . mucronatus specimens . BLAST analysis of the concatened sequence of the three genes revealed 99% of coverage and 90% similarity with the two aforementioned B . bacilliformis strains . Phylogenetic analysis based on the concatened sequences revealed clustering of our Bartonella strain with two B . bacilliformis and B . ancashensis strains ( Fig 6 ) .
Bartonella species are small fastidious gram-negative bacteria belonging to the Alphaproteobacteria class that are able to infect many mammals , including humans [24] . They are mostly transmitted by arthropod vectors such as sandflies ( Lutzomyia verrucarum ) , human body lice ( Pediculus humanus humanus ) , different fleas including cat fleas ( Ctenocephalides felis ) , biting flies and ticks [25] . Among the several Bartonella species , some have been identified as human pathogens , causing well-known vector-borne diseases such as Carrion’s disease ( B . bacilliformis ) , trench fever ( B . quintana ) , cat scratch disease ( B . henselae ) as well as endocarditis [24] . We hereby describe a novel Bartonella genotype , phylogenetically related to several human and animal pathogens , as shown by the phylogenetic analyses . B . bacilliformis , a closely related species , is the causative agent of the first and well-described human bartonellosis called Carrion’s disease [26] . Transmitted through the bite of an infected phlebotomine sand fly , L . verrucarum , this South American endemic bacterium can induce a biphasic illness with two distinct syndromes that can be concomitant or independent . An acute phase known as Oroya fever manifests as a hemolytic fever linked to bacteremia that can range from 10 to 210 days and can be fatal in 40–88% of individuals without treatment . The second syndrome called verruga peruana manifests as blood-filled hemangiomas due to infection of the endothelium [26] . No human cases of B . bacilliformis infection have been reported in French Guiana to date [27] . Our genotype is also closely related to a strain of B . ancashensis , a recently described Bartonella species closely related to B . bacilliformis that was isolated from the blood of two patients diagnosed with a chronic stage of verruga peruana in Peru [20] . All data suggest that B . ancashensis could be a second agent . Our new agent is also closely related to B . bovis strains . Isolated from cats , which are only accidental hosts , this endocarditis [28] . E . mucronatus is a sylvatic triatominae bug involved in the transmission of T . cruzi [11] . It is recognized now as an invasive species as it has been described around and inside houses since 1959 [29] because of its attraction to artificial light sources [30] . They are known to feed on bats , but also on small mammals such as xenarthrans and opossums [11] . Bats are widely reported to be sources of many viral and bacterial pathogens [31] , including Bartonella spp . worldwide , including in French Guiana [32] , Nigeria [33] , Guatemala [34] and Vietnam , for example [35] . Therefore , Bartonella spp . were frequently detected in hematophagous arthropods feeding on bats such as bat flies ( Hippoboscidae , Streblidae , Nycteribiidae ) [36] or Cimex adjunctus [37] . Triatomine vectors belonging to genera Triatoma , Rhodnius , Panstrongylus and Eratyrus can be totally domiciliated or invasive , since they occasionally visit houses as described in Bolivia [38] , Brazil [39] , Argentina [40] and Venezuela [11] . The presence of these bugs around houses has long been known and has justified the establishment of chemical control campaigns , which after 5 years remain a failure in Bolivia [38] . The invasive behavior of E . mucronatus has not yet been described in French Guiana but data from Bolivia suggest that eliminating it once it is settled is challenging [38] . Living in various ecotopes and not host-specific [41] , these bugs can easily feed both on humans and animals [38] , both of them potentially bacteriemic , parasitemic or viremic at the blood meal time point . Triatominae species are well-studied bugs , however , this work provides the first evidence to our knowledge of infection with a bacterium that is not a priori endosymbiotic . The specimens we analyzed were dry , with no information regarding their engorgement status at the time of collection . However , as they were collected using light traps or interception traps , we can assume that they were seeking hosts and therefore probably non-engorged . Thus , we can suppose that we did not detect DNA of a bacterium present in recently ingested blood but a genuine infection . To support this hypothesis , the infection rate was considerable ( 56% ) among triatomines collected in very distant sampling periods , geographically and over time . Ct values were also very low , increasing the possibilities that this bacterium multiplies within the bug's body . However , to evaluate the possibility of transmission of these Bartonella spp . , an experimental model of infection , or at least information regarding the location of the bacteria in the bug , would be necessary . Such information could not be obtained from our samples as they were dry and old . Cultivation of any bacteria or any attempt to localize with immunofluorescence , for example , was not possible . In continuing this work , it would be interesting to collect wild E . mucronatus specimens in order to isolate the bacterium and establish an experimental model of infection with this arthropod/pathogen pair . This would reveal whether the bug is a simple carrier or an efficient vector of this Bartonella . The possible interaction between T . cruzi and this Bartonella spp . in this insect is also unknown and could be investigated by monitoring the trypanosome’s cycle and transmission in co-infected E . mucronatus . Being phylogenetically closely related to two severe human pathogens ( B . bacilliformis and B . ancashensis ) , it would also be important to evaluate its pathogenicity . Because of the huge public health impact of Chagas disease in South America , investigations on Triatominae were limited to the study of their interactions with T . cruzi . In fact , Triatominae bugs may host such bacteria as Bartonella species and , probably , may be its vector .
|
Triatomines are hematophagous insects including vectors of T . cruzi , the agent of Chagas disease , a huge public health issue , especially in South America . Whether these arthropods carry other pathogenic microorganisms is currently unknown . We investigated the presence of different arthropod-borne pathogens , including Bartonella spp . , by quantitative PCR . Bartonella species were identified using ftsZ , gltA and rpoB gene sequencing and a new genotype of Bartonella spp . was detected in Eratyrus mucronatus specimens , an important vector of T . cruzi to humans . This agent is closely related to several human and animal pathogens . Depending on the gene fragment used , this agent shares 84% to 91% of identity with B . bacilliformis , the agent of the deadly Carrion’s disease . The possibility of transmission of potentially pathogenic bacteria could be an additional threat to human health since E . mucronatus bugs are more and more anthropophilic .
|
[
"Abstract",
"Introduction",
"Methods",
"Results",
"Discussion"
] |
[
"invertebrates",
"medicine",
"and",
"health",
"sciences",
"pathology",
"and",
"laboratory",
"medicine",
"pathogens",
"microbiology",
"vertebrates",
"parasitic",
"protozoans",
"animals",
"mammals",
"organisms",
"protozoans",
"phylogenetic",
"analysis",
"molecular",
"biology",
"techniques",
"bacteria",
"bacterial",
"pathogens",
"research",
"and",
"analysis",
"methods",
"sequence",
"analysis",
"artificial",
"gene",
"amplification",
"and",
"extension",
"bioinformatics",
"sequence",
"alignment",
"medical",
"microbiology",
"microbial",
"pathogens",
"molecular",
"biology",
"molecular",
"biology",
"assays",
"and",
"analysis",
"techniques",
"bartonella",
"arthropoda",
"trypanosoma",
"cruzi",
"trypanosoma",
"dna",
"sequence",
"analysis",
"database",
"and",
"informatics",
"methods",
"horses",
"biology",
"and",
"life",
"sciences",
"equines",
"amniotes",
"polymerase",
"chain",
"reaction"
] |
2017
|
Detection of a Potential New Bartonella Species “Candidatus Bartonella rondoniensis” in Human Biting Kissing Bugs (Reduviidae; Triatominae)
|
Intracellular pathogens must egress from the host cell to continue their infectious cycle . Apicomplexans are a phylum of intracellular protozoans that have evolved members of the membrane attack complex and perforin ( MACPF ) family of pore forming proteins to disrupt cellular membranes for traversing cells during tissue migration or egress from a replicative vacuole following intracellular reproduction . Previous work showed that the apicomplexan Toxoplasma gondii secretes a perforin-like protein ( TgPLP1 ) that contains a C-terminal Domain ( CTD ) which is necessary for efficient parasite egress . However , the structural basis for CTD membrane binding and egress competency remained unknown . Here , we present evidence that TgPLP1 CTD prefers binding lipids that are abundant in the inner leaflet of the lipid bilayer . Additionally , solving the high-resolution crystal structure of the TgPLP1 APCβ domain within the CTD reveals an unusual double-layered β-prism fold that resembles only one other protein of known structure . Three direct repeat sequences comprise subdomains , with each constituting a wall of the β-prism fold . One subdomain features a protruding hydrophobic loop with an exposed tryptophan at its tip . Spectrophotometric measurements of intrinsic tryptophan fluorescence are consistent with insertion of the hydrophobic loop into a target membrane . Using CRISPR/Cas9 gene editing we show that parasite strains bearing mutations in the hydrophobic loop , including alanine substitution of the tip tryptophan , are equally deficient in egress as a strain lacking TgPLP1 altogether . Taken together our findings suggest a crucial role for the hydrophobic loop in anchoring TgPLP1 to the membrane to support its cytolytic activity and egress function .
Cellular egress from the host is a crucial step in the infectious cycle of intracellular pathogens . Accordingly , such pathogens have evolved multiple exit strategies , which can be divided into those that leave the host cell intact and those that rupture the host cell . Several bacterial pathogens , including L . monocytogenes , use an actin-based protrusion mechanism that allows a bacterium to enter a neighboring host cell without damaging the original host cell [1] . Other bacteria have developed extrusion or expulsion mechanisms that also leave the host cell intact [2–5] . Also , pyroptotic and apoptotic mechanisms leverage cell-death signaling as a means for intracellular pathogens to exit the host cell [6] . Many apicomplexan parasites , including Toxoplasma gondii , use a cytolytic mechanism of egress that obliterates the infected cell . Cytolytic egress results in direct tissue destruction and indirect collateral damage from the ensuing inflammatory response , a hallmark of acute infection by apicomplexan parasites and a key aspect of disease [7] . The apicomplexan phylum encompasses a variety of parasitic genera important to both human and veterinary health including Plasmodium , Cryptosporidium , Eimeria , and Toxoplasma [8–11] . These parasites contain a unique set of apical secretory organelles , including micronemes and rhoptries , which discharge proteins involved in parasite motility , host cell manipulation , and egress [12–15] . T . gondii is capable of infecting and replicating asexually within virtually any nucleated cell during the acute stage of infection . Its “lytic cycle” can generally be divided into three steps: invasion where the parasite containing vacuole ( parasitophorous vacuole , PV ) is formed , intracellular replication , and finally egress from the vacuole . Whereas invasion and intracellular replication have garnered considerable attention , egress remains the least understood component of the lytic cycle . Efficient egress by T . gondii critically relies on micronemal secretion of TgPLP1 , a member of the membrane attack complex/perforin ( MACPF ) protein family [15] . MACPF proteins play central roles in immunity ( e . g . , perforins and complement proteins ) , embryonic development ( Drosophila torso-like and mammalian astrotactins ) , fungal predation ( Oyster mushroom pleurotolysins A/B ) , and cell traversal or egress by apicomplexan parasites . The domain arrangement for TgPLP1 includes a central MACPF domain that is flanked by an N-terminal domain ( NTD ) and a C-terminal domain ( CTD ) . Both the NTD and CTD have membrane-binding activity , but only CTD is crucial for proper TgPLP1 function [16] . Although structural insight into the TgPLP1 apicomplexan perforin β-domain ( APCβ ) , which comprises a portion of the CTD , was reported recently , [17] how structural features of this domain contribute to the function of TgPLP1 in egress has not been addressed . Overall , the mechanism of MACPF proteins begins with membrane recognition by the CTD [18] . Following membrane binding , MACPF proteins oligomerize into ring or arc shaped complexes and undergo a marked structural rearrangement of the MACPF domain where the so-called CH1 and CH2 helices unfurl and become extensions of the central β-sheets to create a large pore [19] . Previous work has shown that TgPLP1 and other apicomplexans share this general mechanism of pore formation . However , the molecular determinants of membrane recognition and binding remain poorly understood . Here , we present evidence that TgPLP1 has a strong preference for inner leaflet lipids and that binding to such membranes occurs via the CTD . Additionally , we solved the 1 . 13 Å resolution X-ray crystal structure of the TgPLP1 APCβ domain and identified a hydrophobic loop that likely inserts into the target membrane . Finally , we use CRISPR/CAS9 to insert mutations into the hydrophobic loop and show that it is critical for egress competence .
Previous work has identified PLP1 as an important egress factor that can bind membranes through its N-terminal domain ( NTD ) and C-terminal domain ( CTD ) ; however , only the CTD has been shown to be required for lytic activity [16] . In the canonical pore forming mechanism , MACPF/CDC proteins bind to the target membrane via the CTD as a first step . Membrane interactions occur via specific protein [20–22] or lipid [23–26] receptors in the target membrane . To test if TgPLP1 is capable of binding lipid receptors , we generated liposomes that mimic the outer leaflet ( OL ) or inner leaflet ( IL ) of the plasma membrane [27] . We tested the binding of both native TgPLP1 and a series of recombinant His-tagged constructs of TgPLP1 ( Fig 1A ) via membrane flotation . Native and recombinant full length TgPLP1 ( rTgPLP1 ) both showed a striking preference for IL liposomes ( Fig 1B & 1C ) . Recombinant TgPLP1 CTD ( rCTD ) also showed a preference for IL liposomes whereas the recombinant NTD ( rNTD ) bound equally to IL and OL liposomes ( Fig 1B & 1C ) . As a control , we included an unrelated His-tagged recombinant micronemal protein ( rTgMIC5 ) , which failed to bind liposomes . Since rCTD appears to bind preferentially to liposomes that mimic the IL of the plasma membrane we tested if rCTD prefers binding liposomes composed of individual lipids or combinations thereof . Consistent with our previous observation , rCTD does not bind to liposomes comprised of OL phospholipids , namely phosphatidylcholine ( PC ) and sphingomyelin ( SM ) , in the presence and absence of cholesterol ( Fig 1D & S1 Fig ) . Consistent with its preference for certain IL lipids , rCTD bound to liposomes composed of phosphatidylethanolamine ( PE ) or phosphatidylserine ( PS ) , but not phosphatidylinositol ( PI ) ( Fig 1D ) . However , none of the liposomes prepared with individual IL liposomes fully recapitulate the binding to IL mimic liposomes . Together these findings suggest that the preference of TgPLP1 for IL lipids is conferred by its CTD and involves amalgamated binding to PS and PE . TgPLP1 contains a well-conserved central MACPF domain that is flanked by a poorly conserved NTD and CTD ( Fig 2A ) . The CTD includes the apicomplexan specific APCβ domain , which consists of 3 direct repeats with 4 highly conserved cysteines in each repeat and a C-terminal tail ( CTT ) that includes a basic patch . To better understand the molecular determinants that govern membrane binding and lipid specificity in this system we expressed , purified , and crystallized the TgPLP1 CTD . Despite using a construct that encompassed the entire CTD , the resulting crystals contained only the APCβ domain . Whether this discrepancy is due to a lack of electron density or enzymatic cleavage of the CTT remains unclear . The TgPLP1 APCβ domain crystallized in the C121 space group with two monomers in the asymmetric unit . The structure was solved using iodine soaks and single-wavelength anomalous dispersion and subsequently refined to 1 . 13 Å resolution . The crystallographic data and refinement statistics are summarized in Table 1 . The TgPLP1 APCβ structure shows that the β-rich repeat region is comprised of a single globular domain with internal pseudo threefold symmetry wherein each β-rich repeat forms a subdomain ( Fig 2B & 2C ) . Each of the three subdomains contains an internal antiparallel β-sheet and an outer β-hairpin ( S2A & S2B Fig ) . The inner and outer layers of each subdomain are held together , in part , by two disulfide bonds between the highly conserved cysteines . An overlay of the three subdomains highlights the similarity in the core of each subdomain . Indeed subdomains 1 and 2 ( S2C Fig red and green ) overlay with an RMSD of 0 . 728 Å . Subdomain 3 ( S2C Fig blue ) , however , aligns less well with subdomain 1 ( RMSD 3 . 207 Å ) and subdomain 2 ( RMSD 3 . 175 Å ) , with the majority of the alignment differences attributed to two loops that protrude from the “bottom” of subdomain 3 . The longer of the two loops has hydrophobic character , and thus is termed the hydrophobic loop ( Fig 2B , colored cyan ) , whereas the shorter of the two loops has basic character , termed the basic loop ( Fig 2B , colored orange ) . To determine commonality of double-layered β-prism fold we searched for proteins with structural similarity to the APCβ using the Dali server [28] . The Dali server measures similarity by a sum-of-pairs method that outputs a Dali-Z score . Structures with a Dali-Z score above 2 are considered to have structural similarity . With the exception of a recently published structure of a similar TgPLP1 APCβ construct ( Dali-Z score 48 . 5 ) [17] , the top scoring result from the Dali server for the TgPLP1APCβ structure is the C-terminal domain of human proprotein convertase subtilisin/kexin type 9 ( PCSK9 V-domain , Dali-Z score 6 . 0; S1 Table ) . Indeed , this domain contains similar internal pseudo threefold symmetry and is a double-layered β-prism fold with disulfide bonds that link the inner and outer layers . However , the APCβ structure has two outer layer β-sheets whereas the PCSK9 V-domain has three outer layer β-sheets ( S3A–S3C Fig ) . Despite the similarities in the overall fold the APCβ structure has very poor global structural alignment with the PCSK9 V-domain ( RMSD 15 . 706 Å ) but comparison of individual subdomains shows a much closer alignment ( RMSD 3 . 998 Å ) . Additionally , the location of the disulfide bonds that link the inner and outer layers are structurally conserved ( S3D Fig ) . These findings suggest that APCβ adopts an uncommon variation of the β-prism fold , which , together with the PCSK9 V-domain , constitutes a new subgroup of the β-prism family . Given the hydrophobic character of the loop at the bottom of the APCβ structure we reasoned that this loop has the potential to insert into the membrane upon binding . To test this , we took advantage of the intrinsic fluorescence of tryptophan , which is augmented upon exposure to a hydrophobic environment . The TgPLP1 CTD houses four tryptophan residues of which only two are surface exposed ( Fig 3A ) . We recorded the tryptophan fluorescence spectrum of purified rCTD in the presence or absence of liposomes of varying compositions . Incubation with liposomes resulted in an increase in the emission spectrum . Replacement of 10% of PC lipids with PC lipids modified with 1-palmitoyl-2-stearoyl- ( 7-doxyl ) -sn-glycero-3-phosphocholine ( 7-Doxyl ) , a collisional quencher , attenuated the increased fluorescence ( Fig 3B & 3C ) . These observations are consistent with insertion of the hydrophobic loop into the lipid bilayer . We next investigated the importance of the hydrophobic loop in TgPLP1 function . We used CRISPR/CAS9 gene editing to generate mutant parasites expressing TgPLPl with a loop that is symmetrically shortened by two or four amino acids ( PLP1MWF and PLP1W , respectively ) as well as a mutant lacking the loop entirely ( PLP1Δloop ) ( Fig 4A ) . Immunofluorescence microscopy confirmed the micronemal sub-cellular localization of the hydrophobic loop deletion mutants ( S4 Fig ) . Additionally , all three loop mutants are secreted in a calcium-dependent manner as evidenced by excreted/secreted antigen ( ESA ) assays conducted in the presence or absence of the calcium chelator BAPTA-AM ( Fig 4B ) . Next , we tested the egress competence of these loop deletion mutants using four general criteria that have been seen in the TgPLP1 knockout strain ( Δplp1 ) [15]: ( 1 ) the presence of spherical structures in egressed cultures representing failed egress events that may be PVs that have failed to rupture; ( 2 ) formation of smaller plaques relative to wild type parasites; ( 3 ) inability to permeabilize the PV membrane ( PVM ) after calcium ionophore induction; and ( 4 ) delayed egress after calcium ionophore induction . Spherical structures containing parasites that failed to egress were observed in egressed cultures of PLP1W , PLP1MWF , and PLP1Δloop similar with those seen in Δplp1 ( Fig 5A ) . PLP1W , PLP1MWF , and PLP1Δloop parasites form smaller plaques compared to the parental ( WT ) strain , consistent to those observed for Δplp1 ( Fig 5B & 5C ) . Previous studies have shown that Δplp1 parasites immobilized with cytochalasin D treatment fail to permeabilize the PVM after induced egress with a calcium ionophore as compared to WT parasites . We therefore tested the ability of the hydrophobic loop deletion strains to permeabilize the PVM under the same conditions . PLP1W , PLP1MWF , and PLP1Δloop parasites all fail to permeabilize the PVM after the addition of 200 μM zaprinast , a phosphodiesterase inhibitor that activates the parasite protein kinase G to induce egress , consistent with what is observed for Δplp1 parasites ( Fig 5D ) . Finally , we monitored the extent to which deletions to the hydrophobic loop affect egress from the host cell . To address this , we infected HFF monolayers in a 96-well plate with WT , Δplp1 , and each of the loop mutants . Thirty hours post-infection the infected monolayers were treated with zaprinast to induce egress . Culture supernatants were assayed for lactate dehydrogenase ( LDH ) , which is released from infected host cells upon parasite egress . Δplp1 , PLP1W , PLP1MWF , and PLP1Δloop showed a marked decrease in LDH release compared to cells infected with WT parasites ( Fig 5E ) . These data suggest that integrity of the hydrophobic loop is critical for TgPLP1 function . Since shortening the hydrophobic loop resulted in an egress defect that mimics Δplp1 parasites , we tested how the amino acid composition of the loop influences function . We again used CRISPR/CAS9 to generate four mutant strains including a leucine to valine substitution ( PLP1MLWVF ) as well as three alanine substitution mutants ( PLP1MAAAF , PLP1MAWAF , and PLP1MLALF ) to probe the importance of the leucine and tryptophan residues in the hydrophobic loop . We then tested the subcellular localization of these constructs by immunofluorescence microscopy and confirmed that all mutant strains have micronemal localization of TgPLP1 ( S5 Fig ) . Additionally , all four mutants showed calcium-dependent secretion ( Fig 6A ) . Next , we tested the egress competence of these mutant strains using the same criteria described above . Whereas the most conservative mutant ( PLP1MLWVF ) was not retained within spherical structures , all of the alanine substitution mutants ( PLP1MLALF , PLP1MAWAF , and PLP1MAAAF ) were entrapped in such structures , consistent with defective egress ( Fig 6B ) . Next , we analyzed the plaque size of the mutants . PLP1MLWVF formed large plaques similar to those of WT parasites ( Fig 6C ) . PLP1MLALF , PLP1MAWAF , and PLP1MAAAF parasites , however , formed small plaques akin to Δplp1 ( Fig 6C & 6D ) . In PVM permeabilization assays , PLP1MLWVF was the only mutant that retained activity after induction with zaprinast . Neither PLP1MLALF , PLP1MAWAF , nor PLP1MAAAF parasites permeabilize the PVM after induction , thus resembling the defect observed in Δplp1 , PLP1W , PLP1MWF , PLP1Δloop strains ( Fig 6E , S1 Movie ) . Finally , we tested the extent to which amino acid substitutions to the hydrophobic loop affect egress from the host cell using the LDH assay described above . Consistent with the above findings , PLP1MLWVF is the only point mutant that retained the ability to egress from host cells . PLP1MLALF , PLP1MAWAF , PLP1MAAAF failed to egress in the duration of the experiment ( Fig 6F ) . Taken together these results indicate that amino acid identity and hydrophobic character of the loop are important for proper TgPLP1 function .
This paper provides new insight into the structure and function of APCβ , an apicomplexan specific membrane-binding domain associated with parasite egress and cell traversal . Previous studies established that the TgPLP1 CTD , which includes APCβ , has membrane-binding activity and is crucial for TgPLP1 function in egress [16]; however , the lipid binding specificity and structure of this domain remained unknown . The structure solved herein is essentially identical to the TgPLP1APCβ domain reported recently by Ni et al . [17] , but our work additionally defines the lipid binding specificity , provides evidence for insertion of the hydrophobic loop into membranes , and establishes a critical role for this loop in TgPLP1 function during parasite egress . Given the lipid asymmetry found in biological membranes [27] and the “inside-out” directionality of T . gondii egress , we tested the binding preference of TgPLP1 CTD for specific lipid head groups . We found that the TgPLP1 CTD has a preference for binding IL phospholipids , particularly PS and PE , and that it fails to efficiently recognize OL phospholipids such as PC and SM . These results suggest a working model of directional activity wherein phospholipid accessibility enhances TgPLP1 activity during egress and limits activity during subsequent invasion . In this model , micronemal secretion of TgPLP1 during egress delivers it to the PVM for putative binding to PS and PE . Upon escape from the PV , additional secretion of TgPLP1 targets PS and PE in the IL of the host plasma membrane to facilitate parasite exit from the infected cell . The model further posits that a lack of preferred phospholipid receptors on the OL of the target cell limits TgPLP1 activity during invasion , thereby allowing formation of the PV . That T . gondii is capable of wounding host cells in a TgPLP1-dependent manner when microneme secretion and TgPLP1 activity are enhanced in low pH medium supports this model . However , the leaflet-specific phospholipid composition of the PVM remains unknown . Although the initial topology of this membrane upon invasion is such that PE and PS would be exclusively in the cytosolic leaflet , reports have suggested extensive lipid remodeling of the PVM after invasion and during intracellular replication [29 , 30] . Also , T . gondii is known to secrete into the PV a PS decarboxylase , which converts PS to PE , implying the presence of PS in the PVM [31] . We attempted to create transgenic T . gondii that secrete a PS binding probe ( lactadherin C2-GFP ) into the PV , but the protein is largely retained in the parasite endoplasmic reticulum . An additional limitation of the working model is that the extent to which TgPLP1 acts upon the host plasma membrane during egress remains unknown . Nevertheless , other examples of directional activity exist including the predatory fungus Pluerotus ostreatus ( Oyster mushroom ) , which targets nematodes by secreting pleurotolysin A/B , a two-component MACPF pore-forming toxin . The pleurotolysin A subunit requires SM and cholesterol for binding , [32–34] which is consistent with its action , together with the MACPF B subunit , on exposed membranes of the target nematode . Perforin-1 also exhibits directional activity , but its specificity for a target membrane is influenced by the spacing of phospholipids from unsaturation of acyl chains rather than by recognition of head groups [35] . Defining the phospholipid binding specificity and the role of acyl chain spacing for other MACPF proteins might reveal additional examples of leaflet preference consistent with a directional activity model . Identifying molecular determinants of membrane recognition by apicomplexan PLPs is necessary to understand cytolytic parasite egress and cell traversal . To address this , we generated an expression construct of the TgPLP1 CTD and solved the 1 . 13 Å resolution crystal structure of the APCβ domain . Despite using a full-length CTD expression construct the diffraction data only contains the APCβ domain and lacks the CTT . Whether the lack of density is due to flexibility in the CTT or enzymatic cleavage remains unclear . The TgPLP1APCβ structure presented here is an unusual double-layered β-prism with pseudo threefold symmetry . Analysis of the crystal structure shows a hydrophobic loop that protrudes from one of the subunits , which we interrogated for membrane insertion via fluorescence of a tryptophan located within the loop . The observed increase in intrinsic tryptophan fluorescence in the presence of liposomes is consistent with that seen for perfringolysin O and pneumolysin in the presence of liposomes [36–38] . The TgPLP1 APCβ domain contains three other tryptophan side chains , but two are buried within the hydrophobic core and their spectra is likely unchanged by the addition of liposomes . The third tryptophan is surface exposed but is located in between the inner and outer layer of the beta prism , a position that is unlikely to insert into a lipid bilayer . Thus , we conclude that the tryptophan located at the tip of the protruding hydrophobic loop is probably responsible for changes in the fluorescence spectra reported here . These results are consistent with recently published molecular dynamic simulations that theorize insertion of the protruding hydrophobic loop into the membrane [17] . Insertion of the hydrophobic loop into the target membrane might be required for TgPLP1 to gain a strong foothold for the subsequent conformational rearrangements that accompany pore formation . Indeed , a tetra-alanine substitution to the hydrophobic loop or recombinant TgPLP1APC reduced membrane binding in an in vitro liposome sedimentation assay [17] . To further probe the extent to which the hydrophobic loop is important to TgPLP1 function we generated parasite lines that were genetically engineered to harbor mutations that probed the length of the hydrophobic loop as well as the importance of particular residues therein . Previous work has shown that TgPLP1 is required for efficient lysis of the PVM and the subsequent egress from the host cell [15] . Additionally our previous work has also identified the TgPLP1 CTD as being critical for function [16] . The results presented here show that mutations that affect the length and character of the hydrophobic loop recapitulate the egress defect seen with Δplp1 parasites . Namely , these loop mutant strains form spheres in egressed cultures , form small plaques relative to wild type parasites , fail to permeabilize the PVM when motility is arrested , and have a delay in egress after induction . The most striking of these observations is the single alanine substitution of the tryptophan located at the tip of the hydrophobic loop . Tryptophan often contributes to the binding interface of membrane binding proteins because of its properties as both a hydrophobic and polar molecule . Indeed , many MACPF/CDC proteins utilize tryptophan residues for binding target membranes[36 , 39 , 40] That a single tryptophan to alanine substitution recapitulates the egress defect observed in parasites lacking TgPLP1 altogether underscores the importance of the hydrophobic loop for efficient egress . The extent to which a similarly important loop is conserved in other apicomplexan PLPs remains unknown in the absence of other structural studies . The work presented here along with that of Ni et al [17] is a key first step in our understanding of how apicomplexan PLPs recognize and bind to membranes . T . gondii encodes two PLPs , TgPLP1 and TgPLP2 , but only TgPLP1 has been shown to be important for cytolytic egress . TgPLP2 is not expressed in the tachyzoite stage . A related apicomplexan parasite and causative agent of malaria , Plasmodium spp . , has a more complex life cycle and encodes an expanded family of five PLPs , PPLP1-5 [41–46] . Plasmodium PLPs have been implicated in cytolytic egress as well as in cell traversal , a process required for migrating through tissue [47–49] . Despite some sequence similarity between the T . gondii and Plasmodium spp . PLPs ( 55% similarity between TgPLP1 and P . falciparum PLP1 ) the molecular determinants that govern membrane recognition and binding by Plasmodium PLPs remain unclear . More thorough structural and functional studies on Plasmodium PLPs are needed to further our understanding of how apicomplexan PLPs recognize their target membranes and shed light on how these differing activities have evolved within a common protein scaffold . Together with previous findings , the current work brings us closer to a molecular understanding of how TgPLP1 facilitates parasite egress . The work also provides a foundation upon which future work on TgPLP1 and other apicomplexan PLPs are expected to illuminate the molecular basis of differential function in egress and cell traversal .
Lipid compositions were chosen to mimic the inner leaflet or outer leaflet of biological membranes as described [27] . Liposomes were prepared from 10 mg/mL stock solutions of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC , Avanti ) , 1-palmitoyl-2-oleoyl-sn-glycero-3-[phospho-L-serine] ( POPS , Avanti ) , L-α-phosphatidylethanolamine ( POPE , Avanti ) , Sphingomyelin ( SM , Avanti ) , L-α-phosphatidylinositol ( PI , Avanti ) , and cholesterol ( Avanti ) in the ratios described in the figure captions . Two μmol total lipids were dried under a nitrogen stream . Dried lipids were incubated in 1 mL of rehydration buffer ( 100 mM NaCl , 1 mM CaCl2 , 1 mM MgCl2 , 50 mM Tris pH 7 . 4 ) for 30 min and subsequently vortexed until lipid film was completely dissolved . Hydrated lipid suspension was subjected to 3 freeze/thaw cycles , alternating between dry ice/ethanol bath and warm water bath . The rehydrated lipid solution was extruded through 400 nm pore-size filters using a mini-extruder ( Avanti ) to produce liposomes . Two hundred nmol of liposomes were incubated with 0 . 5–1 nmol of recombinant proteins at 37°C for 15 min in a final volume of 200 μL . After incubation the reaction mixture was diluted with 1 mL 85% sucrose and layered with 2 . 8 mL 0f 65% sucrose and 1 mL of 10% sucrose . The reaction was centrifuged at 115 , 000g at 4°C for 16 h ( Sorvall rotor AH-650 ) . Fractions were collected and analyzed by western blot . Band intensity was quantified using Image J . The liposome binding efficiency was calculated as: %Bound=IBoundIBound+IUnbound×100 ( 1 ) High-five or Sf9 cells were expanded to six 1 L volumes . These were used to seed 20 L of media ( Expression Systems ) at 2 . 0 x 106 cells/mL in a 36 L stir tank bioreactor . The culture was infected with recombinant baculovirus at a multiplicity of infection of 5 . The reactor was incubated at 27°C with stirring and sparged with air for 72 h . The culture was pumped out and centrifuged at 1000g and 4°C for 40 min to pellet the cells . The media was collected , and batch bound with 0 . 5 mL per liter of Roche cOmplete His-Tag purification resin for 4 h at 4°C with stirring . Resin was loaded onto a column and washed with 50 mM Tris ( pH 8 . 0 ) , 300 mM NaCl and 20 mM imidazole . Protein was eluted with 50 mM Tris ( pH 8 . 0 ) , 300 mM NaCl and 250 mM imidazole . Wash and elution fractions were collected and run on SDS page to determine purity and protein location . The poly-histidine tag was subsequently cleaved with TEV-protease at 4°C overnight and separated from the CTD by immobilized metal affinity chromatography . CTD fractions were further purified by anion exchange chromatography and concentrated to 22 mg/mL . TgPLP1APCβ crystals were obtained from the Molecular Dimensions Clear Strategy II screen set up with 1 μL of protein and 1 μL of reservoir containing 100 mM sodium cacodylate at pH 6 . 5 , 0 . 15 M KSCN , and 18% PEG 3350 . Hanging drop plates were equilibrated at 20°C . The needle-like crystals were mounted on a cryoloop and transferred to a cryoprotectant solution containing the reservoir solution supplemented with 20% ethylene glycol . Native crystals were flash frozen in liquid nitrogen for data collection . For phase problem solution data sets , a two-hour soak was performed in a synthetic mother liquor supplemented with 20 mM KI and flash frozen in mother liquor supplemented with 20% ethylene glycol . Data sets were collected at LS-CAT beamline 21ID-D and 21ID-G . The crystals belonged to the C121 space group with cell dimensions of a = 101 . 98 Å , b = 50 . 85 Å , c = 105 . 34 Å , α = 90° , β = 90 . 13° , and γ = 90° . Data reduction and scaling were performed with autoPROC [50] . Phasing was performed using the AutoSol in Phenix [51] . Initial solution was obtained by SAD on KI soaked crystals . The partial solution was then used as a molecular replacement model for the high-resolution data sets . Model building was performed in COOT [52] and refinement in PHENIX . The TgPLP1APCβ crystal structure was refined to a crystallographic Rwork of 14 . 07% and a Rfree of 16 . 36% The final structure was analyzed with validation tools in MOLPROBITY . Structural visualization was performed via PyMOL . Intrinsic tryptophan emission intensity was measured on a Biotek Snergy H1 equipped with monochromators for both excitation and emission . The emission spectra were recorded between 320–400 nm with a step size of 5 nm . The excitation wavelength was set to 280 nm . The emission intensity was recorded in the absence and presence of PC , PC/PE , PC/PS liposomes and liposomes that had 10 mol% of the PC lipid replaced with 1-palmitoyl-2-stearoyl-7-doxyl ) -sn-glycero-3-phosphocholine ( 7-Doxyl ) ( Avanti ) . Raw data were corrected for the baseline spectrum . Spectra were collected as three independent replicates with three technical replicates each . All cells and parasites were maintained in a humidified incubator at 37°C and 5% CO2 . Human foreskin fibroblast cells ( HFF , ATCC CRL-1634 ) were maintained in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% Cosmic Calf serum , 20 mM HEPES pH 7 . 4 , 2 mM L-glutamine and 50 μg/mL penicillin/streptomycin . All T . gondii parasites were maintained by serial passaging in HFF cells and were routinely checked for mycoplasma contamination . PLP1 hydrophobic mutant strains were generated by CRISPR-Cas9 gene editing using RHΔku80DsRed [16] as a parental strain such that the mutant strains express DsRed in the parasitophorous vacuole . Briefly , 20 bp of plp1-specific guide RNA sequence targeting the hydrophobic loop was inserted into pCRISPR-Cas-9-Ble using site-directed mutagenesis to generate the pAG1 plasmid . An annealed synthetic oligonucleotide repair template pair that encoded the appropriate mutations and a silent mutation to replace the NGG cut site ( IDTDNA ) was mixed with pAG1 and precipitated by addition of ethanol . Fifty million tachyzoites were transfected by electroporation in a 4 mm gap cuvette using a Bio-Rad Gene Pulser II with an exponential decay program set to 1500 V , 25 μF capacitance and no resistance and immediately added to a confluent HFF monolayer in a T25 flask . Parasites were selected with 50 μg/mL phleomycin 24 h post transfection for 6 h and subsequently added to a confluent HFF monolayer in T25 flask . Parasites were incubated for 36–48 h . Clonal populations were isolated and a 1 kb fragment generated from extracted DNA was sequenced for confirmation of the correct mutation . Induced excreted secreted antigens ( ESA ) were performed as previously described [53] . Briefly , 2 x 107 parasites were incubated at 37°C for 2 min in 1 . 5 mL Eppendorf tubes with DMEM containing 10 mM HEPES , pH 7 . 4 and supplemented with 1% ethanol . Parasites were separated by centrifugation ( 1000g , 10 min , 4°C ) . Samples of the pellet and supernatant were run on a 10% SDS-PAGE gel and analyzed by western blot . ESAs were performed in the presence and absence of 100 μM BAPTA-AM . Infected monolayers were fixed with 4% formaldehyde for 20 min and washed with PBS . Slides were with 0 . 1% Triton X-100 for 10 min and blocked with 10% fetal bovine serum ( FBS ) , 0 . 01% Triton X-100 in PBS ) . Slides were subsequently incubated with rabbit anti-PLP1 ( 1:500 ) and mouse anit-MIC2 ( 6D10 , 1:250 ) diluted in wash buffer ( 1% FBS , 1% Normal Goat Serum ( NGS ) , 0 . 01%Triton X-100 in PBS ) for 1 h at RT . After three washes , slides were incubated for 1 h at RT with Alexa Fluor goat anti-rabbit and goat anti-mouse secondary antibodies ( Invitrogen Molecular Probes ) diluted in wash buffer ( 1:1000 ) . Slides were washed and mounted in Mowiol prior to imaging . Infected monolayers of HFF cells in T25 flasks were washed with phosphate-buffered saline ( PBS ) . Parasites were liberated by scraping the monolayer and passaging through a 27-gauge needle . Liberated parasites were filtered through a 3 μm filter ( Millipore ) and counted on a hemocytometer . Monolayers of HFFs in individual 6-well plates were inoculated with 50 parasites and incubated , undisturbed , for one week and fixed with 2% ( w/v ) crystal violet . Plaque size was quantified using ImageJ . Monolayers were infected with tachyzoites expressing DsRed and allowed to replicate for 30 h . Infected monolayers were washed twice with warmed Ringer’s buffer ( 155 mM NaCl , 3 mM KCl , 2 mM CaCl2 , 1 mM MgCl2 , 3 mM , NaH2PO4 , 10 mM HEPES , 10 mM Glucose , 1% FBS , pH 7 . 40 ) . Ionophore treatment was initiated by addition of an equal volume of 2 μM cytochalasin D and 400 μM zaprinast . Fluorescence image time series were collected every 10 sec for 5 min . Infected monolayers of HFF cells in T25 flasks were washed with phosphate-buffered saline ( PBS ) and liberated by scraping the monolayer and passaging through a 27-gauge needle . Liberated parasites were filtered through a 3 μm filter ( Millipore ) , counted on a hemocytometer , and centrifuged at 1000g for 10 min . Parasites were resuspended to a density of 5 x 105 parasites/mL and each well of a 96-well flat bottom culture plate was inoculated with 5 x 104 tachyzoites of the appropriate strain . Infected cells were washed with warm Ringer’s buffer and treated with zaprinast diluted to 200 μM in Ringer’s buffer , Ringer’s with an equal volume of dimethyl sulfoxide ( DMSO ) , or cell lysis reagent ( BioVision ) diluted in Ringer’s . Treated plates were incubated at 5% CO2 and 37°C for 20 min . Plates were placed on ice and 50 μl of each well was transferred to a 96-well round-bottom plate . Round-bottom plates were centrifuged at 500g for 5 min . Ten μl of solution was tested for lactate dehydrogenase ( LDH ) following the manufacturer’s protocol ( Biovision ) . This article contains supporting information online .
|
The intracellular parasite Toxoplasma gondii infects many hosts including humans . Infected people with a weak immune system can suffer severe disease when the parasite replicates uncontrolled via repeated cycles of cell invasion , intracellular growth , and exit , resulting in cell death . Previous studies showed that T . gondii encodes a pore-forming protein , TgPLP1 , which contains an unusual domain that is crucial for efficient exit from both the parasite containing vacuole and the host cell . However , how TgPLP1 recognizes and binds to the appropriate membrane is unclear . Here we use a combination of biochemistry , structural biology , and parasitology to identify a preference of TgPLP1 for specific lipids and show that a loop within the structure of the membrane-binding domain inserts into the target membrane and is necessary for exit from the parasite containing vacuole . Our study sheds light into the determinants of membrane binding in TgPLP1 and may inform the overall mechanism of pore formation in similar systems .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] |
[
"medicine",
"and",
"health",
"sciences",
"vesicles",
"immune",
"cells",
"chemical",
"compounds",
"crystal",
"structure",
"viral",
"transmission",
"and",
"infection",
"antigen-presenting",
"cells",
"immunology",
"condensed",
"matter",
"physics",
"microbiology",
"parasitic",
"diseases",
"organic",
"compounds",
"amino",
"acids",
"crystallography",
"cellular",
"structures",
"and",
"organelles",
"aromatic",
"amino",
"acids",
"lipids",
"solid",
"state",
"physics",
"animal",
"cells",
"proteins",
"chemistry",
"liposomes",
"cell",
"membranes",
"physics",
"biochemistry",
"host",
"cells",
"cell",
"biology",
"organic",
"chemistry",
"virology",
"tryptophan",
"biology",
"and",
"life",
"sciences",
"cellular",
"types",
"physical",
"sciences"
] |
2018
|
Structural basis of Toxoplasma gondii perforin-like protein 1 membrane interaction and activity during egress
|
Technological advances have unraveled the existence of small clusters of co-active neurons in the neocortex . The functional implications of these microcircuits are in large part unexplored . Using a heavily constrained biophysical model of a L5 PFC microcircuit , we recently showed that these structures act as tunable modules of persistent activity , the cellular correlate of working memory . Here , we investigate the mechanisms that underlie persistent activity emergence ( ON ) and termination ( OFF ) and search for the minimum network size required for expressing these states within physiological regimes . We show that ( a ) NMDA-mediated dendritic spikes gate the induction of persistent firing in the microcircuit . ( b ) The minimum network size required for persistent activity induction is inversely proportional to the synaptic drive of each excitatory neuron . ( c ) Relaxation of connectivity and synaptic delay constraints eliminates the gating effect of NMDA spikes , albeit at a cost of much larger networks . ( d ) Persistent activity termination by increased inhibition depends on the strength of the synaptic input and is negatively modulated by dADP . ( e ) Slow synaptic mechanisms and network activity contain predictive information regarding the ability of a given stimulus to turn ON and/or OFF persistent firing in the microcircuit model . Overall , this study zooms out from dendrites to cell assemblies and suggests a tight interaction between dendritic non-linearities and network properties ( size/connectivity ) that may facilitate the short-memory function of the PFC .
Small , tightly interconnected “clusters” of cortical neurons have recently been discovered in regions such as the visual , somatosensory and prefrontal cortex ( PFC ) [1]–[6] , yet their role in cognitive processes remains unexplored . In the PFC , such microcircuits have been suggested to participate in persistent activity , the cellular correlate of working memory , but this hypothesis has not been rigorously tested [7] , [8] . Towards this goal , Papoutsi and colleagues developed a layer 5 ( L5 ) PFC microcircuit model , heavily constrained against experimental data , and showed that such microcircuits can serve as tunable modules of persistent activity [9] . What remain unclear are the biophysical and anatomical mechanisms that allow the induction ( ON ) and can cause termination ( OFF ) of persistent firing in such modules . Previous studies have uncovered the NMDA synaptic current [10] , [11] and the delayed afterdepolarization ( dADP ) [12] as two important biophysical mechanisms that contribute to persistent activity initiation . However , the structure and size of the network studied varied greatly , from a couple of cells [13] to networks of hundreds to thousands of neurons ( in silico ) [14] , [15] along with networks of unknown size in the slice preparation or in vivo . Moreover , in the majority of these studies , pharmacological manipulations ( e . g . carbahol addition [16] ) and/or unrealistic connectivity properties ( e . g . long conduction delays [13] ) were needed for the phenomenon to emerge . As a result , the minimum size of a network capable of expressing persistent activity under physiological conditions remains unknown , yet critical for understanding the mechanisms underlying its induction [17] . Like network size , the role of dendritic nonlinearities in persistent activity emergence is also ambiguous . First , the NMDA receptors , which are primarily located in the dendrites of cortical pyramidal neurons , where found to be imperative for the in vivo initiation of persistent activity in the PFC [10] . Second , the generation of dendritic plateau potentials at the basal dendrites of L5 PFC neurons [18] , [19] has been suggested to underlie the somatic depolarization observed during Up states [20] . However , while these studies concern a known player – the NMDA receptor - a link between dendritic regenerative events and persistent activity emergence has yet to be established . Finally , little is known about how key characteristic features of persistent activity like stimulus-specificity , resistance to distracters and termination induced by behavioral actions [21] can be implemented by neural tissue . For example , the only candidate mechanism for termination is inhibitory input which was shown to cease Up states [22] . Regarding stimulus-specificity , Sidiropoulou and Poirazi , 2012 used a computational model of a single L5 PFC pyramidal neuron to show that location of activated synapses along the basal dendrites and action potential timing could serve as encoding and decoding mechanisms , respectively , of stimulus-selective induction [23] . Identifying such information in the response pattern of these neurons is particularly important as it may signal the upcoming state transitions to downstream neurons , setting the ground for the subsequent behaviour actions that will terminate persistent firing . However , whether such mechanisms are also relevant at the network level or whether other mechanisms are implicated in not known . Here , we use our recently developed microcircuit model [9] to investigate the mechanisms that underlie persistent activity emergence ( ON ) and termination ( OFF ) at the dendritic , neuronal and network levels and search for the minimum network size required for expressing these states within physiological regimes . Moreover , we search for mechanisms that may underlie persistent activity maintenance upon presentation of distracting stimuli and identify network characteristics that code for the upcoming state transitions .
To investigate the effect of network size on persistent activity emergence , we varied the number of pyramidal neurons in the microcircuit and recorded the result of this manipulation on the probability of induction . We found that reducing the number of pyramidal neurons from 7 to 5 completely abolished persistent activity , whereas adding more neurons increased the probability of induction to 100% ( Figure 1C , diamonds ) . The latter could be due to stronger synaptic drive within the network ( a connectivity effect ) or due to having more neurons that propagate signals ( a size effect ) . To discriminate between these two possibilities , we varied the network size while keeping the number of recurrent connections per neuron fixed to that of a size 7 network ( 31 synapses per neuron: 6×5 pyramidal-to-pyramidal inputs plus 1 autapse ) . Persistent activity emerged in all cases tested ( Figure 1C , squares ) , even in a microcircuit of size 2 . On the contrary , changing the synaptic drive of each neuron ( in a network of size 7 ) had a strong effect: reducing the pyramidal-to-pyramidal synaptic contacts by 20% , from 5 to 4 ( total inputs per neuron: from 31 to 25 ) abolished persistent activity , whereas the respective increase in connections from 5 to 6 ( total inputs per neuron: from 31 to 37 ) led to 100% probability of induction ( Figure 1D , diamonds ) . Varying the connectivity strength between pyramidal and interneurons had less pronounced effects: persistent activity could emerge even when the pyramidal-to-interneuron connections tripled ( Figure 1D , squares ) or the interneuron-to-pyramidal connections doubled ( Figure 1D , triangles ) . The validity of these findings was also tested under conditions that more closely approximate the in vivo situation , where neurons constantly receive synaptic barrages that alter their dynamics . Background synaptic activity as reported in vivo during quiet wakefulness and not under anaesthesia [26] , [27] was added to both pyramidal neurons and interneurons ( Figure S1B ) and the same experiments were repeated . Results were very similar to the previous analysis . Excitatory synaptic transmission was still the determinant factor for persistent activity emergence ( Figures 1E , F ) . The main effects of background synaptic activity where: ( a ) to slightly reduce the synaptic drive required for persistent activity ( emergence in a network with 6 pyramidal neurons instead of 7 , Figure 1E ) and ( b ) to increase the tolerance of persistent activity emergence to changes in inhibitory transmission ( connection strength between pyramidals and interneurons , Figure 1F ) . Overall , these results suggest that the strength of excitatory-to-excitatory transmission , as opposed to the network size , is the crucial factor for persistent activity induction in the microcircuit , under both in-vitro and in-vivo like conditions . Given that excitatory-to-excitatory transmission is crucial for persistent activity induction and NMDA receptors play a key role in shaping excitatory synaptic transmission in L5 PFC pyramidal neurons [19] , [28] , we next examined their contribution to the size vs . synaptic drive argument . Activation of NMDA receptors was recently found to be imperative for persistent activity emergence in vivo [10] and the generation of Up-states in acute slices [20] , [29] , [30] . Since both of these phenomena are characterized by long-lasting depolarizations , it can be assumed that the role of NMDA currents is to provide or sustain these depolarizations through regenerative dendritic events such as NMDA spikes [19] . We thus investigated whether and how the generation of NMDA spikes may influence persistent activity induction in the microcircuit model . We first assessed whether NMDA spikes are inducible in our pyramidal neuron models using four different iNMDA-to-iAMPA ratios: 1 . 1 , 1 . 5 , 1 . 9 , and 2 . 3 . In all cases , only the NMDA current increased and the ratios were calculated under voltage clamp conditions ( Figure S1A ) . Stimulation of increasing number of synapses ( 5–50 , with step of 5 ) at the basal dendrite of a single pyramidal neuron with 2 pulses at 50Hz led to a non-linear increase in the somatic EPSP amplitude for ratios 1 . 9 and 2 . 3 ( Figure S2A ) , that was also evident in the EPSP half width ( Figure S2B ) . Moreover , comparison of model and experimental data regarding the EPSP amplitude and half width ( for a ratio of 2 . 3 ) measured under single-pulse and paired pulse ( at 50Hz ) stimulation revealed a close mapping between simulated and experimental attributes of synaptic integration in basal dendrites of L5 PFC pyramidal neurons [19] ( Figure S2E ) . These findings are characteristic of dendritic NMDA spike generation . In addition , the ability to induce NMDA spikes in a given pyramidal neuron with input from connecting synapses alone ( 6 neurons×5 inputs +1 autapse = 31 synapses ) was assessed . Representative traces resulting from the activation of 31 synapses ( 2 pulses at 50Hz: diamonds or 1 pulse: squares ) in the basal dendrite of a pyramidal neuron model , under blockade of Na+ channels are shown in Figure 2A . The dendritic EPSP half width and amplitude are shown in Figures 2B and 2C , respectively for both the paired ( at 50Hz ) and single pulse protocols . Enhancement of dendritic EPSP amplitude and half width , reminiscent of NMDA spike generation [19] , was seen primarily for a ratio of 2 . 3 ( small increases are also seen for a ratio of 1 . 9 ) under the paired -but not the single- pulse stimulation , suggesting the occurrence of NMDA spikes . These results are in very good agreement with experimental recordings [19] ( Figure S2 ) . Emergence of persistent activity was strongly correlated with the generation of NMDA spikes . As shown in Figure 2D , lack of prominent NMDA spikes ( ratio 1 . 1–1 . 9 ) was associated with zero probability of persistent activity , whereas generation of large NMDA spikes ( ratio 2 . 3 ) was associated with an induction probability of 86% . Importantly , reducing the NMDA decay time constant ( from τ = 107 ms to τ = 18 ms , Figure 2D , triangle ) or blocking the NMDA receptors ( 90% reduction in conductance , while compensating for reduced excitability by increasing the AMPA conductance ) under conditions that normally supported persistent activity ( Figure 2D , square ) also abolished the persistent state . The above experiments were repeated in the presence of background synaptic activity to establish their validity under in-vivo like conditions . The only difference observed was a reduction in the amount of NMDA current required for persistent activity emergence: the induction probability for ratio of 1 . 9 climbed from zero to 0 . 58 ( Figure 2E ) . These findings suggest that background synaptic input facilitates persistent activity induction by enhancing NMDA spikes appearing at a ratio of 1 . 9 ( Figure 2B , Figure S2B ) , which would otherwise be ineffective . Representative traces with and without persistent activity in the presence of background synaptic input for the 1 . 9 ratio are shown in Figure 2F . To investigate whether NMDA spikes are generated during persistent activity , we evaluated whether individual pyramidal neurons express such NMDA spikes when embedded in the microcircuit model and not in isolation , as was done above . Persistent activity induction in the microcircuit was associated with a larger depolarization at the soma ( Figure 3A , top and Figure 3E ) and a larger inward NMDA current ( Figure 3A , bottom and Figure 3B ) , compared to the non-persistent state . Representative traces for the two cases are shown in Figure 3A ( black trace: ratio 1 . 1 , grey trace: ratio 2 . 3 ) . To investigate whether this synaptic current supports NMDA spikes at the basal dendrites , we added to the microcircuit a pyramidal neuron in which we blocked the somatic and axonal fast sodium channels . This ‘silent’ pyramidal neuron received the same synaptic activity as the other pyramidals , but did not contribute to the microcircuit activity . As shown in Figure 3C , the characteristic depolarization plateau potential of NMDA spikes is absent for ratios 1 . 1 and 1 . 5 , emerges with small width for a ratio of 1 . 9 and becomes pronounced for a ratio of 2 . 3 . In the presence of background synaptic activity , these features appear at smaller ratios: small plateau potentials are evident at a ratio of 1 . 5 , they become pronounced at a ratio of 1 . 9 and they always lead to persistent activity for a ratio of 2 . 3 ( Figure 3D ) . These results illustrate that NMDA spikes do emerge in pyramidal neurons participating in the microcircuit , under conditions that enable persistent activity induction ( ratio of 1 . 9 & 2 . 3 ) . Finally , we tested whether the mechanism of action of NMDA spikes is to provide long-lasting somatic depolarizations on top of which persistent activity can ride . Specifically , we measured the somatic depolarization of a pyramidal neuron model during the last 100 ms of stimulus presentation ( Figure 3A , dotted box ) for the four different ratios . Generation of NMDA spikes for a ratio of 2 . 3 ( calculated over 50 trials ) induced , on average , a large plateau potential ( ∼16 mV ) at the soma; smaller plateaus were seen for ratios of 1 . 1–1 . 9 ( Figure 3E ) . This depolarized state , which is not seen for small ratios , is proposed to underlie the persistent spiking activity . To further investigate this hypothesis we asked whether the 100 ms depolarizing potential is also different between trials that led to persistent firing vs . trials that didn't , this time for a fixed iNMDA-to-iAMPA ratio . Indeed , for a ratio of 2 . 3 , this plateau potential was significantly larger in the persistent compared to the transient response trials ( p value<0 . 001 ) ( Figure S4A ) . These results suggest that somatic depolarizations resulting from NMDA spike generation may underlie persistent activity . If this was truly the case , injection of a current at the soma could potentially substitute the need for NMDA spikes at the basal dendrites . To test this hypothesis , we blocked NMDA receptors in all pyramidal neurons and delivered a depolarizing current throughout the stimulus and delay periods . We found that , currents resulting in somatic depolarization potentials similar to the ones seen for a ratio of 2 . 3 ( 16mV ) also supported persistent activity ( Table 1 ) . This raised the issue that other intrinsic mechanisms may support persistent activity in small microcircuits , if they can build a similarly-sized depolarizing plateau potential at the soma . To address this question we blocked NMDA receptors in all pyramidal neuron models and independently enhanced the conductance of each excitatory ionic mechanism by a factor of 2–5 . Examined mechanisms included the Naf , NaP , CaL , CaT , CaR , CaN , as well as the conductance of the h current that has been shown to participate in persistent activity induction [31] . None of these manipulations resulted in persistent activity , reinforcing our previous findings that NMDA spikes at the basal dendrites support persistent spiking activity in small microcircuits , via the build-up of long-lasting depolarizing plateau potentials [32] . The only mechanism able to replace NMDA spikes was the dADP , a mechanism activated by cholinergic input in L5 PFC pyramidal neurons , which was previously linked to persistent activity by us and others [12] , [23] . However , it should be noted that the amplitude of the dADP required for persistent activity emergence under NMDA blockade was 15 mV , namely much larger than the physiologically reported values ( 1–4 mV [23] ) . In agreement with prior work [23] , these results suggest that while intrinsic ionic conductances , and particularly the dADP , can contribute to persistent activity , NMDA receptors are crucial for its emergence . In sum , our simulations predict that dendritic nonlinearities alone , through the generation of NMDA spikes and a subsequent build up of somatic depolarization , act as a switch for entering a sustained firing state in L5 PFC microcircuits . The predicted dependence on NMDA dendritic spikes may seem contradictory to previous work , whereby both large- [15] and small- scale neuronal network models [13] without NMDA receptors supported persistent activity . However , those models were not biophysically constrained in several aspects , including their connectivity properties . Moreover , blockade of NMDA receptors in the PFC was shown to abolish prolonged spiking activity [10] , [33] , suggesting that this dependence may be a region-specific effect . To further investigate this issue , we simulated a large scale network of fully connected 250 neurons ( 200 pyramidal and 50 interneurons ) whereby NMDA and GABAB receptors were completely blocked ( as in earlier reports ) and constraints regarding synaptic delays between pyramidal neurons were relaxed ( delays for excitatory-to-excitatory connections were drawn from a Gaussian distribution with μ = 40 ms and σ = 10 ms ) . We found that the resulting asynchronicity in conjunction with the much larger size of the network were sufficient for persistent activity to emerge ( Figure 4B ) with high probability ( 82% ) , in agreement with earlier work [13] . Note the elimination of the somatic plateau potential during persistent firing generated in our large-scale networks ( Figure 4B ) compared to the microcircuit ( Figure 4A ) . To test whether asynchronicity produced by long conduction delays was sufficient to replace NMDA-induced depolarizations , we blocked NMDA receptors in the microcircuit model and allowed for conduction delays similar to the ones used in Figure 4B . In this case , persistent activity could not be induced in any of the trials tested ( Figure 4C ) . Similarly , we asked if reverberating activity in a large scale network with short conduction delays ( similar to the ones used in the microcircuit ) could support persistent firing under NMDA blockade ( Figure 4D ) and again failed to see sustained responses . To elucidate the mechanisms that underlie persistent firing in the two network configurations , we recorded the net excitatory synaptic current to each pyramidal neuron , under conditions leading to persistent activity emergence with the same probability ( 86% for the microcircuit and 82% for the large-scale ) . As shown in Figure 4E–F , the net excitatory synaptic current per neuron is considerably larger in the microcircuit compared to the large scale network and is in great part mediated by the NMDA receptors ( Figure 4F , red trace ) . Note that the total synaptic current that flows through all pyramidal neurons is , of course , much greater in the large scale network than in the microcircuit . These findings suggest that in small , biophysically constrained PFC microcircuits , where conduction delays are short , synaptic input to pyramidal neurons through NMDA receptors is necessary for persistent activity induction . This necessity disappears in large scale networks whereby long conduction delays in conjunction with multiple reverberating connections are sufficient to bridge depolarizations over time , thus prolonging spiking . The persistent activity recorded during working memory terminates normally upon the execution of motor actions [34] or prematurely as a result of distracting stimuli , in which case performance drops significantly [35] . However , the mechanisms underlying persistent activity termination remain unclear . Inhibition is currently the primary candidate , as it has been found that Up states , a condition similar to persistent firing , are terminated by activation of interneurons [22] . Since PFC interneurons receive feed forward excitation during working memory tasks [36] , we investigated their role in persistent activity termination . Delivery of a second excitatory stimulus ( 10 events at 100Hz ) to the interneuron models one second after induction resulted in termination of persistent with a probability less than 0 . 5 . Representative traces of a terminating and a stable trial are shown in Figure 5A . Increased inhibitory input resulted in a slight increase of the termination probability whereas activation of the dADP mechanism had the opposite effect ( Figure 5B ) . Specifically , dADP activation ( 2 mV ) led to a significant decrease ( ∼21% ) in the termination probability ( Figure 5B , circles ) . These results are the first to propose a role of the dADP mechanism in the stabilization of persistent activity . Since the dADP primarily emerges following acetylcholine or glutamate action and is modulated by dopamine [12] , our data suggest that neuromodulatory effects are likely to have a key role in the maintenance of persistent firing . Finally , termination was significantly harder in the presence of ongoing network activity ( Figure 5B , diamonds ) , suggesting a new role for this activity in stabilizing persistent firing . In sum , these results show that termination of persistent activity depends on the strength of the synaptic input and is negatively modulated by dADP activation . The abovementioned experiments showed that in small , biophysically constrained PFC microcircuits , dendritic events underlined by NMDA spikes build a somatic depolarization plateau on top of which stable persistent activity rides . Termination of this activity can be achieved via an inhibitory input . To substantiate our results regarding the key role of synaptic currents in persistent firing , we asked whether the response properties of the microcircuit ( i . e . synaptic mechanisms and activity features ) during stimulus presentation contained predictive information regarding both the induction and the termination of the persistent state . Termination was caused , as previously , by a second stimulus delivered to the interneurons 1 s after the inducing stimulus ( 100 synapses activated with 10 events at 100Hz ) . Induction and termination were evaluated under control conditions over 500 and 423 trials , respectively . We used a linear SVM classifier ( see Methods for details ) to identify features of the stimulus-induced response that can serve as predictive markers for the microcircuit output . The examined features included measures of a ) network spiking activity , b ) single-cell spiking activity and c ) single-cell synaptic currents . For each feature tested , the SVM was trained with 100 trials ( training set ) exhibiting the desired phenotypes ( i . e . persistent activity vs . transient response or terminated vs . stable persistent activity ) and prediction accuracy , sensitivity and specificity were estimated on a set of 30 previously unseen trials ( test set ) . In all cases , a strict threshold of 70% sensitivity and specificity was used for the identification of informative features . We started by examining the predictive power of synaptic currents measured at a single pyramidal neuron . We found that both NMDA-mediated slow excitation and GABAB-mediated slow inhibition coded for the upcoming state transitions . Specifically , the iGABAB during stimulus presentation predicted induction and termination with an accuracy of 82%±8% ( Figure 6A ) and 88%±5% ( Figure 6B ) , respectively . Although counterintuitive , the total GABAB current was significantly larger in persistent than transient response trials ( p value<0 . 001 , non-parametric parametric U test , Figure S3C ) . The same was true for stable compared to terminated trials ( p value<0 . 001 , non-parametric parametric U test , Figure S3D ) . A similar trend was seen for the NMDA current , which predicted induction and termination with accuracies of 74%±6% ( Figure 6A ) and 80%±9% ( Figure 6B ) , respectively . In this case , the total NMDA current during stimulus presentation was significantly larger in persistent than in transient response trials ( p value<0 . 05 , non-parametric parametric U test , Figure S3E ) but not in stable vs . terminated trials ( p value>0 . 05 , non-parametric parametric U test , Figure S3F ) . Note that , in order to study the impact of their slow kinetics , the NMDA and GABAB currents were filtered using a Butterworth low pass filter . Indicative traces before and after the filtering of the NMDA and GABAB currents are shown in Figure S3A , B . The predictive power of NMDA and GABAB currents during stimulus presentation does not explain how these mechanisms determine the induction and termination of persistent activity . We hypothesize that their interactions shape the somatic plateau potential towards the end of the stimulus as described in Figure 3E . We thus tested whether this plateau potential is significantly different not only between persistent and transient response trials as shown in Figure S4A , but also between stable and terminated trials . Indeed , the somatic depolarization ( measured in the last 100 ms of the stimulus presentation ) in stable trials was significantly larger than in terminated ones ( p value<0 . 001 ) ( Figure S4B ) , suggesting that the magnitude of the somatic membrane depolarization is a determining factor for both the emergence and termination of persistent firing in the PFC microcircuit . Finally , previous work from our lab showed that temporal features ( first spike latency and first inter-spike-intervals ) of the stimulus-induced response of a single L5 PFC pyramidal neuron model code for an upcoming transition to a persistent state [23] . We thus tested whether this type of coding is preserved at the microcircuit level . State transitions in the microcircuit could not be predicted by the first few ISIs of the pyramidal neuron responses . A possible explanation is the lack of a detailed dendritic morphology for pyramidal neurons , which could account for different responses generated by location specific inputs , as previously argued [23] . However , we found that both the induction and the termination of persistent activity could be accurately predicted by the total number of spikes from all pyramidal neurons measured over the first 400 ms of the stimulus presentation . This feature predicted persistent activity emergence with an accuracy of 83%±7% ( Figure 6A ) and termination with an accuracy of 89%±6% ( Figure 6B ) , respectively . In both cases , the total number of pyramidal neuron spikes during stimulus presentation was higher for the persistent vs . transient ( p value<0 . 001 , Figure S4C ) and stable vs . terminated ( Figure S4D ) states , respectively . Overall , our findings show that both spiking characteristic of the network activity and slow synaptic currents through their effect on somatic membrane depolarization , mediate induction as well as termination of persistent activity in the microcircuit model .
Dendrites of PFC pyramidal neurons have distinct NMDA receptor properties with enriched NR2B subunits and slower kinetics compared to sensory areas [28] , [37] . In fact , the expression of mRNAs for NMDA subunits is higher in the PFC than in other regions of the human neocortex [38] . How exactly do NMDA receptors contribute to PFC function ? Computational studies have traditionally considered NMDA current as a slow mechanism that provides stability to persistent activity in large scale networks [11] . This view has recently been challenged as oversimplified , involving asymmetric contributions of NMDA receptors in excitatory versus inhibitory pathways [39] . Regenerative , non-linear dendritic integration that depends on the iNMDA-to-iAMPA ratio [32] has been recorded in vitro in L5 PFC pyramidal neurons , as well as in many other areas [40]–[42] . In these neurons , enhancement of the NMDA conductance needed for the emergence of the non-linear behavior was shown to depend on the recruitment of extrasynaptic NMDA receptors [19] , through glia-related events [20] , or could be due to dopamine-dependent increase in the NMDA conductance [24] . Our study predicts that the regenerative events occurring in the basal dendrites of PFC pyramidal neurons , in particular NMDA spikes , gate the induction of persistent firing in a biophysically validated PFC microcircuit , but not in large scale neuronal networks with relaxation of biophysical constrains . A number of studies have suggested that NMDA receptors are critical for persistent firing [10] , [11] , [39] , however , our study is the first to provide a direct link between NMDA spike generation in dendrites and persistent activity induction . These findings concur with a reported association between NMDA spikes and stimulus-specificity at the single neuron level [23] and with reports that dendritic NMDA spikes are crucial for the generation of Up-states in L5 PFC pyramidal neurons in vitro [20] . In support , recent studies have recorded dendritic spikes in thin dendrites in vivo and have correlated their appearance with a number of region-specific functions [43]–[46] , indicating that dendritic events may be a brain-wide mechanism for neuronal functions . We also claim that the minimum network size required for persistent activity induction is inversely proportional to the synaptic drive of each excitatory neuron: if synaptic input is sufficient to induce NMDA spikes , the network can be reduced down to 2 cells ( albeit under unrealistic conditions for synaptic connections ) . A number of modeling and experimental studies have focused on the effect of network size/synaptic strength/neuronal clustering in the emergence of various physiological phenotypes [17] , [47]–[49] . In fact , persistent activity was generated in very small networks but only under conditions that are far from the physiological ones [13] . This is the first study where a heavily constrained microcircuit model is used to infer a link between network size and dendritic nonlinearities with respect to persistent firing . Finally , we find that relaxation of connectivity and synaptic delay constraints eliminates the gating effect of NMDA spikes , albeit at a cost of much larger networks . Large scale networks were classically assumed necessary for persistent activity induction [15] . Our study suggests that different mechanisms underlie persistent firing in small vs . large scale networks , at least in the PFC: NMDA-dependent dendritic spikes underlie persistent firing in small , biophysically constrained , microcircuits via the generation of long lasting somatic depolarizations; in large scale networks , these plateau potentials are replaced by massive , asynchronous inputs that are sufficient to maintain activity . Our predictions concur with the finding that intracellular application of the NMDA-channel specific blocker MK-801 in monkeys performing a working memory task abolishes persistent activity in the PFC [10] , with in vitro studies in the visual cortex where NMDA blockade does not eliminate Up and Down states [50] , as well as with in silico studies in large-scale networks [15] which support persistent firing without NMDA receptors , under the assumption of asynchronous spiking activity . Our findings are particularly important from an optimization/energy conservation point of view as they suggest that active dendrites enable small microcircuits to express memory related processes such as persistent activity without requiring the recruitment of large neuronal networks and the associated energy costs . We found that inhibition , which was previously suggested to terminate persistent firing , mainly during Up states [22] , could terminate persistent activity with a probability below 0 . 5 . This termination could be significantly reduced by ongoing background synaptic activity as well as by the activation of the afterdepolarization mechanism ( dADP ) mediated by the calcium-activated non-selective cation ( CAN ) current . This finding points to a new functional role for the dADP , which has thus far been suggested to underlie the emergence of persistent activity [12] , [23] rather than its maintenance . We claim that the dADP , regulated by neuromodulators , may play a key role in preventing interfering signals from distracting the animals and thus improving working memory performance . An experimentally testable prediction made by our model based on these findings is that termination would be more easily achieved in the absence of dADP . In addition , we predict that in-vivo ongoing background activity contributes to the stability of the persistent state , possibly by providing wide-spread excitation during the delay period . Finally , we show that network activity ( number of spikes ) and slow synaptic mechanisms ( NMDA and GABAB currents ) , contain predictive information regarding the ability of a given stimulus to turn ON or OFF persistent firing in the microcircuit model . Interestingly , an upcoming ON state can be predicted by the microcircuit spiking activity , several milliseconds before the transition occurs . More importantly , a switch from the ON to the OFF state caused by a second inhibitory input can be predicted by the microcircuit response properties ( total number of spikes ) during the inducing stimulus , which is presented seconds before the termination takes place . This ability to predict ON and OFF states is in agreement with previous modeling ( albeit with a different feature ) work [23] and conforms with experimental work [51] showing that single neurons can encode state transitions , and PFC neurons in particular , can categorize signals in vivo at the onset of stimulus presentation [52] . This information is readily available to downstream regions [53] , [54] , presumably contributing to the preparation of a specific movement . The predictive roles of GABAB and NMDA are in accordance with recent findings that slow synaptic currents mediate persistent activity [55] and stimulus-outcome discrimination [56] , respectively . The finding that both iNMDA and iGABAB code for state transitions , presumably by shaping the somatic plateau potential , indicates that the balance of slow excitation/inhibition is crucial for the stability of the persistent state , as proposed by [55] . In support of this argument , we found that stable persistent activity trials were characterized by both increased NMDA and GABAB currents , effectively stabilizing the microcircuit activity . This is consistent with in vivo experiments in the PFC where Up states are generated through a temporal enhancement of fast excitation , whereas balanced synaptic events promote their stability [57] . Overall , this study zooms out from dendrites to cell assemblies and suggests a tight interaction between dendritic non-linearities and network properties ( size/connectivity ) that may facilitate the short-memory function of the PFC . In addition , it makes a number of novel , experimentally testable predictions regarding the role of dADP in the stability of persistent activity that may guide future studies and shed new light on memory-related processes .
A linear Support Vector Machine ( SVM ) classifier was used to examine whether certain features of the microcircuit responses ( network activity features , single cell response properties and synaptic currents ) could predict state transitions .
|
Working memory , the ability to retain information for a short period of time , is a fundamental cognitive function that shapes behavior . The cellular correlate of working memory is the prolonged spiking ( persistent ) activity of neurons in the prefrontal cortex . Impairments of prefrontal cortex functionalities and working memory have been associated with a variety of cognitive disorders , such as schizophrenia , the attention deficit hyperactivity disorder , and drug addiction . Hence , understanding how neurons embedded in the local circuitry support and maintain persistent activity is of outmost importance . Our work uses a multi-level integrative approach spanning from the dendritic , to the neuronal and network levels to identify the key biophysical and anatomical mechanisms contributing to persistent activity , leading to a number of high impact findings: it predicts a tradeoff between dendritic regenerative events and the size of a network expressing persistent activity . It also proposes when and how the persistent state can be stabilized , opening new avenues for pharmacological interventions . Finally , it describes decoding mechanisms for upcoming ON/OFF state transitions , furthering our understanding of information processing in the PFC and shedding new light on the emergence of anticipatory behaviors .
|
[
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] |
[
"circuit",
"models",
"computational",
"neuroscience",
"single",
"neuron",
"function",
"biology",
"and",
"life",
"sciences",
"computational",
"biology",
"coding",
"mechanisms"
] |
2014
|
Dendritic Nonlinearities Reduce Network Size Requirements and Mediate ON and OFF States of Persistent Activity in a PFC Microcircuit Model
|
In murine neurocysticercosis ( NCC ) , caused by infection with the parasite Mesocestoides corti , the breakdown of the Blood Brain Barrier ( BBB ) and associated leukocyte infiltration into the CNS is dependent on the anatomical location and type of vascular bed . Prior studies of NCC show that the BBB comprised of pial vessels are most affected in comparison to the BBB associated with the vasculature of other compartments , particularly parenchymal vessels . Herein , we describe a comprehensive study to characterize infection-induced changes in the genome wide gene expression of pial vessels using laser capture microdissection microscopy ( LCM ) combined with microarray analyses . Of the 380 genes that were found to be affected , 285 were upregulated and 95 were downregulated . Ingenuity Pathway Analysis ( IPA ) software was then used to assess the biological significance of differentially expressed genes . The most significantly affected networks of genes were “inflammatory response , cell-to-cell signaling and interaction , cellular movement” , “cellular movement , hematological system development and function , immune cell trafficking , and “antimicrobial response , cell-to-cell signaling and interaction embryonic development” . RT-PCR analyses validated the pattern of gene expression obtained from microarray analysis . In addition , chemokines CCL5 and CCL9 were confirmed at the protein level by immunofluorescence ( IF ) microscopy . Our data show altered gene expression related to immune and physiological functions and collectively provide insight into changes in BBB disruption and associated leukocyte infiltration during murine NCC .
The blood brain barrier ( BBB ) separates the peripheral circulation from the CNS and plays a critical role in homeostasis of the CNS environment . In the healthy brain BBB selectively restricts molecular and cellular trafficking between the blood and brain tissue and between blood and cerebrospinal fluid ( CSF ) [1] . The restrictive properties are largely controlled by specialized endothelial cells of the CNS vasculature which differ from those in the peripheral vasculature in terms of polarized expression of various transport systems , low transcytosis activity , high mitochondrial volume and sealing of the paracellular cleft between endothelial cells by continuous strands of interendothelial junction proteins including tight junctions [1] . However , additional components of the BBB are present in different CNS compartments and vary according to their anatomical location in the CNS and nature of the vasculature . The blood vessels present in leptomeninges ( pia ) in subarachnoid space are collectively termed pial vessels . The BBB associated with pial vessels in adult brain are largely devoid of pericytes , astrocytic endfeet processes , additional basement membranes and parenchymal tissue in comparison to that of parenchymal vessels [2] , [3] , [4] . Infection of the CNS leads to changes in barrier properties of the BBB allowing the leakage of serum components ( edema ) and infiltration of leukocytes resulting in CNS pathology [5] , [6] . In addition , the BBB transport system is also affected further disturbing the homeostasis of the CNS environment [1] . Neurocysticercosis ( NCC ) is a CNS infection caused by the metacestode ( larva ) of the tapeworm Taenia solium . It is one of the most common parasitic infections of the CNS and a major cause of acquired epilepsy worldwide [7] . Depending upon the size , location , and number of parasites as well as sex , age and immune status of the host , there are differences in disease severity and pathologies [8] . Epidemiological studies show that among the various forms of NCC , subarachnoid NCC has the worst outcome and is associated with poor prognosis , more resistance to anti-helminthic drugs and more severe inflammation [9] . The chronic inflammation of the vasculature and arachnoid thickening ( chronic basal meningitis ) leads to blockade of CSF further contributing to CNS pathology [8] . Similarly , using a murine model for NCC by infection with the highly related parasite Metacestoides corti , prior studies from our laboratory have demonstrated that breakdown of the BBB and associated leukocyte infiltration depends on many criteria including the anatomical site , type of vascular bed , and infiltrating cell phenotype [6] , [10] , [11] . Assessment of the integrity of the BBB by changes in the architecture of interendothelial junction proteins and leakage of serum proteins revealed that the BBB associated with pial vessels were compromised earlier and to a greater extent in comparison to the BBB associated with vessels present in other CNS compartments [12] , [13] . In addition , previous studies have shown that during murine NCC , the temporal pattern of infiltrating leukocyte subsets is characterized by a large infiltration of macrophages and γδ T cells followed by αβ T cells and lastly B cells [14] . Further characterization of leukocyte subset infiltration in different CNS compartments has established that the majority of the infiltration occurs via pial vessels [13] . There is a lack of detailed analysis of BBB disruption in vivo in a CNS compartment-specific manner . To address this deficiency and to obtain insights into changes occurring only to pial vessels , we designed a microarray-based , comprehensive study to analyze the changes in gene expression associated with the BBB comprised of pial vessels of the leptomeninges and subarachnoid spaces . We utilized laser capture microdissection microscopy ( LCM ) to isolate pial vessels from mock- and parasite-infected mice and performed microarray analyses . Our transcriptome data indicate an altered expression of genes related to the immune response and to physiological function and collectively provide insight into the dysfunction of the BBB during murine NCC associated with pial vessels .
This study was conducted in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the U . S . National Institutes of Health . Experiments were carried out under the approved guidelines of the Institutional Animal Care and Use Committee ( IACUC ) , University of Texas at San Antonio ( approved IACUC protocol number MU003-07/11A0 ) . Female Balb/c mice were purchased from National Cancer Institute program ( Bethesda , MD ) . Parasite maintenance and intracranial infection were performed using a protocol developed earlier [14] . M . corti metacestodes were maintained by serial intraperitoneal ( i . p . ) inoculation of 8- to 12-week-old female BALB/c mice . For intracranial inoculations , parasites were aseptically collected from the peritoneal cavity of mice that had been infected for about 4–6 months . Harvested parasites were extensively washed in HBSS . After that , the metacestodes ( 70 microorganisms ) were suspended in 50 µl of HBSS and injected intracranially into 3–5-week-old female BALB/c mice using a 1-mL syringe and a 25-gauge needle using our protocol developed earlier . The needle was inserted to a 2-mm depth at the junction of the superior sagittal and the transverse sutures . This allows insertion of the needle into a protective cuff avoiding penetration of the brain tissue . Control mice were injected with 50 µl sterile HBSS using the same protocol . Before intracranial inoculation , mice were anesthetized intramuscularly with 50 µl mixture of ketamine HCL and xylazine ( 30 mg/ml ketamine and 4 mg/ml xylazine ) . Animals were sacrificed at 3 weeks after inoculation . Before sacrifice , animals were anesthetized with 50 µl of mixture of ketamine HCL and xylazine . The thoracic cage was opened and 100–125 µl of a Rhodamine Red-X conjugated Ricinus communis agglutinin ( Rh-RCA ) lectin ( Vector Lab ) was injected through the left ventricle in heart . After 2 minutes of Rh-RCA injection , perfusion was performed through the left ventricle with 15 mL of cold HBSS [15] . Perfused brains were immediately removed , embedded in O . C . T . resin ( Sakura , Torrance , CA ) and snap frozen in 2-methyl butane ( Fisher Scientific , Pittsburgh , PA ) contained/cooled in liquid nitrogen and stored at −80°C for later use . 10 µm thick horizontal cryosections were obtained from each brain on polyethylene naphthalate membrane slides ( Leica Microsystems , Wetzlar , Germany ) . The tissues were fixed in −20°C acetone for 20 seconds and kept in dry ice . Subsequently brain sections were dehydrated in 70% ( 10 s ) , 95% ( 20 s ) , 100% ( 3x , 30 s each ) and xylene ( 2x , 30 sec ) . After dehydration , the slides were kept in desiccators until the time of dissection to avoid the humidity . LCM was performed with Leica LMD 7000 micro systems ( Leica Microsystems , Wetzlar Germany ) as described previously [16] . From LCM isolated endothelial cells , RNA was extracted with Pico Pure RNA isolation kit ( Arcturus Bioscience , Mountain View , CA ) according to manufacturer's protocol . DNase ( Qiagen , Valencia , CA ) treatment was performed directly within the purification column to remove any possible genomic contamination during the RNA extraction process . The quality of the RNA was inspected with Agilent 2100 Bioanalzyer and NanoDrop ND-1000 . Samples passing quality control assessment were then subjected to linear amplification and subsequently labeled with NuGEN Ovation Aminoallyl RNA Amplification and Labeling System ( NuGEN Technologies , San Carlos , CA ) as per manufacturer's instructions . Arrays were printed at the Duke Microarray Facility using the Genomics Solutions OmniGrid 100 Arrayer and mouse genome oligo set ( version4 . 0 ) . The Mus musculus Operon v4 . 0 spotted microarray contains 35 , 852 longmer probes representing 25 , 000 genes and about 38 , 000 gene transcripts ( Operon Biotechnologies , Huntsville , AL ) . The amplified and labeled product was hybridized to Mus musculus Operon v4 . 0 spotted microarray according to the manufacture protocol at 42°C with the MAUI hybridization system ( BioMicro Systems , MAUI hybridization System , Salt Lake City , Utah ) . The array was then washed at increasing stringencies and scanned on a GenePix 4000B microarray scanner ( Axon Instruments , Foster City , CA ) . The Genespring 11 program ( Agilent Technologies , Redwood City , CA ) was used to perform data processing and statistical analysis . Intensity-dependent ( Lowess ) normalization was done on the entire data set . To assess the quality of a data set , a principle component analysis was performed on samples on expression of all genes with mean centering and scaling . Datasets were filtered based on values and probe sets with background-subtracted intensity of 44 or less were excluded from the analysis . Subsequently , t-test analysis was performed to calculate the p-values using an asymptotic method and Benjamini-Hochberg , for multiple testing correction . Differentially expressed probe sets were selected based on volcano plot with a 2-fold change and p-value cut off of 0 . 05 . Differentially expressed genes were then clustered using Average Linkage with Pearson Correlation as the similarity measurement . Molecular networks of the selected molecules and specific pathways were analyzed through Ingenuity Pathway Analysis software ( Agilent Technologies , Redwood City , CA ) . RNA obtained from LCM isolated endothelial cells ( as described above ) was subjected to linear amplification by the WT-Ovation Pico System ( Nugen technology , San Carlos , CA ) . Resulting cDNA was loaded onto Taq-Man Low Density Arrays ( Applied Biosystems , CA ) microfluidic cards either preloaded with fluorogenic probes and custom-designed primers and housekeeping genes β-actin , ribosomal 18S , and GAPDH ( glyceraldehyde 3-phosphate dehydrogenase ) [17] or commercially available Mouse Immune Array ( catalog number – 4367786 , Applied Biosystems , CA ) . These plates were then loaded on an ABI Prism 7900 HT Sequence Detection System ( Applied Biosystems , CA ) . The target expression levels were normalized to the levels of the house keeping genes 18S , β-actin and GAPDH in the same sample . Expression of each specific gene in infected samples over mock was calculated by ΔΔCt method and results are represented as ΔΔCt over mock [18] . Tissue preparation and immunofluorescence ( IF ) staining was performed using our protocol as described previously [13] . Animals were sacrificed at 3 weeks after inoculation . Before sacrifice , animals were anesthetized with 50 µl of mouse cocktail and perfused through the left ventricle with 15 mL of cold PBS . Perfused brains were immediately removed , embedded in O . C . T . resin ( Sakura , Torrance , CA ) and stored at −80°C . Serial horizontal cryosections of 10 µm in thickness were placed on saline prep slides ( Sigma-Aldrich , St . Louis , MO ) . The slides were air dried overnight and fixed in fresh acetone for 20 s at room temperature ( rt ) . Acetone-fixed sections were wrapped in aluminum foil and stored at −80°C or processed immediately for immunofluorescence . Briefly , tissues were fixed in −20°C acetone for 10 min and then hydrated in PBS . Non-specific immunoglobulin binding was blocked by 30 min incubation at rt with 10% serum from the same species from which the fluorochrome conjugated antibodies ( secondary antibodies ) were derived . Sections were incubated for 40 min with primary antibodies diluted in 3% serum from the host of secondary antibody . Sections were washed 7× for 3 min each after incubation with specified antibodies . Secondary antibodies were incubated for 30 min at rt when necessary . Then , sections were mounted using fluorsave reagent ( Calbiochem , La Jolla , CA ) containing 0 . 3 µM 4′ , 6′-diamidino-2-phenylindole dilactate-DAPI ( Molecular Probes , Eugene , OR ) . Negative controls using secondary antibodies alone were included in each experiment and found to be negative for staining . Fluorescence was visualized in a Leica microscope ( Leica Microsystems , Wetzlar Germany ) . Images were acquired and processed using IP lab software ( Scanalytics , Inc . , Rockville , MD , USA ) and Adobe Photoshop CS2 ( Adobe , Mountain View , CA ) . The purified primary antibodies goat anti mouse CCL5 ( catalog number AF478 ) and CCL9 ( catalog number AF463 ) were bought from R&D systems and biotinylated CD31 antibody ( catalog number 553371 ) from Pharmingen ( San Diego , CA ) . Rabbit anti Goat labeled with Rhodamine Red- X and donkey anti rabbit rhodamine red X secondary antibodies were purchased from Jackson ImmunoResearch ( West Grove , PA ) [13] . M . corti parasites were collected aseptically from 4–6 months ip infected mice and washed rigorously with HBSS and then incubated with half the volume of HBSS+ gentamycin at 37°C , 4% CO2 for 72 hrs in a 25 CM2 culture flask . After incubation , parasites were removed by filtering with a nylon mesh and the supernatant ( MCS ) was collected and kept at −80°C for future use . bEND . 3 cells were purchased from ATCC and subcultured using DMEM+10%FBS . Cell were seeded in chamber slides and stimulated with parasite supernatant , parasite homogenate or PBS for control . After , 72 hrs of stimulation , IF staining was performed . Briefly , cells were washed with PBS and incubated with 70% ETOH for 10 minutes followed by 3 PBS washes for 3 min each . Subsequently , cells were blocked with 10% serum from the host of secondary antibody , followed by 40 min incubation with primary antibodies and 30 min with secondary antibodies as described in previous ( IF section ) section . Chamber slides were mounted using fluorsave reagent ( Calbiochem , La Jolla , CA ) containing DAPI . Images were acquired and processed as described in the previous section .
We administered Rh-RCA lectin ( Rhodamine conjugated Ricinus communis agglutinin ) systemically at 3 wk post infection ( p . i . ) and mock-infected mice to label the pial vessels as described in Materials and Methods . The 3 wk p . i . time point was used because this is consistently the peak of leukocyte infiltration . Brain sections from in vivo labeled , perfused brain tissues were prepared and analyzed for labeling of the blood vessels after dehydration . We found that 5 µg/mg of body weight was sufficient to label the blood vessels ( Fig . 1 ) . LCM was performed as described previously [16] . RH-RCA labeled Pial vessels , distinctly located in subarachnoid spaces along with leptomeninges were collected by LCM . Subsequently , total RNA was isolated from LCM enabled samples , and linear amplification was done in order to perform microarray experiments as described in Material and Methods . Microarray hybridization experiments were performed to assess differentially expressed genes during infection using operon spotted chip arrays , and the data were processed by Genespring 11 to quantify differentially expressed probe sets ( see Materials and Methods ) . Quality control on samples was done by principle component analysis which showed separation between mock and infected samples based on their gene expression profile while clustering the infected samples and mock samples together respectively ( data not shown ) . In total , 2154 probe sets passed the screen when the probe sets were filtered for intensity with a lower cut off 44 . Out of these , 768 probe sets met a corrected p-value ( Benjamini-Hochberg cut off of 0 . 05 . Of the 768 probe sets , 578 probe sets were found to be differentially expressed with a fold change of ≥2 . Differentially expressed probe sets with a fold change of ≥2 were subjected to hierarchical cluster analysis using Average Linkage with Pearson Correlation as the similarity measurement of gene expression ( Fig . 2 ) . Operon chips contain oligo probe sets representing transcripts belonging to annotated genes as well as Expression Sequence Tags ( EST ) which represent yet to be defined genes . All the 578 differentially expressed probes were uploaded to Ingenuity Pathway Analysis ( IPA ) software to find out known genes associated with differentially expressed probe sets . IPA is a web-based application that uses a knowledge base created by previous findings of molecular interactions in the context of biological events . Once a gene is uploaded into IPA during core analysis , it maps the gene and places them in relevant molecular networks , biofunctions and specific pathways ( https://analysis . ingenuity . com/ ) . Out of 578 probe sets , 380 ( 285 upregulated and 95 down regulated ) were found annotated or mapped by IPA ( Table S1 ) . In order to understand the biological significance of the differentially expressed genes , biofunctions and networks of genes involved in biofunctions were analyzed using IPA . Under biofunction analysis genes were categorized into three different classes of biofunctions such as disease and disorder , molecular and cellular function , and physiological system development and function ( Table 1 ) . The disease and disorder category included Immunological disease , infectious disease , inflammatory response , connective tissue disorders and inflammatory disease ( p = 8 . 17E-24 to 2 . 83E-05 ) . The category of molecular and cellular functions included Cellular function and maintenance , cellular movement , cell death , cellular development and cellular growth and proliferation ( p = 1 . 00E-33 to 3 . 92E-05 ) . Genes in the category of physiological system development and function were associated with Hematological system development and function , tissue morphology , immune cell trafficking , tissue development and humoral immune response ( p = 9 . 38E-32 to3 . 87E-05 ) ( Table 1 ) . Many of the genes were classified in more than one biofunction category due to the broad and overlapping nature of the categories as well as an individual gene influencing multiple biofunctions . We analyzed the differentially expressed genes using IPA to assess how genes interact with each other as part of biological pathways . The resulting networks are generated based on the random selection of focus genes with maximum connectivity and several interconnected focus genes put together as a network in order of high to low scores . Scores are derived from p-values and are calculated through Fisher's exact test which represents the probability of finding the focus genes of a network in a set of n genes randomly selected from a global molecular network of genes . Based on focus genes differentially expressed during infection , 23 networks were identified . 22 networks that yielded a score of more than 3 are shown in Table S2 . Network analysis indicated that genes involved in the metabolism of lipids , carbohydrates and amino acids are affected . Further , immune response related genes were identified in multiple networks along with genes involved in cell growth , death and connective tissue disorder ( Table S2 ) . Pictorial representation of three of the networks is shown in Fig . 3 . Fig . 3 B , C and D show the networks “inflammatory response , cell-to-cell signaling and interaction , cellular movement” “cellular movement , hematological system development and function , immune cell trafficking” and “antimicrobial response , cell-to-cell signaling and interaction , embryonic development” respectively involving immune response related genes . A number of genes were chosen from different functional categories to be verified for their gene expression pattern by Taqman real time polymerase chain reaction ( RT-PCR ) using the amplified cDNA derived from pial endothelial cells isolated by LCM . Results obtained from RT-PCR experiments confirmed the expression pattern of a number of genes . Data showed that SELP , CD274 , LGALS3 , MRC1 , FIZZ1 , β2M , C3 , CCL2 , CCL5 and STAT1 were significantly upregulated ( Table 2 ) similar to microarray . To assess protein expression , brain sections from mock-infected and infected mice ( 3 wk p . i . ) were analyzed by IF microscopy for chemokines including CCL5 ( Fig . 4A ) and CCL9 ( Fig . 4B ) . In sections from mock-infected mice , CCL5 was undetectable . Infection resulted in a substantial up-regulation of CCL5 which co-localized with CD31 , an endothelial cell marker . Similarly , CCL9 was scarcely detected in the blood vessels from mock-infected samples . CCL9 was highly up-regulated as a result of infection and appears to be secreted . In addition , it co-localizes with undefined strand-like structures that appear to form a gradient starting from the outer surface of pial vessels ( abluminal ) towards the direction of infiltrating cells . The degree of CCL9 expression was higher in inflamed vessels exhibiting leukocyte egress . Since chemokines can be secreted and deposited on extracellular matrix , it was important to confirm that endothelial cells can produce these chemokines . To test this , bEND . 3 ( brain endothelial cell line ) cells were stimulated with either M . corti secretory/released antigens ( MCS ) or whole parasite homogenate in HBSS ( WP ) and analyzed for the production of CCL5 and CCL9 . We found that both parasite preparations induced an increased expression of CCL5 and CCL9 by bEND . 3 cells compared with controls in the absence of antigen ( Fig . 5 ) .
The BBB acts as an interface between the periphery and the CNS and tightly regulates the components of the immune response to prevent unnecessary inflammation/pathology in the healthy brain . It is known that the nature of the vasculature and associated functions differ greatly depending upon their location in different CNS compartments [4] . In a number of CNS infections , pial vessels of the BBB are particularly prone to disruption with leakage of leukocytes and serum components leading to meningitis [19] . This increased vulnerability is possibly due to lack of additional barrier components and potential exposure to antigens compared with parenchymal vessels [20] , [21] , [22] . Previously , gene expression analysis of endothelial cells has been performed either in an in vitro setting or with whole brain endothelial cells [23] , [24] , [25] , [26] , [27] , but not with endothelial cells present in specific anatomical compartments . Moreover , the effect of parasitic infection on endothelial cell biology has not been studied . The focus of this study was to characterize the infection-induced molecular signature of LCM isolated pial endothelial cells by evaluating global gene expression by microarray analyses . LCM allowed us to isolate cells present in a specific location which has an added advantage over other marker-based techniques such as FACS . However , one pitfall is that the potential contamination of the BBB endothelium with the leukocyte that may be extravasating or adhering to the endothelial cells . Our data analysis confirmed that differential gene expression data obtained through microarray hybridization experiment is mainly contributed by endothelial cells comprising the BBB as common lymphoid or myeloid cell markers were not detectable in the data set . In addition , the expression of the following BBB specific transporter markers were induced during infection: TFRC ( related to iron metabolism ) , ABCG1 ( cholesterol homeostasis ) , SLC15A3 ( proton oligopeptide co-transporters ) , SLC7A5 ( cationic amino acid transporters and the glycoprotein-associated amino acid transporters ) , ABCC3 ( multidrug resistance associated protein 3 ) and ABCC5 ( multidrug resistance associated protein 5 ) . Other BBB specific markers were downregulated including SLC9A3R2 ( sodium/hydrogen exchanger ) , SLC6A9 ( neurotransmitter transporter , glycine , sodium and chloride dependent neurotransmitter ) [23] , [24] , [25] , [26] , [27] . Network analysis shows that apart from transporters several other sets of immune related genes including MRC1 , complements ( C3 , C6 , and C1R and complement factor properdin ) , TNF super family members and interferon inducible genes including STAT1 are induced in NCC infection which can potentially lead to endothelial cell activation [23] , [24] , [28] . Interferon inducible genes have been shown to be induced in an in vitro study with endothelial cells in HIV and Cryptococcus neoformans infection model [25] , [29] . STAT1 has been shown to promote inflammatory mediators and leukocyte transmigration at the BBB [30] . Interferon signaling mediated through the Jak Stat pathway is critical to induce several of these genes in endothelial cells including chemokines and MHC class I antigen presentation related genes [23] , [24] , [28] . Among immune related genes chemokines play a critical role in leukocyte trafficking , differentiation and angiogenesis or angiostasis [31] , [32] . Leukocyte trafficking is a multistep process in which chemokines induce the migration of leukocytes toward a chemokine gradient . Interaction between chemokines expressed by endothelial cells with their receptors on leukocytes triggers a signaling process that increases the avidity of integrin to their receptors on endothelial cells causing firm adhesion of leukocytes and facilitated transmigration towards chemokine gradient [33] . Chemokines are divided into C , CC , CXC , and CX3C subgroups based on conserved cysteine residues [31] . The present study advances the understanding about chemokine expression profile in endothelial cells comprising the BBB which are the first CNS cells to encounter peripheral leukocytes in vivo . Many of the chemokines upregulated ( Table S1 ) in response to infection are summarized in Table 3 along with their putative receptor and influence on specific leukocyte subsets . Our in vivo and in vitro data shows that CCL9 is expressed abundantly by endothelial cells and appears to coat the strands in a gradient fashion . Such strands have been observed in the areas of inflammation in other disease conditions such as EAE and toxoplasmic encephalitis [34] . The origin and composition of these strands are still not clear . They have been described to extend from blood vessels to parenchyma and are thought to provide structural support for leukocytes migration [34] . In the case of NCC , these strands coated with CCL9 might also provide a physical scaffold structure with a chemotactic signal for migration of leukocytes into the CNS . The functional correlation for CCL9 in terms of leukocyte subset recruitment remains to be defined in the CNS . However , in the periphery CCL9 has been implicated in recruitment of myeloid cells to peyers' patches and osteoclasts through the CCR1 receptor . Furthermore , it is also critical to recruit immature myeloid cells through CCR1receptor during liver metastasis [35] . In addition , CCL17 and CCL22 are also noteworthy as they have been implicated in trafficking of CCR4 positive regulatory and Th2 T cells subsets [33] . Chemokine can selectively influence the trafficking of leukocyte subsets . Therefore , the expression profile of chemokines in the BBB provides insight into the trafficking of different leukocyte subsets such as M1 and M2 macrophages , granulocytes , γδ T cells , αβ T cells and B cells known to infiltrate during NCC [13] , [14] , [36] , [37] , [38] . In summary , our data delineate infection-induced changes in the expression of genes associated with both immunity and disease , and collectively provide insight into the dysfunction of the BBB and mechanisms associated with leukocyte infiltration during murine NCC .
|
Neurocysticercosis ( NCC ) is one of the most common parasitic diseases of the CNS caused by the metacestode ( larva ) of the tapeworm Taenia solium . Epidemiological studies show that among the various forms of NCC , subarachnoid NCC is associated with poor prognosis , more resistance to anti-helminthic drugs and more severe inflammation . The chronic inflammation of the vasculature and arachnoid thickening ( chronic basal meningitis ) leads to blockade of CSF further contributing to CNS pathology . Using a murine model for NCC , we have found that among the different types of vasculature associated with the blood-brain barrier ( BBB ) , pial vessels of BBB are compromised earlier and to a greater extent during NCC . In addition , pial vessels are likely the most important entryway for leukocyte infiltration during NCC . The aim of this study was to characterize infection-induced changes in the genome-wide gene expression of pial vessels . Our approach was to isolate pial vessels of the BBB by in vivo labeling of vessels followed by laser capture microdissection microscopy ( LCM ) . Further , microarray analysis of pial vessels showed infection-induced changes in the expression of genes associated with both immunity and disease , and collectively provides insight into the dysfunction of the BBB and mechanisms associated with leukocyte infiltration during murine NCC .
|
[
"Abstract",
"Introduction",
"Materials",
"and",
"Methods",
"Results",
"Discussion"
] |
[
"immunopathology",
"medicine",
"cerebrovascular",
"diseases",
"emerging",
"infectious",
"diseases",
"neurological",
"disorders",
"neurology",
"immunology",
"biology",
"microbiology",
"host-pathogen",
"interaction"
] |
2013
|
Changes in Gene Expression of Pial Vessels of the Blood Brain Barrier during Murine Neurocysticercosis
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.