doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pntd.0001762 | Whole Organism High-Content Screening by Label-Free, Image-Based Bayesian Classification for Parasitic Diseases | Sole reliance on one drug, Praziquantel, for treatment and control of schistosomiasis raises concerns about development of widespread resistance, prompting renewed interest in the discovery of new anthelmintics. To discover new leads we designed an automated label-free, high content-based, high throughput screen (HTS) to assess drug-induced effects on in vitro cultured larvae (schistosomula) using bright-field imaging. Automatic image analysis and Bayesian prediction models define morphological damage, hit/non-hit prediction and larval phenotype characterization. Motility was also assessed from time-lapse images. In screening a 10,041 compound library the HTS correctly detected 99.8% of the hits scored visually. A proportion of these larval hits were also active in an adult worm ex-vivo screen and are the subject of ongoing studies. The method allows, for the first time, screening of large compound collections against schistosomes and the methods are adaptable to other whole organism and cell-based screening by morphology and motility phenotyping.
| Schistosomiasis is a severe helminth infection affecting an estimated 600 million people. The one drug widely available, praziquantel (PZQ), is not ideal. PZQ kills the adult worms but not the developing juveniles so the treated patient may not be cured long-term. In addition, use of repeated mass treatment campaigns with PZQ to control morbidity raises concerns about the development of drug resistance. Our work is aimed at providing starting points for drug discovery programs for schistosomiasis by screening large compound libraries against whole organisms. Praziquantel and several other known anti-schistosomal drugs are also active in vitro against the adult worms and the larval stages, schistosomula. The latter are ideal for novel drug screening as they can be produced in large numbers in vitro, are small and so are amenable to screening in microwell plates. Drug activity can be assessed visually but this is subjective and laborious. We have built an automated system for assessing drug action involving the collection of images of the larvae and the development of computer algorithms to analyze their morphology and motility, defining them as "hits" or "nonhits." The method is reliable, consistent and efficient, making it feasible, for the first time, to screen large compound collections.
| Infection with parasitic worms (helminths) causes a huge burden of human disease [1] and economic loss to the livestock industry [2]. Currently the major control strategy for the human diseases is by large scale drug administration to schools or by mass drug administration [3]. However, the drugs available are limited in number and efficacy and their increasing use worldwide raises concerns about the development of drug resistance [4], [5].
Schistosomiasis affects an estimated 600 million people [6] but only one drug, praziquantel (PZQ), is commercially available for its treatment and control. PZQ is poorly effective against the immature worms [7] and its increasingly widespread use [8] fuels concerns about drug resistance developing [9]. There have been sporadic reports of treatment failures with PZQ [10], [11], [12] and strains isolated from such cases show lower susceptibility to PZQ [13]. However, since the development of PZQ [14] there has been limited interest in discovery of new schistosomicides apart from the recent identification of oxadiazole-2-oxides as lead compounds [15] and of anti-schistosome activities for some anti-protozoal drugs [16], [17].
The approved anthelmintics invariably were discovered by in vivo screening in animal models. However the low throughput and high costs of these models limits the discovery of new anti-helminth agents. Therefore new high-throughput, in vitro phenotypic screening methods are necessary to advance the discovery of new anthelmintics including anti-schistosome compounds. In the recent past small, focussed, compound collections have been screened against adult schistosomes recovered from rodents [18], [19]. To facilitate screening of larger compound collections microplate-based visual assays were developed using in vitro-derived larval stages, schistosomula, which can be generated in very large numbers [20], [21]. With a view to standardization and automation methods other than manual visual assessment have recently been applied to evaluate drug-induced damage to schistosomula [21], [22], [23], [24]. However bright-field microscopy is simpler to set up, reveals drug-specific morphological effects, and is 100% effective in detecting compounds active in the adult ex-vivo assays [20], [21]. To overcome the need for visual assessment we have developed a label-free, high content screen (HCS) using automatic bright-field image analysis to establish and validate a high throughput screen (HTS) for primary drug screening against schistosomes. Compound efficacy is assessed by a combination of larval motility and larval morphology quantified by Bayesian analysis. The methods make it feasible for the first time to screen very large compound collections against schistosomes and are applicable to other larval helminths.
Experimentation was carried out under the United Kingdom Animal's Scientific Procedures Act 1986 with approval from the London School of Hygiene and Tropical Medicine Ethics committee. CD1 mice supplied by Charles River, UK were maintained at St Mary's Hospital, Imperial College London.
Schistosoma mansoni was maintained by routine passage and schistosomula were prepared and cultured in M169 [25] as previously described [21]. Adult worm ex-vivo drug testing was as previously described [19].
The reference anti-schistosome compounds praziquantel (PZQ) and dihydroartemisinin (DHA) were obtained from Sigma-Aldrich (UK), methylclonazepam (MCZ) and Ro15-5458 (Ro15) were a gift from Dr H. Stohler (Hoffman-La Roche, Basle, Switzerland), oxamniquine (OX) was from Pfizer Ltd (Sandwich, UK) and oltipraz (OPZ) from WHO Special Programme for Research and Training in Tropical Diseases (WHO-TDR; Geneva, Switzerland). Compounds were dissolved in DMSO (Sigma-Aldrich, UK). A 10,041 compound library comprising lead-like compounds was provided by the Division of Biological Chemistry and Drug Discovery, University of Dundee.
The last two columns of each test plate were reserved for controls. The test compound solvent, DMSO, was used as the negative control and added to 16 wells. Our initial testing of the image analysis models revealed that OPZ induced the lowest phenotype and motility scores reflecting the visual assessment that OLT caused the most severe effects of all of the anti-schistosome compounds tested. Therefore, OLT was chosen as the positive reference standard and applied to 8 wells. PZQ, the current therapy for schistosomiasis, induced intermediate phenotype and motility scores and so 4 wells of PZQ were included on each plate as an arbitrary check on plate performance.
Black 384-well clear-bottomed plates were selected for imaging (PerkinElmer, UK Cat no 6007460). Into each well 0.5 µl of test compound or DMSO was dry stamped using the Biomek FXp (Beckman Coulter, High Wycombe, UK). When necessary a prior intermediate dilution step in DMSO was carried out in V-bottomed dilution plates (Greiner bio-one, UK, cat no 781280). Schistosomula (120/well) were added to each well in 80 µl of M169 media using a Matrix WellMate (Thermo Scientific, Basingstoke, UK). Plates were then incubated in a Cytomat C2 automatic incubator (Thermo Scientific, UK) at 37°C, 5% CO2 for 3 days.
A Scara Robot (KiNEDx Robot KX-300-470, Peak Robotics, Colorado, USA) controlled by Overlord 3 (Process Analysis and Automation, Hemel Hempstead, UK) was used for all plate movements. After 3 days culture schistosomula were redistributed and disaggregated by using the Biomek FXp programmed to aspirate and dispense 40 µl of the well contents in each of the 4 corners of each well (×3). Bright-field images were collected using an ImageXpressMicro HCS microscope (IXM; Molecular Devices, Wokingham, UK) fitted with a PhotometricsCoolSnapHQ camera (Roper Scientific, Germany). Focussing of the plate and well bottom was achieved by the IXM high-speed laser auto-focus, with a 25 µm offset to focus on the larvae.
For motility analysis 5×6 sec interval time-lapse images were collected using a 4× S Fluor 0.2NA Nikon objective. For detailed morphology a 10× Ph1 Plan Fluor DL 0.3NA Nikon objective was used to collect 4 adjacent images, which were tiled together to maximise larval numbers for phenotype analysis. After imaging, the plates were visualized by two independent assessors using an inverted microscope (LeitzDiavertWetzlar, Germany).
Differences in phenotype and motility scores were measured by one-way ANOVA with a Dunn's post-test to measure significant differences between DMSO control wells and individual drug treatments. Z factors for both the phenotype and motility scores were measured on a per plate basis in Pipeline Pilot 8.5 (Accelrys Inc., San Diego, USA) with an acceptable score being >0.5 [26].
Following preliminary assessment of appropriate screening concentration/hit rate, the 10,041 compound library was screened at 10 µM. All of the plates were also visually scored by two independent assessors [21]. Using the HCS hit thresholds defined above, all the visual hits (379) apart from four were determined to be hits by the automatic analysis (Figs. 5A & B). Three of the failures were ascribed scores which fell just outside the hit threshold and one failed to segment due to the parasites remaining aggregated. The hit region also contained 109 wells which were visual non-hits (i.e. false positives by HCS). Of these, 86 were wells containing compound crystals. All of these were readily rejected as hits on manual review of the corresponding images in the automated HCS plate reports by marking the “Non-Hit checkbox” (Fig. S3E). Overall, during the manual plate reading, 780 wells were found to have crystals, of which 130 were deemed to be hits. Importantly, all of these fell in the HCS hit region. A novel phenotypic effect (internal vacuolation) was also identified during visual assessment/plate reporting for a number of compounds, a proportion of which were scored as hits by the HCS. Visually, larval viability was not considered sufficiently reduced to designate these as hits and none of the compounds were active in the adult assay. Z factor scores were reviewed to assess plate performance during screening (Fig. 5C) all of which were within an acceptable range.
Hits from the screen were also analysed and grouped according to larval phenotype by the Bayesian categorization model to determine which anti-schistosome compound they most resembled. From the 378 hits, 175 were ascribed to the OPZ treatment class, 60 to PZQ, 13 to MCZ, 83 to Ro15, 34 to OX and 13 to DHA.
The assay was further validated by re-testing a selection (796) of hits and non-hits from the 10,041 compound library along with compounds from the WHO-TDR set. There was a high level of concordance between the initial and repeat testing (92.3% for morphology, Fig. 6A; 95.2% for motility, Fig. 6B).
In vitro testing against ex-vivo adult worms is a crucial secondary screen since the adult worm is the key target of drug action. Preliminary testing of a few of the larval hits in the secondary adult worm assay [19] at 10 µM yielded a very low hit rate and so the hits were all tested at 20 µM which gave 45 adult hits. Plotting the larval phenotype and motility scores for these hits (Fig. 7) showed that the majority corresponded to severe larval phenotypes but a few were scattered throughout the hit threshold region. The number of hits ascribed to different treatment classes were OPZ 28, PZQ 6, DHA 4, Ro15 4, OX 2 and MCZ 1. Subsequent IC50 testing of the adult hits identified 7 compounds which had IC50s of <10 µM and which are the subject of on-going studies. These compounds were attributed the drug treatment classes OLT 5, PZQ 1, OX 1.
Whole organism screens have an advantage over more target-based approaches as hit compounds can be directly translated into new therapeutics and it has been suggested that the lack of these screens has impacted on the discovery of new compounds [28]. We have established the first HTS for whole organism screening of helminths based on bright-field HCS analysis of morphological and motility changes in schistosome larvae. The published examples of primary high content drug screens are predominately cell-based [29], [30]. A few involving whole organism screens for Caenorhabditis elegans [31], zebrafish embryos [32], Leishmania major [33] and Plasmodium falciparum [34] have been reported. These exploit use of fluorescent probes or proteins since fluorescence provides more contrast, sharpness and discrimination compared with transmitted-light imaging. However, use of fluorescent probes often involves more manipulations and use of potentially toxic fluorophores [35]. Furthermore, fluorescent transgenic lines are not available for certain organisms of interest including parasitic helminths.
Transmitted-light imaging and analysis has been developed for whole well segmentation of C. elegans [36] and to demonstrate drug-induced motility changes using optical flow in adult Brugia malayi [37]. Our approach focussed on the development of bright-field imaging of morphology and motility of larval schistosomes directly comparable to the manual visualization used previously [20], [21]. This requires segmentation and analysis of whole organisms as has been described for C. elegans [31], [38], [39]. Our segmentation method differs from previously reported approaches applied to schistosomula, which used a single threshold and aimed to segment touching larvae [40], [41]. The protocol we describe uses an adaptive threshold (relative to regional background intensity) and avoids the need to segment touching organisms due to successful larval resuspension prior to imaging. This resulted in capture of sufficient individual larvae (≥10/well) at 4× and 10× for analysis in 97.8% wells from the 10,041 compound library.
Analysis of cells or whole organisms by HCS has the potential to generate image profiles to define characteristic drug phenotypes [42] which may ultimately be interpretable in relation to compound activity, modes of action and molecular targets, currently undefined for any of the known schistosomicides [43]. Different schistosomicides induce a range of distinct morphological effects in both adults [19] and larvae [20] which led us to develop the treatment class model, grouping test compounds causing similar effects to the anti-schistosome drugs. In screening the 10,041 compound library, larval treatment classes ascribed to adult hits with IC50<10 µM were OLT 5, PZQ 1, OX 1. Using an alternative approach to our Bayesian treatment class model, agglomerative hierarchical clustering or DBSCAN has been recently developed [41] to ascribe larval phenotypes according to defined morphological classes. The utility of such classifications models according to phenotype may become clear as more hits are detected and structure/activity relationships are understood.
The Bayesian models developed here can also be readily modified by uploading larval images of any novel phenotypic effects identified during image review for any compounds of particular interest e.g. those which show adult worm activity. Similarly the models could be modified by addition of images of novel phenotypes deemed on manual review not to warrant hit status e.g. those causing internal vacuolation identified during our screening.
The HCS was validated using several compound collections. At 10 µM, a commonly used concentration for primary screening, the models reliably and reproducibly distinguished all of our reference schistosomicides from the controls with the exception of DHA and Ro15. This is not considered a failure of the primary assay since both of these compounds are inactive in the visual larval and adult worm in vitro assays at 10 µM. Based on the results with the anti-schistosome drugs, hit thresholds of −0.15 for phenotype and −0.35 for motility were established and proved robust in detecting 40 previously tested adult hit compounds [19] and 99.8% of visually assessed hits from the 10,041 compound library. The HCS produced a false positive rate of 1.1%, mostly (78%) due to the precipitation of test compounds in wells which in fact contained healthy parasites and which were readily redefined as non-hits during routine reviewing of the automated plate reports.
The HCS offers significant advantages over other recent approaches to objective quantitation of schistosomula damage. Peak et al. (2010) [23] successfully assessed severe drug effects based on uptake of fluorescent markers but the assay involves multiple wash steps and was unable to detect more subtle effects e.g. caused by PZQ. An assay based on use of the redox indicator, Alamar Blue, was similarly less sensitive than visual assessment for more subtle effects and was influenced by variation in parasite numbers per well [21]. Isothermal microcalorimetry [44], assessment of motility via electrical impedance measurement [45] and optical flow [37] are also able to quantitate drug effects in schistosomes, but are not currently readily adaptable to high throughput applications.
Much of the HTS described is automated and simple to operate. Once test plates have been set up and left for 3 days in the Cytomat incubator, the image capture and analysis systems would run automatically after opening “Overlord 3”, the automation control software and pressing “run”. Thereafter, plates emerge from the Cytomat and are robotically moved around for barcode reading, parasite resuspension, imaging and then return to the cytomat. Subsequent analysis involves opening Pipeline Pilot 8.5 and running the analysis protocol. Once complete the plate reports are accessible within an intranet web-port. To become sufficiently familiar with the current customized system would require only a couple of days of training and completing a few test runs. Basic modification of the analysis protocols e.g. to run fewer test compounds in a plate, would require some familiarity with Pipeline Pilot 8.5 as well as MetaXpress software which controls the IXM microscope. More significant alterations of the protocols would require in-depth knowledge of all the different software involved. Currently the platform takes 2 hrs to image a 384 well plate and a further 2 hrs to analyse the phenotype and motility images. The system can be programmed to start imaging at any time of the day and could run close to continuously. So if test plates were set up on each of 4 days per week, the throughput, limited by plate reading, would be ∼48 plates or 16,896 compounds/week which would require around 2×106 cercariae/week. In fact it is cercarial production which is limiting our current throughput capability to around 10 plates twice per week (∼7,000 compounds/week, 350,000/year).
In conclusion, the HCS described is suitable for primary screening of large compound collections for activity against schistosomes. Further studies are ongoing to adapt this system to screen against several species of nematode larvae of medical and veterinary importance, which may allow parallel testing of libraries against various helminths.
|
10.1371/journal.pntd.0006899 | Continuing evidence of Chagas disease along the Texas-Mexico border | Chagas disease is a chronic parasitic infection that progresses to dilated cardiomyopathy in 30% of human cases. Public health efforts target diagnosing asymptomatic cases, as therapeutic efficacy diminishes as irreversible tissue damage progresses. Physician diagnosis of Chagas disease cases in the United States is low, partially due to lack of awareness of the potential burden in the United States.
The current study tested a patient cohort of 1,196 Starr County, Texas residents using the Hemagen Chagas ELISA Kit as a preliminary screening assay. Samples testing positive using the Hemagen test were subjected to additional confirmatory tests. Two patients (0.17%) without previous Chagas disease diagnosis were identified; both had evidence of acquiring disease in the United States or along the Texas-Mexico border.
The Texas-Mexico border is a foci of Chagas disease human cases, with a local disease burden potentially twice the national estimate of Hispanic populations. It is imperative that physicians consider persons with residential histories along the Texas-Mexico border for Chagas disease testing.
| Chagas disease is a parasitic infection, which in 30% of people results in chronic dilated organomegaly over the course of decades. This progressive disease typically advances subclinically until end stage heart failure is established, at which point treatment is no longer efficacious. Thus, current public health efforts target screening of high-risk populations for early disease detection and clinical management for improved health outcomes. Although historical evidence suggests the Texas-Mexico border as a region with high Chagas disease prevalence, surveillance is nonexistent. Our study indicates this region likely has an increased risk for undiagnosed morbidity. Furthermore, occupational histories were highlighted as key risk factors for both cases in this report. Physicians should consider Chagas screening in persons with residential histories along the Texas-Mexico border.
| Chagas disease is caused by infection with the protozoan parasite Trypanosoma cruzi. Primarily transmitted by infected triatomine vector species in the Americas, congenital transmission, blood transfusion, organ transplantation, or ingestion of contaminated beverage/food products are also known to contribute to human disease. The estimated global disease burden is over 6 million [1], with up to 30% of infected patients experiencing clinical symptoms as a result of infection [2,3]. Over the course of decades, clinical disease manifests as progressive cardiac or gastrointestinal tissue dilation. Early identification of these clinical cases is critical, as the presence of irreversible tissue damage is inversely proportional to drug efficacy [4] and survival [5].
Eleven triatomine species exist in the United States–[6] with historical reports of these species dating back to the early 1800s [7]. Sylvatic transmission cycles, with raccoons, opossums, woodrats and dogs serving as important mammalian reservoirs have been reported in seventeen southern states [8]. Despite continual documentation of T. cruzi positive triatomines and animal reservoir species in the United States [6,9], autochthonous human Chagas disease cases have been rarely described. Of the six states with published evidence of autochthonous vector-borne transmission to humans (n = 43 total cases) (Arizona, California, Louisiana, Mississippi, Tennessee, and Texas), Texas is home to a disproportionate burden of reported cases [6, 10–14].
Specifically, the Rio Grande Valley region has been identified as a hot-spot of positive patient populations based on historical evidence demonstrating that in 1980, 2.4% of residents screened positive for Chagas [15]. This Texas-Mexico border region (Cameron, Willacy, Hidalgo, and Starr counties) has long been theorized to be a high-risk area for vector-borne disease due to pervasive poverty, substandard housing, barriers to healthcare access, and weaknesses in the public health infrastructure [16]. Renewed efforts of Chagas disease surveillance in the state have peaked interest in understanding the contemporary disease burden in this potentially high-risk community. A recent screening of banked sera from Cameron County, Texas revealed that 0.36% of residents had a confirmed Chagas disease infection [12]. With each colonia having different and unique social determinants of health [17], our current study aimed to continue surveillance in neighboring Starr County and to further understand the Chagas disease prevalence and risk factors for infection in this local population.
In February 1981 the Starr County Health Studies program was established by investigators from the University of Texas Health Science Center at Houston with the opening of a field office in Rio Grande City, Texas. Since then and continuing today, a series of studies has been conducted to understand the genetics and epidemiology of type 2 diabetes, its complications and related conditions. In total, some 10,000 local residents have participated in more than 28,000 examinations generating more than 500,000 banked biological samples [18].
For the purposes of our current investigation, banked sera from 1,196 study participants were available for testing from study of type 2 diabetes, sleep apnea, and endothelial function [19]. All samples were initially screened by ELISA using Hemagen Chagas Kit (Hemagen Diagnostics Inc, Columbia, MD). Any samples positive by Hemagen Chagas Kit were further tested using Chagas STAT-PAK assay (Chembio Diagnostic Systems Inc, Medford, NY) and Chagas DPP assay (Chembio Diagnostic Systems Inc, Medford, NY). Lastly, any sample positive by one or more of the described tests was sent to the US Centers for Disease Control (CDC) Parasitic Disease Branch for confirmation testing: Weiner EIA and trypomastigote excreted-secreted antigens (TESA) immunoblot. Confirmed positives were defined as being positive on at least one study assay and at least one CDC confirmation assay. Confirmed positives were invited to take part in a follow-up examination, comprised of a medical chart abstraction, a contemporary health survey, and electrocardiogram (ECG).
Protocols were approved by institutional review boards at the University of Texas Health Science Center at Houston (HSC-SPH-02-042) and Baylor College of Medicine (H-35471), respectively. Follow-up evaluation of Chagas seropositive patients was performed through enrollment under Baylor College of Medicine (BCM)’s protocol. All adult subjects provided informed written consent.
From 2010 to 2014, 1,196 Starr County residents were enrolled in a comprehensive examination that included anthropometric evaluations, electrocardiography (ECG), echocardiography (ECHO), medical and medication histories, an in-home overnight sleep evaluation, end evaluation of endothelial function and aortic stiffness (19). Fasting plasma, serum and urine specimens were obtained from all individuals while those with no previous diagnosis of diabetes having an oral glucose tolerance test and those with type 2 diabetes having a standard meal challenge. Of the 1,196, 602 were classified as having type 2 diabetes. Complete details of the sampling and disease burden are in Hanis et al. (2016) [19].
Participants (n = 1,196) previously enrolled were tested for Chagas and eight screened positive by Hemagen Chagas Kit (Table 1). Of these eight individuals, one was positive by Chagas STAT-PAK assay, two were positive by DPP, two were positive by Weiner EIA, and one was positive by TESA immunoblot. Per our diagnosis criteria, we determined that two participants (Sample IDs #4788 and #5070) were Chagas positive. The follow-up risk factor analysis, clinical assessments, and locations of importance (Fig 1) are listed below.
Case-patient 1 was a 76-year old Hispanic female with a 43-year history of diabetes mellitus and 20-year history of hypertension. On May 11, 2017, she tested positive for Chagas antibodies on Hemagen Chagas Kit, DPP and CDC Weiner EIA, and negative for Chagas antibodies on STAT-PAK and CDC TESA immunoblot. Her ECG on August 23, 2017 indicated the presence of a 1st atrioventricular block and left ventricular hypertrophy with repolarization abnormality. She reported a history of an “enlarged heart” but was unable to provide additional clinical details at the time of assessment nor did she complete an echocardiogram. While she regarded her medical management as acceptable, her clinical complaints at the time of follow-up were continued pedal edema and inability to climb two flights of stairs.
She was born in Mission, Texas where she has lived her whole life with the exception of 15 years during her early adulthood when she lived in Bartow and Winter Haven, Florida. Case-patient 1 was a mother to four children, now all adults, who were unavailable or uninterested in Chagas testing. She reported no travel to Latin American countries except for infrequent shopping day trips at the border town of Reynosa, Mexico. Her risk factors for acquiring Chagas included possible congenital transmission from her mother (born, raised, and lived in Nuevo Leon, Mexico) and occupational exposures. As a child, she reported frequently performing migrant work in Tennessee, west Texas, California, Iowa, and Colorado. She reported staying in “shacks” while working, but had never seen triatomine insects. Based on her discordant testing results, indicating a low antibody titer, we theorize that transmission occurred earlier in her life, in either Texas or a southern state in which she previously worked. This is not surprising since both quantitative (decrease in the number of antibody-secreting B-cells) and qualitative (activity of different B-cell subsets including changes to the specificity of the antibody repertoire) changes are associated with aging [20]. More specifically, decrease in IgG titers with specificity to vaccine antigens have been described [21].The exact timing cannot be determined based on her lack of triatomine recognition and reported lack of triatomine exposure.
Case-patient 2 was a 35-year old Hispanic male with an unremarkable health history. On May 11, 2017, he tested positive on all three in-house tests and both of CDC’s confirmatory tests. At his follow-up appointment on August 23, 2017, his ECG was normal and he had no clinical complaints. He was born in Rio Grande City, Texas, where he resided for 25 years before moving to Alabama for 2 years, and then returning back to Rio Grande City, Texas. From the ages of 17–25, he split his time between Rio Grande City, Texas and San Vicente, Nuevo Leon, Mexico (74 miles away). His mother and maternal grandmother were both from Nuevo Leon, Mexico, posing the potential for congenital transmission.
His family has had ranches in Rio Grande City and San Vicente where he has worked as his primary occupation for over 20 years. His job responsibilities have included fencing, feeding animals, clearing land, and gardening. The family ranch in Mexico has had collective animal housing (goats, chickens, cows, lambs, and goats), where he reported “many times” seeing triatomines in the animal bedding. He also reported seeing triatomines in Rio Grande City on the trees outside his home. He never recalled seeing triatomines inside his home at either location. Given his extensive history of working in areas endemic to triatomines, he likely acquired infection at one of the two residential locations.
Chagas disease is a significant public health threat in numerous parts of the world. It is mainly considered a tropical disease and has received limited attention in the United States. This study adds to the growing body of work establishing Chagas disease as a considerable public health concern along the Texas-Mexico border. Owing to the significant morbidity among those infected, there should be an increased awareness in endemic triatomine areas and increased testing among those with pathologies consistent with infection such as unexplained heart failure. Historical evidence of autochthonous transmission in the state dates back to 1955 and surveillance studies have continually demonstrated high disease burdens among Hispanic foreign-born populations [22]. Despite clear evidence presented here and in our earlier screen in Cameron County [12] that Chagas continues to be present, there is no unified public health set of guidelines for its screening in either the general or specialized clinic population. Consequently, Chagas disease has long been regarded as the most neglected of the neglected tropical diseases [23], and scientific literature suggest that this ‘neglect’ extends to some regions along the Mexico-United States border [24–26]. Lack of awareness by physicians, barriers in healthcare access, a paucity of efficacious treatment options, and substandard public health resources continue to contribute to preventable morbidity and mortality due to Chagas disease nationally [27].
It is critical we clarify the disease burden and epidemiology of transmission to improve physician awareness. An important component of developing enhanced patient profiles for physician education is identifying foci of vector-borne transmission and elevated population disease burdens. Our study demonstrated that Chagas disease infections in Starr County, Texas are higher than the estimated prevalence among foreign-born Hispanic United States residents [0.17% (2/1,196) vs. 0.09% (300,000/323,000,000[28]), respectively]. Furthermore, our study is consistent with a previous study which found 0.36% infection prevalence from human residents sampled from 2005–2008 in neighboring Cameron County, Texas [12]. Similarly, the border state of Nuevo Leon, Mexico has a documented history of autochthonous transmission, with a recent study in 2014 identifying a 2% seroprevalence among residents [2] and a 2017 study identifying 14.5% of sylvatic animals seropositive [29].
An important limitation to our study is our inability to determine the precise incident of transmission. This limitation is inherent due to our serologic diagnostic method of this life-long infection; however, based on the patients’ epidemiologic risk profiles and relatively limited travel histories, the data suggest that both patients likely acquired their infections in the United States or along the Texas-Mexico border. An implication was our finding of discordant testing results, indicating a lack of specificity of commercially available diagnostic assays for diagnosing strains of T. cruzi that evolved in North America compared to South American strains. Chagas disease diagnostics have been suboptimal [30] and our study highlights that Texas-Mexico border patient populations present similar point-of-care testing challenges.
Chagas disease is a neglected tropical disease in the United States. Our research adds to the growing body of evidence that this disease is likely endemic along the Texas-Mexico border with infection prevalence higher than that of the overall foreign-born Hispanic United States resident population. However, due to limitations in the sample size for this study, it is difficult to make assumptions regarding larger populations. Physicians should be aware of the potential elevated disease risk among this geographic patient population.
|
10.1371/journal.pntd.0003952 | Role of sph2 Gene Regulation in Hemolytic and Sphingomyelinase Activities Produced by Leptospira interrogans | Pathogenic members of the genus Leptospira are the causative agents of leptospirosis, a neglected disease of public and veterinary health concern. Leptospirosis is a systemic disease that in its severest forms leads to renal insufficiency, hepatic dysfunction, and pulmonary failure. Many strains of Leptospira produce hemolytic and sphingomyelinase activities, and a number of candidate leptospiral hemolysins have been identified based on sequence similarity to well-characterized bacterial hemolysins. Five of the putative hemolysins are sphingomyelinase paralogs. Although recombinant forms of the sphingomyelinase Sph2 and other hemolysins lyse erythrocytes, none have been demonstrated to contribute to the hemolytic activity secreted by leptospiral cells. In this study, we examined the regulation of sph2 and its relationship to hemolytic and sphingomyelinase activities produced by several L. interrogans strains cultivated under the osmotic conditions found in the mammalian host. The sph2 gene was poorly expressed when the Fiocruz L1-130 (serovar Copenhageni), 56601 (sv. Lai), and L495 (sv. Manilae) strains were cultivated in the standard culture medium EMJH. Raising EMJH osmolarity to physiological levels with sodium chloride enhanced Sph2 production in all three strains. In addition, the Pomona subtype kennewicki strain LC82-25 produced substantially greater amounts of Sph2 during standard EMJH growth than the other strains, and sph2 expression increased further by addition of salt. When 10% rat serum was present in EMJH along with the sodium chloride supplement, Sph2 production increased further in all strains. Osmotic regulation and differences in basal Sph2 production in the Manilae L495 and Pomona strains correlated with the levels of secreted hemolysin and sphingomyelinase activities. Finally, a transposon insertion in sph2 dramatically reduced hemolytic and sphingomyelinase activities during incubation of L. interrogans at physiologic osmolarity. Complementation of the mutation with the sph2 gene partially restored production of hemolytic and sphingomyelinase activities. These results indicate that the sph2 gene product contributes to the hemolytic and sphingomyelinase activities secreted by L. interrogans and most likely dominates those functions under the culture condition tested.
| The spirochete Leptospira causes leptospirosis, a potentially deadly disease of humans and animals. Candidate factors that promote infection include hemolysins encoded by several leptospiral genes. Hemolysins rupture red blood cells in vitro. Some hemolysins are sphingomyelinases, which target sphingomyelin in the host cell membrane. Hemolysins have the potential to disrupt organ function during infection. It is not known which hemolysins and sphingomyelinases are responsible for the hemolytic and sphingomyelinase activities secreted by L. interrogans. We found that the production of hemolytic activity is regulated and is tied to expression of sph2, which encodes a hemolysin with sphingomyelinase, cytotoxic, and fibronectin-binding activities. Hemolytic and sphingomyelinase activities and sph2 expression were higher when the osmolarity of the culture medium was raised to the level found in the mammalian host. Similarly, sph2 expression was substantially higher in an L. interrogans strain that secreted large amounts of hemolytic and sphingomyelinase activities than in a strain that generated negligible amounts. Most importantly, disruption of the sph2 gene eliminated hemolysin production and yielded substantially less sphingomyelinase than the wild-type strain. Our findings indicate that sph2 is a major contributor to the hemolytic and sphingomyelinase activities secreted by L. interrogans and that the hemolytic and sphingomyelinase activities measured in standard L. interrogans cultures may underestimate the levels produced during infection.
| Leptospirosis is a neglected zoonotic disease that afflicts humans and animals [1–3]. Although the disease occurs worldwide, it is observed most frequently in tropical countries, where conditions for environmental survival and transmission are most favorable [4]. The causative organisms are spirochetes belonging to the genus Leptospira, which comprise pathogenic and nonpathogenic species. The bacteria are able to enter the mammalian host through skin abrasions and mucous membranes. Following entry, the spirochetes disseminate via the bloodstream to many organs [2]. Patients exhibit a wide range of signs and symptoms from undifferentiated fever to liver dysfunction, renal insufficiency, hemolytic anemia, bleeding, and respiratory failure [5]. Proposed mechanisms of pathogenesis include vascular damage, which is frequently manifested as hemorrhage in the lungs and other organs on post-mortem examination [6]. Additionally, reproductive failure is observed in livestock animals with leptospirosis [7]. Leptospira bacteria are spread and maintained in the environment by rats and other reservoir hosts, whose renal tubules are persistently colonized by the spirochete. Humans and animals can be infected by direct or indirect contact with contaminated urine or other tissue.
Many Leptospira strains secrete hemolysins and sphingomyelinases during in vitro growth [8–11]. Strains of serovar Pomona secrete the highest levels of hemolytic activity [9] and cause hemolytic anemia in ruminants [12–14]. The single peak of hemolytic activity detected by isoelectric focusing of the culture supernatant fluid of one Pomona strain co-purified with sphingomyelinase C activity, suggesting that hemolysis is due to sphingomyelinase [15]. It is not known which gene or genes encode the secreted hemolytic and sphingomyelinase activities.
Hemolytic activity has been reported for purified recombinant forms of the L. interrogans protein HlyX, HlpA, TlyA, and the sphingomyelinase-like paralogs Sph1, Sph2, Sph3, and SphH [16–19]. Among the sphingomyelinase paralogs, sphingomyelinase C activity has been demonstrated only for Sph2, which cleaves sphingomyelin to ceramide and phosphocholine [20]. Sphingomyelinase activity has also been reported for Sph1, Sph3, and Sph4 [21], although these proteins lack some of the catalytic amino acid residues required for enzymatic activity [22]. The genes encoding the sphingomyelinase-like proteins are missing from the nonpathogen Leptospira biflexa. Similarly, sphingomyelinase activity has been detected only in pathogenic strains of Leptospira [10]. These observations suggest that sphingomyelinase-like proteins function during infection [23].
Hemolysins are assumed to assist microbial pathogens in acquiring iron by lysing host erythrocytes during infection [24]. Heme can be used by L. interrogans as an iron source for growth [25]. They express the hemin-binding protein HbpA, which may participate in transporting heme into the cell, and a heme oxygenase, which is required for heme utilization and for virulence in the hamster model of leptospirosis [26–28]. Iron depletion resulted in increased levels of a sphingomyelinase-like protein, which is consistent with the notion that hemolysins are involved in iron acquisition by L. interrogans [29]. Nevertheless, experimental evidence indicates that leptospiral hemolysins are capable of pathogenic processes that do not involve red blood cells. For example, recombinant Sph2 protein is cytotoxic towards equine endothelial cells, mouse lymphocytes and macrophages, and a human liver cell line [18, 30]. Additionally, L. biflexa ectopically expressing sph2 captures plasma fibronectin in vitro [31].
Sph2 has not been detected in appreciable amounts in cell lysates of L. interrogans grown in the standard culture medium EMJH except in strains of serovar Pomona [32]. However, sph2 expression in the Fiocruz L1-130 strain is dramatically upregulated by environmental conditions that occur during infection. In our previous whole-genome microarray study examining the changes in transcript levels to an increase in osmolarity from low osmolarity found in EMJH medium to physiologic levels found in the mammalian host, sph2 was the second most strongly upregulated gene in the Fiocruz L1-130 strain [33]. Levels of cellular and extracellular Sph2 in the Fiocruz L1-130 strain were also upregulated upon addition of sodium chloride or sucrose to physiological levels of osmolarity [34]. Anti-Sph2 antibody is detected in patients with leptospirosis, suggesting that sph2 is expressed during infection [32]. In a recent RNA-seq experiment, sph2 transcript levels were higher in L. interrogans bacteria growing in dialysis membrane chambers in rats than in those growing in in vitro culture [35]. These observations suggest that upregulation of sph2 is involved in L. interrogans infection.
The enzymatic domain of Sph2 is flanked by unusual sequences missing from bacterial sphingomyelinases whose crystal structures have been determined [20]. Three or four ~20 residue imperfect tandem repeats are present to the amino terminus of Sph2, and the carboxy terminus comprises a unique segment of ~186 amino acid residues [18, 22]. The enzymatic domain may not be sufficient for hemolytic activity of Sph2. A sphingomyelinase with a similar C-terminal extension is found in the sphingomyelinase protein of a fish isolate of Pseudomonas species [36]. Removal of the C-terminal domain eliminates its hemolytic activity despite its enzymatic domain being left intact [36]. L. interrogans releases Sph2 in smaller forms whose apparent molecular masses are ~13 and ~21 kDa less than that of the cell-associated form of Sph2 [34]. These observations raise the question of whether L. interrogans secretes Sph2 in an active form because the hemolytic activity of Sph2 was demonstrated with full-length recombinant protein instead of the shortened forms detected in culture supernatant fluids [16, 18, 34]. One approach to establish that Sph2 contributes to the secreted hemolytic activity is to demonstrate inhibition of hemolysis when Sph2 antibody is added to the spent culture medium. However, both anti-Sph2 antiserum and normal serum inhibited the hemolytic activity secreted by a Pomona strain of L. interrogans [32].
In this study, we extend our earlier findings of osmotic induction of sph2 expression with the Fiocruz L1-130 strain to show that the sph2 genes of four additional strains of L. interrogans are also regulated by sodium chloride, including a Pomona strain that displays high levels of basal sph2 expression. In addition, we demonstrate that the levels of secreted hemolytic and sphingomyelinase activities are regulated by osmolarity. In our attempts to maximize sph2 expression, we also found that rat serum increased Sph2 production further at physiologic osmolarity, although the presence of serum inhibited hemolytic activity in liquid-phase assays. Finally, we present genetic evidence that sph2 contributes to the hemolytic and sphingomyelinase activities secreted from L. interrogans.
The leptospiral strains Leptospira interrogans serovars Manilae (strain L495), Pomona subtype kennewicki (strain LC82-25), and Lai (strains 56601 and L391) were grown at 30°C in EMJH medium assembled with Probumin BSA (Vaccine Grade, lot 103) as described previously [37] or purchased as Probumin Vaccine Grade Solution (lot 103) from EMD Millipore (Temecula, California, USA). The Pomona LC82-25 and Lai 56601 strains were obtained from Rich Zuerner (National Animal Disease Center, Ames, Iowa, USA) and Mathieu Picardeau (Institut Pasteur, Paris, France), respectively. Leptospira interrogans serovar Copenhageni strain Fiocruz L1-130 was grown at 30°C in EMJH medium supplemented with 1% heat-inactivated rabbit serum (Rockland; Gilbertsville, Pennsylvania, USA). The L391 strain and an L391 mutant with a kanr-marked Himar1 transposon inserted in the sph2 gene were provided by Gerald Murray [38]. The sph2 mutant was maintained in EMJH containing 40 μg/mL kanamycin. Culture densities were determined by directly counting the number of Leptospira using an AxioLab A1 microscope with a darkfield condenser (Zeiss Microscopy Division; Pleasanton, California, USA). To examine the influence of environmental conditions on sph2 expression, L. interrogans cells were grown to a cell density of 0.6–1 x 108 cells/mL and then supplemented with 120 mM NaCl, 10% rat serum (Rockland Immunochemicals, Limerick, Pennsylvania, USA), or a mixture of nine parts of EMJH supplemented with 120 mM NaCl and one part of rat serum and allowed to incubate for 4 h. Cultures were harvested by centrifugation at 9,200 x g for 20 min at 4°C in a Sorvall high-speed centrifuge (Thermo Scientific; Marietta, Ohio, USA), and the spent medium was collected for storage. Cells were washed once with cold PBS-5 mM MgCl2. Sodium chloride was added to the spent medium obtained from the EMJH cultures to equalize the osmolarity across all samples. Cell pellets and spent growth medium were stored in -80°C.
For Western blots, the proteins were fractionated on 10% PAGEr Gold precast Tris-Glycine gels (Lonza; USA). Dual Color Precision Plus Protein Standards from Bio-Rad (catalog # 161–0374) (Hercules, California, USA) were included in adjacent lanes. Proteins were transferred from gels on to PVDF Immobilon-P transfer membrane (EMD Millipore; USA) using Trans-blot SD Semi-Dry transfer Cell system (Bio-Rad) (10V for 45 min). The membranes were incubated in blocking solution (5% skim milk in PBS-0.05% Tween 20 [PBS-T]) for 30 min and then incubated with rabbit anti-Sph2 and anti-LipL41 antisera [34, 39] at dilutions of 1:1,000 and 1:10,000, respectively, for 30 min and washed three times (5 min each) with PBS-T. The membrane was then incubated with donkey anti-rabbit antibody (1:5,000) (Amersham Biosciences; Piscataway, New Jersey, USA) or protein A-horseradish peroxidase conjugate (1:3,000) (Amersham Biosciences) in blocking solution for 30 min and again washed three times with PBS-T. The membranes were developed with the ECL Western blot detection system (Thermo Scientific), and the bands were visualized with Hyperfilm (Amersham Biosciences).
Immunoprecipitation of Sph2 from the spent culture medium was performed as described [37], except the culture supernatant fluid was preadsorbed with 25 μL of EZview Red Protein A Affinity gel (Sigma-Aldrich; St. Louis, Missouri, USA) for 60 min at 4°C prior to immunoprecipitation with anti-Sph2 antiserum [32]. The resuspension volumes of the immunoprecipitates with Laemmli sample buffer were adjusted according to the cell densities of the cultures.
Restriction enzymes and T4 DNA ligase were obtained from New England Biolabs (Beverly, Massachusetts, USA). PCR amplifications were conducted with Phusion DNA polymerase (Thermo Scientific). L. interrogans sequences cloned into plasmid vectors were verified by Sanger sequencing (Laragen; Culver City, California, USA).
The sph1, sph3, and sph4 sequences were amplified by PCR with the primer pairs lic12632(Nd)-3F and lic12632(Xh)-4R, lic13198(Nd)-3F and lic13198(Xh)-4R, and lic11040(Nd)-3F and lic11040(Xh)-4R, respectively (Table 1). Forward and reverse primers included NdeI and XhoI restriction sites near the 5' ends. Genomic DNA from L. interrogans Fiocruz L1-130 served as template. Each amplicon was digested with NdeI and XhoI and ligated to pET20b+ (EMD Biosciences; La Jolla, California USA) that had been digested with the same enzymes. The sph1, sph3, and sph4 expression plasmids were named pRAT547, pRAT548, and pRAT549, respectively. The sph2 expression plasmid pTOPO-Sph2(27–623) was described previously [34].
To construct the sph2 complementation plasmid, the oligonucleotides lic12631(Kp)-17F and lic12631(Xh)-18R (Table 1) were used to PCR amplify sph2 along with its flanking regions using Fiocruz L1-130 genomic DNA as template. The amplicon included sequences 327 bp upstream and 130 bp downstream of the start and stop codons, respectively. Restriction sites for KpnI and XhoI were included near the 5' ends of the oligonucleotides. Following digestion of the amplicon with KpnI and XhoI, the sph2-containing fragment was inserted into the similarly-digested plasmid pRAT575 [40] to create pRAT613 for the current and future studies. The KpnI-XhoI fragment containing sph2 was subsequently transferred to the mobilizable plasmid pAL614, which contains KpnI and XhoI sites between the ends of a Himar1 element that also harbored a gene encoding resistance to spectinomycin [41]. The resulting sph2 plasmid was designated pRAT708.
Frozen competent E. coli BLR(DE3)/pLysS (EMD Biosciences) was transformed with the plasmids pRAT547 (sph1), pRAT548 (sph3), pRAT549 (sph4), and pTOPO-Sph2(27–623) (sph2), and transformants were selected on LB plates containing 100 μg/mL carbenicillin. Colonies were inoculated into 10 mL LB containing carbenicillin for overnight growth at 37°C. 5 mL of each overnight culture was then placed into 200 mL LB with carbenicillin and incubated at 37°C. When the OD at 600 nm reached 0.6, 1 M IPTG was added to a final concentration of 0.5 mM. For expression of recombinant Sph1, Sph3, and Sph4, the cultures were incubated for 2 hrs at 37°C. The bacteria were collected by centrifugation, and the cell pellets were stored at -80°C. For expression of recombinant Sph2, the culture was incubated overnight at 16°C and harvested by centrifugation. To lyse the bacteria, cell pellets were suspended in 5 mL (for Sph1, Sph3, and Sph4) or 10 mL (Sph2) of BugBuster (EMD Biosciences) containing 20 unit/mL DNase I (Thermo Scientific) and 0.25 mM phenylmethylsulfonyl fluoride (Sigma-Aldrich), and the suspension was swirled for 20 min at room temperature. The lysates were poured into Corex tubes and subjected to centrifugation at 15,000 x g for 20 min at 4°C in a Sorvall RC-5B Superspeed centrifuge.
To wash the Sph1, Sph3, and Sph4 inclusion bodies, the pellets were suspended in 10 mL of a 10-fold dilution of BugBuster. Following centrifugation at 15,000 x g for 20 min at 4°C, the pellets were suspended in 5 mL of 100 mM NaH2PO4, 10 mM Tris-HCl, and 8M urea, pH 8.0 (Buffer B) and swirled for 15 min at room temperature to solubilize the inclusion bodies. The material was centrifuged at 10,000 x g at room temperature to pellet the cell debris. 1 mL of 50% Ni2+-NTA slurry (Qiagen) was added to the solubilized inclusion bodies, and the suspension was mixed for 60 min at room temperature. The material was then poured into a 5-mL polypropylene column (Qiagen) for purification of the recombinant protein. The column was washed with 4 mL Buffer B that was adjusted to pH 6.3 and then with 2 mL Buffer B at pH 5.9. The recombinant protein was eluted with Buffer B adjusted to pH 4.5, and 0.5 mL fractions were collected. The fractions were neutralized by adding 55.5 μl of 1 M Tris-HCl, pH 8.0, to each.
To purify soluble Sph2, 250 μl of 0.5 M imidazole and 2 mL of the 50% Ni2+-NTA slurry (Qiagen) were added to 10 mL of lysate. The mixture was mixed for 60 min at 4°C and then poured into a 5 mL polypropylene column. After collecting the flow through, the column was washed at 4°C with a total of 30 mL of 20 mM imidazole in Wash Buffer (50 mM NaH2PO4, 0.5 M NaCl, pH 8.0). The protein was eluted at 4°C with 0.5 M imidazole in Wash Buffer, and 1 mL fractions were collected. Protein concentrations of all purified Sph proteins were determined using the Pierce BCA Protein Assay kit (Thermo Scientific) with BSA standards.
For preparation of RNA, 25 mL of culture was transferred into an Erlenmeyer flask, quickly chilled by swirling for 7 s in a dry ice-ethanol bath, and centrifuged at 9,200 x g for 20 min at 4°C. RNA was isolated from leptospires as follows: 1 mL Trizol (Life Technologies; Grand Island, New York, USA) was added to the cell pellet, resuspended thoroughly by pipetting followed by incubation at room temperature for 5 min. 200 μL of chloroform (Sigma-Aldrich) was added, mixed vigorously for 15 s followed by incubation at room temperature for 3 min. The tubes were centrifuged at 12,000 x g for 15 min at 4°C, and the upper aqueous layer was pipetted into fresh tubes. RNA was precipitated by the addition of equal volumes of isopropanol, mixed and incubated for 10 min and subjected to centrifugation at 16,000 x g for 15 min at 4°C. The pellet was washed with 1 mL of 75% ethanol, air dried and dissolved in 84 μL of RNase free water. The dissolved RNA was protected from degradation by addition of 1μL of 20 U/μL SUPERase In RNase inhibitor (Life Technologies). DNA was digested by addition of 10 μL of 10X Turbo DNase buffer and 5 μL of Turbo DNase followed by incubation in 37°C water bath for 2 h. The RNA samples were subjected to clean-up using the RNeasy Mini kit (Qiagen; USA) as described in the manufacturer’s instructions. The concentration of the RNA was determined using the NanoVue spectrophotometer (GE Healthcare; Piscataway, New Jersey, USA) and the quality of RNA was assessed by examining the A260/A280 and A260/A230 ratios and by electrophoresis of 200 ng of RNA onto a 1.2% agarose gel.
cDNA was synthesized using the iScript cDNA Synthesis Kit (Bio-Rad) following the manufacturer’s instructions. The reaction was set up in a microcentrifuge tube in a reaction volume of 20 μL containing 4 μL of 5X iScript Reverse Transcription Supermix, 1 μg RNA, and nuclease-free water. cDNA synthesis was carried out in a Master Cycler gradient thermal cycler (Eppendorf; Hauppauge, New York, USA) with priming at 25°C for 5 min followed by reverse transcription at 42°C for 30 min and enzyme inactivation at 85°C for 5 min. A separate reaction without reverse transcriptase was included as a negative control. The synthesized cDNA was diluted with RNase free water (Ambion) to obtain a working concentration of 10 ng/μL.
The quantification of cDNA was done using real-time PCR using iQ5 SYBR Green Supermix (Bio-Rad) and the iQ5 Multicolor Real-time PCR Detection System (Bio-Rad). Each reaction contained cDNA derived from 50 ng of RNA, 10 pmol of each forward and reverse primers, and 12.5 μL of 2X iQ5 SYBR Green supermix in a total volume of 25 μL. The sph1-, sph2- and lipL41-specific primers (Table 1) were designed using Primer Premier (Premier Biosoft; Palo Alto, California, USA). The amplification protocol consisted of an initial denaturation for 15 min (95°C) followed by 40 cycles of amplification (15 s at 95°C, 30 s at 58°C, 30 s at 72°C) and a final extension of 2 min at 72°C. Standard curves were constructed by 5-fold serial dilutions (50 ng to 0.08 ng) of cDNA as template in triplicate. Amplification efficiency was evaluated from the standard curves by determining the E value; results in the range of 90% to 110% were considered to be acceptable. The CT values were normalized to that of lipL41, and expression levels were calculated using 2ΔΔCT method [42]. The relative fold change was calculated by comparison with the EMJH control. The assay was performed using three biological replicates.
To assess hemolytic activity qualitatively, 5 μL of culture supernatant fluid was spotted onto BBL Trypticase Soy Agar with 5% sheep red blood cells (TSA II) (Becton Dickinson; Sparks, Maryland, USA). Because magnesium is necessary for sphingomyelinase activity, 100 mM MgCl2 was added to the culture supernatant to a final concentration of 10 mM prior to spotting the plates. 0.05 units of Bacillus cereus sphingomyelinase (Sigma-Aldrich) was spotted as a positive control. The plates were incubated for 20 hours at 37°C and then at 4°C for at least three days.
The liquid-phase hemolysis assay was set up in a 96 well round-bottomed microtiter plate as reported earlier [20] with some modifications. Briefly, sheep erythrocytes were procured commercially from Quad Five (Ryegate, Montana, USA) as a 50% (v/v) suspension in Alsever's's solution. Erythrocytes were collected by centrifugation at 400 x g for 10 min at 8°C, washed three times with cold PBS (pH 7.4), and resuspended in cold PBS to a final concentration of 10%. Each reaction mixture (200 μL) contained 10 mM MgCl2 in PBS, with 40 μL 10% washed sheep erythrocytes and 100 μL of culture supernatant fluid. For background measurements, EMJH with 120 mM sodium chloride replaced the spent medium. Three biological replicates were examined. The hemolysis reaction proceeded at 37°C for 90 min followed by incubation at 4°C for 30 min. The plate was centrifuged at 800 x g in an Eppendorf 5430 centrifuge to pellet intact erythrocytes, and the supernatant fluid from each well was transferred to a flat-bottom 96-well ELISA plate. The plate was read in an iMark Microplate Absorbance Reader (Bio-Rad) at 415 nm. Percent hemolysis was calculated by multiplying the PBS background-subtracted absorbance of the sample by 100 and dividing by the absorbance of the osmotically-lysed erythrocytes.
Sphingomyelinase activity was measured by a coupled assay using the Amplex Red Sphingomyelinase assay kit (Molecular Probes, Invitrogen, USA) as described in the manufacturer’s instructions. The reactions were set up in 96-well special optics flat clear bottom black polystyrene Microplate (Corning, product # 3720). The reaction mixture (200 μL) contained 100 μL test sample and 100 μL of 100 μM Amplex red reagent (containing 2 U/mL horseradish peroxidase, 0.2 U/mL choline oxidase, 8 U/mL alkaline phosphatase, and 0.5 mM sphingomyelin) (Life Technologies). The reaction proceeded for 90 min at 37°C. The fluorescence was measured at excitation and emission wavelengths of 530 nm and 590 nm respectively using the Synergy2 Multi-Mode Microplate Reader (BioTek; Winooski, Vermont, USA). The background fluorescence was corrected by subtracting the negative control, which contained the reaction buffer without sphingomyelinase. Standard curves were generated with B. cereus sphingomyelinase, and the measurements were fit by nonlinear regression to a hyperbola (one-site binding) model with GraphPad Prism, version 5.04 (GraphPad Software; La Jolla, California, USA). Experiments were performed with three biological replicates.
For complementation of the sph2 mutant, the mobilizable sph2 plasmid pRAT708 was transformed into the diaminopimelic acid auxotroph E. coli β2163, which expresses the RP4 conjugation machinery [43]. The plasmid was transferred into the L. interrogans sph2 mutant by conjugation as described [44, 45]. Transconjugants were selected on EMJH agar plates containing 40 μg/mL kanamycin and 40 μg/mL spectinomycin. After two weeks of incubation at 30°C, colonies were inoculated into EMJH liquid medium. The insertion site of the sph2-containing transposon was identified by nested PCR using primers TnK1 and Deg1 for the first PCR reaction and primers TnkN1 and Tag for the second [45].
All values for hemolytic and sphingomyelinase activities were log transformed prior to statistical analysis to achieve similar variances across all groups. One-way ANOVA was conducted with R version 3.0.3 [46]. The Tukey post test was used for group comparisons.
Our earlier work demonstrated that sph2 gene expression in L. interrogans strain Fiocruz L1-130 (serovar Copenhageni) increases substantially when incubated in EMJH supplemented with sodium chloride to attain physiological osmolarity [33, 34]. In the current study, we examined sph2 regulation in three additional strains of L. interrogans, L495 (serovar Manilae), 56601 (serovar Lai), and LC82-25 (serovar Pomona subtype kennewicki). The Pomona strain was included because members of the serovar produce larger amounts of hemolysin than other serovars [9]. Immunoblots were performed to determine the effects of NaCl and serum on Sph2 protein levels. As shown in Fig 1 (lane 1), Sph2 was not detected in the Fiocruz L1-130 [34], Manilae L495, or Lai 56601 strain when grown in EMJH. The addition of 120 mM sodium chloride to EMJH increased Sph2 production to detectable levels in these three strains (lane 2). On the other hand, Sph2 was detected easily in the Pomona LC82-25 strain growing in EMJH, and Sph2 levels increased further when the medium was supplemented with 120 mM sodium chloride (Fig 1C, lanes 1 and 2). The presence of rat serum to the culture medium had an effect on Sph2 expression independent of osmolarity; Sph2 levels were noticeably higher than that when only sodium chloride was present, despite the osmolarity of the culture medium being similar (Fig 1, lanes 2 and 4). Relative to LipL41 levels, Sph2 levels were substantially higher in the Pomona LC82-25 strain than in the Manilae L495 strain under all conditions (Fig 1). The apparent molecular masses of Sph2 was higher than the calculated masses of 71.0, 71.0, 70.4, and 73.5 kDa in the Copenhageni, Lai, Manilae, and Pomona strains, respectively. The apparent molecular mass of Sph2 in the LC82-25 strain was ~13 kDa higher than those of the other strains (Fig 1). As noted in our earlier study and by others, anti-Sph2 antibodies also recognized SphH in these samples (Fig 1) [32, 34].
Sph2 breakdown products were visible by immunoblot after growth of the serovar Pomona strain in 120 mM NaCl with and without serum (Fig 1C). To exclude the possibility that some of the bands originated from the other sphingomyelinase-like proteins, we examined the cross-reactivity of our anti-Sph2 antiserum with purified recombinant forms of Sph1, Sph3, and Sph4. Equal masses of Sph1, Sph2, Sph3, and Sph4 were subjected to SDS-polyacrylamide gel electrophoresis (Fig 2A) and immunoblot analysis with the anti-Sph2 antiserum (Fig 2B). As observed for the native form of Sph2, the recombinant form ran more slowly than expected from its calculated molecular mass, whereas Sph1, Sph3, and Sph4 migrated as expected (Fig 2A). The immunoblot shows that the anti-Sph2 antibody does not cross-react with Sph3 or Sph4 and reacts poorly with Sph1 (Fig 2B). These results indicate that the bands observed below the Sph2 species in the Pomona LC82-25 lysate are Sph2 breakdown products, although we cannot rule out the possibility that the bands running below SphH arose from SphH degradation.
Immunoprecipitation of Sph2 was performed with the spent growth medium from cultures of the Manilae L495 and Pomona LC82-25 strains to assess the effects of salt and serum on the levels of extracellular Sph2. We did not detect Sph2 with the L495 strain grown in EMJH. When EMJH was supplemented with sodium chloride, with or without rat serum, Sph2 released from the L495 strain was detected in two smaller forms (Fig 3, lanes 3 and 4), as shown in our earlier study with Fiocruz L1-130 [30]. The Pomona LC82-25 strain exhibited similar regulation of extracellular Sph2 levels by sodium chloride and rat serum, except the amount of Sph2 detected was much greater than that in the L495 strain (Fig 3, lanes 2–4 vs. 5–7). The two Sph2 bands in the serovar Pomona immunoprecipitates were smaller than the species detected in the corresponding cell lysate (Fig 3, lanes 5–8). The apparent molecular masses of the extracellular forms of Sph2 were greater in the LC82-25 strain than those of the L495 strain, as was the case for the cellular form.
We compared sph2 transcript levels in the serovar Pomona strain LC82-85 with those of the serovar Manilae strain L495 in response to environmental conditions. Quantitative RT-PCR measurements revealed that when grown in standard EMJH medium sph2 transcript levels were 21 fold higher in the Pomona strain than in the Manilae strain (Fig 4). In both strains, the addition of 120 mM sodium chloride to the cultures increased sph2 transcript levels by over 100 fold (Fig 4). At the higher osmolarity, levels of sph2 transcript remained nearly 20 fold higher in the Pomona strain than in the Manilae strain. Addition of 10% rat serum increased sph2 transcript levels by four fold. The ~20-fold difference in sph2 transcript levels between the two strains was maintained when both 10% rat serum and the additional 120 mM sodium chloride were present in the culture medium.
The levels of hemolytic and sphingomyelinase activities in the spent growth medium of the Pomona LC82-25 and Manilae L495 strains grown in EMJH and in EMJH supplemented with 120 mM sodium chloride were determined. To assess hemolytic activity qualitatively, the culture supernatant fluid was spotted onto sheep erythrocyte agar plates. The culture medium, which was adjusted to physiological osmolarity with sodium chloride, did not cause detectable lysis of the erythrocytes (Fig 5A). On the other hand, the spent growth medium from both strains, including those obtained from cultures containing rat serum, caused partial clearance of erythrocytes on the plate (Fig 5A).
The amount of hemolytic activity in the spent growth medium was quantified with a liquid-phase assay. In EMJH, the Pomona strain showed greater hemolytic activity for sheep erythrocytes than the Manilae strain (Fig 5B). Both strains released higher levels of hemolytic activity when EMJH was supplemented with 120 mM sodium chloride. In the process of developing the hemolysis assay, we discovered that hemolytic activity in spent medium from the Pomona strain incubated in EMJH with 120 mM sodium chloride was at least 92% lower when 10% rat serum was also present, even though it was detected in sheep erythrocyte plates (Fig 5A). The hemolysis results with the strains grown in EMJH and EMJH with 120 mM sodium chloride were reflected in an assay for sphingomyelin hydrolase activity (Fig 6). The results from the Western blots, hemolysis assay, and enzymatic assay reveal a correlation between sph2 expression and extracellular hemolysis and sphingomyelinase activities of L. interrogans.
To examine the contribution of sph2 to secreted hemolytic and sphingomyelinase activities by a genetic approach, we were provided with the highly-passaged L. interrogans strain L391 (serovar Lai) and an isogenic sph2 mutant, which was generated by transposon insertional mutagenesis [38]. The transposon is located within sph2 near the end of the segment encoding the enzymatic domain of the protein (Fig 7A). The truncated form predicted to be generated by the mutant allele lacks the second catalytic histidine residue in the enzymatic domain. Substitution of this histidine with alanine in Bacillus cereus sphingomyelinase completely abolishes its enzymatic activity [47]. Therefore, the truncated form expressed from the mutant sph2 allele is unlikely to possess residual enzymatic function. Immunoblot analysis with the sph2 mutant and its wild-type parent confirmed that the mutant was unable to produce the full-length form of Sph2 following incubation at physiological osmolarity for six hours (Fig 8A, lane 4). The sph1 coding region lies 915 bp downstream of sph2 (Fig 7A). Quantitative RT-PCR analysis indicates that sph1 transcript levels were 34% lower in the sph2 mutant compared with that in the wild-type strain.
We introduced an intact copy of sph2 into the mutant for complementation studies. We selected the sph2 gene from the Fiocruz L1-130 strain because of the possibility that the Sph2 protein synthesized by the highly-passaged L391 strain was not fully active. The Sph2 protein sequences of the Fiocruz L1-130 and Lai 56601 strains are 99.5% identical (621/624), with none of the three differences occurring in predicted catalytic residues [22]. The sph2 gene was introduced by conjugation on a suicide plasmid carrying sph2 within a Himar1 transposon carrying a gene encoding resistance to spectinomycin (Fig 7B). Following mating and incubation on selective plates, 15 transconjugants were screened for Sph2 production. The transconjugants exhibited various levels of Sph2 following incubation at physiologic osmolarity (Fig 8B), as would be expected from random insertion of the complementing transposon. Transconjugants that produced high (TC14), medium (TC4), and low (TC10) levels of Sph2 were further examined for control of Sph2 production (Fig 7C). TC14 showed higher basal levels of Sph2 than the wild-type strain (Fig 8C, lane 9 vs 1), and Sph2 levels increased further upon incubation at physiologic osmolarity (Fig 8C, lane 10). The complementing transposon in TC14 was located in the hypothetical gene LA3726, with sph2 in the transposon oriented opposite of the disrupted gene (Fig 7C). In TC10, Sph2 levels were regulated by sodium chloride, but the amount produced at physiologic osmolarity was low relative to the levels produced in the wild-type strain (Fig 8C, lane 8 vs 2). The sph2 gene in the complementing transposon was positioned near the promoter region of the hypothetical gene LB225 with the two genes directed in the same orientation (Fig 7C). TC4 generated similar levels of Sph2 as the wild-type strain (Fig 8C, lanes 5 and 6 vs. 1 and 2). The complementing transposon in TC4 was inserted in the promoter region of the hypothetical gene LEPIN2803 with sph2 oriented in the same direction (Fig 7C).
For the hemolysis assay, cultures were incubated for 24 hours to permit accumulation of Sph2. Hemolytic activity was higher in spent medium collected from cultures of the wild-type strain incubated at physiological osmolarity compared to incubation in EMJH alone (Fig 9A). However, hemolytic activity remained at background levels in spent medium collected from cultures of the sph2 mutant (Fig 9A), indicating that sph2 is required for hemolytic activity. For complementation analysis, three transconjugants were selected for their high levels of Sph2 production at physiologic osmolarity: TC4, TC8, and TC14 (Fig 8B). The complementing transposon in TC8 was inserted in a gene encoding a zinc-dependent hydrolase (LA3205), with sph2 oriented opposite of the disrupted gene (Fig 7C). The three complemented strains produced significantly more hemolytic activity than the sph2 mutant (Fig 9A). However, the hemolytic activity secreted by the complemented mutants was less than that generated by the wild-type strain.
The sphingomyelinase activity present in the samples was also quantified. The sph2 mutation resulted in a substantial reduction of sphingomyelinase activity, although activity above background levels was detected following incubation at physiologic osmolarity (Fig 9B). Similar to what we observed with the hemolysis assay, sphingomyelinase activity was partially restored with the three complemented strains (Fig 9B).
In this study we provide the first experimental evidence that demonstrates that Sph2 contributes to the hemolytic and sphingomyelinase activities secreted by L. interrogans. The amounts of transcript and protein generated from sph2 correlated with the levels of hemolysis and sphingomyelinase activities observed in spent culture medium when examining the Manilae L495 and Pomona LC82-25 strains incubated in standard culture medium and in medium supplemented with additional sodium chloride. More importantly, hemolysis associated with the spent growth medium from an sph2 mutant failed to rise above background levels. Additionally, the sph2 mutation eliminated most of the sphingomyelinase activity. These results suggest that Sph2 is the dominant hemolysin secreted by L. interrogans during cultivation in EMJH adjusted to physiologic osmolarity with sodium chloride.
Complementation of the mutant with the sph2 gene partially restored hemolytic and sphingomyelinase activities in three transconjugants producing similar or higher levels of Sph2 as the wild-type strain. These results confirm that sph2 contributes to the secreted hemolytic and sphingomyelinase activities, most likely by encoding a single protein that possesses both activities. However, wild-type levels of hemolytic and sphingomyelinase activities were not recovered in the complemented strains. It is unlikely that all three insertions of the complementing transposon disrupted Sph2 secretion or another process necessary for wild-type expression of the two activities. Partial complementation of the hemolytic activity may be explained in part by the mild reduction of sph1 transcript levels observed in the sph2 mutant. Another possibility for the incomplete dominance of the wild-type sph2 gene is that the truncated form of Sph2 produced by the mutant allele interferes with the function of the wild-type protein. We did not detect a truncated Sph2 protein in our immunoblots with the expected molecular mass of 46 kDa. Nevertheless, the same property of Sph2 that causes it to migrate slower than its calculated mass during electrophoresis could also cause the truncated form to run slowly, causing the protein to be obscured by the intense SphH band in the blot.
The sph2 mutation left behind a trace amount of sphingomyelinase activity above background levels. Our previous sequence analysis of the sphingomyelinase-like paralogs revealed that Sph1 and Sph3 lacked a few of the catalytic residues necessary for sphingomyelin hydrolysis activity [22]. Nevertheless, Sph1 and Sph3 may retain a limited capacity to hydrolyze sphingomyelin that is detected only when the dominant sphingomyelinase, Sph2, is removed genetically or when large amounts of recombinant Sph1 and Sph3 are analyzed [21].
Additional hemolysins may also contribute to hemolytic activity secreted by L. interrogans, depending on growth conditions. The hemolysins Sph1, Sph2, Sph3, HlpA, and TlyA were detected extracellularly in co-cultures with macrophage-like cell lines in tissue culture medium supplemented with 10% fetal calf serum [16]. Despite the secretion of several hemolysins, our findings predict that Sph2 would be a major hemolysin secreted by L. interrogans during colonization of the rat host since rat serum increased Sph2 production beyond what was observed without serum at physiologic osmolarity. This finding indicates that undiscovered factors in rat serum are responsible for enhancing sph2 expression.
As demonstrated in another study [32], we found that a Pomona strain incubated in standard culture medium produced large amounts of Sph2, in contrast to the negligible levels produced by the other four strains tested (Figs 1 and 2 and 6A). For the Pomona and Manilae strains, these results correlated with the secreted hemolytic activity (Fig 4B). The large amount of Sph2 and hemolytic activity observed in vitro with a Pomona strain in our study may be implicated in the hemolytic anemia and hemoglobinuria observed in ruminants infected with Pomona strains [12–14]. High basal sph2 expression may be a consequence of a insertion sequence-like element present near the promoter region of sph2 or the extra 75 nucleotides located within the 5' end of the coding region in the Pomona strain [18, 22, 48]. The latter sequences, which correspond to one of four tandem 25 amino acid residue repeats lying amino-terminal to the enzymatic domain, are likely to account for the higher molecular masses of the cellular and released forms of Pomona Sph2 relative to that of the other strains, which possess only three repeats. Moreover, the difference in the apparent molecular mass of Sph2 from the Pomona strain versus the other L. interrogans strains examined in this study is much greater than the calculated difference of 3 kDa. This observation indicates that the unexpectedly slow migration of Sph2 in acrylamide gels observed in this and in our earlier study is related to the repeats, which cause an increase of the apparent molecular mass of Sph2 relative to its calculated mass of up to 18 kDa [34]. The repeats are rich in proline, which could place conformational constraints on the protein to retard its migration through the acrylamide matrix.
Hemolytic activity was detected easily when spent medium from cultures containing rat serum was spotted onto agar plates containing sheep erythrocytes. However, despite the large amount of Sph2 released by the Pomona strain, hemolytic activity measured with the liquid phase assay was reduced considerably when rat serum was present. The inhibitory effect of rabbit and bovine serum on leptospiral hemolytic activity has been noted by others [8, 32, 49]. The ability to detect hemolytic activity in the presence of rat serum on erythrocyte agar plates suggests that the inhibitor is labile or highly diffusible in agar. The hemolysin inhibitor in bovine serum was shown to be phosphatidylethanolamine, phosphotidylcholine, and sphingomyelin [50]. In addition, because rats are a reservoir host for Leptospira, rat serum may contain anti-Sph2 antibodies that inhibit the function of Sph2. Our results raise the question of whether Sph2 can function as a hemolysin during hematogenous dissemination. It is possible that uncharacterized host or bacterial processes minimize the inhibitory action of phospholipids during leptospiral infection. Alternatively, the critical function of Sph2 during infection may not involve hemolysis. For example, Sph2 may function as an adhesin or trigger host signaling pathways via its sphingomyelinase when expressed by leptospires that colonize the renal tubule, where plasma proteins that carry phospholipids are likely to be excluded by the glomerulus [22, 31].
Several studies have demonstrated that most of the hemolytic activity detected in Leptospira cultures is present in the culture supernatant fluid [8, 9, 11, 32]. Similarly, almost all of the sphingomyelinase activity in another strain of Pomona is extracellular [51]. Although we did not measure the hemolytic and sphingomyelinase activities that remained associated with L. interrogans cells, the earlier studies suggest that the full-length Sph2 protein that is associated with the bacterial cell remains inactive until secretion, despite the full-length recombinant protein being active in vitro [16, 20]. These observations suggest that Sph2 and other hemolysins such as the pore-former SphH [17] are inaccessible to the surface, tied up in the secretion apparatus, or bound by an inhibitor until released from the cell, perhaps to prevent injury to the bacterium. Our findings presented in Fig 3 suggest that processing of Sph2 occurs during its release from the cell and that at least one of the two truncated forms of Sph2 detected extracellularly in L. interrogans cultures are functional rather than degradation products. However, we cannot rule out the possibility that the true active secreted product is the full-length species that degrades to inactive truncated forms during the lengthy immunoprecipitation procedure. Definitive proof will require demonstration of hemolytic and sphingomyelinase activities of purified forms of the truncated Sph2 species.
The pathway leading to secretion and processing of Sph2 remains to be identified. The processing sites leading to the released forms of Sph2 are predicted to lie within its N- or C-terminal domains, outside of the enzymatic domain. Sph2 appears to lack a classic amino-terminal signal peptide. Given the presence of genes encoding TolC, HlyB, and HlyD orthologs in the L. interrogans genome [1], secretion of Sph2 and the other sphingomyelinase-like proteins may involve a type I secretion pathway. In one study, a TolC homolog was co-immunoprecipitated using anti-Sph3 immunoglobulin, providing some experimental support for a TolC-based secretion system for the leptospiral sphingomyelinase-like proteins [29].
In conclusion, we report that sph2 expression and Sph2 secretion are regulated by salt and rat serum. Salt also regulated the levels of hemolytic and sphingomyelinase activities, potentially pathogenic functions of L. interrogans. These findings indicate that the quantities of hemolytic and sphingomyelinase activities measured in standard cultures of L. interrogans may underestimate the levels produced during infection, even for strains that produce large amounts in culture. Partial restoration of hemolytic and sphingomyelinase activities of an sph2 mutant by introduction of an intact sph2 gene strongly suggests that Sph2 is secreted as an active hemolysin and sphingomyelinase. Given the profound response of sph2 to host tissue-like conditions and the potential importance of Sph2 in leptospiral pathogenesis, further studies are needed to better understand the molecular mechanisms of sph2 gene regulation and secretion of its product.
|
10.1371/journal.pcbi.1005090 | When Does Model-Based Control Pay Off? | Many accounts of decision making and reinforcement learning posit the existence of two distinct systems that control choice: a fast, automatic system and a slow, deliberative system. Recent research formalizes this distinction by mapping these systems to “model-free” and “model-based” strategies in reinforcement learning. Model-free strategies are computationally cheap, but sometimes inaccurate, because action values can be accessed by inspecting a look-up table constructed through trial-and-error. In contrast, model-based strategies compute action values through planning in a causal model of the environment, which is more accurate but also more cognitively demanding. It is assumed that this trade-off between accuracy and computational demand plays an important role in the arbitration between the two strategies, but we show that the hallmark task for dissociating model-free and model-based strategies, as well as several related variants, do not embody such a trade-off. We describe five factors that reduce the effectiveness of the model-based strategy on these tasks by reducing its accuracy in estimating reward outcomes and decreasing the importance of its choices. Based on these observations, we describe a version of the task that formally and empirically obtains an accuracy-demand trade-off between model-free and model-based strategies. Moreover, we show that human participants spontaneously increase their reliance on model-based control on this task, compared to the original paradigm. Our novel task and our computational analyses may prove important in subsequent empirical investigations of how humans balance accuracy and demand.
| When you make a choice about what groceries to get for dinner, you can rely on two different strategies. You can make your choice by relying on habit, simply buying the items you need to make a meal that is second nature to you. However, you can also plan your actions in a more deliberative way, realizing that the friend who will join you is a vegetarian, and therefore you should not make the burgers that have become a staple in your cooking. These two strategies differ in how computationally demanding and accurate they are. While the habitual strategy is less computationally demanding (costs less effort and time), the deliberative strategy is more accurate. Scientists have been able to study the distinction between these strategies using a task that allows them to measure how much people rely on habit and planning strategies. Interestingly, we have discovered that in this task, the deliberative strategy does not increase performance accuracy, and hence does not induce a trade-off between accuracy and demand. We describe why this happens, and improve the task so that it embodies an accuracy-demand trade-off, providing evidence for theories of cost-based arbitration between cognitive strategies.
| Theoretical accounts of decision making emphasize a distinction between two systems competing for control of behavior [1–6]: one that is fast and automatic, and one that is slow and deliberative. These systems occupy different points along a trade-off between accuracy and computational demand (henceforth demand), making each one suitable for particular task demands. This raises the problem of arbitration: how does the brain adaptively determine which system to use at any given time? Answering this question depends on models and experimental tasks that embody the accuracy-demand trade-off at the heart of dual-system models.
Recent research formalizes the dual-system architecture in the framework of reinforcement learning [7, 8], a computational approach to value-guided decision-making that we describe in further detail below. The application of reinforcement learning methods to dual-process models of decision-making sparked an explosion of empirical and theoretical developments over the past decade because it offers a computationally precise characterization of the distinction between “automatic” and “controlled” processes for the task of value guided decision-making. Current research assumes that experimental methods grounded in reinforcement learning also capture a trade-off between accuracy (the proportion of value-maximizing actions) and computational demand (the minimization of computational effort and related costs), but this assumption remains largely untested (cf. [9]).
Currently, the dominant method that aims to dissociate mechanisms of behavioral control within the reinforcement learning framework is the “two-step task” introduced by Daw, Gershman, Seymour, Dayan, and Dolan [8] (Fig 1A), which we describe in detail in the next section. This task has proven to be a useful and popular tool to characterize the neural [8, 10–18], behavioral [19–31] and clinical [32–35] implications of dual-process models within the reinforcement learning framework. However, in this paper we argue that the two-step task does not induce a trade-off between accuracy and demand: Our simulations show that the “deliberative” strategy does not increase performance accuracy on the task. These simulations mirror a recent report by Akam, Costa, and Dayan [9], who also show that the two-step task does not embody a trade-off between model-based control and reward. Here, we expand on that result by showing that it holds across an exhaustive range of reinforcement learning parameters. Furthermore, we show that the same shortcoming is present in other, more recent variants of the task that have been reported. We then identify five factors that collectively restrict the accuracy benefits posited to arise from model-based control. Finally, we describe a novel task that induces the accuracy-demand trade-off, while retaining the ability to dissociate formally between distinct processes of behavioral control.
The fundamental problem in reinforcement learning is estimation of state-action values (cumulative future reward), which an agent then uses to choose actions. In the dual-system theory, the fast and automatic system corresponds to a “model-free” reinforcement learning strategy, which estimates state-action values from trial-and-error learning [36, 37]. In essence, this strategy is an elaborated version of Thorndike’s Law of Effect: actions that previously led to reward are more likely to be taken in the future. The strategy is “model-free” because it has no representation of the environment’s causal structure (i.e., the transition function between states and the reward function in each state). Instead, it incrementally constructs a look-up table or function approximation from which values can be quickly computed. However, this strategy can lead to errors if the environment changes, because the entire value function must be incrementally updated to accommodate changes. In addition, the strategy can produce sub-optimal credit assignment [8], a property we explore below. These forms of brittleness illustrate how model-free learning gives rise to “habits”—fast but inflexible response tendencies stamped in by repetition.
The slow and deliberative system corresponds to a “model-based” learning strategy that possesses operating characteristics complementary to the model-free strategy. This strategy learns an explicit causal model of the environment, which it uses to construct plans (e.g., by dynamic programming or tree search). In contrast to the habitual nature of the model-free strategy, the capacity to plan enables the model-based strategy to flexibly pursue goals. While more computationally expensive (hence slower and more effortful) than the model-free approach, it has the potential to be more accurate, because changes in the environment can be immediately incorporated into the model. The availability of a causal model also allows the model-based strategy to solve the credit-assignment problem optimally.
This dual-system framework sketched above can account for important findings in the reinforcement learning literature, such as insensitivity to outcome devaluation following overtraining of an action-reward contingency [7, 38]. Furthermore, the framework has spurred a wealth of new research on the neural [8, 10–13, 39, 40] and behavioral implications of competition and cooperation between reinforcement learning strategies [19–25, 29, 41, 42].
How might the brain arbitrate between model-free and model-based strategies? Since the model-based strategy attains more accurate performance through effortful computation, people can (up to a point) increase reward by engaging this system. However, in time-critical decision making settings, the model-based strategy may be too slow to be useful. Furthermore, if cognitive effort enters into the reward function [43–45], then it may be rational to prefer the model-free strategy in situations where the additional cognitive effort of model-based planning does not appreciably increase reward. It has been hypothesized that this trade-off between accuracy and demand plays a pivotal role in the arbitration between the two strategies [38, 46–50], but so far direct evidence for arbitration has been sparse [51].
Here, we will first describe in detail the design of the Daw two-step task, and the reinforcement-learning model of this task [8]. Next, we will show through computational simulations that model-based planning on this task does not yield increased performance accuracy. Finally, we will discuss several factors that contribute to this shortcoming in the current approach in this two-step task, and related paradigms.
Since its conception, the design of the Daw two-step task has been used in many similar sequential decision making tasks. Given the surprising absence of the accuracy-demand trade-off in the original task, it is important to investigate whether related versions of this paradigm are subject to the same shortcoming.
Despite the substantial differences between these variants of the two-step task, we found that none of them encompasses a motivational trade-off between planning and reward. This observation naturally raises a question: Why does planning not produce an increased reward rate in this task? What characteristics of the paradigm distort the accuracy-demand trade-off?
We investigate five potential explanations. These are not mutually exclusive; rather, they may have a cumulative effect, and they may also interact with each other. First, we show that the sets of drifting reward probabilities that are most often employed are marked by relatively low distinguishability. Second, we show that the rate of change in this paradigm is slow and does not require fast online (model-based) flexibility. Third, we show that the rare transitions in the Daw two-step task diminish the reward-maximizing effect of a model-based choice. Fourth, we show that the presence of the choice at the second stage decreases the importance of the choice at the first stage, which is the only phase where the model-based system has an influence. Fifth, we show that the stochastic reward observations in this task do not carry enough information about the value of the associated stimuli. We use simulations of performance on novel tasks to demonstrate these five points and, as a result, develop a novel paradigm that embodies an accuracy-demand trade-off.
For the participants who completed the Daw task, we found that a reward on the previous trial increased the probability of staying with the previous trial’s choice [t(196) = 7.70, p < 0.001; Fig 16A], but that this effect interacted with the type of transition on the previous trial [t(196) = 5.38, p < 0.001]. This result replicates the basic finding on the original two-step confirming that participants used both model-based and model-free strategies.
For the participants who completed the new paradigm, we found that a positive reward on the previous trial significantly enhanced staying behavior from chance for both similar and different current start states, (p < 0.001 for both effects), but this effect was larger for the same compared to the different start state condition [t(183) = 9.64, p < 0.001; Fig 16B]. This pattern of behavior suggests that the participants did not employ a pure model-based strategy (compare with Fig 15B). However, as described above, it does not allow us to assess the relative contributions of model-based and model-free strategies to control based on these raw stay probabilities: both a purely model-free agent and an agent with a mixture of model-based and model-free strategies choices are predicted to show an increased stay probability after a win in a different start state, since a reward is indicative of history of recently reward trials.
The reinforcement learning models described above incorporates the (decayed) experience on all previous trials to choice and is better able to dissociate the contributions of the two strategies. This model consists of a model-free system that updates action values using temporal-difference learning and model-based system that learns the transition model of the task and uses this to compute action values online. The weighting parameter w determines the relative contribution between model-based and model-free control. The stickiness parameters π and ρ capture perseveration on either the response-level or the stimulus-choice.
We first investigated whether the inclusion of either stickiness parameter (π and ρ) was justified by comparing both the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC), for models that included none, one, or both parameters for both tasks separately (see S1 Table). For the Daw task, we found that both goodness-of-fit measures favored a model that included both stickiness parameters. For the novel task, the BIC favored a model with response stickiness but not stimulus stickiness included, whereas the AIC favored a model that included both stickiness parameters. We decided to favor the more parsimonious model without stimulus stickiness, and parameter fits from this model will be reported in the following, but the results did not qualitatively change when the stimulus stickiness parameter was included.
Second, we used model comparison with both goodness-of-fit measures to analyze whether the hybrid model including the w parameter fit the data better than either a pure model-based or model-free model (see S2 Table). For the Daw task, we found that the AIC favored the hybrid model, but that the BIC favored the pure model-free model. However, there have been many reports in the literature that justify the inclusion of the weighting parameter for this task [8], and so we adopt the hybrid model for consistency with prior work. (Note also that it would be impossible to assess the relationship between model-based control and reward without using the hybrid model). For the new task, we found that both BIC and AIC favored the hybrid model compared to the pure model-based and model-free models. This suggests that human performance in the new paradigm is characterized by a mixture of model-based and model-free strategies.
In summary, the model fits presented below used all six free parameters for the participants that completed the Daw paradigm, but omitted the stimulus stickiness parameters for the participants that completed the novel paradigm. These parameter estimates and their quartiles are depicted in Table 1.
Across participants, we found that the median weighting parameter w was 0.27 for the Daw paradigm and 0.48 for the novel paradigm, indicating that both strategies were mixed in the population for both tasks. However, we found that model-based control was significantly higher for participants in the novel paradigm compared to the Daw paradigm [Wilcoxon two-sample rank sum test, z = 3.31, p < 0.001], suggesting that the existence of the accuracy-demand trade-off in the novel paradigm induced a shift towards model-based control.
Of greatest relevance to our present aims, we found that the weighting parameter was positively related to our measure of the reward rate that controlled for average chance performance for the novel task (r = 0.55, p < 0.001), but not for the Daw paradigm (r = 0.10, p = 0.15; Fig 17). A subsequent multiple regression showed that this relationship was significantly different between groups [t(377) = 4.71, p < 0.001].
Next, in order to quantify the average gain in points across the entire range of w for both tasks, we ran a set of linear regression analyses predicting the reward rate from the weighting parameters for both tasks. For the Daw task, we found a predicted reward rate of 0.52 for w = 0 (i.e., the intercept), and an increase of 0.002 on top of this for w = 1 (i.e., the slope), indicating a 0.42% increase in points (to 0.525). For the novel task, we found a predicted reward rate of 0.58 for w = 0, and an increase of 0.67 on top of this for reward rate w = 1, indicating a 215% increase in points (to 1.25). When we computed these slopes using the corrected reward rates, subtracting the average value of each participant’s reward distribution from their reward rate, we found an average increase in reward rate of 0.01 across the range of the weighting parameter for the Daw task, and an average increase in reward rate of 0.51 for the novel task.
These results validate the accuracy-demand trade-off of the novel two-step paradigm, and also demonstrate that the original Daw two-step paradigm does not embody such a trade-off.
The distinction between planning and habit lies at the core of behavioral and neuroscientific research, and plays a central role in contemporary dual process models of cognition and decision making. Modern reinforcement learning theories formalize the distinction in terms of model-based and model-free control, bringing new computational precision to the long-recognized trade-off between accuracy and demand in decision making. In principle, the model-based strategy attains more accurate performance through increased effort relative to the computationally inexpensive but more inaccurate model-free strategy.
Yet, building on prior work [9], we provide an exhaustive demonstration that the hallmark task for dissociating model-based and model-free control—the Daw two-step paradigm (Fig 3A) and several related variants of this task (Figs 4 and 5)—do not embody a trade-off between accuracy and demand across a wide range of parameter space. Using simulations of reinforcement learning agents on variants of the two-step task, we have identified five features that reduce the reward associated with model-based control to such a degree that pure model-based and model-free agents obtain equivalent reward (Fig 13). By systematically eliminating these features from the task, we developed a novel variant that shows a strong relationship between model-based control and reward both in simulations and in experimental data. In addition to providing new insight into the affordances of distinct experimental paradigms, our findings demonstrate that the magnitude of the accuracy-demand trade-off varies greatly with the specific features of any given task.
First, we found that the trade-off depends on highly distinguishable reward probabilities. Broadening the range of possible reward probabilities (from 0 to 1) contributed a small, but measurable effect on the relationship between model-based control and reward (Simulation 1, Fig 6B). Second, we found that the trade-off depends on the rate of change of the second-stage reward probabilities. Our analyses indicates that the rate of change in the original paradigm was too slow to elicit a reliable accuracy-demand trade-off, because it allowed the model-free strategy to integrate sufficient information over trials to match the performance of the model-based strategy (Fig 7A). Based on this analysis, we showed that a task with larger drift rate produced a stronger relationship between model-based control and reward (Simulation 2; Fig 7B). Third, the trade-off can be limited by the presence of stochastic transitions. In the original two-step task, model-based choices do not always lead to the desired second-stage state, since this paradigm includes rare transitions from the first to the second stage, reducing the efficacy of model-based control. A new transition structure, using deterministic transitions from two different starting states, avoids this issue, and substantially strengthens the accuracy-demand trade-off (Simulation 3; Fig 8A). Fourth, the trade-off is limited when the environment contains a large number of actions bounded by the same rewarded probabilities. Specifically, by reducing the number of second-stage choice options, the average difference in value between the optimal choices of the two second-stage states is increased, which allows the model-based advantage at the first stage to emerge more distinctly. This change to the paradigm further strengthens the accuracy-demand trade-off (Simulation 4; Fig 9). Fifth, the trade-off is limited under conditions of high uncertainty about the reward value of actions. Specifically, we found that the stochastic reward observations in this task do not carry enough information about the value of the associated stimuli. Subsequently, removing the binomial noise from the reward distributions leads to a substantial increase in the strength of the accuracy-demand trade-off in this paradigm (Simulation 5; Fig 11). Moreover, we find that these factors have a superadditive effect on the relationship between model-based control and reward: All five changes to the task are required to establish a reliable accuracy-demand trade-off. We experimentally confirmed these theoretical predictions, demonstrating that the empirical estimate of model-based control in the new task was correlated with reward rate across participants.
It is likely that more than these five factors alone moderate the effect of model-based control on accuracy. For example, in the Akam version of the two-step task, rewards alternate between blocks of opposite reward probabilities, so that one option strictly dominates the other until the next alternation is implemented. As discussed, this change to the paradigm resulted in a strong trade-off between control and reward in a selective region of parameter space. It is plausible that there are alternative versions of the two-step task that embody an even stronger trade-off than those discussed here, and we look forward to a comparison of how those relate to the current paradigm.
In addition to the difference in the strength of the accuracy-demand trade-off between paradigms, we also found that novel two-step task elicited greater average model-based control in our participants than the original Daw two-step task. This result is one of the first pieces of behavioral evidence suggesting an adaptive trade-off between model-based and model-free control. Put simply, participants reliably shifted towards model-based control when this was a more rewarding strategy. This may indicate that participants store “controller values” summarizing the rewards associated with model-based and model-free control. However, there are alternative explanations for this result. For example, it is possible the presence of deterministic transition structure or the introduction of negative reward induced increased model-based control triggered by a Pavlovian response to these types of task features. In other words, the increase in planning might not a reflect motivational trade-off, but rather a simple decision heuristic that does not integrate computational demand and accuracy. Future investigations, where task features and reward are independently manipulated, will be able to provide more conclusive evidence that people adaptively weigh the costs and benefits of the two strategies against each other.
Although the original Daw two-step task does not embody an accuracy-demand trade-off, choice behavior on this task nonetheless reflects a mixture of model-based and model-free strategies. Furthermore, the degree of model-free control on this task is predicted by individual difference measures such as working memory capacity [23], cognitive control ability [24], processing speed [29], age [20, 31], extraversion [30], and even psychiatric pathology [11, 33, 34]. This discrepancy demands explanation. Why does the original task, without a motivational trade-off, still yield meaningful and interpretable results? One possibility is that, in the absence of a reliable signal from the environment, behavior on this task reflects participants’ belief about how model-based control relates to reward maximization in the real world (where the trade-off is presumably more pervasive). Another possibility is that the extensive training of participants on the transition structure of the experiment induces them to assume they should be using it during task performance. In this sense, the absence of a trade-off is not problematic for mapping out individual differences that co-vary with the use of model-based control.
This analysis can help explain the types of experimentally induced shifts in control allocation that have been reported using the two-step task, as well as those that have not. Prior research has demonstrated several factors that increase the control of model-free strategies on decision making. Control shifts to the model-free system with extensive experience [57], under cognitive load [22], and after the induction of stress [23, 28]. Such shifts are rational insofar as there is no advantage to model-based control in the task. Notably, however, few studies report factors that increase the use of model-based control. The exception to this rule is a study in which the underlying neural mechanism was altered by administering dopamine agonists after which control shifted to the model-based system [12]. Apart from this report, no other studies have successfully increased model-based control in the two-step task. Our simulation results suggest an explanation: in the original version of the two-step task, planning behavior does not improve reward, and so there is no incentive to increase the contribution of the model-based system.
Our novel paradigm opens up the possibility of studying the neural mechanism underlying the trade-off between model-based and model-free control. The first and most influential neuroimaging study of the two-step task [8] focused on the neural correlates of “reward prediction error” (the difference between expected and observed reward) that is used by both the model-based and model-free controllers. A host of previous research shows that model-free reward prediction errors are encoded in the striatum [36]. The results of Daw and colleagues [8] were in line with this finding; the reward prediction errors of the model-free system correlated with signal in the striatum. However, despite the distinct computational features of the two systems, the model-based reward prediction errors recruited a similar, indistinguishable, region of the striatum (see also [13]). Our recent simulations may shed light on this surprising finding, insofar as model-based system was not appropriately incentivized. An important area for future research is to identify the neural correlates of model-based control under conditions where it obtains a higher average rate of reward than does model-free control.
One potential limitation of the current paradigm is that it does not afford a simple qualitative characterization of model-based versus model-free control based exclusively on the relationship between reward (vs. punishment) on one trial and a consistent (vs. inconsistent) behavioral policy on the subsequent trial. As depicted in Fig 15B, both strategies predict an increased likelihood of behavioral consistency after a reward in either start state, but also a higher probability of consistency when the current start state is the same as in the previous trial compared to when the current start state is different. Our results reinforce this point. Even though the raw consistency behavior was not able to distinguish between the pure model-free and mixture strategies, our model-fitting procedure showed that most participants employed both model-based and model-free strategies.
Indeed, our exploration of this point revealed an apparent mystery and suggests a potentially illuminating explanation. Although our full model fits of participant data indicate a high degree of model-based control, this trend is not at all evident in their raw stay probabilities, conditioned on reward in the previous trial. Not only do we fail to find the high staying probability we would expect for trials on which the associated stage-one choice was previously rewarded (assuming some influence of model-based control), in fact we find an even lower stay probability than would be expected given a computational model of pure model-free control. How can we explain this divergence between our empirical result and the predictions of our generative model? Recent work on the influence of working memory capacity on reinforcement learning may shed some light on this puzzling finding. Collins and Frank [58] show that the performance accuracy on a reinforcement learning task varied as a function of the number of stimuli that had to be remembered (the load) and the delay between repetitions of the same choice. Behavior in the current task is likely to be subject to similar constraints, since the number of choice options (six) is well above the capacity limit reported by Collins and Frank [58]. Therefore, the smaller-than-predicted probability of staying after a reward in the different start state might be predicted be memory decay, since the average delay of seeing the stimuli in this state is strictly higher than in trials with the same starting state. Exploring these possibilities further, while beyond the scope of the present study, is a key area for further investigation.
Finally, we observed a shift in arbitration between model-based and model-free control when comparing the original and novel versions of the two-step paradigm. Specifically, participants in the novel paradigm were more likely to adopt the model-based strategy compared to those who completed the Daw version of the task. This result is one of the first pieces of evidence that the people negotiate an accuracy-demand trade-off between model-based and model-free strategies, and is consistent with a large body of literature that suggests that increased incentives prime more intense controlled processing [56]. Though tantalizing, this result raises several new questions. For example, how does the brain adapt its allocation between model-based and model-free control? At what time scale is this possible? What is the appropriate computational account of arbitration between the two systems? What neural regions are involved in determining whether one should exert more model-based control? Future investigations, using a combination of neural, behavioral, and computational methods will aim at answering these questions.
In recent years, the Daw two-step task has become the gold standard for describing the trade-off between accuracy (model-based control) and computational demand (model-free control) in sequential decision making. Our computational simulations of this task reveal that it does not embody such a trade-off. We have developed a novel version of this task that theoretically and empirically obtains a relationship between model-based control and reward (a proxy for the accuracy-demand trade-off). The current investigation reveals a critical role for computational simulation of predicted effects, even if these appear to be intuitive and straightforward. It also introduces a new experimental tool for behavioral and neural investigations of cost-benefit trade-offs in reinforcement learning. Finally, it opens new avenues for investigating the features of specific tasks, or domains of task, that favor model-based over model-free control.
|
10.1371/journal.pgen.1003905 | Robust Demographic Inference from Genomic and SNP Data | We introduce a flexible and robust simulation-based framework to infer demographic parameters from the site frequency spectrum (SFS) computed on large genomic datasets. We show that our composite-likelihood approach allows one to study evolutionary models of arbitrary complexity, which cannot be tackled by other current likelihood-based methods. For simple scenarios, our approach compares favorably in terms of accuracy and speed with , the current reference in the field, while showing better convergence properties for complex models. We first apply our methodology to non-coding genomic SNP data from four human populations. To infer their demographic history, we compare neutral evolutionary models of increasing complexity, including unsampled populations. We further show the versatility of our framework by extending it to the inference of demographic parameters from SNP chips with known ascertainment, such as that recently released by Affymetrix to study human origins. Whereas previous ways of handling ascertained SNPs were either restricted to a single population or only allowed the inference of divergence time between a pair of populations, our framework can correctly infer parameters of more complex models including the divergence of several populations, bottlenecks and migration. We apply this approach to the reconstruction of African demography using two distinct ascertained human SNP panels studied under two evolutionary models. The two SNP panels lead to globally very similar estimates and confidence intervals, and suggest an ancient divergence (>110 Ky) between Yoruba and San populations. Our methodology appears well suited to the study of complex scenarios from large genomic data sets.
| We present a new likelihood-based method to infer the past demography of a set of populations from large genomic datasets. Our method can be applied to arbitrarily complex models as the likelihood is estimated by coalescent simulations. Under simple scenarios, our method behaves similarly to a widely used diffusion-based method while showing better convergence properties. In addition, our approach can be applied to very complex models including as many as a dozen populations, and still retrieve parameters very accurately in a reasonable time. We apply our approach to estimate the past demography of four human populations for which non-coding whole genome diversity is available, estimating the degree of European admixture of a southwest African American population and that of a Kenyan population with an unsampled East African population. We also show the versatility of our framework by inferring the demographic history of African populations from SNP chip data with known ascertainment bias, and find a very old divergence time (>110 Ky) between Yorubas from Western Africa and Sans from Southern Africa.
| Reconstructing the past history of a given species is important not only for its own sake, but for disentangling demographic from selective effects [1], [2]. Demography is indeed often estimated on a set of markers and the best neutral model is used as a null for evidencing markers under selection [3], [4] or for finding global patterns of selection across the genome [e.g. 5]. Various methods have been proposed to estimate demography from genetic data, including full-likelihood methods [6]–[9], summary-statistics likelihood based methods [10], [11], or different flavours of Approximate Bayesian Computation [12]–[16]. With some exceptions, these methods are relatively slow and do not scale up very well with new genomic data, as computation time increases with the number of loci. In contrast, recently developed composite-likelihood methods based on the site frequency spectrum [SFS, 17] have computing times that do not depend on the amount of available genomic data [18]–[21], and several approaches have been proposed to estimate demographic parameters from the SFS [e.g. 11], [17], [20], [21]–[24]. Among these latter methods, the most widely used is [21], which estimates the expected joint site frequency spectrum for an arbitrary set of parameters by a diffusion approach. Whereas the estimation of the expected SFS is relatively fast, the optimization of the parameters is still time-consuming, which prevents to tackle models with more than three populations at the same time. While some methods can extract demographic information from single whole-genomes per population [25], [26], SFS-based methods, when applied to multiple individuals, do not require whole genome data because correct estimates of the SFS can be obtained from a few Mb [21]. However, with few exceptions [11], the accuracy of SFS-based methods has not been properly assessed, and their ability to infer demographic parameters has been questioned [27].
One advantage of SFS-based inference methods is that they can handle large next generation sequencing (NGS) data sets [28]–[30]. However, the computation of the SFS from NGS data is not always trivial. An empirical Bayes approach has been proposed to estimate the joint 2D SFS from low coverage data [31] and an unbiased maximum likelihood approach has been developed to recover the SFS for a single population [32]. SFS obtained from low-coverage genomic data often show a deficit of rare alleles because a given allele needs to be observed in several individuals to exclude read errors [28], [33]. These missing low frequency variants can lead to imprecisions and biases in population genetic inferences [34]. Several approaches have been proposed to correct for this bias [32], [35], either during the process of genotype calling itself [e.g. 31], [36], [37] or later by applying quality filters on called genotypes [e.g. 38]. Gravel et al. [28] have also proposed to predict the SFS from low-coverage data by using an overlapping subset of high quality data to derive a generalized correction of the SFS. It appears likely that SFS estimation will improve with higher coverage NGS data, and that such data will become increasingly available and used in the near future.
As an alternative to deep sequencing, one could use information from a few tens of thousands SNP scattered over the whole genome to make demographic inference, but most SNP chips have complex and often unknown ascertainment schemes that bias the SFS if not properly taken into account [39]–[41]. However, a new SNP chip has recently been introduced [42], [43], which implements a known and simple ascertainment scheme where SNPs are selected at random from sites that are heterozygous in a single individual of a given population. Whereas this ascertainment scheme has no major effect on statistics designed to infer admixture [42], it biases the site frequency spectrum [44], [45] and thus potentially alters the estimation of other parameters. Using simple combinatorics, the SFS can be unbiased [44] in a single population, and this strategy could be extended to unbias joint SFS under complex models involving more populations. A diffusion approach has been recently proposed to estimate divergence times between two populations based on the fraction of SNPs having occurred recently in the ascertained population [45], but this approach is currently restricted to the sole estimation of divergence time and cannot be applied if gene flow occurred between populations.
In this paper, we introduce a flexible and robust way to estimate demographic parameters from the SFS inferred from sequence or SNP chip data that we implemented in the fastsimcoal2 software. Our method is based on Nielsen's approach [17], which estimates the expected SFS from simulations under any demographic model. We compare the performance of this approach to [21] under a variety of evolutionary models with simulated data, and we show that it can successfully handle models including more than three populations. We also show how this approach can be extended to deal with ascertained SNP panels by explicitly modelling the ascertainment bias and computing likelihoods based on expected ascertained SFSs. We first apply our method to a large human genomic data set from which we estimate the demography of four populations, and then to two separate Affymetrix ascertained SNP panels [43] from which we estimate the demography of two African populations.
We performed parameter estimations for 10 data sets generated under each of the 3 evolutionary scenarios shown in Figures 1A–1C. We took two approaches for estimating demography: our new approach based on a composite multinomial likelihood where the expected SFS is obtained using coalescent simulations and [21], which computes a composite Poisson likelihood where the expected SFS is obtained by a diffusion approximation. The two approaches have a very similar accuracy under a simple bottleneck scenario (Figure S4) and under a scenario of population isolation with migration [46] (IM model, Figure S5). For both approaches we report the estimates leading to the maximum likelihood obtained among 50 independent runs. Under these conditions, leads to extremely accurate estimations for most data sets. However, in a few cases (1/10 for the bottleneck scenario, and 2/10 for the IM model), the best likelihood obtained from 50 runs led to very divergent estimates, which were not reported in Figures S4, S5. For those cases, the log likelihood appeared orders of magnitude smaller than those inferred for other data sets and could be easily spotted. Although it is possible to recognize that additional runs are necessary to get meaningful estimates, we did not follow this procedure here, as we wanted to allocate similar resources to the two programs and get results using an automated procedure not requiring further user tweaks. Contrastingly, fastsimcoal2 estimations seem to converge to correct values for all data sets in Figure S4 and S5, even though the variances of the estimators are slightly larger than 's for those cases where both approaches agree on the correct demographic model.
Parameter estimations under the more complex scenario of Figure 1C, mimicking a simple model of human evolution, are reported in Figure 2. In this case, results obtained by fastsimcoal2 are again very accurate and close to the true values for all 10 data sets. With , we report results for only 8 data sets due to potential lack of convergence, as explained above. However, even for these 8 data sets, the best estimates can be quite far from the true parameters, especially for parameters related to the ancestral bottleneck. It suggests that for complex scenarios involving three populations and more than 5 parameters, needs to be run from many more than 50 initial conditions and that some iterative refinements of search ranges might be necessary to obtain correct solutions (R. Gutenkunst, personal communication). Note that a lack of robustness of under certain conditions (e.g. high migration rates between populations) had already been reported before [11], [24].
We have estimated parameters for the more complex hierarchical continent-island model shown in Figure 1D, involving samples from 10 different populations (islands), a model that cannot handle. Continent-island models are equivalent to infinite islands models, and have been used to model recent spatial expansions [see e.g. 47]. This model could therefore represent two successive spatial expansions, the first one stemming from an ancestral refuge area, and the second one starting more recently from a single deme belonging to the first expansion wave. The parameters of interest are here the immigrations rates in each sampled deme, the timing of the spatial expansions and the ancestral population size. As shown in Figure 3, all these parameters are extremely well estimated by fastsimcoal2 when we maximize the multiple pairwise composite-likelihood shown in eq. (7). We note that we can also recover very well the immigration rate to the unsampled deme (rightmost column in Fig. 3) from which the second expansion started. The accuracy of the immigration rate estimations is quite remarkable, given that they span over two orders of magnitude and that we specified the same search intervals covering four orders of magnitude for each parameter.
We first applied our methodology to the problem of estimating the past demography of two African, one European and one African-American populations. The multidimensional SFS for these 4 populations was estimated from more than 220,000 non-coding SNPs, each located more than 5 Kb away from its closest neighbour, such as to minimize linkage disequilibrium between SNPs. We examined three evolutionary scenarios shown in Figure 4 to explain observed patterns of diversity. In the first and simplest scenario (Figure 4A), the South Western African American population (ASW) was assumed to have been formed 16 generations ago (around 1600 AD) with initial input from one European (CEU) and two Niger-Congo speaking African populations (Yoruba from Nigeria: YRI; Luhya from Kenya: LWK) having diverged earlier. In order to calibrate the other parameters, we assumed that the European population diverged from the ancestral African population 50 Ky ago [28], [48]. Under this scenario, we find that the ASW population would have initially received 16% (CI95% = [15–17%]) of its gene pool from the CEU population, 83.8% from the YRI population and almost nothing (0.2%) from the LWK population (see Table 1, Model A). This European contribution is in line with previous estimates obtained from SNP-chip allele frequencies (17% for Southwest African Americans [49]). Under model A, the two Niger-Congo populations would have diverged very recently (70 generations ago, CI95% = [56–197]), and the CEU and YRI populations have the smallest effective population sizes (around 4000 individuals), whereas the ASW population has the largest (NASW = 170,000 individuals). The inferred human ancestral population size is relatively small (about 8000 individuals) and there is no real signal of an ancestral bottleneck since the estimated bottleneck size (NBOT = 7083) is only 12% smaller than the ancestral size, in line with recent results showing no evidence for a strong Pleistocene bottleneck in humans [50].
Whereas model A captures some obvious features of the past demography of these populations (see Table S1), it seems relatively unrealistic for some other features (i.e. a direct contribution of the CEU and YRI populations to ASW). We therefore investigated a more realistic but more complex and parameter-rich model involving several other unsampled populations, as shown in Figure 4B (see Material and Methods for a complete description of this model). The multiple continent-island model B1 assumes that the ASW population was founded by migrants originating from a Niger-Congo and from a European metapopulations, from which the two Niger-Congo and the CEU populations currently receive migrants. It also assumes that the Niger-Congo and the European metapopulations passed through a bottleneck when they diverged from an ancestral African population. An even more complex scenario B2 includes a potential admixture of the Luhya population (a Niger-Congo speaking population from Kenya) with an unsampled (potentially East-African) population, which also diverged earlier ago from the ancestral African population.
The model parameters estimates and their confidence intervals obtained by a parametric bootstrap approach are listed in Table 1. The two models show overall very congruent values and overlapping 95% confidence intervals for their common parameters. The agreement is especially good for the human ancestral size (NANC = 12–13,000 individuals), the ancestral African population size (NAFR = 25–27,000), the continental European size (NEUR = 14,500–16,500 individuals), the European strong bottleneck intensity (IBEUR = = 0.42–0.43, where is the bottleneck duration, and is the bottleneck size), the Niger-Congo milder bottleneck intensity (INC = 0.027–0.028), the divergence time of the Niger-Congo metapopulation (TNC = 793–797 generations), the time to the shift to the ancestral human population size (TBOT∼10,000 generations), and the European contribution to the ASW population (aE = 0.16–0.17). The other parameters show different point estimates but all have overlapping confidence intervals.
We have plotted the marginal SFS for each of the four populations in Figure S6, to visualize the fit of the expected and observed SFS for each model. Whereas the expected population specific marginal SFSs show some discrepancies with the observation for the four populations under model A, the fit is much better for model B1, except for LWK, which still shows an underestimation of singletons and doubletons. Model B2, which allows for LWK admixture, leads to a much better fit for the LWK population, as shown by the cumulative distribution of differences between the expected and observed marginal SFS (see 3rd row in Figure S6). Under this model B2, we estimate the LWK population to have 17% admixture from an unspecified but probably East African (see e.g. Figure 1 in ref. [51]) population. This East African population would have diverged from the ancestral African population more than 2200 generations ago (95% CI 1274–3586), thus potentially before the out-of-Africa dispersal. Even though the different models can be conveniently compared on the basis of their marginal SFSs, these 1D SFSs only capture a small fraction of the total (multidimensional) SFS. Therefore the models are better compared on the basis of their likelihood. This is formalized here by a model comparison procedure based on AIC [52], revealing that the relative likelihood of models A and B1 are almost 0 as compared to that of model B2 (see Table S2).
We estimated the parameters of African past demographies shown in Figure 5 based on Yoruba and San samples for which we have independent SNP panels (see Methods section). In model A (shown in Figure 5A), we assumed that the Yoruba and San samples were taken from large populations that expanded after their divergence, and we allowed for a single pulse of gene flow between them at a given time Ta in the past. The model B (shown in Figure 5B) includes the divergence of two-continent island metapopulations, and assume that the sampled populations are each an island attached to these continents and that the two continents exchanged migrants some time ago in a single pulse of gene flow, like in model A, but also earlier in time (see Figure 5B and material and methods for a complete description of the model).
The point estimates of the two models and their associated 95% confidence intervals (CI) inferred from 100 parametric bootstraps are reported in Table 2 for both SNP panels. Overall, the two SNP panels show congruent point estimators and CI widths under the two models. There is only one parameter (NAY) for which the CI do not overlap under model A, which suggests that the two panels provide broadly compatible scenarios of African demography. Estimations from data simulated under the same model for parameter values similar to those inferred in Figure 5A show (see Figure S8) that i) both panels should perform very similarly for estimating parameters, ii) all parameters of the model should be well estimated, except those related to a very recent expansion of one of the ascertained population, iii) ancestral population sizes and divergence times are particularly well estimated, and iv) the addition of a single Denisovan sequence allows one to recover the absolute values of the parameters.
Concentrating on the parameters common to both models, we see in Table 2 that the ancestral size NANC shows very similar estimates across models and panels, with an estimated value around 9,000–9,500 individuals (in line with estimates obtained with non-ascertained data set). The African population size is also consistently estimated to be around 18,000–28,000 individuals across models, and the ancestral Yoruban size appears smaller and between 5,500 and 13,000 individuals. These estimates fit well with previous Bayesian estimations of African demography from nuclear markers under slightly different models. Based on microsatellites, Wegmann et al. [13] estimated the ancestral size of Niger-Congo (NC) populations (to which Yoruba belong) to be 12,500 individuals and that of the ancestral African population to be 15,000 individuals. More recently, the analysis of 40 non-coding regions of 2 Kb [53] led to estimates of NC and African ancestral size to be 17,500 and 11,000 individuals, respectively, as well as a San effective size of the order of 20,000 individuals. The differences between these estimations and ours might be due to the fact that these previous analyses were based on slightly different models that assumed constant sizes for all current populations and the same population size before the split with Denisovans.
In addition, we find evidence for some asymmetrical gene flow between San and Yoruba, around 500–600 generations ago (12.5–15 Kya) under model A, and much more recently (60–80 generations ago) under model B. Interestingly, this is the only parameter common to the two models that shows such drastic difference. Despite this disparity, which could be due to the fact that we allow for earlier migration between the two metapopulations in model B, we obtain very similar estimates for the admixture rates between populations both between panels and across models. Overall, we find a slightly larger extent of gene flow from Yoruba to San than the reverse, but the confidence intervals of the two parameters seem quite overlapping under both models. Under model A, the point estimates for the divergence time TDIV are much more different than what was obtained under our simulations (Figure S8), with a much younger divergence suggested by the San panel (2,600 generations or 65 Kya) than for the Yoruba panel (4,700 generations or 117.5 Kya). Taking the middle of the overlap between the two CI would lead to a divergence time of 4,500 generations or 112.5 Kya (Table 2), in keeping with a recent estimate of the divergence of Khoisan populations obtained by an ABC approach [110 Ky, 53], and compatible with the divergence time estimated between San and other West African population (65–120 Ky in [54], or ∼100 Ky in [55]). Under model B, the two estimates obtained for panel 4 and 5, show a similar discrepancy, but the estimated values are much higher (5,530 and 10,330 generations for panels 4 and 5, respectively), which can also be due to the fact that we authorize some gene flow between the two metapopulations after their divergence. If we again take the middle of the overlap between the two CI, we obtain a value of 7,500 generations (180 Kya), substantially larger than the value obtained under model A (4,500 generations).
An examination of the parameters restricted to model B suggests that the Yoruban continent expanded recently 170–300 generations ago (4250–7500 ya), from a relatively small population of 600–3600 individuals, and that the Yoruban island receives more migrants (around 18 per generation) than the San island (2–3 individuals per generation). The age of the expansion is slightly older than the divergence time between two Western Niger-Congo populations estimated previously (140 generations, [13]), and intermediate between the age of the Niger-Congo languages (∼10 Kya, [56]), and that of the Bantu expansion (∼5 Kya, [57]). The larger immigration rate seen in Yorubans is compatible with the fact that farmer populations generally maintain higher levels of gene flow with their neighbours than hunter-gatherers due to their larger effective size [47]. Note however that all parameter estimates mentioned above assume that the Denisova divergence time is correctly estimated at 16,000 generations or 400 Kya [58], even though there is still a large uncertainty attached to this divergence time, which could range from 230 to 650 Kya [58] or even between 170 and 700 Kya in a more recent study [59]. Reported estimates and CI in Table 2 do not take this uncertainty into account, and should thus be rescaled if a different divergence time between Denisovans and Humans was proposed.
Like in the case of non-ascertained data, we find that the more complex model is much better supported by the data. Even though this better fit is barely visible when considering the marginal 1D expected SFS (see Figure S10), this is more exactly quantified by an AIC analysis (Table S3) revealing that the relative likelihood of model A is close to zero for both panels when compared to model B.
We have introduced a new and flexible simulation-based approach to estimating demographic parameters. For the tested scenarios, our composite-likelihood approach is as precise as [21], which is the current standard in the field. Our approach seems more robust than since it is more likely to converge towards the correct solution when starting from the same number (50) of initial conditions (see Figures 2, 3, S4, S5). In terms of computational speed, point estimates are very quickly obtained by for simple models (on average 15 seconds and 6 minutes for models in Fig. 1A and 1B, respectively, compared to 15 minutes and 2h30 for fastsimcoal2, respectively). However, fastsimcoal2 is much faster for more complex models with three populations and migration (4–5 h per run for fastsimcoal2 for model on Fig. 1C, compared to 34 h on average for ). By maximizing the fit of two-dimensional SFS, fastsimcoal2 can also explore very complex models involving more than 10 populations with migration, which cannot be tackled by any other current method. Since fastsimcoal2 and use a very similar likelihood function (see Figure S3), it seems that the improved convergence of our approach lies in the use of the ECM optimization scheme, which compensates for the use of non-optimal approximate likelihoods. Note that our robust ECM maximization technique and the maximization of the product of pairwise composite likelihoods could also be used by methods deriving the SFS analytically or by a diffusion approximation (like ), thus potentially enabling the analysis of models as complex as those studied here. Also note that recent progress in the computation of joint SFS using coalescent or diffusion approaches [18], [23] have led to the development of promising demographic inference methods applied to the study of relatively complex evolutionary models [see e.g. 24].
Even though different demographic trajectories can lead to exactly the same SFS in a single population [27], we do not find any evidence of parameter non-identifiability in our investigated cases. This is probably because we restricted our search to a limited set of possible histories, defined by few-parameter models. Our results confirm that if the true history lies within the models considered, the parameters of relatively complex scenarios can be well recovered from the (joint) SFS. However, we must keep in mind that histories outside our model family might have identical likelihoods.
One disadvantage of our method (and of any other simulation-based method) is that we are approximating the likelihood, implying that two runs from identical initial parameter values can results in different estimations (see Figure S2). Using more simulations for the estimation of the likelihood would lessen but not totally suppress this problem, but our results show that our maximization procedure leads to almost completely unbiased estimates and converges to correct values. Another disadvantage of our approach is its dependence on composite likelihoods. More powerful full likelihood approaches explicitly take into account linkage disequilibrium (LD) between sites [60], and therefore might reveal useful to infer recent migration events (see e.g. [61]). That being said, our applied data sets consist of SNPs randomly distributed across the whole genome, and so patterns of LD between sites are minimal. Whereas confidence intervals of demographic parameters based on composite likelihood ratios should in principle be too narrow (see e.g. [21], [60], [62], [63]), a study based on short stretches of DNA sequences has empirically shown that they were extremely similar to those obtained by explicitly modeling patterns of recombination [54]. This appears unlikely to be true in general, and certainly not if products of pairwise composite likelihoods were used (as with eq. (7), which was actually not used for our test cases). Similarly, the use of composite likelihoods in model tests based on AIC can overestimate the support for the most likely model [64]. However, the composite likelihoods in our test cases are quasi likelihoods due to the global independence between SNPs, and the differences in relative likelihood of alternative models are so huge (see Tables S2 and S3) that some residual patterns of LD are unlikely to change our conclusions.
As an alternative to our composite likelihood maximization approach, Garrigan [22] has proposed to integrate an approximate likelihood computed in a way similar to ours into an MCMC algorithm, allowing him to get posterior distributions and credible intervals. Whereas MCMC algorithms generally assume that the likelihood is computed accurately, it has been shown that MCMC procedure should lead to correct posterior distributions even if the likelihood is approximated, provided that there is no systematic error in its computation [65], [66]. This Bayesian approach could be worth exploring as a possible extension of our likelihood maximization procedure. However, our current implementation has the advantage of quickly getting point estimates, around which CIs can be obtained later by repeating the estimation on bootstrapped samples. For instance, a point estimate for the IM model shown in Figure 1B is obtained in about 2h30 on a single core machine, whereas 40–80 h are necessary to get posterior distributions for the parameters of a similar IM model from a single MCMC run using a specialized coalescent program on a multi-core machine [see 22].
The additional versatility of our simulation-based likelihood approach is well exemplified by its handling of ascertained SNP chips, and the inference of several parameters from the SFS under complex demographic scenarios. Previous ways of handling ascertained SNP chips either consisted in removing the bias induced by the ascertainment [44] or taking it into account in the estimation procedure [39], [45]. However, these methods are usually not as general as our implementation, as they are either restricted to models including a single population [44], or to the case of the sole estimation of divergence time between two populations [45]. Contrastingly, our method can be applied to various types of demographic models including several populations, bottlenecks and migration.
Our simulation results suggest that parameters of complex models can be correctly recovered when the ascertainment consists of randomly chosen SNPs heterozygous in a single individual (Figures S8 and S9). Interestingly, we find that some parameters of unascertained populations that diverged a long time ago either with (Figure S8) or without (Figure S9) admixture can also be quite well estimated when the model is well specified. This suggests that a given ascertainment panel of the GWHO Affymetrix chip could be used to infer parameters in several related populations. It is also worth noting that our calibration of parameters relied on the assumption that the divergence time with an outgroup population was known, but a different divergence time would only require a rescaling of the estimated parameters. The use of an outgroup species with fixed divergence time is a standard way to calibrate mutation rates (as e.g. in [21]), but we note it could also be used within species for DNA sequence data when some uncertainty exist on mutation rates, which is currently the case in humans [67], [68].
Most parameters inferred from real African populations have very similar estimates and confidence intervals irrespective of which SNP panel is used (Figure 5, Table 2), which agrees with our simulation results (Figures S8, S9). However, a few parameters seem to provide relatively divergent estimates, like the Yoruba and the African ancestral size, as well as the Yoruba-San divergence time, a discrepancy that is not really expected from the simulations. This discrepancy could stem from either an unknown source of ascertainment, from a misspecification of the model for one of the two ascertained population, or from an ascertained individual that is not representative of its population, the latter case being possibly due to inbreeding or admixture. It currently appears difficult to disentangle these cases, and the inclusion of additional parameters in model B only seems to marginally improve the fit of the expected SFS to the data. It suggests that our models still do not capture all aspect of the true demography of these populations, which might also affect our ability to reproduce the ascertained SFS, and have a negative impact on our estimations. We note however that previous estimates of African demography [e.g. 53] are more in line with those inferred from the Yoruba than from the San panel, which could suggest that our demographic models are more appropriate for the Yoruba than for the San population. Overall, our results nevertheless show that meaningful demographic estimates can be obtained from ascertained SNP chips, suggesting a useful and cheap alternative to large scale sequencing for demographic inference.
Our methodology has the potential to infer demographic parameters from large scale genomic data under a much wider range of neutral evolutionary models than either the current implementation of , current Approximate Bayesian Computation (ABC) implementations [69], summary statistics based approaches [11], or other existing likelihood-based methods [22]. Whereas ABC has the potential to be applied to genomic data, it has rarely been done since it usually requires the simulations of data sets as large as those analysed, which is computationally very costly. Our approach could thus be seen as a powerful likelihood-based alternative to the study of complex evolutionary models, which are usually only tackled by ABC approaches [see e.g. 16], [70], [71], with the additional advantage of not having to choose which summary statistics to use for the inference, which is often a problem in ABC [e.g. 13], [72], [73], [74]. Our approach can indeed tackle complex evolutionary models with a relatively large number of populations (see Figs. 1D, 4B and 5B). For instance, the model shown in Figure 4B includes 4 sampled populations, as well as four other unsampled populations, whose demography also needs to be reconstructed. AIC analysis reveals that the cost associated to increasing model complexity is rewarded by a much better fit to the data. One should however make a distinction between the inclusion of additional parameters for a given number of populations (e.g. adding the possibility to have gene flow between populations), and the inclusion of additional populations. The addition of unsampled or ghost populations can not only modify parameter estimations but also alter our interpretation of the results (see e.g. [75], [76]). For instance, the inclusion of continents from which sampled populations received migrants (which is an attempt at taking into account the spatial structure of African populations) in Figure 4B improved the fit of expected SFS (see Table S2), without really modifying our estimation of the level of European admixture, but it radically changed our interpretation of the relationships between African Americans and extant African and European populations. As expected, the inclusion of a potential source of admixture for the Luhya population in model B2 improved the fit of the model and it allowed us to make inference about this ghost population, but it also modified estimated parameter values of this and other populations. These observations suggest that complex models are better studied by considering all populations simultaneously, and that a strategy consisting in estimating population-specific parameters and fixing them when incorporating additional populations would not be optimal.
There are still some limits to the complexity of models that can be studied, and AIC-like approaches can be used to study which modifications sufficiently improve the model to be preserved. However, the question of whether our best model is the true model is not addressed by model comparisons such as likelihood ratios or AIC. One would ideally like to assess how well the model explains the data, which is usually done by some posterior predictive check in a Bayesian setting [77], or by getting the data p-value under a frequentist approach. We have implemented such an approach, where the model p-value was evaluated by comparing an observed G-test statistic [3], [62] to its model distribution. As expected, this approach leads to non-significant p-values when applied to simulated data sets (Figure S11). However, the p-values for all models shown in Figures 4 and 5 are highly significant (p = 0, Figures S12 and S13) suggesting that our implemented models of human evolution are still overly simplistic. This is not surprising given the high-dimensionality of the parameter space and the large amount of SNPs at hand giving us high power to reject inaccurate hypotheses. Since models are generally expected to be wrong, the question is at what point is a model so wrong that it is no longer useful [78, p. 74]. The fact that the addition of plausible source of realism into our models significantly improves the fit to the data (Tables S2 and S3) is reassuring in the sense that we have a methodology to refine our still imperfect evolutionary scenarios.
Nielsen [17] has shown that one could estimate the likelihood of a demographic model , where X is the site frequency spectrum, on the basis of coalescent simulations. This is because the probability of a given derived allele frequency i is simply a ratio of branch lengths of the coalescent tree expected under model as [17]:(1)where is the total length of a set of branches directly leading to i terminal nodes, and T is the total tree length. This probability can then be estimated with arbitrary precision on the basis of Z simulations as [62](2)where is the length of the j-th compatible branch in simulation k (see Figure S1A). Note that the estimator shown in eq. (2) implicitly weights simulations according to the probability that a mutation occurs on the simulated tree. Note that an estimator of the form (as used by Garrigan [22] to estimate the expected SFS) would give each tree the same weight and would thus give an excessive weight to genomic regions with shallow coalescent trees, which can be a problem for recently bottlenecked populations. If some simulated entries of the SFS were zero (because), was set to an arbitrarily small values [as in 22] chosen here as .
We have empirically checked that our procedure gives the correct SFS under two simple scenarios for which the expected SFS can be obtained exactly by the method developed by Chen [18] for cases involving up to two populations and no migration. These scenarios were (i) a bottleneck model (as in Fig. 1A) and (ii) a divergence model without migration (as in Fig. 1B but without migration). We show in Figures S14 and S15 for scenarios i) and ii), respectively, the fit of the SFSs entries (estimated by our approach for different numbers of coalescent simulations) to the true SFS entries. As expected the fit improves with the number of simulations, and the estimated SFS entries are distributed symmetrically around the true values without any visible bias for these two scenarios.
Probabilities inferred from the simulations and eq. (2) can then be used to compute the composite likelihood of a given model as [20](3)where is the SFS in a single population sample of size n, S is the number of polymorphic sites, L is the length of the studied sequence, and is the probability of no mutation on the tree, obtained as assuming a Poisson distribution of mutations occurring at rate .
This formulation can be extended for the joint SFS of two populations as(4)and one can define a v-dimensional SFS for more than two (v) populations as(5)where is a composite index. However, when the number of populations in the model is larger than 2 and sample sizes are relatively large, the number of entries in the v-dimensional SFS can be huge, implying that most entries of the observed SFS will be either zero or a very small number and that the expected values for these low-count entries will be difficult to estimate precisely. In that case, we have chosen to estimate the v-dimensional by collapsing all entries with observed SFS less than a predefined threshold as(6)When v>4, this approach will also prove computationally difficult, and in that case we have chosen to compute a composite composite-likelihood (C2L) obtained by multiplying all pairwise CL's, as(7)where is given by eq. (4).
As the likelihood is obtained by simulations, which incurs some approximation, we cannot use optimization methods based on partial derivatives. Even though other methods would be possible, we have chosen to use a conditional maximization algorithm [ECM, 79], which is an extension of the EM algorithm where each parameter of the model is maximized in turn, keeping the other parameters at their last estimated value. The maximization of each parameter was done using Brent's [80, Chapter 5] algorithm, which is a root-finding algorithm using a combination of bisection, secant and inverse quadratic interpolation [see e.g. 81]. We start with initial random parameter values, and perform a series of ECM optimization cycles until estimated values stabilize or until we have reached a specified maximum number of ECM cycles (usually 20–40). Unless specified otherwise, we used 100,000 coalescent simulations for the estimation of the expected SFS and likelihood for a given set of demographic parameters. Even though a higher precision could be reached with a larger number of simulations, especially for complex models, this number appears like a good compromise between computational efficiency and likelihood estimation accuracy (see Figure S2). Note that the imprecision on the likelihood estimation might also prevent an efficient optimization of our parameters, as a sub-optimal parameter might give by chance a better likelihood than the optimal one during an ECM cycle. Because the composite likelihood surface might have several local maxima and be difficult to explore [e.g. 60], several independent optimizations are performed (between 20 and 40 depending on the model and computation time), each starting from different initial conditions, and the overall maximum composite likelihood solution is retained.
Coalescent simulations, estimation of the SFS, likelihood computations and its maximization were all done with fastsimcoal2, a modified version of the fastsimcoal program [82]. fastsimcoal2 input file format and command lines arguments are briefly described in Supplementary Text S1, and examples of input files are provided in Supplementary Text S2.
We have tested our program ability to recover demographic parameters from DNA sequence data in four relatively plausible but distinct scenarios of population differentiation involving one to ten populations with migration (see Figure 1). In all cases, we simulated with fastsimcoal2 400,000 unlinked regions of 50 bp, thus totaling 20 Mb of DNA sequences, assuming a mutation rate of 2.5×10−8 bp−1 per generation and an infinite-site model. Pseudo-observed SFS were also directly computed with fastsimcoal2. Parameters were estimated independently from ten data sets generated under each model. For each data set generated under models with one to three populations, we performed 50 parameter estimations via ECM maximization, and each time retained the parameter set with maximum likelihood. For the model with 10 populations we only performed 20 estimations per data sets, and used 50,000 simulations instead of 100,000 for the other models to estimate the expected SFS due to long computation times. We describe the four tested models in Figure 1, and the used parameter values are showed as red dots in Figures 2, 3, S4 and S5. Absolute numbers (generations, population sizes) were obtained by assuming that the mutation rate of 2.5×10−8 bp−1 per generation was known.
As a benchmark, we used to infer the demographic parameters in scenarios shown in Figure 1A–1C involving up to three populations. For each generated data set, we performed 50 parameter estimations using the Broyden-Fletcher-Goldfarb-Shanno (BFGS) optimization method implemented in , and we retained the parameters associated with the maximum likelihood. We followed 's manual specification to set reasonable upper and lower bounds of the search ranges of the parameter. In all cases, the expected SFS was estimated by extrapolating the SFS inferred from 3 grid sizes set to 40, 50 and 60, which are in all cases larger than our maximum samples sizes (30 in the IM model case). The composite likelihood was computed using 's multinomial model, which is in fact a product of Poisson likelihoods, where the expected model entries are scaled to sum up to 1. This likelihood also ignores information about the expected and observed numbers of monomorphic and polymorphic sites used in our likelihood formulation (as well as in [20]). Therefore, the ratio should be equal to showing that barring the terms, the two CLs differ by a single constant value. The difference between likelihoods computed with fastsimcoal and is illustrated in Figure S3 for the case of the bottleneck scenarios shown in Figures 1A. It shows that when monomorphic sites are not taken into account, fastsimcoal and indeed produce essentially identical likelihood profiles around true parameters. However, when monomorphic sites are used in the likelihood, the shape of the likelihood profiles differs, making it more or less peaky depending on the parameter. There is thus no clear advantage in using one or the other likelihood form for this scenario, but our use of monomorphic sites allows us to directly get absolute values of the parameters. We report in Figures 2, S4 and S5 only the results obtained for data sets for which 's best log likelihood was less than 10% lower than the largest log-likelihood obtained with the other data sets, and we considered not to have converged for the discarded data sets.
Recently, Affymetrix developed a new SNP array including ∼629,000 SNPs with known ascertainment scheme for population inference (Axiom Genome-Wide Human Origins 1 Array, http://www.affymetrix.com/support/technical/byproduct.affx?product=Axiom_GW_HuOrigin) [43]. This array, abbreviated hereafter GWHO, is made up of SNPs defined in 13 discovery panels. In the first 12 panels, SNPs have been identified by comparing the two chromosomes of an individual from a known population, further quality checks and validation on a large population sample [43]. The 13th panel contains SNPs that are polymorphic when comparing the Denisovan sequence and a random San chromosome. Raw genotypes from 943 unrelated individuals from more than 50 worldwide populations are freely available on ftp://ftp.cephb.fr/hgdp_supp10/.
The ascertainment scheme of this array is simple and homogeneous over a given panel. However, the SFS inferred from this array is biased as only mutations that occur in the ancestry of the two compared chromosomes will be considered (see Figure S1B). We show in Figure S7 the difference between the ascertained and non-ascertained SFS under a few basic demographic scenarios in a single population. The differences between the two SFS can be quite dramatic, implying that the estimation of demographic parameters on ascertained data sets without taking the ascertainment into account is bound to lead to biased estimates. Nielsen et al. [44] have shown how to correct the expected SFS within a given population under such a simple ascertainment scheme, and the ascertained joint SFS could be unbiased in a similar way by taking into account ascertainment probabilities in the ascertained populations. Rather than unbiasing the SFS, we have chosen here to incorporate the bias in the model and to infer demographic parameters directly from the ascertained (joint) SFS, a strategy similar in spirit to that used by Gravel et al. [28] to account for biases in the SFS obtained from low-coverage next-generation sequencing data. It implies we need to model the ascertainment scheme in the coalescent simulations such as to infer the expected ascertained SFS for a given demography. In order to estimate the SFS when SNPs are defined as being sites heterozygous in a given individual, we use the following procedure: 1) we perform conventional coalescent simulations under a given demography, 2) we choose two lineages at random in the ascertained population, 3) we identify the subtree relating the chosen lineages to their most recent common ancestor (MRCA) (highlighted in blue in Figure S1B), 4) we update the numerator in eq. (2) by summing up branch lengths of the blue subtree that are ancestral to i1 lineages in population 1, i2 lineages in population 2, …, iv lineages in population v, 5) The denominator of eq. (2) is updated by summing up the total length of the blue subtree.
Parameter optimization is then performed similarly to the unascertained case, except that the terms depending on the number of monomorphic sites () in eq. (6) are removed from the likelihood since only polymorphic sites are reported on the ascertained chip, which implies that we cannot use a molecular clock. Therefore, parameter absolute estimation should be done relative to an arbitrarily fixed or known parameter (e.g. population size, divergence time). Note however that a molecular clock could be used if the fraction of sites found heterozygous were known in ascertained individuals, as in this case the expected fraction of monomorphic sites would then simply be , where would be the total length of the expected ascertained tree (shown in blue in Figure S1B).
As mentioned in the next section on model test, it might be difficult to accept a simple model with a G-test based on tens of thousands of polymorphic sites, but in that case, it might be better to establish a procedure allowing one to improve models, by progressively adding some realism to simple models [89]. Our likelihood-based approach would in principle lend itself to model comparison through likelihood-ratios for nested models or through Akaike Information Criterion (AIC, [52]) for other model comparisons. However, we are here confronted with two distinct problems. The first one affects all composite likelihood approaches and due to the fact that the distribution of the composite likelihood ratio test (CLRT) is generally unknown. When the SFS is obtained from DNA sequences with relatively large levels of linkage disequilibrium, it has been proposed to obtain an empirical distribution of the CLRT by simulation of DNA sequences with recombination (e.g. [54], [90]). In the case of the AIC, Varin and Vidoni [91] have proposed to replace the number of parameter d of a given model in by an effective number of parameters that needs to be computed from a sensitivity matrix and Godambe Information matrix, which might be difficult to do in practice. We note however that in our two applied examples the SFS is computed from a collection of SNPs randomly distributed across the genome, such that we shall conservatively assume that the CL computed from the multidimensional SFS is close to a true likelihood. Note that this assumption would not be valid if one had computed a composite likelihood based on the product of pairwise composite likelihoods, like in eq. (7). The second problem is linked to the fact that we estimate the likelihood with some error (Figure S2). As noted previously, this can prevent us to efficiently optimize our parameters, but it also means that the likelihood ratios or AIC statistics are imprecisely estimated. To address this problem, we have compared models on the basis of the maximum value of the likelihood obtained over 100 estimations performed for the ML parameters obtained by our optimization procedure. We then calculated the relative likelihood or the Akaike's weight of evidence in favor of the i-th model as (see e.g. [89])where .
Even though we can estimate parameters under any model, it can be useful to check how data fit the chosen model. To this aim we use an approach based on a likelihood ratio G-statistic [3], [62] of the form , where CL0 is the observed maximum composite likelihood where the expected SFS is replaced by the relative observed SFS in eqs. 3, 5 or 7, and CLE is the estimated composite maximum likelihood computed using the procedure described above. We obtain the null distribution of the CLR statistic by the same parametric bootstrap procedure used to infer confidence intervals, where we generate by simulation a number of data sets using the estimated maximum-likelihood parameters of the model, and each time perform parameter estimations and estimate the CLR statistic. We can compute the p-value of the observed CLR statistic from this null distribution. Note that this type of G-test has been used before to find genomic regions under selection [3], [62].
We report in Figure S11 the null distributions of the CLR and the p-values of two data sets generated under models shown in Figure 1A and 1B. In both cases, the p-values are not significant confirming that the data sets are compatible with the tested models.
Note however that a non-significant p-value is not an absolute proof that the tested model is correct, as there could be a large number of models leading to similar SFSs, as was shown previously [27], but it is an indication that the observed SFS is well explained by the model.
However, in applied cases, we actually expect that this test leads to very significant values, since the true history of the populations is completely unspecified and our models are certainly overly simple and potentially far from reality.
|
10.1371/journal.pbio.1001329 | A Systematic Approach to Pair Secretory Cargo Receptors with Their Cargo Suggests a Mechanism for Cargo Selection by Erv14 | The endoplasmic reticulum (ER) is the site of synthesis of secreted and membrane proteins. To exit the ER, proteins are packaged into COPII vesicles through direct interaction with the COPII coat or aided by specific cargo receptors. Despite the fundamental role of such cargo receptors in protein traffic, only a few have been identified; their cargo spectrum is unknown and the signals they recognize remain poorly understood. We present here an approach we term “PAIRS” (pairing analysis of cargo receptors), which combines systematic genetic manipulations of yeast with automated microscopy screening, to map the spectrum of cargo for a known receptor or to uncover a novel receptor for a particular cargo. Using PAIRS we followed the fate of ∼150 cargos on the background of mutations in nine putative cargo receptors and identified novel cargo for most of these receptors. Deletion of the Erv14 cargo receptor affected the widest range of cargo. Erv14 substrates have a wide array of functions and structures; however, they are all membrane-spanning proteins of the late secretory pathway or plasma membrane. Proteins residing in these organelles have longer transmembrane domains (TMDs). Detailed examination of one cargo supported the hypothesis that Erv14 dependency reflects the length rather than the sequence of the TMD. The PAIRS approach allowed us to uncover new cargo for known cargo receptors and to obtain an unbiased look at specificity in cargo selection. Obtaining the spectrum of cargo for a cargo receptor allows a novel perspective on its mode of action. The rules that appear to guide Erv14 substrate recognition suggest that sorting of membrane proteins at multiple points in the secretory pathway could depend on the physical properties of TMDs. Such a mechanism would allow diverse proteins to utilize a few receptors without the constraints of evolving location-specific sorting motifs.
| All cells sense their environment, respond to it, and communicate with neighboring cells. To perform these functions, cells use an impressive array of proteins that they display on their surface membranes and secrete into their external environment. Newly synthesized proteins destined for the surface of nucleated cells, or to be secreted into the environment must enter the secretory pathway through the endoplasmic reticulum. Those that reside there remain behind, but most leave for their next destination as cargo proteins in lipid vesicles. To be packaged into vesicles, many of them require a “cargo receptor,” which recognizes and tethers specific cargo proteins in the vesicles. Our study takes a systematic approach to identify the range of cargo proteins that bind to each of the known receptors in yeast. By using this approach, we both discover new cargo for known cargo receptors and delineate the rule that governs cargo selection for one cargo receptor, Erv14. Thus, our study demonstrates a novel approach to identify the cargo for any receptor or to discover new cargo receptors.
| The endoplasmic reticulum (ER) is the entry site into the secretory pathway, responsible for the folding, maturation, and trafficking of all secreted, membrane-bound, and secretory pathway resident proteins. Once folded, the proteins exit the ER as cargo within COPII-coated vesicles that bud from ER exit sites [1],[2]. Active concentration of proteins into the vesicles [3]–[6] occurs either by direct interaction with the Sec23 and Sec24 subunits of the COPII coat or else are mediated through a diverse group of proteins that function as “adaptors” and have been termed cargo receptors [4],[7]. Cargo receptors allow sorting of cargo that cannot directly bind Sec23/24, or cargo whose exit requires quality control or regulation [8],[9].
The prevalent way to identify cargo for a cargo receptor entails testing selected individual proteins in transport assays or in vitro COPII budding reactions, as was utilized to pair glycosylphosphatidylinositol (GPI)-anchored proteins with their cargo receptors—the p24 family of proteins [10]–[12]. Because of the complexity of these approaches, only a few additional cargo receptors have since been identified (Table S1). Moreover, despite their important function in ER exit and their potential for regulating the flow of traffic in the entire secretory pathway, there is still little information about the entire spectrum of cargos for a given cargo receptor or what defines its cargo specificity. Importantly, no attempt has yet been made to pair large sets of possible cargos with their cargo receptors in a manner that is systematic and unbiased. The lack of systematic data has hindered the identification of the determinants shared by specific sets of cargo that allow their recognition by a particular cargo receptor. Identification of such determinants might also shed light on the purpose and mechanism of action by which a given cargo receptor operates.
Here we describe a systematic approach that aims to complement the traditional methods of cargo and cargo receptor discovery which we call “PAIRS” (pairing analysis of cargo receptors). The PAIRS approach utilizes robotic methodologies to genetically manipulate Saccharomyces cerevisiae libraries containing green fluorescent protein (GFP)-tagged cargo [13]–[15], followed by automated microscopy to identify mutated backgrounds that cause ER retention of cargo. Using PAIRS we have increased the number of known cargos for a set of nine cargo receptors. Since our approach probes a large set of proteins for their receptor requirements it defines both groups that are dependent and that are independent of any given cargo receptor. Combined, this should help to define the rules of specificity for each of the cargo receptors. We demonstrate the utility of this approach by using the set of cargo uncovered for the cargo receptor Erv14 to formulate a hypothesis on its mode of substrate recognition. The large group of cargo that require Erv14 as their cargo receptor do not share a detectable functional similarity or sequence motif. However, all identified cargo resides in late secretory pathway membranes that are populated with proteins of longer transmembrane domains (TMDs) than TMDs of ER resident proteins [16]. This raises the hypothesis that cargo specificity of Erv14 is determined by TMD length. Following up on one substrate, Mid2, we show this to indeed be the case. Thus Erv14 may be able to recognize a diversity of cargo by recognizing a shared physical property rather than a specific sequence. This also suggests a resolution for conflicting findings on the effect of TMD length on protein retention in the ER or Golgi [17]–[22].
To pair as many cargo proteins as possible with their corresponding cargo receptors in a systematic, non-biased approach, we devised a methodology we call PAIRS. PAIRS is based on the idea that when a cargo receptor is missing, then its cargo accumulates in the ER and that this can be visualized by using fluorescently tagged cargo.
The PAIRS approach can be used for two purposes. First, it can be used to uncover the cargo receptor for a specific cargo of interest by expressing that specific cargo fused to GFP on the background of mutations in trafficking-related proteins. Second, it can be used to uncover the spectrum of cargo for a putative cargo receptor by visualizing a large number of strains with various GFP-tagged cargo on the background of mutations in that cargo receptor. The approach relies on systematic creation of genetically modified strains using the synthetic genetic array (SGA) methodology [13],[14],[23], which is followed by acquisition of fluorescent images of all strains using a high-throughput automated microscopy platform. Finally, manual examination of the resulting images uncovers strains in which the mutation causes ER retention of cargo, implying a cargo receptor/cargo pair (Figure 1).
To determine whether our methodology can indeed facilitate the identification of a cargo receptor for an arbitrary cargo, we chose Tpo4, a plasma membrane multidrug transporter involved in polyamine transport, whose ER exit had not been shown to rely on a particular cargo receptor. We first created an SGA compatible query strain expressing Tpo4-GFP from its endogenous promoter. We then collected strains of mutants in trafficking related proteins from either the deletion library (for non-essential genes) [11], or from the decreased abundance by mRNA perturbation (DAmP) library (for hypomorphic alleles of essential genes) [24] (for a full list of strains used and the proposed function of their corresponding protein see Table S6). Using the SGA approach [14], we crossed the Tpo4-GFP into the mutant library creating a new library of haploid yeast strains each expressing Tpo4-GFP on the background of a mutation in a single gene. Visualization of these strains demonstrated that all but one of the strains did not alter Tpo4-GFP's localization (Figure 2). Only the Δerv14 strain displayed ER accumulation of Tpo4-GFP (red arrows in Figure 2). Although Erv14 is a known cargo receptor [25]–[27], it has not been previously implicated in trafficking of Tpo4. Our analysis suggests that Erv14 is the cargo receptor for Tpo4 and demonstrates that the PAIRS methodology can be used to find a cargo receptor for a given cargo of interest.
We next wanted to utilize the PAIRS methodology to map the spectrum of cargos for a cargo receptor of interest. We therefore created nine query strains, each carrying a deletion or a DAmP hypomorphic allele of a putative cargo receptor: Δerv14, Δerv15, Δerv26, Δerv29, Δemp24, Δemp47, Δgsf2, Δchs7, and Shr3-DAmP. To generate a library of GFP-tagged cargo we used the Yeast GFP Fusion Localization Database to identify fusion proteins that were reported to reside in post-ER compartments (Golgi, puncta, vacuole, or cell periphery) [15]. These were filtered to remove all those without TMDs or a signal peptide giving rise to 157 strains. The strains and controls were assembled and all of these were crossed into each of the nine query strains thus generating nine new libraries of GFP-tagged cargo proteins, each lacking an individual putative receptor.
Inspection of the strains showed that the majority of cargos (126 out of 157) managed to exit the ER in all deletion backgrounds, suggesting that they do not solely rely on any one of the nine cargo receptors studied here for their ER exit. This suggests that most proteins can either bind the COPII coat directly, depend on redundant mechanisms for ER exit, rely on as yet undiscovered cargo receptors, or that they are exported out of the ER by spontaneous “bulk flow.” However, for all but one cargo receptor, Erv15, we could find at least one cargo that depended on it. The annotation of Erv15 stems from its high homology to the cargo receptor Erv14 (63%), and it appears to be required to augment the activity of Erv14 in transporting particular cargo in sporulating cells but not under normal growth conditions [25],[26]. The full set of novel cargos found for each of the other seven cargo receptors is shown in Figure 3 (for previously characterized cargo that were verified by the screen see Figure S1). It appears that there is never a complete blockage of ER exit; this may simply reflect the proteins leaking out of the ER once they have accumulated to high levels, but there may also be some functional redundancy in the ER exit machinery.
Since this is the first time that all cargo receptors have been studied in the same system and under the same conditions in a systematic manner, the spectrum of cargo uncovered for each cargo receptor could also be used to start defining the functional rules guiding the recognition mode. For example, all cargo for Erv26 comprised of Golgi-localized mannosyltransferases (Figure 3C) as had previously been suggested [28],[29]. However Erv26 seems to be specific to a subset of this functional group as several additional mannosyltransferases (Mnn1-GFP, Mnn11-GFP, Mnn10-GFP, Anp1-GFP, and Hoc1-GFP) did not accumulate in the ER in this background (unpublished data). Another example for specificity is our finding of only a single novel cargo for Shr3 and the identity of this cargo as an amino acid permease (Figure 3E) as are all previously identified cargo, supporting the notion that Shr3 is a dedicated cargo receptor for amino acid permeases [8],[9],[30],[31]. A similar picture emerges for Gsf2 whose novel cargo all fall into the same functional category of sugar transporters (Figure 3B) as reported previously [32]. Moreover, the sugar transporter Hxt2-GFP previously shown to be independent of Gsf2, is indeed properly localized to the plasma membrane (unpublished data), supporting the notion that Gsf2 is involved in exit of only particular sugar transporters from the ER. Other cases are less clear, such as the three seemingly unrelated cargo that we uncovered for Erv29 (Figure 3A). Previous reports identified three soluble proteins as requiring Erv29 for efficient ER exit (PrA, CPY, and α-factor) [33],[34], and although one of the new proteins is a soluble protein (Pry1), two others (Ear1 and Mam3) are membrane proteins of the vacuole or endosome.
Perhaps the most striking finding is the large number of proteins that require Erv14 for efficient ER exit. Erv14 was identified as being enriched in COPII vesicles and shown initially to be required for the ER exit of the plasma membrane protein Axl2 [26]. Recent work has shown that mutants lacking Erv14 also show ER accumulation of the proteins Sma2 [25], Mid2, Gap1, Hxt1, and Hxt2 [35]. Our PAIRS approach identified that Erv14 is required for the ER exit of 32% of the plasma membrane proteins checked (18 of 57) (Figures 3F and S2A and S2B). Among these proteins are permeases (e.g., Mep2-GFP), transporters (e.g., Hxt2-GFP and Nha1-GFP), multidrug transporters (e.g., Snq2-GFP and Tpo4-GFP), lipid flippases (e.g., Cdc50-GFP and Dnf1-GFP), eisosome components (Sur7-GFP), and proteins involved in cell polarity or cell wall regulation (e.g., Mid2-GFP and Axl2-GFP). Some have a single TMD whilst others are polytopic with up to 12 TMDs. Hence there is no obvious functional or structural similarity between the proteins affected by Erv14.
Consistent with previous work, Erv14 was not required for ER exit of soluble proteins and non-conventional membrane tethered proteins such as GPI-anchored and tail-anchored proteins (Figure S3) [26],[35]. To strengthen the predictions made by our PAIRS methodology we analyzed the physical interactors of Erv14 under the assumption that direct cargo should physically interact with its cargo receptor. To this end, we immunoprecipitated HA-tagged Erv14 (which completely retains the function of the endogenous Erv14 [36]) from microsomes and analyzed the precipitated proteins by mass spectrometry. Using this approach we could corroborate eight out of the 23 cargo predicted by PAIRS as physically interacting (Figure S4A). We also found five interacting proteins that could be cargo; however, they were not examined in our original screen because of mislocalization of the C-terminal tagged fusion. To verify these proteins as cargo we made strains expressing N-terminal GFP fusion proteins and found that two of them are indeed retained in the ER in Δerv14 (Figure S4B).
Live cell imaging confirmed that Erv14's absence decreased the kinetics of ER exit of predicted cargo (Figure S5), raising the question of how it could accelerate the ER exit of such a defined set of diverse proteins. We performed in-depth sequence analysis of Erv14 cargo but could not uncover any identifiable sequence motifs (unpublished data). However, the fact that all cargos of Erv14 are membrane proteins destined to reside in the membranes of the late secretory pathway suggested that inherent characteristics of the membrane-spanning region might be responsible for the recruitment of Erv14. Indeed, a comprehensive comparison of TMDs of bitopic proteins from different compartments has shown that TMDs from post-Golgi compartments are significantly longer, suggesting that the bilayer is thicker [16]. Thus a larger hydrophobic portion, adapted for the apparently thicker bilayer of the plasma membrane, may be the trait that determines potential cargo for Erv14. We thus investigated the dependence of ER exit of an Erv14-regulated cargo on its TMD length.
To assay the effect of TMD length on protein sorting by Erv14 we used the plasma membrane cargo protein Mid2 as a reporter. Mid2 is a non-essential type I membrane protein with a signal peptide and a single, 26–amino acid-long, TMD (Figure 4A). One advantage of using Mid2 is that its maturation along the secretory pathway can be monitored owing to the presence of luminal modifications by a single N-linked glycan and multiple O-linked glycans. Since the extension of the O-linked glycans occurs in the Golgi and results in reduced mobility on SDS gels, this can be used to assay the extent of ER exit [37]. To remove any possible interference that may stem from specific sequences in the TMD we replaced the endogenous TMD with a stretch of 26 leucines (Mid2L26M). The residues at either end of the TMD were modified to be basic to provide sharp ends to the hydrophobic region (Figure 4B), and because basic residues are the most common charged residues at both the cytoplasmic and luminal ends of the TMDs of yeast plasma membrane proteins [16]. To ascertain that these changes did not alter the basic cargo properties of Mid2, we expressed a GFP-tagged form of Mid2L26M in yeast and observed that it localizes to the plasma membrane in a manner identical to the wild-type protein (Figure 4C). This finding indicates that our synthetic reporter is correctly localized and that Mid2 does not require specific motifs within its TMD for its trafficking through the secretory pathway.
We next generated variants of Mid2 in which the polyleucine TMD was shortened in increments of two residues to give a set of variants spanning the range of 14–26 residues. To examine the trafficking of these variants, each one was expressed under the control of a galactose-inducible, glucose-repressible, promoter. By inducing transcription with galactose for 90–120 min followed by termination upon return to glucose, it was possible to generate a pulse of protein whose progress through the secretory pathway could be followed by blotting and microscopy.
When we compared the TMD-length variants by blotting we found that they behaved differently (Figures 4D and S6A). The majority of the longest TMD form was exported from the ER even during the pulse of induction, and the remaining material was rapidly matured during the chase. However, as TMDs became shorter in the variants, the ER form took a longer time to disappear. Similar results were obtained with an independent set of TMD variants in which the polyleucine stretch was flanked by a tryptophan at either end, a residue sometimes enriched in this position (Figure S6B). In fact, in the shorter TMD variants we could not observe Golgi forms accumulating but instead we observed an accumulation of a band corresponding to free GFP. This free GFP likely reflects degradation of the short TMD variants in the vacuole, as it is not present in a strain expressing the 16-leucine variant (Mid2L16M) and lacking the vacuolar hydrolase Pep4 (Figure 4E). Therefore we assume that once out of the ER and in the Golgi, the shorter TMD variants are directed to the vacuole and degraded, whereas the longer TMD forms are trafficked correctly to the plasma membrane. Thus, the steady state pool of the Golgi-modified, but not yet degraded, form of the shorter variants will be low. Quantitation of the ER form confirmed that the longer TMD variants exited the ER much more rapidly, with 20 leucines being the point of transition (Figure 4F).
We next asked what effect TMD length has on Erv14-dependent exit. By repeating the pulse chase experiments in a Δerv14 strain we found that the ER exit of the long form of Mid2 was drastically slowed down, demonstrating a dependence on Erv14 (Figure 4G). However, we could not observe any change in the slow exit rate of the short TMD form Mid2L16M, indicating that this shorter form exits the ER in an Erv14-independent manner (Figure 4H). Taken together with the fact that the polyleucine TMD did not abolish Erv14 dependence of Mid2L24M, this strongly supports the idea that the length of the TMD and not its sequence are a determinant of Erv14-mediated sorting.
The above results are consistent with the idea that Erv14 selectively acts as a cargo receptor on proteins with a long TMD. However it is also possible that a short TMD acts as an ER retention signal by discouraging entry into COPII vesicles or ER exit domains, and hence this effect over-rides the ability of Erv14 to extract the cargo from the ER. To investigate this possibility we asked whether the short TMD forms of Mid2 could exit more rapidly if directed to COPII vesicles by a different mechanism. Thus we created new versions of the Mid2 reporter that were fused with the cytoplasmic tail of the Golgi-localized Sys1 protein that contains a DXE motif for direct binding to Sec24 (Figure 5A) [7],[38],[39].
When these constructs were expressed in wild-type cells we found that the ER exit of the short forms was significantly accelerated in comparison to the form that lacked the DXE motif and was now closer to the rate of the longer TMDs (Figure 5B and 5D). In addition, we observed a reduction in the levels of free GFP, indicating that once the constructs had left the ER, they were no longer rapidly degraded. When the DXE motif in the Sys1 tail was mutated to AXE the rate of ER exit was reduced, confirming that the effect is mediated by COPII binding (Figure 5C and 5E). Interestingly, the stabilization of the shorter TMD variants (i.e., the reduction in accumulation of free GFP during the chase) conferred by the Sys1 tail was retained even after DXE was mutated, suggesting that other sequences in the Sys1 tail are responsible for this effect. Taken together, these results indicate that the reduction in ER exit rate seen upon TMD shortening is a feature of an Erv14-dependent cargo, but not a feature of a cargo protein that is concentrated in COPII vesicles by other mechanisms. This suggests that Erv14 specifically directs the ER exit of proteins with longer TMDs.
In the above experiments we found that the Sys1 tail enabled all of the polyleucine TMD variants to efficiently exit the ER. This allowed us to investigate the effect of TMD length on later trafficking steps without complications from variations in ER exit rate. We thus examined the distribution of the different TMD length forms at the end of the galactose pulse (Figure 6A). As expected, the longer TMDs showed predominantly plasma membrane staining. However, the shorter TMD variants accumulated in intracellular puncta that were also labeled with the late Golgi protein Sec7 (Figure 6A).
This finding suggests that TMD length affected Golgi exit as well as ER exit. To confirm that this difference was not due to kinetic effects, we re-examined the distribution of TMD variants expressed constitutively under the control of the endogenous Mid2 promoter. As observed for the galactose-induced versions, the longer TMD variants show a clear plasma membrane localization, but the variants with TMDs of less than 22 residues again accumulated in puncta with little if any cell surface staining (Figure 6B). These puncta co-localized with the late Golgi marker Sec7, but not with the early Golgi marker Rud3, confirming that the protein was accumulating in trans Golgi compartments (Figure 6B and 6C). The same Golgi accumulation of shorter TMDs was observed in cells lacking the End3 protein that is required for endocytosis [40], negating the possibility that the shorter TMD variants are simply travelling to the surface and being more rapidly endocytosed (Figure 6D) [41].
To determine whether this TMD length-dependent sorting at the Golgi also required Erv14 we expressed the 16 and 24 leucine constructs with Sys1 tails in Δerv14. Both variants displayed the same ER exit rates; however, they still differed in location (Figure 7A and 7B). This indicates that the difference in localization was not dependent on any residual differences in ER exit rate conferred by the Erv14 protein, or by Erv14 chaperoning the protein through the Golgi and having a second role at Golgi exit. Taken together, these results demonstrate that the length of the TMD can determine the rate at which Mid2 leaves the Golgi to travel to the plasma membrane and, unlike the effect of TMD length on ER exit rate, this sorting does not require Erv14.
The PAIRS methodology aims at pairing cargos with their cargo receptors and to elaborate on the existing body of knowledge on cargo extraction from the ER. Our analysis did not include all possible cargos since the proteins in the GFP library are all C-terminally tagged and under their endogenous promoter. This resulted in a number of strains whose proteins were mislocalized due to the tag or had fluorescence levels below detection [15]. This may explain why we could not identify some of the previously recognized cargo/cargo receptor pairs. Despite these caveats, the power of the PAIRS analysis is that it is not biased within the set of pre-selected proteins, allowing us a broad overview of cargo receptors function. This has allowed us to gain insight to the rules governing their specificity and to present the first step towards creation of a cellular “traffickome.”
We demonstrate the value of the PAIRS approach by identifying putative new cargo proteins for most of the known cargo receptors of yeast. This represents 31 of the 157 proteins tested, of which 27 had not been previously linked to a cargo receptor. This is likely to be a slight underestimate of the success rate as some of the proteins stated to have a vacuolar localization in the GFP Database probably reflect ER residents displaced by the tag and hence would not be expected to have a dedicated cargo receptor. One general trend seen across all hits is that removal of a receptor did not result in a complete block of ER exit of its GFP-tagged cargo. This is consistent with previous studies of known cargo/receptor pairs including GPI-anchored proteins [10]. It may be that cargo receptors typically act to accelerate the exit of a particular cargo, and that some bulk flow always occurs, with the volume of this flow increasing for a particular protein if it accumulates in the ER.
The increased knowledge of the range of cargo for each specific cargo receptor should make it easier to generate hypotheses as to what determines the selective recognition of particular cargo by individual cargo receptors. Indeed, examination of the spectrum of cargo relying on the cargo receptor Erv14 suggested that this large and non-homogenous group is recognized on the basis of the length of its TMD. By assessing this hypothesis using one particular cargo, Mid2, we found that TMD length is a major determinant for allowing Erv14 to accelerate exit from the ER. It is still feasible that Erv14 recognizes a more specific motif in Mid2 adjacent to the TMD, and alterations in the length of the TMD affect the position of this region relative to the bilayer and thus prevent Erv14 binding. However, this seems unlikely given the wide range of bitopic and polytopic proteins that are affected when Erv14 is deleted, and their lack of shared sequence motifs. Interestingly, there have been two recent studies reporting that shortening the TMD of a mammalian protein reduces its exit rate from the ER [18],[19]. The mechanism for these effects was not determined, but one reporter used was VSV-G that has been found to depend on the Erv14 paralogue, CNIH4, for normal ER exit [42].
Whether Erv14 enables exit of long TMD-containing proteins from the ER by performing more than just COPII coupling is yet to be uncovered. One option is that it could also act as a chaperone to protect protruding hydrophobic residues on cargo proteins thus enabling them to assume a correct conformation in the shorter ER membranes. Another option is that it sorts long TMD containing proteins into areas of the ER that have thicker membranes thereby enabling their recruitment to vesicles. Regardless, it seems that speed of ER exit may be a major determinant in Erv14's function. In yeast, one of the substrates, Axl2, has been shown to require very rapid ER exit [26], as it must be inserted into the forming yeast bud at a particular point in the cell cycle (AXL2 mRNA is under cell-cycle control) [43]. One of the substrates for the Drosophila paralogue of Erv14, Cornichon, is Gurken, a TGFα-like bitopic protein [44]–[46]. The ability of Gurken to polarize Drosophila oocytes depends on its rapid exit from a restricted region of the oocyte ER following translation from a pool of mRNA that is spatially restricted for a short time during development [47],[48]. Indeed, also the action of Cornichon on Gurken requires that the latter has a TMD [47],[49].
Our analysis of the substrate recognition mode for Erv14 reveals that TMD length-dependent sorting may be a more general principle in cellular trafficking than previously appreciated. Using our DxE-containing Mid2 variant we noticed that the TMD variants also underwent a TMD length-dependent sorting in the Golgi apparatus. This is consistent with a previous study examining the effects of lengthening the TMD of the yeast ER protein Ufe1, although this is the first time it has been shown to occur with a homogenous synthetic TMD rather than a native TMD, which may contain additional cryptic sorting motifs [50].
How might length-dependent sorting occur in the Golgi if it does not involve Erv14? One option is that a dedicated cargo receptor exists at this compartment that has not yet been identified. However, an alternative option is that the vesicle composition itself plays a major role in this step with lipids and/or cargo proteins directing a change in bilayer properties [16],[51]–[53].
In summary, our unbiased approach allowed the formulation of a simple hypothesis for the underlying commonality allowing cargo identification by Erv14. Using Mid2 with a synthetic TMD has allowed us to indeed observe such TMD length-dependent steps both in the ER and the Golgi. The notion that TMD length is used by the cell to sort proteins is appealing [51],[53],[54], since many and diverse membrane proteins must be continuously extracted from the ER following synthesis. If these proteins share a generic feature that reflects their normal environment being different to that present in the ER, in this case TMD length, then it would provide a simple means of sorting of many different proteins without the need for specific linear signals.
More generally, the conceptual methodology that we have put forward here could be applied in a wider context to uncover protein localization changes that occur in the absence of any specific gene in the genome. The notion of the “traffickome” could be extended to other trafficking events such as retrograde Golgi to ER traffic, Golgi to plasma membrane traffic, or Golgi to vacuole traffic. Hence, by pairing high-throughput genetic manipulations with a microscopic output it is now possible to study basic questions of specificity and promiscuity in cell biology that have previously been difficult to tackle.
Cultures were grown at 30°C in either rich medium (1% Bacto-yeast extract [BD], 2% Bacto-peptone [BD], and 2% dextrose [Amresco] or synthetic [S] minimal medium [0.67%] yeast nitrogen base without amino acids [Conda Pronadisa)] and 2% dextrose) containing the appropriate supplements for plasmid selection. When necessary, dextrose was replaced by galactose (2%; Amresco) or raffinose (2%; Amresco). For galactose induction, overnight cultures in SD, SD-LEU, or SD–URA cells were diluted 1/10 and grown at 30°C to early log phase in SD or SD–URA medium, then washed and resuspended in 2% galactose-containing SG or SG–URA medium for 2 h. For pulse-chase lysates, the first time point was obtained directly from the galactose culture. Cells were resuspended in glucose-containing medium for chase time points. When needed as selection markers, G418 (200 µg/ml; Calbiochem) or Nourseothricin (Nat) (200 µg/ml WERNER BioAgents) were added. In cases where G418 was required in a SD-based medium, yeast nitrogen base without ammonium sulfate (Conda Pronadisa) was added and supplemented with mono-sodium glutamate (Sigma) as an alternative nitrogen source. Manipulations of plasmid DNA were performed in Escherichia Coli strains DH5α and TOP10. A complete list of plasmids used in this study can be found in Table S2.
All yeast strains in this study are based on the BY4741 laboratory strain [55]. General laboratory strains and strains created in this study are listed in Table S3. Unless otherwise stated, strains harboring a deletion in a specific ORF were taken from the yeast deletion library [11], while strains harboring a hypomorphic allele of an essential gene were taken from the DAmP library [24]. Strains harboring an ORF endogenously tagged with GFP in its C terminus were taken from the yeast GFP library [15]. Genomic modifications and introduction of plasmid DNA were done as previously described [56]. YMS792, YMS793, and YMS954 were created by targeting the erv14, erv15, and emp24 genes, respectively, for disruption with the kanR gene with pFA6a-KanMX6 [57]. MID2 was cloned by PCR in frame with a GAL1 promoter and monomerized GFP (A207K) into a modified version of pRS416 between HinDIII and Xho I. AclI, SpeI, and BglII sites were introduced into the MID2 sequence to facilitate cloning of overlapping oligonucleotides encoding polyleucine stretches to replace the Mid2 TMD.
Integration plasmids to express TMD chimeras of Mid2-EGFP under the GAL1 promoter or MID2 promoter were constructed as follows: MID2 promoter-NatMX-GAL1 promoter-MID2(chimera)-EGFP-MID2 terminator - in pBluescriptII(KS-). Homologous recombination was performed using a unique SnaB1 site in the MID2 gene for expression from the endogenous promoter, or using the unique sites HpaI and Blp1 for expression from the GAL1 promoter. All constructs were sequenced. All genetic manipulations were performed using the Traffo method for transforming yeast strains [56], and deletions were verified using check PCR to assay for loss of the endogenous gene copy. For a complete list of primers used see Table S4.
For this work we assembled two “mini libraries” by choosing strains of interest from the above commercially available yeast libraries. First we chose 379 strains that represent a variety of possible cargo (the cargo library) from the GFP library in which each ORF is C-terminally tagged with GFP, thus enabling the visualization of the sub-cellular localization of a protein under control of its own promoter [15]. To assemble the library, we hand-picked all possible cargo proteins—those which had been visualized as being localized to either the plasma membrane, Golgi apparatus, vacuolar membrane, vacuolar lumen, COPI vesicles, COPII vesicles, peroxisomes, adiposomes, or endosomes. In addition, we added all proteins that had an undefined punctate localization (a full list of selected cargo strains is available in Table S5). The initial array was visualized and only strains displaying a strong and correctly localized GFP signal were put into the final array. The second library contained strains mutated in ER to Golgi trafficking proteins (the trafficking library), either from the yeast deletion library that contains deletions of all non-essential proteins [11] or from the DAmP library that contains hypomorphic alleles of the essential ones (for a full list of strains included see Table S6) [24].
All genetic manipulations were performed using SGA techniques to allow efficient introduction of a trait (mutation or marker) into systematic yeast libraries. SGA was performed as previously described [13],[14],[23],[58]. Briefly, using a RoToR bench-top colony arrayer (Singer Instruments) to manipulate libraries in high-density formats (384 or 1,536), haploid strains from opposing mating types, each harboring a different genomic alteration, were mated on rich media plates. Diploid cells were selected on plates containing all selection markers found on both parent haploid strains. Sporulation was then induced by transferring cells to nitrogen starvation plates. Haploid cells containing all desired mutations were selected for by transferring cells to plates containing all selection markers alongside the toxic amino acid derivatives canavanine and thialysine (Sigma-Aldrich) to select against remaining diploids. Each SGA procedure was validated by inspecting representative strains for the presence of the GFP-tagged cargo and for the correct genotype using check PCR (primer sequences can be found in Table S4).
Microscopic screening was performed using an automated microscopy set-up as previously described [14]. Briefly, cells were moved from agar plates into liquid 384-well polystyrene growth plates using the RoTor arrayer. Liquid cultures were grown overnight in SD medium, with appropriate auxotrophic selections where applicable, in a shaking incubator (LiCONiC Instruments) in 30°C. A JANUS liquid handler (Perkin Elmer), which is connected to the incubator, was used to back-dilute the strains into plates containing the same medium, after which plates were transferred back to the incubator and were allowed to grow for 3.5 h at 30°C to reach logarithmic growth. The liquid handler was then used to transfer strains into glass bottom 384-well microscope plates (Matrical Bioscience) coated with Concanavalin A (Sigma-Aldrich) to allow formation of a cell monolayer. Wells were washed twice in medium to remove unconnected cells and plates were transferred into an automated inverted fluorescent microscopic ScanR system (Olympus) using a swap robot (Hamilton). The ScanR system is designed to allow auto focus and imaging of plates in 384-well format using a 60× air lens and is equipped with a cooled CCD camera. Images were acquired at excitation at 490/20 nm and emission at 535/50 nm (GFP). After acquisition images were manually reviewed using the ScanR analysis program. Images were processed by the Adobe Photoshop CS3 program for slight contrast and brightness adjustments.
Manual Microscopy was performed using either one of two systems: for Figures S3 and S4 we used an Olympus IX71 microscope controlled by the Delta Vision SoftWoRx 3.5.1 software with ×100 oil lens. Images were captured by a Phoetometrics Coolsnap HQ camera with excitation at 490/20 nm and emission at 528/38 nm (GFP) or excitation at 555/28 nm and emission at 617/73 nm (mCherry/RFP). Images were transferred to Adobe Photoshop CS3 for slight contrast and brightness adjustments.
For Figures 4–8 we used a 100×1.49 NA objective on a Nikon Eclipse TE2000 epifluorescent microscope using a CCD camera (CoolSNAP-HQ2, Roper Scientific) and RFP and GFP filters (Chroma Technology). Images were acquired and analyzed using MetaMorph and ImageJ, and normalized using Adobe Photoshop. For some co-localization studies with Golgi markers both channels were imaged simultaneously using a beam splitter (Cairn Research). For fusions expressed under the MID2 promoter strains were grown in synthetic complete medium to reduce background fluorescence.
For protein purification during the galactose-induced pulse chases we first added 3 OD600 of cells to NaN3 (t = 0). For subsequent time points 1 ml of cells were collected. Cells were resuspended in 500 µl NaOH solution (0.2 M NaOH, 0.2% β-mercaptoethanol) and precipitated in 5% trichloroacetic acid. Pellets were resuspended in sample buffer and 10 µl Tris base. After electrophoresis on 4%–20% gradient gels (Novex, Invitrogen), immunoblots were blotted with mouse anti-GFP (7.1/13.1, Roche), HRP anti-mouse, and ECL (Amersham). For quantitation purposes, gel lane profiles were obtained from scanned autoradiograms and peak areas were determined using ImageJ. The ratio of the ER peak to the sum of the ER, post-ER, and free-GFP peaks was calculated and normalized so that it was 1.0 at the start of the chase (t = 0).
|
10.1371/journal.pntd.0000416 | Optimization of Control Strategies for Non-Domiciliated Triatoma dimidiata, Chagas Disease Vector in the Yucatán Peninsula, Mexico | Chagas disease is the most important vector-borne disease in Latin America. Regional initiatives based on residual insecticide spraying have successfully controlled domiciliated vectors in many regions. Non-domiciliated vectors remain responsible for a significant transmission risk, and their control is now a key challenge for disease control.
A mathematical model was developed to predict the temporal variations in abundance of non-domiciliated vectors inside houses. Demographic parameters were estimated by fitting the model to two years of field data from the Yucatan peninsula, Mexico. The predictive value of the model was tested on an independent data set before simulations examined the efficacy of control strategies based on residual insecticide spraying, insect screens, and bednets. The model accurately fitted and predicted field data in the absence and presence of insecticide spraying. Pyrethroid spraying was found effective when 50 mg/m2 were applied yearly within a two-month period matching the immigration season. The >80% reduction in bug abundance was not improved by larger doses or more frequent interventions, and it decreased drastically for different timing and lower frequencies of intervention. Alternatively, the use of insect screens consistently reduced bug abundance proportionally to the reduction of the vector immigration rate.
Control of non-domiciliated vectors can hardly be achieved by insecticide spraying, because it would require yearly application and an accurate understanding of the temporal pattern of immigration. Insect screens appear to offer an effective and sustainable alternative, which may be part of multi-disease interventions for the integrated control of neglected vector-borne diseases.
| Chagas disease is the most important vector-borne disease in Latin America. Residual insecticide spraying has been used successfully for the elimination of domestic vectors in many regions. However, some vectors of non-domestic origin are able to invade houses, and they are now a key challenge for further disease control. We developed a mathematical model to predict the temporal variations in abundance of non-domiciliated vectors inside houses, based on triatomine demographic parameters. The reliability of the predictions was demonstrated by comparing these with different sets of insect collection data from the Yucatan peninsula, Mexico. We then simulated vector control strategies based on insecticide spraying, insect, screens and bednets to evaluate their efficacy at reducing triatomine abundance in the houses. An optimum reduction in bug abundance by at least 80% could be obtained by insecticide application only when doses of at least 50 mg/m2 were applied every year within a two-month period matching the house invasion season by bugs. Alternatively, the use of insect screens consistently reduced bug abundance in the houses and offers a sustainable alternative. Such screens may be part of novel interventions for the integrated control of various vector-borne diseases.
| Chagas disease is a major vector-borne parasitic disease in Latin America, with 9.8 to 11 million infected people, 60 million at risk of infection [1],[2] and a disease burden of over 800,000 DALYs [3]. International travel and immigration are also turning it into a global disease [4]. It is caused by the protozoan parasite Trypanosoma cruzi, which is transmitted to humans primarily by triatomine vectors. Due to its importance in public health, vector control strategies have been widely implemented through several regional initiatives in the Americas. These interventions are based on the elimination of domiciliated triatomine vectors by residual insecticide spraying and/or housing improvement, and have resulted in a large reduction in house infestation by triatomines (particularly Triatoma infestans), and a corresponding reduction in Chagas disease transmission to humans [1],[2],[5].
However, it has become increasingly clear that several triatomine species do not establish permanent domestic colonies, but can occasionally infest domestic habitats by immigration from peridomestic and/or sylvatic habitats. These species include Rhodnius prolixus in Venezuela [6], Triatoma brasiliensis and Triatoma pseudomaculata in Brazil [7], Triatoma mexicana in central Mexico [8], or Triatoma dimidiata in the Yucatan peninsula, Mexico and Belize [9],[10].
Extensive field collections of T. dimidiata in both rural and urban areas of the Yucatan peninsula revealed a very clear and reproducible seasonal pattern of transient house infestation by predominantly adult triatomines during April-July, associated with a very limited colonization of domiciles [9], [11]–[14]. These data suggested a seasonal dispersal of adult triatomines from nearby peridomestic and/or sylvatic sites, which was confirmed by the analysis of population stage structure [9] and population genetics studies [15]. Mathematical modelling further revealed that dispersal was the dominant parameter involved in this infestation process, while demography was of secondary importance [16],[17]. Finally, analysis of blood-feeding and fecundity of natural populations suggested that foraging for better host-feeding sources may contribute to the seasonal dispersal of T. dimidiata [18], and while nutritional status and fecundity tended to improve in the houses, these remained largely suboptimal and may thus contribute to ineffective colonization [18]. Accordingly, T. dimidiata populations in the Yucatan peninsula behave as typical source-sink dynamical systems [19],[20], with outdoor habitats as sources and houses as sinks [16]. Another important specificity of these populations is the very low bug abundance observed, which suggests that density dependent process may be of little relevance in the dynamics of the sink habitats [16]. Importantly, variations in this infestation pattern may occur elsewhere as T. dimidiata presents extensive ecological, behavioral and genetic diversity [21]–[23].
The control of house infestation by such non-domiciliated triatomine vectors is identified as a major problem and one of the new challenges for Chagas disease control since conventional spraying control strategies may be of limited efficacy in these conditions [2], [24]–[26]. Insecticide spraying has a rather short-lived effect on house infestation in the case of recurring infestation by immigrating peridomestic and/or sylvatic bugs, as we observed in a previous field study on T. dimidiata vector control in the Yucatan peninsula [27]. It is thus of key importance to improve and optimize the efficacy of current insecticide spraying strategies to cope with (re)infestation by non-domiciliated vectors and to investigate the potential of alternative strategies such as insect screens or bednets [26],[28],[29]. This can be achieved by empirical field trials [30],[31], but this costly approach is limited in the number of control strategies that can be evaluated and the follow-up time required. Alternatively, the use of mathematical modelling has proven to be a very efficient approach to explore control strategies in a variety of contexts and diseases [32]–[35]. Although some modelling studies have investigated vector population dynamics [16],[17],[32],[36] and Chagas disease transmission [37], very few have attempted to optimize control strategies [32] and none focused on non-domiciliated vectors, most likely because of the lack of estimates of the required population parameters in this situation [24],[26].
In the present contribution, we use a combination of field and modelling studies to evaluate the efficacy of several strategies for the control of seasonal infestation by non-domiciliated triatomine populations. We took advantage of one of the best documented case of non-domiciliated triatomine vector; the populations of T. dimidiata in the Yucatan peninsula, Mexico. Our modelling shows that the control of non-domiciliated vectors can hardly be achieved by insecticide spraying, but that insect screens may offer an effective and sustainable alternative.
We aimed to construct a model able 1) to reproduce and predict the temporal variations of vector abundance in the absence of control, and 2) to account for various control strategies. We expanded a previous population dynamics model [16] to include mathematical descriptions of different control strategies such as insecticide spraying, insect screens, and bednets, for their evaluation. The model predicts the temporal variations in vector abundance in one house as a function of survival and fecundity of triatomines inside the house, the immigration of bugs from peridomestic or sylvatic habitats, and the effect of the above control strategies on those parameters. Estimates of the parameters in the absence of control intervention were obtained by fitting the model to a first set of field data corresponding to the observed variations in the average vector abundance inside houses of two villages where no control actions were applied. The predictive value of the model was then tested on a second independent data set, corresponding to the observed variations in vector abundance inside houses of three other villages with no control interventions. This parametrized model, combined with the description of the effect of insecticide on vector survival and fecundity, was then fitted to a third data set from a field control trial to estimate the half-life of the insecticide. We then used the model to explore the efficacy of varying the timing of insecticide application within the year, the frequency of spraying, and the dose of insecticide used. Similarly, we evaluated the effect of insect screens and bednets by performing a complete sensitivity analysis of their possible effects. The efficacy of any given strategy was evaluated as the percent reduction in the abundance of vectors, in comparison with the expected abundance in the absence of control interventions as evaluated from the model. Finally, we performed a sensitivity analysis of the effect of the number of immigrant bugs, the domestic demography of the vector, the half-life and the lethal effect of the insecticide on the efficacy of the various interventions.
Data on the dynamics of house infestation by triatomines in the absence of vector control interventions were collected over 3 years of field studies, from October 1999 to December 2001 and from January to December 2003 [9],[11],[13]. Triatomines were collected by a standardized methodology based on community participation in 5 villages from Northern Yucatan, Mexico (Dzidzilche, Tetiz, Eknakan, Suma and Izamal). Participating families provided oral consent prior to their participation, as written consent was waived because the study involved no procedures for which written consent is normally required outside of the research context. Consent was logged in field notebooks. All procedures were approved by the Institutional Bioethics committee of the Regional Research Center “Dr. Hideyo Noguchi”, Universidad Autonoma de Yucatan. Householders from 5 houses per village were instructed to collect any triatomines present inside their houses, and were then visited every 3 months to take the triatomines to the laboratory. This method has been found to be highly reliable [9],[11],[13] and more sensitive than manual collections in the presence of limited colonization [14],[38]. Four houses from two of these villages (Dzidzilche and Eknakan) were sprayed with a standard dose of 50 mg/m2 of cyfluthrin in November 2000, and monitored every 2 weeks for up to 9 months to detect re-infestation using a combination of manual searches, mouse traps and household collections [27]. All field data were expressed as the average number of bugs collected/house-trimester with 95% confidence intervals.
We modelled the dynamics of a non-domiciliated population of T. dimidiata by using the model of Gourbière et al. [16]. In this model, the egg and larval stages are pooled into a single immature stage, which is then divided into a number of sub-stages of equal duration corresponding to the time step of the model. The underlying assumption is that every individual spends a fixed time as an immature, and the outcome is that immature sub-stages are groups of age classes [39]. The matrix describing the demography of the vector within a house is a Leslie matrix, which we denote A. The model also includes a periodic immigration vector M to mimic the seasonal invasion of vectors observed in the Yucatan peninsula. The overall dynamical system can then be written:(1)where N(n) = (n1(n), n2(n), n3(n), nA(n)) included the number of females in three immature age classes and the number of adult females at the nth time step and M(n) = (m1(n), m2(n), m3(n), mA(n)) the number of immigrants of the same categories (Note that we use index n instead of t as in Gourbière et al. [16] to refer to the main time step of the model, and t describes the smaller time-scale variations in the timing of insecticide spraying in this contribution (see below)). The time step of the model was fixed to 3 months to match model predictions with field data, which were determined every trimester, and to account for the average development time from egg to adult consistent with available data (see [16] for details). Accordingly, individuals of the first, second and third immature age classes are aged [0–90[, [90–180[ and [180–270] days, respectively. Because survival of individuals in these three immature age classes are considered identical, the Leslie matrix takes the form:(2)where SI and SA are survival of immature and adults (probabilities per trimester), and F is female fecundity (female immature offspring per female-trimester). Because only adults immigrate into houses and because this only occurs between April and June [9], M(n) = (0,0,0,M) during the migration period, with M being the number of adult female immigrants, and M(n) = (0,0,0,0) during the remaining of the year.
Because the time unit desired to describe the control strategies in a flexible way is much shorter than the three-month time step previously selected, we adapted the above model to account for a daily description of the population dynamics, while keeping the three-month time step of the model. We divided each time step into T = 90 time units (t) and considered that immigrating individuals survive and reproduce proportionally to the time spent in the domestic habitat since their arrival at time τ. The population dynamics model is then divided into two parts, one describing the demography of individuals present in the domestic habitat since the beginning of the time step, and one accounting for the demography of individuals arriving at each time unit of the time step:(3)L(n,τ) are Leslie matrices similar to L, but set up from survival SI(n,τ), SA(n,τ) and fecundity F(n,τ) defined over the time T-τ spent in the domestic habitat within the nth time step. Similarly, M(n,τ) includes the number of immigrants at time τ of the nth time step. We then used Equation 3 to simulate the vector population dynamics with or without control by changing the definition of parameters SI(n,τ), SA(n,τ), F(n,τ) and M(n,τ) according to the control strategies to be considered and the assumptions about their effects on vector demography. Finally, bug collection over the time steps was incorporated by removing a percentage p of individuals at the end of each day. The removed insects were summed over the duration of the time step to obtain a number of collected bugs/house-trimester, which is the model outcome that we compared to field observations. The best fits were obtained for p values 1–10%, with very limited changes in the quality of predictions over this range. For consistency, we thus display all our results for p = 5%.
The model's demographic parameters were first fitted to two years of field data from two villages in the absence of vector control interventions. The optimal parameter values were M = 21.1 immigrants/year, SI = 1/trimester, SA = 0.434/trimester, F = 0.224 female offsprings/female-trimester, and these provided a very good fit of the model to field data for the total bug population (R2 = 0.953, Fig. 1A). This corresponded to a domestic population growth rate of λ = 0.83. In agreement with a previous estimate of λ = 0.20 obtained for another population [16], this confirmed that houses can truly be considered as sinks since λ<1 [19]. All the demographic parameter values were similar to those determined in our previous model [16], except for the survival of immatures. The unrealistically high value obtained is explained by the very low number of immatures in the population, resulting in a negligible weight to S in the overall quality of the fit. Using an immature survival probability of 0 only changed the least square value associated to the fit by 4.6%, whereas decreasing the amount of immigration to M = 1 lowered the quality of the fit by 2256%. This corroborated previous sensitivity analysis, where the effect of SI was found to be 7 to 8 orders of magnitude lower than the effect of M (with Sobol standardized indices equal to 0.000005 and 0.89, respectively) [16]. We further tested the predictions of the model by comparing them with 3 years of independent field data from three other villages, which confirmed its very good predictive value to reproduce the observed seasonal variations in triatomine population (R2 = 0.891, Fig. 1B). All further calculations presented in this study were performed using demographic parameter values providing the best fit, but similar results were obtained when immature survival probability was forced to zero (data not shown). Insecticide spraying was then introduced into the model by reducing bug survival and fecundity values in a dose-dependent manner, and the model output was fitted to field data from a pilot trial to estimate insecticide half-life. The best fit of the model (R2 = 0.985, Fig. 1C) was obtained for a half-life of 38 days, which is in good agreement with the expected and measured half-life of pyrethroids and a lethal residual effect of about 3 months [40],[43],[44].
Once we determined the model's parameters that best fitted field data, we predicted domestic bug abundance as a function of time after various control interventions. We first explored the effect of the timing of insecticide spraying during the year. The effects of a single insecticide spraying (50 mg/m2 at various dates) on bug abundance in the houses was only observed for a few months, and was followed by a rapid return to a normal cycle of infestation as soon as a new season of infestation occurred (Fig. 2A). Also, the timing of spraying during the year was critical for the magnitude of the reduction in bug abundance post-intervention (Fig. 2A and 2B). A maximum reduction in triatomine abundance of 90% for one year was achieved when spraying was conducted at the beginning of April, just before the start of the seasonal infestation. However, this maximum effect was only obtained for a very narrow time window, and efficacy dramatically decreased when spraying was applied before or after this period (Fig. 2B). Insecticide spraying had negligible effects (<5% reduction in bug abundance) when applied between August and December.
Although a standard cyfluthrin dose of 50 mg/m2 is commonly used for triatomine control [25], we evaluated the effect of varying this dose when the application is performed at the optimal time (April). A four-fold increase in insecticide dose (200 mg/m2) only provided a limited improvement in the reduction of bug abundance compared with the standard dose, and was not enough to sustain triatomine control for more than one seasonal infestation cycle (Fig. 2C and 2D). The standard dose of 50 mg/m2 thus provided a nearly optimal vector control. Nonetheless, an insecticide dose as low as 10 mg/m2 sprayed at the beginning of the infestation period was still able to reduce bug population by over 50% for a year (Fig. 2C and 2D).
Because a single insecticide spraying did not allow to achieve a sustainable vector control, we then evaluated the effects of repeated spraying and determined the optimal frequency of application. Our simulations clearly indicated that spraying once a year, just before the start of house invasion by adult triatomines, was the best strategy (Fig. 2E and 2F). Less frequent spraying led to a poor control during the seasons without insecticide application, whereas more frequent spraying did not increase the efficacy of the spraying.
Although our insecticide half-life estimate was in good agreement with expected values, we evaluated the robustness of the results using various half-life values in simulations where 50 mg/m2 are applied with various frequency at the optimal timing (April 1st). As expected, increasing insecticide half-life allowed for a more sustained vector control, leading to about 80% reduction in bug abundance by spraying every two years instead of one. However, a half-life of over 4 months was required for such a frequency of spraying to be effective (Fig. 3A). Similarly, the importance of the timing of insecticide application during the year decreased with longer half-life (Fig. 3B and 3C). Yearly interventions can be performed at any time when spraying insecticide with a half-life of over 4 months (Fig. 3B), but when spraying is conducted every two years, the timing of intervention still has to be considered even for insecticides with the highest half-life (Fig. 3C). Nonetheless, all our initial predictions remained valid for an insecticide half-life shorter than 2 months, for which the optimal strategy required yearly insecticide spraying during a narrow time window, just before the start of the seasonal house invasion by triatomines. These conclusions were valid for a wide range of LD50 of the insecticide, provided the spraying dose is adjusted accordingly, regardless of the level of immigration considered (Table 1, data not shown). Interestingly, the results of the above sensitivity analysis were found similar when considering the demographic parameter estimates from Gourbière et al., [16]. Our main conclusions on strategies of insecticide spraying thus hold for a wide range of non-domiciliated population dynamics because the two population growth rates tested, λ = 0.2 and λ = 0.83, cover most of the range of population growth rate corresponding to the definition of sink population, i.e., 0<λ<1.
Given the importance of dispersal in triatomine population dynamics inside houses, we evaluated the effect of the presence of insect screens on doors and windows by reducing the immigration of bugs inside houses. Reduction in triatomine abundance in the houses was immediate following screens implementation, directly proportional to the reduction in bug immigration rate, and sustained for as long as the screens were maintained (Fig. 4A and 4B). We also simulated the use of non-impregnated bednets by considering that these reduced bug feeding, and thus bug survival and fecundity. While the effect of such bednets was sustained for as long as they were used, a reduction in bug survival and fecundity of up to 90% only accounted for a reduction in bug abundance of about 30% over a year. Smaller effects on survival and fecundity resulted in even smaller effects on bug abundance. The estimated efficacy of insect screens and bednets did not depend on the level of immigration considered and varied only slightly with the demographic parameters. (data not shown).
The integrative studies performed in the Yucatan peninsula provide a rare opportunity to develop mathematical models rooted in several years of field data. It was used here for the first time in an attempt at optimizing control strategies for non-domiciliated vectors of Chagas disease. The quality of the fit and of the predictive value of our deterministic model allowed to produce reliable simulations of a variety of control interventions. Also, while stochastic variations in the number of immigrants, which ultimately determine the number of individuals present in a given house, were not considered in our model, these are unlikely to qualitatively affect our results as indicated by our sensitivity analysis of immigrant numbers.
Simulations aimed at optimizing insecticide spraying clearly indicated that efficacy depended dramatically on the timing and frequency of spraying, both of which had to match closely the immigration season. This implies that a good understanding of the temporal pattern of immigration, which may differ between non-domiciliated triatomine species [6]–[8],[24],[38], is required for optimal control. On the other hand, variations in birth and death rates between individual genotypes or between species of vectors seems of little relevance to tune the optimal strategy of control for such sink populations. As long as the number of immigrant adult triatomines is controlled effectively, there remains virtually no individuals inside the houses after immigration, so that variations in the ability of these remaining insects to reproduce and survive inside the houses has only a minor impact on the percentage of reduction of their year-round abundance. In the case of T. dimidiata in the Yucatan peninsula and current pyrethroids, which have a half-life shorter than 2 months and have thus a residual lethal effect of about 1–6 months depending on the substrates [40],[43], a reduction of at least 80% in bug abundance would require yearly applications within a very short time window (March or April). While this may be feasible on a small scale, implementing such a control strategy on a large scale would require unrealistic logistics and a large cost of money. For example, based on a spraying capacity of 6–10 houses/day by a team of 2 people, spraying the ∼200,000 rural houses of the state of Yucatan in less than 2 months would require the simultaneous work of 400–650 teams during that time, together with a timely supply of insecticide in each village. Using an insecticide with a half-life >4 months would allow to either reduce spraying frequency to once every two years, or spray at any time of the year every year. It is interesting to note that the key factor for insecticide optimization against non-domiciliated triatomine is the half-life of the insecticide rather than its lethal effect or initial dose. This contrasts with the control of domiciliated triatomines, for which effectiveness of pyrethroids rests more on their initial impact rather than their residual effect [25]. Thus, while third-generation pyrethroids seem to be particularly adapted for the control of domiciliated triatomines [25], alternative insecticides with longer half-life such as fipronil [45], bifenthrin [44], or even the previously discarded organochlorines [25] may be more appropriate for the control of non-domiciliated bugs. However, their use may require strict management to avoid undesirable environmental and health impact, as well as the development of insecticide resistance, as already observed in some populations of triatomines [46]–[48].
Our results clearly indicate that none of these insecticide spraying interventions would be sustainable, since as soon as they are interrupted, re-infestation by new immigrant bugs occurs during the next season, implying large costs associated with repeated spraying. Some authors even suggested that control of non-domiciliated triatomines should not be considered, and that resources should rather be devoted to patient detection and care [2]. Nonetheless, alternative strategies may provide a more appropriate level of vector control.
Our simulations of insect screen effects indicate that an effective and sustained control can be achieved when a significant reduction of bug immigration is obtained. While it is difficult to estimate the possible efficacy of such screens in the field, an exclusion of over 90% of other insects has been observed with some greenhouse screens [49]. Also, a pilot field study of impregnated curtains used as a chemical barrier against non-domiciliated R. robustus resulted in a >60% reduction in live bugs collected over one month [30]. Our results are also consistent with the identification of such insect screens as a major protective factor against house infestation by T. dimidiata in urban Merida in the Yucatan [12]. A range of efficacy of insect screens of 70–90% would thus be very comparable to that of a yearly application of pyrethroids, but sustainable and hence less expensive. Even though our model did not take into account a decrease in efficacy of insect screens due to progressive tear-and-wear, it seems reasonable to consider that they would be effective for several years.
On the other hand, we found that bednets had little effect on bug abundance, possibly because triatomine reproductive output inside houses is already low in the absence of interventions [16],[18]. However, the potential of bednets cannot be ruled out from our results, since reduction in vector-human contacts, and thus parasite transmission, is not taken into account in our model, but has been observed in other settings [31],[50],[51]. Also, a number of additional vector control intervention have not directly been tested in this study, but their outcome can be predicted from our results. For example, insecticide-impregnated insect screens and bednets should reduce bug abundance, and their sustainability would depend on the half-life of the insecticide used for impregnation.
In conclusion, our study illustrates well the usefulness of coupling modelling and field studies to design and optimize effective control interventions and develop evidence-based public health policies, as previously done for the control of other diseases [33]–[35]. Our results clearly indicate that pyrethroid spraying is of limited usefulness for the control of non-domiciliated triatomines, while insect screens may be a simple, cost-effective and sustainable intervention. In addition, such screens would have an effect on all vector-borne diseases present, such as dengue, malaria or leishmaniasis [51],[52], and would thus be an excellent example of a high impact multi-disease intervention for the integrated control of neglected diseases [53]. Further field evaluations of the best vector control strategies identified here are warranted to confirm their efficacy and provide information on their implementation, including acceptability by the community and costs.
|
10.1371/journal.pntd.0002114 | Serovar Diversity of Pathogenic Leptospira Circulating in the French West Indies | Leptospirosis is one of the most important neglected tropical bacterial diseases in Latin America and the Caribbean. However, very little is known about the circulating etiological agents of leptospirosis in this region. In this study, we describe the serological and molecular features of leptospires isolated from 104 leptospirosis patients in Guadeloupe (n = 85) and Martinique (n = 19) and six rats captured in Guadeloupe, between 2004 and 2012.
Strains were studied by serogrouping, PFGE, MLVA, and sequencing 16SrRNA and secY. DNA extracts from blood samples collected from 36 patients in Martinique were also used for molecular typing of leptospires via PCR. Phylogenetic analyses revealed thirteen different genotypes clustered into five main clades that corresponded to the species: L. interrogans, L. kirschneri, L. borgpetersenii, L. noguchi, and L. santarosai. We also identified L. kmetyi in at least two patients with acute leptospirosis. This is the first time, to our knowledge, that this species has been identified in humans. The most prevalent genotypes were associated with L. interrogans serovars Icterohaemorrhagiae and Copenhageni, L. kirschneri serovar Bogvere, and L. borgpetersenii serovar Arborea. We were unable to identify nine strains at the serovar level and comparison of genotyping results to the MLST database revealed new secY alleles.
The overall serovar distribution in the French West Indies was unique compared to the neighboring islands. Typing of leptospiral isolates also suggested the existence of previously undescribed serovars.
| Leptospirosis is an emerging zoonotic disease caused by infection with pathogenic strains of Leptospira. Isolation of Leptospira strains is rare, making it difficult to assess their distribution worldwide. In this study, we characterized cultures of Leptospira obtained from more than one hundred leptospirosis patients from the French West Indies by serology and various molecular typing methods to identify the strains circulating in this endemic region. Typing of leptospiral isolates showed that causative agents of leptospirosis in the French West Indies are mainly from the serogroups Icterohaemorrhagiae and Ballum, but we also identified new genotypes. We also found that the distribution of the predominant pathogenic leptospiral serovars differed between the Caribbean islands. A better understanding of the epidemiology of leptospirosis will improve our knowledge in the distribution of this emerging neglected tropical disease worldwide. The identification of the circulating etiological agents of leptospirosis in the French West Indies will also help establish appropriate control and prevention measures in this area where the disease is endemic.
| Leptospirosis is an emerging zoonosis with a worldwide distribution. The World Health Organization (WHO) estimates that there are over 1,700,000 severe cases of leptospirosis worldwide, with an higher incidence in impoverished populations in developing countries and tropical regions [1]–[3]. The disease is transmitted during direct contact with animal reservoirs or, more frequently, water and soil contaminated with their urine [4]. Leptospirosis is found in rural regions because of the higher risk of exposure to animal reservoirs [5], [6] and also in urban slums where inadequate sanitation provides the conditions for rat-borne transmission of the disease [7], [8]. Outbreaks may occur after heavy seasonal rainfall and extreme climatic events such as tropical storms and hurricanes [8], [9]. Leptospirosis causes a broad spectrum of symptoms from subclinical infection to multiple organ failure with a mortality rate of 10 to 50% [3].
Leptospirosis is one of the most important neglected tropical bacterial diseases in Latin America and the Caribbean [10]. This includes the French West Indies which consist of the Caribbean islands of Guadeloupe and Martinique, both French overseas departments. Climate-related changes, such as El Niño events, and periods of heavy rainfall may influence the incidence of leptospirosis in this region [11], [12]. The mean incidence of leptospirosis in the French West Indies is higher than 10/100,000 inhabitants peaking at ∼39/100,000 inhabitants in Guadeloupe in 2011, which is 100 times higher than that in Mainland France (data from the National Reference Center for Leptospirosis, Institut Pasteur).
Pathogenic Leptospira encompass nine species with more than 300 serovars which are the etiological agents of leptospirosis [13]. The taxonomy of the genus Leptospira has been both complex and controversial. Leptospiral serovars are defined by the cross agglutinin absorption test (CAAT) which uses polyclonal antibodies against the lipopolysaccharides (LPS) [4], [14]. However, this test is fastidious to perform and it is restricted to a few reference laboratories. Although the term serogroup has no taxonomic value, it has been used to define a group of antigenically related serovars which can be identified by microscopic agglutination test (MAT). Advancement in molecular techniques has allowed the speciation of members of the genus Leptospira. A significant outcome of the genetic classification scheme was the finding that a serovar may belong to different species [14]. With the emergence of molecular typing methods, it appears that the concept of a “serovar” is no longer fully satisfactory as it may fail to define epidemiologically important strains or genotypes.
Despite its medical significance, the isolation of clinical Leptospira strains is rare due to the fastidious growth in culture of this species, and poor awareness of the disease. Detailed characterization of Leptospira isolates is important for understanding the epidemiology of leptospirosis. Leptospira serovars can be prevalent in a particular geographical area and/or associated with a restricted number of animal reservoirs. Local Leptospira isolates can serve as antigens for the serodiagnosis of leptospirosis. The diverse distributions of Leptospira serovars and genotypes may have implications for vaccine design and efficacy.
The main function of the National Reference Center (NRC) for Leptospirosis, which is also a WHO Collaborating Center, at the Institut Pasteur of Paris is the surveillance of human leptospirosis. This includes the collection of diagnostic data from laboratories around the country, including French overseas territories. The NRC is the only laboratory in France that can confirm leptospirosis diagnosis by means of the MAT, which remains the gold standard for the serological diagnosis of leptospirosis, with an extended panel of antigens. The NRC also identifies clinical isolates both from mainland France and from French overseas territories. As part of a long-term typing project, more than three-quarters of all the Leptospira isolates received from Guadeloupe and Martinique were systematically fingerprinted so as to identify the strains circulating in this region the Americas.
The Leptospira cultures from human patients analyzed in this study were previously isolated by the University Hospitals of Pointe à Pitre (Guadeloupe) and Fort de France (Martinique) and the National Reference Center for Leptospirosis (Institut Pasteur) as part of the national surveillance of leptospirosis. The strains and DNAs derived from these cultures were analyzed anonymously for this research study. All serum samples were initially sampled for diagnostic purpose, and archived at the National Reference Center for Leptospirosis (Institut Pasteur). All sera are de-linked from the patients from whom they originated and analyzed anonymously if used in any research study. The study protocol was approved by the ethical committees of the University Hospitals of Pointe à Pitre (Guadeloupe) and Fort de France (Martinique) and the CNIL (Commission Nationale Informatique et Liberté). This study was part of a protocol approved by the Institut Pasteur and the French Ministry for Education & Research French Ministry for Education & Research (protocol # AC- DC-2010-1197). Rats are listed as invasive mammals on the French West Indies [15] and authorization (through arrêté préfectoral) are regularly published, in agreement with the « Fédération Départementale des Groupements de Défense contre les Ennemis des Cultures » for their captures. Rats were captured in 2002–2003 during a study on schistosomiasis [16], which was supported by the CNRS (PNDBE), the MENRT (PRFMMIP 95) and the French Ministry of Ecology and Sustainable Development (Contract CV 02000071, MEDD, programme “Ecosystemes Tropicaux”). Capture and euthanasia of rats was performed by Dr. A. Theron (University of Perpignan) who was accredited to experiment on rodent (authorization n° C 66.11.02 by the Préfecture des Pyrénées Orientales). Studies on rats were performed in accordance with the European Union legislation (Directive 86/609/EEC).
Serum samples from suspected cases of leptospirosis were subjected to the microscopic agglutination test (MAT) at the National Reference Center for Leptospirosis (NRC) at the Institut Pasteur (Paris, France). MAT was performed using 24 leptospiral antigens (Table S1). A high agglutination titer of the serum with one particular serogroup is taken to identify the presumptive serogroup of the infecting bacterium. For patients presenting symptoms during the first week of infection, total genomic DNA was extracted from plasma collected into EDTA tubes and tested for the presence of pathogenic Leptospira by real-time PCR [17]. Molecular Biology Grade Water (EUROBIO, Les Ulis, France) was used for PCR. Reactions with no template DNA were included as negative controls in each PCR experiment. For patients testing positive by PCR, acute and, if possible, convalescent serum samples were collected for serological testing.
Leptospirosis cases were defined as having clinical signs and symptoms consistent with leptospirosis and a single MAT titer ≥1/400 for a pathogenic serogroup or detection of pathogenic Leptospira by PCR or culture.
A total of 104 clinical isolates of Leptospira isolated from patients in Guadeloupe (85) and Martinique (19) between 2004 and 2012 were studied. Leptospira was cultured by inoculating plasma prepared from heparinized blood from patients into EMJH liquid medium [18] at the University Hospitals of Pointe à Pitre (Guadeloupe) and Fort de France (Martinique). Leptospires positive cultures were then sent to the NRC for Leptospirosis (Institut Pasteur, Paris, France) for typing. Six isolates collected from kidney tissues of Rattus rattus captured in Guadeloupe (mangrove area of Morne à L'eau) in 2002–2003 (André Théron, University of Perpignan) were also included in the study. Reference strains from the collection maintained by the NRC for Leptospirosis were used for comparisons (Table S2). DNA was also isolated from the blood of 36 additional patients from Martinique who tested positive for leptospirosis by PCR during the study.
The microscopic agglutination test (MAT) was used for antigenic characterization of Leptospira isolates, with a standard battery of rabbit antisera against reference serovars representing the 24 serogroups as previously described [18]. Mice monoclonal antibodies F70 C14-10 and F70 C24-20 (WHO/FAO/OIE and National Collaborating Centre for Reference and Research on Leptospirosis, Royal Tropical Institute, Amsterdam, The Netherlands), which react against the serovars Icterohaemorrhagiae and Copenhageni respectively, were also used for some strains as previously described [19].
Genomic DNA was extracted from EMJH cultures or from human plasma (see above). DNA was amplified using Taq polymerase (GE Healthcare) under standard conditions. For species identification, the rrs gene was amplified with the primers LA (5′-GGCGGCGCGTCTTAAACATG-3′) and LB (5′-TTCCCCCCATTGAGCAAGATT-3′), and when necessary, by nested primers LC (5′-CAAGTCAAGCGGAGTAGCAA-3′) and RS4 (5′- TCTTAACTGCTGCCTCCCGT-3′) [20], [21]. Part of the secY gene was amplified with the primers F (5′-ATGCCGATCATTTTTGCTTC-3′) and R (5′-CCGTCCCTTAATTTTAGACTTCTTC-3′) [22]. Sequencing was performed at the Genotyping of Pathogens and Public Health Platform (Institut Pasteur, Paris, France). All molecular epidemiological data were stored and analysed with Bionumerics software (Version 6.5; Applied-Maths, Belgium). Genotyping was also performed by multiple-locus variable-number tandem repeat analysis (MLVA) using the loci VNTR4, VNTR7, and VNTR10 as described by Salaun et al. [23]. In the absence of PCR products, a second round of nested PCR amplification was performed with the inner primers NP 4A (5′-TTGGAGCGCAATCTCTTTTT-3′) and NP4B (5′- TGAGGATACCCCATTTTTACCTT-3′), NP7A (5′-GATGGGCGGAGAAAAGTGTA-3′) and NP7B (5′-TGGATCGGTATTTTGGTTCA- 3′), NP10A (5′-ATTCCAAAACTCAGCCCTCA-3′) and NP10B (5′- TGATGGGATTACCGGAAGAA-3′). For pulsed-field gel electrophoresis (PFGE), cells were embedded in agarose plugs as previously described [24], and the DNA in the plugs digested with NotI. PFGE was performed in a contour-clamped homogeneous electric field DRII apparatus (Bio-Rad Laboratories, Richmond, CA). Restriction fragments were resolved with ramping from 5 to 60 s for 50 h, 1 to 30 s for 40 h, or from 1 to 70 s for 36 h at 6 V/cm.
Nucleotide sequences have been deposited with GenBank under accession numbers JX827500 - JX827597.
Guadeloupe and Martinique are islands situated in the Caribbean archipelago and are 100 miles apart. Guadeloupe and Martinique share common geological environments (although Grande Terre in Guadeloupe is composed of limestone, the islands are mainly volcanic) and are 1,705 and 1,100 km2, respectively. They have similar population sizes (<400,000 inhabitants) and levels of urbanization. The islands are among the most highly developed islands in the Caribbean and their economies depend largely on tourism and agriculture (sugar cane and bananas). The climate is tropical with two distinct seasons, the dry season from December to May and the rainy season from June to November.
Over the last five years (2007–2011), the annual incidence of leptospirosis has ranged from <12 per 100,000 inhabitants in Martinique (2007) to >41 per 100,000 inhabitants in Guadeloupe (2011) (data from the NRC for Leptospirosis, Institut Pasteur, France), which is among the highest reported in the Caribbean (<2 per 100,000 inhabitants in Trinidad and Tobago [25], [26] and <13/100,000 inhabitants in Barbados [27]).
In 2011, the total number of cases was 165 in Guadeloupe and 142 in Martinique, which is two to three-fold more than in 2007. Most of the infections were during the rainy season from August to November (around 70% of all cases in 2011).
Detection of antibodies in patient sera by MAT has shown that the most prevalent Leptospira serogroup in the French West Indies is Icterohaemorrhagiae (<25% in Martinique and <37% in Guadeloupe). The other serogroups each account for less than 12% of cases and include serogroups Ballum (<5% in Martinique and <12% in Guadeloupe), Sejroe (<7% in Martinique and <5% in Guadeloupe), and Canicola (<7% in Martinique and <9% in Guadeloupe) (data from the NRC for Leptospirosis, Institut Pasteur, France).
Results of identification of strains sent to the NRC for Leptospirosis (Institut Pasteur) for serogroup and genotype identification are shown in Table 1 (Table 1). Genomic DNA from 36 acute-phase blood samples that were positive for pathogenic Leptospira by PCR at the University Hospital of Fort de France were also included in this study (Table 2). The geographical distribution of the isolates was as follows: 91 strains isolated in Guadeloupe from 2003 to 2012 (including 6 rat isolates) and 55 strains isolated in Martinique from 2011 to 2012 (including DNA from 36 patients) (Tables 1 and 2).
Serogrouping of isolates was first performed with rabbit antisera against reference serovars. The most frequent serogroups were Icterohaemorrhagiae (58%) and Ballum (25%), consistent with the findings obtained by MAT with human serum samples (see above). Other serogroups detected include Mini (9 isolates), Tarassovi (4 isolates), Australis (1 isolate), and Celledoni (1 isolate). Four isolates scored negative (no agglutination) with the antisera raised against the 24 serogroups. A selection of isolates from the serogroup Icterohaemorrhagiae were subsequently typed to serovar level by MAT with monoclonal antibodies (MAbs) against the serovars Icterohaemorrhagiae and Copenhageni: both serovars were present among the clinical isolates (data not shown).
Molecular typing was then performed by sequencing the 16S rRNA gene (rrs) in genomic DNA from 110 cultures and 36 acute-blood samples [20]. All the samples corresponded to one of five pathogenic species: L. interrogans (44 samples), L. kirschneri (36 samples), L. borgpetersenii (38 samples), L. noguchi (3 samples), and L. santarosai (22 samples) (Tables 1 and 2). Two samples were phylogenetically related to L. kmetyi (Figure 1). The sequences of their 279-nucleotide 16S rRNA PCR products were identical with two mismatches (99% nucleotide identity) to the corresponding variable region of the 16S rRNA sequence of the L. kmetyi reference strain. These L. kmetyi-positive cases showed MAT cross reaction with the saprophyte serovar Patoc (Table 2). The serovar Patoc, which is non-pathogenic, was included in our analysis because it has cross-reactivity with pathogenic serogroups and can be indicative of an infection. The last sample (201102109) was related to both L. kmetyi (273/279 nucleotides) and L. kirschneri (272/279 nucleotides). This DNA may therefore correspond to a variant of L. kmetyi or L. kirschneri.
PFGE has long been the gold standard method for genotyping Leptospira strains [28], [29]. PFGE analysis of NotI-digested genomic DNA revealed at least thirteen distinct patterns for the typed isolates (Figure 2). For each strain, serovar designation was attributed by comparing the patterns with those of reference strains belonging to the identified serogroup and species (Table S2). For example, patterns of isolates identified as belonging to the species L. santarosai and serogroup Mini were compared with reference serovars that belong to the L. santarosai serogroup Mini (i.e. serovars Beye, Georgia, Szwajizak, and Tabaquite). In this case, the “Mini” isolates displayed a PFGE pattern which was similar (less than three band differences) to the type strain of serovar Tabaquite (Figure 2). The “Tarassovi” isolates displayed unique PFGE patterns which were different from the PFGE patterns of the reference strains of L. borgpetersenii and L. santarosai serogroup Tarassovi (serovars Kisuba, Tarassovi, Kanana, Guidae, Tunis, Yunxian, Atchafalaya, Atlantae, Bravo, Chagres, Darien, Navet, Rama, and Sulzerae). The PFGE profile of the “Australis” isolate was similar to serovar Bajan and distinct to the other reference strains from L. noguchi serogroup Australis (serovars Rushan, Peruviana, and Nicaragua). For the “Celledoni” isolate, none of the reference serovars within this serogroup belong to the species L. santarosai. The PFGE patterns of the “Icterohaemorrhagiae” strains, which were subdivided into L. interrogans or L. kirschneri, were identical to the patterns obtained from L. interrogans serovars Icterohaemorrhagiae and Copenhageni, known to be indistinguishable by PFGE and other molecular typing techniques, and L. kirschneri serovar Bogvere (less than three band differences were observed). The “Icterohaemorrhagiae” strains that were isolated from different patients over an eight-year period (2004–2012) and those from rats all presented indistinguishable PFGE patterns. The “Ballum” isolates displayed a pattern that was similar to that displayed by L. borgpetersenii serovars Ballum, Castellonis, Guangdong, Arborea, and Soccoestomes [30] (Figure 2).
MLVA (Multi Locus VNTR Analysis) is a simple and rapid PCR-based method for the identification of most of the serovars of L. interrogans and L. kirschneri [23]. The L. interrogans and L. kirschneri isolates from the French West Indies had a MLVA pattern with VNTR-4, VNTR-7, and VNTR-10 identical to the serovars Icterohaemorrhagiae and Bogvere type strains, respectively. This is in agreement with the clusters determined by PFGE (Table 1), further confirming the identity of the serovars Icterohaemorrhagiae/Copenhageni and Bogvere. Strains from species L. borgpetersenii, L. noguchi, L. santarosai, and L. kmetyi could not be typed by this method because of the absence of one or more of the VNTR loci.
The secY housekeeping gene [22] was also amplified from DNA extracts and sequenced. No PCR products were obtained for DNA from the L. kmetyi strains (here designated as genotype F). This was presumably due to mismatching between the PCR primers and the target gene (due to DNA sequence divergence), preventing PCR amplification [31]. The phylogenetic tree constructed with the secY nucleotide sequences is shown in the Figure 3 (Figure 3). Our 143 sequences (not including the 3 L. kmetyi strains) segregate into five main clades that correspond to the species identified by 16S rRNA sequencing. Thirteen different genotypes were observed and genotypes A (42 isolates), B (35 isolates), and C (33 isolates) were the most prevalent. The secY alleles A, B, and C were associated with serovars Icterohaemorrhagiae/Copenhageni, Bogvere, and Arborea/Castellonis/Ballum/Guangdong/Soccoestomes, respectively. The remaining 32 strains were distributed into nine groups (D, E, G, H, I, J, K, L, and M), including six new alleles not found in the database published by Nalam et al. [32]. Thus, there were thirteen groups in total, and most were present on both Guadeloupe and Martinique. However, some genotypes were found only among isolates from Guadeloupe (group G with 9 isolates) or Martinique (groups H with 3 samples, I with 6 samples, and J with 2 isolates). Clusters A and B contained both, clinical and rat isolates.
Leptospirosis is endemic in the French West Indies. The first human cases were first documented in 1932 in Guadeloupe [33] and 1938 in Martinique [34]. The annual incidence of leptospirosis in the French West Indies was estimated to be approximately 10 cases per 100,000 inhabitants in the 1990s. The incidence of leptospirosis during 2002–2004 was affected by the El Nino phenomenon, which resulted in increases in rainfall and the number of cases in Guadeloupe [35]. A prospective study of patients with acute febrile illness in Martinique and Guadeloupe (InVS, CIRE Antilles-Guyane) in 2011 improved the surveillance of leptospirosis. This increased awareness could explain the record incidence in 2011, peaking at <39 cases per 100,000 inhabitants in both Guadeloupe and Martinique.
Although the use of PCR diagnostic testing is becoming more common in the French West Indies, the diagnosis of leptospirosis is mostly dependent on MAT, which can identify the presumptive serogroup of the infecting bacterium. MAT has been used to show that the most frequent serogroups in Guadeloupe are Icterohaemorrhagiae and Ballum, followed by Sejroe and Canicola [36] (data from the NRC for Leptospirosis). Serogroups Cynopteri, Tarassovi, Panama, Grippotyphosa and Autumnalis are less common. The sensitivity of MAT is low during the acute stage of disease [37] and, because of paradoxical reactions and cross-reactions between serogroups, the accuracy of MAT in identifying the infecting serovar or serogroup can also be poor [38], [39], limiting its epidemiological value. In this study, MAT serological data from culture-positive patients were reviewed retrospectively, allowing the identification of a total of 36 patients with MAT data for serum samples (data not shown). It was possible to infer the serogroup identity of infecting leptospires from the MAT results for 26 of these 36 patients (72%). Similarly, only a small proportion of PCR-positive samples were correctly identified by MAT (Table 2). This further confirms that only the isolation of Leptospira from patients allowed definitive identification of the infecting serovar and is therefore essential for the study of the epidemiology of the disease.
We determined 16S rRNA sequences to identify the isolates to the species level, and then used serogrouping, PFGE, secY sequences, and MLVA to sub-type the species.
For most of the isolates (101/110), the PFGE patterns were mostly consistent with those of known serovars: i.e. serovars Bogvere, Tabaquite, Bajan, and Icterohaemorrhagiae or Copenhageni. For the serogroup Ballum, the PFGE patterns of the reference isolates for serovars Ballum, Castellonis, Guangdong, Arborea, and Soccoestomes were all similar, with fewer than three band differences [30]. Serovar Arborea was previously identified by CAAT as the major serovar from the serogroup Ballum in the Caribbean island Barbados [40], suggesting that our strains may belong to serovar Arborea. For the remaining five isolates which were serogrouped (Tarassovi and Celledoni), comparison of PFGE patterns with reference strains was inconclusive for the serovar. Finally, for four isolates the rabbit antisera used did not lead to agglutination such that comparison with the reference strains was not possible. Surprisingly, none of the isolates in the last ten years from the French West Indies were identified as belonging to serogroups Canicola or Sejroe, although up to14% of MAT-positive sera correspond to these two serogroups. Similar findings were reported in Barbados [40]. This may be due to cross-reactions between serogroups in MAT and/or difficulties in isolating these strains from patients (for example patients not hospitalized because of less severe symptoms or the strains fail to grown in EMJH medium).
Typing by PCR-based methods for amplification of 16S rRNA, secY, and VNTR loci can be used directly on biological samples, thus avoiding culturing of the pathogen. The bacterial load in blood during the acute phase ranges from 102 to 106 Leptospira/ml. The Leptospira count decreases with time, and can be detected for up to 15 days [41]. Thus, if the bacterial load is low, it may be necessary to use nested-PCR for amplification of the target sequences. The classification according to secY sequences was in good agreement with the groupings determined by PFGE and MLVA, further confirming our previous data on clinical isolates from Mayotte [42].
Sequencing of secY in DNA extracted from the clinical isolates and blood samples allowed a simple and rapid first-line screening and the identification of the presumptive serovar. A total of thirteen genotypes were found in our study, a large proportion of strains (75%) being of only three genotypes associated with serovars Icterohaemorrhagiae/Copenhageni, Bogvere, and Arborea. The secY sequences from the L. santarosai isolates showed the highest nucleotide diversity. Six genotypes were not found in the MLST database and may therefore be specific to the French West Indies. Further characterization of these isolates should include the use of the CAAT, which requires the preparation of antibodies against the strain of interest, for definitive identification of the serovar. We also detected the appearance of strains related to the pathogenic species L. kmetyi in Martinique in at least two patients, one of which was probably exposed during canyoneering activities in the tropical forest (Hochedez et al., submitted). To our knowledge, L. kmetyi which was first isolated from soil in Malaysia [43], has never been isolated from a patient with leptospirosis. Further studies are needed to determine the serological and molecular features of these strains and their distribution in the French West Indies.
The distribution of the predominant pathogenic leptospiral serovars differed between Guadeloupe and Martinique. Serovars Bogvere, Arborea, and Icterohaemorrhagiae/Copenhageni made up 35, 31, and 23% respectively of all Leptospira isolates in Guadeloupe since 2004. In Martinique, serovar Icterohaemorrhagiae is the most frequent (35%), followed by Arborea (9%) and Bogvere (6%). In the Caribbean island of Barbados, 140 miles from Martinique, serovars Arborea (14%) and Icterohaemorrhagiae (26%) similarly cause many human infections, but the serovar Bogvere [38], [40], [44], which was first isolated in Jamaica [45] does not. Serovar Tabaquite (serogroup Mini), which was found in Guadeloupe, was first isolated from a patient in Trinidad [46]. Serovar Bim (serogroup Autumnalis) is the most frequently isolated serovar in Barbados (75% of all isolates) [40], was not isolated in the French West Indies. This suggests that some strains circulate throughout the Caribbean islands but others are highly prevalent only in restricted areas. This may be related to the distribution of the animal reservoirs for the different serovar in these islands.
Leptospira can colonize or infect renal tubules of a wide variety of wild and domesticated mammals. In the Caribbean, numerous mammalian species including rodents, opossums, mongoose, bats, pigs, cattle, and dogs have been demonstrated to be hosts of pathogenic Leptospira species [47]. Isolates from the serogroup Icterohaemorrhagiae, including serovars Icterohaemorrhagiae and Bogvere, have been isolated from the kidneys of rats, mice, and mongoose and isolates from the serogroup Ballum were isolated from rats and mice, suggesting predominantly rodent-borne transmission of the disease. Serovar Arborea was reported to be prevalent in both humans and animals in Barbados [40], [48]. Serovar Bajan was originally isolated from toads and frogs in Barbados [49]. In our study, human and rat isolates from Guadeloupe and belonging to serovars Icterohaemorrhagiae/Copenhageni, Bogvere, and Arborea all showed identical genotypes, consistent with rats being responsible for the transmission of the disease.
The identification of the circulating etiological agents of leptospirosis in the French West Indies will help establish appropriate control and prevention measures in this area where the disease is endemic. For example, the reference technique, MAT, requires a panel of live antigens representing a broad range of serogroups. The use of local isolates in the panel of antigens may maximize the chances of detecting an immune response to the infecting bacterium. At the NRC for Leptospirosis (Institut Pasteur), the initial panel of 18 antigens, which already included strains representative of the serogroups Icterohaemorrhagiae, Ballum, Australis and Tarassovi, was thus expanded to include local isolates from serogroups Celledoni and Mini for serum samples originating from the French West Indies. Knowledge of leptospiral epidemiology may also be useful for the development of a whole bacterial vaccine against leptospirosis. Vaccines currently available for use in animals and, in a few countries, in humans generally consist of one, two or more locally prevalent serovars. In France, a human vaccine containing only serovar Icterohaemorrhagiae has been used since 1981 [50]. However, we report here that only one-third of the infections in the French West Indies are due to serovar Icterohaemorrhagiae (not including serovar Bogvere from the serogroup Icterohaemorrhagiae), and the corresponding figure for Barbabos is 22.5% [38]. Immunity is restricted to antigenically related serovars, so the vaccine used in France may not be effective against the majority of strains circulating in the French West Indies.
Further studies should include the analysis of the influence of serovar and strain genetic background on the clinical presentation and outcome of the disease. It would also be valuable to investigate the reasons for differences in the distributions of Leptospira serovars in the Caribbean islands.
|
10.1371/journal.ppat.1006348 | HIV-1 competition experiments in humanized mice show that APOBEC3H imposes selective pressure and promotes virus adaptation | APOBEC3 (A3) family proteins are DNA cytosine deaminases recognized for contributing to HIV-1 restriction and mutation. Prior studies have demonstrated that A3D, A3F, and A3G enzymes elicit a robust anti-HIV-1 effect in cell cultures and in humanized mouse models. Human A3H is polymorphic and can be categorized into three phenotypes: stable, intermediate, and unstable. However, the anti-viral effect of endogenous A3H in vivo has yet to be examined. Here we utilize a hematopoietic stem cell-transplanted humanized mouse model and demonstrate that stable A3H robustly affects HIV-1 fitness in vivo. In contrast, the selection pressure mediated by intermediate A3H is relaxed. Intriguingly, viral genomic RNA sequencing reveled that HIV-1 frequently adapts to better counteract stable A3H during replication in humanized mice. Molecular phylogenetic analyses and mathematical modeling suggest that stable A3H may be a critical factor in human-to-human viral transmission. Taken together, this study provides evidence that stable variants of A3H impose selective pressure on HIV-1.
| Human APOBEC3 family proteins are known as intrinsic defenses against HIV-1, whereas HIV-1 Vif counteracts APOBEC3-mediated anti-viral action. Using a hematopoietic stem cell-transplanted humanized mouse model, we demonstrated that endogenous APOBEC3D, APOBEC3F and APOBEC3G play pivotal roles in restricting HIV-1 replication in vivo. In addition to these three APOBEC3 family proteins, certain haplotypes of APOBEC3H have the ability to control HIV-1 replication in cell culture studies. However, the anti-viral effect of APOBEC3H polymorphism in vivo and in human population is yet to be addressed. Here we use a humanized mouse model to show that acquiring resistance to anti-viral APOBEC3H is necessary for HIV-1 replication. Together with phylogenetic analyses and mathematical modeling, we conclude that APOBEC3H is a critical determinant of HIV-1 replication within infected individuals and we propose that it may also be a factor in human-to-human HIV-1 transmission.
| Apolipoprotein B mRNA editing enzyme catalytic polypeptide-like 3 (APOBEC3; A3) enzymes are cellular single-stranded DNA cytosine deaminases that are specifically encoded in mammals [1,2]. Rodents including mice (Mus musculus) have a single A3 gene, while primates including humans (Homo sapiens), chimpanzees (Pan troglodytes) and Old World monkeys have seven A3 paralogous genes (A3A, A3B, A3C, A3D, A3F, A3G and A3H). Gene duplication is a hallmark of the genes that are under evolutionary selective pressures [3], and indeed, the seven primate A3 genes have been positively selected during evolution [4], These observations suggest that primate A3 proteins play crucial roles in primates including humans. Human A3G was discovered first and was shown to be capable of restricting the replication of human immunodeficiency virus type 1 (HIV-1) in an in vitro cell culture system [5]. Subsequent investigations revealed that several human A3 family proteins exhibit the ability to reduce HIV-1 infectivity [2,6–8]. Moreover, previous studies including ours have demonstrated that A3D, A3F, and A3G, which are endogenously expressed in human CD4+ T cells, are restriction factors potently controlling HIV-1 replication in human CD34+ hematopoietic stem cell (HSC)-transplanted humanized mouse models [9–12]. To antagonize the anti-viral effect of A3 proteins, HIV-1 encodes a protein named viral infectivity factor (Vif). Vif orchestrates cellular ubiquitin ligase complex and degrades anti-viral A3 proteins via ubiquitin/proteasome-dependent pathway in infected cells [2,13].
In addition to A3D, A3F and A3G, human A3H is known as a potent restriction factor against HIV-1. Human A3H is polymorphic and has seven haplotypes [14,15]. Three of them, called haplotypes II, V, and VII, produce stably expressed enzymes that exhibit anti-HIV-1 activity in model cell culture experiments as well as primary T lymphocytes ex vivo [14–16]. In contrast, the other three haplotypes (III, IV, and VI) do not exhibit detectable protein expression [14–16]. Additionally, our recent study has demonstrated that A3H haplotype I (A3H-I) has intermediate stability and clear enzymatic activity [17] (Fig 1A). Importantly, the frequency of each haplotype differs among human population, with a higher frequency of stable A3H in the African-descendant population [14,15]. Furthermore, it is more intriguing that the Vif proteins of certain HIV-1 strains are unable to counteract stable A3H haplotypes, and the ability of Vif to antagonize stable A3H is determined by at least two residues at positions 39 and 48 (Fig 1B) [18–20]. These observations suggest that both the A3H-mediated anti-viral effect and the antagonistic ability of Vif against A3H are co-mingled in the human population, in contrast to the functional relationships between Vif and A3D, A3F and A3G, which appear much less variable. However, the robustness of the effects of stable/intermediate A3H haplotypes on viral replication at an individual scale and a population scale remains unclear, and the dynamics by which HIV-1 may circumvent and/or counteract the anti-viral effect of stable A3H is yet to be addressed.
Here we use an HSC-transplanted humanized mouse model to demonstrate that stable A3H, but not intermediate A3H, which is endogenously expressed in human CD4+ T cells, is a bona fide restriction factor capable of controlling HIV-1 replication in vivo. In addition, we reveal that HIV-1 Vif readily acquires the ability to counteract stable A3H during viral expansion in vivo. Additionally, we use molecular phylogenetic analysis and mathematical modeling to further address the impact of stable A3H on HIV-1 epidemics. Our analyses suggest that stable A3H may control HIV-1 dissemination in both intra- and inter-individual scales.
To address the impact of endogenous A3H haplotypes (Fig 1A) on HIV-1 replication in vivo, two derivatives of the replication-competent CCR5-tropic HIV-1 strain NLCSFV3 [21] were made with differing A3H haplotype II (A3H-II) neutralization capabilities [20]. One virus encodes a Vif protein that is able to counteract stable A3H ("hyper Vif"), while the other encodes a Vif protein that does not ("hypo Vif") (Fig 1B) [20]. Importantly, previous reports demonstrated that the Vif's ability to degrade stable A3H is determined by the two amino acid residues at positions 39 and 48 (Fig 1B) [18–20]. Consistent with a prior study [20], hyper HIV-1 fully counteracted the anti-viral activity mediated by A3H-II, whereas hypo Vif was not able to counteract A3H-II (Fig 1B). In the absence of A3H-II, the infectivity of both of these HIV-1 molecular clones is similar (Fig 1B).
To investigate the impact of endogenous A3H on HIV-1 replication in vivo, a series of hyper versus hypo Vif competition experiments was conducted in humanized mice reconstituted with stable A3H-expressing HSCs. The first experiment used eight humanized mice, which were heterozygous for stable A3H haplotypes (S1 Table): two out of the eight had blood cell compartments reconstituted with haplotypes III and V cells, and the other six mice expressed haplotypes I and II (S1 Table). Next, these eight mice were intraperitoneally co-inoculated with equal amounts of hyper and hypo viruses (1,500 TCID50 each; Fig 1C), and the amount of viral RNA in the plasma and the level of human CD4+ T cells in the peripheral blood (PB) were routinely analyzed for 6 weeks post-infection (wpi). HIV-1 efficiently expanded in the humanized mice, as observed in our previous studies [11,12,22–24], and the level of peripheral CD4+ T cells was significantly reduced compared to mock-infected mice (Fig 1D). At 6 wpi, viral RNA was extracted from the plasma of infected mice and the sequences of the vif gene were analyzed. As anticipated, hyper vif and its derivatives were able to outcompete hypo vif virus in mice expressing stable A3H (73.1% ± 7.7% [286/391] in Fig 1E; see also S2 Table). However, hypo vif-related sequences were the majority in infected mouse no. 5 and were still present at significant levels in all animals (no. 5, 67.3% [35/52] to no. 1, 7.9% [3/38]; Fig 1E & S2 Table). These results raised the possibility that these hypo vif viruses may have adapted in vivo and gained a better ability to counteract stable A3H during viral replication. To address this idea, we subcloned the 13 hypo vif-related open reading frames (ORFs) into the expression plasmid and evaluated their anti-stable A3H activity using in vitro cell culture system. As shown in Fig 1F, adapted Vif proteins with V39F/N48H (26 clones from 4 mice), V39F/N48H/D113N (1 clone from 1 mouse) and V39F/N48H/L148S (1 clone from 1 mouse) mutations, degraded A3H-II and impaired the A3H-II packaging into the released viral particles. Additionally, the HIV-1 restriction capacity of A3H-II was significantly counteracted by these 3 adapted hypo Vif derivatives as evidenced by hyper Vif levels of infectivity (Fig 1G). We verified that these 3 hypo Vif derivatives as well as parental hypo Vif were active in counteracting other HIV-1 relevant A3s such as A3D, A3F and A3G (S1 Fig). Together, the sequencing results and the tests of the functionality of the adapted hypo Vif proteins indicated that 80.3% ± 4.8% (314/391) of the vif sequences in the plasma of infected stable A3H mice are able to counteract stable A3H (Fig 1E–1G & S2 Table). These findings indicate that the ability to antagonize stable A3H is required for efficient HIV-1 replication in humanized mice.
We next investigated whether HIV-1 undergoes selection as a result of pressure from A3H-I in vivo. Six humanized mice were reconstituted with HSCs from three individual donors. Five out of the six mice were homozygotes for A3H-I and one mouse was heterozygous for A3H haplotypes I and VI (S1 Table). These six mice were co-inoculated with hyper and hypo HIV-1s (Fig 2A). All mice exhibited a high level of viremia and a declined level of peripheral CD4+ T cells (Fig 2B). We then analyzed the vif sequences in the plasma of these six infected mice at 6 wpi. In contrast to the observations from animals with at least one copy of stable A3H (hap II or V; Fig 1E), the proportion of hyper/hypo vif sequences varied in each infected mouse, and no obvious replication biases were observed (Fig 2C & S3 Table). On average, the percentage of hyper and hypo vif-derived sequence were similar, 46.5% ± 13.7% (127/273) and, 53.5% ± 13.7% (146/273), respectively (S4 Table).
To assess whether A3H genotype affects the expression level of other HIV-1 relevant A3 genes, we analyzed the expression levels of A3D, A3F and A3G in the splenic CD4+ T cells of humanized mice. Consistent with our previous studies with primary CD4+ T cells ex vivo and with humanized mice [24,25], the mRNA expression levels of these A3 genes in HIV-1-infected mice were significantly higher than those in mock-infected mice (S2A Fig). The expression levels of these A3 genes were comparable between the humanized mice expressing stable A3H and intermediate A3H (S2B Fig), indicating that the A3H genotype is not associated with the expression levels of other HIV-1 relevant A3 genes.
In addition to A3H, single nucleotide variants (SNVs) in human A3G [26–28], A3F [29,30] and A3D [31] have been reported. The variants of A3G [28] and A3D [31] are degraded efficiently by HIV-1 Vif and are therefore unlikely to play significant roles in vivo. In contrast, An et al. have recently reported that an SNV of A3F, V231I, confers partial resistance to Vif-mediated degradation by certain strains of HIV-1 [30]. To address the possibility that A3F V231I mutant affects viral growth and the sensitivity to hyper/hypo HIV-1, we assessed the genomic sequences of A3F. However, this A3F SNV was not detected in the human cells used in our studies (data not shown). Altogether, these findings suggest that no specific selective pressure is elicited against either hyper or hypo HIV-1 in intermediate A3H humanized mice and that virus expansion is occurring in a stochastic manner.
Co-infection studies revealed that hyper Vif HIV-1 dominates over hypo Vif virus in animals humanized with stable A3H expressing cells (Fig 1). These observations suggest that the stable A3H protein, which is expressed endogenously in human CD4+ T cells, exhibits a robust anti-viral effect and impairs the expansion of the viruses without full A3H counteraction abilities (i.e., V39 hypo Vif). We used three HIV-1 strains, NLCSFV3, JRCSF and AD8 to demonstrate that the Vif proteins of these viruses are unable to antagonize A3H-II (S3 Fig). We then inoculated these viruses into 50 humanized mice reconstructed from 16 HSC donors (Fig 3A). The genotyping PCR revealed that 13 out of the 16 HSC donors encode A3H-I, and 3 donors possessed one stable A3H allele (S5 Table). Based on A3H haplotypes, these infected mice were classified into two groups, intermediate A3H (n = 37) and stable A3H (n = 13), and the level of peak viral load in each group was compared. As shown in Fig 3B, surprisingly, the peak viral load was comparable between intermediate A3H mice and stable A3H mice (P = 0.92 by Mann-Whitney U test). Because certain hypo vif derivatives acquired anti-stable A3H activity in the stable A3H mice co-inoculated with hyper/hypo HIV-1 (Fig 1E–1G), we hypothesized that these viruses acquired de novo resistance to stable A3H in vivo (strains NLCSFV3, JRCSF, or AD8; Fig 3B). To test this hypothesis, we analyzed the vif sequences in the plasma of the 4 stable A3H mice infected with HIV-1 (strain NLCSFV3) at 6 wpi. Notably, some Vif sequences were commonly detected in the 4 stable A3H mice infected with NLCSFV3 (Fig 3C; the raw data and mutation matrix are shown in S4 Fig). To investigate whether these mutant variants acquired anti-stable A3H activity de novo, we prepared the expression plasmids of these Vif derivatives and conducted in vitro experiments using our cell culture system. As shown in Fig 3D & 3E, we detected 2 Vif variants, N48H and V13I/N48H/GDAK60-63EKGE, that are able to antagonize A3H-II at the level observed for hyper Vif. In summary, 59.7% ± 5.1% (145/243) of the vif sequences in the plasma acquired the ability to counteract stable A3H (S6 Table), but such mutants were not detected in the intermediate A3H mice infected, solely, with NLCSFV3 (S5 Fig & S7 Table). Taken together, these findings suggest that the ability of HIV-1 Vif to antagonize stable A3H is acquired de novo during viral expansion in vivo.
We previously reported that endogenous A3H mRNA expression levels in primary human CD4+ T cells are significantly lower than those of anti-viral A3 genes such as A3D, A3F, and A3G, and that activation and/or infection stimuli induces higher A3H expression [25,32]. In agreement with these prior works, the activation stimuli driven by anti-CD3/CD28 antibodies induced the expression of CD25, a marker of activated human CD4+ T cells (S6A Fig), and also A3H mRNA expression levels (P = 0.010 by paired t test, S6B Fig). However, it should be noted that human CD4+ T cells in humanized mice [12] and human PB [33] are less activated (S6C Fig) and in a quiescent state. In this regard, previous studies reported that HIV-1 infection induces the activation of CD4+ T cells of infected individuals [34,35]. Therefore, we hypothesized that HIV-1 infection induced CD4+ T-cell activation and augmented A3H expression in humanized mice, and this resulted in robust anti-viral effect by endogenous A3H (Figs 1 & 3). To investigate the immune activation status in detail, we performed RNA sequencing (RNA-seq). Human mononuclear cells (MNCs) were isolated from the spleen of 4 HIV-1-infected mice and 4 mock-infected mice at 6 wpi, and RNA-seq analyses were conducted. As shown in Fig 4A, 93 genes were significantly up-regulated by HIV-1 infection, whereas 16 genes were down-regulated. Parametric gene set enrichment analysis (GSEA) revealed that the genes associated with T-cell/lymphocyte activation, inflammatory response, and positive regulation of T cell activation were significantly up-regulated in the human MNCs of HIV-1-infected mice (Fig 4B; the GESA result is listed in S8 Table). In addition, various interferon-stimulated genes such as RSAD2 (encoding Viperin), DDX58 (encoding RIG-I), EIF2AK2 (encoding PKR), MX1, ISG15, MOV10 and BST2 (encoding tetherin) were up-regulated in HIV-1-infected mice (Fig 4A). As observed in infected patients [34,35], our findings suggest that HIV-1 infection triggers immune activation in humanized mouse model.
We then addressed the possibility that the immune activation caused by HIV-1 infection (Fig 4A & 4B) leads to the up-regulation of A3H in humanized mice. As shown in Fig 4C, the proportion of the splenic CD4+ T cells (CD3+ CD8− cells) of infected mice was significantly lower than that of uninfected mice, and particularly, CD25+ activated CD4+ T cells were severely depleted by HIV-1 infection (P = 0.0039 versus mock infection). Consistent with our previous findings [12], HIV-1 infection led to the severe depletion of activated CD4+ T cells in humanized mice.
Next, we sorted the fractions of non-activated CD4+ T cells (CD45+ CD3+ CD8− CD25− cells) and activated CD4+ T cells (CD45+ CD3+ CD8− CD25+ cells) of mock-infected mice (S7 Fig) and analyzed the mRNA expression level of A3H in each population by real-time RT-PCR. In mock-infected mice, A3H expression in the activated CD4+ T cells was significantly higher than that in non-activated cells (P = 0.0090 by Mann-Whitney U test; Fig 4D). This finding further suggests that the CD4+ T-cell activation augments A3H expression, as observed in in vitro experiments (S6B Fig) and in our previous reports [25,32]. Because CD25+ CD4+ T cells were severely depleted in infected mice (Fig 4C), we sorted only the fraction of CD25− CD4+ T cells (CD45+ CD3+ CD8− CD25− cells) of HIV-1-infected mice for real-time RT-PCR. Interestingly, the A3H expression level in the CD25− CD4+ T cells of HIV-1-infected humanized mice was significantly higher than that of CD25− CD4+ T cells of uninfected mice (P = 0.0062 by Mann-Whitney U test; Fig 4D). Altogether, these findings suggest that the immune activation triggered by HIV-1 infection augments A3H expression in CD4+ T cells of infected humanized mice.
Finally, we addressed how hyper and hypo HIV-1 sequences circulate in the human population. The HIV-1 Vif sequences were obtained from the Los Alamos National Laboratory HIV-1 sequence database (https://www.hiv.lanl.gov/components/sequence/HIV/search/search.html). Fig 5A shows a phylogenetic tree of Vif sequences of HIV-1 group M (n = 2,976), which is a pandemic strain worldwide. The phylogenetic tree indicates that Vif sequences cluster based on subtype (Fig 5A). Interestingly, the sequences of hyper Vif (here we defined "hyper Vif" as a sequence that possesses F or Y at position 39 and H at position 48) scattered in this tree and did not form a unique cluster (Fig 5A). Additionally, the percentage of hyper Vif varied in each subtype (Fig 5B), suggesting that hyper and hypo Vif mutually swap in human population.
To evaluate the counteracting ability of HIV-1 clinical isolates (group M) against stable A3H, we used 15 infectious molecular clones (IMCs): 8 subtype B and 7 subtype C; 10 transmitted/founder (TF) viruses and 5 chronic control (CC) viruses. As shown in Fig 5C, the infectivity of the 3 IMCs (strains AD17, RHPA and MCST) was significantly suppressed by A3H-II with statistical differences, suggesting that these viruses have established new infection as TF viruses in individuals without anti-stable A3H activity. In contrast, other 12 IMCs overcame A3H-II-mediated restriction (Fig 5C), suggesting that these viruses exist in human population as hyper HIV-1. Importantly, the anti-stable A3H ability of these IMCs corresponded well to the amino acid residues positioned at 39 and 48 (S9 Table). These findings suggest that anti-stable A3H ability is not a necessary requirement for certain viruses circulating within individuals.
We then assessed the anti-stable A3H ability of non-pandemic HIV-1 groups N, O, and P. As shown in Fig 5B, the proportion of hyper Vif sequences varied in each group. In particular, HIV-1 group O strains (n = 51) did not encode a hyper vif sequences. However, the cell-based experiments demonstrated that the IMCs of group O (strains BCF183 and RBF206) overcame A3H-II-mediated anti-viral effect (Fig 5C). The IMC of HIV-1 group P (strain RBF168) also counteracted A3H-II, while that of HIV-1 group N (strain DJO0131) did not (Fig 5C). Interestingly, in contrast to the results of HIV-1 group M, the anti-stable A3H ability of HIV-1 group O Vif was not governed by the two residues at positions 39 and 48 (S9 Table) These findings suggest that other residues than those positioned at 39 and 48 determine the ability of Vif proteins of HIV-1 group O to counteract stable A3H.
Furthermore, we assessed the correlation between the frequency of hyper HIV-1 and the proportion of the individuals harboring stable A3H haplotype worldwide. The HIV-1 Vif sequences were obtained from HIV-1 sequence database (S10 Table), and the information of A3H haplotype was obtained from the 1000 Genomes Project (http://www.internationalgenome.org) [36] (S11 Table). As shown in Fig 5A & 5B, the Vif sequences are highly diversified and the logoplot (S8 Fig) further indicated that the amino acids at position 39 and 48 were not highly conserved when compared to the YRHHY motif, which is essential for A3G degradation [24,37]. Additionally, consistent with previous reports [14,15,20], both the percentage of hyper Vif and the proportion of stable A3H haplotype were highest in Africa, particularly in Nigeria (Fig 5D & S12 Table), and these two parameters were correlated each other with statistical significance (Spearman's r = 0.720, P = 0.017 by Spearman rank correlation test; S9 Fig). To further investigate the relationship between hyper HIV-1 and stable A3H haplotype, we conducted a mathematical simulation. As shown in Fig 5E, the frequency of hyper HIV-1 increased dependent on the proportion of the people harboring stable A3H haplotype. Taken together, our analyses at a human population level suggest that stable A3H elicits a selective pressure against HIV-1, and that HIV-1 overcomes stable A3H-mediated anti-viral immunity by acquiring the ability to counteract stable A3H.
In this study, we used a humanized mouse model to show that HIV-1 infection induces immune activation and augments the expression of endogenous A3H in human CD4+ T cells (Fig 4). We also showed that the ability of HIV-1 Vif to counteract stable A3H-mediated anti-viral effect is crucial for efficient viral expansion in vivo when endogenous A3H is expressed stably (Figs 1 & 3). In contrast, the ability of HIV-1 Vif to counteract stable A3H is dispensable when stable A3H is absent in vivo (Fig 2). Furthermore, we addressed the significance of the stable A3H-mediated anti-viral effect on HIV-1 dissemination in human populations using molecular phylogenetic analysis and mathematical modeling. The occurrence of hyper Vif variants and stable A3H haplotypes correlates worldwide, suggesting that the ability of Vif to antagonize stable A3H was acquired during viral spread throughout the human population (Fig 5). These findings suggest that the A3H polymorphism influences HIV-1 dissemination at individual and population levels.
In the stable A3H humanized mice co-inoculated with hyper and hypo HIV-1 infectious clones, several hypo Vif viruses acquired V39F and N48H changes, which resulted in the gain-of-function to counteract A3H-II (Fig 1E–1G). Given that these two amino acids are identical to those in hyper Vif, the emergence of the hypo Vif derivatives, which can potently antagonize A3H-II in the hyper/hypo HIV-1 co-inoculated stable A3H mice (Fig 1), may be due to the recombination between hypo and hyper Vif sequences. However, the results shown in Fig 3 argue against this possibility. We demonstrated that some Vif sequences with the ability to antagonize stable A3H emerge during viral replication in the stable A3H mice within only 6 weeks. In contrast, in the intermediate A3H mice co-inoculated with hyper and hypo HIV-1 clones, the hyper or hypo Vif viruses expanded randomly with no evidence of selection on vif (Fig 2). These findings suggest that the stable A3H, which is endogenously expressed in CD4+ T cells, has a robust anti-viral activity in vivo and that it is feasible for Vif to acquire the counteracting ability against stable A3H de novo. We favor a model in which the starting hypo Vif virus is constrained evolutionarily, likely by needing to counteract A3D, A3F, and A3G, and that de novo (not recombination mediated) amino acid substitutions at positions 39 and 48 provide the most efficient route to optimize anti-A3H activity. Moreover, it is important to note that all the stable A3H humanized mice used in this study were heterozygous for A3H stability (S1 & S5 Tables). It appears that an allele of stable A3H is sufficient to induce a robust selective pressure against HIV-1.
In sharp contrast to the findings in the stable A3H mice co-inoculated with hyper and hypo HIV-1s (Fig 1), hyper HIV-1 was not commonly selected in the intermediate A3H mice co-inoculated with hyper and hypo HIV-1 clones (Fig 2). Also, de novo emergence of hyper Vif was not detected in the intermediate A3H mice infected with NLCSFV3 (S5 Fig). On the other hand, we recently showed that the intermediate A3H (A3H-I) is enzymatically active and contributes to breast and lung cancer mutagenesis despite being expressed at lower levels compared to its stable A3H counterpart [17]. These findings suggest that A3H-I, which is endogenously expressed in human CD4+ T cells, is not sufficient to impose selective pressure on HIV-1 replication in vivo.
Here we detected the emergence of Vif sequences that acquired the ability to antagonize stable A3H (Figs 1 & 3). In contrast, in the humanized mice infected with a vif-mutated HIV-1 (designated 4A HIV-1), which is sensitive to A3D and A3F, we have previously demonstrated that Vif sequences with the ability to antagonize A3D and A3F do not emerge [24]. We confirmed the absence of Vif revertants in the plasma of two 4A-HIV-1 infected mice infected at 6 wpi (S10 Fig). These observations suggest that HIV-1 is able to overcome the restriction mediated by stable A3H but not by A3D and A3F during viral replication in vivo. Two nonexclusive models may explain the observed differences. One possibility is that it might be more feasible for Vif to overcome stable A3H-mediated restriction than A3D/A3F because the anti-viral activity of endogenous stable A3H is lower than those of endogenous A3D and A3F. However, at least four previous studies have demonstrated that the anti-HIV-1 activity of stable A3H (haplotype II) is similar to that of A3F and is higher than that of A3D [18,19,25,38] and argue against this possibility. In addition, it should be noted that the endogenous expression levels of the respective A3 genes in primary human CD4+ T cells are different from each other. Indeed, Refsland et al. have revealed that endogenous expression levels of A3D and A3F mRNAs are higher than that of A3H in primary CD4+ T cells [32]. Another possibility is the number of amino acids responsible for A3 counteraction: only two amino acids at positions 39 and 48 are responsible for counteracting stable A3H [18–20], while there are four that are responsible for counteracting A3D and A3F (known as DRMR motif at position 14–17) [37,39].
The emergence of Vif revertants harboring the ability to counteract stable A3H is reminiscent of the observations that the sub-optimal drug concentrations facilitate the emergence of drug-resistant viruses in infected patients [40,41]. In fact, it appears difficult for Vif to acquire the ability to counteract A3F and A3G de novo during viral replication in cell cultures [42,43] and a humanized mouse model [11,24]. In contrast, previous studies have successfully selected the viruses that acquired the ability to counteract stable A3H in the in vitro culture infection experiments using the human CD4+ T cell lines such as MT-4 cells [18] and SupT11 cells [20] that ectopically express A3H-II. Here we demonstrated that HIV-1 infection induces immune activation in humanized mice, as observed in infected individuals [34,35], and augments the expression of endogenous A3H in the human CD4+ T cells of infected mice (Fig 4). But still, the anti-HIV-1 activity of endogenous stable A3H is not sufficient to control viral expansion in vivo, and therefore, Vif may easily acquire the ability to counteract the restrictive activity of endogenous A3H.
Our findings in infected humanized mice revealed that hyper HIV-1 is predominantly selected in the mice expressing stable A3H (80.3% ± 4.8%; S2 Table), while the viruses replicated in the mice with intermediate A3H were selected stochastically (46.5% ± 13.7%; S4 Table). We also demonstrated the de novo emergence of hyper HIV-1 in the stable A3H mice infected with NLCSFV3 (59.7% ± 5.1%; S6 Table). Based on these findings and numerical parameters, we investigated the dynamic effect of A3H haplotypes on HIV-1 epidemic in the human population through molecular phylogenetic and mathematical modeling and revealed that the occurrence of hyper Vif and stable A3H variants are correlated positively in the human population (Fig 5). This suggests that stable A3H may not just provide an intrinsic immunity at the level of individual patients, as elaborated here in humanized mice, but it may also function to control the dissemination of hypo HIV-1 isolates in the human population [44,45].
All procedures including animal studies were conducted following the guidelines for the Care and Use of Laboratory Animals of the Ministry of Education, Culture, Sports, Science and Technology, Japan. The authors received approval from the Institutional Animal Care and Use Committees (IACUC)/ethics committee of the institutional review board of Kyoto University (protocol number D15-08). All protocols involving human subjects were reviewed and approved by the Kyoto University institutional review board. All human subjects were provided written informed consent from adults.
NOG mice (NOD/SCID/Il2r KO mice) [46] were obtained from the Central Institute for Experimental Animals (Kawasaki, Kanagawa, Japan). The mice were maintained under specific-pathogen-free conditions and were handled in accordance with the regulation of the IACUC/ethics committee of Kyoto University. Human CD34+ hematopoietic stem cells (HSCs) were isolated from human fetal liver as previously described [47]. The humanized mouse model (NOG-hCD34 mouse) was constructed as previously described [11,22,23,48–50]. In the experiments shown in Figs 1 & 2, 14 newborn (aged 0 to 2 days) NOG mice from 7 litters were irradiated with X-ray (10 cGy per mouse) using an RX-650 X-ray cabinet system (Faxitron X-ray Corporation) and were then intrahepatically injected with the human fetal liver-derived CD34+ cells (1.0 × 105 to 2.3 × 105 cells; 5 donors). A list of the humanized mice used in this study is summarized in S1 Table. In the experiments shown in Fig 3, the 35 NOG-hCD34 mice infected with HIV-1 were used in our previous studies [12,23,24] (Fig 3) and the 15 NOG-hCD34 mice were newly infected with HIV-1. These humanized mice were constructed using 16 independent HSC donors with 29 NOG litters (summarized in S5 Table).
HEK293T cells (a human embryonic kidney 293 T cell line; ATCC CRL-3216) and TZM-bl cells (obtained through the NIH AIDS Research and Reference Reagent Program) [51] were maintained in Dulbecco’s modified Eagle's medium (Sigma) containing FCS and antibiotics. Human peripheral CD4+ T cells were isolated human CD4+ T cell isolation kit (Miltenyi) according to the manufacturer’s protocol. These cells were activated with anti-CD3/anti-CD28 dynabeads (Thermo Fisher Scientific) and maintained in RPMI1640 (Sigma) containing FCS and antibiotics with human interleukin-2 (100 U/ml) as previously described [23].
To construct the IMCs of hyper HIV-1 and hypo HIV-1 derivatives (based on a CCR5-tropic strain NLCSFV3 [21]), the hyper and hypo Vif variants of the HIVIIIB A200C proviral constructs [20] were digested with AgeI and EcoRI, then the resultant DNA fragment was inserted into the AgeI-EcoRI site of pNLCSFV3 [21]. The IMCs of HIV-1 strains JRCSF [52] and AD8 [53] were also used. The two vif-mutated derivatives based on pNLCSFV3, vif-deleted virus (pNLCSFV3Δvif) and DRMR/AAAA-mutated virus (4A HIV-1), are constructed in our previous study [11,24]. The IMCs of transmitted/founder (TF) and chronic control (CC) viruses as well as those of HIV-1 groups N (strain DJO0131), O (strains BCF183 and RBF206) and P (strain RBF168) (Fig 5C) were obtained kindly provided by Drs. Beatrice H. Hahn (University of Pennsylvania, USA) and Frank Kirchhoff (Ulm University Medical Center, Germany).
To prepare virus solutions of hyper and hypo HIV-1s, 30 μg of each IMC was transfected into HEK293T cells according to calcium-phosphate method as previously described [11,12,23,24]. At 48 h posttransfection, the culture supernatant was harvested, centrifuged, and then filtered through a 0.45-μm filter (Millipore) to obtain the virus solution. The amount of viral particles was quantified using an HIV-1 p24 (Gag) antigen ELISA kit (Zeptometrix). Virus solutions of hyper and hypo HIV-1 clones (containing 2.5 ng of Gag antigen each) were intraperitoneally co-inoculated into NOG-hCD34 mice. RPMI 1640 was used for mock infection.
PB and plasma were routinely collected as previously described [11,12,22–24]. The mice were euthanized at 6 wpi with anesthesia and the spleen was crushed, rubbed, and suspended as previously described [11,12,22–24]. To obtain splenic human MNCs, the splenic cell suspension was separated using Ficoll-Paque (Pharmacia) as previously described [11,12,22–24]. The amount of HIV-1 RNA in 50 μl plasma was quantified by Bio Medical Laboratories, Inc. (the detection limit of HIV-1 RNA is 800 copies/ml).
In the experiments shown in Figs 1 & 2, genomic DNA was extracted from the PB of NOG-hCD34 mice using a DNeasy Blood & Tissue kit (Qiagen) as previously described [24]. In the experiments shown in Fig 3, genomic DNA was extracted from the splenic MNCs of NOG-hCD34 mice in the same procedure. Genotyping PCR of A3H was performed using PfuUltra High Fidelity DNA polymerase (Agilent) according to the manufacturer’s protocol, and the following primers were used: Exon2_Fwd, 5'-GAA ACA CGA TGG CTC TGT TAA CAG CC-3'; Exon3_Rev, 5'-CGG GGG TTT GCA CTC TTA T-3'; Exon4_Fwd, 5'-AGG AAG GAA GGA TTG TGG CTC A-3'; Exon4_Rev, 5'-GAG TCC TCA TGC TCA GCA CA-3' (see also Fig 1A). For genotyping PCR of A3F, the following primers were used: A3F_exon5_8822_Fwd. 5'-GGT CTC TGC ATT GGG GTT TC-3'; A3F_exon5_9069_Rev: 5'-TGC ATT CCT AGC TGC TTA GC-3'. The resulting DNA fragments were directly sequenced, and, if needed, were cloned using a zero blunt TOPO PCR cloning kit (Thermo Fisher Scientific). The sequence was analyzed with Sequencher v5.1 software (Gene Codes Corporation).
Flow cytometry was performed with FACS Calibur (BD Biosciences) and FACSJazz (BD Biosciences) as previously described [11,12,22–24], and the obtained data were analysed with Cell Quest software (BD Biosciences) and FlowJo software (Tree Star, Inc.). For flow cytometry analysis, anti-CD45-PE (HI30; Biolegend), anti-CD3-APC-Cy7 (HIT3a; Biolegend), anti-CD4-APC (RPA-T4; Biolegend), anti-CD25-APC (BC96; eBioscience), and anti-Ki67-PE (B56; BD Biosciences) antibodies were used. Hematometry was performed with a Celltac α MEK-6450 (Nihon Kohden Co.) as previously described [11,12,23,24,49]. Live cell sorting was performed using FACSJazz (BD Biosciences) according to the manufacture's procedure. The purity of each population was >94% (see also S7 Fig).
Transfection, the TZM-bl assay and Western blotting were performed as previously described [11,12,23,24]. Briefly, in the experiments shown in Fig 1B & S3 Fig, HEK293 cells were cotransfected with an expression plasmid for flag-tagged A3H-II (0, 25, 50 and 100 ng) and the indicated IMCs (1 μg). In the experiments shown in Figs 1F, 1G, 3D & 3E, HEK293 cells were co- cotransfected with an expression plasmid for flag-tagged A3H-II (10 ng), pNLCSFV3Δvif (500 ng) and an expression plasmid for the indicated Vif tagged with HA (500 ng). In the experiments shown in Fig 5C, HEK293 cells were cotransfected with an expression plasmid for flag-tagged A3H-II (50 ng) and the indicated IMCs (1 μg). In the experiments shown in S1 Fig, HEK293 cells were cotransfected with an expression plasmid for flag-tagged A3D (50 ng), A3F (10 ng) or A3G (10 ng), pNLCSFV3Δvif (500 ng) and an expression plasmid for the indicated Vif tagged with HA (500 ng). For Western blotting, anti-Flag antibody (M2; Sigma), anti-HA antibody (3F10; Roche), anti-p24 antiserum (ViroStat), and anti-α-tubulin (TUBA) antibody (DM1A; Sigma) were used.
RT-PCR was performed as previously described [24]. Briefly, viral RNA was extracted from the plasma of infected mice at 6 wpi using a QIAamp viral RNA mini kit (Qiagen), and cDNA was prepared as previously described [24]. RT-PCR was performed using PrimeSTAR GXL DNA polymerase according to the manufacturer’s protocol, and the following primers used are used: Vif-Fwd, 5'-GTT TGG AAA GGA CCA GCA AA-3'; Vif-Rev, 5'-GCC CAA GTA TCC CCG TAA GT-3'. The resulting DNA fragments were cloned using a zero blunt TOPO PCR cloning kit (Thermo Fisher Scientific), and the sequence was analyzed with Sequencher v5.1 software (Gene Codes Corporation).
The vif ORF sequences (Figs 1E, 2C & 3C) were aligned by using MUSCLE [54] implemented in MEGA 6 software [55]. ML phylogenetic trees were constructed using MEGA 5.1 software [55]. The Vif sequences (Fig 5A and 5B & S10 Table; one sequence per patient) were extracted from Los Alamos National Laboratory HIV-1 sequence database (https://www.hiv.lanl.gov/components/sequence/HIV/search/search.html). These sequences were aligned and the phylogenetic tree was constructed as described above.
A series of HA-tagged Vif expression plasmids are based on pDON-AI (Takara) and are constructed in our previous study [24]. To prepare the expression plasmids of Vif derivatives (Figs 1F, 1G, 3D & 3E), the pCRII-blunt-TOPO containing vif ORFs were digested with EcoRI and blunted. The resultant DNA fragments containing vif ORF were subcloned into the HpaI site of pDON-AI (Takara).
Human MNCs were isolated from the spleen of humanized mice as described above and RNA was extracted using QIAamp RNA Blood Mini kit (Qiagen) as described above [11,23,24]. RNA-seq analysis was conducted in Medical & Biological Laboratories, co (Nagoya, Japan). The obtained raw sequence data (.fastq files) were mapped to the human reference genome (NCBI hg19) by Bowtie2 version 2.2.5 [56], followed by spliced junction detection by Tophat2 version 2.1.0 [57]. Several R (versions 3.1.1) and Bioconductor packages were used to further process the gene expression data. Read count data for each sample were extracted by package ‘Rsubread’ [58]. The obtained raw read count data were then normalized by applying repeated edgeR normalization defined in package ‘TCC’ [59]. The normalized read count data were classified into two groups according to infection status (HIV-1 infected, or uninfected as control). The expression data were analyzed to detect differentially expression genes by package edgeR [60]. Top-ranked genes were selected as differentially expressed genes (DEGs) with the following threshold values: False Discovery Rate (FDR) less than 0.001 calculated by the Benjamini-Hochberg method [61], and more than twice up-regulated or less than half down-regulated normalized gene expressions compared with the control (see Fig 4B & S8 Table). DEGs were then used to obtain enriched biological functions by a parametric gene set enrichment analysis by using package ‘gage’ [62]. The method defined in ‘gage’ enabled to extract gene ontology terms associated with up-regulated DEGs. Finally, a distance matrix was calculated from the expression data for DEGs based on the correlation distance [63], and the distance matrix was converted by the Z-transformation defined in package ‘gplots’ to visualize the result with a heatmap (Fig 4A).
Real-time RT-PCR was performed as previously described [20,24] using CFX connect real-time system (Biorad) and the following primers: A3H-Fwd (RSH2757), 5'- AGC TGT GGC CAG AAG CAC-3' and A3H-Rev (RSH2758), 5'-CGG AAT GTT TCG GCT GTT-3'. A3D, A3F, A3G were amplified by using the primers reported previously [32], and the primers for GAPDH were purchased from Thermo Fisher Scientific.
The information of A3H haplotypes of individuals was extracted from the 1000 Genomes Project (http://www.internationalgenome.org) [36]. We obtained the Phase 1 VCF (variant call format) data of 1092 individuals from all available human populations. From this phased variant dataset we extracted the information of 5 A3H SNPs 15, 18, 105, 121, and 178 and estimated the frequencies of A3H haplotypes for each population.
The following simple model describes the HIV-1 transmission among human population:
dS(t)dt=b−dS(t)−βS(t)I(t)N(t), dI(t)dt=βS(t)I(t)N(t)−μI(t),
where S(t) and I(t) represent the number of susceptible and infected individuals, respectively [64]. N(t) is the total population size at time t, and N(0) = b/d is the initial size. Susceptible individuals are born at rate b and removed at rate d, and infected individuals transmit HIV-1 at a rate β during their infectious period of 1/μ. To describe the dissemination of hyper HIV-1 in the human population, we modified the above model as follows:
dSU(t)dt=bU−dSU(t)−βSU(t)N(t){IrU(t)+IoU(t)+IrS(t)+IoS(t)},
dIrU(t)dt=βSU(t)N(t){IrU(t)+IrS(t)}−μIrU(t),
dIoU(t)dt=βSU(t)N(t){IoU(t)+IoS(t)}−μIoU(t),
dSS(t)dt=bS−dSS(t)−βSS(t)N(t){IrU(t)+IoU(t)+IrS(t)+IoS(t)}
dIrS(t)dt=βSS(t)N(t){IrU(t)+IrS(t)+fIoS(t)}−μIrS(t),
dIoS(t)dt=βSS(t)N(t){IoU(t)+(1−f)IoS(t)}−μIoS(t).
The variable SU(t) is the number of susceptible individuals harboring unstable A3H haplotype, and IrU(t) and IoU(t) are the number of infected individuals with hyper and hypo HIV-1s, respectively. On the other hand, the variable SS(t) is the number of susceptible individuals harboring stable A3H haplotype, and IrS(t) and IoS(t) are the number of infected individuals with hyper and hypo HIV-1s, respectively. We assumed that the susceptible individuals harboring unstable and stable A3H haplotype are born at the rates bU and bS = b − bU, respectively. Furthermore, we considered that the fraction, f, of newly infected individuals harboring stable A3H haplotype with hypo HIV-1 become infected individuals with hyper HIV-1 because of adaptive evolution of hyper HIV-1 from hypo HIV-1 in vivo, as we observed in the stable A3H mice infected with NLCSFV3 (Fig 3 & S6 Table).
To investigate how the frequency of hyper HIV-1 at 100 years after the initial infection (i.e., (IrU(100)+IrS(100))/(IrU(100)+IrS(100)+IoU(100)+IoS(100))) is determined depend on the proportion of the people harboring stable A3H haplotype (i.e.,SS(0)/N(0) = (bS/d)/(b/d) = bs/b), we simulated the transmission dynamics of hyper and hypo HIV-1s among 1 million individuals for 0 < bs/b < 1 based on the above modified mathematical model. Here we simply fixed 1/d = 35 years (i.e., adults aged 15–49 years), which implies b = dN(0) = 2.86 × 104 per year. As previously estimated in [65,66], we assumed that β = 4.53 per year, and 1/μ = 35 years corresponding to HIV-1-infected individuals with the mean set-point viral load of 3.2 × 104 RNA copies/ml. The fraction, f, is fixed to be 0.60 in our simulations based on our findings in the stable A3H humanized mice infected with NLCSFV3 (Fig 3 & S6 Table). Our simulations well reproduced that the prevalence of hyper HIV-1 in the human population with different stable A3H proportion (Fig 5E).
The data are presented as averages ± SDs or SEMs. Statistically significant differences were determined by Student's t test, Paired t test, and Mann-Whitney U test. To determine statistically significant correlations (S9 Fig), the Spearman rank correlation test was applied to the data.
An accession number for the data generated in this study is as follows: the RNA-seq data of the splenic MNCs of HIV-1-infected (n = 4) and mock-infected (n = 4) humanized mice (GEO: GSE92262).
|
10.1371/journal.pntd.0002013 | Vectorial Capacity of Aedes aegypti for Dengue Virus Type 2 Is Reduced with Co-infection of Metarhizium anisopliae | Aedes aegypti, is the major dengue vector and a worldwide public health threat combated basically by chemical insecticides. In this study, the vectorial competence of Ae. aegypti co-infected with a mildly virulent Metarhizium anisopliae and fed with blood infected with the DENV-2 virus, was examined.
The study encompassed three bioassays (B). In B1 the median lethal time (LT50) of Ae. aegypti exposed to M. anisopliae was determined in four treatments: co-infected (CI), single-fungus infection (SF), single-virus infection (SV) and control (C). In B2, the mortality and viral infection rate in midgut and in head were registered in fifty females of CI and in SV. In B3, the same treatments as in B1 but with females separated individually were tested to evaluate the effect on fecundity and gonotrophic cycle length. Survival in CI and SF females was 70% shorter than the one of those in SV and control. Overall viral infection rate in CI and SV were 76 and 84% but the mortality at day six post-infection was 78% (54% infected) and 6% respectively. Survivors with virus in head at day seven post-infection were 12 and 64% in both CI and SV mosquitoes. Fecundity and gonotrophic cycle length were reduced in 52 and 40% in CI compared to the ones in control.
Fungus-induced mortality for the CI group was 78%. Of the survivors, 12% (6/50) could potentially transmit DENV-2, as opposed to 64% (32/50) of the SV group, meaning a 5-fold reduction in the number of infective mosquitoes. This is the first report on a fungus that reduces the vectorial capacity of Ae. aegypti infected with the DENV-2 virus.
| Dengue is a worldwide public health problem. There is not an effective vaccine yet; the chemical struggle against its transmitter, the mosquito Aedes aegypti, is onerous and erratic, and the community participation to eliminate vector breeding sites is unconfident. Here, we examined mosquitoes fed on human blood mixed with the Dengue virus, by exposure to the fungus Metarhizium anisopliae, to test whether the fungus halts the viral dissemination from midgut to head in co-infected (CI) insects. We found an overall viral infection rate in CI mosquitoes of 76% but infected or not, most (78%) died before or at day six post-infection; only six (12%) out of 50, survivors had virus in head and were potentially infectious at day seven post-infection. A higher infection (84%) was observed in single-virus infected mosquitoes, but they suffered only 6% mortality after 6 days and 32 (64%) survivors tested positive for virus in head after 7 days. Survival, fecundity and ovaric cycle of CI mosquitoes were reduced in 70, 52 and 40% in comparison to the ones of control. Therefore, if the fungus caused a 5-fold reduction in the number of infectious mosquitoes, it has potential to be evaluated against the Dengue transmitter in field.
| The susceptibility of Aedes aegypti adults to infection with Beauveria bassiana was first reported in the late 1960s [1]. However the potential of entomopathogenic Ascomycetes (Hypocreales) as adulticides of vector mosquitoes was largely overlooked until Metarhizium anisopliae was demonstrated to induce mortality of Culex quinquefasciatus and Anopheles gambiae [2]; and sequentially both M. anisopliae and B. bassiana have been tested against Ae. aegypti and Aedes albopictus [3]. The successful infection of adult female mosquitoes has been made via direct contact [4], [5] and also via auto-dissemination from males to females when mating [6], [7].
The increasing interest in exploring these fungi as biocontrol agents of dengue vectors stems from the fact that they are ubiquitously available, relatively cheap to mass-produce, and kill mosquitoes effectively [8]. In addition to the infection studies, attention has also been focused on other topics such as determining their safety to public health [9], and the effect of different surfaces on the infectivity of conidia to resting mosquitoes [10]. Likewise, some devices with inoculum baited with lures have also been tested for attracting and infecting adults to avoiding domiciliary sprayings [11].
Metarhizium anisopliae pathogenesis to insects has been widely documented [12]. The fungus is hemibiotrophic [13]. Conidia germination and cuticle perforation last around 24 hours [14]. After penetration, the pathogen produces hyphal bodies or blastospores invading the whole host's hemocele, depleting nutrients and killing the insect by starvation, dehydration, and toxemia [15]. It is therefore proposed that the rapid fungal invasion could affect the survival of the DENV virus if both are present in the same female of Ae. aegypti, weakening its vectorial competence. Here, we fed Ae. aegypti females with DENV-2-infected human blood, and/or exposed them to M. anisopliae conidia to produce single-fungus (SF), single-virus (SV) and co-infected (CI) mosquitoes. The parameters evaluated included mosquito survival, fecundity and first gonotrophic cycle (GC) length, plus the viral infection rate in the midgut and head.
The DENV-2 Yuc 18500 strain was isolated from blood of a sick person at Merida city in 2008; it is deposited at the “Collection of Arboviruses isolated at the Yucatán Peninsula” of the Regional Research Center “Dr Hideyo Noguchi”, University of Yucatan (UADY), Mexico; its use in this study was approved by written consent given by Dr. Fernando Andrade-Narvaez, Chair, Bioethics Committee of the Regional Research Center “Dr Hideyo Noguchi”, University of Yucatan (UADY), Merida, Yucatan, Mexico. In addition, all members of the Bioethics Committee provided informed consent. This strain was used to infect C6/36 Aedes albopictus cells at a starting viral titer of 1.5 particles/cell. C6/36 cells were grown in Leibovitz's medium but the infected ones were held in medium containing 2% fetal bovine serum. The Ma-CBG-1 strain of M. anisopliae was isolated from soil collected at rural habitats around the city of Saltillo; cultured on potato-dextrose-agar; and passaged three times through living hosts (Ae. aegypti females) before the study. At 12 days post-infection (PI) the virus was harvested. Mosquitoes used in the study were derived from a colony of Ae. aegypti that was established in 2008 with larvae collected in Monterrey, México. Four to seven day-old female Ae. aegypti mosquitoes were exposed to single infections of either Ma-CBG-1 M. anisopliae strain at 1.6×108 conidia mL−1 (SF) or DENV-2 (SV); and both fungus and virus (CI). Fungal infections were done as described previously [7]. For the virus infections, females were confined in 1-liter glass flasks, and were fed on 2,320 µL defibrinated human blood and 680 µL of virus suspension containing a titer of 1×107 plaque-forming units (PFU) mL−1, for 1 hour via a water-jacketed membrane feeding apparatus [16]. Control mosquitoes (C) and those in the SF group were fed with non-virus infected blood in the same manner. After blood feeding, mosquitoes were anaesthetized by exposing the flask to 4°C for 25 minutes, and then only blood-fed females were transferred to the holding containers for each bioassay.
Three bioassays (B) were conducted. In B1, the survival of Ae. aegypti females was compared between each treatment: CI, SF, SV, and control. Fifty mosquitoes per treatments were used, encompassing two replicates of 25 each in a 1-liter plastic flask. The 25 females per replicate were randomly selected from, those emerged from larvae (200 larvae/liter) of the same plastic tray and all replicates were conformed by adults emerged from larvae of different plastic trays, origin (eggs) and handling. Dead insects in treatments were recorded and removed daily. The cadavers were submerged twice in 1% chlorine solution, washed in distilled H2O, and placed in humid chambers for conidiation. This bioassay was run until the last insect died.
In B2 the viral infection rate in the midgut and head of mosquitoes was examined. Treatments were CI and SV, with also fifty insects per treatment and two replicates as in B1. Dead females were registered daily until six days PI without registering sporulation. At day 7 PI all surviving mosquitoes were cold-killed. Day 7 was the cut-off point because this is the average extrinsic incubation period (EIP) for the DENV-2 in Ae. aegypti [17], [18] and these studies were conducted at 30 (±2)°C. Mosquito midguts and heads were dissected on a glass slide containing 10 µL of phosphate-buffered saline; then were fixed and placed into 0.2 mL Gold-PCR tubes containing 150 µL of 4% paraformaldehyde and kept at 4°C until further analysis.
In B3, sixty females were used per treatment to assess the impact of fungal infection on fecundity and length of the first GC. The same treatments as in B1 were set up but with three replicates of 20 females each. However here, the mosquitoes of each replicate were individually separated into 40-ml capped-vials, containing a small amount of water and cardboard to record daily oviposition for each individual female.
In B2, the viral infection rate in the females sacrificed at 7 days PI, as well as in cadavers collected at 1–6 days, was determined using PCR. Details of protocols for viral RNA purification, cDNA synthesis, primer sequences, PCR conditions, and detection of PCR products by agarose gel electrophoresis, were reported earlier [16]. The preparation of midguts and heads, the immunofluorescent assay, and stain of female's tissues have also been published elsewhere [18].
Daily mortality rate was used to compute the median lethal time (LT50) per treatment with the Kaplan-Meier model; the model was stratified by replicate number to account for dependencies for mosquitoes which were held within each replicate. In B2, two 2×3 cross tabulation analyses by χ2 using Fisher's exact test were applied to the percentages of females with viral infection in the midgut, then with disseminated infection in head and non-DENV-2-infected, across the CI and SV treatments; the first one was an overall analysis while the second was only for the three groups of surviving females across both treatments on 7 days PI. The Fisher's exact test was used because the sample sizes were small. In B3, a 2×2 cross tabulation analysis by χ2 was applied to the percentages of ovipositing females in both CI and SV treatments; moreover, a one-way analysis of variance was applied for fecundity and GC; means were contrasted with a Ryan test. All analyses were performed with SAS [19].
In B1, the overall survival varied among the four groups (χ2 = 237.25, df = 3, p<0.0001); further analysis indicated that there was no statistical difference in the survival of DENV-2 infected (CI) and non-DENV-2-infected Ae. aegypti females (SF) (χ2 = 2.87, df = 1,p>0.05). The LT50 of fungal infected females was 6.93 (range, 6.59–7.27 days, 50 samples) for CI and 7.23 (range, 6.80–7.66 days, 50 samples) for SF. The same occurred for the survival of females infected only with the virus (SV) which was similar to the uninfected controls (χ2 = 0.21, df = 1, p>0.05). The LT50 for SV and control mosquitoes was 24.00 (range, 23.08–24.98 days, 50 samples) and 24.83 (range, 24.06–25.60 days, 50 samples) days respectively. Overall, M. anisopliae reduced the survival (as indicated by the LT50 values) of both CI and SF mosquitoes by ≈70% (Figure 1). The sporulation rates of cadavers collected from the CI and SV treatments was 85% on average for both experimental factors. Therefore, regardless of the virus, the fungus killed 85% of mosquitoes in both treatments.
In B2 at 7 days PI, the mortality rate of SV females was only 6% (3/50; Figure 2a), while in CI mosquitoes there was 78% (39/50; Figure 2b) mortality. For the SV treatment after 7 days, the percentage of surviving mosquitoes with virus in the midgut, head or non-DENV-2-infected was 15% (7), 64% (32) and 17% (8) respectively. For the CI treatment after 7 days, the percentage of surviving mosquitoes with virus in the midgut, head and non-DENV-2-infected was 10% (5), 12% (6) and 0% (0) respectively. As such, for mosquitoes surviving to 7 days PI, there was a reduced proportion which were able to develop a head infection between the CI and SV groups (χ2 = 6.14, df = 2, p<0.05). However, an assessment of vectorial capacity should also account for mortality. When mortality is considered, 64% (32/50) of the mosquitoes in the SV treatment were alive and potentially able to transmit at 7 days PI; compared with only 12% (6/50) in the CI group (χ2 = 17.99, df = 1, p<0.0001). Therefore, there was a 5-fold reduction in the number of potential infective females due to the high mortality of females before they were able to complete the EIP.
In B3, the percentage of fungus-infected individuals which oviposited within 7 days was only 56% (23/41) for CI and 42% (18/42) for SF. This was ≈50% less females than observed for the non-fungus treatments where 100% of females oviposited (SV = 59/59 and Control = 59/59 (χ2 = 294.00, df = 3, p<0.0001). Of the females which laid at least one egg, fungus-infected females generally laid less eggs. The mean number of eggs per female for fungal-infected treatments were 21.75 (range, 19.10–24.40 eggs, 20 samples) in CI and 21.65 (range, 18.19–25.11 eggs, 20 samples) in SF, contrasting with the non-fungus treatments where the mean number of eggs per female was 45.76 (range, 40.45–51.07 eggs, 20 samples) in SV and 46.58 (range, 41.23–51.93 eggs, 20 samples) in CI. Therefore, the fecundity of Ae. aegypti was reduced by 52% (Figure 3) (F = 22.95, df = 3, p<0.001). The infection of females with M. anisopliae was also observed to accelerate the oogenesis. The GC of fungus-infected females was 2.65 (range, 2.48–2.82 days, 20 samples) in CI and 3.46 (range, 3.25–3.67 days, 20 samples) in SF and both of these were shorter than the GC of SV and uninfected mosquitoes which were 5.31 (range, 4.96–5.66 days, 20 samples) and 5.35 (range, 4.91–5.79 days, 20 samples) days respectively (F = 14.15, df = 3, p<0.001). Thus, the fungal co-infection diminished the length of the GC by 40% (from 5 to 3 days).
In Ae. aegypti co-infected with DENV-2 and with a Mexican strain of M. anisopliae there was a 5-fold reduction in the number of mosquitoes which survived the EIP and with potential to transmit dengue. The most sensitive component of vectorial capacity (C) as defined by the Ross-MacDonald model is daily survival [20], [21]. M. anisopliae infection killed the majority of females (78–88%) before surviving the 7-day EIP. Furthermore, of the fungus-infected mosquitoes which did survive the EIP, there was a reduced chance of them becoming infectious after a feeding on DENV-2 infected blood. To express more clearly the impact of the fungus on C of the dengue vector, we computed this index taken our own data and others from literature: Daily survival probability (p) for SV and CI females was computed by regressing the number of survivors [ln (X+1)] on days PI up to day 6, and were 0.98 and 0.75 with determination coefficients of R2 = 0.62 and 0.98, respectively (Figure 4). These different rates mean a reduction in p of 0.23 by the fungal effect. The C model ma3bpn/-ln(p) and their components are defined as follows: m = “daily biting rate”, a3 = “the human biting rate” powered to the number of blood meals per GC, which is at least three for Ae. aegypti [22]; a = 1/GC, b = proportion of infectious females at the EIP which were 0.64 and 0.12 for SV and CI groups, respectively, and pn/-ln(p) = the expected infective life in days after the EIP. Now in Monterrey, MX, the Ae. aegypti annual population is bimodal with the highest peak in October, where a time series of 19 consecutive days of human-landing captures allowed to estimate a m = 37 bites/human/day [23]; then keeping constant m = 37, a = 1/GC days (5 days for SV and 3 for CI) and n = EIP = 7 days, the calculation results in a C = 8.14 and 0.07 for SV and CI, respectively; this means that the fungus reduced the C by 116 times in CI compared to the one of SV-infected females. This would be a drastic impact of M. anisopliae on the vectorial capacity of Ae. aegypti in field.
Comparatively, the 64% of females that fed on DENV-2 infected blood and became infectious within 7 days, is two-fold superior than 27% and 30% recorded in Ae. aegypti from Texas, USA, and Chiapas, Mexico infected with the DENV-2 Southeast Asia strain, respectively [24], but is similar to the competence of Ae. aegypti from Australia [25].
The fungus was able to quickly invade the tissues and cells causing the early death of CI mosquitoes before many of them were able to survive the entire EIP. This observation is supported by previous work which detected M. anisopliae hyphal bodies circulating in the hemolymph of Locusta migratoria manilensis at day 2 PI [26]. Generally cuticle perforation and hemocele invasion last around 20 hours [14]. Once the host is invaded, the fungus starts to consume nutrients and propagate. By day 2–3 PI the fungus releases detectable levels of toxins and enzymes [27], [28], compromising the immune system of the mosquitoes triggering mortality, as was observed in CI and SF females by 3 days PI. Similarly, this competition for nutrients between the host and fungus, could also explain the 52% reduction in fecundity and a shortening of the GC from 5 to 3 days in B3. In a previous study, we conducted with this fungus but with a highly virulent strain (Ma-CBG-2) found that in mycosed females the fecundity was reduced to almost zero [7]. In addition, the Anopheles gambiae female mosquitoes tend to take smaller blood meals after becoming fungus-infected [29], [30] and this also plausibly explains the reduced number of females which actually laid eggs. In the current experiments, we controlled for any possible biases of dengue infection on fecundity by using the four experimental factors (CI, SV, SF and control). It is a general knowledge that entomopathogens usually induce changes in vector behavior and physiology [31], [32]. The shorten of GC observed in B3 could be part of an adaptive strategy by the female mosquitoes in achieving reproduction before the pathogens drastically deplete the nutritive resources required for the egg development. It could also be an adaptive behavior by the mosquitoes which are unlikely to survive the virulence of the pathogens as earlier observed in crickets infected with bacteria and parasites [33], [34], while the oogenesis of Schistocerca gregaria and Ae. aegypti was speeded up by M. anisopliae and B. bassiana [35], [36]. However, instead of curtailment of reproductive potential of the insects, some pathogens have been known to enhance the reproductive success of their host through higher production of offspring early in life; a phenomenon often referred as “fecundity compensation” [37]. These reactions may partly be mediated by the immune system of the insects in response to the infection as earlier posited by researchers [33], [37].
Concerning the impact of the virus, no effect of dengue infection on fecundity was noted, contrary to previous research. There is some evidence that the dengue virus exerts low mortality rates on Ae. aegypti adults, as well on its fecundity, as was recently reported for Brazil, where 4–5 day-old Ae. aegypti females were fed with rabbit blood mixed with 3.6×105 PFU mL−1 of the DENV-2 (strain 16681) [38]. The authors found that the longevity of infected mosquitoes averaged 26 days, which is similar to the 24 days observed here in SV females. They reported a viral infection of 66% which is also similar to the 68% (6 females with virus in head and 5 with virus in midgut) obtained in SV females at 7 days PI that we found; the viral effect on fecundity was examined by a logistic regression; however the authors only reported that the mean number of eggs per GC through five GCs tended to be lower in infected mosquitoes, without mentioning the specific means. The reduced fecundity may be beneficial for mosquito control because the population size of consecutive generations will be proportionally smaller; however this reduction occurred concomitantly with a shorter GC, which not necessarily implies only a faster reproduction. A shorter GC may also alter the natural mortality of populations by increased exposure to predation and other adverse environmental factors when search more often for blood-meals. It is possible that the direct mortality observed in our laboratory study may be compounded such indirect increases of mortality under field conditions.
In conclusion, M. anisopliae has the potential to drastically affecting the vectorial capacity of Ae. aegypti in the field and could be accomplish without necessarily using a highly virulent strain (LT50 = 3–4 days) of the fungus. Whether these conditions could be fulfilled in field is the aim for an ongoing investigation by our team.
|
10.1371/journal.pcbi.1004509 | Beyond the E-Value: Stratified Statistics for Protein Domain Prediction | E-values have been the dominant statistic for protein sequence analysis for the past two decades: from identifying statistically significant local sequence alignments to evaluating matches to hidden Markov models describing protein domain families. Here we formally show that for “stratified” multiple hypothesis testing problems—that is, those in which statistical tests can be partitioned naturally—controlling the local False Discovery Rate (lFDR) per stratum, or partition, yields the most predictions across the data at any given threshold on the FDR or E-value over all strata combined. For the important problem of protein domain prediction, a key step in characterizing protein structure, function and evolution, we show that stratifying statistical tests by domain family yields excellent results. We develop the first FDR-estimating algorithms for domain prediction, and evaluate how well thresholds based on q-values, E-values and lFDRs perform in domain prediction using five complementary approaches for estimating empirical FDRs in this context. We show that stratified q-value thresholds substantially outperform E-values. Contradicting our theoretical results, q-values also outperform lFDRs; however, our tests reveal a small but coherent subset of domain families, biased towards models for specific repetitive patterns, for which weaknesses in random sequence models yield notably inaccurate statistical significance measures. Usage of lFDR thresholds outperform q-values for the remaining families, which have as-expected noise, suggesting that further improvements in domain predictions can be achieved with improved modeling of random sequences. Overall, our theoretical and empirical findings suggest that the use of stratified q-values and lFDRs could result in improvements in a host of structured multiple hypothesis testing problems arising in bioinformatics, including genome-wide association studies, orthology prediction, and motif scanning.
| Despite decades of research, it remains a challenge to distinguish homologous relationships between proteins from sequence similarities arising due to chance alone. This is an increasingly important problem as sequence database sizes continue to grow, and even today many computational analyses require that the statistics of billions of sequence comparisons be assessed automatically. Here we explore statistical significance evaluation on data that is stratified—that is, naturally partitioned into subsets that may differ in their amount of signal—and find a theoretically optimal criterion for automatically setting thresholds of significance for each stratum. For the task of domain prediction, an important component of efforts to annotate protein sequences and identify remote sequence homologs, we empirically show that our stratified analysis of statistical significance greatly improves upon a combined analysis. Further, we identify weaknesses in the prevailing random sequence model for assessing statistical significance for a small subset of domain families with repetitive sequence patterns and known biological, structural, and evolutionary properties. Our theoretical findings in statistics are relevant not only for identifying protein domains, but for arbitrary stratified problems in genomics and beyond.
| The evaluation of statistical significance is crucial in genome-wide studies, such as detecting differentially-expressed genes in microarray or proteomic studies, performing genome-wide association studies, and uncovering homologous sequences. Different biological applications have settled for different statistics to set thresholds on. In biological sequence analysis, accurate statistics for pairwise alignments and their use in database search [1–3] were introduced with the use of random sequence models and E-values two decades ago [4,5]. Sequence similarity searches have evolved further, from the pairwise comparison tools of FASTA [3] and BLAST [5], to sequence-profile [6–8] and profile-profile [9–12] comparisons. While different approaches to detect sequence similarity have relied on a variety of statistics, including bit scores [13,14] and Z-scores [3], most modern approaches are based on E-values.
Detecting sequence similarity in order to uncover homologous relationships between proteins remains the single most powerful tool for function prediction. Many modern sequence similarity approaches are based on identifying domains, which are fundamental units of protein structure, function, and evolution. Homologous domains are grouped into “families” that may be associated with specific functions and structures, and these domain families organize protein space. Domain families are typically modeled with profile hidden Markov models (HMMs) [13,15]. There are many domain HMM databases, each providing a different focus and organization of domain space, including Pfam [14], Superfamily [16], and Smart [17]. Although HMM-based software, such as the state-of-the-art HMMER program [18], has features that make it superior to its predecessors, accurate significance measures arose only recently [19].
At its core, domain prediction is a multiple hypothesis testing problem, where tens of thousands of homology models (one for each domain) are scored against tens of millions of sequences. Each comparison yields a score s and a p-value, defined as the probability of obtaining a score equal to or larger than s if the null hypothesis holds. While a small p-value threshold (for example, 0.05 or smaller) is acceptable to declare a single test significant, this is inappropriate for a large number of tests. Instead, thresholds for domain prediction are typically based on the E-value. The E-value can be computed from a p-value thresholds as E = pN, where N is the number of tests, and yields the expected number of false positives at this p-value threshold. E-value thresholds make sense for a single database search, especially if few positives are expected. However, E-values are less meaningful when millions of positives are obtained, and a relatively larger number of false positives might be tolerated. Moreover, in multiple database query problems, such as BLAST-based orthology prediction [20] or genome-wide domain prediction [21], E-values are usually not valid because many searches are performed without the additional multiple hypothesis correction required.
Control of the False Discovery Rate (FDR) is an alternative and appealing approach for multiple hypothesis testing [22]. The FDR is loosely defined as the proportion of all significant tests that are expected to be false, and can be estimated as the E-value divided by the number of predictions made. The FDR does not increase with the database size N the way the E-value does; thus, predictions do not usually lose significance with the FDR as the database grows. The FDR also does not require additional correction in the case of multiple database queries. The FDR is controlled from p-values using the Benjamini-Hochberg procedure [22]. The q-value statistic is the FDR-analog of the p-value, and it provides conservative and powerful FDR control [23]. The q-value of a statistic t is the minimum FDR incurred by declaring t significant [23]. Thus, q-values vary monotonically with p-values, and they are easily estimated from p-values [23]. While E-values control the number of false positives, q-values control their proportion. The local FDR (lFDR) measures the proportion of false positives in the infinitesimal vicinity of the threshold, and hence it is a “local” version of the FDR [24]; it is also equivalent to the Bayesian posterior probability that a prediction is false [24]. However, q-value estimates are much more robust than lFDR estimates, since the former are based on empirical cumulative densities, which converge uniformly to the true cumulative densities [25,26]. On the other hand, lFDR estimates are local fits to the density, so they are comparably more susceptible to noise, especially on the most significant tail of the distribution. The FDR [22], q-value [23], and lFDR [24] have all been successfully used in many areas of bioinformatics, including gene expression microarray analysis [24,27,28], genome-wide association studies (GWAS) [27,29], and proteomics analysis [30–34].
Here we introduce the first FDR- and lFDR-estimating algorithms for domain prediction. An essential feature of our approach is that statistical tests are stratified by domain family, rather than pooled. We prove that stratified problems are optimally tackled using the lFDR. For domain prediction, we evaluate how well thresholds based on stratified lFDRs and q-values perform using five independent approaches for estimating empirical FDRs. Through extensive benchmarking using the Pfam database and HMMER, we find that using stratified q-values increases domain predictions by 6.7% compared to the Standard Pfam thresholds on UniRef50 [35]. In contrast to theory, we also find that q-values outperform lFDRs. Further, while the empirical FDRs for most domain families agree with our q-value thresholds, some families tend to have larger FDRs; the standard null model appears to be inappropriate for them and yields inaccurate p-values. Specifically, families with larger-than-expected empirical FDRs are enriched for those containing repetitive patterns, such as coiled-coils, transmembrane domains, and other low-complexity regions. When only families with as-expected FDRs are considered, the use of q-values increases domain predictions by 8.8% compared to the Standard Pfam, and lFDRs further outperform q-values, suggesting that further performance improvements are possible if the statistical modeling of repetitive families is improved.
Stratified FDR analyses have been previously explored [36–39], and have been successfully applied to GWAS in particular [29,40,41]. Thus, the same solution we introduce for domain recognition applies to a wide variety of problems in which statistical tests can be analyzed separately, including GWAS (stratifying by candidate or genic regions), orthology prediction (stratifying by each ortholog database search), motif scanning (stratifying by each motif search across a genome), multi-microarray analysis (stratifying by each microarray), and other multi-dataset analyses. Overall, we expect the use of stratified q-values and lFDRs to yield improvements in many applications in bioinformatics and beyond.
We briefly review the relevant FDR definitions; for a comprehensive overview, see [42]. Given a p-value threshold t, let V be the number of false positive predictions, and R be the total number of significant tests. Assuming independent p-values drawn from a two-component distribution of null and alternative hypotheses (Fig 1), V and R have expected values of
E[V(t)]=tπ0N,E[R(t)]=F(t)N,
where π0 is the proportion of tests which are truly null, N is the total number of tests, and F(t) is the cumulative density of p-values [23,24,43]. Note that E[V(t)] gives the E-value.
There are two closely-related versions of the FDR used in our work: the positive FDR (pFDR) and marginal FDR (mFDR) [42,43], defined as
pFDR=E[VR|R>0],mFDR=E[V]E[R].
The advantages of the pFDR compared to the original FDR definition of Benjamini and Hochberg [22] are discussed in [43]. If p-values are drawn independently from the two-component distribution of Fig 1, the pFDR and mFDR were proven to be equivalent to the following posterior probability [43]:
pFDR(t)=mFDR(t)=Pr(H=0|p≤t)=tπ0F(t),
where H = 0 denotes that the null hypothesis holds. This quantity is sometimes called the “Bayesian FDR” [24]. The pFDR and mFDR are also asymptotically equal under certain forms of “weak dependence,” as defined in [44]. Our domain prediction problem has large sample sizes and weak dependence: our dataset contains millions of protein sequences and thousands of HMMs, and null p-values are only dependent for very similar sequences and similar HMMs. Dependent tests represent a very small subset of all hypotheses tested, even on each stratum (for any one HMM). For this reason, we use FDR to refer loosely to all these FDR definitions.
The local FDR (lFDR) is the Bayesian posterior error probability defined as [24]
lFDR(t)=Pr(H=0|p=t)=π0f(t),
where f(t) = F'(t) is the p-value density at t. Thus, while the pFDR is a ratio of areas, the lFDR is a ratio of densities (Fig 1) [45].
The q-value of a statistic t is the minimum pFDR incurred by declaring t significant [23]. Estimated q-values are efficiently constructed from p-values, and conservatively estimate the pFDR [23]. Specifically, q-value and lFDR estimation are based on the above formulas, where π0, F(t) and f(t) are replaced by estimates. See the Supp. Methods in S1 Text for the algorithms for estimating q-values and lFDRs.
Here we prove that the lFDR gives optimal thresholds for stratified problems. For domain prediction, each domain family defines a stratum. We wish to find p-value thresholds ti per stratum i that maximize the number of predictions across strata while constraining the maximum FDR of the strata combined. Optimality of the lFDR here is consistent with the related Bayesian classification problem, where posterior error probabilities are also optimal [43].
Let the FDR model quantities Ni, π0,i, Fi(ti) and fi(ti) be given per stratum i. We desire to maximize the expected number of predictions across strata
∑iFi(ti)Ni,
while constraining the “combined” FDR, which we define as the sum of expected false positives across strata divided by the total number of expected predictions, to a maximum value of Q, or
∑itiπ0,iNi∑iFi(ti)Ni≤Q.
This problem is solved using the Lagrangian multiplier function Λ, with the constraint set to strict equality, in a formulation that avoids quotients:
Λ=∑iFi(ti)Ni+λ(∑itiπ0,iNi−Q∑iFi(ti)Ni)=∑iFi(ti)Ni(1−λQ)+λtiπ0,iNi.
Taking the partial derivative of Λ with respect to tj, we obtain a necessary condition for optimality,
∂Λ∂tj=fj(tj)Nj(1−λQ)+λπ0,jNj=0⇔Q−1λ=π0,jfj(tj)=lFDRj(tj),
which shows that the lFDR of each stratum must be equal, since the last equation has the same value for every j. Optimality of the lFDR also holds when constraining the combined E-value instead of the combined FDR (Supp. Methods in S1 Text).
Each of the 12,273 Pfam domain families was used to scan for domains in each of 3.8 million proteins of UniRef50 (Supp. Methods in S1 Text), resulting in a total of 47 billion tests. Domain predictions are stratified by family (HMM), and each stratum contains p-values from which we estimate q-values and lFDRs. We note that standard q-value and lFDR implementations fail for domain data for two reasons. First, modern HMM software only reports the smallest p-values due to heuristic filters [19]. Second, homologous families (grouped into “superfamilies” [16] or “clans” [14]) produce frequent overlaps that are resolved by removal of all but the most significant match, and thus there are fewer predictions than an independent family analysis would predict, which leads to underestimated FDRs. To address these issues, we remove overlapping domains (keeping those with the smallest p-values), and then estimate q-values and lFDRs with methods adapted for censored p-values (Methods). For comparison, we also use E-value thresholds and the “Standard Pfam” curated bitscore thresholds (also called “Gathering” or “GA” [14]). Note that a stratified E-values approach (separating families) is no different from a combined E-value approach in that the ranking of predictions is preserved, since the number of proteins, or tests, is the same per stratum; the stratified E-value threshold equals the combined E-value threshold divided by the number of strata. Similarly, a combined q-value or lFDR approach (obtained by combining the p-values of all strata) also preserves the E-value rankings.
We estimate the true FDR via “empirical” FDR tests, to compare all methods on an equal footing, but also to test the accuracy of q-value estimates. We created or adapted five tests, each of which labels domain predictions as either true or false positives (TP, FP) using different statistical and biological criteria. The proportion of predictions labeled FP estimates the FDR.
For simplicity, only two tests are described here in detail and are featured in the main figures. First, the ClanOv (“Clan Overlap”) test is based on the expectation that overlapping domain predictions should be evolutionarily related [46]. Pfam annotates related families via clans. In this test, domain predictions are ranked by p-value, highest ranking domains are considered as TPs, domains that overlap a higher-ranking domain of the same clan are removed (since they would not be counted as separate predictions), and domains that overlap a higher-ranking domain of a different clan are considered FPs (Methods, Fig 2A). All FPs in this test would not be predicted by our method when overlaps are removed; nevertheless, this method estimates well the amount of noise. Second, the ContextC (“Context Coherence”) test is based on whether domain pairs predicted within a sequence have been observed together before [47]. For each sequence, domain predictions are ranked by p-value, and the highest ranking domain is always a TP. Subsequently, a domain is a TP if its family has previously been observed with the family of at least one higher-ranking domain, and otherwise it is a FP (Methods, Fig 2B).
The principles behind the other three tests are described here briefly: OrthoC (“Ortholog Set Coherence”) is based on the expectation that orthologous proteins contain similar domains [48], RevSeq (“Reverse Sequence”) estimates noise based on domains predicted on reversed amino acid sequences [49], and MarkovR (“Markov Random”) estimates noise based on domains predicted on random sequences generated from a second-order Markov model (Supp. Methods and Fig A in S1 Text).
Methods are compared at the same empirical FDR based on the number of domain predictions (Fig 3 and Fig B in S1 Text), unique families per protein (Fig C in S1 Text), amino acids covered (Fig D in S1 Text), and proteins with predictions (Fig E in S1 Text), as well as their total “GO information content” scores (derived from the Gene Ontology [50] and MultiPfam2GO [51]; Supp. Methods and Fig F in S1 Text).
Stratified q-value thresholds outperform E-values in all tests (Fig 3, Fig B in S1 Text). While stratified lFDR thresholds are superior to E-values in all tests, they are unexpectedly outperformed by q-values on most tests. We hypothesize that lFDR estimates are less robust than q-values due to errors in p-values; these errors most likely arise because of weaknesses in the standard null model. The Standard Pfam is not evaluated using ClanOv and ContextC (Fig 3); these tests are based on the Pfam clans and observed domain pairs, so the Standard Pfam has zero empirical FDRs in both. However, q-values outperform the Standard Pfam in two of the three fair tests (OrthoC, MarkovR) and perform similarly in RevSeq (Fig B in S1 Text). The same trends hold if the combined empirical E-value is controlled (Supp. Methods and Fig G in S1 Text).
We measure improvements not only of domain counts, which may be inflated for families with many small repeating units, but also of unique family counts. We also measure the information content based on the GO terms associated with domain predictions [51] (Supp. Methods in S1 Text). To have amounts of noise comparable to Pfam, we calculate p- and q-value equivalents to the Standard Pam thresholds for each family (Supp. Results in S1 Text). The medians of these distributions give thresholds of q ≤ 4e-4, and for E-values, p ≤ 1.3e-8 (Supp. Results in S1 Text). Q-values improve all metrics consistently relative to the Standard Pfam (between 4–7%, Fig 4). E-values predict 2% fewer domains than the Standard Pfam, but slightly outperform Pfam in the other metrics (Fig 4).
We also evaluated dPUC, a prediction method based on domain context [48,52]. dPUC also improves upon the Standard Pfam in all cases (Fig 4). dPUC increases domains more than q-values, but their unique family count and amino acid coverage are comparable, and q-values best dPUC for protein counts and GO information content. This is because dPUC predicts more repeat domains (of the same family) and tends to restrict new predictions to proteins that already had Standard Pfam predictions. In contrast, q-values increase domains at the same rate as they increase protein coverage, which increases information the most. Thus, while stratified q-values predict fewer domains than dPUC, those domains tend to be more informative than the dPUC predictions at comparable FDRs.
We find large disagreements between q-values and our empirical FDRs tests (except for MarkovR; Fig 5, Fig H in S1 Text). Interestingly, the disagreement is proportionally larger for smaller FDRs, and shrinks as the FDR grows (Fig 5). We hypothesize that a few families are too noisy at stringent thresholds, and this subset becomes proportionally smaller as all families are allowed greater noise. To test this, we compute empirical FDRs separately per family at a threshold of q ≤ 1e-2 (Methods). This threshold gives a greater FDR than the Standard Pfam (Supp. Results in S1 Text), which is desirable here as many families have few predictions at more stringent thresholds. Since large deviations between the empirical FDRs and q-values may arise due to low sampling, significance is assessed by modeling this random sampling (Methods). We find that most families (92–99%, Table A in S1 Text) have FDRs close to the q-value threshold or have statistically insignificant differences (blue and black data in Fig 6, Fig I in S1 Text).
Four tests (ClanOv, ContextC, OrthoC, and RevSeq) detect many families with significantly larger FDRs than expected (3–8%, Table A in S1 Text). These families are significantly enriched for those containing coiled-coils, transmembrane domains, and low-complexity regions (Fig 7; Methods). There are fewer families with significantly smaller FDRs than expected (0–2%, Table A in S1 Text), and they do not appear to share common patterns. Only the MarkovR test conforms to expectation, with no families having significantly larger FDRs than expected and 0.1% of families having significantly smaller FDRs than expected.
We use the four tests (excluding MarkovR) to assign families into mutually-exclusive classes by majority rule. The “increased-noise” families have significantly large positive deviations (see Methods; red in Fig 6) in at least three tests. The “decreased-noise” families have significantly large negative deviations (green in Fig 6) in at least three tests. Lastly, the families with “as-expected-noise” have small deviations (blue and some black in Fig 6) in at least three tests. There are 327 increased-noise families (2.7% of Pfam, S1 File), one decreased-noise family (HemolysinCabind), and 4433 as-expected-noise families (36%, S2 File). There are 7512 unclassified families in Pfam (61%). Using these classes, we find that the Standard Pfam has more stringent thresholds (in terms of q-values) for increased-noise as compared to as-expected-noise families, but many increased-noise family thresholds remain too permissive (Supp. Results and Fig J in S1 Text).
Empirical FDRs agree more with q-values in as-expected-noise families than in all families combined, although some disagreement remains (Fig K in S1 Text). In these families, lFDRs outperform q-values (Fig 8), as we expect from our theoretical results when the underlying p-values are correct. Compared to the Standard Pfam, domain counts at q ≤ 4e-4 increase from 6.7% in all families to 8.8% in as-expected noise families (similar increases are observed on all metrics; Fig L in S1 Text), and lFDRs further improve upon q-values. Thus, lFDRs may become more useful should p-values for all families improve in the future.
The previous methods describe a single “domain” threshold set via the stratified q-value or lFDR analysis. However, HMMER provides additional information in the form of “sequence” p-values, which score the presence of domain families combining the evidence of repeating domains. Only 2.3% families have different sequence and domain Standard Pfam thresholds [14]. Here we define “two-tier” thresholds using the FDR. In the first tier, we compute q-values from the sequence p-values and set the threshold Qseq. In the second tier, we compute q-values on the domain p-values, only for the domains in sequences that satisfied the sequence threshold, and set the threshold Qdom|seq (corresponding to a FDR conditional on the first threshold). The final FDR is approximately Qseq+Qdom|seq if both thresholds are small and under an independence assumption (Supp. Methods in S1 Text). For simplicity, we only evaluate the case where Qseq = Qdom|seq.
Tiered q-values predict many more domains, at any fixed empirical FDR, than domain q-values and domain lFDRs, our previous two best statistics, consistently and by very large margins (Fig B in S1 Text). Tiered q-values also outperform other methods in predicting new families per sequence (Fig C in S1 Text); the entire signal of these families comes from combining repeating units, none of which is significant by itself. There is also a large increase in amino acid coverage (Fig D in S1 Text), and a smaller increase in protein coverage (Fig E in S1 Text) and GO information content (Fig F in S1 Text). Tiered q-values also compare favorably to dPUC [48], matching the superior domain improvements of dPUC, and outperforming dPUC in all other metrics (Fig M in S1 Text). Thus, tiered q-values retain the strengths of domain q-values while powerfully leveraging the limited context information of repeating domains present in sequence q-values. However, the estimated FDRs of tiered q-values are less accurate than for domain q-values (Fig H in S1 Text), and remain less accurate in as-expected-noise families (Fig K in S1 Text). For this reason, tiered stratified q-values are experimental: although they are more powerful than domain-only q-values, they do not, as described, control the FDR as well.
In multiple hypothesis testing, the FDR and lFDR are straightforward approaches for controlling the proportion of false positives and the posterior error probability, respectively. The q-value is a statistic for controlling the FDR that is less biased and more flexible than previous FDR procedures such as the one from Benjamini and Hochberg [22]. Benchmarks based on empirical FDRs have been a part of recent works studying protein and DNA homology [47,48,52,53]; however, those approaches have used expensive simulations rather than estimating FDRs directly from p-values (or E-values), as q-values do very efficiently. Our work is, to the best of our knowledge, the first attempt at applying q-values and lFDRs to domain identification, thus advancing the statistics of this field.
Our theoretical work revealed that the lFDR, which is the Bayesian posterior probability that a prediction is false, is the optimal quantity to control in stratified problems. Stratified lFDR control has previously been found to optimize stratified thresholds in the related problem of minimizing the combined false non-discovery rate while controlling the combined FDR [37]. The lFDR also arises naturally in Bayesian classification problems [43]. Stratified lFDR thresholds ensure the least confident predictions of each stratum have the same posterior error probability. However, we found that estimated q-values are more robust than our lFDR estimates for domain predictions, where the underlying p-value estimates are imperfect [45] (Fig 3).
We extended the domain stratified q-value approach into what we call tiered stratified q-values, by setting q-value thresholds on both the sequence and domain statistics reported by HMMER. While accurate FDR estimation of this procedure remains a challenge, tiered q-values successfully leverage the additional signal of repeating domains to increase predictions (Fig M in S1 Text). There are other successful approaches, such as dPUC [48] and CODD [52], that use the broader concept of domain context (or co-occurrence) to improve domain predictions. Remarkably, tiered q-values perform as well or better than as dPUC under all metrics (Fig M in S1 Text), even though tiered q-values only utilize the context signal of repeating domains, while dPUC additionally considers context between families [48]. In the future, tiered q-values could be combined with dPUC to yield further improvements in domain prediction.
We introduced a suite of empirical FDR tests to evaluate domain predictions. Altogether, these tests are powerful means for evaluating the correctness of predictions (“Evaluation of empirical FDR tests” Supp. Results in S1 Text). Four of our tests consistently revealed flaws in the estimates of statistical significance for some families. We found a strong enrichment among noisy families for coiled coils, transmembrane domains, and other low-complexity regions. These problematic domain categories have been noted elsewhere [46,54,55], and ad hoc solutions have been proposed [54,56]. However, none of these solutions are implemented by standard software such as BLAST and HMMER [56]. In our view, obtaining correct statistics for these repetitive families should be the top priority of the field of sequence homology. Nevertheless, most families in Pfam appear to have correct statistics, and the advantage of using q-values and lFDRs is clear. In the future, the standard sequence similarity software packages should be able to report these stratified statistics natively rather than as a post-processing step as is done here.
Domain prediction is one case where stratified FDR and lFDR control are desirable, since domain families occur with vastly different frequencies and are thus associated with differing amounts of true signal. However, the same holds for other applications, such as BLAST-based orthology prediction [20], since some ortholog groups are orders of magnitude larger than others. FDR and lFDR control may also improve iterative profile database searches, such as PSI-BLAST [6], as well as numerous other sequence analysis problems.
The basis of our work is a general theorem applicable to naturally stratified statistical tests. Whether the combined FDR or E-value is constrained, equal stratified lFDR thresholds are required to maximize predictions. Besides limits on sample size, the strata may be arbitrary, so our result can be broadly applied to multiple hypothesis testing problems. In motif scanning, for example in silico transcription factor (TF) binding site identification, the position weight matrix of each TF may yield a p-value per match [57], and the number of binding sites per TF may vary by orders of magnitude across different TFs. Here, we recommend computing lFDRs stratified by TF, and setting equal lFDR thresholds across TFs. For protein domains, one could further stratify p-values using taxonomy, since domain family abundances vary greatly across the kingdoms of life (archaea, bacteria, eukarya, and viruses) [58,59]. In sum, we have demonstrated the practical utility of our theoretical contributions to domain prediction, which are likely to influence many applications in bioinformatics and beyond.
A p-value distribution is required to estimate q-values and lFDRs. HMMER reports two kinds of p-values. The “sequence” p-value combines every domain of the same family on a protein sequence, while the “domain” p-value is limited to each domain instance. The sequence p-value thus reports whether the protein sequence as a whole contains similarity to the HMM, whereas the domain p-value scores individual domain units within the sequence. We obtained domain predictions with p-values on UniRef50 [35] and OrthoMCL5 [20] proteins using hmmsearch from HMMER 3.0 and HMMs from Pfam 25 with these parameters: the heuristic filters “--F1 1e-1--F2 1e-1--F3 1e-2” allow sequence predictions with “stage 1/2/3” p-value thresholds of 0.1, 0.1, and 0.01, respectively. Moreover, we obtain p-values using “-Z 1--domZ 1”. Lastly, we remove domains with p>0.01 by adding “-E 1e-2--domE 1e-2”.
For each domain family HMM, we use its HMMER p-values over a protein database to estimate q-values and lFDRs. We use standard methods [27,45] adapted for censored tests since HMMER3 only reports the most significant p-values while standard methods require all p-values. Notably, HMMER3 does not provide complete p-values even if filters are removed [60], and only small p-values are accurate [19], so the full set of p-values is not useful. Moreover, the filters are desirable to reduce HMMER3's runtime. The Supp. Methods (S1 Text) reviews these standard methods for estimating q-values and lFDRs, and details our adaptations for domains. Briefly, we remove overlaps between domain predictions ranking by p-value, before computing q-values and lFDRs; otherwise, the amount of true positive may be overestimated because overlapping domains will be counted double, a common case within Pfam clans. Secondly, the standard approaches require all p-values solely to estimate π0, here roughly the proportion of proteins that do not contain a domain family. We set π0 = 1, which gives slightly more conservative q-values and lFDRs than otherwise. Our software for computing stratified q-values, lFDR estimates and tiered q-values from HMMER3 is DomStratStats 1.03, available at https://github.com/alexviiia/DomStratStats.
We compare new and standard domain prediction approaches over a range of relevant empirical FDRs. We vary thresholds based on stratified q-values and lFDRs, and compare their performances to thresholds varied by E-values and extensions of the Standard Pfam. Stratified domain E-values are computed from the HMMER p-values by multiplying them by the number of proteins in UniRef50, as hmmsearch would compute them. The “Standard Pfam” has two expert-curated thresholds per family, for domain and sequence bitscores respectively (Pfam calls them “gathering” thresholds) [14]. For all methods, domain overlaps are removed ranking by p-value. Overlaps between families in the “nesting” list are not removed (Supp. Methods in S1 Text). All methods use a permissive overlap definition [61] (Supp. Methods in S1 Text), except for the Standard Pfam (there overlaps of even one amino acid are removed [14]). The Standard Pfam thresholds are mapped to p-values, q-values, and lFDRs, and the medians of these distributions are used in comparisons (Supp. Results and Fig N in S1 Text).
We introduce a suite of tests that measure empirical FDRs using biologically-motivated definitions of TPs and FPs. The “standard” biological sequence null model, which most software from BLAST to HMMER use, consists of random sequences generated assuming independent and identically distributed amino acids. Domains predicted on these random sequences produce a distribution of random bit scores from which p-values are computed. The five empirical tests we use instead label every prediction as either a TP or a FP, and these labels are used to compute empirical FDRs and E-values (number of type I errors, or FPs). Each test makes different assumptions, and together they provide independent and complementary evaluations. We describe our two primary tests in detail next; for the other three, see Supp. Methods (S1 Text).
Given domain predictions labeled as either TPs or FPs as above, we compute empirical FDRs at two levels. Briefly, the “method-level” FDR evaluates an entire scoring method (q-values, E-values, etc.) combining all domain families, whereas the “family-level” FDR evaluates the accuracy of q-values separately per family. These quantities are consistent estimators of the corresponding true pFDRs under weak dependence [44]. At a threshold t, let TPij(t) and FPij(t) be the observed number of true positives and false positives, respectively, for domain family j in protein sequence i.
DPUC improves domain prediction by taking into account the “context,” or presence of other domain predictions [48]. A newer version of dPUC now works with HMMER3, among other improvements that will be described elsewhere. Context family pair counts were derived from Pfam 25 on UniProt proteins. The “candidate domain p-value threshold” of dPUC is a tunable parameter, which when set to p ≤ 1e-4 gives comparable empirical FDRs to q ≤ 4e-4 on the MarkovR and OrthoC tests. dPUC is not evaluated in ContextC because both are based on domain context (dPUC would have a zero empirical FDR), nor in ClanOv because dPUC requires overlap removal while ClanOv requires observing overlaps to compute its FDR. DPUC 2.0 is available at http://compbio.cs.princeton.edu/dpuc.
PairCoil2 [62], TMHMM [63], and SEG [64], were run on UniRef50 using standard parameters to predict coiled coils, transmembrane domains, and low-complexity regions, respectively. Each Pfam family observed at least 4 times in UniRef50 was associated with a category if more than half of its domains overlapped the category's predictions. For families with multiple categories, only the one with the greatest amino acid overlap was kept. Unassigned families were categorized as “other”.
|
10.1371/journal.ppat.1002026 | A Yersinia Effector with Enhanced Inhibitory
Activity on the NF-κB Pathway Activates the NLRP3/ASC/Caspase-1 Inflammasome
in Macrophages | A type III secretion system (T3SS) in pathogenic Yersinia
species functions to translocate Yop effectors, which modulate cytokine
production and regulate cell death in macrophages. Distinct pathways of
T3SS-dependent cell death and caspase-1 activation occur in
Yersinia-infected macrophages. One pathway of cell death
and caspase-1 activation in macrophages requires the effector YopJ. YopJ is an
acetyltransferase that inactivates MAPK kinases and IKKβ to cause
TLR4-dependent apoptosis in naïve macrophages. A YopJ isoform in Y.
pestis KIM (YopJKIM) has two amino acid substitutions,
F177L and K206E, not present in YopJ proteins of Y.
pseudotuberculosis and Y. pestis CO92. As compared
to other YopJ isoforms, YopJKIM causes increased apoptosis, caspase-1
activation, and secretion of IL-1β in Yersinia-infected
macrophages. The molecular basis for increased apoptosis and activation of
caspase-1 by YopJKIM in Yersinia-infected
macrophages was studied. Site directed mutagenesis showed that the F177L and
K206E substitutions in YopJKIM were important for enhanced apoptosis,
caspase-1 activation, and IL-1β secretion. As compared to
YopJCO92, YopJKIM displayed an enhanced capacity to
inhibit phosphorylation of IκB-α in macrophages and to bind IKKβ in
vitro. YopJKIM also showed a moderately increased ability to inhibit
phosphorylation of MAPKs. Increased caspase-1 cleavage and IL-1β secretion
occurred in IKKβ-deficient macrophages infected with Y.
pestis expressing YopJCO92, confirming that the
NF-κB pathway can negatively regulate inflammasome activation.
K+ efflux, NLRP3 and ASC were important for secretion of
IL-1β in response to Y. pestis KIM infection as shown using
macrophages lacking inflammasome components or by the addition of exogenous KCl.
These data show that caspase-1 is activated in naïve macrophages in
response to infection with a pathogen that inhibits IKKβ and MAPK kinases
and induces TLR4-dependent apoptosis. This pro-inflammatory form of apoptosis
may represent an early innate immune response to highly virulent pathogens such
as Y. pestis KIM that have evolved an enhanced ability to
inhibit host signaling pathways.
| Pathogenic bacteria in the genus Yersinia use multiple virulence
determinants to counteract innate immunity and facilitate infection. A type III
system in Yersinia translocates an effector called YopJ that
elicits cell death in macrophages. YopJ inhibits the production of survival
factors in naïve macrophages, causing them to die by apoptosis, which is
generally considered to be immunologically silent. However, recent studies show
that caspase-1, a key regulator of pro-inflammatory responses, is activated in
Yersinia-infected macrophages undergoing apoptosis. How
caspase-1 is activated during YopJ-induced macrophage apoptosis is not known. We
have identified a distinct isoform of YopJ in Y. pestis
(YopJKIM) that induces high levels of apoptosis and caspase-1
activation in infected macrophages. In this study, the molecular basis for the
increased activity of YopJKIM was studied with the goal of better
understanding the underlying mechanism of caspase-1 activation. The data show
that YopJKIM has two amino acid changes that give it an enhanced
ability to inhibit survival signals in macrophages. The increased apoptosis may
cause membrane permeability, resulting in efflux of ions and activation of
caspase-1. Therefore, apoptosis of naïve macrophages inflicted by highly
virulent pathogens may not be immunologically silent.
| Microbial pathogens encode numerous types of virulence factors that are used to
circumvent or usurp immune responses within cells of their hosts. A protein export
pathway known as the type III secretion system (T3SS) allows Gram-negative bacterial
pathogens to deliver effector proteins into or across the plasma membrane of host
cells, with the goal of co-opting or disrupting eukaryotic signaling pathways [1], [2]. Infection of
macrophages with T3SS-expressing bacterial pathogens commonly causes cytotoxicity in
the host cell, but the mechanisms of cellular demise and the morphological and
immunological characteristics of cell death can be unique for each microbe [3]. Two types of
macrophage death that can be induced by T3SS-expressing pathogens and distinguished
morphologically and immunologically are apoptosis and pyroptosis [4]. Apoptosis is
traditionally associated with a lack of inflammation while pyroptosis is considered
pro-inflammatory [4], [5]. Apoptosis and pyroptosis can also be distinguished
mechanistically by the fact that only the latter mechanism of cell death is
dependent upon the activity of caspase-1, a pro-inflammatory caspase [4], [5]. Recently,
however it has been determined that caspase-1 can be activated in macrophages dying
of apoptosis [6],
[7], [8], indicating that
pathogen-inflicted apoptosis may not be immunologically silent.
Caspase-1 is synthesized as a 45 kDa inactive zymogen that is cleaved to generate the
active heterotetramer composed of two p10 and two p20 subunits [9]. Activation of caspase-1
occurs through its recruitment to an inflammasome complex [10], [11], [12]. Activated caspase-1 cleaves
pro-IL-1β and pro-IL-18, and promotes secretion of the mature forms of these
cytokines by a non-conventional pathway. Macrophages dying of pyroptosis therefore
release active forms of IL-1β, and IL-18, which are important cytokines for
protective host responses against several pathogens [5]. In addition, pyroptosis can
release intracellular bacteria from macrophages, allowing for clearance of the
pathogens by neutrophils [13].
Inflammasome complexes assemble on a scaffold of NOD-like receptors
(NLRs) [11], [12], [14]. NLRs comprise a family of pattern recognition receptors
(PRRs) that detect cytosolic pathogen-associated molecular patterns (PAMPs) or
infection-associated processes [10], [11], [12]. Well-studied NLR family members include NLRP3 (formerly
NALP3 or Cryopyrin), NLRC4 (formerly IPAF) and NAIP5. NLRP3 in complex with the
adaptor protein ASC induces the activation of caspase-1 in response to a variety of
microbial products as well as endogenous danger signals such as potassium
(K+) efflux or extracellular ATP [10], [11], [12]. NLRC4 recognizes bacterial
flagellin from S. enterica serovar Typhimurium, which is delivered
into macrophages via a T3SS in this pathogen [10], [13], [15], [16], [17]. Another family of PRRs,
the toll-like receptors (TLRs) often function in concert with NLRs to positively
regulate inflammasome activation and function [18]. For example, production of
pro-IL-1β and pro-IL-18 is upregulated by TLR signaling. In addition, production
of NLRP3 is positively regulated by TLR signaling through the NF-κB pathway
[19].
In pathogenic Yersinia species, a plasmid-encoded T3SS delivers Yop
effectors into host cells [2], allowing the bacteria to modulate innate immune responses
[20]. The T3SS
is an essential virulence determinant in these pathogens, which cause diseases
ranging from plague (Y. pestis) to enterocolitis (Y.
enterocolitica) and mesenteric lymphadenitis (Y.
pseudotuberculosis). Naïve Yersinia-infected
macrophages undergo apoptosis via a cell death program that requires TLR4-dependent
activation of initiator and executioner caspases and T3SS-mediated delivery of YopJ,
which inhibits expression of anti-apoptotic factors under regulatory control of MAPK
and NF-κB signaling pathways [21], [22], [23], [24]. Inactivation of the NF-κB and MAPK signaling
pathways via YopJ-mediated inhibition of the inhibitor of kappa B kinase beta
(IKKβ) and MAPK kinases (MKKs) is critical for apoptosis of
Yersinia-infected macrophages [23], [25].
YopJ is the prototypical member of a family of T3SS effectors that inhibit the
NF-κB pathway [26], [27], [28]. These proteins exhibit homology to CE cysteine proteases
[26]. Evidence
has been obtained that YopJ can function as a deubiquitinase [29], [30], [31]. However, more recent studies
indicate that YopJ has acetyltransferase activity, acetylating Ser and Thr residues
critical for the activation of the MKKs and IKKβ [32], [33], [34]. YopJ is an important virulence
factor in Y. pseudotuberculosis
[21], [35] and Y.
enterocolitica, where it is known as YopP [36].
Recently, it has been determined that caspase-1 can be activated in a T3SS-dependent
manner by two distinct pathways in macrophages infected with
Yersinia
[6]. In one
pathway, insertion of channels or pores in the plasma membrane by the T3SS
translocon activates caspase-1 and causes pyroptosis in
Yersinia-infected macrophages [6], [37], [38], [39]. Activation of caspase-1 in
response to the Yersinia T3SS translocon can be counteracted by Yop
effectors including YopE [39] and YopK [6], and therefore this pathway is
inhibited in macrophages infected with wild-type bacteria.
A second pathway of caspase-1 activation that occurs in macrophages infected with
wild-type Yersinia is not inhibited by YopE or YopK and requires
YopJ activity [6], [8]. Although caspase-1 is activated in response to YopJ
activity, caspase-1 is not required for YopJ-dependent macrophage apoptosis [6], [8]. A potential
explanation for the ability of YopJ to cause caspase-1 activation came from the work
of Greten et al. [7], who showed that genetic or pharmacological ablation of
IKKβ resulted in apoptosis, activation of caspase-1 and secretion of IL-1β
from macrophages following stimulation of TLR4 with LPS. Evidence was obtained that
an anti-apoptosis gene product expressed under control of NF-κB, plasminogen
activator inhibitor 2 (PAI-2), negatively regulates apoptosis and caspase-1
activation in LPS-stimulated macrophages [7]. The authors suggested that
inhibition of caspase-1 by the NF-κB pathway represents a negative feedback loop
that allows the innate immune system to activate, via TLR4, a compensatory host
defense response against Gram-negative pathogens that inhibit activation of
NF-κB [7].
Different Yersinia strains display a range of YopJ-dependent
apoptotic activities on macrophages [8], [35], [40], [41]. This difference in apoptotic activity is due to the
expression of distinct YopJ isoforms by different Yersinia strains
[8], [35], [40], [41]. For
example, a Y. enterocolitica strain encoding a YopP protein with an
Arg at position 143 was shown to have higher apoptotic activity and inhibit IKKβ
more efficiently in macrophages then strains having a Ser at this position [40].
Y. pestis KIM, a 2.MED (Mediaevalis) biovar strain, encodes a
YopJ isoform that causes higher levels of apoptosis and caspase-1 activation in
infected macrophages as compared to other YopJ isoforms [8]. The YopJKIM protein
has an Arg at position 143 but in addition has two amino acid substitutions, at
positions 177 and 206, as compared to other isoforms of this effector found in
Y. pestis or Y. pseudotuberculosis
strains.
Here, the molecular basis for the enhanced ability of YopJKIM to cause
apoptosis and activate caspase-1 in Yersinia-infected macrophages
was studied, with the goal of better understanding the underlying mechanism of this
host response. Analysis of YopJKIM in parallel with other YopJ isoforms
indicated that the unique capacity of this effector to cause high-level apoptosis
and caspase-1 activation requires both codon substitutions at positions 177 and 206.
The presence of these codon substitutions also correlated with the enhanced ability
of YopJKIM to inhibit MKK and IKKβ signaling pathways. Infection of
IKKβ-deficient macrophages with Y. pestis confirmed that this
kinase has an important role in negatively regulating caspase-1 activation [7]. Finally,
evidence was obtained that K+ efflux leading to activation of the
NLRP3/ASC/capsase-1 inflammasome is important for secretion of IL-1β and IL-18
from macrophages infected with Y. pestis KIM. These findings
indicate that, by inhibiting production of survival factors under control of the
NF-κB and MAPK pathways, YopJ causes TLR4-dependent apoptosis and caspase-1
activation in macrophages infected with Yersinia. In addition,
Y. pestis KIM causes high levels of apoptosis and caspase-1
activation in macrophages because it has evolved a YopJ isoform with enhanced
inhibitory activity on NF-κB and MAPK pathways.
Sequence comparisons were made between YopJKIM and YopJ proteins from
two other Yersinia strains that display lower apoptosis
activity in macrophages, Y. pseudotuberculosis IP2666 and
Y. pestis CO92. There is one amino acid difference between
YopJKIM and YopJ in Y. pseudotuberculosis
(YopJYPTB), corresponding to L177F in the predicted catalytic
core [29] of the
enzyme (residues 109-194; Figure S1 in Text S1). Comparison of YopJKIM
with YopJ from Y. pestis CO92 (YopJCO92) revealed
two differences, L177F and E206K, the latter of which is located just beyond the
carboxy-terminal end of the predicted catalytic core (Figure S1 in Text
S1).
To determine if the amino acid substitutions at positions 177 and 206 of
YopJKIM affect secretion or delivery of the effector into
macrophages, expression plasmids encoding YopJKIM,
YopJYPTB, or YopJCO92 appended with C-terminal GSK
tags were constructed. An expression plasmid encoding a YopJ isoform with a Leu
at position 177 and a Lys at position 206 (YopJKIME206K) was also
constructed in the same manner. The expression plasmids were introduced into a
ΔyopJ mutant of Y. pseudotuberculosis
(IP26; Table S1 in Text S1). Y.
pseudotuberculosis was used in the experiment because it lacks the
Pla protease of Y. pestis which is known to degrade Yops
secreted in vitro [42]. The resulting strains were induced to secrete Yops
under low calcium growth conditions and immunoblotting of the secreted proteins
showed that YopJKIM, YopJYPTB, YopJKIME206K and
YopJCO92 were exported at equal levels (Figure S2 in Text
S1).
A translocation assay was performed using the phospho-GSK reporter system [43]. IP26
strains expressing the different YopJ isoforms fused to GSK were used to infect
bone marrow derived macrophages (BMDMs) for 2 hr. Delivery of the effector into
host cells was measured by anti-phospho-GSK immunoblotting [43]. The results showed that
YopJKIM, YopJYPTB, YopJKIME206K and
YopJCO92 isoforms were translocated at similar levels (Figure 1A). Samples of the
same lysates analyzed in Figure
1A were subjected to immunoblotting with anti-caspase-1 antibody to
measure the level of caspase-1 cleavage. Consistent with previous results [6], cleavage
of caspase-1 was detected in BMDMs infected with Y.
pseudotuberculosis expressing YopJYPTB (Figure 1B, lane 2). However,
caspase-1 cleavage was comparatively higher with expression of
YopJKIM (lane 1) and lower with expression of
YopJKIME206K or YopJCO92 isoforms (lanes 3 and 4,
respectively). These results suggest that the ability of YopJKIM to
trigger maximal caspase-1 activation requires both the F177L and K206E
substitutions, and these codon changes impart an activity to the protein that is
manifested following its delivery into the host cell.
YopJ-mediated apoptosis in response to Yersinia infection
requires stimulation of TLR4 in naïve macrophages to activate a death
response pathway [25], [44]. It is not known if TLR signaling is required for
YopJ-dependent activation of caspase-1 in Yersinia-infected
macrophages. When BMDMs lacking the two major TLR adaptors, MyD88 and Trif, were
infected with wild-type Y. pseudotuberculosis IP2666 for 2 hr,
activation of caspase-1 was substantially reduced (Figure S3 in Text S1).
Cleavage of caspase-1 was not diminished in IP2666-infected BMDMs missing only
MyD88 or Trif (data not shown), indicating that TLR signaling through either of
these adaptors is important for the downstream events that lead to activation of
caspase-1 in conjunction with YopJ activity. YopJ-dependent caspase-1 activation
and IL-1β secretion were inhibited when BMDMs were treated with LPS prior to
infection with Y. pseudotuberculosis (Figure S3 in Text S1)
[6] or
Y. pestis
[8]. Thus,
macrophages pre-stimulated with LPS are desensitized to undergo YopJ-dependent
apoptosis [38] and caspase-1 activation upon
Yersinia infection. Desensitization occurs because the TLR4
signaling pathway contains a negative feed back mechanism operating via
NF-κB that upregulates expression of proteins that inhibit apoptosis and
activation of caspase-1 [7].
To demonstrate that the polymorphisms in YopJKIM at positions 177 and
206 were important for the activity of this effector in the native context of
Y. pestis, a L177F codon change was introduced into the
sequence of yopJKIM on the virulence plasmid pCD1 by
allelic exchange, converting it to yopJYPTB. In
addition, an E206K codon change, a double L177F/E206K codon change, and a C172A
codon change were introduced into pCD1, creating
yopJKIME206K, yopJC092, and
yopJC172A, respectively. The resulting
strains (referred to as Yp-YopJYPTB, Yp-YopJKIME206K,
Yp-YopJC092 and Yp-YopJC172A)(Table S1 in Text S1)
were phenotypically analyzed. As shown by immunoblotting of whole bacterial
lysates, YopJKIM, YopJYPTB, YopJKIME206K and
YopJCO92 were expressed at equal levels in Y.
pestis (Figure S4 in Text S1). The ability of Y.
pestis strains expressing the different YopJ isoforms to induce
apoptosis and cytokine secretion in BMDMs was then determined after a 24 hr
infection. As shown in Figure
2A,B, the amounts of lactate dehydrogenase (LDH) released (used as a
marker of cell death) and IL-1β secreted were significantly lower in
macrophages infected with Yp-YopJYPTB, Yp-YopJKIME206K or
Yp-YopJCO92 as compared to Yp-YopJKIM. A similar trend
was seen for secretion of IL-18 (Figure 2C).
Caspase-1 was required for the processing and release of IL-1β from
macrophages under these infection conditions as shown by infecting wild-type or
casp-1-/- BMDMs with Yp-YopJKIM and
isolating IL-1β from infection supernatants by immunoprecipitation. Mature
IL-1β was absent in supernatants isolated from
casp-1-/- BMDMs infected with
Yp-YopJKIM (Figure S5 in Text S1), indicating that the processing and
release of IL-1β during infection of wild-type macrophages with
Yp-YopJKIM occurred in a caspase-1-dependent manner.
As a control, levels of TNF-α, which is secreted independent of caspase-1
activity, were measured. Macrophages infected with Yp-YopJC172A or
Yp-YopJCO92 secreted significantly higher levels of TNF-α as
compared to Yp-YopJKIM, whereas the other mutants tested produced
intermediate results (Figure
2D). Overall, these results indicate that amino acid substitutions at
positions 177 and 206 are important for the ability of YopJKIM to
induce high levels of macrophage apoptosis, caspase-1 activation and secretion
of mature IL-1β and IL-18 in Y. pestis-infected
macrophages. Conversely, the amino acid substitutions at positions 177 and 206
are important for the ability of YopJKIM to inhibit TNF-α
secretion in macrophages under the same conditions.
To determine if YopJKIM has higher affinity for IKKβ as compared
to other YopJ isoforms, several different YopJ proteins were assayed for the
ability to bind this kinase in cell lysates. Purified GST-YopJ fusion proteins
or GST alone bound to beads were incubated in HEK293T cell lysates that
contained overexpressed IKKβ. The amounts of IKKβ and GST proteins
recovered on the beads after washing was measured by quantitative
immunoblotting. IKKβ bound to beads coated with GST-YopJKIM but
not to beads coated with GST alone (Figure 3A, compare lanes 2 and 3). There was reduced binding of
IKKβ to GST-YopJCO92 as compared to GST-YopJKIM (Figure 3A, compare lanes 3 and
5). When the amount of bound IKKβ was normalized to the amount of GST fusion
protein recovered, it was estimated that 10-times less IKKβ bound to
GST-YopJCO92 as compared to GST-YopJKIM (Figure 3B). A GST fusion
protein encoding YopJC172A bound ∼5 times less IKKβ as
compared to GST-YopJKIM (Figure 3A, compare lanes 3 and 4, Figure 3B), suggesting that the catalytic Cys
residue contributes to binding between IKKβ and YopJKIM. Overall,
these results suggest that YopJKIM has higher affinity for IKKβ
as compared to YopJCO92.
To determine if YopJKIM is a better inhibitor of IKKβ than
YopJCO92, the amount of phosphorylated IκBα
(p-IκBα) in BMDMs was measured after a 1 hr infection. As shown in Figure 3C, significantly lower
levels of p-IκBα were present in macrophages infected with
Yp-YopJKIM as compared to BMDMs infected with
Yp-YopJCO92. In addition, significantly lower levels of
p-IκBα were present in macrophages infected with Yp-YopJKIM
as compared to BMDMs infected with Yp-YopJC172A (Fig. 3C), confirming that acetyltransferase
activity is important for YopJ to inhibit the NF-κB pathway. Because
IκBα is directly phosphorylated by IKKβ, these results are
consistent with the idea that YopJKIM more efficiently inhibits
IKKβ activity as compared to YopJCO92.
Greten et al. have shown that treatment of IKKβ-deficient macrophages with
LPS causes activation of caspase-1 and secretion of IL-1β [7]. If IKKβ
activity is important to suppress activation of the inflammasome in macrophages
infected with a live Gram-negative pathogen, than increased caspase-1 activation
and IL-1β secretion should be observed in IKKβ-deficient as compared to
wild-type BMDMs infected with Y. pestis. The effect of genetic
inactivation of Ikkβ on caspase-1 activation in Y.
pestis-infected macrophages was therefore investigated.
IKKβ-deficient BMDMs were generated by conditional Cre-lox-mediated deletion
of a “floxed” Ikkβ gene (referred to as
IkkβΔ BMDMs; Materials and Methods). The
IkkβΔ BMDMs or wild-type control
IkkβF/F macrophages were left uninfected or
infected with Yp-YopJKIM,
Yp-YopJCO92 or
Yp-YopJC172A for 4 hr. Quantitative RT-PCR
(qRT-PCR) of Ikkβ message was used to estimate the
efficiency of Cre-lox mediated deletion of the Ikkβ gene in
the BMDMs. Results indicated that ∼50% of the
Ikkβ genes had been deleted in the population of
IkkβΔ cells (Figure S6A in Text S1).
The impact of this partial deficiency in Ikkβ on the
expression and secretion of cytokines in the Y. pestis infected
macrophages was determined. As compared to the
IkkβF/F macrophages, the
IkkβΔ BMDMs were compromised for
infection-induced expression of mRNA for the cytokines IL-18, TNFα and
IL-1β, as shown by qRT-PCR (Figure S6B–D in Text S1).
This result was expected since the NF-κB pathway positively regulates
expression the il-18, tnf and il-1b genes. Accordingly, the
IkkβΔ BMDMs secreted lower levels of
TNFα as compared to IkkβF/F macrophages
after a 24 hr infection (Figure
4A). In addition, during infection with Yp-YopJCO92 or
Yp-YopJC172A, higher amounts of IL-1β were secreted from
IkkβΔ BMDMs as compared to
IkkβF/F macrophages (Figure 4B), consistent with the idea that the
NF-κB pathway negatively regulates processing and secretion of IL-1β via
control of caspase-1 activation [7]. Unexpectedly, the amount of IL-1β secreted
following infection with Yp-YopJKIM appeared to be lower in
IkkβΔ BMDMs as compared to
IkkβF/F macrophages, although the observed
difference was not statistically significant (Figure 4B). The interpretation of this latter
result was complicated because of the fact that there was only partial
deficiency in Ikkβ in the
IkkβΔ BMDMs, but one possible
explanation was that synthesis of pro-IL-1β was reduced due to the extremely
low level il-1b message in the IkkβΔ BMDMs
infected with Yp-YopJKIM (Figure S6D in Text
S1).
Activation of caspase-1 was measured by immunoblotting to detect the cleaved
enzyme in lysates prepared 2 hr after infection of
IkkβΔ or
IkkβF/F BMDMs with Yp-YopJKIM,
Yp-YopJCO92 or Yp-YopJC172A. Caspase-1 activation in
uninfected BMDMs or in macrophages treated with LPS and ATP was determined in
parallel for comparison. Increased caspase-1 cleavage occured in
IkkβΔ macrophages infected with
Yp-YopJKIM or Yp-YopJCO92 as compared to
IkkβF/F BMDMs infected with the same strains
(Figure 5A, compare
lanes 7 and 8 with 2 and 3). Cleaved caspase-1 was below the limit of detection
in IkkβΔ macrophages infected with
Yp-YopJC172A (Figure
5A, lane 9). Activation of caspase-1 was also measured by a
microscopic assay utilizing FAM-YVAD-FMK, a fluorescent probe for active
caspase-1, in IkkβΔ or
IkkβF/F BMDMs infected for 9 hr. The results
showed overall higher levels of caspase-1 positive cells in
IkkβΔ as compared to
IkkβF/F macrophages (Figure 5B and C). Taken together, these
results show that loss of IKKβ activity can increase caspase-1 activation in
macrophages infected with Y. pestis, and are consistent with
the idea that IKKβ is an important target of YopJ for activation of the
inflammasome.
In addition to binding to and acetylating IKKβ, YopJ binds to and acetylates
other members of the MKK superfamily including MKK1, MKK2, MKK3, MKK4, MKK5, and
MKK6 [26], [32], [33]. There is
evidence that YopJ binds to a site conserved on members of the MKK-IKK
superfamily [45].
Since we had previously obtained evidence that inhibition of MAPK signaling was
critical for YopJ-induced macrophage apoptosis [23], we sought to determine if
YopJKIM could more efficiently inhibit MAPK phosphorylation as
compared to YopJCO92. BMDMs were left uninfected or infected for 30
or 60 min with Yp-YopJKIM, Yp-YopJCO92,
or Yp-YopJC172A and ELISA was used to measure
phosphorylation of the MAPKs ERK (substrate of MKK1/2), p38 (substrate of
MKK3/6) and SAPK/JNK (substrate of MKK4/7) (Materials and Methods). As shown in Figure 6A, ERK was not phosphorylated to a
large degree at either time point in macrophages infected with
Yp-YopJC172A and therefore it was not possible
to evaluate the degree to which ERK phosphorylation was inhibited by either
YopJKIM or Yp-YopJCO92. In contrast, p38 and JNK did
show increased phosphorylation upon infection with
Yp-YopJC172A, especially at the 30 min time
point (Figure 6B and C,
respectively). There was in general reduced phosphorylation of p38 and JNK in
BMDMs infected with Yp-YopJKIM as compared to YopJCO92,
especially at the 30 min time point, and the difference was statistically
significant in the case of JNK (Figure 6B and C). These results suggest that YopJKIM more
efficiently inhibits the activities of MKK3/6 and MKK4/7 as compared to
YopJCO92.
The importance of several different inflammasome components for Y.
pestis-induced secretion of IL-1β and IL-18 was investigated
using NLRP3 (Nlrp3-/-)-, ASC (Asc-/-)- or NLRC4
(Nlrc4-/-)-deficient BMDMs. The mutant BMDMs or wild-type control
macrophages were infected with Yp-YopJKIM or Yp-YopJC172A.
Tissue culture supernatants were collected and analyzed by ELISA to measure the
levels of IL-1β and IL-18 present after 24 hr of infection. NLRP3- or
ASC-deficient BMDMs infected with Yp-YopJKIM secreted significantly
lower levels of IL-1β and IL-18 as compared to wild-type macrophages
infected with Yp-YopJKIM (Figure 7A,B; Figure S7A, B in Text S1).
NLRC4-deficient macrophages released similar levels of these cytokines as
compared to wild-type BMDMs (Figure
7A, Figure S7A in Text S1), suggesting that NLRC4 does not play
a significant role in caspase-1 activation and cytokine secretion during
Yp-YopJKIM infection. Both Yp-YopJKIM and
Yp-YopJC172A stimulated infected BMDMs to secrete TNF-α,
although higher levels (∼2 to 3 fold) of TNF-α were secreted from
macrophages infected with Yp-YopJC172A regardless of macrophage type
infected (Figure 7C, D).
Thus, NLRP3 and ASC, but not NLRC4, are involved in the secretion of IL-1β
and IL-18, but not TNF-α, from Yp-YopJKIM -infected
macrophages.
To determine if NLRP3, NLRC4 or ASC play a role in YopJKIM-dependent
apoptosis, wild-type BMDMs or BMDMs deficient for these inflammasome components
were left uninfected or infected with Yp-YopJKIM or
Yp-YopJC172A. Tissue culture supernatants were collected 24 hr
post-infection and analyzed for LDH. Similar levels of LDH were released from
NLRP3, NLRC4 or ASC-deficient BMDMs as compared wild-type macrophages after
Yp-YopJKIM infection (Figure 7E, F). Low levels of LDH release
occurred in all macrophages infected with Yp-YopJC172A. These results
demonstrate that apoptosis can occur in Yp-YopJKIM -infected
macrophages in the absence of NLRP3, NLRC4 or ASC, consistent with our previous
data showing that macrophage apoptosis during Yp-YopJKIM infection is
independent of caspase-1 [8].
Efflux of intracellular K+ has been implicated in the activation
of the NLRP3/ASC/caspase-1 inflammasome [10], [11], [12]. To assess a role for
intracellular K+ efflux in caspase-1 activation and IL-1β
release during infection with Y. pestis, BMDMs were infected
with Yp-YopJKIM or Yp-YopJC172A, and then incubated in
cell culture media supplemented with 30 mM KCl, 30 mM NaCl or no supplement.
Cell culture supernatants were collected at 8 hr and 24 hr time points and
analyzed for the presence of IL-1β and TNF-α by ELISA. Significantly
lower levels of IL-1β (∼5-fold) were secreted from macrophages infected
with Yp-YopJKIM in the presence of 30 mM KCl as compared to untreated
macrophages at 8 hr post-infection (Figure 8A). Macrophages infected with Yp-YopJKIM in the
presence of 30 mM NaCl appeared to secrete IL-1β to slightly lower levels as
compared to untreated infected macrophages at 8 hr post-infection, but this
difference was not significant (Figure 8A). A similar trend of IL-1β secretion was observed at
the 24 hr time point when macrophages were infected with Yp-YopJKIM
in the presence or absence of KCl or NaCl (Figure 8C). Macrophages infected with
Yp-YopJC172A secreted similar low levels of IL-1β regardless
of treatment (Figure 8A, C).
Secretion of TNF-α from Yp-YopJKIM- or
Yp-YopJC172A-infected macrophages was not affected by the presence of
30 mM KCl or NaCl (Figure 8B,
D). In addition, the presence of 30 mM KCl did not diminish LDH
release from BMDMs infected with Yp-YopJKIM (data not shown). BMDMs
deficient for the purinergic receptor, P2X7, secreted similar levels
of IL-1β and IL-18 as did wild-type macrophages infected with
Yp-YopJKIM, indicating that this receptor does not play a
significant role in inducing the secretion of these cytokines (data not shown).
Taken together, these results suggest that a K+ efflux that
occurs independent of P2X7R is important for activation of the
NLRP3/ASC/caspase-1 inflammasome in macrophages infected with
Yp-YopJKIM.
To examine how Y. pestis infection and KCl treatment affected
steady state levels of pro-IL-1β, lysates of macrophages left untreated or
treated with KCl or NaCl were prepared at 8 hr post-infection and analyzed by
immunoblotting for pro-IL-1β or actin as a loading control. As shown in
Figure 8E, infection
stimulated production of pro-IL-1β, with steady state levels of
pro-IL-1β slightly lower in macrophages infected with Yp-YopJKIM
as compared to Yp-YopJC172A (compare lanes 2 and 3, 5 and 6 and 8 and
9). Similar amounts of pro-IL-1β were detected in macrophages infected with
Yp-YopJKIM in the absence or presence of 30 mM KCl or 30 mM NaCl
(Figure 8E, compare lane
2 with 5 and 8). These results indicated that reduced detection of IL-1β in
supernatants of macrophages infected with YopJKIM and treated with
exogenous KCl was not due to KCl inhibiting production of pro-IL-1β.
It was previously shown that caspase-1 was activated during YopJ-induced apoptosis of
macrophages infected with Y. pseudotuberculosis
[6]. In addition,
it was demonstrated that YopJKIM had increased capacity to cause
macrophage apoptosis and activate caspase-1 as compared to other YopJ isoforms [8]. However, the
mechanism of YopJ-induced caspase-1 activation and the molecular basis for enhanced
apoptosis and activation of caspase-1 in macrophages by YopJKIM was
unknown. The results of studies reported here indicate that several of the
requirements for YopJ-induced apoptosis and caspase-1 activation are the same, and
therefore it is likely that these two processes are mechanistically connected.
First, it is known that TLR4 signaling is important for YopJ-induced macrophage
apoptosis [21],
[22], [23], [24] and we
show here that the two major TLR adaptors, MyD88 and Trif, are important for
YopJ-induced caspase-1 activation. Second, desensitization of macrophages by
pretreatment with LPS decreases YopJ-induced apoptosis [38] and caspase-1 activation.
Third, comparison of the activities of different YopJ isoforms showed a direct
correlation between apoptosis, caspase-1 activation and inhibition of MAPK and
NF-κB signaling pathways. Forth, when macrophages in which
Ikkβ was conditionally deleted were infected with Y.
pestis, caspase-1 activation increased, providing genetic evidence that
IKKβ is an important target of YopJ for caspase-1 activation, as well as
apoptosis [25].
Inhibition of MAPK and NF-κB pathways by YopJ is thought to reduce expression of
survival factors (e.g. FLIP, XIAP), thereby potentiating TLR4 signaling to trigger
apoptosis [22],
[23], [24].
Inactivation of the MAPK and NF-κB pathways by YopJ could also prevent
expression of putative negative regulators of caspase-1 (e.g. PAI-2) [7]. It is important
to point out that there is no direct evidence that PAI-2 inhibits caspase-1
activation independently of blocking apoptosis, rather the data show that PAI-2
overexpression reduces both apoptosis and caspase-1 activation [7]. It is possible that PAI-2
inhibits apoptosis and that events triggered downstream of TLR4-dependent programmed
cell death are required for caspase-1 activation. We suggest that caspase-1
activation is a normal outcome of a type of apoptosis that is triggered in
naïve macrophages by TLR4 signaling combined with pathogen interference with
MAPK and NF-κB pathways.
Data presented here suggest that YopJKIM triggers increased apoptosis and
caspase-1 activation because it is a better inhibitor of macrophage survival
pathways than other YopJ isoforms. YopJKIM could function as a better
inhibitor of macrophage signaling pathways if it had a longer half-life in the host
cell, or had higher affinity for substrates. The F177L polymorphism could increase
protein stability, although it is not immediately clear why a Leu at position 177
rather than a Phe would increase protein half-life. The K206E mutation could
increase half-life, which is reasonable since Lys residues can be subject to
ubiquitination. Although not mutually exclusive of the preceding ideas, we favor the
hypothesis that the F177L and K206E substitutions allow YopJKIM to bind
more tightly to substrates, thereby making acetylation of targets more efficient at
limiting enzyme concentrations. We obtained two pieces of evidence supporting this
hypothesis. First, YopJKIM had higher apparent affinity for IKKβ than
YopJCO92 when these interactions were measured in cell lysates by a
GST pull down assay. Second, macrophages infected with Yp-YopJKIM had
lower levels of phosphorylated IκBα and MAPKs as compared to macrophages
infected with Yp-YopJCO92, indicating that there was increased inhibition
of IKKβ and MAPK kinase activity by Yp-YopJKIM.
The results suggest a model whereby the canonical yopJ allele in
Y. pseudotuberculosis (yopJYPTB)
was inherited by an ancestral Y. pestis strain, from which it
evolved to encode an isoform with higher apoptotic and caspase-1-activating
potential, YopJKIM, by the F177L mutation. The predicted sequence of a
YopJ protein in Y. pestis biovar 2.MED strain K1973002
(ZP_02318615) is identical to the sequence of YopJKIM, suggesting that
the phenotype observed is not an artifact resulting from a mutation acquired during
laboratory passage, but is associated with a unique yopJ genotype
associated with 2.MED strains. It is also hypothesize that the
yopJCO92 allele evolved from
yopJYPTB to encode an isoform with lower
cytotoxic and caspase-1 activating potential (YopJCO92) by the E206K
codon substitution. How these polymorphisms in YopJ affect Y.
pestis virulence and or the host response is not known but is an
important question to address in future studies.
The importance of different inflammasome components for YopJ-dependent activation of
caspase-1 in macrophages infected with Y. pseudotuberculosis has
recently been examined [6]. This study showed that NLRP3 and ASC were not required
for activation of caspase-1 as measured by immunoblot analysis of caspase-1 cleavage
[6]. Those
results would appear to be in conflict with findings presented here showing a role
for NLRP3 and ASC in secretion of IL-1β and IL-18 from macrophages infected with
Yp-YopJKIM. However, recent studies suggest that multiple distinct
caspase-1 activation pathways with different biological outcomes can operate in
macrophages infected with a bacterial pathogen. For example, evidence has been
obtained that Legionella pneumophila stimulates two distinct
pathways of caspase-1 activation in macrophages [46]. ASC is required for secretion of
active IL-18 from L. pneumophila-infected macrophages, but is not
required for caspase-1 dependent induction of pyroptosis [46]. In addition, the multiplicity
and temporal stage of infection of macrophages with a bacterial pathogen can affect
the requirements for cell death and activation of caspase-1. Shigella
flexneri infection of macrophages at low MOI (<10) for short periods
of time induces NLRC4-dependent pyroptosis [47], [48], while infection at higher MOI
(50) for longer time periods induces NLRP3-dependent pyronecrosis [48]. Two different
infection procedures for examining YopJ-induced caspase-1 activation in macrophages
have been used in this study and previous publications [6], [8]. A high MOI (20) followed by 1 hr
of bacterial-host cell contact before addition of gentamicin results in detectable
YopJ-dependent apoptosis and caspase-1 activation within 2 hr of infection (Figure 5A, Figure S3 in Text S1) [6] but no
detectable secretion of IL-1β by this time point (data not shown) [6] . A low MOI
(10) followed by 20 min of bacterial-host cell contact before addition of gentamicin
results in detectable apoptosis and caspase-1 activation by 8–9 hr (Figure 5B,C) [8], at which time secreted IL-1β
and IL-18 are first detected [8]. High amounts of secreted IL-1β and IL-18 are detected
at 24 hr post infection under the low MOI procedure (e.g. Figure 2) [8]. The high and low MOI infection
procedures may result in different requirements for NLRs to activate caspase-1, as
shown by a requirement for ASC and NLRP3 in the latter but not former method.
Interestingly, the low MOI procedure appears to slow down the kinetics of apoptosis
and caspase-1 activation, which is likely important to allow for synthesis of NLRP3
[19] and
the pro-forms of IL-1β and IL-18.
Under the low MOI conditions the presence of 30 mM KCl in the infection medium
inhibited the secretion of IL-1β and IL-18 from macrophages infected with
Yp-YopJKIM, suggesting an important role for K+
efflux in caspase-1 activation. Efflux of intracellular K+ mediated
by the P2X7R is critical for ATP-induced caspase-1 activation in
macrophages primed with LPS [49]. However, like other NLRP3 activators such as nigericin,
caspase-1 activation in response to Yp-YopJKIM infection did not require
P2X7R. One possibility is that pore formation during
YopJKIM-induced apoptosis leads to K+ efflux,
resulting in activation of the NALP3/ASC/caspase-1 inflammasome. One limitation of
this model is that it remains to be determined if K+ efflux acts as
a proximal activating signal of the NALP3/ASC/caspase-1 inflammasome. A second
limitation of this model is that apoptosis is generally associated with maintenance
of an intact plasma membrane, until late stages of cell death [4]. Future experiments will need to
address the possibility that YopJ-induced apoptosis of
Yersinia-infected macrophages can be associated with rapid membrane
permeability, resulting in K+ efflux and caspase-1 activation.
All animal use procedures were conducted following the NIG Guide for the Care and
Use of Laboratory Animals and performed in accordance with Institutional
regulations after review and approval by the Institutional Animal Care and Use
Committee at Stony Brook University.
Y. pestis and Y. pseudotuberculosis strains
used in this study are listed in Table S1 in Text S1.
Y. pestis strains used in this study are derived from KIM5
[8], which
lacks the pigmentation locus (pgm) and are exempt from select
agent guidelines and conditionally attenuated. Introduction of codon changes
into yopJ in KIM5 (Table S1 in Text S1)
was performed using the suicide plasmid pSB890 and allelic exchange as described
[50]. The
arabinose inducible plasmid encoding YopJKIM (pYopJ-GSK) has been
described [43]). Codon changes were introduced into
yopJKIM on this plasmid using Quikchange
(Invitrogen), yielding pYopJYPTB-GSK, pYopJKIME206K-GSK,
and pYopJCO92-GSK. The resulting plasmids were used to transform IP26
(IP2666 ΔyopJ) using electroporation and selection on LB
agar plates containing ampicillin (100 µg/ml) [8].
Bone marrow derived macrophages (BMDM) were isolated from the femurs of 6- to
8-week-old C57BL/6 female mice (Jackson Laboratories),
Casp-1-/- mice [8], P2X7
receptor-deficient mice [51], Ikkβf/f or
Ikkβf/f;MLysCre mice [52], [53], NLRC4-
(Nlrc4-/-), ASC- (Asc-/-) or NLRP3-
(Nlrp3-/-) deficient mice [54], and MyD88-, Trif- and
MyD88/Trif-deficient mice [55] and cultured as previously described [56], [57].
Y. pestis cultures were grown overnight with aeration in HI
broth at 28°C. The next day the cultures were diluted to an OD600
of 0.1 in the same medium supplemented with 2.5 mM CaCl2 and
incubated for 2 hr at 37°C with aeration. Twenty-four hours before
infection, BMDM were seeded into wells of 24-well plates at a density of
1.5×105 cells/ml. Macrophage infections were performed in
37°C incubators with 5% CO2 at a multiplicity of infection
(MOI) of 10 as previously described [8]. After addition of bacteria,
plates were centrifuged for 5 minutes at 95 xg to induce contact between
bacteria and macrophages. After incubation at 37°C for 15 minutes,
macrophages were washed once with pre-warmed PBS to remove any bacteria that
have not been taken up. Fresh infection medium containing 8 µg/ml of
gentamicin was added for 1 hr at 37°C. After 1 hr, macrophages were washed
once with PBS and a lower concentration of gentamicin (4.5 µg/ml) in fresh
tissue culture media was added for the remaining incubation times. To inhibit
potassium efflux from infected macrophages, potassium chloride (KCl) was added
to a final concentration of 30 mM concurrently with the media exchanges
containing 8 µg/ml gentamicin and 4.5 µg/ml gentamicin [8]. Sodium
chloride (NaCl) was used as a control and added as above at a concentration of
30 mM above baseline. Amounts of IL-1β, TNF-α or IL-18 secreted into
tissue culture media during infection assays were measured by ELISA as described
[8].
Supernatants from infected macrophages were collected and analyzed for LDH
release as described [8]. Staining with
6-carboxyfluorescein–YVAD–fluoromethylketone (FAM-YVAD-FMK;
fluorescent inhibitor of apoptosis (FLICA)) (Immunochemistry Technologies) to
detect active caspase-1 in infected macrophages was performed using fluorescence
and phase microscopy as described [8] with the exception that the procedure was performed 9
hr post-infection, and the anti-Yersinia immunolabeling step
was omitted. Quantification of percent caspase-1 positive BMDMs was performed by
scoring macrophages for positive signal in three different randomly selected
fields (∼50–100 cells per field) on a coverslip.
At 8 hr post-infection, macrophage lysates from triplicate wells were collected
in 100 µl of 1X lysis buffer (50 mM Tris-HCl, 5 mM EDTA, 150 mM NaCl,
1% Triton X-100, 2 mM DTT and a protease inhibitor cocktail
[Complete Mini, EDTA-Free, Roche]). Proteins were resolved by SDS-PAGE
and transferred to a nitrocellulose membrane. To detect IL-1β, membranes
were blotted with goat anti-IL-1β (R&D Systems). A secondary antibody,
Hamster anti-goat IRDye 700 antibody (Rockland) was used to detect samples, and
blots were viewed on the Odyssey Infrared Imaging System (LI-COR). To control
for loading, blots were probed with a rabbit anti-actin antibody
(Sigma-Aldrich).
BMDMs (106 cells per well) were seeded in 6-well plates. Y.
pestis cultures were grown as above and used to infect BMDM at a
MOI of 50. 1 hr post infection, cells were washed with ice-cold PBS and
incubated in 150 ul of 1X Lysis Buffer (Cell Signaling) for 5 min. Cells were
scraped on ice and sonicated twice for 5 seconds each. Lysates were centrifuged
at 4°C for 10 min and 100 µl of supernatant was used for ELISA.
Phospho-IκBα levels were determined using a PathScan
Phospho-IkappaB-alpha (Ser32) Sandwich ELISA kit according to
manufacturer's protocol (Cell Signaling).
BMDMs (106 cells per well) were seeded in 6-well plates. Y.
pestis cultures were grown in HI at 28°C overnight and diluted
1∶20 next day in the same medium supplemented with 20 mM NaOX and 20 mM
MgCl2. Cultures were shaken at 28°C for 1 hr and switched to 37°C for 2
hr. Cells were infected at an MOI of 20 and incubated for 30 or 60 min without
adding gentamicin. Macrophages were harvested and lysed as above. The PathScan
MAP Kinase Multi-Target Sandwich ELISA kit was used to determine phosphor-ERK,
-p38 and –JNK levels according to manufacturer's instruction (Cell
Signaling).
Y. pseudotuberculosis strains were grown in 2xYT at 26°C
overnight and diluted 1∶40 in the same medium supplemented with 20 mM
NaOX, and 20 mM MgCl2. Cultures were shaken at 26°C for 1 hr and
shifted to 37°C for 2 hr. BMDMs were seeded into wells of 6-well plates at a
density of 106 cells/well. Bacteria were harvested, washed with DMEM
and added to BMDMs at an MOI of 20. After 1 hr of infection gentamicin was added
to a final concentration of 100 µg/ml. To induce expression of YopJ-GSK
proteins, arabinose (0.2%) was maintained during grown in 2xYT at
37°C and in the cell culture medium used for infection. Y.
pestis strains were grown and used to infect macrophages as above
except that HI broth was used and arabinose was omitted. Two hr post-infection,
infected BMDMs were washed with PBS and lysed in buffer containing 50 mM
Tris-HCl pH 8.0, 5 mM EDTA, 2% Triton X-100, and 0.02% sodium
azide with protease inhibitors. In some experiments the macrophages were
incubated with 50 ng/ml of LPS for 3 hrs and then exposed to ATP at final
concentration of 2.5 mM for 1 hr as a positive control for caspase-1 cleavage.
Proteins were resolved by 10% SDS-PAGE, transferred to a PVDF membrane
and probed with anti-phospho-GSK-3β primary antibody (Cell Signaling). In
some experiments the blots were stripped and re-probed with rabbit polyclonal
anti-caspase-1 antibodies (Santa Cruz) or directly developed with this antibody.
As a loading control blots were reprobed with an anti-actin antibody
(Sigma-Aldrich, clone AC15). Goat anti-rabbit HRP conjugated secondary antibody
was used. Blots were detected with ECL reagent (Perkin Elmer Life Sciences,
Inc.).
Plasmids for expression of GST-YopJ fusion proteins were constructed from pLP16
[58]. The
pLP16 vector was derived from pGEX-2T and codes for YopJYPTB with an
N-terminal glutathione-S transferase (GST) affinity tag and a C-terminal M45
epitope tag. Quikchange mutagenesis (Invitrogen) was used to introduce codon
changes into pLP16 to generate pGEX-2T-YopJKIM,
pGEX-2T-YopJKIMC172A and pGEX-2T-YopJCO92, which
encode GST-YopJKIM, GST-YopJC172A and
GST-YopJCO92, respectively. The plasmids pGEX-2T,
pGEX-2T-YopJKIM, pGEX-2T-YopJKIMC172A and
pGEX-2T-YopJCO92 were used to transform E. coli
TUNER cells (Novagen). Cultures of TUNER cells harboring the above plasmids were
grown in LB at 37°C to OD600 of 0.2. IPTG was added to 0.1 mM final
concentration and cultures were grown at 18°C with shaking for 4 hrs. The
bacterial pellet obtained from 40 ml of each culture was resuspended in PBS
supplemented with protease inhibitor cocktail (Roche) and sonicated on ice. The
solubility of proteins in the sonicates was increased by incubation in the
presence of a buffer containing 10% sarkosyl at 4°C overnight [59]. After
centrifugation, the supernatant of the bacterial lysate was diluted 5 times with
a buffer containing 4% Triton X-100 and 40 mM CHAPS at final
concentrations. Thirty µl of glutathione beads (GST Bind Kit, Novagen)
were added and the mixture was shaken at 4°C for 1 hr. Beads were washed 4
times with 1 ml of GST Bind Kit buffer and used for pull down assays.
Cell lysates containing overexpressed IKKβ were prepared from HEK293T cells
tranfected with a retroviral construct (pCLXSN-IKKβ-IRES-GFP) [25]. HEK293T
cells were seeded in 10 cm dishes and grown to reach 70% confluence. The
culture medium was replaced with serum free DMEM and the HEK293T cells in each
dish were transfected with 10 µg of pCLXSN-IKKβ-IRES-GFP using a
calcium phosphate method. Six hrs post transfection, the culture medium was
replaced with DMEM containing 10% FBS. Cells were harvested 48 hrs post
transfection, sonicated in PBS and centrifuged. Supernatants were stored at
−80°C until use.
Beads containing bound GST proteins were incubated with 250 µl of cell
lysate supernatants from transfected HEK293T supernatant for 4 hrs at 4°C
with constant rotation. The beads were then washed 4 times with 1 ml of PBS each
and proteins bound to the beads were eluted in boiling 2X Laemmli sample buffer.
Samples of the eluates were subjected to SDS-PAGE and immunoblotting. Rabbit
polyclonal anti-IKKβ antibodies and mouse monoclonal anti-GST antibodies
were purchased from Cell Signaling and Santa Cruz, respectively. Immunoblot
signals representing IKKβ and GST or GST fusion proteins were quantified
using an Odyssey imaging system.
Experimental data analyzed for significance (GraphPad Prism 4.0) were performed
three independent times. Probability (P) values for multiple comparisons of
cytokine, phospho-IκBα ELISA and LDH release data were calculated by
one-way ANOVA and Tukey's multiple comparisons post-test. P values for two
group comparisons of cytokine, phospho-IκBα, and phospho-MAPK ELISA were
calculated by two-tailed paired student t test. P values were considered
significant if less than 0.05.
|
10.1371/journal.pcbi.1005588 | Mitochondrial respiration and ROS emission during β-oxidation in the heart: An experimental-computational study | Lipids are main fuels for cellular energy and mitochondria their major oxidation site. Yet unknown is to what extent the fuel role of lipids is influenced by their uncoupling effects, and how this affects mitochondrial energetics, redox balance and the emission of reactive oxygen species (ROS). Employing a combined experimental-computational approach, we comparatively analyze β-oxidation of palmitoyl CoA (PCoA) in isolated heart mitochondria from Sham and streptozotocin (STZ)-induced type 1 diabetic (T1DM) guinea pigs (GPs). Parallel high throughput measurements of the rates of oxygen consumption (VO2) and hydrogen peroxide (H2O2) emission as a function of PCoA concentration, in the presence of L-carnitine and malate, were performed. We found that PCoA concentration < 200 nmol/mg mito protein resulted in low H2O2 emission flux, increasing thereafter in Sham and T1DM GPs under both states 4 and 3 respiration with diabetic mitochondria releasing higher amounts of ROS. Respiratory uncoupling and ROS excess occurred at PCoA > 600 nmol/mg mito prot, in both control and diabetic animals. Also, for the first time, we show that an integrated two compartment mitochondrial model of β-oxidation of long-chain fatty acids and main energy-redox processes is able to simulate the relationship between VO2 and H2O2 emission as a function of lipid concentration. Model and experimental results indicate that PCoA oxidation and its concentration-dependent uncoupling effect, together with a partial lipid-dependent decrease in the rate of superoxide generation, modulate H2O2 emission as a function of VO2. Results indicate that keeping low levels of intracellular lipid is crucial for mitochondria and cells to maintain ROS within physiological levels compatible with signaling and reliable energy supply.
| Lipids are main sources of energy for liver and cardiac and skeletal muscle. Mitochondria are the main site of lipid oxidation which, in the heart, supplies most of the energy required for its blood pumping function. Paradoxically, however, lipids over supply impair mitochondrial function leading to metabolic syndrome, insulin resistance and diabetes. In this context, scientific debate centers on the impact of lipids and mitochondrial function on diverse aspects of human health, nutrition and disease. To elucidate the underlying mechanisms of this issue, while accounting for both the fundamental role of lipids as energy source as well as their potential detrimental effects, we utilized a combined experimental and computational approach. Our mitochondrial computational model includes β-oxidation, the main route of lipid degradation, among other pathways that include oxygen radical generation and consumption. Studies were performed in heart mitochondria from type 1 diabetic and control guinea pigs. Model and experimental results show that, below a concentration threshold, lipids fueling proceeds without disrupting mitochondrial function; above threshold, lipids uncouple mitochondrial respiration triggering excess emission of oxidants while impairing antioxidant systems and the mitochondrial energy supply-demand response. These contributions are of direct use for interpreting and predicting functional impairments in metabolic disorders associated with increased circulating levels of lipids and metabolic alterations in their utilization, storage and intracellular signaling.
| Fatty Acids (FAs) are main sources of cellular energy affecting mitochondrial energetics and redox balance. The lipid energy content becomes available from β-oxidation as reducing equivalents and acetyl CoA (AcCoA) of which the latter, after further processing in the tricarboxylic acid cycle, also supplies most of the energy as NADH and FADH2, which, in turn, fuel the buildup of the proton motive force for oxidative phosphorylation (OxPhos). Under physiological conditions, the non-esterified forms of FAs represent an important fuel supply in many tissues. However, persistent excess of FAs and accumulation of triacylglycerols in non-adipose tissues are associated with metabolic disorders like diabetes, hyperlipidemia and lipodystrophies [1,2].
Preserving the intracellular redox environment is crucial for vital functions such as division, differentiation, contractile work and survival, amongst many others [3,4,5,6,7,8,9,10,11]. Mitochondria are main drivers of intracellular redox [12,13,14,15,16], playing a central role in the development of diabetes and obesity complications [17,18,19,20,21]. Hearts from diabetic subjects are particularly prone to excess ROS because sympathetic hyper-activation and -glycemia are present in a large cohort of these patients [22,23]. These two conditions may alter cardiac and skeletal muscle redox conditions [5,6] endangering mitochondrial function [7,8]. Perturbations of cardiac mitochondrial energetics and increased mitochondrial ROS emission can account for tissue redox imbalance [8,11,12,13] and abnormal cardiac contractility leading to systolic and diastolic dysfunction in diabetic patients [17,18,19,20,21]. These abnormalities are common features in T1DM and type 2 diabetes mellitus (T2DM) patients [1,9,10] and they underlie diabetic cardiomyopathy, a major life-threatening complication that limits life quality and expectancy [3,19].
Although available evidence indicates the participation of oxidative stress in the etiology of T1DM, obesity-induced insulin resistance and T2DM [10,17,24,25,26], the role of dysfunctional β-oxidation per se as an underlying cause of metabolic disorder remains a topic of active research and debate [10]. Prevailing wisdom indicates that the myocardial shift from glucose to FA utilization occurring in diabetes may aggravate mitochondrial dysfunction, fueling contractile deficit [25,27]. Dysfunctional lipid metabolism in diabetes has been implicated in the development of cardiac impairment [28] and lipotoxicity resulting from accumulation of triacylglycerols and free FAs in the cytoplasm, which lead to the generation of apoptosis inducers such as diacylglycerol and ceramide [29]. In contrast, other studies have reported that FAs may actually benefit cardiac function in the course of metabolic syndrome [17,30,31]. In T1DM [32] and T2DM animal models [18,21] exhibiting impaired heart function when subjected to metabolic stress caused by hyperglycemia and elevated energy demand, it was shown that, unlike insulin, palmitate was able to rescue contractile function from the detrimental action of hyperglycemia. The beneficial effect of palmitate was concomitant with a higher content of reduced glutathione (GSH) and augmented mitochondrial ROS-scavenging capacity [18,21].
Together with peroxisomes, mitochondria represent the main subcellular compartments where lipid degradation occurs. Yet, the impact of dietary lipids on mitochondrial redox status and ROS emission, and their downstream effects on energetics are not fully elucidated. Thus, we investigated the basic mechanisms underlying the impact of lipid-precursor availability for β-oxidation on the energetic and redox responses from heart mitochondria of a previously described animal model of T1DM in STZ-treated GP that harbor glucose levels similar to those found in human T1DM [32,33]. More specifically, we analyzed how the relationship between mitochondrial respiration and ROS emission is altered as a function of PCoA in T1DM GPs and Sham controls. The experimental results are interpreted with the help of a two-compartment mitochondrial energetic-redox computational model [15] that includes β-oxidation [34] functionally linked to main redox couples and scavenging systems distributed in mitochondrial matrix and extra-matrix compartments, and transport between compartments of ROS species and GSH (Fig 1).
We quantified VO2 and H2O2 emission in isolated heart mitochondria from Sham and diabetic GPs under β-oxidation conditions with PCoA, in the presence of 0.5mM malate and 0.5mM L-carnitine, and in the absence (state 4) or presence (state 3) of 1mM ADP. As a caveat, Mal is needed to feed the TCA cycle to enable the efficient regeneration of Coenzyme A from acetyl CoA and cycling of β-oxidation [34].
Fig 2 depicts the results obtained in VO2 (Fig 2A and 2B) and H2O2 emission (Fig 2C and 2D) under states 4 and 3 respiration and as a function of PCoA concentration (0 to 800nmol PCoA/mg mito prot equivalent to 0 to 40μM PCoA: see Fig 3) in control (Sham) and diabetic (STZ) groups. State 3 respiration increased, attaining an apparent plateau level of ~125nmol O2/min/mg mito prot at 600nmol PCoA/mg mito prot in mitochondria from both Sham and diabetic GPs (Fig 2B). At amounts > 600nmol PCoA/mg mito prot VO2 further augmented an additional ~25% suggesting uncoupling of respiration (Fig 2B). State 4 respiration was stable at ~15nmol O2/min/mg mito prot up to 400nmol PCoA/mg mito prot, subsequently increasing with PCoA concentration (Fig 2A).
Mitochondrial ROS release remained approximately constant at ~ 150-200pmol H2O2/min/mg mito protein up to 200nmol PCoA/mg mito protein, in Sham and diabetic groups, and for both states 4 and 3 respiration (Fig 2C and 2D). Thereafter, a PCoA concentration-dependent increase in H2O2 emission occurs at PCoA > 200nmol/mg protein that in states 4 and 3 respiration plateaus in Sham at ~ 600pmol H2O2/min/mg mito protein (Fig 2C and 2D), In contrast, H2O2 emission from diabetic mitochondria under state 3 respiration increased almost 2-fold higher compared to Sham controls (Fig 2D), whereas in state 4 lower values than Sham were attained (Fig 2C).
The relationship between the rates of respiration and H2O2 emission in heart mitochondria from Sham and diabetic GPs is also shown in Fig 2. In state 3 respiration, the ROS efflux stays approximately constant in the VO2 range from 50 to 100 nmol O2/min/mg mito prot but ~2-fold higher in diabetic as compared to Sham (Fig 2F). At higher state 3 respiratory fluxes, ROS emission increases steadily as a function of VO2 up to ~ 130nmol O2/min/mg mito prot, and plateauing at VO2 > 150nmol O2/min/mg mito prot, although higher in diabetic than in Sham GPs (Fig 2F). At both ends of low and high respiration exhibited by mitochondria exposed to different PCoA concentrations (Fig 2A–2D), the H2O2 emission expressed as a percentage of the total O2 consumption flux [13] were for Sham/diabetic GP, respectively, 0.45%/0.89% and 0.38%/0.63% in state 3 (Fig 2F) and 0.82%/0.82% and 2.2%/1.96% in state 4 respiration (Fig 2E). Unlike in state 4 mitochondria from diabetic GPs, in which H2O2 emission augmented steadily as a function of VO2, in Sham ROS release remained independent from VO2 to increase only after a certain threshold of respiratory flux was overcome (Fig 2E).
Together, the experimental results obtained so far show that at PCoA < 200nmol/ mg prot, VO2 and H2O2 emission remained constant, while both increased as a function of the lipid precursor within the range 200–600nmol PCoA/mg mito prot, with overt (non-compensated) respiratory uncoupling and excess ROS emission happening at > 600nmol/ mg mito prot.
To help interpret the mechanisms underlying the observed increase in VO2 and ROS efflux from mitochondria, our computational model was utilized to simulate the experimental data under conditions mimicking those employed with isolated mitochondria. The simulations shown in Fig 3 reproduce the shape of the increase in VO2 as a function of PCoA concentration (0 to 40μM PCoA equivalent to 0 to 800nmol PCoA/mg mito prot: see Fig 2) observed in the experiments corresponding to states 4 (compare Fig 2A with Fig 3A) and 3 respiration (compare Fig 2B with Fig 3B). In state 4 respiration the model results reproduce the rise of VH2O2 at PCoA concentration above 20 μM while further showing that the extent of the increase in VH2O2 can be modulated by the scavenging capacity of mitochondria, i.e., achieving a higher VH2O2 at lower scavenging levels (simulated with different concentrations of glutathione reductase, GR; compare Fig 3C with Fig 2C). In state 3 respiration, model simulations are able to reproduce the rise and saturation of VH2O2 and, additionally, that the response can be modulated by the antioxidant capacity of mitochondria (compare Fig 3D with Fig 2D). Consequently, as suggested by the model simulations, the difference between Sham and diabetic mitochondrial H2O2 emission may be due to the lower scavenging capacity of the STZ-treated GPs (Fig 3).
Mechanistically speaking, our model simulations attribute a direct role to uncoupling of the mitochondrial inner membrane triggered by PCoA > 20μM (or > 400nmol PCoA/mg mito prot) as a main determinant of the modulation of the overall shape of the relationship of VO2 and VH2O2 vs. PCoA (Fig 2A–2D), along with the apparent sigmoidal behavior exhibited by these two fluxes under state 3 respiration when plotted together (Fig 2F). According to the model, the plateau of VH2O2 at high VO2, corresponding to PCoA > 30μM (or > 600nmol PCoA/mg mito prot; compare panels F from Figs 2 and 3) can be explained from a PCoA concentration-dependent uncoupling of mitochondria at high lipid concentration, reducing H2O2 emission through decreased ROS generation by the respiratory chain. On the other hand, overwhelming and/or causing impairment of the ROS scavenging systems can also modulate (up or down) the relationship between both respiratory and ROS fluxes (Figs 2 and 3).
To investigate the impact of lipid uncoupling on mitochondrial energetics, we performed experiments with isolated mitochondria consuming PCoA while monitoring NAD(P)H [35], and the results are depicted in Fig 4. As a measure of mitochondrial energetics, we monitored NAD(P)H levels during β-oxidation with PCoA 0–20μM (equivalent up to 400nmol PCoA/mg mito prot). Within the PCoA concentration range evaluated, mitochondria conserve the state 4→3 transition triggered by 5mM G/M followed by 1mM ADP, as a lipid-independent way to assess the energetic response [21,36], but start to show some impairment at 20μM PCoA, as can be judged from the response to ADP (Fig 4A). Computational simulations mimicking the experimental protocol show, firstly, that the experimentally determined initial NAD(P)H response to PCoA addition corresponds to the expected redox rise from β-oxidation while its subsequent decrease could be explained by PCoA consumption via β-oxidation (Fig 4B), as indicated by the extent of the NAD(P)H peak vs. PCoA (Fig 4A); secondly, that at PCoA > 20μM an uncoupling effect by the lipid becomes noticeable, leading to a diminished response to ADP during the state 4→3 transition (Fig 4A and 4B). The differences between experimental and model simulations data after PCoA, but before G/M, addition, can be explained by the fact that, unlike in the experiment, PCoA is “clamped” in the model at the indicated concentration. This explains that the more pronounced NADH oxidation observed in the model at 25μM compared to 20μM PCoA corresponds, to a certain extent, to the uncoupling effect of the lipid whereas in the experiments PCoA is consumed faster via β-oxidation thus the uncoupling effect is less prominent. Besides it is worth mentioning that the model could simulate the respiratory coupling ratio (RCR) observed experimentally to an acceptable approximation (~4 vs. 6, theoretical vs. experimental, respectively). As a caveat, while experimentally the RCR decreased from 6 to ~5 at 40μM PCoA (= 800nmol/mg mito prot) (Fig 2A and 2B), in the model it dropped from ~4 to ~2 at 10μM and >30μM PCoA, respectively. Thus, in the model the effect of uncoupling is higher than in the experiments.
Mechanistically, the PCoA uncoupling effect was taken into account by the model through an increase of the mitochondrial leak via the PCoA-dependent increase in proton conductance (Fig 5A and 5B). Under parametric conditions in which the model was able to reproduce the overall shape of H2O2 emission as a function of mitochondrial respiration (compare panel F from Figs 2 and 3), Fig 5 depicts changes in protein conductance and redox components of the glutathione and thioredoxin antioxidant systems as a function of PCoA, within the experimentally assayed concentration range, in both states 4 (Fig 5A and 5B) and 3 respiration (Fig 5C and 5D). At PCoA > 20μM, a significant increase in the mitochondrial H2O2 emission flux (VH2O2) happens in state 4 respiration (Fig 5A and 5B), together with a parallel decrease in mitochondrial glutathione (GSHm), accompanied by a slight decrease in the reduced pool of thioredoxin (Trx[SH]2), followed by NAD(P)H oxidation at higher PCoA concentration (Fig 5B). In state 3 respiration, VH2O2 describes a biphasic curve of increase as a function of proton conductance (Fig 5C), in which the initial phase occurs associated with an abrupt decrease in GSHm whereas the second, smoother phase appears to be determined by oxidation of Trx[SH]2) and NAD(P)H at relatively higher PCoA concentrations (Fig 5D).
Together, the qualitative behavior of the experimental and model results converge in indicating the combined involvement of uncoupling and the ROS scavenging systems in the mitochondrial energetic-redox response to PCoA during β-oxidation. The explanation offered by the model, i.e., that the concomitant increase in respiratory and H2O2 emission fluxes elicited by PCoA concentration (> 20μM) under state 3 respiration is associated with interdependent actions of lipid-elicited uncoupling and overwhelming/impairment of the matrix GSH and thioredoxin (Trx) antioxidant systems (Fig 5), was further tested experimentally. We asked whether mitochondria from diabetic GPs possessed a different protein expression profile of the antioxidant systems as compared to Sham controls, or the lipid was eliciting enzymatic activity impairment leading to loss of antioxidant capacity. The protein expression level of main mitochondrial ROS scavenging systems by Western blot revealed no significant differences between seven components of the antioxidant systems in mitochondria from Sham and diabetic GPs (Fig 6A and 6B). A similar outcome was found between wild types and two different animals models of type 2 diabetes, db/db mice [21] and Zucker diabetic fatty rats [18]. Since these results pointed out differences in activity as responsible for the results observed, we analyzed two of the main branches of the antioxidant defenses, GSH and Trx systems. Their antioxidant capacity was estimated by quantitating H2O2 emission in the absence or in the presence of 1-chloro-2,4 dinitrobenzene (DNCB) and auranofin (AF), two specific inhibitors of GSH/Trx, respectively [13,16], when mitochondria from Sham or diabetic GPs were consuming PCoA/malate or glutamate and malate (G/M). Judging from the specific mitochondrial H2O2 emission when GSH/Trx are inactive (presence of AF+DNCB) or active (absence of inhibitors), with substrates PCoA/malate, the amount of ROS generated under state 4 respiration that was scavenged was 87% and 79% in Sham and diabetic, respectively, and 83% and 73% in state 3 respiration (Fig 6C). With G/M in state 4 respiration, the amount of ROS generated that was scavenged was 97% and 95% in Sham and STZ, respectively, whereas it represented 98% and 98% in state 3 respiration (Fig 6D). The results obtained indicate that in the presence of PCoA the mitochondrial GSH/Trx scavenging capacity was lower than in G/M, and more so in diabetic than in Sham mitochondria, suggesting that the lipid oxidation or intermediates from the β-oxidation pathway reduced the activity of the antioxidant systems evaluated, resulting in their being overwhelmed. Comparatively, and with the exception of state 3 respiration in the presence of AF+DNCB, mitochondria from diabetic GPs released significantly more ROS than controls, both with PCoA (Fig 6C) or G/M (Fig 6D). The lower ROS emission from diabetic as compared to Sham mitochondria, when both GSH/Trx systems were inhibited with AF+DNCB, unveils previously described deficits in the GP animal model [32] in both the electron flow through the respiratory chain (Complex II and IV) and the phosphorylation system, which may ultimately constrain ROS generation. Model simulations of the increase in mitochondrial H2O2 emission in response to inhibition of the glutathione reductase from the GSH/Trx system, were able to reproduce semi-quantitatively the experimental results in the presence of PCoA as substrate (compare panels C-E in Fig 6). Irrespective of the changes in ROS efflux, VO2 did not change at low or high antioxidant capacity (Fig 6F).
Taken together, experimental and modeling results enable us to conclude that, compared to Shams, the larger ROS emission exhibited by heart mitochondria from diabetic GPs is due to a lipid-dependent decrease in antioxidant activity (namely GSH and Trx), and an associated mitochondrial uncoupling. These results also indicate that PCoA modulates the relationship between respiration and ROS emission from mitochondria within the concentration range of 200-600nmol/mg mito prot, with respiratory uncoupling and energetic-redox impairment occurring at PCoA > 400nmol/mg mito prot.
The main contribution of the present work is to propose and validate some of the mechanisms involved in the beneficial and detrimental consequences of lipid oxidation on mitochondrial function. A combined experimental and computational approach was applied to assess the effects on redox and energy metabolism of varying levels of PCoA, the CoA-activated form of palmitate. As a tool for analyzing the complex functional impact of lipids, we used a comprehensive computational model of mitochondrial energy-redox and ionic processes that includes compartmentation of ROS scavenging systems and the supply of AcCoA from β-oxidation (Fig 1). This computational model was built on the basis of an extensively validated model of mitochondrial energetics and redox balance [15] integrated to a β-oxidation model, based on the one developed by van Eunen and colleagues [34]. Among the existing hypotheses regarding the effects of lipids on mitochondrial physiology which were tested by the present experimental-computational approach, three of them stand out as specific contributions from the present work: (i) delineation of the combined roles of lipid on OxPhos uncoupling and antioxidant systems, namely glutathione and thioredoxin, and their modulation of the relationship between mitochondrial respiration and H2O2 emission fluxes; (ii) the lipid-mediated impairment of the ROS generation and scavenging activity, and its quantitative impact on mitochondrial H2O2 emission under both states 3 and 4 respiration, and (iii) detection of a critical lipid concentration threshold, below which energetic and redox mitochondrial functions proceed in a controlled manner, but above which these functions can be derailed. These contributions are of direct use for interpreting as well as predicting functional impairments in metabolic disorders associated with increased circulating levels of lipids and metabolic alterations in their utilization and storage as well as in ROS-dependent intracellular signaling. In the liver, Van Eunen and coworkers [34] have proposed that, under FAs overload, competition between intermediate metabolites from β-oxidation can lead to metabolic disease.
Specifically, we characterized H2O2 emission during β-oxidation-driven respiration in mitochondria isolated from diabetic and control Sham GP hearts. Experimental and modeling studies confirmed that noticeably increased respiratory uncoupling, ROS emission and energetic impairment start to occur at PCoA > 400nmol/mg mito prot (equivalent to PCoA > 20μM), due to the action of this lipid precursor of β-oxidation on uncoupling. Model simulations (Figs 3–5) and experiments (Figs 2, 4 and 6) further showed that high lipid concentration is responsible both for uncoupled respiratory flux and enhanced H2O2 emission concomitantly with impairment of the state 4→3 transition, and for overwhelmed matrix GSH/Trx scavenging systems likely due to inhibition of enzyme activity rather than differential protein level of mitochondrial antioxidant systems (Fig 6). Broadly speaking, these results are agreement with a wealth of existing knowledge while expanding on the impact of lipid excess on the impairment of GSH/Trx ROS scavenging systems and mitochondrial redox-energetics, especially under diabetic conditions [27,37]. Additionally, the present data also show that, within a certain concentration range, lipids can fulfill their energetic role without either impairing mitochondrial energetics or eliciting excessive release of ROS.
Through β-oxidation, FAs are main metabolic fuels for heart and skeletal muscle function [38]. In the heart, two thirds of the cellular ATP is generated from FAs which provide reducing equivalents (NADH and FADH2) via mitochondrial β-oxidation. The higher energy delivered by the saturated FA palmitate in the form of reducing power (i.e., three times higher than from glucose in ATP/mol substrate), provides electrons to antioxidant systems and mitochondrial respiration [17,39]. It has been shown that energization of mitochondria by substrate oxidation increases the antioxidant potential of the thioredoxin system in the mitochondrial matrix where Trx(SH)2 rose in parallel with NAD(P)H and GSH as well as mitochondrial membrane potential (ΔΨm) after glutamate/malate addition and remained high both in states 3 and 4 respiration [16,21].
The rate of β-oxidation is led by demand, implying that increased work rate and ATP demand drives faster OxPhos and tricarboxylic acid cycle activity [38]. Although existing evidence favors the idea that during T1DM the myocardial shift from glucose to FA utilization may aggravate mitochondrial and contractile dysfunctions [27,40], recent studies show that FAs may actually benefit cardiac function, at least acutely, in the course of metabolic syndrome. Cardiac myocytes from T1DM GPs exposed to high glucose and adrenergic stimulation with isoproterenol were not able to fully contract and relax, an effect that was found associated with mitochondrial oxidized redox status leading to impaired ATP synthesis along with altered Ca2+ handling and myocyte mechanical function [32]. In this T1DM GP animal model [32], as well as in db/db mice exposed to high glucose and β-adrenergic stimulation [21], and in the Zucker Diabetic Fatty rat, where hyperglycemia had a significant negative impact on contractility of heart trabeculae [18], palmitate was able to rescue contractile performance via higher antioxidant capacity of the GSH/Trx systems.
The pathogenesis of diabetes involves alterations in lipid oxidation by mitochondria. Inherited or acquired mitochondrial dysfunction may cause slow FA degradation driving the accumulation of intramyocellular lipids [41,42]. Also, mismatch between excess lipid supply with respect to demand may generate excess ROS [10]. Although acceleration of the β-oxidation flux could improve insulin sensitivity, disease may also ensue from inappropriately elevated β-oxidation flux in the absence of demand. Central to any of these possible situations is to determine the mechanisms through which mitochondria control ROS release as a function of lipid availability, and how this affects their energetic function. In this regard, the present work shows that, in the heart, mitochondria can increase their ROS release as a function of the rate of β-oxidation dependent respiration, but also that impairment of mitochondrial energetics-redox function would only start to happen after a certain threshold of PCoA concentration (> 400 nmol/min/mg prot) is crossed, triggering, for example, progressive uncoupled respiration and ROS emission in both states 4 and 3 respiration in Sham and diabetic mitochondria (Figs 2 and 3) and impairment of the state 4→3 transition (Fig 4). Enhanced H2O2 emission was caused by concomitant OxPhos uncoupling with decreased activity of matrix GSH/Trx ROS scavenging systems according to experimental (Fig 6) and modeling evidence (Fig 5). In agreement with previous reports [17,18,21,32,43,44,45], these results support the notion that in the diabetic heart the antioxidant capacity is lower, thus explaining, at least in part, the increased levels of oxidative stress observed.
Besides their metabolic role in energy provision, long-chain FAs affect cellular membranes and enzyme catalysis [46]. Non-esterified and esterified FAs interfere with mitochondrial OxPhos in vitro [47] acting as weak uncouplers [48] by increasing state 4 respiration [49,50]. Under reverse electron transport, FAs dramatically decrease mitochondrial ROS generation by an as yet unknown mechanism [51]. In contrast, the relatively low ROS emission by mitochondria under forward electron transport is significantly increased in the presence of FA [48,51]. In the present work, we show that the latter is likely true at relatively high concentrations of lipid precursor whereas at relatively lower concentrations (≤ 400nmol PCoA/mg mito prot) H2O2 emission stays constant and low, although higher in diabetic as compared to control mitochondria (Fig 2).
Mitochondria modulate both the release as well as scavenging of H2O2 from the cytoplasm thus playing a key role in cellular redox conditions and redox-dependent signaling, vital for normal cell function [17,52,53]. Using targeted viral gene transfer vectors expressing redox-sensitive GFP fused to sensor domains to measure H2O2 or oxidized glutathione in H9c2 cells, and selective knockdown (by 50%-90%) or overexpression of antioxidant enzymes, Dey and colleagues [14] showed that ROS scavenging by mitochondria significantly contributes to cytoplasmic ROS handling. Knockdown of the cytosolic antioxidant enzymes had no statistically significant effect on mitochondrial matrix H2O2, in agreement with the idea that the mitochondrial scavenger reserve capacity was high enough to buffer H2O2 diffusing into the matrix even when the cytoplasmic system was impaired [14].
Keeping a proper cellular/mitochondrial redox environment is vital for optimal excitation-contraction (EC) coupling as well as energy supply in the heart [53,54,55]. Intracellular redox balance affects Ca2+ handling by functionally stabilizing a wide range of proteins implicated in EC coupling [30] including the sarcoplasmic reticulum (SR) Ca2+ release channels, the SR Ca2+ pumps, and the sarcolemmal Na+/Ca2+ exchanger [56].
Consistent with the concept of a prominent role of lipids on governing the intracellular redox status, it was shown that palmitate determines a transition from oxidized-to-reduced redox state coupled to a marked GSH rise that abated ROS levels drastically in cardiomyocytes from T1DM and T2DM hearts. As a consequence of its favorable effect on cellular redox balance, palmitate significantly improved contractile performance in cardiomyocytes from STZ-treated GPs [32], db/db mice [21] and heart trabeculae from Zucker rats [18].
The findings described herein suggest that keeping the intracellular levels of FAs low is critical to avoid detrimental oxidative stress. Under lipid surplus, development of tissue lipotoxicity and dysfunction are linked to alterations in LD biogenesis and regulation of hydrolysis of triacylglycerols [57]. In pathologic states lipotoxicity may occur over time [29], despite triacylglycerol accumulation, when either the cellular capacity for triacylglycerol (TAG) storage is exceeded or when triglyceride pools are hydrolyzed, resulting in increased cellular free FA levels. Thus, the duration and extent of lipid overload may determine if a cell is protected or damaged.
Lipid storage and utilization appears to be a tightly regulated cellular process (reviewed in [39]). Perilipins are involved in modulation of LD storage-utilization dynamics [57]. Reduced expression of perilipins may promote both lipolysis and fat oxidation, resulting in reduced intracellular TAG and adipose mass whereas excessive lypolysis and defective lipid storage may promote insulin resistance and impaired cardiac function through chronic mitochondrial FA overload. As a matter of fact, excessive triacylglycerol catabolism by perilipin5-deficient hearts is paralleled by increased FA oxidation and enhanced ROS levels leading to age-dependent decline in heart function. Consequently, uncontrolled lipolysis and defective lipid storage may impair cardiac function through chronic mitochondrial FA overload [58,59].
Proper mitochondrial function is needed to sustain energy supply reliably while releasing ROS levels compatible with signaling. However, lipids in excess can derail both of these critical functions. In keeping with the results reported herein, cytoplasmic mechanisms for “sequestering” FAs (and those from lipotoxic intermediates) to keep their concentration low become relevant. Metabolic channeling of lipid transport and β-oxidation, involving direct delivery into mitochondria, may represent a reliable and efficient way to ensure energy supply and redox control. Such a mechanism could avoid exceeding the limit of lipid storage capacity and help in hindering lipotoxicity, which is relevant under heavy influx of FAs as happens in skeletal muscle or heart in matching energy supply with demand when subjected to high workload.
All procedures on guinea pigs to render them diabetic were approved by the Animal Care and Use Committee of Hilltop Lab Animals, Inc., and adhere to NIH public health service guidelines. For mitochondrial isolation, GPs were heparinized (500 IU) and euthanized with sodium pentobarbital (180 mg/kg intraperitoneal), following the requirements of the Institutional Animal Care/Use Committee at JHU, adherent to NIH guidelines.
GPs were rendered diabetic by Hilltop Lab Animals, Inc. (Scottsdale, PA), following the procedure that we described previously [32]. Briefly, male GPs (200-250g) were made diabetic by a single intra-peritoneal injection of buffered streptozotocin (STZ group, 80 mg/kg in citrate buffer pH 4.5). Littermate animals received an equivalent volume of vehicle (citrate buffer pH 4.5) (Sham). According to our protocol, elsewhere described [32], Sham and STZ animals were utilized after 4 weeks of STZ administration, the minimum time needed to observe the T1DM phenotype.
Diabetic GPs had 36% higher levels of glucose in blood (in mg/dl±S.E.M: Sham 153±3.2 vs. STZ 208±6.5; p<0.001, i.e., from ~8 to ~12mM glucose, n = 26 and n = 24, respectively). No significant differences in body weight between the two groups of animals were detected [32].
Procedures for the isolation and handling of mitochondria from guinea pig hearts were performed as previously described [12,35].
High-throughput–automated 96-well microplate reader analyses of respiration (XF96 extracellular flux analyzer; Seahorse Bioscience) [13], and ROS (H2O2) emission (Flex Station 3, Molecular Devices), were performed in parallel in freshly isolated mitochondria from GP heart.
The rate of O2 consumption, VO2, was evaluated in mitochondria (using the equivalent of 10μg of mitochondrial protein) under β-oxidation fueled conditions in the presence of PCoA in a dose-response manner (5–40μM, corresponding to 100–800 nmol PCoA/mg. mito prot), 0.5mM malate and 0.5mM L-carnitine, in a medium (buffer B, 200μl final assay volume) containing (in mM): 137 KCl, 2 KH2PO4, 0.5 EGTA, 2.5 MgCl2, and 20 HEPES at pH 7.2 and 37°C in presence of 0.2% fatty acid–free BSA [13].
The VO2 corresponding to states 4 and 3 respiration was determined before and after addition of 1mM ADP, respectively. Respiratory Control Ratios (state3/state4) of 5 or higher were obtained. The experimental rates are expressed in pmol min-1 mg-1 mitochondrial protein to enable comparison with data in the literature. However, modeling results are expressed in mM s-1 for VO2 and μM s-1 for VH2O2. A conversion factor of 1 μmol min-1 mg-1 protein = 16.67 mM s-1 relates both flux units based on a mitochondrial volume of 1 μl per mg of mitochondrial protein.
The day before the experiment, 120μl polyethylenimine (1:15000 dilution in buffer B of a 50% solution of polyethylenimine) were added to the wells of the XF96 plate and incubated overnight at 37°C. For comparison purpose, internal controls were run in the absence of β-oxidation with NADH-linked substrates (G/M, 5/5 mM each). Before the experiment, the solution of polyethylenimine was removed. After transfer of appropriate amounts of mitochondrial suspension into each well (10μg of mitochondrial protein), the microplate was centrifuged at 3,000 x g for 7 min at 4°C using a swinging bucket rotor (S5700; Beckman Coulter). To avoid temperature inhomogeneity effects, the plate was incubated at 37°C for 20 min before starting the assay in the Seahorse Bioscience equipment.
Using mitochondria from the same preparation (10μg mitochondrial protein), parallel fluorescence measurements of H2O2 emission with the Amplex Red kit (Invitrogen) (λexc = 530 nm and λem = 590 nm) were performed with a Flex station, under the same aforementioned substrate and buffer conditions, with the exception that BSA was not added and the plate wells were not coated with polyethylenimine. The H2O2 emissions corresponding to states 4 and 3 respiration were quantified before and after addition of 1mM ADP, respectively. The specific rate of H2O2 emission was determined in each well using an internal standard H2O2 solution, and calculated from the slopes of the normalized Amplex Red signal with respect to initial fluorescence (F0), before the addition of substrate. This procedure automatically discounts any drift of the baseline due to effects not specifically produced by the corresponding treatment [12,13].
Both VO2 and ROS emission measurements were performed in the linear range of detection with respect to the amount of mitochondrial protein utilized. Examples of raw data from O2 consumption rate and H2O2 emission measurements are displayed in Fig C in S1 Text.
Mitochondrial protein was determined using the bicinchoninic acid method protein assay kit (Thermo Fisher Scientific).
Since the experiments reported herein are performed in aqueous media to avoid introducing additional hydrophobic chemicals (i.e., diluent) other than the lipid itself and the natural substrates of mitochondrial respiration, we chose PCoA because its critical micellar concentration (CMC) is in the range of 40–60μM at 23°C, which also depends upon ionic composition and pH [60,61]. In our mitochondrial assay buffered medium, at 37°C and in the absence of mitochondria, PCoA starts to form micelles at 40μM concentration, as checked by fluorometry using 90o light scattering. To rule out unspecific surfactant effects elicited by PCoA within the concentration range utilized (up to 40μM), we measured mitochondrial respiration in the absence of malate (needed for feeding the tricarboxylic acid [TCA] cycle to enable β-oxidation to proceed). Under these conditions, VO2 with PCoA was very low, up to 3 and 7nmol O2.min-1.mg-1 prot in states 4 and 3, respectively. Comparatively, under the same conditions but in the presence of malate, respiration with PCoA was up to 22 and 125nmol O2. min-1. mg-1 prot in states 4 and 3, respectively. Importantly, as direct evidence that the mitochondria are coupled, at all PCoA concentrations there are clear differences between states 3 and 4 respiration (Fig 2A and 2B and Fig C in S1 Text), i.e., in uncoupled mitochondria states 4 and 3 respiration would be similar. Together, these controls rule out uncoupling due to unspecific permeabilization of mitochondrial membranes by PCoA.
With the aim of both controlling for putative binding effects of PCoA to albumin and its impact on mitochondrial H2O2 emission, and the reproducibility of the PCoA dose-response, we performed experiments in two different experimental setups (high throughput Flex station plate reader and low throughput fluorometric monitoring), and with the same freshly prepared mitochondria, in the absence or presence of 0.2% free FA BSA. These controls show that the results are reproducible, independently from the experimental set up, and that the results are not significantly different in the absence or presence of BSA in states 4 and 3 respiration (Fig D in S1 Text). We also present controls of mitochondrial respiration with 5mM G/M, in the absence or presence of 0.2% free FA bovine serum albumin, conducted in Seahorse XF96 equipment, showing that VO2 is not significantly affected by BSA.
NAD(P)H, and H2O2 emission with Amplex Red, were determined as previously described [12,13,35], and monitored simultaneously with a wavelength scanning fluorimeter (QuantaMaster; Photon Technology International, Inc.) using the same above mentioned medium for measuring respiration and a multidye program for simultaneous online monitoring of different fluorescent probes.
Heart tissue from Sham or STZ-treated GPs were homogenized using a polytron homogenizer into 5 volumes/weight of extraction buffer (50 mM Bis-Tris (pH 6.4), 2% SDS with protease inhibitor cocktail (EDTA Free, Roche). Protein concentration was determined using bicinchoninic acid [BCA] method protein assay kit (Thermo Fisher Scientific). Equal protein loads of extract were run on 4–12% Acrylamide Bis-Tris Gels (Life Technologies). Gels were transferred (using Biorad wet transfer apparatus) to nitrocellulose membranes with Tris/Glycine buffer and membranes blocked with Odyssey blocking reagent (Li-Cor Biosci.) in TBS buffer. Membranes were then probed with primary antibodies raised to the following antioxidant proteins: Thioredoxin 2 (Trx2, Rabbit Polyclonal, Abfrontiers), Thioredoxin Reductase 2 (TrxR2, Rabbit polyclonal, Abfrontiers), Glutatione Reductase (GR, Rabbit polyclonal, Ab Frontiers); Superoxide Dismutase 2 (SOD2, Rabbit Polyclonal, Santa Cruz Biotechnol.), Nicotinamide nucleotide transhydrogenase (NNT, Rabbit polyclonal, Aviva Systems Biology), Glutathione Peroxidase 4 (Gpx4, Rabbit polyclonal, Abcam), Peroxiredoxin 3 (Prx3, Rabbit polyclonal, Abfrontiers). Fluorescent secondary antibodies labeled with either IRDye 800CW or IRDye 680RD was used to visualize protein bands utilizing an Odyssey Infrared Scanner (Li-Cor Biosci.) and bands quantitated using Odyssey software.
L-carnitine and palmitate were from Sigma, and palmitoyl Coenzyme A ammonium salt from Avanti Polar Lipids, Inc.
Data were analyzed with the software GraphPad Prism [Ver. 6; San Diego, CA] or MicroCal Origin. Significance of the difference between treatments was evaluated with one-way ANOVA using Tukey's multiple comparison test, or with a t test [small samples, paired t test with two tail p values] and the results presented as mean±SEM [95% confidence interval].
The scheme of the two-compartment mitochondrial energy-redox model including the β-oxidation pathway is depicted in Fig 1. The model is based on our previous two-compartment model of mitochondrial energetics [15], encompassing the use of AcCoA in the TCA cycle and the provision of glutamate that replenish TCA cycle intermediates (Fig 1). Also included in the model are pH regulation, ion dynamics [36], and main scavenging systems—glutathione [GSH], thioredoxin [Trx], superoxide dismutase [SOD], catalase—distributed in mitochondrial matrix and extra-matrix compartments, four main redox couples (NADH/NAD+, NADPH/NADP+, GSH/GSSG, Trx(SH)2/TrxSS), and transport between compartments of ROS species (superoxide, O2.-, hydrogen peroxide, H2O2), and GSH [15]. In the present work the model takes into account the AcCoA supply from β-oxidation as described in the following section.
The present model (Fig 1) accounts for the β-oxidation pathway from PCoA, which was modeled based on van Eunen et al. [34]. The model formulation considers the transport of PCoA from the cytoplasmic to the mitochondrial matrix via carnitine palmitoyltransferase I (CPT1), carnitine acylcarnitine translocase (CACT) and carnitine palmitoyltransferase II (CPT2) (Eqs. S1-S3 in S1 Text). As a caveat, our formulation differs from that of van Eunen and colleagues, since the only substrates of CPT2 considered are PCoA and palmitoyl carnitine. Thus, the competition of CPT2 and CPT1 through the various acyl-carnitine and acyl-CoA species was not taken into consideration in our β-oxidation model (section 1 in S1 Text).
The β-oxidation model describes the catabolism of PCoA through the recursive action of four enzymes: very long-, long-, middle- and short-chain fatty acyl-CoA dehydrogenases, catalyzing consecutive steps in cycles, where in each of seven cycles 2-carbon units (i.e., AcCoA) are released. β-oxidation reactions occur coupled to the reduction of either flavin adenine dinucleotide (FAD) in the steps catalyzed by the fatty acyl-CoA dehydrogenases, or NAD+ by β-hydroxy-acyl CoA dehydrogenases (equations S4-S22 and S31-38, respectively, in S1 Text). A more detailed description of the β-oxidation model equations and parameters can be found in S1 Text.
The coupling between β-oxidation, the TCA cycle and mitochondrial electron transport chain is accomplished through NADH, AcCoA and FADH2; the latter reduces the electron transferring protein FAD that, in turn, donates electrons to ubiquinone in the respiratory chain [62].
The role of PCoA in the present model can be both as a substrate—providing AcCoA and reduction equivalents to feed the TCA cycle and the respiratory chain through the electron carriers NADH and FADH2—and as an uncoupler at high concentration (above 200nmol/mg mito prot). The PCoA-mediated uncoupling is modeled as an increase in the proton conductance, gH (Eq. S137 in S1 Text) as a function of the cytoplasmic PCoA concentration to the fourth power. The need for a fourth power dependence can be attributed to the system approaching the critical micellar concentration (CMC) and more molecules of PCoA being incorporated into the mitochondrial membrane, altering its permeability. A more detailed description of the β-oxidation model equations and parameters can be found in S1 Text.
Model simulations were run with a code written in MATLAB (The Mathworks, Natick, MA) using the ODEs15 integrator. In S1 Text, the system of ordinary differential equations (ODEs) (section 2 Appendix) and the Matlab code for the full computational model as well as parameters (Tables A-M), and initial conditions (Table N) are listed. Results reported correspond to steady state behavior, when the relative time derivative of each variable is ˂ 1.10−10 sec-1.
|
10.1371/journal.ppat.1004379 | Neutrophil Crawling in Capillaries; A Novel Immune Response to Staphylococcus aureus | Methicillin-resistant Staphylococcus aureus (MRSA), particularly the USA300 strain, is a highly virulent pathogen responsible for an increasing number of skin and soft tissue infections globally. Furthermore, MRSA-induced soft tissue infections can rapidly progress into life-threatening conditions, such as sepsis and necrotizing fasciitis. The importance of neutrophils in these devastating soft tissue infections remains ambiguous, partly because of our incomplete understanding of their behaviour. Spinning disk confocal microscopy was used to visualize the behaviour of GR1-labelled neutrophils in subcutaneous tissue in response to GFP-expressing MRSA attached to a foreign particle (agarose bead). We observed significant directional neutrophil recruitment towards the S. aureus agarose bead but not a control agarose bead. A significant increase in neutrophil crawling within the capillaries surrounding the infectious nidus was noted, with impaired capillary perfusion in these vessels and increased parenchymal cell death. No neutrophils were able to emigrate from capillaries. The crawling within these capillaries was mediated by the β2 and α4 integrins and blocking these integrins 2 hours post infection eliminated neutrophil crawling, improved capillary perfusion, reduced cell death and reduced lesion size. Blocking prior to infection increased pathology. Neutrophil crawling within capillaries during MRSA soft tissue infections, while potentially contributing to walling off or preventing early dissemination of the pathogen, resulted in impaired perfusion and increased tissue injury with time.
| Methicillin-resistant Staphylococcus aureus (MRSA) is a highly virulent pathogen responsible for a significant portion of skin and soft tissue infections throughout the world. We investigated the role of neutrophils in soft tissue infections, as these immune cells have been shown to be both essential for clearance of this pathogen but also for increasing tissue injury associated with S. aureus infections. We visualized the behaviour of neutrophils in the subcutaneous tissue following the introduction of a localized infectious stimulus. In addition to a profound neutrophil recruitment into the infectious nidus, significant neutrophil crawling in capillaries surrounding the region was also noted, a region of vasculature which has not previously been associated with neutrophil recruitment during infection. The neutrophils were not seen to emigrate from the capillaries but rather were retained in these vessels and maintained a crawling behaviour via β2 and α4 integrins. Blocking these integrins released the neutrophils from the capillaries, reinstituted capillary perfusion, and reduced the surrounding cell death leading to reduced lesion size following infection. Neutrophil crawling within capillaries during MRSA soft tissue infections, while potentially contributing to walling off or preventing dissemination of the pathogen, resulted in impaired perfusion and increased tissue injury.
| Staphylococcus aureus is a Gram-positive, facultatively anaerobic bacterium that poses considerable challenges to human health as a re-emerging pathogen in both hospital and community settings. As a commensal bacterium, approximately 50% of the general population carry S. aureus in the anterior nares [1]. Despite its commensal status, S. aureus is a serious pathogen, responsible for approximately 18,500 deaths per year in the United States, more than all deaths caused by AIDS, influenza, or viral hepatitis [2]. S. aureus infections, particularly those due to methicillin-resistant Staphylococcus aureus (MRSA) have been increasing in frequency in recent years, and now account for the majority of all clinical skin and soft tissue infections in the United States [3]. Importantly, these infections can cause serious complications, such as necrotizing fasciitis, necrotizing pneumonitis and sepsis [4]. A single MRSA strain, pulsotype USA300 is the dominant community acquired strain in North America [5], [6], [7], [8].
During S. aureus soft tissue infections, pattern recognition receptors such as NOD2 and TLR2, as well as complement fragments, induce signalling pathways that promote neutrophil recruitment critical for abscess formation and clearance of the bacteria [9]. The importance of neutrophils in S. aureus infections cannot be understated; neutrophils are the first to arrive at the local infectious nidus, migrate out of the vasculature, and attempt to eradicate the pathogen through an armamentarium of defenses that include oxidant production, as well as the release of proteases, defensins and various other toxins [10], [11]. Neutropenia leads to uncontrolled infection in mice, impaired healing, and increased likelihood of S. aureus dissemination that can lead to sepsis [12]. Additionally, neutrophil deficiencies (either genetic, or due to treatments such as chemotherapy or corticosteroids) make individuals highly susceptible to infection with S. aureus [11]. Paradoxically, these same defenses so critical to survival can also injure host tissues [13]–[15]. In fact, delayed neutropenia can actually provide some benefit to tissue repair associated with S. aureus soft tissue infections. Additionally, S. aureus can survive when phagocytosed by neutrophils [16] and the neutrophil may act as a “Trojan horse”, allowing the bacteria to disseminate from the point of infection and cause additional damage to the host [17]. Therefore, early neutrophil recruitment is critical to protect the host from the bacterial infection, but later neutrophil recruitment leads to additional bystander tissue damage, and may actually be a mechanism by which S. aureus enhances its virulence [18].
Neutrophil recruitment to a site of infection occurs exclusively from the post-capillary venules with no published reports of recruitment from other vascular structures such as arterioles or capillaries. The first step of the cascade subcutaneously is tethering and rolling, mediated largely by P- and E-selectin on endothelial cells binding with P-selectin glycoprotein ligand-1 (PSGL-1) on neutrophils [19]. This is followed by firm adhesion to the endothelium, typically mediated by the integrin LFA-1 [20]. Neutrophils then crawl inside the vessel, migrating along the vessel wall, usually perpendicular to or against blood flow via Mac-1 [20]. Although α4β1 (VLA-4) has also been reported to have a minor contribution in mouse neutrophils [21], in humans it appears to be upregulated and contributes primarily in severe infections such as sepsis [22]. Following adhesion and crawling, neutrophils emigrate predominantly via a junctional, paracellular pathway or at times transcellularly using integrins and intracellular adhesion molecules (ICAMs). In addition, platelet/endothelial cell adhesion molecule (PECAM-1, also known as CD31), junctional adhesion proteins (JAMs), and endothelial cell selective adhesion molecules (ESAM) play important roles in neutrophil emigration from the vasculature [23], [24].
This study made use of spinning disk intravital microscopy to visualize the behaviour of neutrophils in the first few hours following a localized nidus of S. aureus infection, introduced here as a small foreign inert particle, an agarose bead. The use of the agarose beads ensured that each mouse received a limited amount of bacteria to a very localized area that would best mimic the most common cause of S. aureus skin and soft tissue infection, namely a post puncture wound localized soft tissue infection. The approach unveiled a novel mechanism of neutrophil crawling within capillaries that we had not observed previously with intradermal injection of vast amounts of S. aureus disseminated over large areas of tissue. The random back and forth crawling of neutrophils within capillaries around the nidus of infection may occur in an attempt to prevent dissemination of bacteria through these vessels or to reduce pH in the area. Because we identified the molecular mechanisms of this capillary walk, we were able to inhibit this phenomenon and noted that this may also contribute to impaired capillary perfusion, increased cell death and increased lesion size of the classical open wound noted in patients with S. aureus infection.
In order to visualize using spinning disk confocal microscopy, neutrophil recruitment to a localized S. aureus infection as might happen following a puncture wound with secondary infection, an agarose bead was inserted into the subcutaneous tissue layer, beneath the connective tissue of the skin via a fine needle to deliver a very small and reproducible amount of bacteria. The infected bead was easily visualized, due to the green fluorescent protein (GFP)-expressing bacteria and the sterile bead was visualized due to fluorescence nanoparticles (Figure 1a). This permitted us to examine the entire process of immune cell recruitment into the infected site, allowing for effective localization of the pathogen, and clear visualization of changes in neutrophil behaviour over time. Addition of the foreign particle with S. aureus was important as this has been shown to be more pathogenic [25]–[27], thus requiring fewer colony-forming units (CFUs) to induce an infection, and permitted optimal modelling of neutrophil responses to MRSA [28]. Since the size of bead linearly predicted the amount of bacteria delivered, we used beads 250–350 µm in diameter that delivered ∼106 bacteria. Sterile beads used as negative controls had no CFUs (Fig. 1b).
Within the first few minutes, increased numbers of neutrophils could be seen rolling along the side of the vessels adjacent to infected but not non-infected beads. Many of these neutrophils adhered and emigrated. There were significantly more neutrophils emigrating as early as 1 hr after introduction of the beads that contained S. aureus when compared with sterile beads (p = 0.031). This is quantified in Figure 1c, and illustrated in a series of panels in Figure 1d and Videos S1 and S2. As demonstrated in Video S1, many neutrophils emigrated outside of the vasculature (quantified in Figure 2a) towards the direction of the bead and migrated in that direction.
In every experiment, we identified a subset of neutrophils that were dramatically deformed, elongated like sausages and crawled back and forth in a linear fashion over a distance of a few hundred microns surrounding the bead insertion site. PECAM-1 staining to delineate venules, capillaries, and arterioles in the skin revealed that this population of neutrophils was crawling in the smallest vascular structures, namely capillaries with diameters less than 10 µm (Figure 2b, Figure 2c and Video S3). This behaviour was not noted with sterile beads (Figure 2b). For these experiments, we used the RB6-8C5 anti-Ly6g antibody, which can label Ly6c (a molecule found on monocytes as well as neutrophils). Since monocytes but not neutrophils have previously been described to crawl in capillaries, 1A8, an antibody known to only label Ly6g and thus neutrophil specific [29], was also tested and confirmed that the crawling cells were indeed neutrophils (Supplementary Figure S1). The neutrophils could have been crawling on top of the capillaries using them as scaffolds; however z-stack imaging and 3D image reconstructions clearly demonstrated the neutrophils were inside and not outside the capillaries (Figure 2d).
The neutrophils adhered directly in the mainstream of blood in capillaries, bypassing any rolling event. The neutrophils then immediately began crawling but no neutrophils were ever seen to emigrate from the capillaries in the twelve experiments assessed. In the S. aureus infected beads, there were approximately 7.5 neutrophils crawling per 10 viewed capillaries although it was not unusual to see multiple neutrophils in one capillary such that approximately 35–40% of capillaries were laden with crawling neutrophils (Figure 2e, p = 0.007) while sterile beads had fewer than one neutrophil per 10 capillaries.
Since neutrophil crawling in capillaries has not been described previously, antibodies to LFA-1 and Mac-1 or to the common β subunit (CD18) of both molecules were used in an attempt to block this event. These are the two major integrins on neutrophils. Rather than pretreatment that could affect other parameters in this process, a very stringent approach of administering antibodies two hours after the infection when crawling was already at its peak was implemented. An antibody to CD18 blocked 60% of the capillary crawling (Figure 3a). Surprisingly the adhesion molecule thought to be dominant for crawling in venules, Mac-1, played no role in crawling in skin capillaries (Figure 3b) while LFA-1 antibody had a trend to reduced crawling (Figure 3c), only blocking the beta chain of the CD18 integrin reached significance (Figure 3a). The drop in recruitment we observed following the blocking of CD18 did not result in a complete inhibition of neutrophil recruitment to the capillaries; close to 40% of neutrophil recruitment remained unexplained. Inhibition of the integrin VLA-4 reduced more than 50% of the recruitment into the capillaries (Figure 3d). Tandem blockade with VLA-4 and the β2 integrin antibodies had additive inhibitory effects on the recruitment of neutrophils to the capillaries almost entirely ablating this event (p = 0.0256 Figure 3d). Each antibody intervention was compared to its own IgG isotype control (Figure 3).
We then sought to determine whether the common ligands for these integrins were binding to partner molecules in the capillaries. Our primary targets were ICAM-1 and VCAM-1. These molecules were also targeted because they are known to be expressed on endothelial cells during inflammatory conditions, and have been shown to be upregulated on endothelial cells in vitro after stimulation with components of the S. aureus cell wall [30]. Blocking ICAM-1 via anti-ICAM-1 antibodies resulted in a significant reduction in the number of neutrophils recruited to the capillaries (p = 0.0246, Figure 3e). When VCAM-1 was blocked, there was also significant decrease in the number of recruited neutrophils (p = 0.009, Figure 3e). However, blocking both VCAM-1 and ICAM-1 resulted in no significant differences in recruitment compared with either ICAM-1 or VCAM-1 alone (p = 0.482, Figure 3e), suggesting that either CD18 or α4-integrin adheres to additional ligands besides VCAM-1 and ICAM-1.
The profound deformation of neutrophils in the capillaries, suggested the potential for vessel occlusion. An intravenous injection of FITC-albumin was used to visualize perfusion through individual blood vessels [31], [32]. All vessels were first labelled with Alexa 647 conjugated to anti-CD31 antibodies (blue). Injection of FITC-albumin turned perfused vessels green (Figure 4a). Almost all of the capillaries in mice treated with sterile beads were perfused with virtually no occlusion of any vessels. Mice treated with S. aureus beads had ∼35–40% of the capillaries occluded a value significantly greater than sterile beads (Figure 4b). In general, no neutrophils were lodged inside perfused capillaries (white arrowheads, Figure 4a), and only capillaries, not venules nor arterioles were occluded. When all neutrophil sequestration in capillaries induced by S. aureus was prevented with anti-CD18 and anti-α4 antibodies, capillary occlusion was significantly reduced (Figure 4b). Capillary occlusion in the antibody-treated S. aureus mice was not significantly different from mice treated with the control sterile beads (Figure 4b) suggesting that neutrophil recruitment was causally related to vessel occlusion.
Propidium iodide was used to investigate the degree of cell death in skin. It is worth noting that under non-inflammatory conditions there is always some basal cell death that was not increased by sterile beads (Figure 4c). There was increased cell death with S. aureus beads, as compared to sterile beads (Figure 4c). When mice were treated with blocking antibodies (anti-CD18 and anti-α4) there was a significant (Figure 4d) reduction in the number of dead cells, compared to isotype control treatment.
S. aureus beads containing 1×106 CFU induced a lesion at 48 hours. Blocking neutrophil recruitment two hours after infection (anti-CD18 and anti-α4) reduced lesion size at 48 hours (Figure 5a). Importantly, blocking neutrophil recruitment two hours before infection resulted in increased lesion size compared to mice that only received S. aureus beads (Figure 5b), suggesting the initial recruitment of neutrophils is critical. Indeed, the tandem inhibition of both CD18 and anti-α4 integrins prevented all neutrophil recruitment to the infectious site, which included neutrophil adhesion (Figure 5c) and emigration in postcapillary venules (Figure 5d). However, administration of antibodies to CD18 and anti-α4 integrin two hours after infection did not affect the huge influx of neutrophils into the infectious nidus via postcapillary venules.
Recent research has focused on the complex and often paradoxical role that the innate immune system plays in S. aureus infection [2], [11]. Neutrophil recruitment is considered critically important to eradicate S. aureus infections, since deficiencies in neutrophil function can impair the host's ability to combat S. aureus infections, as demonstrated in both patients and experimental mouse models [33], [34]. However, neutrophil recruitment has also been shown to be highly cytotoxic, causing substantial bystander tissue damage in the process of controlling infections [13], [14]. In this study, we combined the use of spinning disk intravital microscopy and a novel model of S. aureus (MRSA USA 300) subcutaneous infection and visualized the complex interaction between neutrophils and this pathogen. Significant recruitment of neutrophils towards localized S. aureus infection was noted, despite the use of orders of magnitude fewer bacteria than previously reported. Although many molecules have been described to be released by and allow S. aureus to evade detection by neutrophils [35], [36], in this study very robust recruitment of neutrophils occurred. This perhaps highlights the ability of neutrophils to overcome these evasion mechanisms or highlights differences between in vivo and in vitro results. The latter may reflect the use of S. aureus that have upregulated their evasion mechanisms. Therefore, our model permitted systematic examination of bacteria localized around a foreign particle, and analysis of neutrophil behaviour around this nidus of infection.
Although our model corroborates the results of other studies [11], [12], [37]–[40] that demonstrate that neutrophils are actively recruited to the site of S. aureus infection via postcapillary venules, we observed a very novel behaviour of neutrophils adhering and subsequently crawling inside the capillaries close to the S. aureus beads. The neutrophils were physically deformed taking the shape of the capillaries, and often moved to and fro inside the vessels. Three dimensional reconstruction confirmed that the neutrophils were inside the capillaries. Further evidence supporting that neutrophils were inside the capillaries was their direct inhibitory effect on perfusion of blood through the capillaries. This recruitment occurred in response to the S. aureus-infected bead, and not due to sham surgery or the bead alone. It is unlikely that the capillaries functioned as a thoroughfare to deliver neutrophils to the site of infection as no neutrophil was ever observed to emigrate out of these vessels. It is possible that neutrophils were recruited to capillaries to occlude perfusion of the infected tissue and thereby prevent any bacterial dissemination via the vasculature from the initial infectious nidus. The reduced perfusion could also reduce pH making the environment less conducive to survival of the bacteria. Alternatively, the neutrophil sequestration in capillaries could be a defense mechanism induced by S. aureus that limits the ability of neutrophils to infiltrate the tissue. Reduced perfusion of tissue could lead to more anaerobic conditions conducive to survival of the pathogen (a facultative anaerobe), and increase tissue damage [17]. Indeed, inhibition of neutrophil recruitment into capillaries resulted in improved perfusion, reduced cell death and significantly reduced lesion size.
One complication with definitively establishing the importance of neutrophil recruitment to the capillaries was the associated inhibition of neutrophil recruitment from the venules. However, allowing neutrophils to infiltrate the tissue in significant numbers over the first 2 hours and then reversing neutrophil recruitment into capillaries reduced some of the pathogenesis associated with the S. aureus infection. Presumably, sufficient numbers of neutrophils were recruited to surround the infectious nidus and further neutrophil recruitment was unnecessary and perhaps even toxic. By contrast, preventing all recruitment of neutrophils by pretreating animals with the two anti-integrin antibodies caused greater tissue injury and more bacteria in blood, consistent with the observations by others [35], [36]. In the first few hours of infection, neutrophils are thus absolutely critical to limit bacterial dissemination. It also suggests that when unchecked by neutrophils, the bacteria can cause injury due to their release of many potent toxins. Herein, we demonstrate that in addition to surrounding the infectious nidus via emigration from venules, plugging surrounding capillaries very early might also contribute by preventing bacterial entry into the mainstream of blood. However, this latter event does cause hypoxia, cell death and increased lesion size, so eventually the occlusion of vessels causes pathophysiology and therapeutic intervention would be beneficial.
The presence of α4 integrin on neutrophils is controversial [21], [22], [41] and thought to perhaps play a greater role in mouse than human. Although neutrophil recruitment to tissues like muscle, skin and brain are primarily via the CD18 integrin [20], [42], recruitment to tissues like liver or lung can occur independent of this β2-integrin. In addition, in both mouse and human, it has been shown that neutrophils can also use the α4 integrin VLA-4 in extreme conditions such as systemic infections associated with sepsis. Plasma from a septic human patient could induce the expression of α4 integrin on the surface of neutrophils from healthy patients [22]. In addition, this molecule induced functional adhesion to its ligand VCAM-1, although other ligands for α4 integrin also exist. In chronic adjuvant arthritis inflammation which was associated with a systemic vasculitis, α4 integrin was important in neutrophil recruitment, but VCAM-1 was not involved [43].
Herein, in a localized S. aureus infection, the recruitment to the capillaries was mediated by both the β2 integrins and the α4 integrin. Blocking either Mac1 or LFA-1 alone in our model did not have significant effects on neutrophil recruitment within the capillaries, suggesting that these subunits likely play overlapping roles in the capillaries, and both must be blocked in addition to α4 integrin in order to prevent neutrophil recruitment. Perhaps not surprisingly, a role for VCAM-1 was revealed for some of the neutrophil recruitment into capillaries as this molecule is expressed constitutively in murine skin endothelium [44]. The fact that ICAM-1 and VCAM-1 did not completely block recruitment to capillaries suggests that other molecules also are used by the integrins. This is not surprising since these integrins can adhere to many different ligands.
With the discovery of neutrophil crawling in capillaries a number of new issues arise. First, how much capillary occlusion is necessary to cause tissue injury. Although 35–40% of capillaries were occluded in our study, it was impossible to exclude the possibility that injury also occurred due to the proteases and oxidants released by neutrophils that infiltrated the injury site via the post-capillary venules. It is also important to note that all capillary beds are different, raising the importance of imaging the skin when studying skin infections and imaging the liver when studying liver infections. Indeed, CD44 and not integrins are used by neutrophils in the sinusoids of the liver postinfection [45]. Moreover, capillaries of other organs may not have neutrophil crawling or capillary plugging. Finally the reason for why neutrophils crawl in capillaries is unclear. However millions of years of evolutionary pressure directing the fight between this common pathogen and the host may have evolved an important anti-microbial process or an important bacterial evasion mechanism that is still not entirely understood.
In conclusion, in this study a novel neutrophil behaviour has been identified in response to subcutaneous infection due to a virulent strain of S. aureus. Using spinning disk confocal microscopy, we noted significant neutrophil recruitment into capillaries surrounding the infectious nidus. We determined that the molecules responsible for this behaviour were the β2 and α4 integrins, binding in part with ICAM-1 and VCAM-1, and causing occlusion of the capillary microvasculature. Blocking this recruitment at a delayed time point, reduced the malperfusion, cell death and lesion size that developed several days after infection with the S. aureus infected bead. As S. aureus becomes more resistant to antibiotics, understanding the mechanisms that underlie the pathogenesis of this infection will enhance the likelihood of non-antibiotic therapeutic intervention.
C57BL6 male mice (Jackson, Bar Harbour), aged 6–8 weeks were used for all experiments.
All animal protocols were submitted to the animal care committee of the University of Calgary under the protocol number AC12-0222. All animal protocols approved by the animal care committee of the University of Calgary and complied with the Canadian Animal Care guidelines.
Green fluorescent protein (GFP)-expressing S. aureus was made from a previously isolated clinical strain USA300-2406, described previously [46]. Bacteria were grown in 5 ml of Brain Heart Infusion (BHI) media (Becton and Dickenson, Sparks, MD), and were incubated overnight at 37°C. GFP-expressing S. aureus (strain USA300-2406) was grown in 20 µg/ml chloramphenicol (EMD Biosciences, La Jolla, CA).
Agarose beads were used to deliver bacteria on a foreign particle, based on an existing model of cystic fibrosis [47]. S. aureus were grown overnight in BHI (20 µg/ml chloramphenicol) at 37°C. The next morning, 5 ml of overnight media was mixed with 45 ml of fresh BHI (20 µg/ml chloramphenicol), and grown for a further two hours. S. aureus was then centrifuged at 2000 rpm for 10 minutes, and resuspended in 250 µl of 1× phosphate buffered saline (PBS). 10 µl of PBS containing bacteria were serially diluted and plated, to measure CFUs. The remaining PBS was then added to 2.25 ml of liquid 1.5% TSA agar. The TSA/PBS/S. aureus solution was then slowly injected into a mixture of 40 ml of mineral oil (Sigma-Aldrich, St Louis, MO) and 400 µl of Tween 20 (Sigma-Aldrich, St Louis, MO), which was gently stirred at 4°C, yielding spherical agarose beads embedded with S. aureus. After 15 minutes, the solution was centrifuged at 2000 rpm for 10 minutes. The mineral oil layer was removed, and beads were washed with PBS and resuspended, then spun again at 2000 rpm. This wash step was repeated three times. Beads were then washed in a 100 µl filter, and resuspended with PBS. Beads were stored at 4°C for up to 6 days. Plating of 1–6 day old beads on fresh agar showed no loss of CFU within this timeframe. For sterile beads (control), bacteria were replaced with 2 µl of Fluoresbrite plan yg 1.0 micron microspheres (Polysciences, Warrington, PA)
Images were analyzed by removing light collected from the 488 and 561 and 649 nanometer channels of the spinning disc confocal microscope. The contrast and brightness used to analyze data was held constant for analysis of each set of experiments. The number of neutrophils at 30, 60, 90, and 120 minutes was counted by using the point tool function of Volocity (Perkin-Elemer, Waltham, MA). For analysis of location of neutrophils, the 649 channel was used to examine the vasculature. Neutrophils that co-localized with CD31 labelled vessels were determined to be inside the capillaries if the vessels did not exceed 10 µm in width. Neutrophils both inside and outside the capillaries were counted using the point tool function of Volocity.
Data was analyzed using the Students t-tests to compare two different conditions. When more than one comparison was made in the same graph, a bonferroni correction was used to correct for false positives. When three variables were all compared with one another in the same graph, a one-way analysis of variance (ANOVA) with a Bonferroni correction was used. All statistical analysis was performed using the statistical software GraphPad prism 4, version 4.03 (GraphPad Software Inc., La Jolla, CA).
|
10.1371/journal.pntd.0005489 | Pre-control relationship of onchocercal skin disease with onchocercal infection in Guinea Savanna, Northern Nigeria | Onchocerca volvulus infection can result in blindness, itching and skin lesions. Previous research concentrated on blindness.
A clinical classification system of the cutaneous changes in onchocerciasis was used for the first time in this study within the context of an early ivermectin drug trial in the savanna region of Kaduna State, northern Nigeria. Skin examinations were performed in 6,790 individuals aged 5+ years in endemic communities and 1,343 individuals in nonendemic communities.
There was increased risk for all forms of onchocercal skin disease in endemic communities with the most common finding being the presence of nodules (1,438 individuals, 21.2%), followed by atrophy (367, 6.1% of those < 50 years), acute papular onchodermatitis, APOD (233, 3.4%), depigmentation (216, 3.2%) and chronic papular onchodermatitis, CPOD (155, 2.3%). A further 645 individuals (9.5%) complained of pruritus but had completely normal skin. APOD was more common in males whereas atrophy, hanging groin and nodules were more common in females. After controlling for age and sex, microfilarial positivity was a risk factor for CPOD, depigmentation, hanging groin and nodules (OR 1.54, p = 0.046; OR 2.29, p = 0.002; OR 2.18, p = 0.002 and OR 3.80, p <0.001 respectively). Comparable results were found using presence of nodules as the marker for infection. Microfilarial load showed similar, though weaker, results. A total of 2621(38.6%) endemic residents had itching with normal skin, or had one or more types of onchocercal skin disease including nodules, which may be considered as a composite index of the overall prevalence of onchocercal skin disease.
Significant levels of onchocercal skin disease were documented in this savanna area, which subsequently resulted in a reassessment of the true burden of skin disease in onchocerciasis. This paper represents the first detailed report of the association of onchocercal skin disease with markers for onchocercal infection.
| Onchocerciasis is a tropical parasitic infection caused by the nematode worm Onchocerca volvulus. The disease mainly occurs across tropical Africa and infection can result in blindness, debilitating itching and a variety of skin changes. Initial research concentrated mainly on the problem of blindness. A number of studies on onchocercal skin disease were performed but were difficult to interpret and compare because of the use of inconsistent terminology. Within the setting of one of the early trials of ivermectin in a savanna area of northern Nigeria, where there were known high rates of onchocercal blindness, we used a novel clinical classification of the skin changes in onchocerciasis. We identified significant levels of itching and various forms of onchocercal skin disease within these endemic communities. A positive skin-snip result proved to be a significant risk factor for the presence of chronic papular onchodermatitis (CPOD), depigmentation, hanging groin and onchocercal nodules. Comparable results were found when the presence of nodules was used as the marker for infection and similar, though weaker odds ratios were found with microfilarial load per se. The findings triggered a reassessment of the true burden of skin disease in onchocerciasis. It is the first detailed report of the association between onchocercal skin disease and markers of infection.
| Onchocerciasis affects approximately 17 million people worldwide [1] with the main burden of disease occurring throughout tropical Africa. The consequences of infection with the nematode Onchocerca volvulus include blindness, debilitating pruritus and skin lesions. Initial research concentrated on blindness because of its devastating socio-economic impact. Prior to any control activities in West Africa, it was common to see entire villages near rivers, which were the breeding sites for the Simulium vector, completely abandoned for less fertile land elsewhere [2].
It is established that the two epidemiological patterns of ocular onchocerciasis result from two strains of the O. volvulus parasite. These can be differentiated by DNA sequencing [3;4]. A severe form of ocular disease occurs primarily in savanna areas, where communities often suffer from a high prevalence of onchocercal blindness. Conversely, in rainforest areas onchocercal blindness is less common. The Onchocerciasis Control Programme, OCP (1974–2002) was a large-scale, multi-country programme which aimed to control the vector in countries with high rates of blinding onchocerciasis by regular aerial larviciding of rivers. At its peak, the programme included eleven countries in West Africa and, although expensive, was very successful in interrupting transmission [5].
In 1987, Merck & Co., Inc. announced the donation of Mectizan (ivermectin) for the treatment of onchocerciasis worldwide for as long as necessary. The early large-scale ivermectin trials, including the study described here, were conducted in non-OCP endemic areas, known to have high rates of onchocercal blindness. As ivermectin is mainly microfilaricidal it is necessary to continue annual treatment of the human host population throughout the adult female worms' life-span of up to 14–16 years.
Up until this time, there had been a number of studies on onchocercal skin disease [6–14] but they were difficult to compare because of the lack of a scheme to describe the cutaneous changes. The true burden of skin disease caused by onchocerciasis in any endemic area and globally was therefore unclear. Based on previous field work in Sudan (R.J. Hay, C.D. Mackenzie and J.W. Williams unpublished observations) and in Ecuador [15], a new skin classification scheme [16] was used for the first time in the study described here.
The categories in the clinical classification used for the first time in this study were "consistent with" rather than specific for onchocerciasis. Each subtype has its own clinical differential diagnosis. The objectives of this study were to field test the skin classification scheme on a large scale and to determine the prevalence of onchocerciasis-related skin changes in this savanna setting, a region which was already known to have high rates of onchocercal blindness. In addition, we aimed to explore the association of factors such as age, gender and markers of infection with various forms of onchocercal skin disease.
The data were collected during 1988–1989 prior to commencement of ivermectin therapy. The urgent focus of the work at the time concerned the ocular findings and impact of ivermectin therapy. At the time the skin classification system was published and was adopted for a WHO multi-country study that established the true public health importance of onchocercal skin disease. Following the multi-country study, the African Programme for Onchocerciasis Control, APOC, using community-directed treatment with ivermectin, was established in many countries. Issues such as ivermectin treatment needing to be repeated annually for many years, possible ivermectin resistance and side-effects in areas co-endemic with Loa-loa have encouraged the continued search for novel therapies for onchocerciasis. The subsequent widespread use of ivermectin has now rendered our baseline dataset unique in having skinsnip results combined with detailed and validated records of skin pathology in an untreated population. The association of onchocercal skin changes with levels of infection is critical in improving understanding of pathogenesis and aiding the search for new treatments for onchocerciasis. We have therefore further examined the data and report the findings here and wish to ensure the dataset is available to the wider research community.
The study was approved by the Medical Ethical Committee of Ahmadu Bello University, Zaria, Nigeria, Project Number ESC/89/00024. Approval for the parent project was also approved by the Ministry of Public Health, Nigeria.
As previously described [17], the study was conducted in the guinea savanna of Kaduna State, northern Nigeria in a subsistence farming mesoendemic area. The initial selection of communities was based on entomological data on Simulium blackfly breeding sites, the vectors being S. damnosum ss [dominant] and S. sirbanum. In 1988, 36 villages were mapped and residential compounds numbered. All individuals aged 5 years and above were registered and photographed and skin snips taken. The criterion for inclusion of communities in the trial was a microfilarial prevalence of at least 30% among those aged 20 years or above. Thirty four of the communities met the criteria for inclusion in the study. The overall prevalence of positive skin snips among villagers aged 5 years and above was 49% and 72% among those aged 20 years and above.
A further two nonendemic communities with similar ethnic and socio-economic characteristics as the endemic communities were selected in Fatika, Zaria Local Government Area of Kaduna State. These communities served as control communities on the basis of a very low prevalence of onchocercal infection (0.3% positive skin snips in those aged 5 years and above).
Two skin snips were taken from both iliac crests in each individual, using a 2mm Holth corneoscleral punch. Snips were incubated for 24–30 hours in normal saline, then weighed and fixed in formal saline. Emergent microfilariae were later counted in Kaduna by independent counters (3 for endemic and 2 for nonendemic communities). In case of disagreement between counters the well was re-examined by an independent observer. The microfilarial load per mg of skin (mf/mg) was calculated for each individual and the community microfilarial load (CMFL) was calculated for each village.
Within one year of registration, the study team performed ophthalmic [17] and dermatological examinations on individuals aged 5 years and above. The skin examinations were conducted from 1988–1989. Individuals were asked in the local language (Hausa) about the presence of itching. Skin examinations were conducted privately in natural daylight. The presence or absence of palpable onchocercal nodules and the various types of onchocercal skin disease (OSD) were noted on a form using a standard clinical classification system [16]. Acute papular onchodermatitis (APOD), chronic papular onchodermatitis (CPOD), lichenified papular onchodermatitis (LOD), atrophy, depigmentation (DPM), hanging groin and onchocercal nodules were documented. APOD, CPOD and LOD were collectively termed reactive skin lesions. In order to avoid confusion with senile or age-related atrophy of the skin, onchocercal skin atrophy was only recorded as an abnormal finding in individuals aged less than 50 years. The categories in the clinical classification are "consistent with" rather than specific for onchocerciasis and each subtype has its own clinical differential diagnosis [16]. The presence of non-onchocercal skin disease was also recorded. All skin examinations were conducted masked to skin snip and eye examination results.
Following preliminary discussions with the village heads, informed consent for skin snipping and subsequent eye and skin examinations was obtained in the local Hausa language and confirmed with a signature or finger-print.
Data were double-entered onto computers with the software package DBase III+ and were cleaned with DBase III+ and SAS/PC. Analyses were done using STATA/IC 12.0 (http://www.stata.com). Logistic regression, both univariable and multivariable, was undertaken to investigate the association between various forms of onchocercal skin disease and onchocercal infection (using microfilarial (mf) positivity, microfilarial load and nodules as separate markers), age and gender. Account was taken for clustering by village in all analyses by using linearization-based variance estimators.
A substantial effort was made to minimise inter-observer variation before the start of data collection. An inter-observer variation study for recording itching and each of the categories of onchocercal skin disease was conducted under similar lighting conditions by two general physicians (HNY and NN) on 291 individuals. Good Kappa values were obtained as follows: pruritus with clinically normal skin = 0.68 (95% confidence intervals (CI) 0.59–0.76); APOD = 0.72 (CI 0.60–0.85); CPOD = 0.79 (CI 0.56–1.03); LOD = 1.0 (CI 0.98–1.02); atrophy = 0.84 (0.68–1.0) and nodules = 0.83 (0.75–0.90). There were insufficient numbers of individuals with DPM (n = 3) or hanging groin (n = 3) to allow comparision. However, these are two of the most easily identified forms of onchocerciasis-related pathology.
A total of 8,140 individuals aged 5 years and above were registered at the time of the census and 7,072 were present at the time of examinations one year later. Of these 6,790 consented to a skin examination. Skin snip data was available for 6,643 (97.8% of those examined, Table 1). Overall 3,276 (49.3% of those snipped) had positive skin snips. Of those who were skin snip positive, the majority (1872/3276, 57.1%) had low microfilarial loads (≤10.00 mf/mg skin).
A total of 1,886 individuals aged 5 years and above were registered at the time of census. Skin examinations were conducted on 1,343 individuals. Skin snip data was available for 1,342 (99.9% of those examined, Table 1). Only 4 persons (0.3%) in the non-endemic villages had positive skin snips, all with microfilarial loads ≤10.00 mf/mg skin.
The prevalences of skin changes consistent with onchocerciasis in endemic and nonendemic communities are summarised in Table 2. In endemic communities the most common clinical sign was the presence of palpable onchocercal nodules in 1438 (21.2%) of the examined population. The next most frequent finding was atrophy (367; 6.1% of those aged <50 years) followed by APOD (233, 3.4%), depigmentation (216, 3.2%) and CPOD (155, 2.3%). Overall, including nodules, 1976 (29.1%) persons had one or more forms of OSD. A further 645 individuals (9.5%) complained of pruritus but had completely normal skin. A total of 2621(38.6%) of the examined endemic population were found to complain of itching with normal skin or had one or more types of OSD including nodules, which may be considered as a composite index of the overall prevalence of onchocercal skin disease.
In the nonendemic communities all forms of onchocercal skin disease and the presence of itching with normal skin were rare. No cases of LOD were identified. The most common finding potentially associated with onchocercal infection was atrophy in 47 persons (3.9% of those aged < 50 years).
There was an increased risk for all forms of OSD in endemic compared with nonendemic communities. The highest risk was seen for the presence of nodules (OR 89.94), followed by hanging groin (OR 19.04), depigmentation (OR 11.0), APOD (OR 9.51), Reactive Skin Lesions (OR 4.83), CPOD (OR 2.83, all p<0.001) and atrophy (OR 1.6, p = 0.001). Endemic communities also had a higher risk of itching alone with clinically normal skin (OR 5.53) and itching alone or one or more OSD- associated findings including nodules (OR 8.45, both p<0.001).
Figs 1–5 show age and gender-specific prevalence of onchocercal skin disease and markers of onchocercal infection in endemic communities. Fig 1 shows the age and gender-specific prevalence for itching with clinically normal skin. The highest prevalence was seen in the youngest age group of 5–14 years, with an overall trend to reduce with age.
The age and gender-specific prevalence of APOD, CPOD and reactive onchocercal skin lesions in endemic communities are shown in Fig 2. There were only five cases of LOD, four of whom were female. Three were aged 15–24 and the remaining two were older. The highest prevalence of APOD was seen in those aged 5–14. Whilst the prevalence was fairly constant in older females, it possibly rose again in males aged 45 years or more. CPOD was more common in those aged 45 years or more. The overall prevalence of reactive skin disease reflects these findings. Among females there was a small rise in prevalence with age. Among males prevalence was lowest between 15 and 44 years of age.
Fig 3 shows the age and gender distribution of markers of chronic disease. All markers of chronic disease increased strongly with age: atrophy and depigmentation from 25 years onwards and hanging groin from 35 years onwards. Atrophy and hanging groin had a consistently higher prevalence in females.There was little gender difference with depigmentation.
A composite index of the overall prevalence of onchocercal skin disease was created of itching alone or any onchocercal skin lesion including nodules. Fig 4 shows the age and gender-specific prevalence of this index. There was a marked increase in prevalence with age for both sexes, with a trend for prevalence in females aged 25 years and above. In both sexes the prevalence plateaued from age 35 years onwards, reaching around 70% in females and 60% in males.
Fig 5 shows age- and gender-specific prevalence of infection as determined by mf positivity and presence of nodules. The prevalence of mf positivity rose steeply with age, reaching a plateau of approximately 70% from the age of 15 years onwards in males and 25 years onwards in females. In contrast there was a steady rise in nodule prevalence until the age of 45 years. In the older age groups nodules were consistently more prevalent in females. Nodule prevalence plateaued at approximately 40% in males and 50% in females.
The prevalence of the various types of onchocercal skin disease were examined by two markers of onchocercal infection, skin snip positivity and presence of palpable onchocercal nodule (S2 Table). CPOD, atrophy, depigmentation and hanging groin were more prevalent among those with positive skin snips and among those with nodules. Neither APOD nor itching without skin lesions were more common among those with these markers of onchocercal infection.
Multivariable logistic regression analyses were performed to investigate the association of each form of onchocercal skin disease with infection after controlling for age and gender (Table 3). Three models were created using mf positivity, mf load or nodules as the marker of infection. Mf positivity was associated with increased risk of CPOD (OR = 1.54, p = 0.046), depigmentation (OR = 2.29, p = 0.002), hanging groin (OR = 2.18, p value = 0.002) and nodules (OR = 3.80, p<0.001).
Comparable increased risks were noted when the presence of nodules was used as the marker of infection (CPOD OR = 1.59, p = 0.016; depigmentation OR = 1.70, p = 0.001; hanging groin OR = 2.76, p <0.001). Nodules were also a risk factor for atrophy (OR 1.53, p = 0.001).
Similar, though weaker, increased risks were found when mf load per se was used as the marker of infection (atrophy OR = 1.24, p = 0.014; depigmentation OR = 1.43, p <0.001; hanging groin OR = 1.30, p <0.001; and nodules OR = 1.86, p <0.001).
After controlling for age and sex in the multivariable analysis, mf load was protective for the presence of itching alone with clinically normal skin (OR 0.80, p = 0.026). After controlling for age and sex there were no significant associations for APOD with microfilarial positivity, mf load or presence of nodules.
In the multivariate analysis the composite index of onchocercal skin disease (itching alone with clinically normal skin or any onchocercal skin disease including nodules) was significantly associated with mf positivity (OR = 2.21, p<0.001). The association was weaker if mf load was used as the marker of infection (OR = 1.56, p<0.001).
The prevalence of non-onchocercal skin disease by meso- and nonendemic communities is shown in Table 4. Overall the prevalence of non-onchocercal skin disease was lower in the endemic communities (53.7% vs 69.1%, OR 0.52, p = 0.024).
The most common non-onchocercal skin disease was acne which was only half as prevalent in endemic communities. In contrast pyoderma, scabies and miliaria were more prevalent in the endemic communities.
Findings of pityriasis versicolor, dermatophyte infection and insect bites were all less common in the endemic communities. A large number of other skin conditions were also recorded, which were generally less prevalent in the endemic communities. These included keloids, eczema, burn scars, warts and erythema ab igne.
This study reports the first use of the classification of the cutaneous changes associated with onchocerciasis on a large scale. It is worth emphasizing that this scheme, which is based purely on clinical signs, was applied without the physician being aware of the individual's skin snip status or results of their eye examination. The clinical signs of each type of onchocercal skin disease are relatively non-specific and hence each category has its own list of clinical differential diagnoses [16]. The most commonly observed onchocercal skin finding in the endemic communities was onchocercal nodules (21.2%) followed by cutaneous atrophy (6.1% of individuals aged <50 years), APOD (3.4%) and depigmentation (3.2%). LOD was rare in this savanna area with only five cases identified. Careful clinical examination highlighted a further group of individuals (9.5% of endemic residents) who complained of itching but had clinically normal skin. Thus, within this endemic region, a total of 38.6% of the population aged 5 years and above had itching with normal skin or one or more forms of OSD including nodules. This demonstrates a remarkably high overall prevalence of onchocercal skin disease in these communities. Although savanna areas of sub-Saharan Africa were known to have high burdens of blinding onchocercal eye disease, this was the first time high levels of onchocercal skin disease had been documented in a savanna region.
The prevalence of onchocercal skin disease in these communities is an underestimate since a diagnosis of atrophy was limited to those aged less than 50 years in order to avoid confusion with senile atrophy of the skin. APOD, CPOD and LOD are by definition, itchy conditions and the degree of the burden of itching suffered by these residents has not been captured. The true prevalence of onchocercal-induced atrophy and itching in the community will therefore be higher.
As mentioned previously, the various skin changes consistent with onchocerciasis may be clinically non-specific. The strong positive associations of CPOD, depigmentation and hanging groin with microfilarial positivity, independent of age and sex, show these classification sub-groupings are relevant to onchocercal infection. It is possible that the relationship of these clinical findings with onchocercal infection may be even stronger since it is known that skin snip sensitivity is increased with higher numbers of skin snips. Sensitivity for infection may be even further increased by the use of PCR or LAMP of skin snips [18,19]. CPOD, atrophy, depigmentation and hanging groin were also associated with the presence of nodules, and atrophy, depigmentation, hanging groin and nodules were all associated with microfilarial load.
It is possible that individuals with pruritus but clinically normal skin had early, light infections which were not always detectable by the routine number of two skin snips performed in this study and this may explain why no association with mf positivity could be documented in the multivariable regression analyses. Paradoxically mf load was found to be inversely associated with itching alone. The reason for this is unclear.
Similarly APOD, which was more common in 5–14 year olds who presumably also had early, light, onchocercal infections, did not show associations with infection in this study. Microfilarial remnants have previously been demonstrated in epidermal microabscesses in skin biopsies of APOD (Murdoch et al. Brit J Dermatol 1990: 123 (Suppl 37):28).
There were only five cases of LOD, three of whom were microfilaria positive. LOD is associated with hyperimmune host immune responses, skin snips are often negative and microfilariae are difficult to find on skin biopsy [11]. Atrophy, hanging groin and nodules were all more common in females.
There were only 4 persons skin snip positive in the nonendemic villages. The prevalence of all forms of skin disease consistent with onchocercal infection was very low in nonendemic villages, supporting the clinical classification scheme as consistent with onchocercal infection. A caveat is that the clinical observers were aware that they were in a nonendemic area. It is possible that this may have produced an element of observer bias, but the observers were masked to all skin snip and eye examination results. The higher prevalence of pyoderma in endemic communities might be a result of pruritus due to either onchocerciasis or the higher prevalence of scabies causing excessive excoriation and secondary bacterial infection. The increased prevalence of other skin diseases in the nonendemic communities might be explained by small changes such as a wart or keloid being easier to see if the skin was otherwise clear.
At all ages, microfilarial positivity was an earlier and more sensitive marker of onchocercal infection than prevalence of nodules (Fig 5). Rapid epidemiological mapping of onchocerciasis (REMO) followed by rapid epidemiological assessment (REA), the examination of samples of 30–50 adult men for the presence of nodules [20,21], is now a well established and useful method to quickly assess levels of onchocercal endemicity in areas to decide priorities for mass drug treatment. Nodule palpation in adult males underestimates the prevalence of infection compared with the more time-consuming and costly process of skin snipping and quantitative models have been developed to describe the association and estimate microfilarial `prevalence from measured nodule prevalence [22]. The results presented here suggest a possible further underestimate as adult females consistently carry the larger burden of disease due to nodules. The practical difficulties, however, of undressing women in privacy for palpation of nodules probably outweigh any benefits that might be gained from changing practice.
Since this study was conducted, the skin classification scheme been successfully used in rain-forest onchocerciasis-endemic areas in Africa [23] as well as in a variety of mass drug treatment [24;25], psychosocial and economic [26–28], genetic [29] and immunological studies [30;31].
The multi-country rainforest study [23], which used prevalence of nodules as a marker of endemicity, confirmed that onchocercal skin disease was a significant public health problem in affected areas with an overall prevalence of onchocercal skin lesions (excluding nodules) in those aged 5 years and above of 28%. The prevalence of APOD, CPOD, LOD and depigmentation was higher in the rainforest areas compared to the findings reported here from a savanna region. Excluding nodules, the most common form of OSD in rainforest areas was CPOD at 13%, whereas atrophy was the most common in this savanna area. Interestingly atrophy was the only type of OSD which was more common in the savanna than rain-forest regions.
This savanna study and the later multicountry rainforest skin survey prompted a reassessment of the skin disease burden of onchocerciasis. The rainforest study's results raised the possibility that many other endemic rainforest areas across Africa had significant levels of onchocercal skin disease, and hence merited mass drug treatment, even though they had low levels of blinding onchocercal eye disease. In 1995, a new control programme, the African Programme for Onchocerciasis Control (APOC), was launched. APOC used a sustainable strategy of community-directed treatment with ivermectin (CDTI) whereby the communities themselves implemented annual ivermectin distribution. APOC covered 20 countries and closed in 2015.
A multicountry skin survey after five or six years of annual ivermectin therapy in meso-and hyperendemic communities [25] revealed significant reductions in the odds ratios of itching (with and without accompanying OSD), APOD, CPOD, LOD, reactive skin lesions, depigmentation and nodules. Atrophy was not assessed.
The aim of the skin classification system is to facilitate standardisation of data collected by different observers in different geographical settings and enable comparisons of results. Observers in the multi-country rainforest study [23] were all trained by the same author (MM) and underwent an inter-observer variation study prior to data collection. Furthermore several of the same clinicians collected the data in the multi-country study performed after five or six years of ivermectin therapy [25], again following an inter-observer variation study. It is reasonable to claim therefore that the results from the current study and from these latter two studies are comparable.
In contrast to the Onchocerciasis Elimination Programme in the Americas (OEPA), which aimed to eliminate onchocerciasis from foci in Central and Southern America [32], APOC's original objective was to try to control onchocerciasis as a public health problem. It was unclear whether ivermectin could interrupt transmission and eventually eradicate onchocerciasis in Africa where the vectors were known to be more efficient. Studies in Mali and Senegal [33], however, have shown that after 15–17 years of annual or six monthly ivermectin therapy the prevalence of microfilariae and vector infectivity rates were either zero or below postulated thresholds for elimination, which triggered revision of APOC's stated objective to one of elimination of onchocerciasis in Africa. It is exciting to note that follow up studies in the same savanna communities reported here reveal that after 15–17 years of annual ivermectin therapy the community prevalence of mf positivity has fallen to 0%. All 3,703 individuals examined were skin snip negative [34]. This represents the first evidence from an APOC operational area that ivermectin treatment alone could eliminate onchocerciasis infection and potentially disease transmission in endemic areas in Africa.
The need for repeated treatments of ivermectin over many years has led to concerns of development of ivermectin resistance [35]. Research continues for a macrofilaricidal drug [36,37].
The pathogenesis of the cutaneous changes in onchocerciasis is still not fully understood. There is a spectrum of immune response to infection, with some infected persons showing a minimal immune response to parasite antigens, allowing the proliferation of microfilariae in the absence of clinical symptoms, while others have an intact and symptomatic immune response [38]. An immunogenetic basis for this clinical spectrum has been suggested [29;39;40] and differing isotypic antibody responses [30] and cellular immune responses [31;41–45] may play a role. The endosymbiotic bacteria Wolbachia are essential for the pathogenesis of O. volvulus keratitis in a mouse model [46]. The relative Wolbachia DNA burden was previously thought to explain the difference in ocular pathogenicity of the two strains of O. volvulus [47] but recent whole-genome data challenges this concept [48]. It is hoped that improved understanding of the pathogenesis of onchocercal skin disease, including clinico-pathological correlations of how it is related to human host age, sex and microfilarial load as delineated here, may help in the endeavour to identify novel treatments.
The Global Burden of Disease Study has estimated 15,531,500 prevalent cases of onchocerciasis remaining in 2015, representing a 29.1% reduction in global prevalence since 2005. Based on the skin clinical classification, the disease burden from onchocercal skin disease has been now included alongside onchocercal eye disease to form an overall global estimate of 1,135,700 years lived with disability (YLDs) due to onchocercal infection [49].
In summary the skin classification scheme for the cutaneous changes in onchocerciasis was easy to use in the field, reproducible and a useful tool to assess the prevalence of onchocerciasis skin disease in this savanna region of northern Nigera. We report that the most common onchocercal skin finding was nodules, followed by atrophy, APOD, depigmentation and CPOD. APOD was more common in males whereas atrophy, hanging groin and nodules were more common in females. Microfilarial positivity and the presence of nodules were associated with CPOD, depigmentation and hanging groin. Nodules were also a risk factor for atrophy whereas microfilarial load showed similar, though weaker associations.
The use of the skin classification scheme in other prevalence and socio-economic studies has contributed towards an ever-growing body of research which aims to estimate the true global disease burden of onchocerciasis, which takes into account not only ophthalmological, but also cutaneous effects of the disease on its unfortunate sufferers.
|
10.1371/journal.pgen.1003225 | Loci Associated with N-Glycosylation of Human Immunoglobulin G Show Pleiotropy with Autoimmune Diseases and Haematological Cancers | Glycosylation of immunoglobulin G (IgG) influences IgG effector function by modulating binding to Fc receptors. To identify genetic loci associated with IgG glycosylation, we quantitated N-linked IgG glycans using two approaches. After isolating IgG from human plasma, we performed 77 quantitative measurements of N-glycosylation using ultra-performance liquid chromatography (UPLC) in 2,247 individuals from four European discovery populations. In parallel, we measured IgG N-glycans using MALDI-TOF mass spectrometry (MS) in a replication cohort of 1,848 Europeans. Meta-analysis of genome-wide association study (GWAS) results identified 9 genome-wide significant loci (P<2.27×10−9) in the discovery analysis and two of the same loci (B4GALT1 and MGAT3) in the replication cohort. Four loci contained genes encoding glycosyltransferases (ST6GAL1, B4GALT1, FUT8, and MGAT3), while the remaining 5 contained genes that have not been previously implicated in protein glycosylation (IKZF1, IL6ST-ANKRD55, ABCF2-SMARCD3, SUV420H1, and SMARCB1-DERL3). However, most of them have been strongly associated with autoimmune and inflammatory conditions (e.g., systemic lupus erythematosus, rheumatoid arthritis, ulcerative colitis, Crohn's disease, diabetes type 1, multiple sclerosis, Graves' disease, celiac disease, nodular sclerosis) and/or haematological cancers (acute lymphoblastic leukaemia, Hodgkin lymphoma, and multiple myeloma). Follow-up functional experiments in haplodeficient Ikzf1 knock-out mice showed the same general pattern of changes in IgG glycosylation as identified in the meta-analysis. As IKZF1 was associated with multiple IgG N-glycan traits, we explored biomarker potential of affected N-glycans in 101 cases with SLE and 183 matched controls and demonstrated substantial discriminative power in a ROC-curve analysis (area under the curve = 0.842). Our study shows that it is possible to identify new loci that control glycosylation of a single plasma protein using GWAS. The results may also provide an explanation for the reported pleiotropy and antagonistic effects of loci involved in autoimmune diseases and haematological cancer.
| After analysing glycans attached to human immunoglobulin G in 4,095 individuals, we performed the first genome-wide association study (GWAS) of the glycome of an individual protein. Nine genetic loci were found to associate with glycans with genome-wide significance. Of these, four were enzymes that directly participate in IgG glycosylation, thus the observed associations were biologically founded. The remaining five genetic loci were not previously implicated in protein glycosylation, but the most of them have been reported to be relevant for autoimmune and inflammatory conditions and/or haematological cancers. A particularly interesting gene, IKZF1 was found to be associated with multiple IgG N-glycans. This gene has been implicated in numerous diseases, including systemic lupus erythematosus (SLE). We analysed N-glycans in 101 cases with SLE and 183 matched controls and demonstrated their substantial biomarker potential. Our study shows that it is possible to identify new loci that control glycosylation of a single plasma protein using GWAS. Our results may also provide an explanation for opposite effects of some genes in autoimmune diseases and haematological cancer.
| Glycosylation is a ubiquitous post-translational protein modification that modulates the structure and function of polypeptide components of glycoproteins [1], [2]. N-glycan structures are essential for multicellular life [3]. Mutations in genes involved in modification of glycan antennae are common and can lead to severe or fatal diseases [4]. Variation in protein glycosylation also has physiological significance, with immunoglobulin G (IgG) being a well-documented example. Each heavy chain of IgG carries a single covalently attached bi-antennary N-glycan at the highly conserved asparagine 297 residue in each of the CH2 domains of the Fc region of the molecule. The attached oligosaccharides are structurally important for the stability of the antibody and its effector functions [5]. In addition, some 15–20% of normal IgG molecules have complex bi-antennary oligosaccharides in the variable regions of light or heavy chains [6], [7]. 36 different glycans (Figure 1) can be attached to the conserved Asn297 of the IgG heavy chain [8], [9], leading to hundreds of different IgG isomers that can be generated from this single glycosylation site.
Glycosylation of IgG has important regulatory functions. The absence of galactose residues in association with rheumatoid arthritis was reported nearly 30 years ago [10]. The addition of sialic acid dramatically changes the physiological role of IgGs, converting them from pro-inflammatory to anti-inflammatory agents [11], [12]. Addition of fucose to the glycan core interferes with the binding of IgG to FcγRIIIa and greatly diminishes its capacity for antibody dependent cell-mediated cytotoxicity (ADCC) [13], [14]. Structural analysis of the IgG-Fc/FcγRIIIa complex has demonstrated that specific glycans on FcγRIIIa are also essential for this effect of core-fucose [15] and that removal of core fucose from IgG glycans increases clinical efficacy of monoclonal antibodies, enhancing their therapeutic effect through ADCC mediated killing [16]–[18].
New high-throughput technologies, such as high/ultra performance liquid chromatography (HPLC/UPLC), MALDI-TOF mass spectrometry (MS) and capillary electrophoresis (CE), allow us to quantitate N-linked glycans from individual human plasma proteins. Recently, we performed the first population-based study to demonstrate physiological variation in IgG glycosylation in three European founder populations [19]. Using UPLC, we showed exceptionally high individual variability in glycosylation of a single protein - human IgG - and substantial heritability of the observed measurements [19]. In parallel, we quantitated IgG N-glycans in another European population (Leiden Longevity Study – LLS) by mass spectrometry. In this study, we combined those high-throughput glycomics measurements with high-throughput genomics to perform the first genome wide association (GWA) study of the human IgG N-glycome.
We separated a single protein (IgG) from human plasma and quantitated its N-linked glycans using two state-of-the-art technologies (UPLC and MALDI-TOF MS). Their comparative advantages in GWA studies were difficult to predict prior to the conducted analyses, so both were used - one in each available cohort. We performed 77 quantitative measurements of IgG N-glycosylation using ultra performance liquid chromatography (UPLC) in 2247 individuals from four European discovery populations (CROATIA-Vis, CROATIA-Korcula, ORCADES, NSPHS). In parallel, we measured IgG N-glycans using MALDI-TOF mass spectrometry (MS) in 1848 individuals from another European population (Leiden Longevity Study (LLS)). Descriptions of these population cohorts are found in Table S11. Aiming to identify genetic loci involved in IgG glycosylation, we performed a GWA study in both cohorts. Associations at 9 loci reached genome-wide significance (P<2.27×10−9) in the discovery meta-analysis and at two loci in the replication cohort. The two loci identified in the latter cohort were associated with the analogous glycan traits in the former cohort as detailed in the subsection “Replication of our findings”. Both UPLC and MS methods for quantitation of N-glycans were found to be amenable to GWA studies. Since our UPLC study gave a considerably greater yield of significant findings in comparison to MS study, the majority of our results section focuses on the findings from the discovery population cohort, which was studied using the UPLC method.
Among the nine loci that passed the genome-wide significance threshold, four contained genes encoding glycosyltransferases (ST6GAL1, B4GALT1, FUT8 and MGAT3), while the remaining five loci contained genes that have not been implicated in protein glycosylation previously (IKZF1, IL6ST-ANKRD55, ABCF2-SMARCD3, SUV420H1-CHKA and SMARCB1-DERL3). As a rule, the implicated genes were associated with several N-glycan traits. The explanation and notation of the 77 N-glycan measures is presented in Table S1. It comprises 23 directly measured quantitative IgG glycosylation traits (shown in Figure 1) and 54 derived traits. Descriptive statistics of these measures in the discovery cohorts are presented in Table S2. GWA analysis was performed in each of the populations separately and the results were combined in an inverse-variance weighted meta-analysis. Summary data for each gene region showing genome-wide association (p<27.2×10−9) or found to be strongly suggestive (2.27×10−9<p<5×10−8) are presented in Table 1. Summary data for all single-nucleotide polymorphisms (SNPs) and traits with suggestive associations (p<1×10−5) are presented in Table S3, with population-specific and pooled genomic control (GC) factors reported in Table S4.
The most statistically significant association was observed in a region on chromosome 3 containing the gene ST6GAL1 (Table 1, Figure S1A). ST6GAL1 codes for the enzyme sialyltransferase 6 which adds sialic acid to various glycoproteins including IgG glycans (Figure 2), and is therefore a highly biologically plausible candidate. In this region of about 70 kilobases (kb) we identified 37genome-wide significant SNPs associated with 14 different IgG glycosylation traits, generally reflecting sialylation of different glycan structures (Table 1). The strongest association was observed for the percentage of monosialylation of fucosylated digalactosylated structures in total IgG glycans (IGP29, see Figure 1 and Table S1 for notation), for which a SNP rs11710456 explained 17%, 16%, 18% and 3% of the trait variation for CROATIA-Vis, CROATIA-Korcula, ORCADES and NSPHS respectively (meta-analysis p = 6.12×10−75). NSPHS had a very small sample size in this analysis (N = 179) and may not provide an accurate portrayal of the variance explained in this particular population (estimated as 3%). Although the allele frequency is similar between all populations, in the forest plot (Figure S1A) although NSPHS does overlap with the other populations, the 95% CI is much larger. It is also possible that there are population-specific genetic and/or environmental differences in NSPHS that are affecting the amount of variance explained by this SNP. After analysis conditioning on the top SNP (rs11710456) in this region, the SNP rs7652995 still reached genome-wide significance (p = 4.15×10−13). After adjusting for this additional SNP, the association peak was completely removed. This suggests that there are several genetic factors underlying this association. Conditional analysis of all other significant and suggestive regions resulted in the complete removal of the association peak.
We also identified 28 SNPs showing genome-wide significant associations with 11 IgG glycosylation traits (2.70×10−11<p<4.73×10−8) at a locus on chromosome 9 spanning over 60 kb (Figure S1B). This region includes B4GALT1, which codes for the galactosyltransferase responsible for the addition of galactose to IgG glycans (Figure 2). The glycan traits showing genome-wide association included the percentage of FA2G2S1 in the total fraction (IGP17), the percentage of FA2G2 in the total and neutral fraction (IGP13, IGP53), the percentage of sialylation of fucosylated structures without bisecting GlcNAc (IGP24, IGP26), the percentage of digalactosylated structures in the total neutral fraction (IGP57) and, in the opposite direction, the percentage of bisecting GlcNAc in fucosylated sialylated structures (IGP36–IGP40).
A large (541 kb) region on chromosome 14 harbouring the FUT8 gene contained 167 SNPs showing significant associations with 12 IgG glycosylation traits reflecting fucosylation of IgG glycans (Figure S1C). FUT8 codes for fucosyltransferase 8, an enzyme responsible for the addition of fucose to IgG glycans (Figure 2). The strongest association (1.08×10−22<p<2.60×10−17) was observed with the percentage of A2 glycans in total and neutral fractions (IGP2, IGP42) and for derived traits related to the proportion of fucosylation (IGP58, IGP59 and IGP61; all in the opposite direction). In summary, SNPs at the FUT8 locus influence the proportion of fucosylated glycans, and, in the opposite direction, the percentages of A2, A2G1 and A2G2 glycans which are not fucosylated.
On chromosome 22, two loci were associated with IgG glycosylation. The first region, containing SYNGR1-TAB1-MGAT3-CACNA1I genes, spans over 233 kb. This region harboured 60 SNPs showing genome-wide significant association with 17 IgG glycosylation traits (Figure S1D). Association was strongest between SNP rs909674 and the incidence of bisecting GlcNAc in all fucosylated disialylated structures (IGP40, p = 9.66×10−25) and the related ratio IGP39 (p = 8.87×10−24). In summary, this locus contained variants influencing levels of fucosylated species and the ratio between fucosylated (especially disialylated) structures with and without bisecting GlcNAc (Figure 2). Since MGAT3 codes for the enzyme N-acetylglucosaminyltransferase III (beta-1,4-mannosyl-glycoprotein-4-beta-N-acetylglucosaminyltransferase), which is responsible for the addition of bisecting GlcNAc to IgG glycans, this gene is the most biologically plausible candidate.
Bioinformatic analysis of known and predicted protein-protein interactions using String 9.0 software (http://string-db.org/) showed that interactions between the clusters of FUT8-B4GALT1-MGAT3 genes and ST6GAL1-B4GALT1-MGAT3 genes had high confidence score: FUT8-B4GALT1 of 0.90; FUT8-MGAT3 of 0.95; ST6GAL1-B4GALT1 of 0.90; and ST6GAL1-MGAT3 of 0.73. The glycosyltranferase genes at the four GWAS loci - ST6GAL1, B4GALT1, FUT8, and MGAT3 – are responsible for adding sialic acid, galactose, fucose and bisecting GlcNAc to IgG glycans, thus demonstrating the proof of principle that a single protein glycosylation GWAS approach can identify biologically important glycan pathways and their networks. Interestingly, ST6GAL1 has been previously associated with Type 2 diabetes [20], MGAT3 with Crohn's disease [21], primary biliary cirrhosis [22] and cardiac arrest [23], and FUT8 with multiple sclerosis, blood glutamate levels [24] and conduct disorder [25] (Table 2). We have recently shown changes in plasma N-glycan profile between patients with attention-deficit hyperactivity disorder (ADHD), autism spectrum disorders and healthy controls, and identified loci influencing plasma N-glycome with pleiotropic effects on ADHD [26], [27].
In addition to four loci containing genes for enzymes known to be involved in IgG glycosylation, our study also found five unexpected associations showing genome-wide significance. In the second region on chromosome 22 we observed genome-wide significant associations of 10 SNPs with 20 IgG glycosylation traits. The region spans 49 kb and contains the genes SMARCB1-DERL3 (Figure S1E). The strongest associations (8.63×10−17<p<3.00×10−13) were observed between SNP rs2186369 and the percentage of FA2[6]BG1 in total and neutral fractions (IGP9, IGP49) and levels of fucosylated structures with bisecting GlcNAc (IGP66, IGP68, IGP70, IGP71 in the same direction and IGP72 in the opposite direction). Thus, the SMARCB1-DERL3 locus appears to specifically influence levels of fucosylated monogalactosylated structures with bisecting GlcNAc (Figure 2). DERL3 is a promising functional candidate, because it encodes a functional component of endoplasmic reticulum (ER)-associated degradation for misfolded luminal glycoproteins [28]. However, SMARCB1 is also known to be important in antiviral activity, inhibition of tumour formation, neurodevelopment, cell proliferation and differentiation [29]. The region has also been implicated in the regulation of γ-glutamyl-transferase (GGT) [30] (Table 2).
A locus on chromosome 7 spanning 26kb contained 11 SNPs showing genome-wide significant associations with 13 IgG glycosylation traits (Figure S1F). The strongest association (p = 1.87×10−13) was observed between SNP rs6421315 located in IKZF1 and the percentage of fucosylation of agalactosylated structures without bisecting GlcNAc (IGP63). Thus, SNPs at this locus influence the percentage of non-fucosylated agalactosylated glycans, the fucosylation ratio in agalactosylated glycans (in opposite directions for glycan species with and without bisecting GlcNAc), and the ratio of fucosylated structures with and without bisecting GlcNAc (Figure 2). The IKZF1 gene encodes the DNA-binding protein Ikaros, acting as a transcriptional regulator and associated with chromatin remodelling. It is considered to be the important regulator of lymphocyte differentiation and has been shown to influence effector pathways through control of class switch recombination [31], thus representing a promising functional candidate [32]. There is overwhelming evidence that IKZF1 variants are associated with childhood acute lymphoblastic leukaemia [33], [34] and several diseases with an autoimmune component: systemic lupus erythematosus (SLE) [35]–[37], type 1 diabetes [38], [39], Crohn's disease [40], systemic sclerosis [41], malaria [42] and erythrocyte mean corpuscular volume [43] (Table 2).
SNPs at several other loci also showed genome-wide significant association with a number of different IgG glycosylation traits (Figure S1G–S1P). Chromosome 5 SNP rs17348299, located in IL6ST-ANKRD55 was significantly associated (6.88×10−11<p<2.39×10−9) with six IgG glycosylation traits, including FA2 and FA2G2 in total and neutral fractions (IGP3, IGP13, IGP43, IGP53) and the percentage of agalactosylated and digalactosylated structures in total neutral IgG glycans (IGP55, IGP57) (Figure 2). The protein encoded by IL6ST is a signal transducer shared by many cytokines, including interleukin 6 (IL6), ciliary neurotrophic factor (CNTF), leukaemia inhibitory factor (LIF), and oncostatin M (OSM). Variants in IL6ST have been associated with rheumatoid arthritis and multiple myeloma, but also with components of metabolic syndrome [44]–[46].
The chromosome 7 SNP rs2072209 located in LAMB1 was strongly suggestively associated with the percentage of fucosylation of digalactosylated (with bisecting GlcNAc) structures (IGP69; p = 1.16×10−8) (Figure 2). LAMB1 (laminin beta 1) is a member of a family of extracellular matrix glycoproteins that are the major non-collagenous constituent of basement membranes. It is thought to mediate the attachment, migration and organization of cells into tissues during embryonic development by interacting with other extracellular matrix components. It has been associated with ulcerative colitis in several large-scale studies in European and Japanese populations, suggesting that changes in the integrity of the intestinal epithelial barrier may contribute to the pathogenesis of the disease [47]–[51] (Table 2).
Another particularly interesting finding was the suggestive association between rs404256 in the BACH2 gene on chromosome 6 and IGP7, defined through proportional contribution of FA2[6]G1 in all IgG glycans (p = 7.49×10−9). BACH2 is B-cell-specific transcription factor that can act as a suppressor or promoter; among many known functions, it has been shown to “orchestrate” transcriptional activation of B-cells, modify the cytotoxic effects of anticancer drugs and regulate IL-2 expression in umbilical cord blood CD4+ T cells [52]. BACH2 has been previously associated with a spectrum of diseases with autoimmune component: type 1 diabetes [53]–[56], Graves' disease [57], celiac disease [58], Crohn's disease [21] and multiple sclerosis [59] (Table 2).
The chromosome 11 SNP rs4930561 located in the SUV420H1-CHKA gene was associated with percentage of FA1 in neutral (IGP41; p = 8.88×10−10) and total (IGP1; p = 1.30×10−8) fractions of IgG glycans. SUV420H1 codes for histone-lysine N-methyltransferase which specifically trimethylates lysine 20 of histone H4 and could therefore affect activity of many different genes; it is thought to be involved in proviral silencing in somatic and germ line cells through epigenetic mechanisms [60]. CHKA has a key role in phospholipid biosynthesis and may contribute to tumour cell growth. We recently reported a number of strong associations between lipidomics and glycomics traits in human plasma [61]. Thus, an enzyme involved in phospholipid synthesis is also a possible candidate because the lipid environment is known to affect glycosyltransferases activity [61].
Three further loci were identified as strongly suggestive through GWAS and deserve attention for their possible pleiotropic effects. SNP rs9296009 in PRRT1 (proline-rich transmembrane protein 1) was associated with IGP23 (p = 3.79×10−08) while variants in PRRT1 previously showed associations with nodular sclerosis and Hodgkin lymphoma [62]. Moreover, rs1049110 in HLA-DQA2-HLA-DQB2 was associated with IGP2 and IGP42 (p = 1.64×10−08 and 4.44×10−08, respectively). This SNP is in nearly complete linkage disequilibrium with two other SNPs in this region that have previously been associated with SLE and hepatitis B [63] (Table 2). Another SNP in this region has been linked with narcolepsy [64]. Finally, rs7224668 in SLC38A10, a putative sodium-dependent amino acid/proton antiporter, showed significant association with IGP31 (p = 3.33×10−08). Although the function of this gene is not understood, it has been associated with autism and longevity [65], [66].
The remaining three signals implicated ABCF2-SMARCD3 region (rs1122979 was associated with IGP 2, 5, 42, 45, with p-value ranging between 2.10×10−10<p<1.89×10−9), RECK (rs4878639 was suggestively associated with IGP17; p = 3.51×10−8) and PEX5 (rs12828421 suggestively associated with IGP41; p = 4.48×10−8). The function of ABCF2 (ATP-binding cassette, sub-family F, member 2) is not well understood. SMARCD3 stimulates nuclear receptor mediated transcription; it belongs to the neural progenitors-specific chromatin remodelling complex (npBAF complex) and the neuron-specific chromatin-remodelling complex (nBAF complex). RECK is known to be a strong suppressor of tumour invasion and metastasis, regulating metalloproteinases which are involved in cancer progression. PEX 5 binds to the C-terminal PTS1-type tripeptide peroxisomal targeting signal and plays an essential role in peroxisomal protein import (www.genecards.org).
The parallel effort in the outbred Leiden Longevity Study (LLS) was based on a different N-glycan quantitation method (MS). While UPLC groups glycans according to structural similarities, MS groups them by mass. Furthermore, MS analysis focused on Fc glycans while UPLC measures both Fc and Fab glycans, thus traits measured by the two methods could not have been directly compared. Glycosylation patterns of IgG1 and IgG2 were investigated by analysis of tryptic glycopeptides, with six glycoforms per IgG subclass measured. The intensities of all glycoforms were related to the monogalactosylated, core-fucosylated biantennary species, providing five relative intensities registered per IgG subclass (Tables S5 and S6). The analysis identified two loci as genome-wide significant - implicating MGAT3 (p = 1.6×10−10 for G1FN, analogous to UPLC IGP9; p = 3.12×10−8 for G0FN, analogous to UPLC IGP5), and B4GALT1 (p = 5.4×10−8 for G2F, analogous to UPLC IGP13) confirming GWAS signals in the discovery meta-analysis.
We then sought a separate independent replication of the other 14 genome-wide significant and strongly suggestive signals identified in the discovery analysis, which was performed in the LLS cohort, appreciating that the quantitated N-glycan traits do not exactly match between the two cohorts. SNPs were chosen for replication based on initial meta-analysis results of genotype data prior to imputed analysis. All five traits measured in LLS cohort were tested for association with all the selected SNPs (Table S6). We were able to reproduce association to ST6GAL1 (p = 8.1×10−7 for G2F, substrate for sialyltransferase) and SMARCB1-DERL3 (p = 1.6×10−7 for G1N, analogous to UPLC IGP9). Weaker, though nominally significant associations were confirmed at IKZF1 (p = 2.3×10−3 for G1N), SLC38A10 (p = 4.8×10−3 for G2N), IL6ST-ANKRD55 (p = 1.3×10−2 for G0N) and ABCF2-SMARCD3 (p = 2.7×10−2 for G2N). The fact that we did not replicate associations at the other 8 loci was not unexpected, because those 8 loci showed association with UPLC-measured N-glycan traits that do not compare to any of the traits measured by MS (see Table S5 for comparison of MS and UPLC traits).
IKZF1 is considered to be the important regulator governing differentiation of T cells into CD4+ and CD8+ T cells [67]. Since glycan traits associated with IKZF1 were related to the presence and absence of core-fucose and bisecting GlcNAc, we analysed the promoter region of MGAT3 (codes for enzyme that adds bisecting GlcNAc to IgG glycans) in silico and identified two binding sites for IKZF1 that were conserved between humans and mice, while recognition sites for IKZF1 were not found in the promoter region of FUT8 (which codes for an enzyme that adds core-fucose to IgG glycans). Since the promoter regions of MGAT3 were conserved between humans and mice, we used Ikzf1 knockout mice [68] as a model to study the effects of IKZF1 deficiency on IgG glycosylation. IgG was isolated from the plasma of 5 heterozygous knockout mice and 5 wild-type controls. The summary of the results of IgG glycosylation analysis is presented in Table 3, while complete results are presented in Table S7. We observed a number of alterations in glycome composition that were all consistent with the role of IKZF1 in the down-regulation of fucosylation and up-regulation of the addition of bisecting GlcNAc to IgG glycans; 12 out of 77 IgG N-glycans measures showed statistically significant difference (p<0.05) between wild type and heterozygous Ikzf1 knock-outs, where 5 mice from each group were compared (Table 3). The empirical version of Hotelling's test demonstrated global significance (p = 0.03) of difference between distributions of IgG glycome between wild type and Ikzf1 knock-out mice, where 5 mice from each group were compared. While the tests for differences between individual glycome measurements did not reach strict statistical significance after conservative Bonferroni correction (p = 0.05/77 = 0.0006), we observed that 12 out of 77 (15%) IgG N-glycans measures showed nominally significant difference (p<0.05) between wild type and heterozygous Ikzf1 knock-outs (Table 3). Significant results from the global difference test ensure that difference between the two groups does exist, and it is most likely due to the difference between (at least some of) the measurements which demonstrated nominal significance. Observed alterations in glycome composition were all consistent with the role of IKZF1 in the down-regulation of fucosylation and up-regulation of the addition of bisecting GlcNAc to IgG glycans.
Given that IKZF1 has been convincingly associated with SLE in previous studies [35]–[37], and that functional studies in heterozygous knock-out mice in our study showed clear differences in profiles of several IgG N-glycan traits, we explored an intriguing hypothesis: whether the same IgG N-glycan traits that were significantly affected in Ikzf1 knock-out mice could be demonstrated to differ between human SLE cases and controls. If this were true, then pleiotropy between the effects of IKZF1 on SLE and on IgG N-glycans in human plasma, revealed by independent GWA studies, would lead to a discovery of a novel class of biomarkers of SLE – IgG N-glycans – which could possibly extend their usefulness in prediction of other autoimmune disorders, cancer and neuropsychiatric disorders, through the same mechanism.
To test this hypothesis, we measured IgG N-glycans in 101 SLE cases and 183 matched controls (typically two controls per case), recruited in Trinidad (see materials and methods for further details). Table 4 shows the results of the measurements: for 10 of 12 N-glycan traits chosen on a basis of the experiments in mice (Table 3). The entire dataset for all glycans can be found in Table S8. There was a statistically significant difference (p<0.05) between SLE cases and controls, which was generally not the case with other groups of N-glycans (data not shown). Moreover, the significance of the difference was striking in some cases, e.g. p<10−14 for IGP48, p<10−13 for IGP8, and p<10−6 for IGP64. Furthermore, the differences in the direction of effect in mice were strikingly preserved in humans (Table 4). The most significant differences observed across all 77 IgG N-glycans measurements between SLE cases and controls (Table 4) were overlapping well with the 12 N-glycan groups that were significantly changed in functional experiments in Ikzf1 knock-out mice.
To strengthen our findings and control for possible bias, we repeated the analysis excluding all the cases on corticosteroid treatment at the time of interview (77/101) and subsequently all the cases that were not on corticosteroid treatment at the time of interview (24/101). Although the power of the analysis decreased due to reduced number of cases, the results did not change and they remained highly statistically significant. We also hypothesized that the observed glycan changes may not be specific to SLE, but may be caused by corticosteroid treatment, or secondary to any inflammatory process. For this reason, and in SLE cases only, we investigated whether corticosteroid treatments and/or CRP measurements, were associated with IgG N-glycan traits. Analysis for CRP was repeated with CRP treated as a binary variable (with cut-off value at 10 mg/L). In all these analyses, the initial results held and were not changed: the association of IgG N-glycans and SLE remained striking, while the association with corticosteroid treatment and CRP was not (Table S9). Finally, we also repeated the analysis adjusting for percent African admixture, as it has been reported that SLE in Afro-Caribbean population is associated with African admixture [69]. However, this adjustment only had a minor and non-systemic effect on the previous results, and the reported observations remained.
We then validated biomarker potential of IGP48, the IgG N-glycan trait most significantly associated with SLE status, in prediction of SLE in 101cases and 183 matched controls. We used the PredictABEL package for R (see materials and methods) [70]. As shown in Figure 3, age, sex and African admixture did not have any predictive power for this disease, but addition of IGP48 substantially increased sensitivity and specificity of prediction, with area under receiver-operator curve (AUC) increasing from 0.515 (95% confidence interval (CI): 0.441–0.590) to 0.842 (0.791–0.893). It is likely that further additions of other IgG N-glycans could provide even more accurate predictions. To cross-validate this result, we split our dataset with SLE cases and controls into a “training set” (2/3; 67 cases and 122 controls) and “test set” (1/3; 34 cases and 61 controls). Area under ROC-curve (AUC) was calculated for the test dataset. The whole process was repeated 1000 times, to allow computation of the mean AUC (and 95% CI) in the test datasets. Mean AUC was virtually unchanged compared to AUC obtained when using the complete dataset and no training, which suggests that the predictive power of IGP48 on SLE is very robust.
This study clearly demonstrates that the recent developments in high-throughput glycomics and genomics now allow identification of genetic loci that control N-glycosylation of a single plasma protein using a GWAS approach. This progress should allow many similar follow-up studies of genetic regulation of N-glycosylation of other important plasma proteins, thus bringing unprecedented insights into the role of protein glycosylation in systems biology. As a prelude to this discovery, we recently reported the results of the first GWA study of the overall human plasma N-glycome using the HPLC method. Although the study was of a comparable sample size (N∼2000), it only identified genome-wide associations with two glycosyltransferases and one transcription factor (HNF1a) [71]. We believe that the power of our initial study was reduced because N-glycans in human plasma originate from different glycoproteins where they have different functions and undergo protein-specific, or tissue-specific glycosylation. In this study the largest percentage of variance explained by a single association was 16–18% where as in the N-glycan study this was 1–6%. Furthermore, concentrations of individual glycoproteins in plasma vary in many physiological processes, introducing substantial “noise” to the quantitation of the whole-plasma N-glycome.
In this study we avoided both problems by isolating a single protein from plasma (IgG), which is produced by a single cell type (B lymphocytes), thus effectively excluding differential regulation of gene expression in different tissues, and the “noise” introduced by variation in plasma IgG concentration and by N-glycans on other plasma proteins. The only remaining “noise” in our system was the incomplete separation of some glycan structures (which co-eluted from the UPLC column) and the presence of Fab glycans on a subset of IgG molecules, but for the majority of glycan structures this “noise” was well below 10% [19]. We expected that the specificity of our phenotype and precision of the measurement provided by novel UPLC and MS methods should substantially increase the power of the study to detect genome-wide associations. Prior to analysis we could not predict which quantitation method would work better in GWA study design (UPLC vs. MS), so we used them both, each in one separate cohort of comparable sample size (N∼2000).
The UPLC method yielded many more, and much stronger, genome-wide association signals in comparison to our previous study of the total plasma N-glycome in virtually same sample set of examinees [27], [71]. Sixteen loci were identified in association with glycan traits with p-values<5×10−8 and nine reached the strict genome wide threshold of 2.27×10−9. The parallel study in the LLS cohort using MS quantitation has independently identified two of those 16 loci, showing genome-wide association with N-glycan traits. MS quantitation also allowed us to replicate 6 further loci identified in the discovery analysis, using comparable N-glycan traits measured by the two methods. However, in this follow-up analysis we were unable to replicate associations for the remaining 8 loci. This was not unexpected, because those glycosylation traits correspond to different fucosylated glycans; since fucosylation was not quantified by MS, the association between glycans measured by MS and those regions should not be expected.
Among the nine loci that reached genome-wide statistical significance, four involved genes encoding glycosyltransferases known to glycosylate IgG (ST6GALI, B4GALT1, FUT8, MGAT3,). The enzyme beta1,4-galactosyltransferase 1 is responsible for the addition of galactose to IgG glycans. Interestingly, variants in B4GALT1 gene did not affect the main measures of IgG galactosylation, but rather differences in sialylation and the percentage of bisecting GlcNAc. These associations are still biologically plausible, because galactosylation is a prerequisite for sialylation, and enzymes which add galactose and bisecting GlcNAc compete for the same substrate [72]. A potential candidate for B4GALT1 regulator is IL6ST, which codes for interleukin 6 signal transducer, because it showed stronger associations with the main measures of IgG galactosylation than B4GALT1 itself. Molecular mechanisms behind this association remain elusive, but early work on IL6 (then called PHGF) suggested that it may be relevant for glycosylation pathways in B lymphocytes [73].
Core-fucosylation of IgG has been intensively studied due to its role in antibody-dependent cell-mediated cytotoxicity (ADCC). This mechanism of killing is considered to be one of the major mechanisms of antibody-based therapeutics against tumours. Core-fucose is critically important in this process, because IgGs without core fucose on the Fc glycan have been found to have ADCC activity enhanced by up to 100-fold [74]. Alpha-(1,6)-fucosyltransferase (fucosyltransferase 8) catalyses the transfer of fucose from GDP-fucose to N-linked type complex glycopeptides, and is encoded by the FUT8 gene. We found that SNPs located near this gene influenced overall levels of fucosylation. The directly measured IgG glycome traits most strongly associated with SNPs in the FUT8 region consisted of A2, and, less strongly, A2G1 and A2G2. These associations are biologically plausible as these glycans serve as substrates for fucosyltransferase 8. Interestingly, SNPs located near the IKZF1 gene influenced fucosylation of a specific subset of glycans, especially those without bisecting GlcNAc, and were also related to the ratio of fucosylated structures with and without bisecting GlcNAc. This suggests the IKZF1 gene encoding Ikaros as a potential indirect regulator of fucosylation in B-lymphocytes by promoting the addition of bisecting GlcNAc, which then inhibits fucosylation. The analysis of IgG glycosylation in Ikzf1 haplodeficient mice confirmed the postulated role of Ikaros in the regulation of IgG glycosylation (Table 3). The effect of Ikzf1 haplodeficiency on IgG glycans manifested mainly in the decrease in bisecting GlcNAc on different glycan structures. The increase in fucose was observed only in a subset of structures, but since very high level of fucosylation was present in the wild type mouse (up to 99.8%), a further increase could not have been demonstrated.
Nearly all genome-wide significant loci in our study have already been clearly demonstrated to be associated with autoimmune diseases, haematologic cancers, and some of them also with chronic inflammation and/or neuropsychiatric disorders. Although the literature on those associations is extensive, we tried to highlight only those associations that were identified using genome-wide association studies in datasets independent from our study. We gave prominence to associations arising from GWA studies because they are typically replicable; GWA studies have sufficient power to detect true associations, and require stringent statistical testing and replication to avoid false positive results. They have been reviewed and summarized in Table 2. The table implies abundant pleiotropy between loci that control N-glycosylation (in this case, of IgG protein) and loci that have been implicated in many human diseases. Autoimmune diseases (including SLE, RA, UC and over 80 others) are generally thought to be triggered by aggressive responses of the adaptive immune system to self antigens, resulting in tissue damage and pathological sequelae [38]. Among other mechanisms, IgG autoantibodies are responsible for the chronic inflammation and destruction of healthy tissues by cross-linking Fc receptors on innate immune effector cells [75]. Class and glycosylation of IgG are important for pathogenicity of autoantibodies in autoimmune diseases (reviewed in [76]). Removal of IgG glycans leads to the loss of the proinflammatory activity, suggesting that in vivo modulation of antibody glycosylation might be a strategy to interfere with autoimmune processes [75]. Indeed, the removal of IgG glycans by injections of EndoS in vivo interfered with autoantibody-mediated proinflammatory processes in a variety of autoimmune models [75].
Results from our study suggest that IgG N-glycome composition is regulated through a complex interplay between loci affecting an overlapping spectrum of glycome measurements, and through interaction of genes directly involved in glycosylation and those that presumably have a “higher-level” regulatory function. SNPs at several different loci in this GWA study showed genome-wide significant associations with the same or similar IgG glycosylation traits. For example, SNPs at loci on chromosomes 9 (B4GALT1 region) and 3 (ST6GAL1 region) both influenced the percentage of sialylation of galactosylated fucosylated structures (without bisecting GlcNAc) in the same direction. SNPs at these loci also influenced the ratio of fucosylated monosialylated structures (with and without bisecting GlcNAc) in the opposite direction. SNPs at the locus on chromosome 9 (B4GALT1), and two loci on chromosome 22 (MGAT3 and SMARCB1-DERL3 region) simultaneously influenced the ratio of fucosylated disialylated structures with and without bisecting GlcNAc. SNPs at loci on chromosome 7 (IKZF1 region) and 14 (FUT8 region) influenced an overlapping range of traits: percentage of A2 and A2G1 glycans, and, in the opposite direction, the percentage of fucosylation of agalactosylated structures.
Finally, this study demonstrated that findings from “hypothesis-free” GWA studies, when targeted at a well defined biological phenotype of unknown relevance to human health and disease (such as N-glycans of a single plasma protein), can implicate genomic loci that were not thought to influence protein glycosylation. Moreover, unexpected pleiotropy of the implicated loci that linked them to diseases has changed this study from “hypothesis-free” to “hypothesis-driven” [77], and led us to explore biomarker potential of a very specific IgG N-glycan trait in prediction of a specific disease (SLE) with considerable success. To our knowledge, this is one of the first convincing demonstrations that GWA studies can lead to biomarker discovery for human disease. This study offers many additional opportunities to validate the role of further N-glycan biomarkers for other diseases implicated through pleiotropy.
A new understanding of the genetic regulation of IgG N-glycan synthesis is emerging from this study. Enzymes directly responsible for the addition of galactose, fucose and bisecting GlcNAc may not have primary responsibility for the final IgG N-glycan structures. For all three processes, genes that are not directly involved in glycosylation showed the most significant associations: IL6ST-ANKRD55 for galactosylation; IKZF1 for fucosylation; and SMARCB1-DERL3 for the addition of bisecting GlcNAc. The suggested higher-level regulation is also apparent from the differences in IgG Fab and Fc glycosylation, observed in human IgG [78], [79] and different myeloma cell lines [80], and further supported by recent observation that various external factors exhibit specific effects on glycosylation of IgG produced in cultured B lymphocytes [81].
Moreover, this study showed that it is possible to identify loci that control glycosylation of a single plasma protein using a GWAS approach, and to develop a novel class of disease biomarkers. This should lead to large advances in understanding of the role of protein glycosylation in the future. This study identified 16 genetic loci that are likely to be part of a much larger genetic network that regulates the complex process of IgG N-glycosylation and several further loci that show suggestive association with glycan traits and merit further study. Genetic variants in several of these genes were previously associated with a number of inflammatory, neoplastic and neuropsychiatric diseases across ethnically diverse populations, all of which could benefit from earlier and more accurate diagnosis based on molecular biomarkers. Variations in individual SNPs have relatively small effects, but when several polymorphisms are combined in a complex pathway like N-glycosylation, the final product of the pathway - in this case IgG N-glycan - can be significantly different, with consequences for IgG function and possibly also disease susceptibility. Our results may also provide an explanation for the reported pleiotropy and antagonistic genetic effects of loci involved in autoimmune diseases and hematologic cancers [39], [77].
All research in this study that involved human participants has been approved by the appropriate ethics committees: the Ethics Committee of the University of Split Medical School for all Croatian examinees from Vis and Korcula islands; the Local Research Ethics Committees in Orkney and Aberdeen for the Orkney Complex Disease Study (ORCADES); the University of Uppsala (Dnr 2005:325) for all examinees from Northern Sweden; the Leiden University Medical Centre Ethical Committee for all participants in the Leiden Longevity Study (LLS); and the Ethics Committee of the London School of Hygiene and Tropical Medicine for all SLE cases and controls from Trinidad. All ethics approvals were given in compliance with the Declaration of Helsinki (World Medical Association, 2000). All human subjects included in this study have signed appropriate informed consent.
All population studies recruited adult individuals within a community irrespective of any specific phenotype. Fasting blood samples were collected, biochemical and physiological measurements taken and questionnaire data for medical history as well as lifestyle and environmental exposures were collected following similar protocols. Basic cohort descriptives are included in Table S11.
The CROATIA-Vis study includes 1008 Croatians, aged 18–93 years, who were recruited from the villages of Vis and Komiža on the Dalmatian island of Vis during 2003 and 2004 within a larger genetic epidemiology program [82]. The CROATIA-Korcula study includes 969 Croatians between the ages of 18 and 98 [83]. The field work was performed in 2007 and 2008 in the eastern part of the island, targeting healthy volunteers from the town of Korčula and the villages of Lumbarda, Žrnovo and Račišće.
The Orkney Complex Disease Study (ORCADES) was performed in the Scottish archipelago of Orkney and collected data between 2005 and 2011 [84]. Data for 889 participants aged 18 to 100 years from a subgroup of ten islands, were used for this analysis.
The Northern Swedish Population Health Study (NSPHS) is a family-based population study including a comprehensive health investigation and collection of data on family structure, lifestyle, diet, medical history and samples for laboratory analyses from peoples living in the north of Sweden [84]. Complete data were available from 179 participants aged 14 to 91 years.
DNA samples were genotyped according to the manufacturer's instructions on Illumina Infinium SNP bead microarrays (HumanHap300v1 for CROATIA-Vis, HumanHap300v2 for ORCADES and NSPHS and HumanCNV370v1 for CROATIA-Korcula). Genotypes were determined using Illumina BeadStudio software. Genotyping was successfully completed on 991 individuals from CROATIA-Vis, 953 from CROATIA-Korcula, 889 from ORCADES and 700 from NSPHS, providing a platform for genome-wide association study of multiple quantitative traits in these founder populations.
The Leiden Longevity Study (LLS) has been described in detail previously [85]. It is a family based study and consists of 1671 offspring of 421 nonagenarian sibling pairs of Dutch descent, and their 744 partners. 1848 individuals with available genotypic and IgG measurements data were included in the current analysis. Within the Leiden Longevity Study 1345 individuals were genotyped using Illumina660 W (Rotterdam, Netherlands) and 503 individuals were genotyped using Illumina OmniExpress (Estonian Biocentre, Genotyping Core Facility, Estonia).
In the discovery population cohorts (CROATIA-Vis, CROATIA-Korcula, ORCADES, and NSPHS), the IgG was isolated using protein G plates and its glycans analysed by UPLC in 2247 individuals, as reported previously [19]. Briefly, IgG glycans were labelled with 2-AB fluorescent dye and separated by hydrophilic interaction ultra-performance liquid chromatography (UPLC). Glycans were separated into 24 chromatographic peaks and quantified as relative contributions of individual peaks to the total IgG glycome. The majority of peaks contained individual glycan structures, while some contained more structures. Relative intensities of each glycan structure in each UPLC peak were determined by mass spectrometry as reported previously [19]. On the basis of these 24 directly measured “glycan traits”, additional 54 “derived traits” were calculated. These include the percentage of galactosylation, fucosylation, sialylation, etc. described in the Table S1. When UPLC peaks containing multiple traits were used to calculate derived traits, only glycans with major contribution to fluorescence intensity were used.
In the replication population cohort (Leiden Longevity Study), the IgG was isolated from plasma samples of 1848 participants. Glycosylation patterns of IgG1 and IgG2 were investigated by analysis of tryptic glycopeptides using MALDI-TOF MS. Six glycoforms per IgG subclass were determined by MALDI-TOFMS. Since the intensities of all glycoforms were related to the monogalactosylated, core-fucosylated biantennary species (glycoform B), five relative intensities were registered per IgG subclass [86].
Genotyping quality control was performed using the same procedures for all four discovery populations (CROATIA-Vis, CROATIA-Korcula, ORCADES, and NSPHS). Individuals with a call rate less than 97% were removed as well as SNPs with a call rate less than 98% (95% for CROATIA-Vis), minor allele frequency less than 0.02 or Hardy-Weinberg equilibrium p-value less than 1×10−10. 924 individuals passed all quality control thresholds from CROATIA-Vis, 898 from CROATIA-Korcula, 889 from ORCADES and 656 from NSPHS.
Extreme outliers (those with values more than 3 times the interquartile distances away from either the 75th or the 25th percentile values) were removed for each glycan measure to account for errors in quantification and to remove individuals not representative of normal variation within the population. After phenotype quality control the number of individuals with complete phenotype and covariate information for the meta-analysis was 2247, consisting of 906 men and 1341 women (802 from CROATIA-Vis, 851 from CROATIA-Korcula, 415 from ORCADES, 179 from NSPHS).
In Leiden Longevity Study, GenomeStudio was used for genotyping calling algorithm. Sample call rate was >95%, and SNP exclusions criteria were Hardy-Weinberg equilibrium p value<10−4, SNP call rate<95%, and minor allele frequency <1%. The number of the overlapping SNPs that passed quality controls in both samples was 296,619.
To combine the data from the different array sets and to increase the overall coverage of the genome to up to 2.5 million SNPs, we imputed autosomal SNPs reported in the Haplotype Mapping Project (release #22, http://hapmap.ncbi.nlm.nih.gov) CEU sample. Based on the SNPs that were genotyped in all arrays and passed quality control, the imputation programmes MACH (http://www.sph.umich.edu/csg/abecasis/MACH/) or IMPUTE2 (http://mathgen.stats.ox.ac.uk/impute/impute_v2.html) were used to obtain ca. 2.5 million SNPs for further analysis.
For replication of genome-wide significant hits identified in the discovery meta-analysis, all SNPs listed in were used and looked up in LLS. The only exception was rs11621121, which had low imputation accuracy and did not pass quality control criteria. For this SNP, a set of 11 proxy SNPs from HapMap r. 22 (all with R2>0.85) was studied. All studied SNPs had imputation quality of 0.3 or greater.
In the discovery populations, genome-wide association analysis was firstly performed for each population and then combined using an inverse-variance weighted meta-analysis for all traits. Each trait was adjusted for sex, age and the first 3 principal components obtained from the population-specific identity-by-state (IBS) derived distances matrix. The residuals were transformed to ensure their normal distribution using quantile normalisation. Sex-specific analyses were adjusted for age and principal components only. The residuals expressed as z-scores were used for association analysis. The “mmscore” function of ProbABEL [87] was used for the association test under an additive model. This score test for family based association takes into account relationship structure and allowed unbiased estimations of SNP allelic effect when relatedness is present between examinees. The relationship matrix used in this analysis was generated by the “ibs” function of GenABEL (using weight = “freq” option), which uses genomic data to estimate the realized pair-wise kinship coefficient. All lambda values for the population-specific analyses were below 1.05 (Table S4), showing that this method efficiently accounts for family structure.
Inverse-variance weighted meta-analysis was performed using the MetABEL package (http://www.genabel.org) for R. SNPs with poor imputation quality (R2<0.3) were excluded prior to meta-analysis. Principal component analysis was performed using R to determine the number of independent traits used for these analyses (Table S10). 21 principal components explained 99% of the variance so an association was considered statistically significant at the genome-wide level if the p-value for an individual SNP was less than 2.27×10−9 (5×10−8/22 traits) [88]. SNPs were considered strongly suggestive with p-values between 5×10−8 and 2.27×10−9. Regions of association were visualized using the web-based software LocusZoom [89] to display the linkage disequilibrium (LD) of the region based on hg18/1000 Genomes June 1010 CEU data. The effect of the most significant SNP in each gene region expressed as percentage of the variance explained was calculated for each glycan trait adjusted for sex, age and first 3 principal components in each cohort individually using the “polygenic” function of the GenABEL package for R. Conditional analysis was undertaken for all significant and suggestive regions. GWAS was performed as described above with the additional adjustment for the dosage of the top SNP in the region for only the chromosome containing the association. Subsequent meta-analysis was performed as described previously and the results visualised using LocusZoom to ensure that the association peak have been removed.
In LLS, all IgG measurements were log-transformed. The score statistic for testing for an additive effect of a diallelic locus on quantitative phenotype was used. To account for relatedness in offspring data we used the kinship coefficients matrix when computing the variance of the score statistic. Imputation was dealt with by accounting for loss of information due to genotype uncertainty [90]. For the association analysis of the GWAS data, we applied the score test for the quantitative trait correcting for sex and age using an executable C++ program QTassoc (http://www.lumc.nl/uh, under GWAS Software). For further details we refer to supplementary online information.
The Ikzf1+/− mice harbouring the Neo-PAX5-IRES-GFP knock in allele were obtained from Meinrad Busslinger (IMP, Vienna) and backcrossed to C57BL/6 mice. Both wild-type and Ikzf1Neo+/− animals at the age of about 8 months were subjected to retro-orbital puncture to collect blood in the presence of EDTA. Samples were centrifuged for 10 minutes at room temperature and plasma was harvested. IgG was isolated and subjected to glycan analyses.
Statistical significance of the difference in distributions of IgG glycome between wild type and the Ikzf1+/− mice was assessed using empirical version of the Hotelling's test. In brief, the empirical distribution of the Hotelling's T2 statistics was worked out by permuting the group status of the animals at random without replacement 10,000 times. This empirical distribution was then contrasted with the original value of T2, with the proportion of empirically observed T2 values greater than or equal to the original T2 regarded as the empirical p-value.
A total of 101 SLE cases and 183 controls from Trinidad were studied. The inclusion criteria for cases and controls in Trinidad were designed to restrict the sample to individuals without Indian or Chinese ancestry. Cases and controls were eligible to be included if they were resident in northern Trinidad (excluding the southern part of the island where Indians are in the majority) and they had Christian (rather than Hindu, Muslim or Chinese) first names. Identification of cases was carried out by contacting all physicians specializing in rheumatology, nephrology and dermatology at the two main public hospitals in northern Trinidad and asking for a list of all SLE patients from their out-patient clinics. At the main dermatology clinic a register of cases since 1992 was available. Furthermore, a systematic search of: (a) outpatient records at the two hospitals, (b) hospital laboratory test results positive for auto-antibodies (anti-nuclear or anti-double-stranded DNA antibody titre >1∶256) and (c) histological reports of skin biopsy examination consistent with SLE was performed. Lastly, SLE cases were also identified through the Lupus Society of Trinidad and Tobago (90% of those patients were also identified through one of the two main public hospitals). For each case, randomly chosen households in the same neighbourhood were sampled by the field team to obtain (where possible) two controls, matched with the case for sex and for 20-year age group. Cases and controls were interviewed at home or in the project office by using a custom made questionnaire.
The case definition of SLE was based on American Rheumatism Association (ARA) criteria [91], applied to medical records (available for more than 90% of cases), and to the medical history given by the patient. Informed consent for blood sampling and the use of the sample for genetic studies including estimation of admixture was obtained from each participant. Initial case ascertainment identified 264 possible cases of SLE. Of these, 72 (27%) were excluded either on the basis of their names or because their medical history did not meet ARA criteria for the diagnosis of SLE. Of the remaining 192 individuals, 54 had incomplete addresses or were not resident in northern Trinidad, four were too ill to be interviewed, eight were aged less than 18 years and two refused to participate. For 80% (99/124) of cases, two matched controls were obtained: the response rate from those invited to participate as controls was 70%. The total sample consisted of 124 cases and 219 controls aged over 20 years who completed the questionnaire. Blood samples were obtained from 122 cases and 219 controls and DNA was successfully extracted from 93% (317/341) of these. IgG glycans were successfully measured in 303 individuals. Age at sampling was not available for 17 individuals and 2 individuals were lost due to the ID mismatch.
To test predictive power of selected glycan trait, we fitted logistic regression models (including and excluding the glycan) and used predRisk function of PredictABEL package for R to evaluate the predictive ability.
|
10.1371/journal.ppat.1006966 | The impact of serotype-specific vaccination on phylodynamic parameters of Streptococcus pneumoniae and the pneumococcal pan-genome | In the United States, the introduction of the heptavalent pneumococcal conjugate vaccine (PCV) largely eliminated vaccine serotypes (VT); non-vaccine serotypes (NVT) subsequently increased in carriage and disease. Vaccination also disrupts the composition of the pneumococcal pangenome, which includes mobile genetic elements and polymorphic non-capsular antigens important for virulence, transmission, and pneumococcal ecology. Antigenic proteins are of interest for future vaccines; yet, little is known about how the they are affected by PCV use. To investigate the evolutionary impact of vaccination, we assessed recombination, evolution, and pathogen demographic history of 937 pneumococci collected from 1998–2012 among Navajo and White Mountain Apache Native American communities. We analyzed changes in the pneumococcal pangenome, focusing on metabolic loci and 19 polymorphic protein antigens. We found the impact of PCV on the pneumococcal population could be observed in reduced diversity, a smaller pangenome, and changing frequencies of accessory clusters of orthologous groups (COGs). Post-PCV7, diversity rebounded through clonal expansion of NVT lineages and inferred in-migration of two previously unobserved lineages. Accessory COGs frequencies trended toward pre-PCV7 values with increasing time since vaccine introduction. Contemporary frequencies of protein antigen variants are better predicted by pre-PCV7 values (1998–2000) than the preceding period (2006–2008), suggesting balancing selection may have acted in maintaining variant frequencies in this population. Overall, we present the largest genomic analysis of pneumococcal carriage in the United States to date, which includes a snapshot of a true vaccine-naïve community prior to the introduction of PCV7. These data improve our understanding of pneumococcal evolution and emphasize the need to consider pangenome composition when inferring the impact of vaccination and developing future protein-based pneumococcal vaccines.
| Pneumococcal disease caused by the bacteria Streptococcus pneumoniae remains a significant cause of morbidity and mortality despite the existence of an effective vaccine. This is because the vaccines only target a small proportion of the total pneumococcal population. Introduction of vaccine in the United States removed vaccine serotypes leaving an open niche that was rapidly filled by non-vaccine serotypes. Forecasting which serotypes, and more generally which pneumococcal lineages, will increase in frequency in carriage and disease is an active area of research with significant public health importance. Here, we investigate the evolutionary impact of vaccination on the pneumococcal population using genomic data from a collection of 937 pneumococcal isolates collected from 1998–2012 among Native American communities. We find the impact of vaccine on the pneumococcal population could be observed in reduced diversity and changing frequencies of genes. Diversity subsequently rebounded through expansion and in-migration of non-vaccine lineages. Further, frequencies of genes coding for protein antigens important to host-pathogen interaction were initially disrupted but later returned to pre-vaccine values, suggesting selection may have acted in maintaining frequencies. These data improve our understanding of pneumococcal evolution and emphasize the need to consider genome composition when inferring the impact of vaccination.
| Pneumococcal conjugate vaccines (PCV) target capsular serotype-specific polysaccharides of the respiratory pathogen Streptococcus pneumoniae, which causes substantial morbidity and mortality [1,2]. Since the heptavalent PCV and 13-valent PCV were introduced in the United States (US) in 2000 and 2010, respectively, their effectiveness in reducing pneumococcal carriage and invasive disease has been well documented [3–6]. In communities where PCV has been introduced, the prevalence of vaccine serotypes (VT) in carriage and invasive disease consistently decreases, resulting in an overall reduction in pneumococcal disease. However, in a process called “serotype replacement,” non-vaccine serotypes (NVT) subsequently increase in carriage after vaccine introduction, leading to slight increases in NVT-associated disease in almost all populations where the vaccine is introduced [7,8]. Because polysaccharide serotypes change rarely during pneumococcal evolution, common pneumococcal lineages typically contain only one or a few serotypes. As a result, PCV implementation removes lineages containing only VT from the population, while lineages including both VT and NVT experience genetic bottlenecks [9–11].
Forecasting which serotypes, and more generally which pneumococcal lineages, will increase in frequency in carriage and disease is an active area of research with significant public health importance. For S. pneumoniae, the most commonly used vaccines globally target a fraction of the more than 93 recognized capsular serotypes [12]. The bacteria’s capsule (CPS) is the most important determinant of virulence and the strongest predictor of prevalence [13], as well as the target of PCVs; thus, changes in CPS serotype frequency have been the focus of many analyses of vaccine effect. However, selection acts on genes outside the operon determining CPS serotype. Whole-genome sequencing data has enabled investigation of variation in multiple genomic loci and genome content among pneumococci, focused on loci involved in host immunity and niche adaption. We focus here on two categories of proteins. The first is antigens (hereafter, when we use the generic term antigen we refer to proteins that elicit an immune response, not to the polysaccharide capsule). Antigens such as pneumococcal surface proteins A and C (pspA and pspC) and pilus are of specific interest as possible targets for non-capsular polysaccharide based vaccines [14]. Together with other components of the pneumococcal genome, the capsule and non-capsular antigens comprise the overall antigenic profile of a pneumococcus [15–17]. Moreover, evolution among metabolic genes gives rise to distinct metabolic-profiles among pneumococcal lineages, which may be adapted for specific metabolic niches [18,19]. Thus, multiple loci may interface with the host, affecting the overall evolutionary success of a lineage at a population level.
Gene content varies tremendously among pneumococcal lineages [20,21]. The pneumococcal pangenome consists of “core genes” shared by ≥99% of strains and “accessory genes” present at frequencies ≤99%. Accessory genes may include polymorphic antigens, phage and plasmid-related chromosomal islands, and integrative and conjugative elements (ICE) harboring antimicrobial resistance genes. The latter are mobile genetic elements (MGE), which are often acquired through horizontal gene transfer (HGT) and may remain stable in pneumococcal lineages [21]. Variations in gene content among lineages of a bacterial species are associated with ecological niche specialization and are important for adaptation to changing environments, including selection by vaccine-induced and natural host immunity [21–23]. For the pneumococcus, MGE affect the bacteria’s ability to recombine (i.e., competence) [24], antimicrobial susceptibility [25], and carriage duration [26]. Accessory loci may also be acted upon by negative frequency dependent selection (NFDS), hinting at their underlying role in non-serotype-specific immunity and S. pneumoniae ecology [27]. Taken together, gene variation beyond the capsular polysaccharide loci may significantly impact virulence, fitness, transmission, and, in turn, the overall epidemiology and ecology of pneumococcal strains.
Before PCV introduction, Navajo and White Mountain Apache (N/WMA) Native American communities in the Southwestern US experienced rates of invasive pneumococcal disease (IPD) 2–5 times higher than the general US population [28,29]. Pneumococcal carriage prevalence among N/WMA pre-PCV7 was 50% among all ages and 75% among children <2 years of age, significantly higher than the general population [30]. Thirty-eight percent of all pneumococcal carriage isolates were PCV7 serotypes [30]. After introduction of PCV7, carriage prevalence of PCV7 VT declined, and the rate of IPD among N/WMA caused by VT decreased by 89% [31]. However, carriage prevalence of NVT strains increased, resulting in no overall change in pneumococcal carriage prevalence among children or adults [5]. Also, despite increased NVT carriage there was no corresponding increase in the rate of IPD caused by NVT. After introduction of PCV13 in 2010, carriage of PCV13-specific serotypes declined by 60% among children <5 years of age within the first two years [6]. Yet, overall IPD rates among N/WMA still remain higher than those in the general US population [32].
Here, we analyze a sample of 937 pneumococci collected over 14 years and spanning before, during, and after the introduction of PCV7 and PCV13 vaccines among N/WMA. To understand the evolutionary impact of vaccination and characterize the shift from VT to NVT, we assessed the recombination, evolution, and pneumococcal population history, classified by serotype and by whole-genome sequencing data, across vaccine introduction periods. Furthermore, we investigated metabolic loci variation and pangenome composition over time, with a focus on pneumococcal antigens.
This study included pneumococci isolated from a subset of participants of three prospective, observational cohort studies of pneumococcal carriage among N/WMA families described elsewhere (hereafter, “parent” studies) [1,6,30]. Briefly, participants living on reservations in the southwest USA were enrolled during three periods: 1998–2001, 2006–2008 and 2010–2012. Nasopharyngeal (NP) swab specimens were obtained during visits to Indian Health Services (IHS) facilities or the participants’ home to determine pneumococcal carriage status (S1 Fig) [30]. A random subsample of isolates was selected from each time period, with an oversampling of isolates post-PCV7 (S2 Fig). A single isolate was chosen from each participant; however, previous pneumococcal carriage history was not assessed. With the exception of a subset of isolates collected from 2006–2008, all isolates were obtained from children ≤5 years of age.
Genomic DNA from S. pneumoniae isolates were sequenced on the Illumina HiSeq, yielding ≥30-fold coverage per isolate. Paired-end 100 bp reads were filtered by quality and length. Serotypes were determined by mapping reads to concatenated CPS locus sequences of 93 pneumococcal serotypes using SRST2 [12,33]. Serotypes for isolates identified as serogroup 6 were further resolved using PneumoCaT [34]. Multilocus sequence type (MLST) was determined through a similar approach using SRST2. De novo genome assemblies were generated with Velvet [35] and annotated using Prokka v1.11 [36]. After annotation, the pangenome was analyzed with Roary, and a concatenated alignment of clusters of orthologous genes (COGs) shared among ≥99% of all strains (i.e., core genome) was abstracted [37]. Pneumococcal population structure was assessed using core genome SNPs with hierBAPS, which was run three times using maximum clustering sizes of 20, 40, and 60 [38]. A maximum likelihood (ML) phylogeny was estimated using RAxML v8.1.5 with GTR+Γ nucleotide substitution model and 100 bootstrap replicates [39]. Sequence clusters (SCs) (i.e., lineages) identified using hierBAPS were annotated on the core genome phylogeny. For the study period during which pediatric and adult isolates were collected (2006–2008), the proportion of isolates by SC was compared between age groups to 10,000 random deviates of a Dirichlet distribution [40].
A subsample of isolates from each SC and 25 publicly available reference genomes were aligned using Parsnp and visualized using Gingr to identify the most appropriate genome for reference-based mapping [41]. The phylogenetically closest genome was selected for reference-based mapping of isolates belonging to that SC. For four out of 27 SCs, a monophyletic match was not available; therefore, we generated references by refining, ordering, and concatenating the best draft assembly in the SC. A second de novo assembly was generated with SPAdes and assemblies were then merged using Zorro [42]. After this, SSPACE and GAPFILLER were used to scaffold the assembly and remove Ns [43,44]. Final contigs were ordered using Mauve, manually curated using ACT, and concatenated [45]. Filtered Illumina reads from isolates comprising each SC were mapped to the selected reference using SMALT v0.7.6 and SNPs were identified using SAMtools v1.3.1 [46]. SNPs were filtered requiring a depth of coverage of five and a minimum alternate allele frequency of 0.75. The output was analyzed as previously described to generate whole-genome multiple sequence alignments for each SC [9,47].
Next, we identified recombination among SCs using Gubbins [48]. Gubbins identifies SNPs introduced through recombination and allows censoring for downstream phylogenetic analysis. Results of Gubbins analyses were visualized using Phandango [49]. For SCs in which over 50% of the genome was censored due ancestral recombination events, we either sub-clustered SCs clearly delineated monophyletic clades (e.g., SC19 which was comprised of serotypes 15A and 17F) or removed divergent isolates that were significantly affected by recombination. Sub-clustered SCs were annotated on the ML phylogeny and then reanalyzed with Gubbins.
For comparison between vaccine periods, isolates were subdivided into three epochs and six sub-epochs by year of collection: pre-PCV7 sub-epochs 1A (1998) and 1B (1999–2001); post-PCV7 sub-epochs 2A (2006) and 2B (2007–2008); PCV13 sub-epochs 3A (2010) and 3B (2011–2012) (S2 Fig). Collection years were grouped to balance sample sizes among sub-epochs. To determine the representativeness of the genomic sample to the parent studies from which the sample was drawn, we compared the serotype distribution and serotype diversity (Simpson’s D) of unique carriage isolates from the three parent studies of pneumococcal carriage [1,6,30], by epoch, to that of the sample. Core genome alignments were generated for isolates in each sub-epoch using Roary, and population genomic statistics including Tajima’s D [50], Watterson’s estimator (Θw) [51], and nucleotide diversity were calculated for each period using 0-fold and 4-fold degenerate sites. The ratio of diversity at non-synonymous sites to synonymous sites (πN/πS) was also calculated as a measure of selection. The same statistics were calculated for each SC. Code for calculating population genetic statistics using Roary output is available at http://github.com/c2-d2/Projects/NWMA_Pneumo/.
ML phylogenies of SCs, inferred from recombination-censored alignments, were used to test temporal signal by assessing correlation between strain isolation date and root-to-tip distance. SCs with poor root-to-tip correlation were assessed for residual recombination and phylogenetic signal. SCs determined to have sufficient temporal signal were analyzed with BEAST v1.8.2 [52]. For each SC or sub-SC a combination of strict and relaxed molecular clock models and constant and Gaussian Markov random field (SkyGrid) demographic models [53] were tested using recombination-free SNP alignments, ascertainment bias correction [54,55], and HKY nucleotide substitution model. For SCs in which the coefficient of variation for relaxed molecular clock models was high (i.e., significant rate heterogeneity across the tree), a random local clock (RLC) model was also tested [56]. Markov chain Monte Carlo lengths for each model run ranged from 150 million to 1 billion depending on the size of the SC and length of the SNP alignment. MCMC chains were sampled to obtain 10,000 trees and 10,000 parameter estimates in the posterior distribution. Effective sampling size (ESS) values were assessed to determine sufficient mixing using Tracer v1.6.0, and runs with ESS values of 200 for all parameters were accepted. Marginal likelihood estimates (MLE) were obtained for each model using path-sampling and stepping-stone analysis, and models were compared using Bayes Factors [57,58]. Parameter estimates for the evolutionary rate, root height (i.e., TMRCA), and Ne were obtained from the best-fit model. For SCs in which SkyGrid demographic models were fit, the slope of the Ne change over time was calculated to determine directionality, and the 95% highest posterior density (HPD) was used to determine significance.
To assess the impact of PCV7 on the pneumococcal pangenome we compared frequencies of polymorphisms in core genes and accessory genome COGs among sub-epochs, focusing on antigens and metabolic loci for the core genome and on antigens for the accessory genome analysis. We identified metabolic genes using coding sequences found in S. pneumoniae reference strain D39 (RefSeq: NC_008533.1) that were annotated as “Metabolism” according to KEGG Orthology (KO) groupings of the KEGG database (http://www.genome.jp/kegg/) and were assigned to a known metabolic pathway (KEGG pathway spd01100). Pangenome analysis using Roary was repeated including D39, and COGs found in the core genome (i.e., present among all 937 taxa) with ≥90 BLAST identity to metabolic genes were abstracted. A concatenated alignment of core metabolic COGs was then constructed, and biallelic SNP sites were identified. To assess changes to the accessory genome, we obtained the binary presence-absence matrix of accessory COGs present in frequencies ranging from 5–95% among all taxa. This frequency range was conservatively selected to mitigate the effect of genome assembly and annotation errors in COG identification. Last, we used a previously described method to identify the variants of 19 polymorphic antigens [15]. These antigens have measurable interactions with the host immune system, and therefore are thought to be under the greatest population level host immune pressure. Ten additional antigens were evaluated (lysM, lytB/C, pcpA, pcsB, phtE, piaA, piuA, psaA, SP2027, pce) but were excluded because they were deemed nearly monomorphic due to their low nucleotide diversity.
Using the concatenated nucleotide alignment of metabolic loci and a binary presence absence alignment accessory COGs and antigen variants, ML phylogenies were inferred using RAxML with GTRGAMMA (nucleotide) or GTRCAT (binary) substitution model and 100 bootstrap replicates. The cophenetic (patristic) distances of each phylogeny were read into R, and the meandist function in the package vegan was used to calculate within-group distances for three population groupings: serogroup, serotype, and SC. Within-group distances for population stratifications were then compared. For each set of genomic loci (metabolic, accessory COGs, and antigens), frequencies were computed for each of the six sub-epochs. Mean squared errors (MSEs) were then calculated to assess changes in frequencies from Epoch 1A. This was done by subsampling 75 individuals with replacement from each sub-epoch and performing 1000 bootstrap replicates of each comparison (e.g., Epoch 1A vs. 1B, 1A vs. 2A, 1A vs. 2B, and so on). The significance of changes in antigen distributions among epochs was additionally tested by comparing the proportion of antigen variants between Epochs 1–3 to 10,000 random deviates of a Dirichlet distribution.
The Navajo Nation, White Mountain Apache tribe and the IRBs of the Johns Hopkins Bloomberg School of Public Health, the Navajo Nation and the Phoenix Area IHS approved this study. During the original pneumococcal carriage studies from which these isolates were obtained, written informed consent was obtained from adult participants and from caregivers of child participants. Assent was obtained from children 7–17 years. Isolates were obtained from NP swabs, as previously described, and de-identified for analysis.
We analyzed genomic data from a total of 937 pneumococcal carriage isolates collected from N/WMA Native Americans in Southwestern US between 1998 and 2012. All isolates were obtained from children ≤5 years of age with the exception of 125 isolates (13.3% of total) collected from individuals 6–76 years of age during 2006–2008. Isolates collected from 1998–2001 (n = 274) were obtained from communities that served as the control for cluster-randomized PCV7 trials and therefore represent a vaccine naïve population. Isolates collected during 2006–2008 (n = 398) represent the post-PCV7 pneumococcal population, and isolates from 2010–2012 (n = 265) were sampled during the implementation of PCV13 (S2 Fig). Whole-genome sequencing data has been deposited in NCBI sequence read archive (SRA) under accession number ERP009399, BioProject PRJEB8327. Individual accession numbers are provided in supplementary file 1.
Pangenome analysis of de novo genome assemblies identified 8,674 COGs, of which 1,111 were present in ≥ 99% of strains (i.e., the core genome). Analysis of population structure using hierBAPS identified 27 SCs, two of which (SC27 and SC4) were polyphyletic in the ML phylogeny (Fig 1). SC27 was comprised of low frequency genotypes whereas SC4 contained three distinct monophyletic clades that were bifurcated by branches with low bootstrap support. Based on recombination analysis using Gubbins and assessment of temporal signal (i.e., molecular clock), SC4 as well as 10 other SCs were further subdivided, as it was evident that substantial ancestral recombination events occurred on branches separating dominant monophyletic clades. This subdivision is consistent with the biological definitions of lineages or sub-populations [59,60]. Subsequent analysis focused on 33 SCs or sub-SCs that varied in size from 10 to 71 isolates (Table 1). The proportion of isolates belonging to each SC differed between age groups for only four of 27 SCs, among isolates collected from 2006–2008. SC07 (serotype 35A) and SC15 (serotype 15A) were more common among children ≤5 years of age, 0.8% and 1.6% adults compared to 3.3% and 4.4% children, respectively (p = 0.03 and 0.05). SC08 (serotype 35B) and SC26 (serotypes 19A/15C) were more common among adults, 5.6% and 14.4% adults compared to 2.2% and 8.4% children, respectively (p = 0.05 and 0.04).
For temporal comparison, we divided study periods into three epochs and six sub-epochs (1A/B, 2A/B, 3A/B) (S2 Fig). To verify representativeness of isolates used for genome sequencing in this study, we obtained prevalence data on 3,868 carriage events from children ≤5 years of age in the parent N/WMA carriage studies from which the genomic sample was drawn. This included 1227 events from Epoch 1, 1038 from Epoch 2, and 1603 from Epoch 3. For the major epochs, the proportions of NVT, PCV7, and PCV13 serotypes in our sample were comparable with the serotype dynamics characterized by the three N/WMA parent studies (Fig 2A). The exception was the proportion of NVT and PCV7 VT in Epoch 1, which was due to differences between serological and genomic assignment of serogroup 6 isolates. In Epoch 1, serotypes 6B and 6C were both assigned to serotype 6B by the Quellung reaction used in the parent carriage study. This was subsequently resolved in the current study using a genomic approach to determine serotype, and later carriage studies were able to distinguish 6B from 6C. In pre-PCV Epoch 1, 26.3% of the sample was comprised of PCV7 VT, mostly serotypes 23F, 9V, 14, and 19F. Post-PCV7, the proportion of PCV7 VT in Epoch 2 fell to 1.8%. The prevalence of PCV13 VTs declined steadily from 17.5% in Epoch 1 to 11.3% in Epoch 3. The reduction in PCV13-specific VT after the introduction of PCV7 was likely due to the cross-reactivity of the 6B component of PCV7 with serotype 6A [61], which can be inferred from the elimination of SC17 (serotype 6A) after Epoch 1 (Fig 3).
Fluctuations in serotype distribution were reflected in measures of serotype diversity. Simpson’s D, which summarizes diversity as the probability that two isolates chosen at random are different, increased from Epoch 1 to 2, reflecting an increase in previously low-frequency NVT serotypes as well as the introduction of previously unobserved serotypes (Fig 2B). Fig 3 illustrates how the composition of the 27 main SCs changed during each of the three epochs. Of two lineages containing PCV7 VT only in Epoch 1, one (SC12) disappeared after vaccination, and another remained, with only PCV7 NVT isolates in Epochs 2 and 3. In SCs containing both PCV7 VT and NVT, the VT lineages were largely eliminated. After Epoch 1, the composition of the pneumococcal population in our sample and parent carriage studies shifted to a predominance of NVT and PCV13 VT, with the largest increases in serotypes 23B and 15C. While in most cases the NVT increases arose from serotypes previously observed in Epoch 1, serotypes belonging to SC10, SC22, and SC24 were not detected until Epoch 2. PCV13 VTs in our sample were not significantly impacted between Epoch 2 and 3. Further comparison of PCV13 implementation data from N/WMA communities during Epoch 3 sampling demonstrated incomplete vaccine coverage and persistence of PCV13 vaccine serotypes (S3 Fig). This finding is consistent with the previous observation that the impact of PCV13 on carriage among underimmunized children was not detected until vaccine coverage in the community reached 58% [6]. This coverage level was not attained until February 2011, at which point 52% of the Epoch 3 sample had been collected. As a result, our assessment of the impact of PCV13 on the overall pneumococcal population was limited.
We used Watterson’s theta (ΘW)–proportional to the number of polymorphic sites—and Tajima’s D to assess the impact of vaccine on population level genetic diversity and population size. Under neutrality and constant population size, ΘW = 2Neμ, where Ne is the effective population size and μ is the mutation rate [51]. Selective removal of several clusters of related strains, such as lineages or sub-lineages associated with VT, should lead to a reduction in ΘW. A related measure, Tajima’s D, tests for evidence of population growth, with negative values suggesting population expansion (due to the presence of rare variants at high frequencies) and positive values suggesting balancing selection or population contraction [50]. Consistent with our expectations, ΘW decreased from Epoch 1B to 2A, illustrating an overall decrease in pneumococcal genomic diversity, while the average number of pairwise differences (π) was unaffected (Fig 2C). Tajima’s D values computed for the polymorphic nucleotide sites in the core genome increased from -0.59 in Epoch 1B to 0.07 in 2A, signifying a removal of rare variants consistent with a species-wide population bottleneck (Fig 2D). By Epoch 3B both ΘW and Tajima’s D returned to pre-PCV7 levels while π increased. No discernible changes in either measure were associated with PCV13 introduction.
After the population genetic bottleneck induced by PCV7’s removal of VT, genetic diversity (i.e., ΘW) may have been augmented by 1) clonal expansion of NVT lineages due to selection or genetic drift (to increase ΘW such lineages would have to have been so rare post-bottleneck that they were not sampled), 2) introduction of new lineages, or 3) recombination. We hence examined evidence for each of these among individual SCs. Recombination rates (r/m) varied among SCs, ranging from 0 to 15.0, averaging 4.25 (Table 1 and S4 Fig). While coalescent analysis found SCs varied in mutation rates (S5 Fig), there was no significant difference between the median evolutionary rates of NVT and VT SCs (95% CI: -1.06e-06–8.54e-06, F(1,29) = 2.55, p = 0.12). Therefore, high evolutionary rates among NVT lineages were not solely responsible for recovering the diversity lost due to the removal of PCV7 VT.
To investigate the contribution of introduction of new lineages or expansion of previously unsampled ones, we estimated the TMRCAs (i.e., lineage age) of SCs. Overall, the median TMRCA was 1955 and ranged from 1839 (SC21: 6A/C ST473) to as recent as 2000 (SC10: 19A ST320) (S6 Fig). Two SCs that were not identified during Epoch 1 sampling emerged following vaccination: SC10 (S7 Fig), which is all type 19A and ST320, and SC24, largely comprised of serotype 23A (S8 Fig) related to PMEN clone Colombia23F-26. Estimated TMRCA for SC10 was 2000 [95% HPD: 1996–2004]. The lineage age, taken together with its low level of genetic diversity (Θw = 0.0006) and negative Tajima’s D value (-2.15), suggests that this SC was introduced after the implementation of PCV7 among southwest Native Americans and is currently experiencing population expansion. SC24 was first identified in 2006 during Epoch 2, but its most recent common ancestor was estimated at 1958 [95% HPD: 1928–1980], near the median TMRCA among all SCs. Considering its prevalence in Epoch 2 and moderate level of diversity (Θw = 0.003), it is likely that SC24 was not recently introduced and that its was present in the population before PCV7 but at a sufficiently low frequency not to be sampled until 2006, by which time its frequency may have increased. Furthermore, SC24’s low Tajima’s D value (-1.63) is consistent with population expansion.
We hypothesized that post-PCV7 changes in pneumococcal populations would be visible as decreases in the effective population size (Ne) of predominantly VT lineages and increases in those of predominantly NVT lineages. The effective population size can be interpreted as the number of genomes contributing offspring to the next generation, and changes in Ne can be used to measure population growth or contraction. Inferring demography among SCs identified that over half (56%) fit constant population size models based on MLEs (Table 1). Furthermore, while the remainder of SCs best fit a fluctuating Ne model (i.e., Skygrid), assessment of Ne trajectories identified only three that were significantly different from a constant size based on HPDs. These three SCs (SC11, SC17, and SC26-A) were found to be decreasing throughout the study period; one was PCV13 VT (SC 17) and two were NVT (SC11 and SC 26-A). To assess bias potentially introduced by removing recombination, we tested the association between recombination rates and inferred demography, which we found to not be significant (F(1,30) = 0.44, p = 0.51) [62]. Overall, these findings show that the relatively subtle increases in sample frequencies of individual SCs containing NVT are not visible as departures from a constant Ne.
To test the hypothesis that selective removal of PCV7 VT disrupted accessory genome content, we compared accessory size and frequencies of 2370 COGs and 53 variants of 19 antigens between pre-PCV7 Epoch 1 to post-PCV7 epochs. Further, we tested the concurrent effect on metabolic loci by assessing frequencies of 22,434 biallelic SNPs found among 256 metabolic genes present in the core genome. For metabolic loci, accessory COGs, and antigen variants, within-group diversity was minimized when SC population groupings were assigned, compared to serogroup and serotype (S9 Fig). The introduction of PCV7 resulted in an overall reduction in pangenome size, illustrated by the difference in logarithmic pangenome curves for Epochs 2A and 3B (S10 Fig). A comparison of pre-PCV7 Epochs 1A and 1B provided a baseline estimate of stochastic, temporal fluctuations in frequencies in the absence of an effect of vaccine. Plotting COG frequencies in subsequent epochs demonstrated perturbation in pneumococcal accessory COGs frequencies following introduction of PCV7 (S11 Fig). This perturbation is characterized by the dispersion of frequency scatterplots comparing Epochs 1A vs. 2A [R2 = 0.96, MSE = 8.26x10-3 (95% CI: 8.32x10-3–8.40x10-3)] and 2B [R2 = 0.98, MSE = 6.65 x10-3 (95% CI: 6.60x10-3–6.70x10-3)] (Figs 4 and S11). This effect was also observed when comparing the frequencies of polymorphic antigens and metabolic loci between epochs (Figs 5, S12 and S13). For all sets of genomic loci, MSE in comparison to Epoch 1A are smaller for 1B than for any of the subsequent epochs, illustrating the disruption caused by PCV7. While this observation alone could be explained by drift leading to increasing divergence in frequencies over time, a further observation cannot: in each example, MSEs decreased from Epoch 2 to 3, indicating metabolic loci, accessory COGs, and antigen frequencies were trending back toward pre-PCV7 values (Fig 5). This trend was observed when isolates collected from individuals >5 years of age were removed from Epoch 2 and the analysis repeated. This led us to compare Epoch 3A (post-PCV7/pre-PCV13) to previous sub-epochs to determine whether the pre-PCV7 Epochs 1A/B or the immediately preceding Epoch 2B were better predictors of COG/antigen frequencies. For accessory genome COG frequencies and metabolic loci, Epoch 2B was a better predictor of 3A frequencies; however, for antigens, pre-PCV7 Epoch 1B was the best predictor of Epoch 3A frequencies (S14 Fig). Taken together, we found that antigen variant frequencies largely returned to pre-PCV7 values; however, some perturbations were not resolved (Fig 6). This was due largely to pspC groups 1/5 (p = 0.01) and srtH Var-I (p = 0.004), which remained at higher frequencies at Epoch 3, and rrgA Var-I (p<0.001), which was completely removed from the population.
The impact of PCV7 introduction on pneumococcal serotype distributions has been well-characterized in the N/WMA and other communities, but the pneumococcal genome-wide impact has been investigated in fewer populations [3,63]. We studied genomes from a sample spanning the introduction of PCV7 and PCV13, which, based on serotype distribution, were representative of the full set of data from which the sample was drawn. Beyond the expected impact on serotypes, we find the effect of vaccine on the pneumococcal population could be observed as changes in population level diversity, metabolic loci, size of the pneumococcal pangenome, and frequencies of accessory genes including polymorphic antigens. We further illustrate how pneumococcal genomic diversity and frequencies of accessory genome COGs rebounded after the population bottleneck induced by the selective removal of VT lineages by PCV7. These findings help explain how the frequency distribution of polymorphic antigens, for example, largely return to baseline frequencies after being disrupted by vaccine.
The post-PCV7 pneumococcal population in N/WMA saw the complete removal of two SCs and a significant reduction in prevalence of three. The population bottleneck was characterized by changes in levels and patterns of genomic diversity, decreasing ΘW and increasing Tajima’s D (Fig 2). Subsequently, the removal of VT pneumococci was counterbalanced by the expansion of SC9 and the emergence of two previously unobserved SCs, SC10 and SC24. In Epoch 2, we identified minor variations in the distribution of SCs by age group for four SCs. As none of the SCs contained PCV7 VT, differences likely resulted from variation in acquired serotype-specific immunity among children and adults [64]. Overall, population structure of SCs was comparable, consistent with pneumococcal transmission dynamics and the wide-ranging impact of the PCV7 vaccine on carriage in children and adults [5]. Despite the changes in the prevalence of SCs over time, no consistent pattern of change in the Ne of these SCs was detectable through coalescent analysis of individual SCs (Table 1). This lack of signal may be due to a number of factors. It may be that where vaccine pressure was strong enough to drastically change the population size of an SC, it was eliminated (e.g., SC12), so the temporal signal was lost; where changes were more modest, e.g. in SC including both VT and NVT, the method may have been too insensitive to detect a change. While assessment of Ne did not clearly identify consistent changes, we did detect the post-PCV7 emergence of two SCs. By comparing TMRCA and core genome diversity, we infer that that the first, SC10, appears to have been recently introduced among N/WMA, while the second, SC24, appears to have become detectable due to the vaccine [8,65]. It is worth noting that assessing Ne and other population genetic parameters of pneumococcal lineages makes implicit assumptions about defining SCs as populations and a collection of SCs as a metapopulation, which, to varying degrees, may compete or interact with one another through recombination. Indeed, this definition is more complex and requires consideration of competition, gene flow, and niche overlap among lineages [60,66,67]. Here, we statistically define SCs and find that these populations are often good predictors of serogroup, metabolic profile, and gene content, thus generally demonstrate genomic coherence consistent with the concept of a bacterial population.
Pneumococcal genomic data from carriage studies in the US are limited [9]. The N/WMA sample provides an opportunity to assess post-vaccine changes in the pneumococcal populations across demographically and geographically varied regions and, at large, the generalizability of bacterial pathogen population dynamics. Comparable analysis of population structure of 616 carriage isolates from Massachusetts collected between 2001 and 2007 found less structure (15 monophyletic SCs (n = 616)) compared to the N/WMA sample (25 monophyletic SCs (n = 937)) [9], and unlike Massachusetts, where the post-PCV7 population emerged largely from the pre-existing serotype diversity, in the N/WMA sample we observed seven previously unidentified serotypes and two entire SCs post-PCV7. Considering carriage data from the larger parent studies, 13 previously unidentified serotypes, excluding 6C, were observed post-PCV7. This difference aside, SC composition and pneumococcal population dynamics were consistent between N/WMA and Massachusetts. For example, SC9 (also SC9 in the Massachusetts study [9]) experienced a near identical population shift post-PCV7 (S15 Fig). This SC, which is comprised of VT 23F and NVTs 23A and 23B, is thought to have arisen through multiple serotype-switching events. In the N/WMA sample, it was one of the most successful in terms of overall prevalence in Epoch 1. As observed in Massachusetts, PCV7 effectively removed 23F isolates from the SC; however, SC9 NVTs subsequently increased 3.5% from Epoch 1 to 3. This shows that these changes were not restricted to the Massachusetts population, but were replicated in a very different setting, and may suggest that SC9 occupies a specific niche. Consistent with this hypothesis, we find that the antigen profiles for VT 23F and the NVT 23B population that replaced it, to be largely consistent with the exception of zmpA (S16 Fig). Taken together, we observe similar pneumococcal population dynamics in two geographically and demographically distinct populations that share common vaccine histories, suggesting that response to population shaping processes are relatively consistent.
We find that each SC is defined by a unique profile of metabolic loci, accessory COGs, and antigen variants. These profiles are most resolved at the SC level rather than serotype or serogroup, as the same serotype can be found in multiple SCs due to switching events. Moreover, within an SC, these genomic loci show significant linkage disequilibrium despite appreciable recombination among pneumococci [68]. Consistent with this linkage, we observed a coincident impact of PCV7 on genetic diversity, accessory COG frequencies, polymorphic antigens, and metabolic loci. The population genomic perturbation that resulted from the removal of PCV7 VT was significantly mitigated by Epoch 3, with frequencies of antigen variants, in particular, returning to pre-vaccine values. A recently proposed model of NFDS provides one putative mechanism for the maintenance of antigen variants and accessory COGs at optimal frequencies [27], and variant-specific host immunity provides a biologically plausible mechanism for NFDS on antigens. Early evidence of balancing selection among pneumococci was the reemergence of strains possessing a type 1 pilus after PCV7 significantly reduced piliated serotypes [69]. In the current study, we also observe the reemergence of type 1 pilus driven by serotype 19A ST320 (SC10). And while the observation with the pilus involved a change in presence-absence frequency, we now see the same dynamic extending to frequencies of antigen variants. Yet, due to linkage it is difficult to untangle which loci are being acted upon by selection and which reflect hitchhiking. Alternatively, balancing selection could be acting upon metabolic loci which are important to niche adaption and have been implicated in post-vaccine metabolic shifts [18]. In an effort to identify which loci may be driving post-vaccine success of SCs, we considered the frequencies of metabolic loci, accessory COG, and antigen variants separately. We find PCV13-era (Epoch 3) frequencies of polymorphic antigens are better predicted by pre-PCV7 (Epoch 1) frequencies than the immediately preceding period. In addition, we observe that overall COG frequencies seemed to trend toward pre-PCV7 norms with increasing time since vaccine introduction, while frequencies of metabolic loci remained disrupted. This does not rule out variation in metabolic loci or other core genes such as GroEL as driving forces for pneumococcal population structure [70]; however, it remains difficult to assign fitness differences based on observed genetic variation. For example, two SCs may be divergent in metabolic loci but capable of exploiting the same metabolic niche.
Previous models have proposed that recombination is the mechanism underlying the post-vaccine shift in metabolic, virulence, and antigenic loci [18]. However, we argue that in our sample, recombination has likely not had enough time to shuffle antigen variants or other COGs into different genomic backgrounds. For example, if we again consider the replacement of VT 23F by NVT 23B belonging to SC9, we observe that both populations possess similar antigenic profiles (S16 Fig). Yet, the TMRCA of the 23B population, and all associated recombination events, predate the introduction of PCV7 (S15 and S16 Figs). This illustrates that at least in this case, an existing population possessing a near identical antigenic profile contributed to the rebalancing of the distribution of antigen variants in the overall pneumococcal population. Overall, the pneumococcal accessory genome is comprised of varying types of MGE (e.g., phages and antigens), and it is likely that their distribution is controlled by many different, yet interconnected, processes [17]. As such, the underlying dynamics maintaining antigenic variant and accessory COG frequencies require further investigation.
Through comprehensive analysis of serotype distribution and population dynamics of S. pneumoniae spanning the introduction of PCV7 and PCV13 among N/WMA communities, we gain a broad understanding of the impact of vaccine on population structure, serotype distribution, and pangenome composition. After the introduction of PCV7, we observe clonal replacement of VT by NVT as well as clonal expansion of vaccine-associated serotypes during a period when carriage prevalence remained unchanged. Further, we show PCV7 significantly disrupted accessory COG frequencies, including frequencies of polymorphic antigens important to host-pathogen interactions. This post-PCV7 period of ‘flux’ in serotype diversity and accessory COG distribution was normalized by Epoch 3, demonstrating rapid adaption to the post-vaccine landscape. Moving forward, continued genomic surveillance will be required to monitor the emergence of new lineages and to investigate the impact on post-PCV13 pneumococcal populations. Last, as balancing selection appears to be an integral component of pneumococcal adaption and considerable serotype-lineage-accessory genome linkage exists, the joint effect of removal of vaccine serotypes and linked antigens on host-susceptibility to extant lineages merits further study, as it has significant implications for the future of protein-based pneumococcal vaccines. For example, protein-based vaccines should consider the prevalence of polymorphic variants across host populations and either include multiple variants of the same antigen or target those in greatest frequency.
|
10.1371/journal.pcbi.1002672 | Interplay of Gene Expression Noise and Ultrasensitive Dynamics Affects Bacterial Operon Organization | Bacterial chromosomes are organized into polycistronic cotranscribed operons, but the evolutionary pressures maintaining them are unclear. We hypothesized that operons alter gene expression noise characteristics, resulting in selection for or against maintaining operons depending on network architecture. Mathematical models for 6 functional classes of network modules showed that three classes exhibited decreased noise and 3 exhibited increased noise with same-operon cotranscription of interacting proteins. Noise reduction was often associated with a decreased chance of reaching an ultrasensitive threshold. Stochastic simulations of the lac operon demonstrated that the predicted effects of transcriptional coupling hold for a complex network module. We employed bioinformatic analysis to find overrepresentation of noise-minimizing operon organization compared with randomized controls. Among constitutively expressed physically interacting protein pairs, higher coupling frequencies appeared at lower expression levels, where noise effects are expected to be dominant. Our results thereby suggest an important role for gene expression noise, in many cases interacting with an ultrasensitive switch, in maintaining or selecting for operons in bacterial chromosomes.
| In some species, most notably bacteria, chromosomal genes are arranged into clusters called operons. In operons, the process of transcription is physically coupled: a single pass of the RNA polymerase enzyme reading that region of the chromosome simultaneously produces messenger RNA encoding multiple proteins. So far, we do not have a satisfying explanation for what evolutionary forces have maintained operons on bacterial chromosomes. We hypothesized that different types of interactions between operon-coded proteins affect how strongly operons are selected for between two genes. The proposed mechanism for this effect is that operons correlate gene expression noise, changing how it manifests in the post-translational network depending on the type of protein interaction. Mathematical models demonstrate that operons reduce noise for some types of interactions but not others. We found that operon-dependent noise reduction has an underlying dependence on surprisingly high sensitivity of the network to the ratio of proteins from each gene. Databases of genetic information show that E. coli has operons more frequently than random if operons reduce noise for the type of interaction various gene pairs have, but not otherwise. Our study thus provides an example of how the architecture of post-translational networks affects bacterial evolution.
| The organization of genes into operons is a prominent feature of bacterial chromosomes [1] that appear in some eukaryotes as well [2]. An operon is typically characterized as a promoter followed by multiple genes that are cotranscribed so that each transcription initiation event produces a polycistronic messenger RNA (mRNA) encoding multiple gene products [3]. Hypotheses explaining the emergence and maintenance of operons include proportional coregulation [4], [5], [6], [7], [8], [9], horizontal transfer of intact “selfish” operons [10], emergence via gene duplication [11], coproduction of physically interacting proteins to speed their association [12], [13], evolvability of co-regulation for interacting protein products [14], and reduction of intrinsic noise [15]. Current evidence favors some hypotheses more than others, but fails to indicate a definitive explanation for how operons are maintained in bacterial chromosomes [4], [11], [12], [13].
Arising from stochasticity of individual biochemical reactions and low copy numbers of reactants per cell, intrinsic noise plays a central role in network dynamics [16], [17]. In bacteria, intrinsic noise is most evident in gene expression, caused by translational bursting arising from small numbers of mRNA producing many proteins [18], [19] and transcriptional bursting arising from slow activation-deactivation cycles of transcriptional activity by unknown mechanism [20], [21]. Extrinsic noise, caused by uncertainties in global parameters and states including those characterizing transcriptional and translational machinery, also contributes to overall biochemical noise [22]. Noise in protein levels is commonly characterized by coefficient of variation (CV), the normalized root-mean square deviation of the protein levels from their mean value (CV = σ/μ, where σ is the standard deviation and μ is the mean) but other measures such as autocorrelation and covariance between concentrations of different proteins can give additional insights.
The effects of intrinsic noise on operon maintenance are not well characterized, but covariance between protein levels arising from intrinsic noise depends on transcriptional coupling (co-expression from an operon) of the corresponding genes [15]. The order of genes within an operon may also affect noise [23]. Therefore, we hypothesize that noise-related effects contribute to the evolutionary maintenance of operons. Studies of several specific systems corroborate that correlative effects of transcriptional coupling alter posttranslational dynamics [6], [24], [25]. However, it is still not clear how different classes of protein interactions and co-expression from an operon may interact to alter biochemical noise.
In this study we assessed these effects for different types of posttranslational interaction between gene products. In several types of interactions, the noise difference between cotranscribed and uncoupled configurations was amplified by the existence of a zero-order ultrasensitive switch [26]. We related the results to an intact naturally occurring system with simulations of cotranscribed and uncoupled configurations of the lac operon. To test our predictions bioinformatically, we classified naturally occurring interacting pairs of E. coli proteins by their type of interaction and analyzed the effect of chromosomal distance between pairs of genes with interacting products. Finally, we used single-cell protein copy number data to determine differences in operon frequencies at high and low expression levels in E. coli for physically interacting protein pairs.
Intrinsic gene expression noise is correlated in a cotranscribed two-gene configuration, but this correlation was not seen in an uncoupled configuration. Relationships between the fluctuations of two proteins can be quantitatively characterized by the covariance of concentrations for proteins A and B (σAB). Using the linear noise approximation (LNA; see Materials and Methods) [18], , we calculated a normalized covariance(1)where angle brackets represent average copy-number of each molecular species, and τmRNA and τprotein are the characteristic timescales of mRNA and protein decay. Increased covariance of cotranscribed genes is preserved regardless of whether the translation processes are coupled (with a single ribosome binding site for multiple genes; Figure 1A–C) and regardless of the source of intrinsic noise (from translational bursting only or from both transcriptional and translational bursting; Figure 1B–C). We hypothesized that positive covariance can increase or decrease CV in relevant network outputs depending on the nature of the interactions between two proteins.
In addition to covariance, the effect of operons on correlation between protein copy number fluctuations can be quantified by another measure, the degree of decorrelation (Text S1). This measure is useful to characterize the effect of gene expression level on noise differences between cotranscribed and uncoupled proteins (Table S1), and to assess the effects of expression level on frequencies of operon occurrence in bacterial genomes.
We surveyed databases of E. coli biochemical networks [29], [30], [31] to identify simple two-gene modules of larger networks that represent different ways that two proteins can directly or indirectly interact. The modules represent simple models of the following interactions: catalysis of subsequent steps in a linear metabolic pathway (Figure 2A), redundant catalysis of the same metabolic step (Figure 2B), catalysis of diverging reactions following a branch point in a metabolic pathway (Figure 2C), redundant transcriptional regulation of a downstream gene (Figure 2D), physical binding between two proteins (Figure 2E), and covalent modification of one protein by another (Figure 2F). The list may not be fully comprehensive, but represents several classes of interactions between proteins that are building blocks of larger networks. For each module we constructed a mathematical model to calculate CV for interacting proteins transcribed from the same and different operons (hereafter referred to as cotranscribed and uncoupled configurations, respectively). We then determined differences in CV for relevant network outputs between cotranscribed and uncoupled configurations. The simulations were controlled by keeping the same mean and CV for total protein from each gene between configurations. The CV calculations were performed at stationary state, both numerically (stochastic simulation algorithm; [32]) and analytically (LNA [28] using Paulsson's [18] normalization). The results of the simulations demonstrate that predicted differences in CV for each metabolic module depend on the type of interaction between proteins (Figure 2).
For the linear metabolic pathway module, cotranscription of two enzymes from the same operon results in lower CV for metabolic intermediate. Without transcriptional coupling, metabolic intermediate concentrations are prone to large spikes (Figures 2A and S1A, B, J, K). Notably, no significant differences between cotranscribed and uncoupled configurations are evident in metabolic product CV (Figure 2A), indicating that metabolic flux is not significantly different between the two groups. Intuitively, a spike occurs when flux from the upstream enzyme exceeds the maximal flux capacity of the downstream enzyme resulting in large increase of metabolic intermediate concentration. This increase exceeds the saturation point for the enzyme converting it to product, making product concentration insulated from these spikes.
In contrast, the metabolic modules with redundant enzymes (Figures 2B and S1C, L) and with a branch point (Figures 2C and S1D, M) show an increase in metabolite CV when the two enzymes are in the same operon. In these cases, lower correlations between enzyme fluctuations reduce the chance of simultaneous stochastic drops in concentration of both enzymes.
Similarly, cotranscription of multiple (redundant) gene regulators from the same operon results in increased CV in the regulated gene as compared to the uncoupled regulator configuration (Figures 2D and S1E, N). Here, we assumed that the gene regulatory logic was an OR gate (i.e., each regulator by itself or both together would have the same effect). Noise in the output from AND gate logic (i.e., a multi-subunit regulator) is expected to follow the noise pattern of the physical interaction module (below).
Consistent with a previous study [25], the physical protein interaction module under transcriptional coupling shows a strong reduction in fluctuations of monomer concentrations. (Figures 2E and S1F, G, O, P). With strong binding, the concentration of each free monomer changes from nearly zero when its partner is in excess to a finite value when the monomer itself is in excess. These fluctuations are more common when the binding partners are not in the same operon, so the noise is therefore high. Cotranscription slightly increases heterodimer CV compared to the uncoupled configuration (species AB; Figure 2E), but to a much lesser extent than its reduction of CV in monomer concentrations. In the limit of strong binding, nearly all of one monomer is bound, so the effect on monomer noise is dominant.
For the covalent modification module (Figures 2F and S1H, I, Q, R), different gene configurations cause small changes in the CV of the modified form of protein A (A*) that can be of either sign depending on parameter values, whereas the unmodified form (A) has consistently lower CV in the cotranscribed configuration (Figures 2F and S1H, Q).
The stochastic simulation approach (Figure 2) gives decisive results, but only for the parameter values tested. To determine how generally the simulation results hold in the face of different parameter values, we used the LNA to analytically determine noise differences (here quantified as CV2) between cotranscribed and uncoupled forms of each network module. For each molecular species denoted by j, we calculated the noise difference between cotranscribed () and uncoupled () configurations as . If the value is positive, the cotranscribed configuration has lower CV2 (and therefore, lower CV); if it is negative, the uncoupled configuration has lower CV2. A more complete analysis for each module is presented in Text S2. Here we highlight the main results.
To summarize, we found that differences in noise between cotranscribed and uncoupled configurations in stochastic simulations are qualitatively consistent with the analytical approach. Notably, in all the cases the magnitude of the differences in CV2 between two configurations is proportional to the value of covariance , but in many cases the coefficient of proportionality is very large. This qualitatively suggests posttranslational interactions in some modules are capable of amplifying noise differences between cotranscribed and uncoupled proteins. However, the LNA method likely underestimates the magnitude of non-linear amplification. We further explore these amplification mechanisms in the next section.
Timecourse simulations predict that uncorrelated fluctuations of the two enzymes in a linear metabolic pathway result in large bursts of metabolic intermediate (Figure 3A, B). This suggests that higher noise in the transcriptionally uncoupled linear metabolic pathway arises at least in part from the increased probability of occasionally crossing an ultrasensitive threshold. Indeed, a sharp threshold in the intermediate of the linear metabolic pathway arises when the enzyme-mediated consumption of a product saturates, leading to non-linear degradation [33]. The ultrasensitive threshold is crossed when the downstream enzyme saturates, and the flux from the upstream enzyme exceeds its maximal value (V+/V−>1 in Figure 3C). Because V+ and V− are proportional to their enzyme levels, the numerator and denominator of the ratio fluctuate together when both enzymes are in the same operon. Therefore transcriptional coupling lowers noise in the flux ratio and making it unlikely to cross the threshold V+/V− = 1. When the enzymes are uncoupled, simulations show more variability in the V+/V− ratio, allowing the ratio to cross the threshold with consequent large spikes in metabolic intermediate. Thus, the ultrasensitive switch amplifies noise differences already present between cotranscribed and uncoupled configurations.
Differences in noise between cotranscribed and uncoupled configurations of all of the non-redundant modules can be amplified by ultrasensitive switches in a similar manner (Figure S2). The metabolic branch point module undergoes the same type of non-linear degradation effect as the linear metabolic pathway, but in the branch point transcriptionally coupled enzyme pairs are more likely to fluctuate downward and saturate simultaneously than uncoupled enzymes. This effect leads to a higher likelihood of substrate buildup in the cotranscribed configuration (Figure S2B). The physical interaction and covalent modification modules undergo molecular titration [34], resulting in an ultrasensitive switch for monomers (physical interaction module) or unmodified protein (covalent module) that depends on the ratio of protein production fluxes (Figure S2C, D). Cotranscription of the two genes prevents the switch from amplifying transcriptional noise by reducing fluctuations in this ratio. Sensitivity analysis of mean-field models shows that the existence of ultrasensitive switches does not depend on strict parameter regimes (Text S3).
To explore how conclusions drawn from models of simple network modules apply to a more complicated realistic network, we implemented stochastic simulations of a detailed lac operon model that is based on a previous deterministic model [35]. The stochastic model includes enzymatic steps reminiscent of a linear metabolic pathway with permease-mediated lactose import and conversion by β-galactosidase to allolactose and β-d-galactose+β-d-glucose (Figure 4A, Tables S7 and S8). Feedback and gene regulation are present with derepression of lacY and lacZ expression caused by allolactose binding to LacI (Figure 4A, Tables S7 and S8).
We simulated three inducer concentrations representing minimal lac operon induction (1.39 µM extracellular lactose concentration or 835 molecules/femtoliter), intermediate induction (83.0 µM or 50,000 molecules/femtoliter), and excess inducer with maximal lac operon induction (∼5,000 µM or 3×106 molecules/femtoliter). Timecourses suggest that transcriptional coupling between lacY and lacZ (wild-type situation) eliminates the large fluctuations in allolactose (Figures 4B and S3) and intracellular lactose (not shown) observed in the transcriptionally uncoupled form of the system. This is consistent with a reduction in the correlation between permease and β-galactosidase (lacY and lacZ gene products, respectively) in time (Figure 4C).
At all inducer concentrations, the uncoupled configuration displays higher CV in allolactose than did the cotranscribed configuration (Figure 4D). This difference is most pronounced in the minimal induction region and gradually reduced with increasing lacY-lacZ induction. At the same time, there is little difference in protein CV between cotranscribed and uncoupled configurations of the model at most inducer levels. In both configurations the CV monotonically decreases with higher expression.
The primary consequence of cotranscription of lac proteins in the same operon is a reduction in fluctuations of intracellular lactose and allolactose. These fluctuations may prevent disruption of other sugar uptake pathways by, for example, interfering with inducer exclusion mechanisms [36]. Physiological benefits of noise reduction are also consistent with reports that excessive lactose import is associated with significant lowering of growth rate in E. coli [37 and references therein,38]. Thus, there may be a selective pressure to maintain high covariance between permease and β-galactosidase resulting from the wild-type genetic structure of the lac operon.
To determine if global operon organization in E. coli correlates with predicted noise differences, we characterized frequencies of gene membership in the same operon bioinformatically (Table 1). We first assigned membership of known E. coli K12 MG1655 [39] biochemical networks into patterns corresponding to the 2-gene modules (Figure 2) using data on E. coli operons [31], metabolic pathways [29], gene regulation networks [31], covalent modification [30], and physical protein interactions [40], [41]. Many natural networks fall into more than one class (e.g., common bacterial signal mediators, two-component systems, have physical interactions between the sensor and the regulator [42] and are also in the covalent modification class). For the metabolic and gene regulation network modules, we eliminated physically interacting pairs to ensure that those included had true functional overlap and were not acting as subunits of a larger enzyme or regulator. Thus, the only systems that are members of more than one class are in members of both the covalent modification and physical interaction modules. In each class we created controls with randomized operon assignment of the genes (see Materials and Methods).
Proteins in the linear metabolic pathway, physical interaction, and covalent modification modules appear in the same operon significantly more frequently than do randomized controls (p<<10−6; Table 1). On the other hand, redundant metabolic nodes and multiple gene regulators are significantly less likely to be in the same operon than randomized controls (p<<10−6; Table 1). Metabolic branch points show a bias toward being uncoupled, but falls just short of being statistically significant (p = 0.071). These findings hold even after we divide each class into essential and nonessential genes using data from Taniguchi et al [43]; Table S2). Thus operon overrepresentation, when it occurs, is present in essential genes, consistent with previous results contradicting the selfish operon hypothesis [12]. Our results establish a correlation between operon organization of protein pairs and their function that is consistent with noise minimization and avoidance of ultrasensitivity.
To separate the specific effect of noise from that of other factors affecting selection for operons, such as proportional coregulation, we considered whether the tendency toward operon membership of posttranslationally interacting protein pairs is related to gene expression levels [23]. Intrinsic noise is stronger for genes with low expression levels [43], covariance of protein concentrations is more pronounced (Equation 1) and the degree of decorrelation is higher (Table S1). Therefore, if noise is an evolutionary factor driving operon formation, levels of gene expression may be inversely correlated with operon patterns. On the other hand, if coregulation of mean expression levels is the dominant factor in selecting for operons, the frequency of transcriptional coupling may be directly correlated with gene expression levels because the cost of differential regulation would be highest at the highest expression levels. As a result, any trend in coupling frequencies with gene expression levels would favor one hypothesis and disfavor the other.
We used a dataset of average single-cell mRNA and protein copy numbers in E. coli [43] to explore this trend for constitutively expressed physically interacting protein pairs (other network modules have insufficient data for such analysis). Because different conditions can shift gene expression levels and the dataset is only available for one condition, we chose to focus on the subset of interacting proteins that are constitutive, i.e., not predicted to undergo any regulation in RegulonDB. Each gene's protein or mRNA copy number was considered once, along with a binary variable indicating whether or not the protein product interacts with a same or non-same operon protein. Further details are given in Materials and Methods.
We divided the set into two subsets of expression level, one below and one above the median copy number (Figure 5). The fraction of protein pairs sharing the same operon is higher in the low-expression subset for protein (bootstrap test p<0.01) and mRNA (bootstrap test p<0.05) copy numbers. This suggests that evolutionary selection against decorrelation (Table S1) significantly contributes to maintenance of operons in the chromosome.
A longstanding question in evolutionary biology is how non-transcriptional dynamics [44] affect selection of particular genetic architectures. By relating chromosomal patterns to protein network structures in E. coli, we see a compelling case for post-translational dynamics altering the probability of operon membership of genes depending on the nature of their interaction.
Because enzymes often operate close to saturation [45], resolving metabolic flux imbalances may prevent widespread accumulation of intermediate, which is potentially toxic [46], [47], [48]. Simulations of a detailed lac operon model in our study corroborate the results of the simpler linear metabolic module, suggesting a role for intrinsic noise in selecting for lac operon architecture (in addition to the stochastic effects previously examined in this system [49]). Simulations that include extrinsic noise as a correlating factor indicate that it does not reduce metabolite noise as well as the stronger correlations caused by cotranscription (Text S4, Figure S4).
Many metabolic operons are large (and with complex evolutionary histories; [50]), but the length of a metabolic pathway is often longer than that of a typical operon, leading to the question of where optimal operon break points for metabolic pathways may lie. Our results suggest that break points occur predominantly where the intermediate is not toxic or where it is processed by multiple downstream enzymes, such as at branch points and metabolic steps with redundant enzymes. Metabolite spikes could also potentially be buffered by reversibility of catalytic reactions, though the reversible step in the lac operon did not prevent intermediate spikes. Furthermore, if portions of metabolic pathways that are divided by intermediates with relatively low toxicity undergo upregulation as needed, there may be a trade-off between reduction of toxic intermediate spikes and just-in-time transcription [51] in the evolution of metabolic networks.
Our analysis suggests that pairs of enzymes after a branch point can have lower noise (CV) if they are not cotranscribed (Figure 2C), but with a less consistent CV difference between cotranscribed and uncoupled configurations than the other modules (Figure S1 D,M). Therefore, the noise hypothesis predicts patterns of transcriptional coupling to be weaker than in other modules, as we observe to be the case in E. coli (Table 1).
The simple physical protein interaction module in our study (Figure 2E) may result in one of two different types of physiologically meaningful output variables: an active heterodimer, in which the genes make up subunits of a functional complex, or an active monomer, in which its activity is negatively regulated by the binding partner (as with sigma-antisigma systems [52]). In either case reduction in monomer noise is justified; in the latter case, to reduce noise in the physiologically relevant output. In the former case, lower noise represents a reduction in inefficient protein production that can reduce promiscuous interactions with other parts of the network. Heterodimer noise is smaller for the uncoupled configuration because upward fluctuations in its concentration are limited to being no larger than the minimum of [A] and [B] and those concentrations are less likely to simultaneously fluctuate upward simultaneously.
The covalent modification system (Figure 2F) in its uncoupled configuration has reduced fluctuations in the unmodified protein (A) compared with the uncoupled configuration. Noise effects of transcriptional coupling may therefore be important in covalent modification systems where the unmodified form of the protein is capable of interacting with other systems (Text S2).
Higher-order chromosome structure, such as bacterial chromatin [53], [54] and regulatory factors such as bidirectional promoters and transcriptional terminators [55] affect the spatial proximity of genes. Operons could also play a role in spatial proximity, as suggested by the selfish operon hypothesis [10]. We explored whether chromosomal proximity can explain operon membership in linear metabolic and physically interacting gene pairs. Our bioinformatic analysis suggests that the prevalence of operons cannot be solely explained by a proximity bias of interacting gene pairs in the E. coli chromosome (Text S4, Figure S5).
A striking feature of the non-redundant protein interaction modules is that they all contain a zero-order ultrasensitive switch, which arises as a side-effect of saturation. This effect amplifies differences in CV between cotranscribed and uncoupled forms of the modules (Figures 3 and S2) and may degrade performance when its threshold is crossed. In each two-gene module, the operon architecture that avoids crossing the ultrasensitive threshold is significantly over-represented in E. coli (Table 1). Signatures of selection against noise in these modules thus likely represent selection against performance-degrading ultrasensitivity as well.
Gene pairs encoding constitutive physically interacting proteins are significantly more likely to be in the same operon if their expression levels are low (Figure 5). This trend could be explained by slow protein diffusion in the crowded intracellular environment, as cotranscribed gene products are more likely to be present at the same subcellular location. However even slow diffusion (<1 µm2/s) across a typical bacterial length of ∼1 µm is much faster than the expected time lag between translation of two proteins given typical ribosomal speeds of 12–21 AA/s [56]. Therefore, increased biochemical noise (here, measured as decorrelations between uncoupled proteins) at low expression levels are the most likely explanation of the observed trend. We argue that these noise effects are detrimental to the performance of some protein interaction networks.
The opposite trend would be expected if proportional expression of mean concentrations or other mechanisms are the primary selective pressure on operon maintenance. In general, genes with high expression levels may operate under greater evolutionary pressure than genes with low expression levels [57], [58] and therefore their deviation from optimal chromosomal organization is less likely. Arguably, noise minimization is the only selective force that is expected to be more important for genes with low expression levels than for genes with high expression levels [23].
Partial functional redundancy of proteins allows one protein to compensate for a downward fluctuation in concentration of the other protein, thereby reducing noise with uncorrelated protein fluctuations (Table 1; Figure 2). Therefore, just as noise minimization may explain operon membership for non-redundant interactions, it may also explain the lack of redundant proteins in operons. Differential regulation of the genes can additionally play an important role in keeping redundant interactions transcriptionally uncoupled. In yeast metabolic pathways, apparently redundant enzymes are differentially expressed in different pathways depending on external conditions [59]. This type of mechanism, if present in E. coli, may also explain why no redundant enzymes are in the same operon. Similarly, different growth conditions may result in different regulators affecting downstream expression of the same genes. Further work is necessary to distinguish the noise reduction hypothesis more decisively from differential gene regulation as a selective force in redundant pairs; differential regulation may be physiologically important in some cases and not in others.
Improvement of dynamic performance of simple networks arising from cotranscription of interacting genes from the same operon raises the question of why operons are rare in eukaryotes. Eukaryotic cell volumes are much higher than prokaryotes, likely lowering the effect of intrinsic noise relative to the dominant effect of extrinsic noise [60]. Nevertheless, such benefits may still be present in some systems, and there are mechanisms that allow correlating gene expression noise in eukaryotic cells without polycistronic loci. Genes located near each other have correlated transcriptional bursts that likely arise from chromatin decondensation [61], [62]. Clusters of co-expressed genes, particularly metabolic genes, appear in eukaryotic chromosomes at a rate higher than would be expected randomly [63], [64]. Co-expressed, functionally related genes at distant genomic loci also appear to migrate together for co-transcription from discrete transcription initiation complexes [65], [66], [67]. These mechanisms, arising from the increased size and structural complexity of eukaryotic chromatin over prokaryotic chromosomes, can correlate gene expression noise with similar dynamic benefits to operons.
We have developed a theory predicting that operon membership can increase or decrease noise in different types of protein interactions. Bioinformatic analysis finds that naturally occurring operon patterns in E. coli correlate with reduction of biochemical noise. Nevertheless, it would be interesting to explore operon coupling frequencies in bacterial stress response systems known to favor population-level heterogeneity, such as stress responses in B. subtilis; the amplification of noise by underlying ultrasensitive switches in non-redundant network modules may be a potential mechanism of population-level heterogeneity.
The existence of implicit ultrasensitive switches also underscores the idea that dramatic non-linearities are likely present in many simple protein interaction networks. Our results suggest that ultrasensitive switches are likely undetectable in the wild-type configurations of well-adapted systems as a result of selection against them, but may be present in conditions with lower selective pressure, or recent evolutionary events. These switches nevertheless have important implications for genome evolution. Their effects, and the mechanisms for avoiding them, may in turn shape larger biochemical networks by changing global noise properties, and will be an important factor in designing synthetic networks.
Symbolic manipulations and data analysis were performed in Mathematica 7.0 and 8.0 (Wolfram Research, Champagne, IL). We predicted intrinsic noise characteristics with stochastic simulations at stationary state using the StochKit (http://engineering.ucsb.edu/~cse/StochKit/) tau-leaping routine for 10,000 runs of each model (except in the lac operon model, for which 1,000 runs of each condition were done). Initial model construction and test runs were done with Copasi (www.copasi.org). Simulations with an extrinsic noise representation were done in Copasi as detailed below. All models were represented with elementary reaction steps; in models involving gene regulation, we defined a promoter variable as always present at one copy per cell.
Unless otherwise specified, each network module was tested with promoter-mediated noise, represented by promoters switching between “on” and “off” states of mRNA production. This process has estimated switching rates of kgoff = 0.0028 s−1 for switching to the “off” state and kgon = 0.00045 s−1 for switching to the “on” state [20]. In models including a gene regulation step, we assumed binding and unbinding of regulators to be independent of promoters switching between on and off states. Without promoter-generated bursting, gene expression noise largely arises from low mRNA copy numbers per cell and the effect of transcriptional coupling is qualitatively similar (Figure 1). Furthermore, analytical results from LNA do not include the effects of bursty transcription, showing that we arrive at qualitatively similar results without transcriptional bursts.
We distinguish between three types of coupling between production of two proteins in stochastic simulation reaction schemes. Transcription may be coupled or uncoupled (i.e., proteins in the same or separate operons) and when transcription is coupled, proteins may be cotranslated (single ribosome binding site for both) or translationally uncoupled (two ribosome binding sites). These three cases represent simplified extremes; intermediate translational linkage (e.g., read-through from multiple ribosome binding sites) is possible but was not further considered here. Figure 1 illustrates the three cases with promoter-mediated gene expression noise. For simplicity of presentation, we compare transcriptionally uncoupled with cotranslated models in the main text.
Tables S4, S5, S6, S7, S8 give reaction schematics and parameters for the gene expression and posttranslational models used in the main text. Parameter values were chosen to be of the correct order of magnitude for realistic expression levels and binding kinetics. To ensure a fair comparison between cotranscribed and uncoupled configurations, production and degradation rates of mRNA species for proteins A and B are identical. The degradation rate kdeg corresponds to the value expected from a dilution rate for typical E. coli doubling every half hour.
The basis of noise differences between networks with proteins in the same operon and those with proteins in separate operons is the covariance between the expressed proteins. We used LNA to analytically characterize noise and covariance as follows. For the mean values of copy numbers (denoted by angular brackets):(8)where i = 1 with proteins A and B in the same operon, and i = 2 with proteins A and B in separate operons. Then we solved the fluctuation-dissipation matrix equation at steady state (Mσ+σMT+ΩN = 0) for σ, where M is the Jacobian of the (macroscopic) system, Ω is cell volume, and N is the diffusion matrix [18]. Characterizing intrinsic noise as and with indices i and j taking values corresponding to molecular species (A, B, m1 and m2), we follow the methods of [18] to obtain:(9)and as in Equation 1. Note that σAA is the variance, or the square of the standard deviation.
To analytically approximate noise of physiologically relevant variables in the simple network modules (Figure 2), we made the following simplifications to make the systems tractable. For metabolic steps with a substrate as a dependent variable, we assumed a Michaelis-Menten propensity. For the covalent modification module, we assumed a simple mass-action with no saturation or complexes. For the multiple gene regulator module, we used a Hill equation propensity for regulated mRNA production. Details of the analysis are in Text S2. Mean-field models are given in Table S3.
Bioinformatic analyses used pairs of interacting genes extracted from databases of E. coli K12 MG1655 as described below. To determine the randomized control, we needed to account for potential biases resulting from dataset size and other features of chromosome organization that we were not attempting to test. For instance, if we randomly assigned genes to extant operons in E. coli across the entire chromosome, the less frequently occurring modules would have much less same-operon membership than the modules with larger numbers of members, and would not be a useful control. We chose to randomize the genes extracted from the pairs within each module to set the random control for each class. Thus, for a list of gene pairswe determined a randomized case by flattening g intorandomly permuting the order of the genes and then re-pairing them to determine the frequency. This process was repeated 1000 times to determine the parameters of the randomized distribution.
We extracted single-cell mRNA expression data (RNAseq) and protein copy number data from Taniguchi et al [43]. To ensure a meaningful comparison of expression levels, we considered only genes predicted to be unregulated in RegulonDB. Only the physical interaction module left enough data for analysis. For instance, the number of unregulated pairs in the same operon for the linear metabolic pathway dataset was 5, insufficient to distinguish the established operon membership pattern from noise when partitioned between high and low expression. Each average single-cell mRNA or protein copy number was used, along with physical interaction status (1 = same operon; 0 = non-same operon). Proteins with multiple interaction partners within and between operons were represented twice, once for same-operon and once for non-same-operon interaction. We then divided the set into above- and below-median subsets and compared the fraction of same-operon interactions in the subsets using a standard bootstrap resampling test. We resampled 10,000 times with replacement and computed the difference in coupling frequencies between low and high expression as the test statistic. To compute error bars, we used bootstrapping of each bin by sampling each bin with replacement up to the bin size, repeated 1,000 times.
|
10.1371/journal.pgen.1007408 | The effector of Hippo signaling, Taz, is required for formation of the micropyle and fertilization in zebrafish | The mechanisms that ensure fertilization of egg by a sperm are not fully understood. In all teleosts, a channel called the ‘micropyle’ is the only route of entry for sperm to enter and fertilize the egg. The micropyle forms by penetration of the vitelline envelope by a single specialized follicle cell, the micropylar cell. The mechanisms underlying micropylar cell specification and micropyle formation are poorly understood. Here, we show that an effector of the Hippo signaling pathway, the Transcriptional co-activator with a PDZ-binding domain (Taz), plays crucial roles in micropyle formation and fertilization in zebrafish (Danio rerio). Genome editing mutants affecting taz can grow to adults. However, eggs from homozygous taz females are not fertilized even though oocytes in mutant females are histologically normal with intact animal-vegetal polarity, complete meiosis and proper ovulation. We find that taz mutant eggs have no micropyle. Taz protein is specifically enriched in mid-oogenesis in the micropylar cell located at the animal pole of wild type oocyte, where it might regulate the cytoskeleton. Taz protein and micropylar cells are not detected in taz mutant ovaries. Our work identifies a novel role for the Hippo/Taz pathway in micropylar cell specification in zebrafish, and uncovers the molecular basis of micropyle formation in teleosts.
| In many fish, sperm enters eggs through a specialized channel called the ‘micropyle’. The micropyle is formed by a special follicle cell, the ‘micropylar cell’, which sits on the top of the developing egg during oogenesis, and forms the sperm entry canal. The underlying mechanisms of this process are unknown. We find that Taz, an effector of an important signaling pathway, the Hippo pathway, is specifically enriched in micropylar cells in zebrafish, and regulates formation of the micropyle. Loss of Taz function in females results in no micropylar cells, failure to form a micropyle on eggs, which are consequently, not fertilized. Our study identifies a new role for the Hippo/Taz pathway in cell fate specification in the ovary, and reveals a potential mechanism for forming the sperm entry port. Similar mechanisms might operate in other fish as well.
| In vertebrates, fertilization occurs by two major strategies. Amniotes such as reptiles, birds and mammals, undergo copulation and internal insemination to ensure gamete fusion. The acrosome reaction is necessary for sperm to penetrate the zona pellucida, a protective egg envelope, and entry of sperm can occur at any position in the egg surface [1–3]. By contrast, most teleosts (bony fish) undergo external fertilization. Without a recognizable acrosome reaction, sperm entry in teleosts relies entirely upon a specialized funnel-like structure, the micropyle, in the chorion, an acellular coat of the egg [4–6]. Morphological and physiological studies of the micropyle in a variety of different teleost species suggest that channel formation results from the transformation of a special micropylar cell in mid-oogenesis [7–12]. The micropylar cell is morphologically distinct from other follicle cells surrounding the oocyte. Positioned over the oocyte animal pole, the micropylar cell is bigger in size and appears like an inverted cone in shape, in contrast to the flattened appearance of follicle cells, sometimes called ‘mushroom’-like [11–13]. The unique shape of the micropylar cell is gradually achieved during oogenesis. A cytoplasmic extension from the micropylar cell expands and extends through the developing vitelline envelope, till the extension tip contacts the oocyte membrane, as the vitelline envelope grows and perforation proceeds, the cytoplasmic extension becomes slim and long [14]. Finally, the micropylar cell degenerates, leaving a narrow canal called the ‘micropyle’ between the chorion and the egg [15, 16]. Previous studies in other teleosts revealed potential drilling forces of the micropylar cell. The aggregation and elongation of microtubules and tonofilaments in the cytoplasmic bulge of the micropylar cell provides internal forces [11, 14], and two opposing rotations between the oocyte and the covering follicle cell layer are thought to provide the external force for the micropylar cell to bore through the chorion [13, 17, 18]. Although these studies described the morphological process of micropyle formation, little is known about the molecular mechanisms underlying formation of this essential structure. A key line of evidence comes from studies of a zebrafish maternal-effect mutant bucky ball (buc), in which oocyte polarity fails to be established. This mutant has multiple micropyles in each egg, arising from the expanded animal identity in buc mutant oocytes [19, 20].
Hippo signaling plays a variety of roles in development, regeneration, tissue homeostasis, and stress response [21, 22]. The WW domain-containing transcription regulator protein 1 (Wwtr1) is a transcriptional co-activator with a PDZ-binding domain (Taz). Taz, together with Yes-associated protein (Yap), are downstream effectors of Hippo signaling. As a transcriptional co-activator, Taz usually exerts its functions by binding to transcription factors, such as Teads and Smad2/3, which modulates transcription of downstream genes [23]. As an oncoprotein, TAZ has been found up-regulated in many kinds of human cancers. TAZ also promotes epithelial-mesenchymal transition (EMT), migration and invasion of cancer cells, where cell morphology is altered and cytoskeleton is inevitably rearranged [24]. Yap/Taz can regulate cytoskeleton dynamics. In medaka (Oryzias latipes), Yap regulates cortical actomyosin activity and tissue tension by the downstream Rho GTPase activating protein ARHGAP, and mutants affecting the inhibitor of F-actin polymerization, hirame/yap, display reduced cortical actomyosin tension and a collapsed body shape [25]. Interestingly, similar to medaka hirame, the establishment of posterior body shape is disrupted in zebrafish yap1; taz double mutants [26, 27], which suggests that Yap1/Taz regulates cytoskeleton dynamics. In turn, Yap/Taz can be activated by environmental mechanical signals, for example matrix rigidity, which are usually transduced by the cytoskeleton [28, 29].
Tumors affecting two female organs, breast and ovaries, have been used extensively to study TAZ functions [30, 31]. However, to date, the role of TAZ in normal oogenesis and ovary differentiation has not been investigated. In a study of Taz function in zebrafish, we have unexpectedly found that Taz is required for the formation of micropyle during oogenesis. We show that taz transcripts are expressed maternally. When taz is knocked out, some homozygous taz mutants can survive to adulthood, and mutant females produce eggs with no micropyle. Our results suggest that Taz might regulate micropylar cell specification and morphogenesis during zebrafish oogenesis.
To study Taz function, we knocked out taz by targeting the first exon using CRISPR/Cas9 genome editing and recovered two mutant alleles, tazΔ10 and tazΔ1, both of which produce mutant transcripts that encode truncated proteins with 148 and 145 amino acids, respectively (Fig 1A, 1B and 1C). Importantly, no Taz protein was detected in tazΔ10/Δ10 mutant embryos (Fig 1D), suggesting that the lesion in tazΔ10/Δ10 results in a null mutant. Consistent with a previous report [32], tazΔ10/Δ10 mutant embryos displayed relatively normal morphology with exception of a smaller swim bladder than wild type and weak pericardial edema at 4.5 day post fertilization (dpf) (Fig 1E and 1F). However, it did not lead to embryonic lethality as indicated by the expected incidence of homozygous mutants from intercrosses of heterozygotes (~25% at 7 dpf), and in accordance with Mendelian segregation (Fig 1G). Some tazΔ10/Δ10 mutants could grow into adulthood, although the survival ratio was much lower than expected. Interestingly, while tazΔ10/Δ10 adult males were fertile (Fig 2B and 2E), all embryos from mating of tazΔ10/Δ10 adult females were arrested at the one-cell stage regardless of the genotype of the male (Fig 2C and 2E), and even though the females ovulated normally and produced eggs. We found similar phenotypes with the tazΔ1/Δ1 allele (Fig 2D), and all subsequent studies reported in this work were done using tazΔ10/Δ10 mutants. Since tazΔ10/Δ10 adult females are infertile, this suggests that taz is indispensable for producing normal eggs. Therefore, we surveyed if taz was expressed in oocytes. In situ hybridization showed that taz transcripts were found in the cortex of oocytes and the attached follicle cells (Fig 2F). Moreover, in one-cell stage embryos, where mRNAs are deposited in eggs by the mother, taz transcripts are abundant (Fig 2H), which is consistent with transcriptomic datasets [33] (http://www.ensembl.org/Danio_rerio/Location/View?r=22%3A38049130-38114599). Together, these data reveal that taz, a maternally expressed gene, is essential for fertilization.
To determine the basis of the failure of tazΔ10/Δ10 eggs to progress beyond the one-cell stage, we examined the ovaries and oogenesis in mutant females. Compared with ovaries at the same stage in wild type females, the ovary of an 8-month old taz mutant female was grossly normal in the size, tissue composition and intraperitoneal position (Fig 3A and 3B), and there were no apparent morphological defects in the color, size and shape of oocytes (Fig 3C and 3D). Histological analysis showed that all stages of oocytes (stage I to IV) were present in taz mutant ovaries, and had no obviously difference from that in wild type controls (Fig 3E and 3F), indicating that the oogenesis was largely normal in taz mutants.
The establishment of animal-vegetal polarity in oocytes is a key event during oogenesis, and determines the formation of two major embryonic axes, the dorsal-ventral and left-right axis, in vertebrates [34]. Therefore, we examined if the failure of taz mutant oocytes to be fertilized was due to defects in animal-vegetal polarity. The Balbiani body (Bb) is the earliest vegetal structure in stage I oocytes, and can be marked by the expression of dazl transcripts [35]. In stage I oocytes of taz mutants, the Balbiani body appeared similarly as in wild type oocytes (Fig 3G and 3H). Furthermore, while dazl transcripts were found located in the vegetal pole, cyclinB, an animal pole marker, was distributed on the opposite side of oocytes in both taz mutant and wild type (Fig 3I and 3J), indicating that the animal-vegetal polarity was normally established in taz mutant oocyte. Similarly, expressions of other two polarity markers, pou2 and brl, were not altered in mutant oocytes (S1 Fig). Taking together, we conclude that taz is not necessary for oogenesis or for the establishment of oocyte polarity in zebrafish.
Since oogenesis seemed normal in taz mutant ovaries, next we checked if fertilization of mutant eggs was normal. In teleost eggs, the micropyle is a narrow canal for sperm entry through the chorion during fertilization. While all wild type eggs had a single, animal-pole localized micropyle (Fig 4A), no micropyle was detected in taz mutant (Fig 4B). Furthermore, a single obvious cytoplasmic projection from the plasma membrane to the micropyle was present in wild type eggs shortly after egg activation (Fig 4C), whereas no protrusion was found in taz mutant eggs (Fig 4D). These observations strongly suggest that the lack of micropyle in taz mutant eggs results in their not being fertilized as sperm likely cannot enter the egg.
Once zebrafish eggs are activated, the second meiotic division is quickly completed, and the second polar body is extruded [36], a hallmark of the completion of the meiosis. To examine if the failure of fertilization of taz mutant eggs is due to no sperm entry, we performed DAPI and Phalloidin staining in activated eggs. While the pronucleus (from egg or sperm) is only stained by DAPI, the polar body from egg, surrounded by Actin, is detected by both DAPI and Phalloidin. After fertilization, sperm DNA enters the egg, and two pronuclei, from the egg and sperm, and one polar body were found in wild type eggs (Fig 4E and 4E’). However, in taz mutant eggs, a polar body and only one pronucleus were observed (Fig 4F and 4F’), indicating that meiosis is complete but there is a lack of sperm entry. These data demonstrate that oocyte meiosis can be completed without Taz, and the failure of fertilization in taz mutant egg is due to lack of the micropyle.
Our analysis suggests that oogenesis in taz mutant appears normal except for the lack of micropyle formation. In addition to oocytes, follicle cells are another group of cells that are essential for oogenesis to progress. In teleost eggs, follicle cells surround oocytes to provide nutrition. Some follicle cells specify into unique micropylar cells, which form one micropyle on each oocyte during stage III oogenesis in zebrafish [12, 37].
To assess follicle cells during oogenesis, wild type and taz mutant ovaries were sectioned and stained with haematoxylin and eosin (HE). Compared with wild type, in taz mutant ovaries, follicle cells surrounding oocytes of all stages had no obvious difference in size, shape or numbers (Fig 3E and 3F). Interestingly, while follicle cells around oocytes showed basal levels of Taz expression, one particular cell was found highly enriched with Taz from late stage II to late stage III oogenesis (Fig 5A, 5B and 5C). This cell became larger than other follicle cells, and displayed a unique morphology change from flattened to ‘nail’-like shape (Fig 5A’, 5A”, 5B’, 5B”, 5C’ and 5C”). The micropylar cell depressed and eventually perforated the developing vitelline envelope (Fig 5A’”, 5B’” and 5C’”). Referring to morphological criteria, this Taz-enriched cell is the micropylar cell. Notably, Taz was predominantly distributed in the nucleus of micropylar cells, and the levels gradually decreased with progression of micropylar cell development (Fig 5A, 5B and 5C). The nuclear localization of Taz suggests that it might exert its function by transcriptional regulation of target genes. Moreover, high levels of Taz were found in the tip of cytoplasmic extension of the micropylar cell, especially in middle stage III oocytes (Fig 5B and 5B”). In sectioned wild type ovaries, the high Taz expressing micropylar cell is located on the top of the animal pole of oocyte marked by cyclinB (Fig 5D, 5D’ and 5D”). However, in taz mutant ovaries, neither the micropylar cell nor the invagination on the developing vitelline envelope was detected (Fig 5E, 5E’ and 5E”). We also performed Taz immunostaining in whole mount oocytes, and detected a single high Taz expressing micropylar cell on the top of animal pole in wild type stage III oocytes, but not in taz mutants (Fig 5F and 5G). These data suggest that Taz is required for the specification of micropylar cell, and the enrichment of Taz in micropylar cell agrees with an indispensable role for Taz in micropyle formation.
Interestingly, in sections of wild type ovaries, the shape of the micropylar cell nucleus sometimes looked like two closely juxtaposed nuclei (Fig 5A). To examine the micropylar cell membrane and nucleus in detail, we performed co-immunostaining with Taz and β-Catenin in whole oocytes, while DAPI was used to label DNA. During oogenesis between late stage II to late stage III, two DAPI signals in close proximity within one cell are identified in almost all micropylar cell nuclei (Fig 6A–6C’”), being readily detected in late stage II/ early stage III oocytes, and gradually fading in late stage III oocytes. Co-labeling with an antibody towards Nup107, a nuclear pore marker, also showed two lumps of DAPI signals surrounded by a continuous nuclear membrane in micropylar cells from late stage II to late stage III oogenesis (Fig 6D–6F’”). However, we did not find clearly separated nuclei in all the oocytes (n = 132) that we assessed.
In many teleosts, formation of the micropyle is thought to require drilling of the vitelline envelope by the micropylar cell. During this process, the micropylar cell shape undergoes extensive changes [11, 13, 14], and the cytoskeleton might participate in this process. To assess the possible role of Taz in regulating cytoskeletal changes during micropyle formation, we performed co-staining of Taz with F-actin or α-Tubulin in wild type oocytes. We found that Actin filaments were enriched at the leading edge of oocyte cortex and the leading tip of micropylar cell, towards the indentation (Fig 7A). As oocytes mature, more Actin filaments were found deposited (Fig 7B and 7C, S2 Fig). Tubulin was also enriched in the cytoplasm of the micropylar cell, and in the cytoplasmic extension into the vitelline envelope (Fig 7D, 7E and 7F, S2 Fig). Considering the role of Yap1/Taz in regulating cytoskeleton in medaka and zebrafish [25, 27], the high expression of Taz and dynamic Actin and Tubulin in the micropylar cell suggests that Taz may regulate cytoskeletal arrangements during formation of a functional micropyle.
Taken together, we have revealed a unique function of Taz in formation of the micropyle in zebrafish which is summarized in a model (Fig 8). In oocytes from late stage II to late stage III, the micropylar cell, sitting on the animal pole, becomes bigger and changes into a ‘nail’ shape. Taz is highly expressed in the micropylar cell. F-actin is deposited in the leading tip of the micropylar cell and the leading edge of oocyte cortex, and Tubulin is enriched in the micropylar cell cytoplasm and protrusion into the vitelline envelope. The dynamic cytoskeleton might facilitate perforation of the developing vitelline envelope. Without Taz, the micropylar cell is not specified, and no micropyle forms in taz mutant eggs.
The most interesting finding in this study is that mutations affecting Taz, a key effector of the Hippo signaling pathway lead to loss of a cell required for formation of the micropyle, the sperm entry port on eggs. Our findings identify the first molecular component in the establishment of this unique cell in zebrafish ovary. In taz mutant, loss function of Taz does not affect ovary and oocytes development, egg ovulation and the second meiosis of oocyte, but leads to failure of formation of the micropyle. Such a specific phenotype is due to the restricted high expression of Taz in the micropylar cell.
The high expression of Taz in one particular follicle cell at the animal pole in mid-oogenesis identifies Taz as the first molecular marker for the micropylar cell. With the aid of high Taz expression, the micropylar cell is easy to be identified. Besides the known characteristics, such as the big size and unique shape [12], most micropylar cells are found to have bilobed nuclei. These findings raise several interesting questions to be addressed in the future: Is DNA segregation incomplete in micropylar cells or are there two closely juxtaposed nuclei in micropylar cells? What leads to this: is this owing to incomplete cell division, cell fusion or proliferation? In preliminary experiments, we did not detect any pH3 signal, a maker for G2/M cell cycle phase, in the micropylar cell (S3 Fig), suggesting that cell proliferation probably does not underlie the bilobed nuclei. We also observed that there is a gradual down-regulation of Taz in the developing micropylar cell, with high expression levels of Taz in micropylar cells during middle oogenesis and lower levels from late stage III onwards. One possible explanation is that the signals that maintain Taz expression might be decreased as micropylar cell development progresses. It is also possible that Taz is not required when the micropylar cell becomes mature and finally degenerates.
Our study has demonstrated that Taz is required for micropylar cell specification from a follicle cell. At this stage, it is hard to distinguish if the high expression of Taz is a cause or consequence of micropylar cell specification. In a parallel study, Dingare et. al. demonstrate that Taz is required for micropylar formation in zebrafish, which agrees with our conclusion, and they also find that Taz is highly expressed in ectopic micropylar cells formed in buc mutant oocytes [38]. This suggests that once the micropylar cell is determined by other factors, it will express Taz. However, we cannot exclude the possibility that Taz is induced first. Identification of signals that induce Taz expression in follicle cells will help to address if Taz is a cause or consequence of micropylar cell specification. Besides upstream components in Hippo pathway which regulate Taz stability [21, 23], two aspects of oogenesis, which precede micropylar cell specification, need attention. One is the establishment of animal-vegetal polarity of the oocyte. It is widely accepted that follicle cells close to animal pole of oocyte contribute towards micropyle formation in many teleosts [14–16], suggesting that i) the animal pole determines the group of follicle cells competent for micropylar cell specification, and ii) animal pole specific-mRNAs could be inducers of Taz. The second is the growth of oocyte. Yap/Taz are known to act as mechanosensors [28, 29]. The volume expansion of oocytes may produce mechanical signals and activate Taz. These events, prior to micropylar cell specification, might induce Taz expression. Nonetheless, how a single follicle cell acquires micropylar cell fate is not clear. Inducible loss-of-function and overexpression of taz could address if Taz is a cause or consequence of micropylar cell specification. Lineage tracing in wild type and taz mutant ovaries, combined with single-cell gene expression profiling might also be informative [39–42].
Although we cannot identify if Taz is a cause or consequence of micropylar cell specification, our data reveal an essential role for Taz in this process. How does the micropylar cell exert its function by Taz expression? The most dramatic behavior of the micropylar cell is the shape change from a follicular epithelium into a highly polarized cell with a prominent projection, a process that is overtly similar to EMT in cancer. Besides EMT, Taz, as an oncoprotein, also promotes migration and invasion of human cancer cells, where cell shape changes are prevalent [24]. The micropylar cell is thought to bore through the developing vitelline envelope to form a channel, a process during which the cell shape must change greatly. Both cancer and micropylar cells are dynamic in shape, and therefore, it is reasonable to speculate that Taz works in a similar way in both processes. Yap/Taz regulates the cytoskeleton [25, 27], and nuclear localization of Taz in the micropylar cell may transcriptionally regulate Actin and Tubulin to drive morphogenesis of the micropylar cell, although the direct target downstream genes are unknown yet. In support of this possibility, a previous study in medaka showed that bundles of microtubules and tonofilaments are formed and elongated in the protruding cytoplasm of the micropylar cell during its penetration of the developing vitelline envelope [13]. In addition to its expression in the nucleus, Taz is also expressed in the cytoplasm of the micropylar cell, and enriched in the leading tip of cytoplasmic extension, where F-actin is extremely abundant. It is worthy of investigating if cytoplasmic Taz regulates the cytoskeleton, and the mechanism of regulation. In experiments to examine if the cytoskeleton is required for micropylar cell maintenance, Latrunculin B or Blebbistain was used to transiently inhibit Actin polymerization and Myosin II ATPase activity, respectively. Both inhibitors don’t have obvious effects on the morphology of micropylar cells (S4H” and S4I” Fig). However, dissociation of F-actin results in delocalization of Taz from nucleus to cytoplasm in the micropylar cell (S4H” Fig), while perturbation of Myosin II does not (S4I” Fig). These results are similar to observations in mammalian cell culture [43], indicating the cytoskeleton is required for maintenance of nuclear localization of Taz.
The first molecular evidence of regulation of micropyle formation comes from studies in a zebrafish mutant bucky ball, which have revealed that proper animal-vegetal polarity of the oocyte is essential for micropyle formation [19, 20]. In zebrafish buc mutant oocytes, the vegetal Balbiani body never forms, leading to an expansion of animal pole-specific gene expression (e.g. vg1) and multiple micropyles form in buc mutant eggs [19, 20]. Previous studies also found that extra territories of vg1 transcripts coincide with the locations of ectopic micropylar cells in buc mutant oocytes [19]. By contrast, in taz mutant oocytes, animal-vegetal polarity is normal and yet, no micropyle forms. Therefore, the polarity of the oocyte alone is insufficient to determine micropyle formation, and additional mechanisms must govern micropyle cell fate. Our work identifies a new view of regulation during specification of this cell, and shows that follicle cells at the animal pole induce the formation of the micropyle in a Taz-dependent manner.
Zebrafish (Danio rerio) were raised and maintained in the fish facility in accordance with standard procedures [44] under approval from the Institutional Review Board of Southwest University (Chongqing, China).
ABtü strain and subsequently generated taz mutant lines (tazΔ10/Δ10 and tazΔ1/Δ1) were used in this study. Embryos or oocytes were collected and staged as described [12, 45].
Embryos (or tail fin clips) were lysed in the lysis buffer (10 mM Tris pH 8.2, 50 mM KCl, 0.3% Tween-20, 0.3% Nonidet P40, 0.5 μg/μl Proteinase K (Fermentas)) at 55°C for 14 hours, followed by enzyme inactivation at 94°C for 20 minutes.
The target sequence of taz gRNA, 5’-GGAGTCTCCCGGGGCTCGG-3’ (PAM site underlined), was located in exon 1 of zebrafish taz gene. Zebrafish Cas9 mRNA and the taz gRNA were synthesized respectively according to the descriptions [46, 47]. After ZCas9 mRNA (300 pg) and taz gRNA (50 pg) co-injection into one-cell stage wild type embryos, the lysate of 10 embryos at 24 hour post fertilization (hpf) was used as template for PCR with primers taz fw (5’-AGACCTGGACACGGATCTGGA-3’) and taz rv (5’-CACTGTATGCACTCCACTAACTGGT-3’). PCR products were sequenced to examine potential indels created in the taz gRNA target region. Embryos co-injected with functional taz gRNA and ZCas9 mRNA were raised to adults (F0). F0 fish were screened to identify founders with progeny harboring the indels in taz gene previously found. Offsprings of identified F0 were raised. Individual F1 adults was reconfirmed by PCR using genomic DNA from tail fin clips, and indel types in fish were determined by sequencing.
To detect taz Δ10 genotype, primers were designed to amplify specific bands by PCR with a common primer, taz fw2 (5’-CGATCGGACGCAGGAGGAACAA-3’), and two reverse primers, taz wt rv (5’-CGGGTGTGGGAGTGGAGTC-3’) and taz Δ10 rv (5’- CGGGTGTGGGAGTGGAGCT-3’). For taz Δ1 genotyping, the above taz fw and taz rv primers were utilized to obtain PCR products for sequencing.
For preparation of zebrafish protein samples, 5 dpf embryos were homogenized in cold PBS with protease inhibitors (Roche) using syringe (1 ml) and needle (size 23G). The deyolked body fragments were collected and heated in whole cell lysis buffer (20 mM NaF, 1 mM DTT, 1 mM EDTA, 0.1 mM Na3VO3, 10% glycerol, 0.5% Nonidet P40, 280 mM KCl, 20 mM Hepes pH7.9) at 100°C for 10 minutes. Lysate supernatant was used for western blot analysis according to the standard protocol [48]. In this study, primary antibodies, anti-Taz (CST, 1:1000) and anti-β-Tubulin (Thermo, 1:1000) were used, while anti-mouse-IgG-HRP (Thermo, 1:5000) and anti-rabbit-IgG-HRP (Thermo, 1:5000) worked as secondary antibodies.
Adult females were euthanized by overdose tricaine treatment according to the guidelines of experimental animal welfare from ministry of science and technology of People’s Republic of China (2006), and abdominal tissue are removed by sharp scissors. Images were taken under a stereo microscope. Dissected ovaries were fixed in 4% PFA under room temperature for 2 hours, followed by images acquisition.
Wild type and mutant ovaries were dissected from 8 month old females, and fixed overnight in saturated picric acid at room temperature. Fixed tissues were embedded in paraffin and sections were collected at 5-μm thickness using a microtome (Leica). Haematoxylin and eosin staining was performed according to standard protocol.
Ovaries were dissected from adult abdomen and fixed in 4% PFA at room temperature for 2 hours. After overnight immersion in 30% sucrose in PBS at 4°C, the tissues were embedded in O. C. T. compound (Sakura) and frozen in ethanol at -80°C. Frozen tissues were sectioned at 10-μm thickness using a Cryotome (Leica). Serial sections were used for in situ hybridization as described previously [49]. Antisense RNA probes cyclinB [37, 50] and pou2 [37, 50], labeled by fluorescein and digoxigenin, respectively, were used for marking animal poles in oocytes, while digoxigenin labeled dazl [35, 51] and brl [52, 53] were used to indicate vegetal poles. Whole mount and section in situ hybridization were performed according to methods used in a previous report to examine gene expression patterns of taz [54].
A nonspecific protein-staining dye, 2% Coomassie Brilliant Blue R (CB), was dissolved in DMSO. Prior to staining, the stock buffer was diluted in PBS (1:10). Eggs were collected in 5 minutes after activation, stained for 3 minutes, and rinsed thoroughly in PBS [55]. Stained eggs were examined and photographed under a stereo microscope.
For oocyte activation, Stage V oocytes were gently extruded from adult females, and activated by water. For in vitro fertilization, sperms were collected from adult males into Hank’s buffer and performed fertilization according to standard procedure [44].
For Immunohistochemistry, dissected ovaries were fixed in 4% PFA for 2 hours at room temperature, and oocytes were isolated in PBS by sharp forceps. For in vitro oocyte culture, dissected ovaries were gently dissociated by a Pasteur pipette in 90% L15 (pH9.0) medium (Gibco) with 0.5% BSA (Sigma).
Isolated oocytes were cultured in 90% L15 (pH9.0) medium with 0.5% BSA and 2ug/ml 17α-DHP (Sigma) at 28.5°C for 8 hours. Latrunculin B (Santa Cruz) or (-)-Blebbistain (MedChem Express), dissolved in DMSO, were supplemented into culture medium to final concentrations at 7.6 μM and 600 μM to inhibit Actin polymerization and Myosin II ATPase activity respectively.
The frozen sections of ovaries were prepared as mentioned above. Ovary sections and oocytes were performed for immunohistochemistry as described previously [49]. Anti-Taz (CST; 1:200), anti-β-Catenin (Sigma; 1:200), anti-Nup107 (BioLegend; 1:400) and anti-α-Tubulin (Sigma; 1:200) were used as primary antibodies, and subsequent visualization was achieved by the application of secondary antibodies Alexa Fluor 488, Alexa Fluor 555 and Alexa Fluor 647 (Life Technology; 1:400). A solution of 4% BSA in PBS was used for blocking and diluting antibodies. FITC-Phalloidin (Sigma, 1:200) was used to detect F-actin. In some experiments, the animal pole was first labeled by in situ hybridization using cyclinB probe, followed by immunohistochemistry according to standard procedure. Before covering with Vectashield (Vector lab), DAPI (Roche) was employed to stain the nuclei. Images were acquired on a Zeiss LSM700 confocal microscope. Brightness of green fluorescence was slightly digitally enhanced to clearly show cytoplasmic expression of Taz on sectioned late stage III oocytes (see Figs 5C, 7C’ and 7F’).
|
10.1371/journal.pcbi.1006031 | Cytosolic proteins can exploit membrane localization to trigger functional assembly | Cell division, endocytosis, and viral budding would not function without the localization and assembly of protein complexes on membranes. What is poorly appreciated, however, is that by localizing to membranes, proteins search in a reduced space that effectively drives up concentration. Here we derive an accurate and practical analytical theory to quantify the significance of this dimensionality reduction in regulating protein assembly on membranes. We define a simple metric, an effective equilibrium constant, that allows for quantitative comparison of protein-protein interactions with and without membrane present. To test the importance of membrane localization for driving protein assembly, we collected the protein-protein and protein-lipid affinities, protein and lipid concentrations, and volume-to-surface-area ratios for 46 interactions between 37 membrane-targeting proteins in human and yeast cells. We find that many of the protein-protein interactions between pairs of proteins involved in clathrin-mediated endocytosis in human and yeast cells can experience enormous increases in effective protein-protein affinity (10–1000 fold) due to membrane localization. Localization of binding partners thus triggers robust protein complexation, suggesting that it can play an important role in controlling the timing of endocytic protein coat formation. Our analysis shows that several other proteins involved in membrane remodeling at various organelles have similar potential to exploit localization. The theory highlights the master role of phosphoinositide lipid concentration, the volume-to-surface-area ratio, and the ratio of 3D to 2D equilibrium constants in triggering (or preventing) constitutive assembly on membranes. Our simple model provides a novel quantitative framework for interpreting or designing in vitro experiments of protein complexation influenced by membrane binding.
| In a multitude of cellular processes, including cell division and endocytosis, proteins must bind to one another to form large multi-protein complexes. To initiate the formation of these critical multi-protein assemblies at the right time and the right place, the constituent proteins must be present at sufficient concentrations. We show here that membrane localization offers a powerful way of controlling protein concentrations by reducing the dimensionality of the protein’s search space. We present a simple and practical analytical theory that determines the significance of membrane localization for triggering protein-protein interactions. We show that protein binding partners will often form substantially more complexes when both partners can localize to surfaces, and thus localization can regulate the timing of multi-protein assembly. We collect in vitro binding data and cellular concentrations of proteins and lipids involved in pathways including clathrin-mediated endocytosis to demonstrate how cellular proteins could exploit membrane localization to regulate assembly.
| When clathrin, the essential cytosolic protein of clathrin-mediated endocytosis (CME), self-assembles into multi-protein cages, the same protein-protein contacts are used regardless of whether clathrin is in solution or on the membrane [1–3]. However, more binding [2] is observed on the membrane. A fundamental phenomenon for explaining this change is dimensionality reduction: if proteins on the membrane search a smaller space, then this increases their relative concentration; higher concentration of proteins (reactants) shifts the equilibrium to produce more protein-protein complexes (products) as defined by LeChatelier’s principle. The question we address is, how significant a role can this dimensionality reduction play for driving protein-protein interactions between cytosolic proteins in vitro? Understanding this role can help interpret mechanisms of assembly in vivo. Despite the wide-ranging cellular processes such as cell division and viral budding that could exploit this phenomenon, it has so far lacked a predictive theoretical framework. Hence while the concept that membrane localization can enhance binding may be familiar or intuitive, we here make that concept quantitative for soluble binding partners. In contrast, theory for understanding reduction of dimensionality in chemoreception and receptor mediated signaling (where it can also be functionally significant [4]) has been studied for decades [5, 6]. Membrane localization can accelerate a ligand’s search for membrane bound targets [5–9] and increase activation of intracellular receptors, influencing downstream response [8–10]. However, in these cases, a soluble protein always targets a membrane bound receptor. Here we capture the dynamic cases where both binding partners are soluble and target lipids present in limited concentrations, as occurs, for example, in CME. Our theory determines how binding enhancement depends on protein and lipid concentrations, protein-protein and protein-lipid affinities, the volume-to-surface area ratio, and the change in binding affinities from 3D to 2D. Quantifying this behavior is critical to understanding assembly on surfaces because 2D localization can strengthen binding reactions regardless of whether additional factors, such as curvature generation [11], membrane microdomains [12, 13], or conformational switches [1], also influence binding.
We show here that membrane localization offers a powerful way of controlling protein concentrations and therefore of regulating the timing of multi-protein assembly. In many cases, we find that the power of membrane localization to drive binding is highly robust; strong and weak protein-protein interactors, at high or low concentrations, will all benefit significantly from membrane localization. The analytical theory we present describes a relatively simple model at equilibrium where a pair of soluble binding partners can form complexes in solution and also can both bind and continue to form complexes on the surface of a membrane (Fig 1). Thus it is useful as a tool to quantify protein-protein interactions that, while physiologically relevant, are being studied in vitro. Without accounting for the complex array of factors present in vivo, such as variability in membrane composition, competition for protein and lipid binding from diverse proteins, spatial distributions of proteins or lipids, and non-equilibrium dynamics, we can only speculate about the behavior in the cell. However, the theory provides a novel and valuable metric for interpreting how important membrane localization can be given the concentrations and binding properties of component proteins, and it isolates the role of membrane localization from other factors. Since even most in vitro experiments contain more components and complexity than is captured in our simple model, we discuss how it can still be used as a quantitative guide for estimating how membrane heterogeneity, competition for binding, and mutations would influence the parameters of the model (volume, surface area, binding affinities and concentrations) and thus the proteins’ subsequent response to localization. We specifically address in our results how lipids such as PI(4,5)P2 can be targeted by many proteins at any time [14, 15], how some membrane binding domains such as BAR domains bind membranes with widely varying lipid composition and in a curvature dependent manner [16, 17], and how mutations and multiple protein binding partners would alter protein complex formation. Despite the limitations of applying an equilibrium theory to understand complexation that ultimately occurs in the nonequilibrium cell, we believe the theory represents a well-defined starting point from which to probe more complex systems, just as using in vitro studies provide a useful guide for interpreting behavior in the cell. It is also a reference point for studying the time-dependence of assembly through computer simulation, as we do here, and a starting point from which to build further complexity into the model.
We apply the theory here to characterizing, within a quantitative framework, the role of membrane localization for enhancing 55 binding interactions involving 33 distinct protein pair interaction sets (S1 Table). Through simulation, we also move beyond the model illustrated in Fig 1 of only pairs of soluble binding partners to show how complexation involving non-membrane binding scaffold proteins such as clathrin, or how formation of higher-order oligomers, which is functionally important for driving membrane remodeling [11, 17, 18], can also be regulated by membrane localization (13 additional interaction sets in S2 Table). Our theory only applies to the pair interactions illustrated in Fig 1. We include 22 proteins involved in CME in both human and yeast cells, as well as 15 proteins involved in lipid regulation, vesicle formation on endosomes, budding, and morphogenesis in yeast cells (Table 1). We collected concentration and cellular geometry data based on in vivo values to better connect to physiologic regimes (S3 Table and S4 Table). Although our theory represents an approximate solution to the full model shown in Fig 1, we show through extensive simulations using both systems of ordinary differential equations and single-particle reaction-diffusion, [19, 20] that it is highly accurate. Through simulation, we additionally find that membrane localization alters the timescales of protein-protein assembly, but that the result is not dominated by changes in protein diffusion between solution and the membrane. Rather, for physiologic binding strengths, the rate-limiting step is the speed of binding to the membrane surface from solution. Finally, a practical application of our simple formula is that it can be used to experimentally fit protein-protein binding affinities on surfaces (Ka2D), which are rarely measured [21, 22]. The advantage of the formula is that it applies to in vitro experiments where a pair of proteins can reversibly bind to the membrane, thus avoiding the need to restrict proteins to the surface.
In our primary model, we consider two proteins P1 and P2, that bind in solution with equilibrium constant KaPP,3D=[P1P2]eq[P1]eq[P2]eq=[P1P2]eq([P1]0−[P1P2]eq)([P2]0−[P1P2]eq), where total concentrations of the proteins are [P1]0 = [P1P2] + [P1] and same for [P2]0. If these proteins can also reversibly bind to membranes via targeting a specific lipid M, and continue to bind one another, their binding equilibrium will shift as a total of nine distinct species can form (Fig 1, Methods). The bound protein-protein complexes can either be in solution or on the membrane, [P1P2]sol+mem = [P1P2]sol + [P1P2]mem = [P1P2]+[P1P2M]+[MP1P2] [MP1P2M], and unbound species are similarly defined [P1]sol + [P1]mem = [P1]+[MP1], and [P2]sol + [P2]mem = [P2]+[P2M], where M indicates a copy of a target membrane lipid bound to P1 or P2. The model thus assumes each protein binds membrane via targeting a single copy of a specific lipid type. Proteins on the membrane must be able to diffuse to bind one another, which is consistent with experimental observations [23] even of RNA-protein complexes (>6000kDa) that are anchored via multiple lipid binding sites along with myristoyl groups [24]. Each of the nine distinct species will be constrained to preserve detailed balance at equilibrium, as defined by the 10 pairwise binding interactions of Fig 1 (see Methods), and the total concentrations of proteins is fixed at the same values as above, but now [P1]0 = [P1P2]sol+mem+[P1]sol+mem, and the same for [P2]0. Similarly, [M]0 = [M]+[MP1]+[P2M]+[P1P2M]+[MP1P2]+2[MP1P2M]. We note that species on the membrane have concentrations normally of μm-2, matching the units of equilibrium constants in 2D (Ka2D)-1. All species copy numbers, whether on or off the membrane, however, can be solved for in volume units when the appropriate Solution volume/Membrane surface Area (V/A) conversion factor scales the 2D binding constants, so we always report volume units for concentrations. To quantify the change in bound protein-protein complexes as a function of membrane localization we will define an effective equilibrium constant
Kaeff=([P1P2]eqsol+[P1P2]eqmem)([P1]eqsol+[P1]eqmem)([P2]eqsol+[P2]eqmem)=([P1P2]eqsol+mem)([P1]0−[P1P2]eqsol+mem)([P2]0−[P1P2]eqsol+mem).
(1)
This is not a true equilibrium constant, as both bound and unbound states as defined above contain several species that do not all stepwise interconvert with one another. However, from Kaeff and initial protein concentrations [P1]0 and [P2]0, one can immediately solve for bound complex concentration using Eq 1. If proteins cannot bind to the membrane, the value of Kaeff will revert to the solution bound value, KaPP, and thus the ratio of Kaeff / KaPP determines the extent to which membrane localization either enhances or diminishes protein-protein complex formation. As we discuss further below, in the extreme limits where all proteins are either in solution or on the membrane, Kaeff reduces to a true equilibrium constant. The strength of Kaeff is that it also quantitatively describes all the conditions in between these limits. Thus, our Kaeff definition offers a valuable metric for quantifying the equilibrium of the model in Fig 1, which must otherwise be defined by multiple quantities.
We derive below an exact expression for Kaeff based on the 10 individual equilibrium relations for each reversible binding reaction (Fig 1, Methods). The value of Kaeff for any protein pair will depend on volume V, surface area A, total protein [P1]0, [P2]0, and lipid concentrations [M]0, and all six true equilibrium constants between protein and lipid interactions in 3D (KaPP,KaP1M,KaP2M) and in 2D (Ka2D,PP,Ka2D,P1M,Ka2D,P2M). Importantly, the Ka2D values (with units μm2/mol, e.g.) are different from the corresponding 3D values, but they are related through Ka2D = Ka3D/(2σ). The variable σ, with units of length, is a thermodynamic property of each binding pair that captures changes to binding free energy as a result of surface restriction and changes in standard state units. Changes to free energy are largely entropic, due to altered rotational freedom and protein flexibility [22], although limitations on the orientation of the binding interfaces could alter the enthalpy. The bending rigidity of the membrane can also affect σ, by controlling the relative orientations that the binding pairs can adopt [25]. The variable σ thus represents an independent variable that is specific to each protein pair studied. It is possible that even if concentrations increase on the membrane surface, a decrease in Ka2D will cause less complex formation, and we quantify this regime in the Results section. To keep track of these distinct 3D and 2D equilibrium constants, we explicitly retain the 2D superscript for 2D binding, otherwise Ka (including Kaeff) describes a 3D constant. To derive a simple analytical expression for Kaeff, we input the pairwise equilibria (Methods) into Eq 1 and after canceling terms, we use the equilibrium expression
KaPnM=[PnM]eq[M]eq[Pn]eq,
(2)
with n = 1 or 2, to complete the derivation (Methods).
Our main result is then a surprisingly simple and exact analytical relationship that quantifies the equilibrium solution of our model (Fig 1) via Kaeff and the enhancement relative to KaPP.
KaeffKaPP=γKaP1MKaP2M([M]eq)2+(KaP1M+KaP2M)[M]eq+1(1+KaP1M[M]eq)(1+KaP2M[M]eq),
(3)
where γ is a dimensionless constant V/(2Aσ), σ = KaPP/2Ka2D,PP, [M]eq is the unbound lipid concentration at equilibrium, and all equilibrium constants (including Kaeff) and concentrations are in volume units (Fig 2). [M]eq is a function of all the model parameters, and can only be solved exactly using numerical methods (Methods); we therefore derive an additional approximate analytical equation for [M]eq described below (Fig 2B). However, in the regime where lipids are in excess, the result of Eq 3 is particularly simple because [M]eq~[M]0, the total concentration of lipids (Fig 2A). Critically, this means that the initial experimental conditions then directly determine the enhancement. In this regime only two factors control enhancement, the ratio V/(2Aσ), and the dimensionless strengths of membrane localization, KaPnM[M]0, which report the ratio of membrane bound versus solution proteins (Eq 2) and which we term the membrane stickiness. Hence the volume-to-surface-area ratio, KaPP/Ka2D,PP, and membrane stickiness play a primary role in triggering (or preventing) constitutive assembly on membranes. The right-hand side of Eq 3 is also constant for all KaPP values (Fig 2A). In this regime, our Eq (3) can also be applied to extract binding affinities on membranes (Ka2D) from experiments where binding occurs both on membranes and in solution. This practical application of our result should help simplify the relatively rare experimental characterization of protein-protein affinities on surfaces, as the proteins need not be restricted to the surface for it to work. The condition of excess lipids can be satisfied even with a lipid recruiter such as PI(4,5)P2, present at 2.5x104μm-2 in the plasma membrane [15], or ~1% of lipids [13] (Fig 2A), as we explore further below for proteins involved in CME.
We derive an additional approximate equation for [M]eq to provide a complete equilibrium theory of complex formation applicable to all experimental regimes, and we validate this equilibrium theory through extensive simulations of ordinary differential equations (ODEs) (Fig 2 and Fig 3, Methods). To briefly outline the derivation, we consider two limiting conditions for localization to the membrane: either there are no protein-protein interactions (KaPP = 0), giving [M]eq0, or complete protein-protein complex formation (KaPP = ∞), giving [M]eqCoop (Fig 1D and 1E). These two bounds are indicated by dashed lines in Fig 2B and both limits are independent of KaPP. We can continuously interpolate between them (Fig 2B) using the definition
[M]eq=[M]eq0(1−λ)+[M]eqCoopλ,
(4)
where [M]eq0 is the root of a quadratic equation, [M]eqCoop is the root of a cubic equation, and λ is the function of KaPP (also the root of a quadratic) that smoothly interpolates between them (see Methods for detailed derivations). This theory then provides a complete description of the equilibrium concentrations of all species, as from Kaeff, one can directly calculate the total complexes formed. The partitioning of complexes between solution and the membrane can subsequently be derived utilizing the equilibrium relations of Fig 1 (S1 Text section 2A). The larger the enhancement, the more complexes must be on the membrane (S1E and S1F Fig).
Using our main result, Eq 3, one can predict when and by how much membrane binding will enhance complex formation of binding pairs without performing any simulation or experiment. Further, we can assess whether this enhanced complex formation is robust to perturbations in binding affinities or concentrations. To establish possible values for Kaeff, we first ask: are there any cases where membrane binding will reduce protein complexation, i.e. Kaeff<KaPP? To answer this, we consider the case where all proteins are on the membrane (KaPnM[M]0→∞), such that we have pure 2D binding and Kaeff = γKaPP. Reduced protein complexation will occur only if γ<1, or V/A<2σ. The size of σ controls the relative strength of Ka2D,PP vs. KaPP and varies from one binding pair to another, but experiment and theory indicate it is of the nanometer length-scale [21, 22, 26]. We therefore collected the V/A ratios for a wide range of cell types and organelles to illustrate that in nearly all cells, V/A>2σ (~20nm) and thus membrane localization will enhance binding (S2 Fig). Indeed, targeting the plasma membrane in most cells results in γ values much greater than 1, in the range 10–1000 (S2 Fig). Ultimately, the V/(Aσ) ratio is absolutely central in controlling observed enhancement, as it sets the maximum achievable Kaeff. In most cases proteins will end up mixed between solution and membrane, and from Eq 3 this gives us KaPP≤Kaeff<γKaPP (Fig 2). In the cases where only one protein binds to lipids, all 2D localization benefits are lost and no enhancement occurs: Kaeff = KaPP (Fig 2E).
An important feature that our analysis captures is the coupling that emerges between protein-protein affinity and protein-lipid binding due to membrane localization. If a bound protein-protein complex localizes to the membrane by binding one lipid, binding to a second lipid then becomes a 2D rather than a 3D search [12] (Fig 1D and 1E). Thus, stabilization of proteins on the membrane is achieved not only through strong protein-lipid interactions, but by feedback from strong protein-protein interactions. This cooperative effect for lipid binding (binding of one lipid changes the affinity for the second lipid) produces the unexpected result that the number of proteins bound to the membrane is dependent on the protein-protein interaction strength (Fig 2B). To contrast, one could instead consider lipid binding as simply partitioning proteins between solution and membrane, after which they form complexes as in Fig 1A and 1C. This simplified interpretation, shown in gray dashed lines in Fig 2, does not capture the cooperative effect (Fig 1D and 1E) and is clearly wrong for strong binding proteins that rely on cooperativity to stabilize complex formation (see S1 Text section 2D for details).
With our theory, it is possible to directly probe how changes in cell geometry, binding affinities, or concentrations will regulate enhancement. Although the model is too simple to fully describe multi-component assembly, by varying the input parameters to mimic competing cytosolic factors, one can evaluate the relative importance of concentration, affinities, and geometry. This is particularly true of equilibrium in vitro experiments. For changes to geometry, we first note that the equilibrium results depend only on the ratio and not the absolute size of V or A. The membrane does not have to surround the solution volume like the plasma membrane but can reflect binding to the outsides of liposomes, for example, which allows for studying any V/A ratio. Although it may seem that increasing V/A (through γ in Eq 3) always increases Kaeff, this is not the case when [M]0 is kept constant in its natural units of μm-2 (it is then converted into Volume units by multiplication by A/V). V and A thus also control the initial copy numbers of proteins and lipids separately, such that large values of V/A have a great excess of proteins over lipids. This drives Kaeff⟶KaPP (Fig 3A and 3B). For most physiologic V/A values (~0.05–20μm S3 Table) however, and physiologic concentrations of proteins (1nM-10μM) or lipids (103−105μm-2 S4 Table), proteins are not in great excess, meaning significant enhancement is achievable depending on the membrane localization strength (Fig 3A and 3B).
Our theory (Eq 3) only describes the equilibrium state of the model. However, we can determine speeds of assembly via simulation. Now, the binding rates and the absolute values of V and A (not just the ratio) will influence the kinetics (all simulation inputs in S2, S3, and S4 Datasets). For these time-scales, we find that protein-lipid affinities KaPM are most often shown to be critical in controlling the overall time-scales of complexation, even driving slow-downs in speeds relative to solution binding (Fig 3E, S4 Fig). Changes in diffusion from solution to the membrane (about 100 times slower) affect the magnitude of association and dissociation rates and are captured implicitly in our ODE simulations (Methods), and explicitly in our spatially resolved reaction-diffusion simulations [19, 20]. However, the influence of diffusion on the reaction rates is rarely a dominant factor in physiological rate regimes (S4 Fig), indicating it is the binding strengths rather than slow 2D diffusion that determine assembly speeds. However, we note that our comparison of ODE and RD kinetics was performed in relatively small RD systems due to simulation costs, and it is true that as spatial dimensions increase, times to diffuse to reach the membrane will influence the overall equilibration times. The timescales we calculated for protein pairs and scaffold mediated systems (S4B and S4C Fig) were performed using ODEs at their corresponding cellular dimensions (S3 Table): V = 1200 μm3 (human) and V = 37.2 μm3 (yeast). Performing RD simulations at these dimensions would produce slower relaxation times, particularly for human cells, due to the time required to reach the surface. Crowding would also lower effective diffusion constants of proteins, although the decrease in time-scales to equilibrate would be negligible unless binding rates were strongly diffusion-influenced (Methods).
To test the biological relevance of membrane localization for driving complex formation and assembly, we collected biochemical (Table 1, S1 Table, S2 Table), concentration (Table 1, S4 Table), and cellular geometry data (S3 Table) for interactions among 37 membrane targeting proteins in yeast and human cells, including 22 proteins involved in clathrin-mediated endocytosis (CME). We first study only individual protein pairs that can bind according to our model of Fig 1, (S1 Table) shown in Fig 4A: the membrane binding proteins AP-2, DAB2, ARH, FCHo1, FCHo2, HIP1, HIP1R, PICALM, SH3GL2, EPN, AP180, SLA2, and SYP1. In Fig 4B and 4C we show results of binding between specific pairs. We used cytosplasmic concentrations of the proteins (Table 1) and the targeted lipids (S4 Table), and the relative Volume and Area from their respective cell types (S3 Table). Binding constants are collected from previous experimental studies (Table 1, S1 Table, S3 Dataset), and for 2D binding constants we test values of σ = 1nm (Fig 4) and 10nm (S5 Fig). Our results thus provide quantitative insight into how these pairs in isolation would use membrane localization at physiologic conditions to drive their protein-protein interactions. For some proteins, such as AP-2, solution KaPP values have been measured with partners (Fig 4A, S1 Table), but further experiments indicate that the proteins undergo minimal binding in solution due to conformational regulation [1]. Despite this additional regulation, membrane localization will still increase complex formation relative to what is observed in solution (γ>1), so the effect is quantified here using the measured KaPP value (S1 Table). Using our theory along with simulations for verification and time-scales, we find that affinities of these CME binding pairs can be enhanced 10–1000 fold by binding to membranes (Fig 4B, S5 Fig for results with σ = 10nm). With limited binding in solution for most pairs, membrane localization then triggers a dramatic increase in complex formation (Fig 4C, S5 Fig). The central adaptor protein AP-2 is responsible for many of these interactions, showing the capacity to trigger assembly with nearly all of its binding partners (Fig 4C, green bars). Even though we assume binding is possible for AP-2 in solution, it is still quite limited prior to localization. Not surprisingly, knockdown of AP-2 in mammalian cells causes severe disruption of endocytosis [28], underlining its secondary importance only to the irreplaceable clathrin [28] and PI(4,5)P2 [29]. We note that AP-2 can potentially bind up to three PI(4,5)P2 copies [30, 31], meaning that there will be less free lipids available for each AP-2. With fewer lipids, enhancement and complexation will be reduced, but is still quite large (S5 Fig). For some proteins such as FCHo1 (SYP1 in yeast), binding affinities (KaPP or KaPM) are not available, and this F-BAR protein does not target a single lipid specifically. However, by considering ranges of membrane stickiness values, we can use our method to identify which combinations (S6 Fig) best describe the experimental observation that these proteins only localize effectively to membranes when they can bind other proteins [18, 32]. We find for this protein, membrane stickiness values of ~0.5 produce membrane targeting that is sensitive to protein-protein interactions, whereas once values exceed ~1, no partners are needed to target the membrane effectively (S6 Fig).
We further interrogate two additional mechanisms for stabilization at the membrane by lipid binding proteins such as AP-2, epsin, and Dab2 [3]. First, they each bind transmembrane cargo after membrane localization, which acts to effectively increase the KaPM by increasing their residence time on the membrane. KaPM is a factor of ~40 higher for AP-2 binding to PI(4,5)P2 when cargo is available [30]. Interestingly, these cargo stabilized interactions (Fig 4B, light green) do not make a significant impact on complexation when we assume the full 1% PI(4,5)P2 concentration is free to bind, as the numerous lipids outweigh a need for stronger binding (Fig 4C, light green). However, when we evaluate complexation with PI(4,5)P2 pools diminished by a factor of 10 due to assumed competition from other PI(4,5)P2 binders, now cargo stabilization via higher KaPM does help recover strong complexation on the membrane (S5 Fig). This suggests that cargo binding, which is a main functional goal of CME, becomes a significant regulator of adaptor stabilization when competition from multiple adaptors limits PI(4,5)P2 binding. Second, when these adaptors can bind multiple partners with distinct appendage domains [33], we see more proteins on the membrane due to the increased difficulty of un-tethering from the membrane domain (S6B Fig).
In the cellular environment, CME proteins are of course not in isolation and can both compete and cooperate with one another to form higher order assemblies, induce conformational changes, and occupy lipid binding sites on the membrane. Thus we can only speculate about the role of localization in nucleating clathrin-coated pits in vivo. However, based on the above analysis showing that, physiologically, γ is greater than 1, membrane localization will drive clathrin towards complex formation. The initial nucleation of clathrin-coated pit sites is difficult to resolve experimentally because of the challenges in tracking the many participatory proteins simultaneously, and because prior to cage formation, the density of molecules is, by definition, low. Experiments have tracked the role of AP-2 and clathrin in nucleating sites [34], which we discuss below.
To go beyond our Fig 1 model of pairwise protein binding and thus characterize how scaffold proteins (Table 1: ITSN1, EPS15, EDE1 and SLA2) stabilize complex formation at the membrane despite not directly interacting with the lipids (model in S7 Fig, list of interactions in S2 Table), we simulate systems of ODEs, as Eq 3 no longer applies (Methods). We thus simulate interactions involving three proteins, two of which can bind lipids but not each other, and the third that binds both peripheral membrane proteins but not the membrane (Fig 5A). Our results in Fig 5B and 5C show that while scaffold mediated complexes can still capitalize on 2D localization for binding (Fig 5B), because localization is now mediated by peripheral membrane proteins that are at much lower concentrations than the lipid recruiters, we find that the increase in complex formation is less robust (Fig 5C), and is limited by concentration of the scaffold protein (S8A Fig).
Thus far we have not discussed clathrin, the central component of the CME vesicles that does not actually bind to lipids itself. In vitro experiments find that clathrin polymerization on the membrane (via adaptor binding) is more robust than occurs in solution (with adaptors still present), supporting a role for membrane localization in its nucleation and assembly [2]. Clathrin is a trimeric protein with three binding sites to target peripheral membrane proteins. It polymerizes with itself into hexagonal lattices without competition from the peripheral membrane proteins. Thus, its interactions with peripheral membrane proteins not only increase the quantity of protein bound to the membrane, it can help drive 2D polymerization between clathrin trimers. Through (non-spatial) stochastic simulations (Methods) set-up to mimic recent in vitro experiments [1], we explored a range of clathrin-clathrin interaction strengths to show how membrane recruitment by the AP-2 adaptor [1] can enhance clathrin polymerization yield (S8 Fig). Although these simulations lack molecular structure or spatial resolution, they can track formation of multi-protein complexes and the important role of affinity and concentration in controlling these complexes. We find that clathrin localizes to the membrane first via AP-2 binding before assembling into cages in 2D for the most reasonable Kds of 10–100μM [35]. This result is supported by evidence from in vivo experiments that probe the early stages in the nucleation of clathrin coated pit sites through tracking of AP-2 and clathrin [34]. They found that clathrin arrives at the membrane most frequently (75%) as a single trimer, and bound to at least one but most often two AP-2 molecules [34]. Nucleation can then initiate in two ways: (A) another clathrin trimer localizes to the membrane via AP-2 and these trimers dimerize in 2D or (B) another clathrin trimer is directly recruited by the clathrin on the surface. Although the subsequent clathrin dimerization events were not resolved experimentally, preventing definitive evidence of membrane localized clathrin-clathrin assembly, the fact that each clathrin is bound to AP-2 suggests that AP-2 binding of clathrin is a prerequisite for initial clathrin dimerization. From our simulations, the (A) nucleation process is markedly dominant. There is a strong driving force both from affinity and from concentration for AP-2 to bind any of the 25000 PI(4,5)P2/um2, and correspondingly minimal drive for a solution clathrin to bind a small number of clathrin trimers localized to the surface. We note that because clathrin also arrives at the membrane as dimers or higher order complexes 25% of the time, solution binding of clathrin also contributes to nucleation of pit sites, but to a much lower extent [34]. Interestingly, once pit sites have formed, assembled clathrin cages exchange with solution clathrin with the aid of ATP-consuming proteins that facilitate remodeling of the clathrin cage [36]. Thus clathrin-clathrin interactions from solution certainly play an important role in the cell in maturing the pit sites [36].
CME proteins with BAR domains that dimerize, appear to oligomerize only on the membrane, and are functionally important for driving membrane deformation [17, 18] can also exploit localization to drive their binding interactions. We study isolated FCHO1/2 oligomerization and endophilin (SH3GL2) oligomerization, again using non-spatial stochastic simulations (S9 Fig). Here again we consider a range of KaPM values to capture uncertainty in the membrane stickiness of these domains. We find the stoichiometry of the dimerization pair (homo or hetero) is central in determining whether large oligomers form. With matched pairs, homodimers such as endophilin form larger oligomers that feedback into higher stabilization at the membrane, whereas the disparity in FCHo1 and FCHo2 concentration (S1 Table) produces more isolated dimers. Experiments have shown that BAR domains exhibit stronger binding to curved membranes [17]. Because we lack this cooperative feedback in our model between oligomers tubulating membranes and thus potentially increasing affinity for subsequent proteins, our result can be interpreted as a lower bound on observed oligomerization.
In all cases, an important outcome of these strong binding interactions on the membrane is that they are difficult to disassemble, consistent with findings that unproductive assembly events observed in vivo [37] require the ATP-driven uncoating machinery for disassembly [2, 38]. Our results demonstrate that establishing the physiologic significance of these polymerization observations depends not only on protein concentrations and solution conditions, but also the V/A ratio. Thus, this ratio should be regarded as a critical factor in designing in vitro experiments to better reflect in vivo behavior.
Lastly, our analysis motivates why diverse proteins that target membranes in yeast can follow pathways to assembly both similar and distinct from the CME proteins. In particular, the CME protein pairs produce limited protein-protein complexes when isolated in solution, but experience large enhancements due to membrane localization, triggering widespread protein-protein interactions only after binding to the membrane (Fig 4). For 15 yeast proteins involved in daughter cell budding, lipid regulation, and morphogenesis, we studied their pairwise binding in using cytoplasmic (yeast) concentrations (Table 1), lipid concentrations (S4 Table), cytoplasmic V/A ratios (S3 Table) and experimentally measured protein-lipid affinities (Table 1). We find binding enhancements are high (100–1000), similar to the CME proteins, indicating that binding will be promoted once proteins are on the membrane (S10 Fig, S2 Dataset). Although enhancements were readily measured for these yeast binding pairs because they were independent of KaPP values (S2 Dataset), we could not directly compare complexation for these interactions as we did for the CME interactions because they lacked any KaPP data. For binding enhancements, we found an exception in the coat forming proteins targeting endosomes (VPS5, VPS17, SNX4, SNX41), which only exhibit enhancements <20. These proteins target the PI(3)P lipid but most bind only weakly (KdPM>100μM) [39], limiting their enhancements despite a favorable V/A ratio at the endosome (S10 Fig, S3 Table). Unlike in CME, however, these coat proteins form stable interactions in solution [40]. Thus, rather than membrane binding triggering protein interactions, we would first expect the reverse: strong protein interactions in solution function to target and stabilize protein at the membrane through the cooperative effect (Fig 1D and 1E, Fig 2B). We test how the binding of the retromer components VPS5 and VPS17 to the endosome will be significantly enhanced by forming a higher order assembly in solution with the strong lipid binding cargo adaptor, SNX3 (S9 Fig). SNX3 targets PI(3)P with stronger affinity (~2μM) than either VPS5 or VPS17 [39], and is known to improve recruitment of the retromer to endosomes [40]. Once these small pre-assembled coat subunits are on the membrane, they can then continue to exploit localization to form larger protein coats.
We conclude by noting that assembly on membranes is regulated to occur at specific times or sub-cellular locales, and our theory provides a useful aid in predicting the changes in local protein, lipid concentrations, and affinities that are necessary to trigger (or prevent) such assembly. Ultimately, our theory is most powerfully applied to interpreting in vitro results, due to the simplifying assumptions of the model, and can improve the design and quantitative interpretation of assays probing multi-protein complexation at membrane surfaces. Also, given known protein-protein and protein-lipid binding affinities, our theory can quantitatively predict the results of in vitro experiments that mimic Fig 1, thus avoiding the need for such measurements. Our results indicate that even relatively low lipid concentrations (i.e. PI(4,5)P2 at ~1% of plasma membrane lipids) can be sufficient in many cases to stabilize proteins to membranes and drive protein-protein interactions. We found that additional factors, such as cargo binding by adaptor proteins in CME, are only strong regulators of membrane localization or protein interactions under specific conditions. Since cargo-binding is known to influence the success of vesicle formation in vivo[41, 42], this suggests that the condition where total PI(4,5)P2 concentration is reduced to mimic competition from other proteins is more physiologically relevant. A fruitful means of exploring in more detail the role of cytoplasmic factors, as well as spatial heterogeneity, crowding, and non-equilibrium dynamics, is through reaction-diffusion simulations, although we note the results will then be dependent on many additional parameters. Overall, the theory we provide here offers a general and useful quantitative guide for predicting when or if membrane localization plays a role in the cellular control of self-assembly.
To simulate the systems of ordinary differential equations (ODEs) for Fig 1 (S1E Fig, S1A Text section), we need macroscopic rates, and to simulate the single-particle reaction-diffusion system (RD), we need microscopic rates (also known as intrinsic rates in the Smoluchowski theory [43]) in both 3D and 2D. The macroscopic rates emerge based on the dynamics of the more detailed microscopic system, and can therefore be constructed to optimally match the kinetics of the ODE simulations to the RD simulations. We note that these definitions are specific to the kinetics, as the equilibrium of both simulation approaches will be identical due to their matching equilibrium constants.
The ODE simulations do not account for space or explicit diffusion. Here, we define their rates to implicitly account for changes to diffusion and thus best match the RD simulation kinetics. That way, discrepancies between kinetics of ODE and RD results can be attributed to explicit spatial heterogeneity influencing the binding interactions. Macroscopic association (on-) rates can be defined in 3D from the intrinsic rate of the Smoluchowski model via the relation [44]:
kon3D=(1ka3D+14πσDtot3D)−1,
(9)
where ka is the intrinsic association rate that captures the barrier to complex formation for species in contact at binding radius σ, and Dtot is the sum of both species’ diffusion constants. The macroscopic off-rate can be defined in all dimensions via
koff=kon/Ka.
(10)
The intrinsic dissociation rate kb is defined via the corresponding equation, kb = ka/Ka, with all off-rates having the same units in all dimensions of s-1. In 2D, there is no single macroscopic rate constant independent of the system size or concentrations [20]. However, one can define a macroscopic 2D rate, built on theory from Szabo et al [45], that provides optimal agreement with the corresponding spatial reaction-diffusion simulations via [20]:
kon2D=(1ka2D+18πDtot2D[4log(b(ρ)/σ)(1−σ2/b(ρ)2)2−2(1−σ2/b(ρ)2)−1])−1
(11)
where
b(ρ)=2A/(πmax(NP1,NP2)+σ2)
(12)
is a length scale that is defined based on the more concentrated of the reacting species P1 or P2 in the surface area A.
The important interpretation of Eqs 9 and 11 is that, unless ka is large, even substantial (factor of 10 or more) changes to the diffusion constant will have a relatively small impact on the macroscopic rate. It is not until macroscopic rates reach values of ~106-107M-1s-1 that they become strongly diffusion influenced and thus sensitive to changes in diffusion.
Our 2D intrinsic rates are defined relative to our 3D rates via
ka2D=ka3D/(2σ),
(13)
and unbinding rates
kb2D=kb3D,
(14)
which produces the equilibrium relation defined in the main text, Ka2D = Ka3D/(2σ). We assume here that the dissociation rates are the same from 3D to 2D. It is the association rates that capture two species finding one another in a specific spatial dimension. This definition of Eq 13 also can be shown to preserve the reactivity of the binding interaction in the Smoluchowski model from 3D to 2D, independent of changes to diffusion (S1 Text section 4A). For the macroscopic 2D rates, kon2D, we used Eq 13 in Eq 11, which allows us to capture effects of diffusion towards timescales of binding in kon2D, as D2D is ~100 times lower than D3D. Transitioning from solution to the membrane via binding lipid or protein involves a 3D search, and thus uses the corresponding 3D rates. See S1 Text section 3B for further discussion.
Ultimately, the results of Kaeff are only sensitive to equilibrium constants such as Ka2D and therefore the size of σ, rather than sizes of relative rates. This length scale σ encodes thermodynamic properties of the molecules involved in the binding reaction and is of the nanometer range [22]. In general, the value of σ therefore depends on the proteins involved, but σ (or Ka2D), is almost never measured. We extract σ~7nm (from V/A = 6.7μm and Kaeff/KaPP≈500) in the experimental measurement of 2D binding between calmodulin and a target peptide [21]. Smaller σ values have been observed [26]. For simulations, we thus used either 1 or 10nm. We used the same value for the protein-protein (σPP) or protein-lipid (σP1M, σP2M) 2D binding interactions, although only σPP appears in Eq 3. The size of these values is constrained to ensure an equilibrium steady-state is reached, and the simplest solution has that σPP=σP1M=σP2M.
In Table 1 we list all the human and yeast proteins for which we were able to collect sufficient biochemical data on lipid and protein interactions. The 20 lipid-binding yeast proteins were retained from a larger list of 139 peripheral membrane proteins (PMP) identified from the Uniprot database as having lipid binding activity in yeast (S1 Dataset). Between this set of 139 PMPs, we found 396 interactions via BioGRID, however, only 17 pairs (S1 Table) involved partners with known KaPM’s. The 15 human proteins studied are all involved in CME and their biochemical data (Table 1, S1 Table, S3 and S4 Datasets) was collected via extensive literature curation. To study scaffold-mediated interactions (S2 Table), we identified all possible interactions that involved a non-membrane binding protein that could simultaneously and non-competitively bind to two of our PMPs. For the yeast proteins, these interactions could be identified from the manually curated interface interaction network for CME proteins [48]. There was a relatively small number of examples where a single scaffold protein was capable of bridging two PMPs (S2 Table). These interactions in humans/yeast involved clathrin/clathrin, eps15/ede1, or itsn1/sla1.
In S3 Table we collected volume and surface areas for cells and organelles with justifications provided. Because the cytoplasmic volume typically constitutes 50–60% of the total cell volume in mammalian cells, our V/A ratios set the solution volume as 60% of the total cell volume for all cell types. Lipid concentrations are collected in S4 Table. The concentrations of specific lipids on specific membranes have only been quantified in a few cases, such as PI(4,5)P2 having an average concentration of 2.5x104μm-2 on the plasma membrane in mouse fibroblasts [15]. We used this concentration as a gold standard, due to its relative consistency across measurements [13, 15], and other phosphoinositide concentrations were quantified relative to this one. We curated literature to collect the necessary copy numbers of each lipid in the cell, and their distributions across organelles. Lastly, protein concentrations were defined from copy numbers measured in yeast [49] and human cells [50] (Table 1).
|
10.1371/journal.pgen.1002706 | Loss of Gnas Imprinting Differentially Affects REM/NREM Sleep and Cognition in Mice | It has been suggested that imprinted genes are important in the regulation of sleep. However, the fundamental question of whether genomic imprinting has a role in sleep has remained elusive up to now. In this work we show that REM and NREM sleep states are differentially modulated by the maternally expressed imprinted gene Gnas. In particular, in mice with loss of imprinting of Gnas, NREM and complex cognitive processes are enhanced while REM and REM–linked behaviors are inhibited. This is the first demonstration that a specific overexpression of an imprinted gene affects sleep states and related complex behavioral traits. Furthermore, in parallel to the Gnas overexpression, we have observed an overexpression of Ucp1 in interscapular brown adipose tissue (BAT) and a significant increase in thermoregulation that may account for the REM/NREM sleep phenotypes. We conclude that there must be significant evolutionary advantages in the monoallelic expression of Gnas for REM sleep and for the consolidation of REM–dependent memories. Conversely, biallelic expression of Gnas reinforces slow wave activity in NREM sleep, and this results in a reduction of uncertainty in temporal decision-making processes.
| REM and NREM sleep are two distinct stages of the sleeping brain that are involved in the modulation of metabolic, physiological, and cognitive processes. Clinical evidence suggests that epigenetic mechanisms, such as genomic imprinting, play a role in sleep. Here we show that REM and NREM brain states are differentially modulated by the maternally imprinted gene Gnas. In particular, a mutation resulting in loss of imprinted expression of Gnas enhances NREM–dependent physiologic and cognitive functions while repressing REM and REM-linked functions. This is the first experimental demonstration of a specific effect of genomic imprinting on sleep states and their associated effects on cognition.
| Mammalian evolution from the reptile lineage involved important changes in gene regulation and sleep. Many genetic mechanisms play important roles in various electrophysiological and behavioral traits that set the three major states of mammalian life: Wakefulness, Rapid Eye Movement (REM) and Non-REM (NREM) sleep [1], [2]. However, here we focus on a particular gene regulation, namely genomic imprinting. Genomic imprinting is an epigenetic mechanism that results in allele-specific expression of some genes according to parental origin and, in vertebrates, is unique to mammals.
Clinical observations of neurodevelopmental disorders of sleep suggest a role of genomic imprinting on various measures of sleep [3], [4]. For example, Prader-Willi syndrome (PWS) and Angelman syndrome (AS), both neurodevelopmental syndromes, exhibit opposing imprinting profiles and opposing sleep phenotypes. PWS is associated with maternal duplications/paternal deletions of alleles on chromosome 15q11–13 and is characterized by temperature control abnormalities and excessive sleepiness as well as REM sleep abnormalities [5]–[8]. Conversely, AS is associated with paternal duplications/maternal deletions on chromosome 15q11–13 and is characterized by severe mental retardation and reductions in sleep. The UBE3A gene, that resides in the PWS/AS imprinting region, has been associated to sleep abnormalities. Ube3a deficient mice are characterized by reduced NREM sleep, deteriorated REM sleep, and an increased frequency of waking during the dark–light transition [9].
Interestingly, the serotonin (5-HT) 2A receptors, which mediate aminergic inhibition of REM-on cells in the parabrachialis lateralis region [10], are primarily expressed from maternal alleles [11], [12]. The above reviewed findings support the idea that epigenetic regulatory mechanisms, such as genomic imprinting, influence sleep-associated mechanisms.
REM and NREM sleep underlie important metabolic, physiological and cognitive processes. REM sleep influences early postnatal developmental behaviors; it facilitates the acquisition of resources from the mother (i.e. by means of suckling behavior) and it promotes the release of hormones such as prolactin and oxytocin which are pivotal in the development of attachment behavior [4]. From an evolutionary and behavioral perspective, REM sleep has been associated with adult reproductive success [4]. NREM sleep is associated with more stable metabolic and autonomic responses compared to REM sleep. Furthermore, the presence of REM-like and NREM-like states has been associated with nutritive behaviors in offspring and mother, respectively [4].
Recent advances in functional studies of sleep strongly suggest that memory consolidation benefits from a slow (<1 Hz) highly synchronized cortical activity in NREM and from subcortical theta (5–9 Hz) rhythms in REM sleep [13]. Despite these advances in functional understanding of REM and NREM sleep states, genetic and epigenetic mechanisms of sleep-dependent plasticity and memory processing are not currently well understood.
Here we report, for the first time, experimental evidence for a role of an imprinted gene, Gnas, in the modulation of REM/NREM sleep physiology. Gnas encodes the stimulatory G-protein subunit Gsα, which is involved in the generation of intracellular cyclic AMP and plays a crucial role in energy expenditure and metabolism by mediating sympathetic effects on many tissues [14]. It is biallelically expressed in most tissues including adult white adipose tissue [15] but it is predominantly maternally expressed and paternally repressed in a subset of tissues such as neonatal brown adipose tissue (BAT) [16], although it is not known if it shows imprinted expression in adult BAT. BAT produces heat by fatty acid oxidation and serves a pivotal thermoregulatory function within the organism [17]. Thanks to the rich presence of mitochondria and by means of the uncoupling protein-1 (UCP1), BAT is implicated in non-shivering thermogenesis [15], [18].
Imprinted expression of Gnas is controlled by a cis-acting differentially methylated region (DMR): the Exon1A-DMR [16]. Paternal transmission of a deletion of the Exon1A-DMR (Gnastm1Jop, hereafter called Ex1a, see Figure 1a) causes derepression of the normally repressed paternal Gnas allele in imprinted tissues resulting in biallelic Gnas expression and loss of imprinting [16]. We show here, for the first time, that loss of imprinting of Gnas results in specific abnormalities in sleep, cognition and thermoregulation in adult mice.
On paternal inheritance of the deletion, we have observed that in +/Ex1a mice the level of Gnas mRNA expression is increased, compared to littermate controls, in adult BAT (Figure 1b) indicating a derepression on the paternal allele due to a loss of Gnas imprinting, as previously reported in newborn mice BAT [16]. Moreover, we show here that Ucp1 mRNA in BAT is significantly higher in +/Ex1a mice compared to littermate controls (Figure 1b). As expected from the observation of increased Ucp1 levels a significantly higher body temperature was found in mutant animals (Figure 1c). The major increase of temperature in +/Ex1a mice compared to littermate controls occurs at the end of the subjective day and at the beginning of the subjective night, when there is a strong urge to sleep (Figure 1c).
In order to investigate the sleep-wake profile and the behavioral performance in these mice, we subjected adult +/Ex1a mice and +/+ littermate controls to behavioral and electrophysiological investigation in the home-cage environment. Interestingly, the total REM sleep was significantly reduced in +/Ex1a mice compared to littermate controls (Figure 2a) while the total NREM was unaffected between the two groups (Figure 2b). However, at the electrophysiological level, the contribution (power density) of delta (1–4 Hz) frequencies, the main synchronized rhythm in NREM sleep, was higher in the +/Ex1a mice compared to +/+ mice (Figure 3).
A 6-hour sleep deprivation protocol triggered, immediately after deprivation, a significantly increase (rebound) of REM in +/Ex1a mice, which testifies the need for a homeostatic recovery of REM, but not for NREM sleep, in mutants (Figure S1 and Figure S2). No differences in body temperature occurred during sleep deprivation and during the following recovery period between the two groups (Figure S2).
Thus, we extended our investigation into specific REM/NREM-dependent behavioral functions. We subjected mice to a classical memory task, the fear conditioning (FC) test, which affects REM sleep homeostasis [19]. After the first (conditioning) day of the FC protocol, we have observed an increase of REM sleep but not of NREM sleep in +/+ mice (Figure S1a–S1b). The presence of REM and its theta density were significantly higher in +/+ compared to +/Ex1a mice (Figure S1b). We suggest that this lack of REM increase in mutants is the causal mechanism explaining the reduced freezing behavior in +/Ex1a mice, compared to controls, that occurred when the animals were exposed to the same context (Figure 4) the following day. Indeed, the consolidation of fear responses has been previously associated with REM mechanisms that occur during sleep [19]. Because no difference between the two groups was observed in the cue condition (day 3) we reason that the deficit in in +/Ex1a mice is restricted to context-dependent mechanism, which has been previously associated with REM/fear memory consolidation [20], [21].
A different subset of mice was tested in their home-cage with a cortico-striatal cognitive task, the “Switch-test” [22]. The test requires the animal to decide to nose-poke in different hoppers within the home-cage in response to a short- versus a long-light signal to obtain a reward (Figure 5a). Optimal performance in this task implies that the animal has learnt to distinguish between a short-signal associated with the reward coming at the hopper in one of the two locations and a long-signal associated with the reward coming at the hopper in the other location. This test of cognitive performance assesses whether the animal has an accurate representation of both endogenous/implicit (the subjective estimation of the signal duration) and exogenous/explicit (the ratio between short and long signals) temporal variables. In one condition (the “Switch” condition) all trials resulted in a reward if the animal responded correctly. In a second condition (the “Probes” condition) a percentage of trials was never rewarded regardless of the response of the animal. This latter condition was to measure the animal's uncertainty and its cognitive performance during a temporal decision making process [23]. Interestingly, +/Ex1a mice performed better in each phase of the experiment showing higher accuracy and time precision compared to controls (Figure 5b–5d). As expected, this 24-hour home-cage cognitive effort triggered a significant NREM increase and a higher delta power in +/Ex1a mice compared to +/+ mice in both “Switch” and “Probes” conditions (Figure S1a–S1c). Notably, the “Probes” condition, which added a degree of uncertainty to the expectation of obtaining a reward, resulted in an more severe augmentation of NREM sleep respect to the “Switch” condition. REM did not change significantly following both conditions neither in +/+ or +/Ex1a mice (Figure S1a–S1c).
We have shown here clear evidence for an involvement of the imprinted Gnas transcript in the modulation of sleep and sleep-dependent behaviors.
Our study demonstrated a close link between sleep and thermoregulation. Temperature control is a well-known mechanism that modulates the expression of REM/NREM sleep in humans and mice [24]. When the thermoregulatory demand increases, REM sleep diminishes in rodents [24]–[26]. Hence, the increase of body temperature in +/Ex1a accounts for the significant reduction of REM sleep that we have observed in these mutants. Moreover, several studies have shown that an increase in body temperature is associated with an increase in NREM sleep propensity [24], [27]–[29]. Thus, the hyperthermic phenotype is responsible for the REM defect-NREM improvement that we observed in +/Ex1a mice, indicating a link between imprinted Gnas, thermoregulation and REM/NREM sleep [24]. The molecular pathway joining Gnas and thermoregulation in BAT is well understood. Gsα is considered an important mediator of the activity of the sympathetic nervous system (SNS) on many functions of BAT, included thermogenesis and energy expenditure [30]. SNS stimulation of BAT leads to thermogenesis. Gsα is a constituent of transmembrane G-protein-coupled receptors that translate adrenergic SNS stimulation, in BAT, to the activation of UCP1 via specific intracellular changes and a signaling cascade that involves the production of cAMP and activation of protein kinase A (PKA). Our results showing increased Gnas - Ucp1 expression are in line with a specific SNS-molecularly mediated cAMP-PKA role in non-shivering thermogenesis and therefore sleep. Indeed, the intracellular activity of cAMP-PKA is inversely related to the urge to sleep. In particular, an increased PKA affects sleep according to the specific cells in which it is expressed [31]–[33]. A reduced REM sleep would, presumably, be associated with an elevated cAMP-PKA activity.
We have shown that functional monoallelic expression of Gnas, due to imprinting of this gene, is important, in mice, for the physiology of REM sleep and for the consolidation of fear conditioning contextual memories. Many experimental observations support the idea that hippocampus is crucial for the development of contextual fear conditioning [20], [34]. However, fear conditioning responses are mediated by a complex interaction within limbic and prelimbic areas and this is also modulated by circadian genes [35]. As Chen and colleagues [36], have shown that Gnas is not imprinted in the hippocampus, our study implies that this specific cognitive phenotype must be due to imprinting of Gnas in brain regions other than hippocampus. Within this same study it was shown that Gnas is imprinted in the paraventricular nucleus (PVN) of the hypothalamus, a brain region that subserves important metabolic functions [36]. The PVN receives important projections from neurons located within the suprachiasmatic nucleus (SCN) of the hypothalamus, the master clock for circadian rhythms [37]. The interaction between PVN and SCN is important in orchestrating circadian rhythms and to set proper neuroendocrine responses to stressors [37]. Thus loss of Gnas imprinting, within these structures, can be envisaged as playing a role in both sleep and behavior. The results of our study confirm the idea that REM sleep is fundamental in the consolidation of fear responses [19], likely, involving a complex network of brain activity.
We have also shown that NREM functions are sensitive to Gnas dosage but, in this case, loss of imprinting of Gnas results in a reduction of uncertainty in temporal decision making. This improvement is paralleled by an increased contribution of slow wave activity in NREM sleep. Our result is consistent with the idea that cortical slow oscillations, by modulating synaptic circuits between subcortical and cortical structures, are responsible for the consolidation of daily memories during sleep [13].
In conclusion, in our study, loss of imprinting of Gnas, inhibits REM and primitive REM-linked functions, such as the fear response to a threatening context. Conversely, it enhances NREM physiology and high-level cognitive functions that developed alongside a progressively complex brain.
However, from a behavioral point of view an increased precision in interval timing estimation does not necessarily result in a better performance in other behavioral responses. Indeed, if the subject is less certain about the time of the foot-shock, its fear conditioning response may start earlier and stop later [38], hence resulting in a higher freezing time. Perhaps, evolution has developed a balanced mechanism between temporal uncertainty and fear behavioral responses and then loss of imprinting, in +/Ex1a mice, involves a reduction of freezing because of a higher timing precision. Thus, REM/NREM sleep expression may favor this well-adjusted mechanism.
The results of our study indicate a specific role for the imprinted gene Gnas in thermoregulation, which in turn affects REM/NREM sleep and then, cognitive performance. In addition we also reported a novel effect of specific cognitive mechanisms on sleep, by showing that a specific decision making process, the consolidation of an interval timing task, influences NREM sleep. Furthermore, in our mouse mutant model, the particular NREM physiology, exacerbates the effect of cognition on sleep homeostasis. This study attests the relevance of Gnas in brain functions and that loss of imprinting of Gnas affects cognitive processes.
Another transcript within the Gnas locus, the paternally expressed transcript Gnasxl, is expressed in specific sleep-related brain areas including the locus ceruleus and cholinergic laterodorsal tegmental nuclei [39]. The activity of neurons in the cholinergic laterodorsal tegmental nucleus is particularly important in the regulation of REM sleep [40], [41]. Mice with mutations in paternally derived Gnasxl transcripts show phenotypic deficits associated with growth and development, which is a critical stage for REM sleep across many species [4]. Thus Gnasxl may also play a role in sleep.
Initial Ex1a stock mice were produced in MRC-Harwell on 129/SvEv background and then transferred to IIT. In IIT mice were bred and maintained, through paternal inheritance, for several generations on C57BL/6J background, as this is a favorite background for behavioral studies. The genotyping of the mice was conducted following the assay as indicated in [42]: Exon1aF 5′cagtcgcgtcggcaccgcggag3′ and Exon1aR 5′gacgcactcacacgcaaagcag3′.
All the behavioral and electrophysiological experiments, in adult +/Ex1a mice and +/+ littermate controls, were conducted in the home-cage environment. In addition, mRNA expression profiles for Gnas and Uncoupling Protein (Ucp)1 were made in naïve adult mice. Each experiment included 8 male mice (10-weeks old) for each genotype and all procedures were done under the guidance issued by the UK authority (Project Licence Numbers 30/2526) and under the Italian Policy (licence issued on 19/06/2009, decreto N°106/2009-B).
Total RNA was extracted from about 0.2 g of snap frozen brown adipose tissue (BAT) and homogenized using Pestel with Trizol Reagent (Invitrogen, Carlsbad, CA) to isolate total RNA according to the manufacturer's procedure. Q-PCR was conducted essentially as previously described [43], [44]. Specific primers were: UCP1:f5′-GTCCCCTGCCATTTACTGTCAG-3′, r5′-TTTATTCGTGGTCTCCCAGCATAG-3′; GNAS: f5′-AGAAGGACAAGCAGGTCTACCG-3′, r5′-GTTAAACCCATTAACATGCAGGA-3′; β-actin, f5′-GGCACCACACCTTCTACAATG-3′, r5′-GGGGTGTTGAAGGTCTCAAAC-3′. Thermal cycling parameters were: denaturation 95°C for 5 min followed by 40 cycles of denaturing- annealing and extending (95°C for 15 sec, 60°C for 30 sec and then 70°C for 1 min). The results were calculated by the comparative Ct method according to the Applied Biosystems ABI-PRISM-7700 User Bulletin#2. Each sample was run in quadruplicate to obtain average Ct values and a ΔCt value for the target gene of the same sample, normalizing each sample to β-actin. The expression relative to β-actin was determined by calculating 2−ΔCt. Mean comparison was performed with unpaired Student's t-test.
Mice were subjected to a long-term investigation in home-cage environment after the implant of a wireless system (Data Sciences) that enables to record electroencephalography (EEG), electromyography (EMG), locomotor activity and body temperature for off-line sleep-stage analysis (see [19]). Automated sleep scoring followed by visual manual inspection was performed using all sleep criteria for mice [19]. We performed Fast Fourier Transform analysis of the EEG signals with SleepSign software. The contributions of EEG frequencies was expressed as power densities in each frequency bin in all NREM and REM sleep epochs (as described in [45]).
A 2 week post-surgery period of recovery were given to each mouse to ensure a full recovery of normal sleep. At the end of the recovery period, we started recording all the physiological signals uninterruptedly for 48 consecutive hours (sleep baseline). Then, mice went under 6-hour sleep deprivation (SD) and 6-hour recovery period. After an additional 1-week the mice underwent fear conditioning (FC) or timing learning (Switch task) conditions.
|
10.1371/journal.pntd.0004182 | Using Co-authorship Networks to Map and Analyse Global Neglected Tropical Disease Research with an Affiliation to Germany | Research on Neglected Tropical Diseases (NTDs) has increased in recent decades, and significant need-gaps in diagnostic and treatment tools remain. Analysing bibliometric data from published research is a powerful method for revealing research efforts, partnerships and expertise. We aim to identify and map NTD research networks in Germany and their partners abroad to enable an informed and transparent evaluation of German contributions to NTD research.
A SCOPUS database search for articles with German author affiliations that were published between 2002 and 2012 was conducted for kinetoplastid and helminth diseases. Open-access tools were used for data cleaning and scientometrics (OpenRefine), geocoding (OpenStreetMaps) and to create (Table2Net), visualise and analyse co-authorship networks (Gephi). From 26,833 publications from around the world that addressed 11 diseases, we identified 1,187 (4.4%) with at least one German author affiliation, and we processed 972 publications for the five most published-about diseases. Of those, we extracted 4,007 individual authors and 863 research institutions to construct co-author networks. The majority of co-authors outside Germany were from high-income countries and Brazil. Collaborations with partners on the African continent remain scattered. NTD research within Germany was distributed among 220 research institutions. We identified strong performers on an individual level by using classic parameters (number of publications, h-index) and social network analysis parameters (betweenness centrality). The research network characteristics varied strongly between diseases.
The share of NTD publications with German affiliations is approximately half of its share in other fields of medical research. This finding underlines the need to identify barriers and expand Germany’s otherwise strong research activities towards NTDs. A geospatial analysis of research collaborations with partners abroad can support decisions to strengthen research capacity, particularly in low- and middle-income countries, which were less involved in collaborations than high-income countries. Identifying knowledge hubs within individual researcher networks complements traditional scientometric indicators that are used to identify opportunities for collaboration. Using free tools to analyse research processes and output could facilitate data-driven health policies. Our findings contribute to the prioritisation of efforts in German NTD research at a time of impending local and global policy decisions.
| Neglected tropical disease research has changed considerably in recent decades, and the German government is committed to addressing its past neglect of NTD research. Our aim was to use an innovative social network analysis of bibliometric data to map neglected tropical disease research networks that are inside of and affiliated with Germany, thereby enabling data-driven health policy decision-making. We created and analysed co-author networks from publications in the SCOPUS database, with a focus on five diseases. We found that Germany's share of global publication output for NTDs is approximately half that of other medical research fields. Furthermore, we identified institutions with prominent NTD research within Germany and strong research collaborations between German institutions and partners abroad, mostly in other high-income countries. This allowed an assessment of strong collaborations for further development, e.g., for research capacity strengthening in low-income-countries, but also for identifying missed opportunities for collaboration within the network. Through co-authorship network analysis of individual researcher networks, we identified strong performers by using classic bibliometric parameters, and we identified academic talent by social network analysis parameters on an individual level.
| In recent decades, global efforts against neglected tropical diseases (NTDs) have undoubtedly been successful in raising awareness and implementing ambitious treatment programs [1]. NTD research and development (R&D) efforts have shown a substantial increase [2], and necessary resources have largely been made available by the philanthropic and public sectors [3]. Although advances have been made, funding for global NTD R&D has been flatlining more recently, and significant need gaps in diagnostic and treatment tools prevail [4]. The global strain on public finances as well as questions about research inputs, e.g., regarding the coordination of international efforts, appropriate prioritisation and the return on investment (RoI), are leading to an increasing pressure to use limited resources most efficiently [5,6]. A Global Health R&D observatory has been proposed by the World Health Assembly to identify research needs to provide better information on where, by whom and what type of research is conducted, and to match limited resources with public health priorities more effective and efficiently [7].
With respect to R&D for NTDs, defining and measuring research output, research productivity and value in return for funding have received comparatively little attention. Furthermore, research processes and collaborations, such as scientific networks, have seldom been used for evidence-driven policy analysis. However, new tools and software are increasingly available to facilitate analysis in this field [8,9].
Bibliometric metadata of scientific publications are used to map and visualise scientific activity within countries or regions [10,11], and co-authorship network analysis is used to explore and quantify R&D collaboration between authors, institutions or countries [12].
In co-authorship networks, authors in the network are linked as nodes via co-authored scientific publications. These networks can be visualised as graphs in which each author represents a node in the network and each co-authored publication is represented by links, or edges, between the nodes. Measurements can describe the network structure, e.g., by density or centrality of authors. For an explanation and description of terminology used in social network analysis (SNA), see Table 1. In particular, a high betweenness centrality of individual author nodes indicates that they are connecting parts of a network that would only be poorly connected otherwise, or are not connected at all, and these nodes are interpreted as innovation hubs within networks [13,14].
Along with other governments and non-governmental organisations, the German government acknowledges the fight to tackle NTDs as a continuing major global challenge and sees improved R&D efforts and capacity building as a way forward [18].
However, German global health and NTD research has been considered to be scarce in comparison with activities in comparable industrialized countries [19,20]. Other EU countries that usually show a similar or smaller amount of publication output in other research fields, such as the United Kingdom or France, have been shown to outperform Germany considerably in the field of NTD research [10]. As a late but committed entrant into this global effort, Germany can gain a lot by using tools effectively and efficiently to prioritize its resources and capabilities.
By using innovative tools such as network analysis, we aimed to both demonstrate the potential of this tool and to identify and map existing NTD research outputs and processes within Germany and with its partners abroad to enable informed, evidence-driven policy making that addresses the needs of patients with NTDs.
Initially, separate searches using the SCOPUS (www.scopus.com) database were performed for individual diseases to allow for the systematic identification of all relevant publications published between 2002 and 2012 on 11 kinetoplastid and helminthic diseases, and the NTD groups receiving the highest German government research funding [21,22]. Their metadata was extracted to build co-authorship research networks (Fig 1).
The search string used here included the disease name or a combination of the scientific name and common name as found on the WHO Neglected Tropical Disease Website (http://www.who.int/neglected_diseases/diseases/en/). A list of diseases and the search string used for further analysis are listed in Table 2.
No filtering was performed for the research type (e.g., basic, clinical, and operational research) to acknowledge the diversity of NTD research needs [23]. To allow a comparison between overall international and German research output, the searches were first performed without a location filter, and during a second step, an affiliation filter for Germany was applied. Only publications with at least one co-author affiliated with a German institution were included.
The results of the SCOPUS database search with the Germany affiliation filter were exported. We further analysed the datasets for the five diseases with the highest number of publications.
OpenRefine software (www.openrefine.org) was used for data cleaning for all exported datasets. The given clustering algorithms were used to remove duplicates and resolve name disambiguation issues [24] for both authors and research institutions. Additional data cleaning was conducted manually afterwards.
Bibliometric data provided by the SCOPUS database was used to calculate the h-index for individual authors in each set of publications for individual diseases to establish measures for individual author output productivity (number of publications) and impact (number of citations by other researchers) for further analysis. To create the h-index, citations for each author's publications were collected from the SCOPUS exports of each disease, and individual author h-indices were calculated specifically for each disease as defined by Hirsch [25].
For the spatial visualisation of the international research organisation networks, individual author affiliations from the bibliometric data were manually harmonised and geocoded using the OpenStreetMaps (www.openstreetmaps.org) application programing interface (API) within OpenRefine.
By using the web-based Table2Net tool [26], the processed database was set up to create co-authorship networks [27]. Networks for each disease were created for both collaborating co-authors and collaborating research institutions using author names or research organisations that were named as affiliations to identify nodes within the network (S1–S10 Datasets) [28]. The title of the co-authored publication was used to identify edges connecting individual nodes in both sets of networks. If authors or institutions collaborated more than once, the edges were weighted by the number of collaborations between the same authors or institutions. The individual disease networks represent all authors or institutions with contributions to NTD research publications that have at least one co-author working at a German research institution.
The Gephi software was used to calculate the occurrence counts and the network analysis parameter betweenness centrality (see Table 1). Authors were then ranked by their occurrence count, h-index and betweenness centrality. For the visualisation of co-author networks, the ForceAtlas 2 layout was used, and the global positioning system (GPS) coordinates for research institutions were used to visualise institutional collaborations on a world map [29].
All software used to analyse the data exported from SCOPUS is open-source and available free of charge.
An overview of the search results, bibliometric data and the following analysis of institutional collaborations and individual author research networks is presented in Fig 2.
Combined searches for publications on the 11 kinetoplastid and helminth NTDs revealed more than one thousand publications with at least one German author affiliation, representing 4.4% of the total number of publications published internationally about the different NTDs (Table 3).
For the 11 diseases included in our analysis, international research efforts were predominantly related to three diseases, whereas the German NTD research efforts were more equally spread across five diseases. The three most researched NTDs internationally (Leishmaniasis, Chagas disease and Schistosomiasis) comprise 68.6% of the cumulative international publications, compared with 60.7% for those with German author affiliations. In contrast, for the top five most frequently researched NTDs (the top three diseases plus Sleeping Sickness and Onchocerciasis), the proportion was 77.2% internationally and 82.1% for those with German author affiliations. More than a fifth of the publications with German author affiliations were in the fields of Sleeping Sickness (11%) and Onchoceriasis (10%) compared with less than a tenth of international publications (Sleeping Sickness 5%; Onchocerciasis 3%).
For the five diseases that formed the focus of our analysis (see Fig 3), we found a total of 972 publications with author affiliations from German institutions for 2002–2012. Of these, 908 publications were unique to one disease, and 64 publications appeared in two or more searches. Among the articles, 4711 authors and 3803 research institutions were identified. After duplicates were removed and name disambiguation was performed through computational and manual data cleaning, we identified 4568 authors and 1502 research institutions, but because some authors and research institutions were named in publications for more than one of the five diseases, we eventually identified 4007 individual authors and 863 research institutions.
Of the 972 publications, 96.5% were published in collaboration with two or more authors; 6,854 signatures were identified, with an author-per-paper index of 7.05. Five separate co-authorship networks were built through the identification and analysis of the 4,568 author nodes and 32,647 co-authorship edges (see Table 1 for network analysis terminology). These publications were published in 312 different scientific journals. A cumulative 50% of the publications were published in 28 different journals. The remaining publications were scattered throughout 284 journals, with 190 journals publishing only one paper on the five diseases during the study period (2002–2012).
Findings on institutional work and collaboration at the individual disease level are listed in Table 4, and more detailed information can be obtained in the S1 Table.
The number of collaborating institutions and their distribution across countries varied for the different disease networks. A map of the collaboration patterns for the five diseases is presented in Fig 3. The United States and the United Kingdom were the only countries in the top five collaborating countries for all five diseases. Over 60 percent of the contributions from outside of Germany came from countries classified as high-income countries (HIC) by the Organisation for Economic Co-operation and Development (OECD) for all diseases but Chagas disease, for which the contribution was 47 percent (S2 Table). The network around Chagas disease had the highest percentage of contributions from research institutions outside of Germany coming from upper-middle-income countries (49.7%). Among the emerging economies of Brazil, Russia, India, China and South Africa (BRICS) affected by NTDs, Brazil clearly outperformed China and South Africa, with both contributing to only approximately one-tenth of the collaborations as Brazil. India is a country with high Leishmaniasis prevalence [30] and it was ranked seventh for Leishmaniasis research partners, with almost no other published research collaborations for any of the other diseases analysed here. Co-authorship collaborations with partners in low-income countries (LIC), which carry the majority of the NTD burden, were overall only a tenth of those with partners from HICs. Among the LICs, co-authorships were spread across 27 countries.
When looking at continents, not income groups, the Onchocerciasis network was the only research network in which researchers from Africa contributed more than researchers from any other continent. It was also the only disease network in which African countries were among the top five research countries. Across all the diseases, co-authorships between researchers on the African continent and Germany showed that research collaborations were spread across 11 countries; however, few of these collaborations were very active.
Among the five diseases analysed here, the German research network was spread most widely, and it was the most international for Leishmaniasis (491 institutions, 64 countries) and the least for Onchocerciasis (134 institutions, 26 countries). In those German research networks, the contributions from German research institutions themselves amount to approximately half of the overall contributions for the five diseases in focus.
For Germany, 220 different research institutions were identified as author affiliations in publications on the five NTDs under study. Ranking individual research institutions by the number of publications in which they were named as affiliations uncovered variations between the different NTDs. Among the top five German research institutions for each of the diseases, there were 12 different institutions. The Charité—Universitätsmedizin Berlin was the most frequently quoted affiliation for Chagas disease and Leishmaniasis research, whereas the Ruprecht-Karls-University of Heidelberg was the most quoted affiliation for Schistosomiasis and Sleeping Sickness research. The Bernhard-Nocht-Institute for Tropical Medicine in Hamburg was the most quoted affiliation for Onchocerciasis and the most quoted affiliation overall. Outside of Germany, 630 different research institutions were identified as affiliations in publications that had at least one co-author affiliated with a German research institution. Non-German research institutions that have frequently collaborated with German research institutions were also found among the most frequently named affiliations in the German NTD research network (Table 4).
Individual researchers were identified by their number of co-authored publications, h-index and betweenness centrality (Table 5). For each disease network, the comparison of authors by either their number of publications or h-index revealed a similar top five ranking. Author results for betweenness centrality (see Table 5) also appeared to be related to the traditional indicators, as mentioned above, for the majority of leading authors in each disease network. This finding allowed us to identify leading researchers in each research field. However, betweenness centrality also enabled the identification of authors who would not have been identified with traditional indicators. For example, 13 of the 25 authors among the top five for betweenness centrality were not among the top five for any of the traditional ranking parameters. The number of authors who contributed to each disease-based co-authorship network ranged from 446 (Onchocerciasis) to 1904 (Leishmaniasis).
The co-authorship networks organised by disease differ in terms of their density, number of components, share of authors who were part of the giant components and average degree (for an explanation of social network analysis terminology, see Table 1). In general, the Onchocerciasis and Leishmaniasis networks displayed similar characteristics, despite the former being much smaller. By contrast, the Chagas network displayed quite different network characteristics; for example, the Leishmaniasis network was the most dense (Density 0.09) and the Chagas disease network (0.01) was the least dense (Table 5). Similarly, the number of individual network components that are not connected to each other range from 18 (Onchocerciasis) to 67 (Chagas disease).
We also identified one giant component for each of the networks (Fig 4). Each network’s giant component represented a varying number of authors ranging from 24.7% (Chagas disease) to 80.6% (Leishmaniasis) of all the authors in each network. For Chagas disease, this finding led to a large proportion of authors not being included in the giant component (Fig 5), but instead they were scattered among a large number of components. By contrast, the Leishmaniasis network included a similar number of components as the Chagas network despite the fact that the former included twice as many authors but a far higher proportion of authors were represented by the giant component (Table 5).
Likewise, both the Onchocerciasis and Leishmaniasis networks had a high average degree (17.0 and 19.9, respectively), which indicated that the authors in these networks have published with a large number of co-authors. This finding contrasted with the low average degree of the Chagas disease network (9.07), which indicated that those authors published with fewer co-authors on Chagas disease.
We found that until the present, the NTD research share of publications with German affiliations was considerably lower relative to medical research in general. Hence, it appears that German NTD research is lagging behind the nation’s role in other fields of health research. For research on the 11 diseases included in our study, 4.4% of global NTD publications were affiliated with Germany. This percentage is less than half the share of German research output (9%) for all research publications in the medical field [31]. Regarding NTD research specifically, Germany is not performing as strongly as other countries that show similar characteristics in other fields of research, such as the United Kingdom or France [10]. This lack of comparative strength for Germany has already been acknowledged by the German government in their recent health research framework program [32].
An analysis of co-authorship patterns with research partners abroad revealed that the majority of all contributions come from HIC, and they are dominated by the United Kingdom and the United States; the next most prolific collaborating country was Brazil. The total number of co-authorship collaborations with partners in LICs is 10 times lower than it is for HICs. The research landscape in Germany does not have clearly dominant NTD research institutions but instead spreads its endeavours across different individual universities and research units.
Research cooperation with low- and middle-income countries has been the focus of the German government by facilitating collaboration and capacity strengthening through their support for initiatives such as the European Clinical Trial Development Partnership (EDCTP), and yet they have not been systematically analysed nor specifically analysed with social network analysis tools. Our analysis of German NTD researcher co-authorship patterns revealed that publication output remains dominated by the global north, and it is particularly dominated by collaborations of researchers from well-known research powerhouses such as the United States and the United Kingdom.
The middle-income country with the most productive collaborating researchers was Brazil, with more co-authorships than all other BRICS countries combined, and except for Onchocerciasis research, Brazilian researchers performed strongly in all other diseases analysed here. This finding reflects the Brazilian government’s strategic commitment to support NTD research [27], and it is even more remarkable considering that not all of the five diseases on which we focused here have high prevalence in Brazil [33].
Overall, the total number of co-authorships from German research partners in low-income countries was meagre, being only a tenth of the number of those from high-income countries. Analysing co-authorships between researchers on the African continent and Germany specifically showed that there are only a few very active collaborations. For all diseases, the ratio between the numbers of collaborations from the continent per number of contributing countries is lower for Africa than for Europe, North America and South America, which indicates that collaborations among high- and middle-income countries focus on a few key research partners, and collaborations with researchers from the African continent were more dispersed.
This finding suggests that capacities for research collaborations exist within a range of different countries, even though the overall number of contributions (as with signatures for co-authorships) from low-income countries remained comparatively low. The analysed publications showed collaborations between many authors within high-income countries and only a few with single authors or institutions in low-income countries. Other findings showed that among publications with African co-authors, the largest number of authors still came from countries in the global north [34]. Our data suggest that co-authorship analysis could further help identify targets for much-needed research capacity strengthening [35] and spur research productivity through diversification via collaboration [36]. In the German context, these findings are particularly pertinent because they could drive policy making for research capacity strengthening through programs that are already in place, such as the Research Networks for Health Innovations in the Sub-Saharan Africa Initiative. [18]
As expected, almost all of the publications were written in collaboration between several co-authors, with the research being published in a broad array of journals and PLoS NTD leading, though it was only launched during the study period.
When analysing the NTD publication output at the disease level, even though the actual publication output in Germany is lower compared to other high-income countries, the focus on Leishmaniasis, Schistosomiasis and Chagas disease research in Germany is similar to that of international NTD research. However, the relatively higher publication output on Sleeping Sickness and Onchocerciasis is at a fifth for authors with an affiliation to Germany compared with a tenth internationally, which could indicate comparatively strong research expertise in Germany that is worthy of additional support.
Our analysis of the German NTD research landscape revealed that no single research institution in Germany is dominating or leading NTD R&D, but the research is spread across different individual universities and publicly funded research entities. This finding differs from NTD research in other high-income countries, where research efforts are more concentrated within fewer institutions. Although the publicly funded Bernhard-Nocht-Institute for Tropical Medicine in Hamburg showed the highest number of co-authorships, it is closely followed by a number of universities with research foci on different NTDs. This finding likely reflects the federal system in Germany with its rather broad approach to university-based research and education compared with the tradition of more centralised structures that exist in other countries. In addition to the relatively small amount of NTD research in Germany, this fragmentation adds to the rather low international visibility of German global health research [19,37].
In contrast to the institutional environment, an analysis on the level of individual researchers suggested that German NTD R&D network hubs were dominated by a few individuals. This finding was substantiated by the fact that some leading authors were ranked highly in terms of both traditional bibliometric indicators and betweenness centrality. When analysing co-author research networks, which include German researchers and their partners abroad, we found that among the top researchers identified through traditional bibliometric indicators such as the number of publications or h-index, one (Schönian G) is an emeritus researcher and one (Büttner D) has already passed away [38]. Brun R, despite being a collaborator who works at the Swiss Tropical and Public Health Institute, has contributed substantially enough to the German NTD research network, making him the leading researcher for both Sleeping Sickness and Chagas disease.
It appears that a considerable amount of NTD R&D expertise is held among German NTD researchers approaching retirement age, and therefore the field is at risk for capacity and expertise loss. This finding emphasizes the need for a knowledge transfer to a younger generation of researchers.
It has been suggested that applying social network analysis to research, and using betweenness centrality in particular, supports the identification of researchers who are most likely to produce a higher h-index in the future, through the analysis of today’s research network structure [39]. Through the identification of individual researchers within the networks that already show a high betweenness centrality, and not yet having established a high number of publications or a high h-index, social network analysis could facilitate the identification and hence the targeting of support for younger, well-connected researchers that have not yet accumulated the years of experience and publications that bias traditional indicators towards older academics.
Stratifying the German NTD research landscape by disease allowed the identification of noticeable differences in co-authorship networks for the different diseases. These differences highlighted specific expertise and the most productive research collaborations, which might be worthy of particular support. For example, the Leishmaniasis and Onchocerciasis networks were characterized by a large number of collaborations between authors, as indicated by a high average degree and network density. Although this finding was no surprise with regards to Leishmaniasis, which was by far the most researched disease of those analysed here, it was more remarkable for Onchocerciasis, supporting the evidence that this disease could be a comparative strength in German NTD research. Conversely, co-authorship network analysis allowed the identification of ‘gaps’ or missed opportunities, e.g., the Chagas disease research network showed great potential because of its strong contributions from middle-income countries such as Brazil, but the network remained scattered among many components at a low density, which indicates that there is room for improved collaborations between the actors who are already involved.
It is of further concern to us that the potential looming public sector NTD R&D capacity loss among aging researchers in Germany is mirrored by the current near-absence of infectious disease R&D capacities within German pharmaceutical companies, which were once global leaders in infectious and neglected tropical disease research (a fact that was even used as propaganda under dubious circumstances [40]).
We only used data from the SCOPUS bibliometric database, which was found to have the widest coverage of NTD literature [41]. Future research network analyses should consider the exploration of other literature databases, for example, Web of Science and MEDLINE, to identify additional publications.
Not including agent names in our search string may have limited the number of hits related to basic research studies; however, we wanted to focus our search strategy on disease names to consider the most relevant outcomes from the perspective of patients who were affected by NTDs.
Although network analysis tools are manifold, it is important to note that we use quantifiable data such as publication output as the best available proxy measure for researcher knowledge or expertise. Additionally, the value of information gained here can only be as good as the data that is available for comparison. We considered this not only a limitation but also a call for further investigation into the structure of other NTD research networks around the world. Because our method is based solely on open source software, it can easily be reproduced in other contexts and might help to put our findings into a broader perspective.
Current German government policy clearly pursues an increasing role in global health, and recent studies have acknowledged a growth in German public sector funding [22], even labeling Germany as an ‘emerging leader’, though its own funding program for NTD research expired at the end of 2014 and a renewed call for proposals is pending for 2015. It remains to be seen if and how the German government's political will, as expressed for example by putting neglected and poverty-related diseases on the agenda for the G7 Summit in Germany in June 2015 [42], is going to be reflected in measurable research output from Germany. As it hands over the G7 presidency to Japan, we urge the German Government to make good on the promises made in the G7 leaders' declaration [43].
Our findings underline the G7 national academies' of science call for policy changes [44], particularly for promoting research collaborations and technology transfer in LMIC and to intensify research within the G7 countries themselves.
Providing process and output-based insights in NTD R&D, such as those provided here, will have an important role in the realisation of the G7 goals and for WHO’s Global Health Research and Development Observatory [45].
The first systematic assessment of the German health and medical research landscape for NTD using authorship networks based on bibliometric metadata demonstrated not only the potential of social network analysis as a tool to apply to the R&D field, but it also revealed valuable findings when used to assess German research capacities in selected NTDs. Our findings showed that 4.4% of all NTD publications worldwide involve an author from a German research institution. This rather low output of German R&D activities on NTD is scattered across numerous publicly funded research institutions without single outstanding centres. Most publications that included researchers from Germany were related to other high-income countries and the emerging economy in Brazil.
Our results could contribute to identify research strengths that can be enhanced, e.g., by expanding targeted collaborations for research capacity building in LMIC, or for weaknesses to amend, for example, through encouraging collaboration in areas of shared expertise that were missed until now.
Future research should provide further in-depth analysis of individual researcher and network productivity, scientific impact and translational success in the development of new products for NTDs. Similar analyses could also include qualitative approaches (e.g., focus groups or semi-structured interviews) with key researchers and policy makers to identify barriers, e.g., limiting factors for collaboration with partners in low- and middle-income countries.
Notwithstanding the apparent political will of the current German government, our network analysis shows that NTD R&D in Germany is scattered and at risk of expertise loss. For a renewed German NTD research-funding program, it appears to be crucial to analyse the existing R&D landscape empirically to inform future research funding decisions. This analysis could be strengthened through innovative tools such as network analysis. Mapping research collaborations with partners abroad can support decisions on the selective strengthening of research capacity. Furthermore, a social network analysis could provide valuable insights into which specific diseases could be prioritised based on where comparative advantages in research networks are found. This direction is essential for developing a data-driven research strategy to expand Germany’s research activities in the field of NTDs.
|
10.1371/journal.ppat.1002066 | Lymphoadenopathy during Lyme Borreliosis Is Caused by Spirochete Migration-Induced Specific B Cell Activation | Lymphadenopathy is a hallmark of acute infection with Borrelia burgdorferi, a tick-borne spirochete and causative agent of Lyme borreliosis, but the underlying causes and the functional consequences of this lymph node enlargement have not been revealed. The present study demonstrates that extracellular, live spirochetes accumulate in the cortical areas of lymph nodes following infection of mice with either host-adapted, or tick-borne B. burgdorferi and that they, but not inactivated spirochetes, drive the lymphadenopathy. The ensuing lymph node response is characterized by strong, rapid extrafollicular B cell proliferation and differentiation to plasma cells, as assessed by immunohistochemistry, flow cytometry and ELISPOT analysis, while germinal center reactions were not consistently observed. The extrafollicular nature of this B cell response and its strongly IgM-skewed isotype profile bear the hallmarks of a T-independent response. The induced B cell response does appear, however, to be largely antigen-specific. Use of a cocktail of recombinant, in vivo-expressed B. burgdorferi-antigens revealed the robust induction of borrelia-specific antibody-secreting cells by ELISPOT. Furthermore, nearly a quarter of hybridomas generated from regional lymph nodes during acute infection showed reactivity against a small number of recombinant Borrelia-antigens. Finally, neither the quality nor the magnitude of the B cell responses was altered in mice lacking the Toll-like receptor adaptor molecule MyD88. Together, these findings suggest a novel evasion strategy for B. burgdorferi: subversion of the quality of a strongly induced, potentially protective borrelia-specific antibody response via B. burdorferi's accumulation in lymph nodes.
| Acute Lyme Disease is one of the most important emerging diseases in the US. People with acute Lyme disease often develop swollen lymph nodes, or lymphadenopathy, but we do not know why this happens or what effect it has on the course of the disease. We show here that when mice are infected with live Borrelia burgdorferi spirochetes (the bacteria that cause Lyme disease), live spirochetes collect in the lymph nodes. These lymph nodes then swell up and start producing large numbers of antibody-producing cells. Although many of these antibodies can recognize the bacteria, they apparently lack the quality to clear the infection. We hypothesize that by moving into the lymph node, usually a site in which strong immune responses are induced, Borrelia evades the immune response: it goes to the lymph nodes and tricks the immune system into making a very strong but inadequate response.
| Lyme borreliosis, caused by Borrelia burgdorferi transmitted by Ixodes spp. ticks, is the most common arthropod-borne illness in the US and Europe, and is increasing in prevalence and expanding in geographic distribution in the US [1], [2]. Clinical manifestations are highly varied, including involvement of the cutaneous, cardiovascular, musculoskeletal, and nervous systems [3]–[5]. A frequent, but largely under-studied manifestation is massive and systemic lymph node enlargement (lymphadenopathy), observed particularly in the regional lymph node near the site of infection in humans, and in experimentally-infected dogs [4], [6]. The lymph node enlargement that arises in both humans and dogs is characterized by increased cellularity and the accumulation of large pleomorphic IgM- and IgG-positive plasma cells [6]–[8]. Despite these unusual characteristics, the lymphadenopathy of Lyme borreliosis has not been well investigated.
Several in vitro studies have shown that culture-grown B. burgdorferi can act as mitogens when co-cultured with human or murine naive B cells [9]–[16]. Therefore, the unusual lymphadenopathy of Lyme borreliosis might be a manifestation of non-specific B cell activation. Massive lymph node enlargement has also been seen in wildtype but not TLR4 gene-targeted mice during infection with Salmonella typhimurium [17] and others have shown a role for TLR-independent, TNF-independent [18] or TNF-dependent [19] involvement of mast cells in non-specific induction of lymph node enlargement. Thus, innate immune activation might account for the lymphadenopathy observed during infection with B. burgdorferi.
On the other hand, there is ample evidence for the induction of specific immune responses following B. burgdorferi infection. Both following experimental and natural infections, B. burgdorferi-specific IgM and IgG antibodies are induced in the serum of infected humans [5], [20]–[24], dogs [25], and mice [26], among other host species. Importantly, passive transfer of immune-serum from chronically infected wildtype or T cell-deficient mice, from naturally infected dogs, and from human patients with chronic Lyme disease can protect mice from a challenge infection with B. burgdorferi [26]–[29], demonstrating that specific and protective antibodies are induced during the course of infection. However, once infection is established, the immune response is incapable of clearing infection [26], [30]. Thus, understanding the host immune response is critical to understanding and treating Lyme borreliosis.
The present study was undertaken to identify the mechanisms involved in the lymphadenopathy induced by infection with B. burgdorferi and to determine the nature and specificity of the reactive B cell response. Using a mouse model of infection with host-adapted spirochetes that faithfully recapitulates experimental and natural infections with ticks, we show that B. burgdorferi actively migrates into the lymph nodes, where it causes a largely specific, but unusual B cell response.
Four to six week old female C3H/He, C57BL/6 and severe combined immunodeficient C57BL/B6.C-Prkdcscid (SCID) mice were obtained from The Jackson Laboratory, Bar Harbor, ME, and maintained at UC Davis in isolator cages under conventional housing conditions. Breeding pairs of C57BL/6.129P2/Ola-MyD88tm1Aki (MyD88 −/−) mice [31] were a generous gift of Richard Flavell (Yale University), given with kind permission from Shizuo Akira (Osaka University). The MyD88−/− mice were rederived and bred in the specific pathogen free barrier facility at UC Davis, and then transferred to conventional housing prior to experiment onset.
Mice were infected with B. burgdorferi in two ways: for tick-borne infections, five B. burgdorferi-infected nymphal ticks (or non-infected control ticks) were placed on the dorsal thoracic midline of mice and allowed to attach and feed to repletion. To generate host-adapted B. burgdorferi, SCID-mice were infected s.c. via syringe inoculation with 104 B. burgdorferi spirochetes grown to mid-log phase (day 5 of culture) in 0.1 ml of sterile medium. For infection with host-adapted spirochetes, 3 mm2 punch biopsies from infected SCID mice were obtained from the hairless, ethanol-cleaned ear pinnae. Biopsies were transplanted subcutaneously on the lateral side of the right tarsal joint of recipient naïve C57BL/6 mice. Ear transplants contained a mean of 1.8×104 spirochetes, based upon quantitative DNA analysis [32]. Control mice were transplanted at the same location with similar tissue from uninfected SCID mice (sham infection).
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All protocols involving animals were approved by the Animal Use and Care Committee at UC Davis (Permit Number: #15330).
A clonal strain of B. burgdorferi sensu stricto (cN40) was grown in modified Barbour-Stoenner-Kelly (BSK II) medium [33] at 33°C and enumerated with a Petroff-Hauser bacterial counting chamber (Baxter Scientific, McGaw Park, IL). Heat-inactivation of B. burgdorferi was done at 56°C for one hour followed by sonication. Aseptically collected samples of lymph nodes, spleen, inoculation site, and urinary bladder were taken at necropsy and cultured for 7 and 14 days in BSK II medium to assess the presence of spirochetes under dark field microscopy.
Uninfected larval Ixodes scapularis ticks were obtained from field-collected adults in southern Connecticut (kindly provided by Durland Fish, Yale University). All larvae for the experiments described in this study were derived from a single cohort. A sample of the cohort was confirmed to be B. burgdorferi flaB negative by PCR. To generate infected nymphs, larvae were allowed to engorge on C3H mice that had been infected with B. burgdorferi for 2 weeks following syringe inoculation, as described previously [34]. Following feeding and molting, cohort analysis of the infected nymphal ticks revealed that 97% of the ticks were confirmed to be PCR positive for B. burgdorferi flaB as previously described [34].
Lymph nodes were fixed in neutral buffered formalin, embedded in paraffin and sectioned at 4 mm and stained with hematoxylin and eosin or by immunohistochemistry. Sections for immunohistochemistry were processed at room temperature and placed on positively charged slides, air-dried, de-paraffinazed and re-hydrated. Endogenous peroxidase activity was eliminated by incubation in 3% H2O2 in methanol for 20 minutes. Non-specific binding was reduced with biotin blocking solution (Vector) for 15 minutes and Power Block (InnoGenex) for 15 minutes. Immunohistochemical labeling of B. burgdorferi was performed by treating sections with 0.5 mg/ml protease type VIII (Sigma Aldrich) for 10 minutes, followed by 30 minutes incubation with 1∶1000 dilution of a polyclonal immune serum from B. burgdorferi-infected rabbits (infected for two months following inoculation with 104 spirochetes). Antigen detection utilized a three-step streptavidin-horseradish peroxidase technique with the substrate DAB (Vector). For other antigens, antigen retrieval was enhanced by microwaving tissue sections for 6 minutes in citrate buffer at pH 6.0. Sections were then incubated with antibodies to B220 (CD45R, RA3-6B2), CD138 (281-2, BD Biosciences), or Ki-67 (NeoMarkers), followed by incubation with biotinylated secondary antibodies (Vector), streptavidin conjugated Alexa 488 and Alexa 594 (Molecular Probes) or streptavidin-horseradish peroxidase followed by DAB (Vector), and mounting with Prolong Antifade (Molecular Probes).
Live cell counts of single cell suspensions of lymph nodes were obtained using a hemocytometer and trypan blue exclusion of non-viable cells. Staining was performed using aliquots of 6.25×105 cells in “staining medium” (buffered saline solution: 0.168 M NaCl, 0.168 M KCl, 0.112 M CaCl2, 0.168 M MsSO4, 0.168 M KH2PO4, 0.112 M K2HPO4, 0.336 M HEPES, 0.336 M NaOH, containing 3.5% heat-inactivated, filtered newborn calf serum and 1 mM EDTA) for 20 min on ice. The following antibody-conjugates were used at previously determined optimal concentrations: CD19-Cy5PE, CD3-APC Efluor780 (both e-biosciences), CD4-FITC, and CD8a-Cy5.5PE (both in-house generated) after blocking Fc receptor with anti-CD16/32 (2.4G2). Dead cells were discriminated with a live/dead violet staining kit (Invitrogen). Data acquisition was performed on a 13-color FACSAria instrument (BD Biosciences) [35]. Data were analyzed using FlowJo software (kind gift from Tree Star Inc.).
To probe for B. burgdorferi-specific antibody-producing cells by ELISPOT, 96-well plates (#MAHAS4510, Mixed Cellulose Ester Membrane; Millipore) were coated with 2.5 µg/mL of four recombinant non-lipidated B. burgdorferi N40 proteins: decorin binding protein A (DbpA), outer surface protein C (OspC), arthritis-related protein (Arp), and borrelia membrane protein A (BmpA) in PBS overnight. After blocking with PBS/4% BSA, lymph node cell suspensions were 2-fold serially diluted in medium (RPMI 1640, 292 µg/mL L- glutamine, 100 µg/mL of penicillin and streptomycin, 10% heat inactivated FCS and 0.03 M 2-ME) and cultured overnight at 37°C with 5% CO2. Cells were lysed with water and binding was revealed by incubation with biotin conjugated anti-IgM (Southern Biotech) or anti-IgH+L (Southern Biotech) for 2 hours in 2% BSA in PBS. This was followed by SA-HRP incubation for 1 hour (Vector Laboratories) in PBS/2% BSA and by 3-amino-9-ethylcarbazole (Sigma-Aldrich). Plates were washed and dried and mean spots were counted in all wells with visible spots and calculated as mean spot numbers per input cell number.
Genes encoding non-lipidated B. burgdorferi N40 proteins, previously identified by genomic expression library analysis to react with serum from B. burgdorferi-infected mice, as described [36], were amplified by PCR from B.burgdorferi N40 DNA using oligonucleotide primers based on their DNA sequences (Supplemental Table S1). Template DNA from the original reactive clone was denatured at 94°C for 1 min, annealed at 55°C for 1 min, and extended at 72°C for 1 min. This process was repeated for 30 cycles. The amplified genes were cloned in frame with the glutathione S-transferase (GT) gene into pMX, derived from a pGEX-2T vector (Pharmacia, Piscataway, N.J.) with a modified polylinker. The PCR-amplified DNA sequences were confirmed by sequence comparison with the original inserts.
E. coli DH5α cells transformed with the recombinant pMX vectors were grown to an optical density of 0.5 at 600 nm and the recombinant GT fusion proteins were induced with 1 mM IPTG for 2 h. Bacteria were centrifuged at 3,310 g for 20 min, pellets were washed with PBS and bacteria lysed with PBS/1% Triton X-100. The mixtures were sonicated and centrifuged at 35,000 g. Supernatants containing recombinant proteins were loaded onto glutathione-Sepharose 4B columns (Pharmacia), 25 U of thrombin was added to remove the GT partner, and purified proteins were eluted after 2 h.
B. burgdorferi was grown to log-phase (8–10 days), pelletted by centrifugation, resuspended in cold PBS plus MgCl2 and centrifuged repeatedly for 5 min at 4°C 17,500 g. Samples were stored in aliquots at −20°C. Protein concentration was determined using Bradford assay (Bio-Rad).
B cell hybridomas were created from enlarged lymph nodes collected at various times after infection with host-adapted B. burgdorferi. Three independent fusions were performed using standard protocols. Briefly, single cell suspensions from mechanically disrupted lymph nodes were fused with P3-X63Ag8.653 mouse myeloma cells (ATCC CRL-1580) using PEG 1450 (ATCC). Hybridomas were selected by incubating cells in HAT medium. Supernatants of all wells with visible cell growths were screened by ELISA for the presence of mouse Ig as previously described [37]. Hybridoma lines were established from all Ig-producers and tested further for reactivity against B. burgdorferi- specific recombinant antigens and whole B. burgdorferi lysate. Some hybridomas were then subcloned. The Ig heavy and light chain isotype profiles of the lines and clones were determined using the Mouse Immunoglobulin Cytometric Bead Array Kit, (BD Biosciences, Cat Number 550026).
Statistical analysis was performed using the two-way ANOVA or Student's t-test with help of Prism 5 software (GraphPad Software). A p-value of <0.05 was considered statistically significant.
Since lymphadenopathy has not been documented in laboratory mice following B. burgdorferi infection, we first sought to determine if and when lymphadenopathy developed in laboratory mice infected experimentally with B. burgdorferi via the natural route, i.e. via tick-bite. For that, B. burgdorferi-genetically susceptible C3H/He mice [38] , were each infested with either 5 infected nymphal ticks or with 5 uninfected nymphal ticks (sham-infected). All ticks were placed on the dorsal cervico-thoracic midline. However, the ticks subsequently migrated and attached to different regions of the body, particularly the head and neck region. The most common tick attachment sites were the ear pinnae and face.
Axillary, brachial, lumbar and inguinal lymph nodes, among others, were collected at various times after infection and examined for visible signs of enlargement (not shown) and to determine cell number counts. Lymph node enlargement was noticed for all lymph nodes from mice exposed to B. burgdorferi infected ticks but not uninfected ticks (Figure 1A and data not shown). By day 14 following infestation with infected ticks, the lymph nodes closest to the tick-attachment site (axillary and brachial) were visibly enlarged and contained significantly increased numbers of cells in comparison to the same lymph nodes collected from the sham-exposed mice. Lymph nodes more distant from the attachment site (inguinal and lumbar) showed a slightly delayed increase in cellularity (Figure 1A). Thus, infection of laboratory mice with tick-borne B. burgdorferi faithfully recapitulates the lymphadenopathy observed in naturally infected humans and dogs, and suggests a relationship between time of lymph node enlargement and proximity to the site of infection.
To directly assess the spatio-temporal relationship between the kinetics of the lymph node enlargement and the site of B. burgdorferi infection, a different infection modality was needed. Ticks change their attachment location in ways that varied significantly between mice, precluding targeted analysis of specific lymph nodes. Direct inoculation of mice with culture-grown B. burgdorferi, on the other hand, introduces untoward experimental variables due to the significant antigenic changes that B. burgdorferi undergoes as it adapts to the vertebrate host. One example is the antibody-response to the major outer surface protein A (OspA), which is strongly expressed in vitro and in ticks [39], but virtually absent in mice infected with B. burgdorferi via tick-infestation or following transplantation of tissue from infected mice containing host-adapted spirochetes [27]. Thus, infection via injection of culture-grown bacteria may favor distinct immune responses that differ from those seen after tick-infection. We therefore transplanted punch biopsies of ear pinnae from infected SCID mice, containing host-adapted spirochetes under the skin of the right tibiotarsus area of congenic, naïve C57BL/6 mice. The right inguinal lymph nodes were evaluated as the regional lymph nodes.
Infection of C57BL/6 mice with host-adapted B. burgdorferi resulted in a rapid enlargement of their regional inguinal lymph nodes (Figures 2A, 2B). These increases closely resembled the lymphadenopathy observed at the site of tick-attachment following tick-borne infection, albeit with somewhat faster kinetics (Figure 2C), possibly due to the increased time between tick-attachment and actual infection, and/or the time it takes for B. burgdorferi to adapt to the host-environment prior to dissemination [34]. Similar to tick-borne infection, infection with host-adapted spirochetes caused a generalized lymphadenopathy, with lymph nodes more distant from the infection-site increasing slower in cell numbers compared to those closest to the site of infection (Table 1). Thus, this infection model faithfully recapitulated tick-borne B. burgdorferi-induced lymphadenopathy with the advantage that we can consistently identify the lymph nodes draining the site of infection. The spleen was not increased in size or cellularity following either tick-borne infection (not shown) or following infection with host-adapted spirochetes (Table 1).
Since proximity to the infection site was correlated with an increase in lymph node cellularity, we investigated next if and when B. burgdorferi could be cultured from the lymph nodes. Within 24 h following infection with host-adapted spirochetes, B. burgdorferi was cultured from the closest draining right inguinal lymph nodes, but not any other lymph nodes (Table 1). By 48 h, the right lumbar lymph nodes became culture-positive. The right axillary lymph nodes yielded positive culture results two days later and before any of the contralateral lymph nodes on the left side of the mouse. A few days after the lymph nodes became culture-positive (about 4–6 days), a marked increase in cellularity of the lymph nodes was consistently observed for all lymph nodes, but not to the degree as the most proximal regional lymph nodes (Table 1). Once culture-positive, the lymph nodes remained so for the 90-day study period. Culture results from the spleen did not reveal B. burgdorferi until day 10 and then also only intermittently thereafter (Table 1).
These data suggested that the lymphadenopathy observed during Lyme borrreliosis is caused by a massive increase in lymph node cellularity triggered by the accumulation of live B. burdorferi spirochetes into the lymph nodes. Alternatively, it was possible that the culture results were a mere reflection of the presence of B. burgdorferi in the lymph node capsule, given that the spirochete travels along connective tissues. In that case, the increase in lymph node cellularity would be an indirect consequence of the infection-induced inflammation rather than the presence of the spirochetes in the lymph nodes. To distinguish between these possibilities, immunohistochemistry was utilized to determine the precise tissue-location of the spirochetes in lymph nodes of mice infected for 8 days with host-adapted B. burgdorferi. The results demonstrated the consistent presence of B. burgdorferi spirochetes in the sub-capsular sinus and superficial cortex of infected lymph nodes (Figures 3A, 3B). Interestingly, the spirochetes were found in the lymph nodes extracellularly appeared intact with characteristic spiral morphology. Together with the results from the culture experiments (Table 1), this suggests a degree of persistence of B. burgdorferi in lymph nodes.
Because lymphadenopathy was correlated with the presence of viable spirochetes, we determined next if lymphadenopathy could also be induced with inactivated spirochetes. For that, mice were either infected directly by subcutaneous inoculation in the right tarsal region with 104 viable cultured spirochetes or by the same number of spirochetes after inactivation by sonication. Inactivation was confirmed by culture of the sonicate. Right inguinal lymph nodes were collected at days 0, 10 and 21 days post infection and either cultured in BSK II medium or assessed for cellularity. At days 10 and 21 post infection, B. burgdorferi was cultured from the right inguinal lymph node of mice infected with live spirochetes but as expected, not from mice inoculated with inactivated spirochetes. Importantly, increases in lymph node cellularity were not observed in mice receiving inactivated B. burgdorferi, but were clearly induced in mice inoculated with viable B. burgdorferi (Figure 3C).
Since B. burgdorferi might replicate in vivo and thus the results might reflect application of differing amounts of bacteria or bacterial antigen between these two groups, the analysis was repeated by giving 100-fold higher amounts of inactivated bacteria (106 organisms). While the increased amount of inactivated bacteria resulted in lymph node enlargement compared to control mice (Figures 3C, 3D), the enlargement was significantly less (p = 0.001) than that seen with viable Borrelia (Figure 3D). Thus, we conclude that lymphadenopathy during B. burgdorferi infection is caused by the accumulation of viable spirochetes in lymph nodes.
Next, the cause of the increase in cellularity of the lymph nodes was investigated. Immunohistochemistry on day 10 after infection demonstrated stark differences in lymph node organization compared to lymph nodes from uninfected mice (Figures 4A, 4B). Morphologically, the cortex of infected regional lymph nodes consisted of tightly packed extrafollicular lymphocytes and very few scattered, poorly demarcated germinal centers without a distinct mantel zone. Indeed follicular structures appear largely absent in these lymph nodes (Figures 4B, 4C). The majority of the lymphocytes in the cortex were positively labeled by B220 and thus identified as B cells (data not shown). The largest extrafollicular B cells were frequently arranged in distinct clusters interpreted as antibody forming foci. These large B cells were characterized by an open euchromatic nucleus with marginated chromatin and a large prominent nucleolus and moderate amounts of cytoplasm, characteristics of plasmablasts.
To distinguish between increased trapping of migrating cells versus expansion of cells within the regional lymph nodes, cells were labeled for Ki-67 antigen to identify proliferating cells. The staining demonstrated large numbers of B cells in the cortex that were actively dividing (Figure 4D). Cells in the paracortex that were negative for B220 rarely expressed Ki-67, suggesting that cell division was restricted to the B cell population only (data not shown). Dual fluorescent labeling of Ki-67 and B220 identified the dividing cells as B cells (Figure 4E). The exclusive and extensive expansion of B cells was further confirmed by flow cytometry. Whereas the lymph nodes showed no significant increases in either CD4 or CD8 T cells compared to lymph nodes from non-infected mice, CD19+ B cells had expanded dramatically by day 10 of infection with live B. burgdorferi (Figure 4G). The massive increases in B cells were absent in mice injected with inactivated B. burgdorferi (Figure 3E).
In addition, the medullary cords of the lymph nodes from the infected mice showed the presence of large numbers of plasma cells, identified by staining with CD138 (Figure 4F), suggesting a strong induction of antibody secretion in the affected lymph nodes. Indeed, ELISPOT analysis on regional lymph nodes infected for up to 60 days with B. burgdorferi showed the presence of large numbers of antibody-forming cells (AFC), with peak responses noted around day 10 of infection (Figure 4H). Depending on the day of study between 1–3% of the AFC secreted antibodies were bound to the Borrelia lysate (Figure 4H). The strong antibody secretion within the lymph nodes following infection with host-adapted spirochetes was in magnitude and kinetics very similar to the induction seen following tick-borne infection with B. burgdorferi (Figure 1B), but was much larger than seen after injection of inactivated bacteria (Figure 3F). In summary, expansion of the lymph node cortex by reactive B cells and extrafollicular antibody forming foci constitutes the morphological basis of lymphadenopathy in Lyme borreliosis.
The finding of active migration of B. burgdorferi into lymph nodes, i.e. an organ responsible for immune response induction, appeared counter-intuitive for an organism that aims to establish persistent infection. Therefore, we aimed to determine next whether B. burgdorferi might cause immune subversion in these lymph nodes. In particular we asked whether it was inducing massive non-specific B cell expansion and differentiation to antibody-secreting cells at the expense of an effective Borrelia-specific antibody response. Addressing this question is complicated by the fact that protein expression of culture-grown spirochetes does not fully resemble Borrelia in the host, i.e. the usefulness of protein lysates from culture-grown bacteria is limited as a source of antigen for ELISPOT analysis.
Initial studies were therefore conducted to identify a number of Borrelia antigens that are expressed in the host and induce robust antibody responses. A screen of available recombinant Borrelia-expressed antigens by ELISPOT analysis with lymph node cells from day 14 Borrelia-infected mice showed that lymph nodes had measurable reactivity against all of the recombinant antigens tested. Interestingly, DbpA had the highest level of reactivity, while the Borrelia lysate, included as a “positive” control, identified a much smaller fraction of Borrelia-reactive AFC (Figure 5A). Further analysis showed that it was possible to pool various Borrelia antigens for ELISPOT analysis without losing sensitivity of reactivity against each antigen (data not shown). A pool of four recombinant antigens, consisting of DbpA, OspC, Arp, BmpA was used as a means of measuring the Borrelia-specific antibody response. While it clearly underestimates the number of total Borrelia-specific responses, testing with the pool of recombinant proteins that are expressed during infection was found to be more sensitive than testing with Borrelia lysate.
A time course analysis of C57BL/6 mice infected for up to 90 days with host-adapted B. burgdorferi showed the robust induction of antibody-secretion within the regional lymph nodes (Figure 5B). The kinetics of the Borrelia-specific response was identical to that of the total antibody responses measured at the site (compare Figure 4H with Figure 5B). Depending on the day of analysis between 4–13% of AFC were shown to be specific for one of the four recombinant proteins included as antigens in the analysis. The isotype profile of the specific response showed a broad representation of all measured isotypes. More than half of the antibody-secreting cells appeared to generate IgM antibodies and IgG antibody isotypes classically associated with T-independent responses (IgG2b and IgG3, Figure 5C). Overall, the isotype profile of the Borrelia-specific response suggested that a considerable proportion of the B cell response might be T-independent, consistent with previous observations [29], [40].
The ELISPOT results suggested that a significant fraction of the induced B cell response was specific and directed against B. burgdorferi. However, given that we probed with only some of the many other Borrelia proteins that are potentially expressed selectively in vivo, assessment of the relative contribution of the specific over the non-specific response was difficult. Therefore, another series of experiments was conducted in which hybridomas were generated from the regional lymph nodes to assess the fraction of hybridomas directed against Borrelia-specific antigens with an expanded list of recombinant Borrelia proteins. Three successful fusions were conducted, including one on lymph nodes at day 8 of infection, and two on day 18. The overall results from these three fusions were similar (Figure 6A). Initial screening of roughly 1000 wells per fusion identified between 150–350 wells that showed antibody-secretion. Further screening of the antibody-secreting lines indicated that between 14–24% of the hybridoma lines generated antibodies that could be identified to react against an expanded list of recombinant Borrelia antigens (DbpA, OspC, Arp, BmpA, P23, P29, P32, P61 (defined in Supplemental Table S1) [41] and/or Borrelia lysate from cultured spirochetes.
From these fusions, 132 hybridoma lines were cultivated. Of these, 45 were specific for B. burgdorferi antigens. The isotype profile of the 45 hybridoma lines matched very closely that observed by ELISPOT on Borrelia-infected lymph nodes, indicating that the hybridoma lines were recapitulating the responses in vivo (Figure 6B). Furthermore, the largest fraction of the 45 Borrelia-specific hybridomas (11/45) recognized DbpA (data not shown). This is consistent with the initial specificity screen by ELISPOT (Figure 5A). Given that the recombinant antigen pool was likely to underestimate the frequencies of antigen-specific B cells/hybridomas, we conclude that a sizable fraction of the massive B cell response induced during Lyme borreliosis is specific against B. burgdorferi.
Non-specific mitogenic stimulation of B cells with Borrelia lipoproteins in vitro has been reported previously [9]–[16]. OspA, a surface lipoprotein that is strongly expressed by Borrelia in culture, but down-regulated upon infection of a mammalian host, was shown to be responsible for at least some of the mitogenic activity [11], [14]. While host-adapted spirochetes are not expected to express significant amounts of OspA, other proteins or lipids may provide mitogenic signals to B cells in vivo. Therefore, we determined the role of the adaptor protein MyD88, important in TLR and IL-1-mediated innate signaling, in regulation of initial B cell activation and/or the lymph node enlargement. A previous study found impaired pathogen-clearance and alterations in the antibody-isotype profile of serum antibodies in mice lacking MyD88 [42]. MyD88−/− mice and congenic control mice were infected with host-adapted spirochetes for ten days. The analysis revealed no role for MyD88 in the quality or magnitude of the lymphadenopathy. Regional lymph nodes from MyD88−/− mice had similar cell numbers on day 10 of infection (Figure 7A), with similar predominance of CD19+ B cells compared to control mice (Figure 7B). Furthermore, there was no difference in the number of Borrelia-specific IgM or total Ig secreting cells in the lymph nodes (Figures 7C, 7D). Thus, MyD88-dependent innate signaling is not driving the induction of lymphadenopathy, nor the massive activation of B cell responses associated with Lyme borreliosis. Together with the strong antigen-specific B cell responses measured by hybridoma generation, the results suggest that Borrelia-infection induces a specific, albeit largely extrafollicular B cell response as a result of the accumulation of live B. burgdorferi in lymph nodes.
This study provides new insights into the pathogenesis of lymphadenopathy during the early stages of human Lyme borreliosis. The results demonstrate for the first time the extracellular accumulation of B. burgdorferi in the cortical regions of lymph nodes and implicate the direct association of migrating B. burgdorferi spirochetes with a marked and specific but unusual B cell response in the lymph nodes, but not the spleens, of mice infected with tick-borne or host-adapted spirochetes. The strong accumulation of proliferating B cells in the cortical areas of the lymph nodes, in the absence of a simultaneous accumulation of CD4 T cells (Figures 3, 4), the lack of strongly demarcated lymph node follicles and germinal centers in the lymph nodes (Figures 4B, 4C), and the strongly IgM and IgG3/2b-driven specific antibody response (Figures 5, 6) indicate that this pathogen drives the borrelia-specific B cell response towards T cell-independence. From these results we hypothesize that these effects of B. burgdorferi on the Borrelia-specific B cell response constitute a novel immune-evasion strategy.
The active migration of B. burgdorferi into sites of immune induction appears counter-intuitive for an organism that aims to establish persistence. Their presence in the lymph nodes and the strong responses their presence evokes thus indicate the intricate balance this pathogen achieves between immune induction and immune evasion. The nature of the observed B cell response is clearly distinct from that observed following acute infections with other non-persistent pathogens, or following immunizations with various protein antigens that induce mainly T cell-dependent extrafollicular and germinal center responses [43], [44]. In particular we note a lack of clearly demarcated follicles, with it an apparent lack of germinal centers and the accumulation of proliferating B cells in the follicles. The presence of T-independent B cell responses to B. burgdorferi had previously been indicated by measurements of strong antibody-responses in the serum of T cell-deficient mice [29]. While we cannot fully exclude the possibility that it is the nature of the expressed Borrelia-antigens that drive the B cell response towards its extrafolicular nature (Figure 4C) and apparent T-independence, we believe that this cannot fully explain our observations. A major component of the B cell response induced to infection with host-adapted spirochetes was directed against decorin-binding protein (DbpA) (Figures 5, 6). When administered in adjuvant, a strong germinal center-response was observed in the draining lymph nodes and DbpA-specific antibody responses were strongly induced against this protein, suggesting that this major B. burgdorferi immunogen is capable of inducing T-dependent responses in the right context (unpubl. observations). Furthermore, immunization with DbpA does induce protective antibody responses [36].
Also, the strongly B cell-driven lymphadenopathy seen following B. burgdorferi infection was not observed following immunization of mice with culture-grown heat-killed and sonicated spirochetes (Figures 3C, 3E), although such immunization increased both lymph node size (Figure 3D) and induced moderate frequencies of B. burgdorferi-specific antibody-secreting cells (Figure 3F). Thus, either live infection and/or the presence of live extracellular bacteria and bacterial proteins in the cortex areas of the lymph nodes appear to trigger this unique B cell response to spirochetes, or alternatively, the response is triggered by an antigen(s) not present on the culture-grown bacteria used for immunization. Since neither the lymphadenopathy nor the B cell response were significantly different following infection of MyD88−/− mice compared to controls (Figure 7), TLR-mediated inflammatory responses can be excluded as potential triggers of this response, in contrast to apparently similar TLR-4-mediated alterations following S. typhimurium infection [17].
From this, it is tempting to speculate that it is the expression of specific immune-subversion antigens by B. burgdorferi in the mammalian host that induce overshooting and potentially aberrant T cell independent B cell responses that are neither of sufficient high-affinity nor induce memory responses able to combat primary and repeat infections. The analysis of candidate antigens must await the development of techniques that allow us to comprehensively compare protein expression by culture-grown and tissue-adapted spirochetes within the context of specific tissue sites, such as lymph nodes.
While the induced B cell response to B. burgdorferi is unable to clear the infection, it does provide immune protection from overt disease. This is indicated by studies in B cell- or CD40L-deficient mice that showed increased signs of tissue-inflammation and disease progression compared to controls [29], [40], [45]. Furthermore, passive transfer of immune serum from infected mice confers immune protection from infection when injected prior to pathogen challenge [26], [30]. Thus, understanding the mechanisms that induce and regulate the borrelia-specific B cell response is of importance. Assessing the specificity of the B cell response to B. burgdorferi is challenging, however, due to differences in the antigenic structure of B. burgdorferi cultured in artificial media versus those grown in the mammalian host [39], [46]–[50]. Thus, lysates or extracts from culture-grown spirochetes do not reflect antigens expressed in the mammalian host. Furthermore, B. burgdorferi differentially expresses antigens during the various stages of its life cycle in the flat tick, the feeding tick and the host [39] . We therefore utilized an infection protocol that mimics tick-borne infection and avoids induction of immune responses to Borrelia antigens not expressed in vivo, by infecting mice with mammalian host-adapted spirochetes via tissue transplant.
For detection of B. burgdorferi-specific antibodies by ELISA and ELISPOT, we used a cocktail of recombinant antigens, including OspC, DbpA, Arp and BmpA, each of which are expressed during infection of the mammalian host [36], [47], [49], [51]–[53]. Furthermore, each of these antigens have been shown to induce protective or disease-resolving immune responses in mice [41], [47], [54]–[57]. We did not include the VlsE protein in our studies, a surface-protein thought to subvert the immune response to B. burgdorferi through extensive genetic variation within the host. However, the N40 strain of B. burgdorferi, which we have used here, does not seem to express this protein, based on transcriptional analysis of the IR6 region of vlsE. Moreover, we found no evidence of seroconversion to the C6 antigen of vlsE from strain B31 (S. W. Barthold, unpublished). Recent sequence analysis of the N40 genome has confirmed that N40 vlsE and BBK01 are on different plasmids and that the vlsE locus is indeed significantly different compared to B31, the commonly used VlsE-expressing Borrelia-strain.
Using only a handful of such in vivo-expressed and immunodominant antigens, we demonstrated the induction of a strong B. burgdorferi-specific antibody response in the lymph nodes of infected mice (Figures 5, 6) in a manner that is independent of MyD88 (Figure 7). We furthermore showed that nearly a quarter of hybridomas generated from lymph nodes of acutely B. burgdorferi-infected mice are specific for this pathogen (Figure 6). Together with earlier studies that demonstrated the protective and disease-resolving capacity of immune sera from long-term infected mice [26], [30], [58], we can conclude that a strong and borrelia-specific B cell response is induced in these lymph nodes.
B. burgdorferi causes spirochetemia, but its primary means of dissemination is via migration through host connective tissues and extracellular matrix [59] . This is consistent with our finding of progressive involvement of the ipsilateral, but not the matching contralateral lymph nodes of the host (Table 1). Furthermore, the spleens of infected mice did not differ in size or in frequencies of B. burgdorferi-specific antibody-secreting cells compared to spleens from uninfected mice, and were only sporadically culture-positive for spirochetes (Table 1). In apparent contrast, a previous study in mice reported the involvement of marginal zone B cells in the response to B. burgdorferi infection [60]. This difference to our study might well be due to the difference in the route of infection, i.e. intra-cutaneously with culture-grown bacteria at the back of mouse versus infection with host-adapted spirochetes by tissue-transplantation in the tarsus region. It has been well documented that the course of infection and organ involvement varies with the site of inoculation in mice [61], [62]. Furthermore, it is notable that tick-borne infections also failed to induce a significant B cell response in the spleen (data not shown).
In conclusion, by accumulating in the extracellular cortical spaces of the lymph node, B. burgdorferi seems to both induce and subvert an important arm of the adaptive immune response. Rather than fully suppressing the activity of B cells, B. burgdorferi appears to shift the major B cell response towards the production of antibodies generated in extrafollicular foci. It thereby seems to support the production of antibodies that provide immune protection from disease, while subverting the induction of more strongly protective, possibly T-dependent B cell responses that could confer bacterial clearance.
|
10.1371/journal.pcbi.1003611 | Histone Modifications Are Associated with Transcript Isoform Diversity in Normal and Cancer Cells | Mechanisms that generate transcript diversity are of fundamental importance in eukaryotes. Although a large fraction of human protein-coding genes and lincRNAs produce more than one mRNA isoform each, the regulation of this phenomenon is still incompletely understood. Much progress has been made in deciphering the role of sequence-specific features as well as DNA-and RNA-binding proteins in alternative splicing. Recently, however, several experimental studies of individual genes have revealed a direct involvement of epigenetic factors in alternative splicing and transcription initiation. While histone modifications are generally correlated with overall gene expression levels, it remains unclear how histone modification enrichment affects relative isoform abundance. Therefore, we sought to investigate the associations between histone modifications and transcript diversity levels measured by the rates of transcription start-site switching and alternative splicing on a genome-wide scale across protein-coding genes and lincRNAs. We found that the relationship between enrichment levels of epigenetic marks and transcription start-site switching is similar for protein-coding genes and lincRNAs. Furthermore, we found associations between splicing rates and enrichment levels of H2az, H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K9me3, H3K27ac, H3K27me3, H3K36me3, H3K79me2, and H4K20me, marks traditionally associated with enhancers, transcription initiation, transcriptional repression, and others. These patterns were observed in both normal and cancer cell lines. Additionally, we developed a novel computational method that identified 840 epigenetically regulated candidate genes and predicted transcription start-site switching and alternative exon splicing with up to 92% accuracy based on epigenetic patterning alone. Our results suggest that the epigenetic regulation of transcript isoform diversity may be a relatively common genome-wide phenomenon representing an avenue of deregulation in tumor development.
| Traditionally, the regulation of gene expression was thought to be largely based on DNA and RNA sequence motifs. However, this dogma has recently been challenged as other factors, such as epigenetic patterning of the genome, have become better understood. Sparse but convincing experimental evidence suggests that the epigenetic background, in the form of histone modifications, acts as an additional layer of regulation determining how transcripts are processed. Here we developed a computational approach to investigate the genome-wide prevalence and the level of association between the enrichment of epigenetic marks and transcript diversity generated via alternative transcription start sites and splicing. We found that the role of epigenetic patterning in alternative transcription start-site switching is likely to be the same for all genes whereas the role of epigenetic patterns in splicing is likely gene-specific. Furthermore, we show that epigenetic data alone can be used to predict the inclusion pattern of an exon. These findings have significant implications for a better understanding of the regulation of transcript diversity in humans as well as the modifications arising during tumor development.
| Molecular processes such as alternative splicing and transcription start-site switching are primary drivers of transcript diversity. About 95% of the ∼23,000 human genes are estimated to produce more than one mRNA isoform [1]. Beyond the genes with protein-coding potential, recent discoveries suggest that the approximately 8,000 large intergenic noncoding RNAs (lincRNAs) found in the human genome generate on average 2.3 isoforms per lincRNA locus [2].
The analysis of transcript diversity regulation has traditionally focused on splicing factors and RNA sequence features such as splicing enhancers and silencers [3], [4]. In recent years, however, experimental studies have expanded to include other regulatory factors such as histone modifications, suggesting that epigenetic features may have the ability not only to determine when and in which tissues certain genes are expressed, but also to influence how these transcripts are processed. Genome-wide analyses indicate that nucleosomes and histone modifications are not randomly distributed, but often coincide with exon boundaries [5]–[7]. This observation, combined with recent evidence that most events of alternative splicing in human cells occur co-transcriptionally [8], [9], strongly suggest a regulatory potential of histone marks [2], [10].
While the connection of epigenetic regulation with overall gene expression has largely been elucidated [11]–[14], it is much less clear whether and how epigenetic marks determine relative isoform abundance. Qualitative and quantitative models have been built to predict expression on the level of genes using histone modification enrichment information alone [15]. Interestingly, a quantitative prediction model based on histone modification enrichment outperforms models based on transcription factor binding [15]. However, a systematic evaluation of the association of epigenetic marks with transcription start-site switching and splicing frequency is still lacking in the literature. Work by Ernst et al. [16], [17], who classified chromatin states to functionally annotate the genome, identified a combination of histone modifications, which were associated with transcription start site and spliced exons. However, since in this work, the histone mark ChIP-seq tag counts were processed into binary presence and absence calls and since isoform abundance was not estimated from the expression data, the critical question remains whether different levels of epigenetic enrichment are associated with the rates of transcription start-site switching and splicing.
In addition to elucidating the epigenetic regulation of transcript diversity, further open questions remain. These questions pertain for instance to the genome-wide prevalence of epigenetic regulation of transcript diversity generated via alternative splicing or transcription start-site switching. Furthermore, it is unclear to what extent the involvement of epigenetic marks in the regulation of transcript diversity is gene-specific, ie. whether individual genes respond to different histone marks or whether there is a “universal” set of marks for alternative splicing. Several studies aimed at deciphering the association between histone modifications and alternative splicing on a genome-wide scale [18]–[22] but relied solely on gene annotation for the assignment of alterative splicing events rather than on a comprehensive transcription analysis [22], or with no more than three cell lines lacked the breadth of conditions analyzed [18], [19], [21]. Finally, the association of epigenetic patterning with transcript diversity in cancer cells has not been analyzed methodically in a genome-wide manner; however, understanding the prevalence of this phenomenon is of particular importance in cancer where cells are known to undergo vast epigenetic aberrations [23]. Indeed, epigenetically divergent regions in cancer cell lines are enriched for cancer-associated genes (Module S1 in Text S1).
Here, we sought to perform a detailed study investigating the association between histone modification enrichments and the processes that influence isoform abundance – transcription start-site switching and splicing – on a genome-wide level (Fig. 1A and Table 1). We further developed a novel approach that identified a set of 840 genes for which transcription start-site switching and splicing was strongly associated with at least one epigenetic mark. We also showed that histone modification enrichment alone can predict exon splicing and transcription start-site switching with up to 92% accuracy in an independent sample set. Our work strongly suggests a broad-scale involvement of epigenetic factors in transcription start-site switching and alternative splicing.
We examined RNA-seq data from nine human cell lines (Gm12878, Hsmm, Huvec, Hepg2, Helas3, K562, H1hesc, Nhek, Nhlf) (http://genome.ucsc.edu/ENCODE/), of which six were normal (Gm12878, Hsmm, Huvec, H1hesc, Nhek, Nhlf) and three were cancer cell lines (Hepg2, Helas3, K562). For all nine cell lines, we obtained information of the genome-wide patterns of the following twelve histone marks: H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K9me1, H3K9me3, H4K20me1, H3K27ac, H3K27me3, H3K36me3, H3K79me2, H2az.
Our analysis of the association between histone enrichment and transcript diversity utilized two different approaches: (i) a genome-wide approach, and (ii) an exon-specific approach. The genome-wide method analyzes each cell line individually and investigates all exons with a given characteristic (i.e. spliced, not spliced, transcription start site exon, etc.) at once, irrespective of the gene of origin. The exon-specific approach, in contrast, analyzes one exon at a time across multiple cell lines. The latter approach is able to identify candidate exons or genes with potential epigenetic regulation of transcription diversity and is analogous to an experimental setup in which each cell line represents an experimental condition (i.e. varying levels of histone modification enrichment) resulting in a particular exon inclusion or transcription start site outcome. The genome-wide approach requires a set of assumptions (see Discussion section); however, due to the large sample size of exons, it may uncover associations that would otherwise not be significant at a single gene level. With sufficiently many samples and sequencing depth, the patterns of associations uncovered by both approaches converge.
To assess the level of transcript diversity in the human genome, we analyzed RNA-seq data from nine human cell lines and quantified the abundance of specific mRNA isoforms for each protein-coding gene and lincRNA. We mapped and assembled the transcriptome of each cell line using the TopHat2 and Cufflinks2 softwares [24], [25], respectively, using merged UCSC reference annotation with lincRNA annotation from Cabili and colleagues [2] as a set of assembly models (see Methods). In order to minimize confounding issues, for instance with the misalignment of RNA-seq reads, we excluded paralogs that were more than 95% identical on the DNA sequence level. Exons were grouped into four categories: transcription start site, internal, transcription end site, or overlapping exons. Only internal and transcription start site exons were used for further analysis. The level of expression of an internal and transcription start site exon was quantified by calculating the splicing exon inclusion rate (SEIR, ranging from 0 to 1) and transcription start site inclusion rate (TSSIR, ranging from 0 to 1) respectively, both of which reflect the proportion of transcripts containing a given exon at a given gene locus (Fig. 1B). An SEIR of 0 implies that a given exon is always spliced in all expressed isoforms of a gene, whereas an SEIR of 1 implies that a given exon is always retained. A TSSIR of less than 1 signifies that a given exon occasionally represents the first exon of an expressed isoform, whereas a TSSIR of 1 indicates that a given exon serves as the transcription start site for all expressed isoforms. The SEIR and TSSIR measures therefore identify exons contributing to transcript diversity of a given gene.
We hypothesized that, if histone modification enrichment patterns play a significant role in transcript diversity, then the levels of transcription start-site switching and splicing should correlate with the enrichment levels of certain histone modifications within each cell line analyzed. We therefore investigated the transcriptome-wide association between histone mark enrichment and TSSIR and SEIR within each cell line. To address the possibility that transcript diversity in cancer cell lines is regulated differently as compared to that in normal cell lines, we quantified the level of association in the normal cell lines first and then assessed the degree of similarity in this pattern between normal and cancer cell lines. To this end, we determined the expression profiles as well as histone modification enrichment for all annotated exons of protein-coding genes and lincRNAs in the normal cell lines (Methods). Out of the twelve histone marks examined, seven (H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K27ac, H3K79me2, and H2az) showed a strong positive association with transcription start-site switching for both protein-coding genes and lincRNAs (Fig. 1C and 1D). Although the involvement of H3K4me2 and H3K4me3, H3K9ac, and H3K27ac in transcription initiation was expected given the findings of previous studies [16], [17], the presence of H3K79me2 and H2az was not anticipated. These results suggest that the transcription initiation of both protein-coding genes and lincRNAs is probably regulated via similar molecular mechanisms.
The transcription profiles of the nine cell lines revealed that many protein-coding genes as well as lincRNAs undergo alternative splicing. Given the fact that transcription start-site switching occurs in a similar epigenetic background for protein coding genes as well as lincRNAs, we then sought to investigate whether splicing in protein-coding genes and lincRNAs is also associated with a similar set of histone marks. We found that splicing in protein-coding genes was most strongly positively correlated with the enrichment of H3K36me3 and negatively correlated with H3K4me2 and H3K4me3 (Fig. 2A). H3K36me3 has been previously found to mark actively transcribed regions and to regulate the splicing of FGFR2 [26], thus confirming our results. However, splicing of lincRNAs did not reveal any association with histone mark enrichment (Fig. 2B), suggesting that splicing of non-coding RNAs is either independent of the epigenetic background, involves sequence-specific regulation, and/or occurs post-transcriptionally.
We then aimed to investigate whether this pattern was consistent when taking into account exon number per gene, gene expression patterns, and genomic features such as simple repeats, microsatellites, and conserved elements. Controlling for these factors, the correlations between TSSIR and H3K4me2 as well as H3K9ac were very robust, varying for instance in the Gm12878 cell line between 0.35<ρ<0.37 for H3K4me2 (uncontrolled correlation ρ = 0.37) and between 0.35<ρ<0.38 for H3K9ac (uncontrolled correlation ρ = 0.37). Similarly, controlling for H3K9ac enrichment reduced the correlation between TSSIR and H3K4me2 by only 0.5%, and controlling for H3K4me2 enrichment reduced the correlation between TSSIR and H3K9ac by only 3.18%. These observations suggest that, while the interplay between transcript diversity and epigenetics probably involves many other factors, which might occlude the signal, the association between the SEIR and specific histone marks is genuine.
Recently, a study examining the alternative splicing of CD45 showed that molecular interactions as far as 1 kb downstream of exon 5 affected its inclusion rate [27]. To investigate how epigenetic marks at a distance from exons influences transcript diversity on a genome-wide scale, we analyzed histone enrichment profiles at distances of 1 kb, 2 kb, and 5 kb immediately upstream and downstream of the exon locus (Methods). We identified pronounced differences in spatial patterns of correlation strength between the previously identified histone marks H3K4me1 and H3K79me2 and the TSSIR of protein-coding genes in normal cells (Fig. 1C). For example, the correlation between TSSIR and H3K4me1 at the exon locus was very weak (z0 kb = 0.09) but rose to much higher levels as close as 1 kb upstream and downstream of the spliced exon (z−1 kb = 0.28, z1 kb = 0.28); this level of correlation was also observed for distances of 2 kb and 5 kb upstream and downstream of the exon (z−5 kb = 0.22, z−2 kb = 0.29, z2 kb = 0.33, z5 kb = 0.30). Interestingly, a very different spatial pattern was observed for the histone mark H3K79me2, for which the correlation between TSSIR and histone enrichment upstream and at the exon locus was weak (z−5 kb = 0.03, z−2 kb = 0.07, z−1 kb = 0.08, z0 kb = 0.15), but became much stronger at distances of 1–5 kb downstream of the exon (z1 kb = 0.26, z2 kb = 0.27, z5 kb = 0.26). The spatial pattern of correlation between H3K4me1 enrichment and TSSIR for lincRNAs was less pronounced (Fig. 1D), showing lower levels of correlation at the exon locus compared to up- and downstream regions (z−5 kb = 0.17, z−2 kb = 0.18, z−1 kb = 0.15, z0 kb = 0.13, z1 kb = 0.19, z2 kb = 0.20, z5 kb = 0.19).
The only spatial pattern evident for an association between histone enrichment and SEIR was observed for H3K36me3 (Fig. 2A). While the correlation outside the exon boundaries ranged from 0.30<z<0.36, the correlation at the exon locus itself was slightly diminished to z0 kb = 0.26. It remains unclear which factors drive the spatial distribution of H3K36me3; for example, Luco et al. showed that H3K36me3 interacts with the FGFR2 pre-mRNA via the MRG15/PTB chromatin-adaptor complex, which regulates the inclusion rates of alternatively spliced IIIb and IIIc exons [28]. Work by others has further showed that additional proteins can act as “chromatin-adaptors” [29]–[32]. The question remains to what extent different chromatin adaptor complexes regulate splicing and which nucleosomes they interact with. A possible explanation of why the correlation of H3K36me3 with SEIR is diminished at the exon locus may lie in the position, relative to the exon, where different chromatin adaptors assemble and interact with H3K36me3 to regulate splicing. Further complicating the situation is a recent report demonstrating opposite causality, where alternative splicing was shown to modulate the levels of H3K36me3 enrichment [33], [34]. We observed no obvious spatial patterns between histone enrichment and splicing for lincRNAs (Fig. 2B).
Overall, our observations suggest that histone mark enrichment is associated with transcription start site exon inclusion and splicing and has a strong spatial signature. In addition to these analyses, we performed several control studies to establish that our results are genuine. First, our findings were robust even after controlling for gene expression, exon number, and genomic features such as simple repeats, microsatellites, and evolutionary conservation. Although the overall correlation between both TSSIR and SEIR and various histone modifications was moderate transcriptome-wide, the rapid change of correlation over short distances from exons and consistent patterns across multiple cell lines suggest an authentic relationship.
Since cells accumulate many genetic and epigenetic aberrations during tumorigenesis [23], [35]–[37], normal and cancer cells may differ substantially in their epigenetic regulation of transcript diversity. To investigate this possibility, we studied whether the association between TSSIR, SEIR and histone modifications in cancer cell lines followed similar patterns as those observed in the normal cells. We thus repeated the analyses described above using the cancer cell line data and tested for significant differences between the results using normal and cancer cell data for both protein-coding genes as well as lincRNAs. Remarkably, protein-coding genes in cancer cell lines displayed very similar patterns of association between the TSSIR and histone modifications as normal cell lines; the histone marks H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K27ac, H3K79me2, and H2az, which we previously found to be highly correlated in normal cell lines, were also highly correlated with TSSIR in cancer cells (Fig. 2C and 2D). Their correlation profiles across upstream and downstream exon regions also did not significantly differ from those of normal cell lines (T-test, 0.13>p>0.89 across all −5 kb, −2 kb, −1 kb, 0 kb, 1 kb, 2 kb, and 5 kb regions). Similarly, the other comparisons between normal and cancer cells, for both protein-coding genes and lincRNAs, did not show significant differences either (see Fig. 2, Table S1–S4 in Text S1, and Figure S5 in Text S1). These findings imply that the same histone modifications are associated with transcript diversity in both normal and cancer cells and that perturbation of the epigenetic environment via experimental manipulation in normal cells would potentially be informative of cancer cells.
So far, our transcriptome-wide and within-cell line approach identified an association between TSSIR, SEIR and histone enrichment across all exons but was unable to identify individual candidate genes with epigenetically regulated transcript diversity. We thus aimed to complement our investigation with a method that analyzes each exon individually across multiple cell lines. This approach is able to determine candidate genes with potential epigenetic regulation of transcript diversity and is analogous to an experimental setup where each cell line represents an experimental condition (i.e. varying levels of histone modification enrichment) resulting in a particular exon inclusion outcome. For example, the gene HPS4 (Hermansky-Pudlak syndrome gene 4) is expressed in all nine cell lines; its 3rd exon is always excluded (SEIR = 0) in all HPS4 isoforms in H1hesc, Helas3, Hsmm, Huvec, and Nhlf cells, but is only occasionally included (0.03<SEIR<0.15) in Gm12878, Hepg2, K562, and Nhek cells (Fig. 3). Interestingly, the cell lines that always exclude this exon do not show a significant enrichment for H3K4me2 within exon boundaries (Fig. 3), whereas the remaining cell lines do and the difference between these two groups is significant (T-test, FDR-corrected p<0.003, Methods).
We thus analyzed all exons across all cell lines in a similar fashion, first only taking into account histone enrichment at the exon locus. Given the TSSIR and SEIR values across cell lines, each exon may be constitutively excluded (TSSIR = 0 and SEIR = 0), occasionally excluded (TSSIR>0 and SEIR<1), or retained (TSSIR = 1 and SEIR = 1). We then directly compared the histone modification levels for the inclusion pattern of a given exon across all available cell lines. The three possible two-way comparisons are: i) cell lines in which a given exon is always excluded versus retained (TSSIR = 0 vs. TSSIR = 1 or SEIR = 0 vs. SEIR = 1), ii) cell lines in which a given exon is retained versus occasionally excluded (TSSIR = 0 vs. 0<TSSIR<1 or SEIR = 0 vs. 0<SEIR<1), and iii) cell lines in which a given exon is occasionally excluded versus retained (0<TSSIR<1 vs. TSSIR = 1 or 0<SEIR<1 vs. SEIR = 1). Unfortunately, since the number of cell lines with available histone modification was limited, the power of this test was low. Nonetheless, given our stringent criteria (Methods), we identified 840 genes for which transcript diversity was significantly associated with histone modification enrichment at the exon locus (Supplementary Dataset S1). Specifically, 399 and 473 genes displayed a significant association between splicing and transcription start-site switching, respectively. Note that a single gene can be significant for the association between epigenetic patterning and both splicing and transcription start-site switching. To understand whether obtaining 840 candidate genes was a result of chance, we performed 1000 permutations by randomly reassigning exon labels for TSSIR and SEIR while keeping the epigenetic background of a gene constant. Observing 840 candidate genes in total was significantly higher (p<0.001) as compared to what was expected by chance (Fig. 4). These 840 genes were enriched for several GO terms (Table S6 in Text S1) including the regulation of the response to stimulus and development process. Thirty three of these genes were cancer-associated genes (Supplementary Dataset S1) (http://www.sanger.ac.uk/genetics/CGP/Census/).
We then aimed to predict exon inclusion patterns in an independent sample set. Specifically, given the histone enrichment levels and the inclusion pattern in the nine previously studied cell lines, we sought to determine, in independent cell lines, whether a given exon was always retained (SEIR = 1), always excluded (SEIR = 0), or occasionally excluded (0<SEIR<1) with regard to splicing or transcription start-site switching (TSSIR = 1, TSSIR = 0, or 0<TSSIR<1, respectively). These predictions were performed in the Hmec and Monocytes CD14 cell lines, for which more complete epigenetic information became available (http://genome.ucsc.edu/ENCODE/downloads.html). We limited our predictions to the 840 candidate genes identified above, since the cell lines previously analyzed provided evidence for an involvement of epigenetic marks in transcript diversity for only 840 candidate genes; attempting to predict exon inclusion based on epigenetic information for genes that are not epigenetically regulated would thus not be appropriate.
To illustrate our approach, consider exon 5 of the ETV1 gene in the Hmec cell line; for this exon, we generated a matrix containing enrichment values for all histone modifications, which were significantly associated with SEIR (in this case H3K9ac, H3K4me3, H3K4me2, and H3K27ac) for the original cell line set (Gm12878, Hsmm, Huvec, Hepg2, Helas3, K562, H1hesc, Nhek, and Nhlf), and identified the SEIR of this exon in each cell line. All ETV1 isoforms in Gm12878 and Hepg2 cell lines lacked exon 5 (SEIR = 0) whereas some isoforms expressed in H1hesc, Hsmm, Huvec, K562, Nhek, and Nhlf cell lines contained exon 5 (SEIR range 0.43–0.69) (Fig. 5A). The difference in histone enrichment between these groups was striking: the Gm12878 and Hepg2 cell lines completely lacked enrichment in H3K9ac, H3K4me3, H3K4me2, and H3K27ac while the remaining cell lines were strongly enriched in those marks (Fig. 5A). We then calculated the pairwise Euclidean distance between all cell lines and the first validation line, Hmec, and determined the three nearest-neighbor cell lines signified by the smallest Euclidean distance (Methods). Since Hmec was enriched for all four histone marks in exon 5, its epigenetic profile was closest to that of the Nhlf, K562, and Huvec cell lines. We therefore predicted that in Hmec, exon 5 of ETV1 was occasionally excluded from some fraction of isoforms (0<SEIR<1), which was validated by the finding that in this cell line, SEIR = 0.74. When extending this approach to all candidate genes, we predicted the correct exon inclusion category with an accuracy of 91.82% and 84.65% for Hmec and Monocytes CD14 cell lines, respectively (Fig. 5B). To establish whether such high prediction accuracy can be established across all cell lines, we performed leave-one-out cross-validation following the approach described above. The accuracies for individual cell lines ranged from 72.1% in the Helas3 cell line to 91.8% in the Nhek cell line, with an average accuracy of 87.2% (Fig. 6). We also calculated the overall accuracy separately for splicing and for transcription start-site switching, which was 90.16% and 85.81%, respectively. Although the 0<EIR<1 vs. EIR = 1 comparison is the most frequent (76%), the accuracy for all comparisons consistently were high, at 90.00%, 95.00%, and 87.28% for EIR = 0 vs. 0<EIR<1, EIR = 0 vs. EIR = 1, and 0<EIR<1 vs. EIR = 1, respectively. Details regarding the fraction of genes that could be assigned into comparative groups and the number of significant genes for each validation step are displayed in Table S7 in Text S1. These findings suggest that the histone modification enrichment levels alone can be used to predict the inclusion pattern of an exon.
In this study, we analyzed the association between transcription start-site switching, spliced exon inclusion rates and histone modification patterns across multiple normal and cancer cell lines for protein-coding genes and lincRNAs. Unlike previous studies [8], [16], [17], which established the relationship between epigenetic patterning and gene expression levels, we addressed the association of the epigenetic background of a gene with its transcript isoform diversity. The main difference between ours and previous investigations therefore is that our study investigates relative isoform diversity of expressed genes, and not actual expression levels.
We used two approaches to address this issue. The first approach correlated transcriptome-wide (ie “within cell line”) transcription start site inclusion rates and spliced exon inclusion rates with histone enrichment levels. The second approach investigated gene-specific associations between transcription start site inclusion rates, spliced exon inclusion rates and histone enrichment levels. The shortcomings and assumptions made by each method are discussed below. Overall, our study led to four main findings. (i) The role of epigenetic patterning in transcription start-site switching is likely to be common across the genome for both protein-coding genes as well as lincRNAs. (ii) The role of epigenetic patterns in splicing is likely gene-specific, with the exception of H3K36me3 (discussed below). (iii) Our gene-specific approach led to the identification of 840 candidate genes whose exon inclusion rates for transcription start-site switching and splicing were strongly associated with patterns of histone modifications. (iv) Lastly, histone modification data alone can be used to predict the inclusion pattern of an exon.
Our first and second findings are based on the observation that both transcriptome-wide and gene-specific approaches identified a common set of histone marks that were associated with transcription start-site switching (H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K27ac, and H2az), whereas the results of these two methods differed for the case of splicing. Transcriptome-wide analysis for splicing showed a pronounced association of splicing inclusion rates with H3K36me3 whereas the gene-specific approach identified H2az, H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K9me3, H3K27ac, H3K27me3, H3K36me3, H3K79me2, and H4K20me1 as significantly associated marks. This discrepancy is likely a result of a bias by the transcriptome-wide approach to detect common genome-wide trends and the gene-specific approach to identify unique relationships for each exon.
Observing both common and gene-specific histone marks associated with splicing is in line with the proposed models of epigenetic regulation of splicing: the kinetic model and the chromatin-adaptor model [38]. According to the kinetic model, chromatin structure affects the elongation rate of RNA polymerase, which in turn influences the competition between weak and strong splice sites for the recruitment of splicing factors [38]. The chromatin-adaptor model, on the other hand, describes an interaction between specific histone marks and pre-mRNA molecules through a chromatin-adaptor complex, which aids in the recruitment of splicing factors to pre-mRNA splicing sites [26], [30], [39]. Since these two models are not mutually exclusive, one can imagine H3K36me3, known to be associated with transcription elongation [16], [17], to act as a common factor in splicing genome-wide, while other histone marks can act in a gene-specific manner. Interestingly, histone marks traditionally associated with transcription initiation and transcription repression, such as H3K4me3 and H3K9me3, respectively, were also found in our study to be associated with splicing gene-specifically. This observation is in line with experimental studies describing splicing chromatin-adaptor complex for H3K4me3 [40] and for H3K9me3 [41]. Further extending the realm of epigenetic regulation of transcript diversity is a recent work by Mercer and colleagues, which presented evidence for the role of 3-dimensional DNA conformation in splicing [42]. According to this study, exons sensitive to DNase I are spatially located close to transcription factories near promoter regions containing initiating Pol II as well as other general transcription and splicing factors.
Interestingly, a large fraction of alternatively spliced exons are DNase I sensitive [42]. This finding suggests that the epigenetic background of an exon cannot only interact with splicing factors via chromatin adaptor complexes, but potentially also induce 3-dimensional DNA conformation changes that enhance the likelihood of interactions with general transcription factors, and perhaps thus influence the splicing frequency. This 3-dimensional conformation is likely enhanced via particular sets of histone modifications. Interestingly, our second analysis, testing individual exon across all cell lines, revealed that alternatively spliced exons were frequently associated with different enrichment levels of histone marks well known to be associated with promoters and enhancers, such as H3K4me1, H3K4me2, H3K4me3, H3K27ac, and H3K9ac [16], [17] (Fig. 7). Accounting for such a 3-dimensional model could further strengthen the association found between histone modification enrichment and transcription start-site switching and splicing,
There are however shortcomings to both approaches. The transcriptome-wide method makes two assumptions that may be violated in cells. First, correlating transcription start site inclusion rates and spliced exon inclusion rates with histone mark enrichment assumes that (i) transcript diversity of all genes is associated with their epigenetic background, and additionally (ii) these rates are associated with the same histone modification. Likely, it is for these reasons that the correlations between exon inclusion rates and histone mark enrichment are rather moderate. As mentioned above, however, because of the rapid change of these correlations over short distances from exons and the consistent patterns across multiple cell lines, these associations suggest a genuine relationship. The shortcoming of the gene-specific approach lies in the low statistical power of eleven cell lines analyzed and the natural tendency to miss tissue specific exon behavior. This is particularly the case for lincRNAs, of which 30%, according to recent estimates, have tissue specific expression [2].
Since a large body of experimental data indicates that aberrant splicing of gene transcripts significantly contributes to many areas of cancer biology, including metabolism, apoptosis, cell cycle control, invasion and metastasis [43]–[45], it is imperative to further our understanding of the regulatory and/or potentially disruptive role of epigenetic patterning in alternative splicing and transcription start site selection in tumorigenesis. Significant effort has been devoted to the discovery of DNA aberrations that drive cancer progression [35]–[37], [46]; surprisingly, however, there is only a small number of recurrent genomic changes within and across cancer types, with few prominent exceptions [47]–[49]. While the identification of affected pathways rather than individual genes affected by DNA mutations might lead to more informative results, the possibility remains that aberrant phenotypes in cancer are largely driven by the epigenetic component of gene expression and transcript deregulation [23], [50], [51]. Our study identified several histone modifications (H3K4me1, H3K4me2, H3K4me3, H3K9ac, H4K27ac, H3K36me3, and H3K79me2) that are strongly associated with transcript diversity across multiple independent cell types as well as 840 candidate genes for which there is evidence of epigenetic co-regulation of transcript diversity. Our work represents a step towards identifying the functional consequences of histone modifications on transcript diversity and suggests a rational methodology for the analysis of modern, large-scale datasets, which can be applied to any sample sets.
Exons were grouped into four categories: transcription start site, internal, transcription end site, or overlapping exons. We then quantified the presence of each exon type. Only internal and transcription start site exons were used for further analysis. The relative presence of transcription start site exons (TSSIR – transcription start site inclusion rate) and spliced exons (SEIR – splicing exon inclusion rate) was calculated from the Cufflinks .gtf output file and reflects the fraction of all isoforms from a given gene that contain a given exon. The inclusion rates therefore have ranges of 0<TSSIR≤1 and 0≤SEIR≤1.
Using raw signal read counts of histone marks and reference samples (input DNA) for each cell line, we calculated the presence of histone mark enrichment using a Fisher's test statistic and considered enrichment significant [17] if p<0.0001. The level of enrichment was calculated as . RPKM is defined as , where r represents the number of reads mapped to a given exon, R is the total number of reads mapped, and L defines the length of a given exon. RPKM therefore denotes the number of reads per kilobase of exon per million reads mapped.
Prior to further analysis, we filtered our exon set to contain only internal exons and exons of genes that express more than one isoform in at least one normal or cancer cell line. In order to avoid potential problems with mapping RNAseq reads to closely related genes, we further excluded genes with paralogs more than 95% identical on the DNA level to generate the final curated exon dataset. We then calculated the Spearman rank correlation (which is more robust for asymmetrical distributions of TSSIR and SEIR and a large fraction of ties than Pearson's correlation) between TSSIR and SEIR and histone enrichment values, excluding all exons for which Fisher's test for histone enrichment was not significant (p>0.0001).
To assess the spatial patterns of correlation between histone modifications and SEIR as well as TSSIR, we calculated the extent of histone enrichment inside 1 kb, 2 kb, and 5 kb blocks immediately upstream or downstream of exons. The upstream 1 kb, 2 kb, and 5 kb regions extended from the upstream exon coordinate a given distance whereas the downstream 1 kb, 2 kb, and 5 kb regions extended from the downstream exon coordinate for a given distance. The Spearman rank correlation was then determined between each upstream or downstream block and the corresponding exon TSSIR or SEIR.
To allow for direct comparisons between correlation coefficients of different cell lines and histone modifications, we transformed the Spearman p using the Fisher transformation formula, .
To identify genes with epigenetically regulated transcript diversity, we analyzed each exon in the context of the nine cell lines (Gm12878, Hsmm, Huvec, Hepg2, Helas3, K562, H1hesc, Nhek, Nhlf). We categorized the exon inclusion rate into three groups: SEIR = 0, 0<SEIR<1, and SEIR = 1. We followed the same approach for TSSIR. Next, we tested whether any histone modification displayed a statistically significant difference in its enrichment in any possible two-group comparison, given an exon's SEIR values across the nine cell lines. For example, if the pattern of SEIR values for a given exon allowed us to separate the nine cell lines into two groups that showed either SEIR = 0 or SEIR = 1, we used T-test to determine whether the respective histone modification enrichment among the two groups of cell lines was statistically different. All p-values were corrected for false discovery rate (FDR) [54]. To discover cell-specific events, we allowed for comparisons where only one cell line versus many could be assigned to an SEIR or TSSIR group. Naturally, given the lower power of this test, most of these did not pass our 5% FDR cutoff. This approach identified 840 genes, for which at least one exon showed a statistically significant association between SEIR and at least one histone modification (ie. statistically significant difference in histone modification enrichment between two SEIR groups for a given exon).
We limited our predictions of exon exclusion or retention in the Hmec and Monocytes CD14 cell lines to the 840 candidate genes that showed significant association between the TSSIR or SEIR and histone modification enrichment in the original set of nine human cell lines (Gm12878, Hsmm, Huvec, Hepg2, Helas3, K562, H1hesc, Nhek, Nhlf). For a given exon, we constructed a Euclidean distance matrix with the formerly identified set of histone modifications for all cell lines, including Hmec and Monocytes CD14. Next, we determined the three closest neighbors of Hmec and Monocytes CD14 from among the original set of nine cell lines (Gm12878, Hsmm, Huvec, Hepg2, Helas3, K562, H1hesc, Nhek, and Nhlf). Because the exon inclusion rates for a given exon were known in the original set of nine cell lines, we separated these cell lines into three comparison groups: i) cell lines in which a given exon was always excluded versus retained (SEIR = 0 vs. SEIR = 1), ii) cell lines in which a given exon was retained versus occasionally excluded (SEIR = 0 vs. 0<SEIR<1), and iii) cell lines in which a given exon was occasionally excluded versus retained (0<SEIR<1 vs. SEIR = 1). The inclusion status - retained, occasionally excluded, or always excluded - of a given exon in the Hmec or Monocytes CD14 cell lines was then determined based on what comparison group the majority of the three closest neighbors belonged to. For example, if the majority of Hmec's three closest neighbors (based on the Euclidian distance matrix) belonged to the group SEIR = 1, then we would predict that particular exon in the Hmec cell line was always retained, ie. a given gene was expressing only those isoforms that included our exon of interest. We applied the same approach to transcription start site exons and their respective TSSIR values.
|
10.1371/journal.pgen.1005704 | Retromer Is Essential for Autophagy-Dependent Plant Infection by the Rice Blast Fungus | The retromer mediates protein trafficking through recycling cargo from endosomes to the trans-Golgi network in eukaryotes. However, the role of such trafficking events during pathogen-host interaction remains unclear. Here, we report that the cargo-recognition complex (MoVps35, MoVps26 and MoVps29) of the retromer is essential for appressorium-mediated host penetration by Magnaporthe oryzae, the causal pathogen of the blast disease in rice. Loss of retromer function blocked glycogen distribution and turnover of lipid bodies, delayed nuclear degeneration and reduced turgor during appressorial development. Cytological observation revealed dynamic MoVps35-GFP foci co-localized with autophagy-related protein RFP-MoAtg8 at the periphery of autolysosomes. Furthermore, RFP-MoAtg8 interacted with MoVps35-GFP in vivo, RFP-MoAtg8 was mislocalized to the vacuole and failed to recycle from the autolysosome in the absence of the retromer function, leading to impaired biogenesis of autophagosomes. We therefore conclude that retromer is essential for autophagy-dependent plant infection by the rice blast fungus.
| The rice blast fungus Magnaporthe oryzae utilizes key infection structures, called appressoria, elaborated at the tips of the conidial germ tubes to gain entry into the host tissue. Development of the appressorium is accompanied with autophagy in the conidium leading to programmed cell death. This work highlights the significance of the Vps35/retromer membrane-trafficking machinery in the regulation of autophagy during appressorium-mediated host penetration, and thus sheds light on a novel molecular mechanism underlying autophagy-based membrane trafficking events during pathogen-host interaction in rice blast disease. Our findings provide the first genetic evidence that the retromer controls the initiation of autophagy in filamentous fungi.
| Rice blast is one of the most serious and recurrent diseases destroying rice production worldwide [1,2]. The ascomycete fungus Magnaporthe oryzae infects rice tissues by forming a dome-shaped and melanized infection structure called appressorium [3,4]. Differentiation of appressorium is regulated by cell cycle progression that is accompanied by autophagy in the conidium leading to its programmed cell death [5,6]. In this process, most of the stored glycogen and lipids are quickly transported from the conidium into the developing appressorium in order to establish a high turgor pressure necessary for successful host penetration by the mature appressorium [7–9]. Subsequently, a narrow penetration peg emerges from the mature appressorium and enters the rice epidermis and differentiates into bulbous, branched invasive hyphae, which are bound by the host plasma membrane in the invaginated cell, allowing the fungus to proliferate within the living plant cells [4,10].
Macroautophagy is a highly conserved bulk degradation process required for stress response and nutrient signaling in eukaryotes. Autophagy requires the formation of double-membrane bound autophagosomes that engulf bulk cytoplasm (nonselective) or specific target cargos (selective autophagy). The autophagosomes fuse with endosomes or the vacuoles to form autophagolysosomes to deliver the sequestered material for recycling and/or degradation [11,12]. A set of evolutionarily conserved autophagy-related genes (ATG genes) was initially identified in yeast [13,14]. In total, 22 ATG genes were identified in the rice blast fungus, and MoATG8 expression has been used to investigate the spatial pattern of autophagy induction during infection-related development [5]. Punctate autophagosomes are found to be enriched in infection-related structures such as conidia, germ tubes and appressoria [15,16]. Deletion of MoATG8 led to significant reduction in conidiation and defects in glycogen autophagy during conidiogenesis [15,16]. Furthermore, the MoATG8-deficient appressoria are nonfunctional and noninfectious due to an inability to undergo autophagic cell death and nuclear degeneration in conidia [5]. Genome-wide characterization of autophagy genes [15] further supports the critical role of autophagy in conidial cell death and the function of the appressorium in M. oryzae. However, the mechanisms regulating autophagy in M. oryzae remain elusive.
The retromer complex is a conserved vital element of the endosomal protein sorting machinery [17]. It consists of two subcomplexes: a trimer of Vps35, Vps29 and Vps26 for cargo selection, and a dimer of Vps5 and Vps17 for tubule or vesicle formation [18]. The retromer complex is known to participate in intracellular retrograde transport of cargos from the endosome to the proper organelles [19,20]. Loss or malfunction of retromer is associated with various pathological states due to protein mistargeting [17]. Serving as the core of retromer, Vps35 directly interacts with cargo proteins for sorting [19,20]. Recently, it was found that retromer played a role in the degradation of autophagic cargo to maintain lysosome function in Drosophila [21]. To our knowledge, the role of Vps35 or retromer in regulating autophagy and plant infection in plant fungal pathogens has not been assessed thus far.
In this study, we report a crucial role for the retromer cargo recognition subcomplex in regulating the autophagic process during appressorial development and pathogenesis in the rice blast fungus. We demonstrate that loss of any of these retromer components, MoVps35, MoVps26, or MoVps29, led to similar defects in fungal conidiogenesis and pathogenesis, which phenocopied the defects of ATG mutants including reduced turgor pressure, delayed turnover of glycogen and lipid bodies, and failure in autophagic cell death during conidial germination, and compromised pathogenicity of the blast fungus. Furthermore, our data suggest that MoVps35 regulates the autophagic process through retrieving the cleaved form of MoAtg8 from the vacuole after autolysosome formation. Therefore, our findings uncover a new function of retromer and shed light on the regulation of autophagy biogenesis in one of the most important fungal pathogens of rice and cereal crops.
We identified a single Vps35 ortholog MoVps35 in the M. oryzae proteome using BLASTP analysis. MGG_05089 (hereafter MoVps35) showed 57% sequence identity to the yeast Vps35 (S1 Table). To determine its function, two MoVPS35 null mutants were generated through targeted gene replacement with the hygromycin resistance cassette in the Δku70 background (S1A and S1B Fig). Phenotypic analyses revealed that the mutant ΔMovps35 grew marginally slower (about 72.4% of the WT, P<0.01) than the wild type on various culture media (S1C Fig). This suggests that the loss of MoVPS35 likely reduces vegetative growth and/or overall fitness of M. oryzae. Given an important role for the cell wall in maintaining hyphal development and adaptation to the environment [22], we further investigated the growth under cell wall or membrane stress conditions. In such growth assays, the ΔMovps35 showed decreased resistance to Calcofluor White (CFW), congo red (CR) and sodium dodecyl sulfate (SDS) compared to the wild-type strain (S2 Fig), suggesting that MoVps35 is involved in maintaining the integrity of the cell wall. We also found that in comparison to the wild type, ΔMovps35 showed a 19-fold decrease in conidiation (S1D Fig and S1E Fig; P<0.01). The differentiation of conidiophores is critical for conidial development [23,24]. Although the sympodial arrangement of the resultant mutant conidia remained unchanged (S1D Fig), conidiophore differentiation was highly reduced in ΔMovps35 at 24 h post conidial induction (S1F Fi.). Thus, the dramatic reduction of conidiogenesis in the ΔMovps35 likely results from decreased conidiophore formation. Genetic complementation via introduction of MoVPS35 restored proper growth and conidiation in the ΔMovps35 strain (S1 Fig and S2 Fig). We conclude that MoVps35, and by inference the retromer function, is essential for proper vegetative growth and asexual development in the blast fungus.
We then assessed the pathogenicity of ΔMovps35 mutant on rice seedlings (Oryza sativa cv. CO39). When spray-inoculated on rice seedlings, the wild type as well as the complementation strain caused numerous typical blast lesions on leaves, whereas the ΔMovps35 mutant caused only a few small and isolated lesions (Fig 1A). ΔMovps35 formed only 1.2 ± 1 lesions per 5 cm of leaf (P<0.01), whereas 69.7 ± 15.3 lesions were evident in leaves inoculated with wild-type conidia (Fig 1A). Likewise, the barley infection assays (cv. Golden Promise) showed severe blast symptoms seven days after inoculation with wild-type conidial suspension or mycelia, whereas the ΔMovps35 mutant failed to cause blast disease in barley seedlings (Fig 1B and Fig 1C). To investigate if the pathogenicity defects in ΔMovps35 were due to a block in penetration or invasive growth, we inoculated mycelia from the wild type or ΔMovps35 through abraded barley leaves. This allows for invasive growth independent of appressorium function. We found that the ΔMovps35 mycelia were able to invade the wounded tissue and caused weak lesions on the wounded leaves compared to the WT (Fig 1D). These results suggested that the ΔMovps35 mutant were unable to infect rice and barley, owing mainly to their inability to penetrate the plant cuticle.
In yeast, plant and mammals, Vps26p and Vps29p form the cargo-selective subcomplex of the retromer via interaction with Vps35p [25,26]. Therefore, we assessed whether these proteins cooperate in common pathogenic pathways in M. oryzae. Using orthologous yeast sequences, we identified MGG_04830 and MGG_02524 as MoVps26 and MoVps29, respectively, in M. oryzae (S1 Table). We first investigated whether such a retromer subcomplex occurs in M. oryzae. In the yeast two-hybrid assay, MoVps35 was found to physically interact with MoVps26 and MoVps29 (Fig 1E). Furthermore, we constructed ΔMovps26 and ΔMovps29 deletion mutants by gene replacement. The ΔMovps26 and ΔMovps29 deletion mutants (S2 Table) were identified by PCR and confirmed by DNA gel blot analysis (S3 Fig). Like the ΔMovps35 mutant, ΔMovps26 and ΔMovps29 mutants were also impaired in conidiation and pathogenicity on the seedlings of the susceptible rice cultivar CO39 (S4 Fig, Fig 1F and Fig 1G). Finally, we expressed WT MoVPS26 and MoVPS29 gene in the corresponding null mutants. As expected, the complementation strains showed suppression of mutant defects in conidiation and pathogenicity (S4 Fig, Fig 1F and Fig 1G). The data suggest that MoVps35, MoVps26 and MoVps29 function together in the retromer pathway and play a key role in plant infection by the rice blast fungus.
To understand why the retromer subcomplex (MoVps35, MoVps26 and MoVps29) is required for pathogenicity in M. oryzae, we first chose the core retromer component MoVps35 for a detailed functional analysis. Based on the above observations, we reasoned that MoVps35 either controls proper appressorium formation or appressorium-mediated infection in M. oryzae. We first assayed for appressorium formation on artificial hydrophobic surfaces, wherein the ΔMovps35 conidia produced abundant melanized appressoria (Fig 2A). No obvious morphological defects in appressorium formation were evident in ΔMovps35 (Fig 2A–2C). The ΔMovps35 exhibited normal appressorium formation on onion epidermal cells as well (Fig 2B). However, such ΔMovps35 appressoria were defective in the penetration of onion epidermal cells and subsequent differentiation into invasive hyphae (Fig 2B). By 48 h, the wild-type strain penetrated and formed invasive hyphae in onion epidermal cells (Fig 2B). A vast majority of the appressoria (95%, P<0.01) failed to penetrate the onion epidermal cells even at 72 hpi. Aniline blue staining and further quantification of penetration efficiency was carried out on barley leaves inoculated with conidia of WT or ΔMovps35 (Fig 2C). The efficiency to form penetration pegs on barley leaves was about 62% in WT, while only 3.6% (P<0.01) appressoria showed host entry in the ΔMovps35 mutant at 48 h time point. Even upon extended incubation, ΔMovps35 appressoria were still unable to penetrate the host surface (Fig 2D and 2H). Similar defects were also evident in invasive growth in ΔMovps35 inoculated on barley leaves (Fig 2E). We conclude that MoVps35 is not required for appressorium formation but is essential for appressorium-mediated host penetration by the rice blast fungus.
Since establishment and maintenance of high internal turgor pressure is necessary for appressorium-mediated host penetration by M. oryzae [9], we examined the turgor pressure in ΔMovps35 using the incipient cytorrhysis assays [27]. These appressorial collapse assays revealed that the ΔMovps35 appressoria generate significantly lower turgor compared to wild type (Fig 3; P<0.01). In 2 M glycerol, 75% of ΔMovps35 appressoria collapsed at 24 h compared to 27% and 25% of the appressoria in the wild type and complementation strains, respectively (Fig 3B). Upon increasing the glycerol to 3 M and extending the incubation, appressoria of ΔMovps35 mutant remained severely collapsed compared to those from the wild-type or the complemented strain (Fig 3B). Further analysis revealed that the wild-type strain completely transferred the cytoplasm from conidia into the appressoria, leading to collapsed conidial morphology, while a large proportion of the cytoplasm was still intact in the mutant cells and consequently the conidia remained intact and turgid at 24 h after incubation (Fig 3A; black arrows).
The delayed mobilization of cytoplasmic content into the appressoria prompted us to further investigate the germinating conidia. In M. oryzae, conidia contain several sources of stored energy such as glycogen and lipids [7], and effective transfer of such materials is required for appressorial maturation and appressorium-mediated host penetration [7,28]. We therefore examined the cellular distribution of glycogen and lipid bodies during appressorium development. Upon iodine-staining abundant glycogen was seen in conidia, germ tubes, and incipient appressoria of the WT and ΔMovps35 from 0 h to 4 h during germination of conidia on a hydrophobic surface (Fig 4A). However, mobilization of glycogen was notably retarded in ΔMovps35 mutant after 8 h conidial germination (Fig 4A). Even after 24 h, a significantly higher proportion of mutant conidia contained glycogen (Fig 4B). Therefore, glycogen catabolism/hydrolysis was greatly delayed in the ΔMovps35 mutant (Fig 4A–4C). Next, we investigated the distribution of lipid bodies by Bodipy staining and confocal microscopy. The distribution of lipid bodies showed the same pattern as glycogen in the ΔMovps35 mutant, as shown in Fig 4D and Fig 4E. The data indicate that the mobilization of glycogen and lipid bodies from conidia to the appressoria is greatly reduced or blocked in the ΔMovps35 mutant.
Like MoVps35, the other two components, MoVps26 and MoVps29, of the retromer subcomplex were also found to be necessary for proper initiation of blast disease, as judged by the similar phenotypic defects in glycogen distribution, lipid droplet turnover, and appressorial turgor generation shown by the requisite ΔMovps26 and ΔMovps29 (S5 Fig). Taken together, these results indicate that the cargo-recognition subcomplex of the retromer comprising of MoVps35, MoVps26 and MoVps29, functions in mediating critical physiological/metabolic processes associated with pathogenic differentiation in M. oryzae.
Studies in M. oryzae have shown that proper conidiation and appressorium formation/function requires autophagy-assisted utilization of carbohydrate(s), glycogen or stored lipids [16,29,30]. Autophagy-deficient mutants (Δatg1, Δatg2, Δatg4, Δatg5, Δatg8, Δatg9 and Δatg18) show delayed breakdown of glycogen and lipid bodies, reduced turgor pressure and complete loss of pathogenesis in M. oryzae. [29,31–35]. Similar defects in ΔMovps35 prompted us to investigate whether MoVps35 is directly involved in regulating autophagy function(s) during appressorium-mediated host penetration. To test this idea, an Hh1-GFP (encodes nuclear localized Histone H1) was introduced into the wild type and ΔMovps35 strains to allow live cell imaging of nuclear degeneration associated with autophagic cell death [5] (S2 Table). In WT, the number of Hh1-GFP marked nuclei gradually decreased due to autophagy-based degeneration in conidial cells during appressorial maturation. As a result, a single nucleus remains intact in the mature appressorium of the wild-type strain at 24 hpi (84%, P<0.01) (Fig 5A). However, the majority of the Hh1-GFP expressing ΔMovps35 conidia (71%) contained more than one nucleus (Fig 5A). These data suggest that MoVps35 deficiency affects the autophagic cell death in conidia of M. oryzae.
A key event in autophagy is the formation of a double-membrane autophagosome, which engulfs portions of cytosol and entire organelles [36,37]. Thus, we sought to determine whether retromer plays a role in the formation of autophagosomes during appressorium development. We used the RFP-MoAtg8 as an epifluorescent marker for autophagosomes [16,38–40]. MoAtg8 is a ubiquitin-like protein that marks autophagosomal structures and is required for the formation of autophagosomes. RFP-MoAtg8 was expressed under the control of the endogenous MoATG8 promoter in the ΔMoatg8 or ΔMovps35 background. Analysis of ΔMoatg8 RFP-MoATG8 showed typical punctate autophagosomes and vacuolar autolysosomes widely distributed in conidia, germ tubes and appressoria (Figs 5B and S6). However, the ΔMovps35 RFP-MoATG8 strain showed no obvious punctate autophagosomes in germ tubes and appressoria, except for aggregated red epifluorescence signal in the vacuoles (Figs 5B and S6). We reasoned that RFP-MoAtg8 was probably retained in the vacuole and subsequently degraded by the vacuolar hydrolase upon loss of MoVps35 function. The dynamics of autophagic structures was investigated using time-lapse microscopy. In ΔMoatg8 RFP-MoATG8 strain, mobile spherical autophagosomes (about 1 μm diameter) fuse with vacuolar structures (2–5 μm diameter) and also dissociate from these structures (S1 Movie), indicating that autophagosomes cooperatively act to form autophagolysosomes or are recovered once autophagy is completed. By contrast, there are no obvious spherical autophagosomes in the ΔMovps35 RFP-MoATG8 strain (S2 Movie). These results suggest that MoVps35 is necessary for the proper localization of MoAtg8 and consequently required for autophagosome formation during appressorial maturation.
To investigate the mechanism by which retromer participates in the regulation of autophagosome formation, we monitored the dynamics of MoVps35 trafficking in M. oryzae using a ΔMovps35 pMoVPS35::MoVPS35-GFP strain (S2 Table). The pMoVPS35::MoVPS35-GFP construct complemented all the defects found in ΔMovps35 mutants (Fig 1), indicating that MoVps35-GFP is fully functional. MoVps35-GFP exhibited a mostly punctate pattern at or near the vacuolar membrane in conidia and mycelia (Fig 6A and 6B). The association with vacuoles was confirmed by staining with the lipophilic styryl dye FM4-64 (Fig 6A and 6B). Furthermore, we investigated the spatiotemporal dynamics of MoVps35-GFP during infection-related development. MoVps35-GFP consistently localized to small punctate/vesicular compartments (approximately 0.5–2.0 μm) in conidia, germ tubes and nascent appressoria by 2 h and 4 h, respectively (S7 Fig). During 8–24 h, the fluorescent signal was predominant in developing appressoria and gradually diminished in the conidia, consistent with conidial autophagic cell death (S7 Fig). Next, we monitored the dynamics of MoVps35-GFP movement in conidia and appressoria during conidial germination upon staining with FM4-64. MoVps35-GFP was present on highly mobile punctate structures in germinated conidia and developing appressoria (Fig 6C; S3 Movie). Strikingly, the movement of MoVps35-GFP-containing structures was not random but inherently associated with vacuolar membranes or late endosomes as judged by FM4-64 staining (Fig 6C; S3 Movie). We interpret these epifluorescence traces of MoVps35-GFP as evidence for vesicular trafficking in the late endosomal compartments. Moreover, the mobility of MoVps35-GFP depends on microtubules but not the F-actin cytoskeleton, because it was disrupted by the microtubule-destabilizing agent MBC (Methyl Benzimidazol-2-yl-Carbamate) but not by actin-depolymerizing drug LatA (latrunculin A) (S8 Fig and S4–S6 Movies). In addition, like MoVps35-GFP localization, MoVps26-GFP and MoVps29-GFP both exhibited a similar dynamic and punctate pattern on the vacuolar membrane (Fig 6D and 6E). Spatiotemporal dynamic of MoVps26-GFP and MoVps29-GFP distribution during infection-related development was also reminiscent of the MoVps35-GFP localization (S9 Fig, S7 and S8 Movies). These data suggest that retromer may function in the retrieval of cargo associated with vacuoles or autolysosomes in M. oryzae.
Given the essential role of retromer in the formation of autophagosomes and the apparent association with late endosomes, we suspected that MoVps35-GFP motility might contribute to the retrieval of MoAtg8 to pre-autophagosomal structures and autophagosomes. To test this hypothesis we first determined whether MoVps35 colocalizes with MoAtg8. pMoVps35::MoVPS35-GFP was introduced into the ΔMoatg8 RFP-MoATG8 strain (S2 Table) for localization and dynamic association analysis. In fresh harvested conidia, most of MoVps35-GFP vesicles were arranged adjacent to RFP-MoAtg8 labeled organelles, implying a potential association between these two compartments (Fig 7). Remarkably, a proportion of MoVps35-GFP punctae colocalized with the cytosolic RFP-MoAtg8 compartments as determined by line-scan and 3D reconstruction analysis (Fig 7A and 7B, see also S9 Movie). In order to test whether the colocalization existed during other developmental stages of M. oryzae, conidia from the dual-labeled ΔMoatg8 RFP-MoATG8 MoVPS35-GFP strain were incubated in vitro to observe germination and appressoria formation using confocal microscopy. At 2 h, many germ tubes initiated appressorium formation. In addition to the partially colocalized/overlapping RFP and GFP fluorescent signals detected in conidia, a small proportion of such colocalized signals were also apparent in the germ tubes (S10A Fig). A similar localization pattern was also found in developing appressoria (S10B Fig). Furthermore, RFP-MoAtg8 partially co-localized with MoVps35-GFP in vegetative hyphae under nitrogen starvation conditions that induce autophagy (S10C Fig). In order to directly record spatial and temporal association between MoVps35-GFP and RFP-MoAtg8, a real time imaging was applied. S10 Movie or time-lapse Fig 7C shows a conidium undergoing autophagy, RFP-MoAtg8 fluorescent were highly associated with oblong vacuoles (approximately 2–5 μm diameter) and spherical structures (approximately 1 μm diameter). Interestingly, we observed that mobile RFP-MoAtg8 puncta showed a rapid dissociation from the adjacent vacuoles, and at the same time the MoVps35-GFP also displayed very close colocalization with the punctate RFP-MoAtg8 (Fig 7C, arrows). This suggests that MoVps35 might play an important role for retrieving MoAtg8 from the vacuole, avoiding unnecessary degradation by vacuolar hydrolases.
To test whether MoVps35 contributes to the retrieval of MoAtg8 in vivo, we applied a GFP-trap/co-immunoprecipitation assay to pull down MoVps35-GFP and found that anti-RFP antibody was able to specifically detect a clear band of about 37 kD, the size of the truncated variant of RFP-MoAtg8 (Fig 8A). No unmodified full-length RFP-MoAtg8 band was detected from the proteins pulled down with MoVps35-GFP (Fig 8A). In the control experiment, both truncated RFP-MoAtg8 variant (approximately 37 kD) and full-length RFP-MoAtg8 (approximately 48 kD) were detected with an anti-RFP antibody with input protein isolated from the ΔMoatg8 RFP-MoATG8 MoVPS35-GFP strain (Fig 8A). These results indicate that MoVps35 is able to specifically interact with the truncated variant of MoAtg8, which is consistent with the time-lapse microscopy results of MoVps35-GFP and RFP-MoAtg8 in M. oryzae. Taken together, the MoVps35 acts through a direct interaction with truncated variant of MoAtg8 and contributes to its retrograde transport in M.oryzae.
Autophagy can be measured by examining the intracellular levels of cleaved and lipidated Atg8, viz Atg8PE, which is the key protein known to associate specifically with autophagosomes, as its levels correlate with the number of autophagosomes [38–40]. If MoVps35 is required for the retrieval of MoAtg8 to phagophore assembly sites (PAS)/autophagosomes, we would anticipate that the level of cleaved and lipidated MoAtg8 (ie., MoAtg8PE) would be reduced in ΔMovps35 under the conditions that induce autophagy. Therefore, we further used immunoblot assays to analyze the levels of the cleaved MoAtg8 protein in ΔMoatg8 and ΔMovps35 strains expressing RFP-MoATG8 (Fig 8B). We detected two forms of RFP-MoAtg8 using anti-RFP antibody, inferring that these represented the full-length and lipidated form, MoAtg8PE. Under nitrogen starvation condition, the amount of RFP-MoAtg8PE gradually increased in the ΔMoatg8 RFP-MoATG8, which correlated well with the extent of autophagosome formation (Fig 8B). However, in the ΔMovps35 RFP-MoATG8 strain, the levels of RFP-MoAtg8PE were greatly reduced and delayed in both CM and MM-N medium compared to the ΔMoatg8 RFP-MoATG8 strain, indicative of fewer autophagosomes (Fig 8B). Interestingly, the autophagy flux was not completely blocked in the ΔMovps35 since some RFP-MoAtg8PE still accumulated in the cells upon nitrogen starvation (Fig 8B). These findings are consistent with the microscopy results of scarce spherical autophagosomes in the ΔMovps35 RFP-MoATG8 strain.
Besides, the compromised expression levels of MoATG8 in the ΔMovps35 mutant could also lead to the lack of MoAtg8-marked autophagosome. Therefore, we further used qRT-PCR to assess the expression levels of the MoATG8 gene in ΔMovps35 and wild-type strains under the MM-N conditions. Compared to the wild-type strain, the expression levels of MoATG8 do not differ significantly in the ΔMovps35, but appear to be mildly upregulated (S11A Fig). We further analysed the expression levels of MoATG4 which is a key cysteine protease responsible for the cleavage of the carboxy terminus of MoAtg8 during the biogenesis of autophagosomes in M. oryzae [31]. Similar to MoATG8, the expression of MoATG4 was slightly higher in the ΔMovps35 as compared to the wild type (S11A Fig), suggesting that MoVps35 regulates the biogenesis of punctate autophagosomes primarily via modulating the retrieval of MoAtg8, but not by compromising the expression of MoATG8 and MoATG4 in M. oryzae. These results also explain why increased expression of MoATG8 is unable to rescue the phenotypic defects in ΔMovps35 mutant (S111B–S11F Fig).
Overall, the retrograde transport function of retromer complex, the close association between retromer core component MoVps35 and the key autophagy protein MoAtg8, and their tight functional link in autophagocytosis, asexual differentiation and plant infection provides an insight into a novel function of regulating the biogenesis of autophagosomes by retrieving cleaved MoAtg8 from the vacuolar compartments for targeting to the proper structures in the rice blast fungus (Fig 9).
In this work, we have addressed the outstanding question about the mechanism of Atg8 retrieval during autophagosome biogenesis. It was previously reported that the majority of Atg8 molecules were released into the cytoplasm before autophagosome–vacuole fusion, suggesting that Atg8 is retrieved for the formation of autophagosomes [41,42]. However, the molecular mechanisms for Atg8 retrieval remained unclear. Our results show that MoVps35, a retromer core component that functions in endosomal sorting, directly interacts with MoAtg8 and is associated with MoAtg8 retrieval process from the periphery of vacuoles (Fig 9). Loss of retromer function leads to the mistargeting of RFP-MoAtg8 to the vacuole and thus the impairment of the biogenesis of autophagosomes. Moreover, the phenotype of all retromer component mutants mimic the morphological disturbance observed with ΔMoatg8, including failure to undergo autophagic cell death during conidial germination, and defects in fungal appressorium-mediated pathogenesis. Taken together, our results provide the first clear linkage between the retrograde transport mediated by retromer complex and the autophagy-dependent plant infection.
The highly conserved retromer is well known to function in the retrieval of recycling cargos to TGN in the retrograde pathway, but its unconventional roles are now beginning to emerge. In Arabidopsis thalliana, the retromer complex components Vps35, Vps26 and Vps29 localize to the prevacuolar compartment (PVC) and are essential for normal PVC morphology [43,44]. Mutations in VPS35 or VPS29 in A. thalliana lead to a dwarf phenotype and defects in PIN protein repolarization, embryogenesis, plant growth, and leaf senescence [45,46]. In Drosophila melanogaster, Vps35 function is necessary for normal endocytic trafficking and organization of the F-actin cytoskeleton [47]. Loss of Vps35 severely affects endocytosis and the localization of a number of endocytic proteins, causes defects in signaling pathways in haemocytes and at the neuromuscular junction, and leads to increased levels of F-actin [47]. Interestingly, a recent report found that the mammalian retromer regulates trafficking and subsequent incorporation of HIV-1 envelope glycoprotein (Env) into virions [48]. Inactivating retromer alters Env localization, cell surface expression and incorporation into virions, and the binding of retromer to the Env cytoplasmic tail is required for these functions [48]. Our study suggests an important role in retromer is essential for the autophagy-dependent plant infection in M. oryzae, thus expanding the functions of retromer. Interestingly, although the retromer is important for appressorium-mediated host penetration, it is not essential for colonization therein. This suggests that the retromer and efficient autophagy is not required for suppression of host defense, invasive growth within the rice cells, or spread from cell-to-cell in the host. It will be interesting to investigate whether the retromer plays a similar role in autophagy and infection-related development in other fungus-plant pathosystems.
Our cell biological, biochemical and genetic analyses demonstrate that the retromer is essential for autophagy by activating the formation of autophagosomes through the Vps35’s direct interaction with and retrieval of MoAtg8 in the M. oryzae. This role of the retromer could be widespread throughout the eukaryotic kingdom. Recently, two studies have implicated a role for the retromer in autophagy in yeast and mammalian cells. Dengjel and his colleagues used quantitative proteomics to identify Vps35/retromer as a stimulus-dependent interacting partner of autophagosomes in human breast cancer cells [49]. A further test for autophagic response revealed that a significantly lower autophagic activity in the yeast vps35 null mutant [49]. Another study identified that the mutant VPS35 allele that causes Parkinson’s disease (PD), VPS35 D620N, also impairs autophagy and alters the trafficking of the multi-pass transmembrane autophagy protein Atg9A in mammalian cells [50]. However, none of these studies addressed the mechanism by which the retromer regulates autophagocytosis. Here, we provide genetic, live-cell imaging and biochemical evidence that MoVps35-GFP positively regulates the autophagy process via recycling or retrieving truncated MoAtg8 to the appropriate compartments, likely PAS, for regeneration of autophagosomes. In a previous report in M. oryzae, EGFP-MgAtg9 and DsRed2-MgAtg8 displayed significant colocalization [35], suggesting that these components interact in conidia. Although our data identified that MoAtg8 mislocalization is a likely contributor to the impaired autophagosome formation in the ΔMovps35 mutant, it is conceivable that other MoAtg proteins are subjected to the retromer complex mediated retrograde transportation too. Given the conservation of the retromer and its role in autophagy, this mechanism likely provides a paradigm for a novel role of retromer in the regulation of autophagy in various eukaryotic organisms.
Autophagy plays an important role in recycling cellular components in eukaryotic cells. Although more than 30 genes have been described for involvement in autophagy, in yeast [13,14], the mechanisms for the activation of this process and for the recycling of its components for autophagosome formation remain poorly understood. A ubiquitin-like system mediates the conjugation of the cleaved Atg8 to the lipid phosphatidylethanolamine (PE), and this conjugate (Atg8PE) is then tethered to autophagosome membrane, where it is necessary for phagophore expansion during autophagosome formation [36,39]. In addition to the lipidation of Atg8, delipidation of Atg8 is also required for autophagosome maturation [41]. In yeast, the cleaved Atg8 has a C-terminal glycine residue that exists in two different forms: a lipidated (Atg8PE) and a deconjugated form [51]. The former is involved in autophagosome formation; the role of latter is to release outer membrane-bound Atg8 upon completion of the autophagosome, presumably for reuse in subsequent rounds of autophagosome formation [41]. However, how the deconjugated Atg8 gets to the PAS for autophagosome formation is currently unknown. Our data indicate a direct interaction between vacuolar membrane conjugated MoAtg8 and Vps35 and suggest a novel mechanism for lipidated MoAtg8 recycling via retromer-mediated transport in M. oryzae. It is possible that the retrieval machinery and/or the cargo-recognition complex of the retromer recognize the lipidated MoAtg8 as a specific cargo/substrate.
All strains used in this study were listed in S2 Table. M. oryzae wild-type Δku70 [15] and all mutant strains were grown on complete medium (CM), starch yeast medium (SYM), oatmeal agar medium (OAT) and rice-polish agar medium (RPA) for mycelial growth assays and on RPA medium for conidiation assays as previously described [52]. To test the sensitivity against cell-wall-disrupting agents, vegetative growth of fungal strains was monitored on CM plates with 200 μg/mL congo red (CR) or 200 μg/mL Calcofluor White (CFW) or 0.01% sodium dodecyl sulfate (SDS). For oxidative stress sensitivity assays, M. oryzae strains were grown on CM containing 5 mM H2O2 and the sensitivity was evaluated by measuring the colony diameter of 6-day-old cultures. Experimental results were verified with a minimum of two strains of the same genotype. All experiments were repeated at least three times.
The M. oryzae protoplast preparation and fungal transformation were performed by following standard protocols [53]. Hygromycin- or neomycin-resistant transformants were selected on media supplemented with 250 μg/mL hygromycin B (Roche Applied Science) or 200 μg/mL G418 (Invitrogen).
To generate ΔMovps35 mutant, a 1,176-bp fragment upstream from MoVPS35 was amplified with primers MO05089AF and MO05089AR, and this amplicon was subsequently cloned into the XhoI and EcoRI sites upstream of the hph cassette on pCX63 [24]. Then, 1.14 kb fragment downstream of MoVPS35 was amplified with primers MO05089BF and MO05089BR, and cloned into the BamHI and XbaI sites downstream of HPH cassette, and this plasmid was transformed into protoplasts of the wild-type Δku70 strain. Hygromycin-resistant transformants were screened by PCR with primers MO05089UA and H853 and primers MO05089OF and MO05089OR (S3 Table). At least two isolates that tested positive with PCR were further verified by Southern blot analysis performed with the DIG-High Prime DNA Labeling and Detection Starter Kit I Roche (Roche, Mannheim, Germany).
The split-marker approach was used to generate gene replacement constructs for other components of the retromer complex [54]. Primers used to amplify the flanking sequences of MoVPS26 and MoVPS29 are listed in S3 Table. Each construct was transformed into protoplasts of Δku70 to generate the ΔMovps26 and ΔMovps29 deletion mutants (S2 Table). Putative knockout mutants were identified by PCR screening and confirmed by DNA gel blot analysis.
The MoVps35-GFP fusion vector, named pGM-MoVps35-GFP, was constructed by amplification of 5,027-bp fragment including 2,944-bp MoVps35 coding sequence and a 2,083-bp native promoter region using primers MO05089CF and MO05089CR (S3 Table). The 5,027-bp PCR product was then cloned into pGEM-T easy vector (Promega) to generate pGM-MoVps35. The GFP allele [55] was amplified using primers Bgl II-GFPF and Bgl II-GFPR (S3 Table), then cloned into pGEM-T easy vector. It was subsequently digested with BglII to release the GFP allele with BglII sticky ends, which was inserted into BglII site of pGM-MoVps35 to create pGM-MoVps35-GFP. We verified the orientation of GFP insertion and in-frame fusion by sequencing the pGM-MoVps35-GFP vector. To generate MoVps35-GFP strain, pGM-MoVps35-GFP construct was co-transformed into protoplasts of the target mutant along with a vector harboring neomycin-resistance marker (pKNT). Transformants carrying a single insertion were screened by PCR with requisite primer pairs (S3 Table) and further confirmed by Southern blot analysis. The same approach was used to generate gene fusion GFP constructs for other components of the retromer complex. Primers used to amplify the complementation sequences of MoVPS26 and MoVPS29 are listed in S3 Table.
For appressorial assays, conidia were harvested from 10-day-old OAT or RPA cultures. Aliquots (30 μL) of conidial suspensions (5×104 conidia/mL in sterile water) were applied on the hydrophobic side of Gelbond film (Cambrex Bio Science) and incubated under humid conditions at room temperature. Conidial germination and appressorium formation were examined at 0.5, 1, 2, 4, 8 and 24 h post incubation. For penetration assays, conidial suspension in sterile water was inoculated on onion epidermal cells or barley leaf and assessed after 24 h, 48 h and 72 h. Penetration pegs and infection hyphae were detected by staining for papillary callose deposits using Aniline blue [56]. For pathogenicity assays, two-week-old seedlings of rice (Oryza sativa L.) cultivar CO39 were used for spray inoculation assays as described [52]. Eight-day-old seedlings of barley cultivar Golden Promise were also used for drop inoculation and mycelial plug assays on the non-wounded or wounded barley leaves [57].
For glycogen staining, the M. oryzae conidia were inoculated on hydrophobic plastic coverslips for different time points and stained with a solution consisting of 60 mg/mL of KI and 10 mg/mL of I2 in distilled water [7]. Yellowish-brown glycogen deposits became visible immediately in bright field. For lipid bodies staining, samples were stained with Bodipy (D3922, Invitrogen) to detect neutral lipids. Bodipy was used at 10 μg/mL (stock 1 mg/mL in ethanol) in PBS buffer. All samples were examined and photographed by using an Olympus-BX51 fluorescence microscope with a cooled CCD camera (DP72, Olympus, Japan). Calcofluor White (Sigma-Aldrich, USA) was used at 3 μg/mL to visualize cell wall and septa of conidia. To visualize the vacuolar membrane, conidia, vegetative hyphae and germinated conidia were treated with 4 μg/mL FM4-64 solution for 30–60 min before observed under the confocal microscope. To examine the effects of microtubule inhibitor methyl 1-(butylcarbamoyl)-2-benzimidazolecarbamate (MBC) or the actin inhibitor latrunculin A (LatA) on trafficking ability of MoVps35 in cells, The MoVps35-GFP strain was inoculated on hydrophobic plastic coverslips and treated for 30 minutes with MBC (final concentration 10 μM) or LatA (final concentration 10 μM) at the hooking stage. A 0.1% DMSO solvent control was used in these assays.
Standard molecular manipulations were performed as described [58]. Total RNA was isolated from mycelia with TRIzol reagent (Invitrogen). Purified RNA was treated with DNase (Takara) and was verified as DNA free by using it directly as template in a PCR assay. First-strand cDNA was synthesized with the M-MLV reverse transcriptase (Invitrogen), and qRT-PCR was performed with the Eppendorf Mastercycler ep Realplex2 PCR system using SYBR Premix Ex TaqTM (RR420A, Takara). Primers used to amplify selected genes in qRT-PCR reactions are listed in S3 Table.
Yeast two-hybrid assay was carried out as indicated in MATCHMAKER GAL4 Two-Hybrid System 3 (Clontech). The full-length cDNA of MoVPS35 was amplified with the primer pair MO05089BDF/MO05089BDR (S3 Table) and cloned into pGBKT7 as the bait vector BD-Movps35. The full-length cDNAs of MoVPS26 and MoVPS29 were amplified with the primer pairs MO04830ADF/MO04830ADR and MO02524ADF/MO02524ADR (S3 Table), respectively, and were cloned into pGADT7 as the prey vectors AD-Movps26 and AD-Movps29. The resultant bait and prey vectors were confirmed by sequencing and were co-transformed into the yeast strain AH109. The Leu+ and Trp+ yeast transformants were isolated and assayed for growth on SD-Trp-Leu-His-Ade medium at specified concentrations. Yeast strains for positive and negative controls were as described in the Matchmaker kit.
For total protein extraction, mycelia grown in liquid CM and MM-N medium (used for nitrogen starvation, 0.5 g/L KCl, 0.5 g/L MgSO4, 1.5 g/L KH2PO4, 10 g/L glucose, pH 6.5) were ground into a fine powder in liquid nitrogen and resuspended in 0.6 mL extraction buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 2 mM PMSF and 1% Triton X-100). The supernatants were centrifuged 15,000 g for 25 min at 4°C to remove cell debris. Total protein concentration was measured using the Bio-Rad Protein Assay and separated on a 12.5% SDS PAGE gel and transferred to PVDF membranes for Western blot analysis. The expression of RFP-Atg8 was detected with anti-RFP (Clontech, USA). The horseradish peroxidase–conjugated secondary antibody and the ECL kit (Amersham Biosciences, Germany) were used to detect the chemiluminescent signals.
For the immunoprecipitation of GFP-fusion-proteins from cellular extracts, equal concentration of total proteins were isolated and incubated with 20–30 μL of GFP-Trap agarose beads (ChromoTek, Germany) and carried out as manufacturer’s instructions. Proteins eluted from GFP-Trap agarose beads were analyzed by immunoblot detection with the anti-RFP (Clontech, USA), anti-GFP (Sigma-Aldrich) antibodies and anti-Actin (Sigma-Aldrich).
Conventional epifluorescence and differential interference contrast (DIC) microscopy was performed with a Olympus BX51 microscope (Olympus, Japan), using a 40x 1.3 NA (numerical aperture), 60x 1.35 NA or 100x 1.40 NA Olympus oil immersion objective lens. Images were acquired using an Olympus DP72 camera and analyzed with DT2-BSW image-processing software. Fluorescence was observed with Olympus U-RFL-T mercury lamp source. The filter sets used were: DAPI, GFP and RFP or FM4-64. Alternatively, confocal microscopy was used for time-lapse or live cell fluorescence imaging by using the Nikon TiE system (Nikon, Japan) as described [59]. The elapsed time is indicated in seconds. Image processing and figure preparation was performed using Image J, Adobe Photoshop, PowerPoint and Microsoft Excel.
Sequence data from this article can be found in the GenBank/EMBL databases under the following accession numbers: S. cerevisiae VPS35 (NP_012381), S. cerevisiae VPS29 (NP_011876), S. cerevisiae VPS26 (NP_012482), S. cerevisiae VPS17 (NP_014775), S. cerevisiae VPS5 (NP_014712), M. oryzae MoVPS35 (XP_003712611), M. oryzae MoVPS29 (XP_003709334), M. oryzae MoVPS26 (XP_003713759), M. oryzae MoVPS17 (XP_003714383), M. oryzae MoVPS5 (XP_003709457).
|
10.1371/journal.ppat.1005794 | Thermoregulation of Meningococcal fHbp, an Important Virulence Factor and Vaccine Antigen, Is Mediated by Anti-ribosomal Binding Site Sequences in the Open Reading Frame | During colonisation of the upper respiratory tract, bacteria are exposed to gradients of temperatures. Neisseria meningitidis is often present in the nasopharynx of healthy individuals, yet can occasionally cause severe disseminated disease. The meningococcus can evade the human complement system using a range of strategies that include recruitment of the negative complement regulator, factor H (CFH) via factor H binding protein (fHbp). We have shown previously that fHbp levels are influenced by the ambient temperature, with more fHbp produced at higher temperatures (i.e. at 37°C compared with 30°C). Here we further characterise the mechanisms underlying thermoregulation of fHbp, which occurs gradually over a physiologically relevant range of temperatures. We show that fHbp thermoregulation is not dependent on the promoters governing transcription of the bi- or mono-cistronic fHbp mRNA, or on meningococcal specific transcription factors. Instead, fHbp thermoregulation requires sequences located in the translated region of the mono-cistronic fHbp mRNA. Site-directed mutagenesis demonstrated that two anti-ribosomal binding sequences within the coding region of the fHbp transcript are involved in fHbp thermoregulation. Our results shed further light on mechanisms underlying the control of the production of this important virulence factor and vaccine antigen.
| The bacterium Neisseria meningitidis is exquisitely adapted to survive in the human host, and possesses several mechanisms to interact with host cells in the upper airway and to circumvent immune responses. However, the mechanisms that govern the expression of factors that contribute to colonisation and disease are incompletely understood. In this work, we further characterise how temperature influences the production of factor H binding protein (fHbp) by the meningococcus; fHbp recruits human complement proteins to the surface of the bacterium, and is an important vaccine antigen. We show that thermoregulation of fHbp occurs gradually over a physiological range of temperatures found in the upper airway, the site of colonisation. This regulation does not require specific meningococcal transcription factors, and sequence analysis indicates that fHbp mRNA forms a secondary structure which could act as an RNA thermosensor. Additional studies demonstrate that there are two specific sequences within the coding region of fHbp mRNA are important for thermosensing and could base-pair to the ribosome binding site, thus blocking translation of this protein. As fHbp is thermoregulated, vaccines that target this antigen might not impose a high level of selective pressure on the bacterium at the mucosal surface, thereby limiting herd immunity induce by fHbp containing vaccines.
| Neisseria meningitidis is a harmless member of the human nasopharyngeal flora in a significant proportion of healthy individuals [1]. However in some instances, the bacterium spreads from the upper airway into the systemic circulation, where it can replicate and spread to the rest of the body [2], especially the cerebrospinal fluid, resulting in meningitis. Therefore the meningococcus remains an important human pathogen in infants and young adults [3].
To survive in the human host, the meningococcus has evolved several mechanisms that enable it to evade the immune system [4]. In particular, the complement system is critical for protection against systemic N. meningitidis infection, evident from the increased susceptibility of individuals with defects in the complement system and findings from a genome wide association study [5,6]. The bacterium evades complement mediated killing by expressing a polysaccharide capsule, sialylation of its lipopolysaccharide, and by binding complement factor H (CFH), the major negative regulator of the alternative complement pathway [7]. CFH is recruited by high affinity interactions with factor H binding protein (fHbp) [8]. CFH competes with complement factor H related protein-3 (CFHR3) for binding to fHbp on the meningococcus [9]; CFHR3 is a competitive inhibitor of CFH binding to the meningococcal surface, and relative levels of CFH and CFHR3 in individuals are likely to determine host genetic susceptibility to meningococcal disease in the general population [9]. Furthermore fHbp is a surface lipoprotein which is an important component of two vaccines which are now licensed for preventing meningococcal disease [10,11].
Within the upper airway, the bacterium is exposed to gradients in ambient temperature that occur in relation to the anatomical location the phase in the respiratory cycle, and the presence of local inflammation [12,13]. The nasal epithelium contains a complex vascular network with a relatively high blood flow, which forms a heat exchanger, to condition inspired air; the temperature of air entering the nasopharynx rises rapidly from the nostrils, increasing to around 32–34°C by the time it reaches the glottis at the end of inspiration [14,15]. During colonisation, the meningococcus is found both on the epithelial surface as well as deep in the submucosal layer surface [16], so the bacterium will be will be exposed to a range of temperatures in these sites [17]. Additionally, during the development of invasive disease, N. meningitidis passes from the lower temperatures in the upper airway to the core body temperature of 37°C or higher with a febrile response to infection [18]. Therefore, temperature is likely to be an important environmental cue for the meningococcus during colonisation and the development of disease.
We have shown previously that three meningococcal genes, cssA, fHbp, and lst which encode factors contributing to capsule biosynthesis, fHbp, and LPS sialylation, respectively, are subject to thermoregulation. Of note, the sialic acid-capsule biosynthesis operon is controlled by an RNA thermosensor [19]. RNA thermosensors are usually located in the 5´-untranslated region (5´-UTR) of an mRNA, and form a secondary structure at lower temperatures that prevents protein translation by blocking access of ribosomes to the nascent mRNA. As the temperature rises, the secondary structure undergoes a conformational change, exposing the ribosome binding site (RBS), and allowing translation in response to the elevated temperature. RNA thermosensors have been described in an increasing number of microbes [20], especially enteric pathogens which are subject to a large fluctuation in temperature upon ingestion by mammalian hosts from the external environment.
Here we describe the mechanisms governing thermoregulation of fHbp. Expression of fHbp in N. meningitidis is initiated from two transcriptional start sites, resulting in a bi-cistronic transcript (including the upstream gene, nmb1869), or a mono-cistronic transcript from the fHbp promoter, PfHbp [21]. It has been shown that transcription from PfHbp is responsive to oxygen limitation, and regulated in an FNR-dependent manner [21]. We demonstrate that fHbp levels are governed by an RNA thermosensor and demonstrate that sequences in the coding region of fHbp contribute to thermosensing and translation efficiency; two sequences within the fHbp coding region, which are complementarity to the ribosome-binding site, are necessary for the thermoregulation of this key virulence factor and vaccine antigen.
We have shown previously that levels of fHbp in the meningococcus increase following growth at higher temperatures [19]. To determine whether this affects the level of fHbp on the bacterial surface, we analysed bacteria grown at different temperatures by flow cytometry using αfHbp pAbs (Fig 1A). The results demonstrate that a rise in temperature is associated with a significant increase in the amount of fHbp on the meningococcus between bacteria grown at 30°C and 42°C.
To establish whether this change is mediated by an alteration in the transcription of fHbp, N. meningitidis was grown to mid-log phase at 30°C, 37°C or 42°C, and fHbp mRNA detected by Northern analysis. Of note, levels of the bi- and mono-cistronic transcripts of fHbp mRNA were unaffected by temperature, demonstrated by Northern blot analysis (Fig 1B). However, analysis of samples from the same experiment demonstrate that there was an accompanying clear increase in fHbp levels at higher temperatures (Fig 1C) as described previously [19], indicating that thermal regulation of fHbp occurs at the post-transcriptional level.
To examine whether the increase in fHbp at higher temperatures results from an alteration in protein stability, N. meningitidis was grown at different temperatures to mid-log phase then de novo protein synthesis was inhibited by adding spectinomycin to cultures. Western blot analyses of samples taken at times afterwards demonstrate that there is no detectable difference in the stability of fHbp in N. meningitidis at 30 and 37°C (Fig 1D). Taken together, these results indicate that the thermoregulation of fHbp does not result from a change in transcription or protein degradation.
Thermoregulation of fHbp was analysed in Escherichia coli using a plasmid (pfHbp) containing the mono-cistronic fHbp and its promoter, PfHbp (Fig 2A and [19]); results demonstrate that the bi-cistronic mRNA is dispensable for fHbp thermoregulation. To further define the mechanism of thermoregulation, we examined the level of fHbp mRNA in E. coli harbouring pfHbp grown at 30°C, 37°C and 42°C; consistent with results obtained with N. meningitidis, there was no discernible change in transcript levels following growth at different temperatures (Fig 2B), even though fHbp was subject to clear thermoregulation in the same samples (Fig 2C). To identify sequences that contribute to thermoregulation, we generated a plasmid (pfHbpΔPfHbp) that includes the 5´-UTR of the transcript but lacks its native promoter, PfHbp; in this plasmid, transcription of fHbp occurs from the T7 promoter in the vector (Fig 2A) or the SP6 promoter in the opposite orientation. Again a temperature-dependent increase in fHbp levels was observed in E. coli harbouring fHbp in either orientation in this plasmid (Fig 2D and S1 Fig), demonstrating that PfHbp is not required for fHbp thermoregulation. In contrast, deletion of the 5´ UTR from pfHbpΔPfHbp (generating pfHbpΔPfhbpΔUTR) led to loss of fHbp thermoregulation, indicating that the 5´-UTR of the monocistronic fHbp mRNA is required for thermoregulation in this system lacking the bicistronic transcript (Fig 2E).
To examine the relevance of these findings in N. meningitidis, we substituted the open reading frame (ORF) upstream of fHbp (i.e. nmb1869, 1065 bp in length) with the 1120 bp kanamycin resistance marker in N. meningitidis to i) leave the promoter for the bi-cistronic transcript intact, and ii) disrupt Pfhbp promoter (Fig 3A and 3B), generating strain mutPfHbp-10. Northern analysis confirmed loss of the monocistronic fHbp mRNA (Fig 3C), and thermoregulation was still observed in the meningococcus with the bi-cistronic transcript alone (Fig 3D). Together with the findings in E. coli (Fig 2), the results provide evidence that the monocistronic PfHbp is dispensable for fHbp thermoregulation.
Next we generated plasmids containing the 5´-UTR with the first one, five or nine codons of fHbp fused to GFP as a reporter (pfHbp1C-gfp, pfHbp5C-gfp and pfHbp9C-gfp, respectively, Fig 4A); we were unable to generate constructs with more codons, probably because of toxic effects of the fHbp leader sequence in E. coli. The plasmids lack a T7 promoter [22], so expression is governed by the native fHbp promoter. We observed a clear correlation between number of fHbp codons in the plasmid and levels of the GFP reporter, with approximately 3.5-fold higher GFP in bacteria harbouring pfHbp9C-gfp compared with fHbp1C-gfp following growth at 37°C (Fig 4B). The increase in GFP levels observed with pfHbp9C-gfp is not caused by a change in protein or RNA stability (S2 Fig), but instead correlates with the predicted minimum free energy values (using the Vienna RNAfold package, Fig 4C) and putative secondary structures (S3 Fig).
To determine whether thermoregulation of fHbp was also observed in a cell free system, in vitro transcription/translation assays were performed with pfHbp1C-gfp, pfHbp5C-gfp and pfHbp9C-gfp. DNA from these plasmids was transcribed and translated in a continuous manner at 30°C or 37°C, and GFP production assayed by Western blot analysis (Fig 5A). There was increased GFP generated in assays performed at the higher temperature with pfHbp5C-gfp and pfHbp9C-gfp; however thermoregulation was not observed in assays containing pfHbp1C-gfp. Overall levels of GFP produced were higher in assays with plasmids harbouring more codons, consistent with sequences in the ORF enhancing translation efficiency. Therefore, the fHbp9C-GFP construct was used in subsequent experiments.
To further examine the dynamics of fHbp thermoregulation, in vitro transcription/translation reactions were conducted with pfHbp9C-gfp at temperatures between 30°C and 38°C. Equal amounts of plasmid were used as the substrate for in vitro transcription/translation reactions performed at 30°C, 34°C, 36°C, 37°C and 38°C for one hour, and reaction products were analysed by Western blotting. The results demonstrate that levels of the GFP reporter rise gradually in response to increasing temperature over a physiological range (Fig 5B) similar to the cssA thermosensor [19].
Next, we performed mutagenesis to identify key nucleotides involved in the formation of putative RNA secondary structures in the fHbp mRNA that mediate thermoregulation. Several point mutations were introduced upstream of the fHbp GTG start codon in pfHbp9C-gfp (Fig 6A), and their effect examined during growth of E. coli at 30°C, 37°C and 42°C. Proteins were isolated from the bacteria and levels of GFP expression examined by Western blot analysis. However, none of the single base substitutions in the 5´-UTR disrupted thermoregulation (Fig 6B), indicating that these changes are insufficient to perturb RNA secondary structures that are necessary for thermosensing.
On further inspection, we identified two copies of a 6 bp repeat (CUGCCU) within the fHbp coding region, close to the start codon and potentially able to base-pair with the ribosome binding site (RBS); we termed these repeats αRBS-1 and αRBS-2 (Fig 6A). We reasoned that the two cytosines could base-pair with the guanines in the RBS (AGGAG) and prevent ribosomes binding to the nascent transcript. Therefore, site-directed mutagenesis was performed to modify the two cytosines to guanines in each αRBS (i.e. CUGCCU to CUGGGU) either singly (generating mutαRBS-1 and mutαRBS-2), or together (mutαRBS-1/2). Constructs were introduced into E. coli, which was then grown at 30°C, 37°C and 42°C, and the level of GFP examined by Western blot analysis. All three constructs displayed reduced fHbp thermoregulation when compared to the plasmid containing wild-type sequences. Comparing cultures grown at 30°C or 42°C, the level of GFP increased by 1.5-fold and 1.6-fold for cells harbouring mutαRBS-1 and mutαRBS-2, respectively, whereas cells containing mutαRBS-1/2, showed a 1.2-fold increase (Fig 7A). In contrast, GFP levels increased by 5.3-fold with the wild-type sequence in the plasmid. RecA levels were stable in all conditions, and modification of the αRBS sequences did not affect the stability of the mRNA (S4 Fig). We also introduced the αRBS mutations into pfHbp, and found that modifications impaired the thermoregulation of fHbp (S5 Fig), although changing the fHbp sequence affected its detection by western blot analysis. It is likely that the lack of detection of fHbp resulted in changes from the protein sequence (S5 Fig); mutαRBS-1 results in an Ala to Gly substitution, while mutαRBS-2 alters a Cys to Trp, and a Leu to Val (S5 Fig).
Next we examined whether the α-RBS affect the translation of fHbp. in vitro translation reactions were performed with equal amounts of total cellular RNA extracted from E. coli strains harbouring the different constructs, and assays conducted at 30°C, 37°C or 42°C. The in vitro synthesised proteins were extracted and assayed by Western blot analysis. The wild-type construct showed thermoregulation, whereas the three constructs with altered α-RBS did not display thermoregulation (Fig 7B), similar to results found in E. coli; RecA levels were the same for all conditions examined. Taken together, these results demonstrate that both α-RBS sequences contribute to fHbp thermosensing. Additionally, we attempted to introduce the α-RBS mutations into N. meningitidis either under the control of the mono-cistronic or bi-cistronic promoter (S6 Fig). However similar to results in E. coli (S5 Fig), fHbp was undetectable by Western blot analysis of cell lysates of the meningococcus containing mutαRBS-1/2, presumably because the amino acid changes affected the processing of the N-terminal region of fHbp and affected its stability. We considered introducing single C to G mutations into α-RBS1 and/or α-RBS2, but these nucleotide changes of the α-RBSs are not predicted to impact the overall structure of fHbp mRNA (S7 Fig).
The meningococcus has successfully evolved to survive in the human nasopharynx, which is its only natural habitat. Colonisation with this bacterium is frequent in young adults (up to 70% among those living in institutionalised settings such as prisons and barracks) [1], and a single strain can persist in the same individual for months in the upper airway [23]. Within this environment, there are several temperature gradients. For example, the temperature on the surface of the anterior nares is around 30°C at the end of inspiration, and rises to around 34°C in the posterior nasopharynx and tonsillar region [13]. Both these sites on the mucosal surface are significantly cooler than the core body temperature of 37°C, where the bacterium replicates during invasive disease. Additionally, fluctuations in the local temperature will be generated by acute inflammation (resulting in increased blood flow) and systemic illnesses, such as influenza, which provoke a febrile response [24].
We have shown previously that temperature is an important environmental cue for the meningococcus [19]. The production of fHbp and enzymes necessary for capsule biosynthesis and LPS sialylation are increased as the temperature rises from 30°C through to 42°C. The enhancement of immune evasion by the bacterium may reflect transition from the cool anterior portion of the nasopharynx (lined by keratinised squamous cells with little lymphoid tissue) to the warmer posterior nasopharynx, where the meningococcus is found in tonsillar tissue [16]. Furthermore, increased temperature could act as a signal of local inflammation [18] or transition to the sub-mucosal layer, where the bacterium will be exposed to immune effectors.
A notable feature of the meningococcus is its capacity to evade exclusion by the human immune system [4]. Surface structures expressed by N. meningitidis, including Type IV pili, undergo frequent antigenic variation in certain strains [25], while the serogroup B capsule is a molecular mimic of a human post-translational modification [26]. Additionally, fHbp mediates high affinity interactions with the complement regulator, CFH using ligand mimicry [8], and a CFH antagonist CFHR3 [9].
We have shown previously that the biosynthesis of sialic acid-containing capsules is governed by an RNA thermosensor in N. meningitidis [19]. A fundamental advantage of RNA thermosensors is that they operate at the post-transcriptional level [20], so do not require a dedicated sensing pathway or de novo transcription to exert their effect. Therefore, RNA thermosensors are an energetically efficient strategy that offers rapid responses to abrupt changes in the temperature, as seen during the onset of inflammation.
For many bacteria, an increase in temperature is a key signal during their acquisition and ingestion by a mammalian host. For example, in enteric pathogens, RNA thermosensors are known to modulate expression of a transcriptional regulator, which in turn orchestrates the expression of a suite of genes that enable the pathogen to survive in its new environment [20]. Examples include the RNA thermosensors in prfA in Listeria monocytogenes [22], agsA in Salmonella [27], and lcrF in Yersinia spp. [28,29]. Thermosensors also govern iron acquisition systems in Shigella and E. coli [30], and Vibrio cholerae [31]. We found that the prfA thermosensor operates as an ON:OFF switch [32], similar to an RNA thermosensor in Cyanobacteria [33]. The abrupt thermodynamic response of the prfA thermosensor is consistent with Listeria undergoing large temperature fluctuations during its transition from the external environment to the intestinal tract. However, our in vitro transcription/translation assays reveal that the fHbp thermosensor exhibits a gradual change over a range of physiologically relevant temperatures i.e. 32–35°C [17], similar to the N. meningitidis capsule thermosensor [19]. The careful calibration of responses to temperature may be a feature of thermosensors in microbes that exist in close association with thermal gradients within hosts, particularly in the upper airway.
Previous work demonstrated that the expression of fHbp is mediated by bi-cistronic and mono-cistronic transcripts, with the shorter mono-cistronic fHbp transcript controlled by the global regulator of anaerobic metabolism, FNR [21]. Several lines of evidence demonstrate that the mono-cistronic transcript is sufficient but not necessary for fHbp thermoregulation, which is also detected with the bi-cistronic transcript in N. meningitidis. Thermoregulation was detected with the mono-cistronic transcript alone in E. coli and using in vitro transcription/translation assays. Additionally we generated a N. meningitidis mutant lacking the bi-cistronic transcript (by inserting a kanamycin resistance cassette with a terminator upstream of PfHbp). Even though fHbp levels are lower than in the wild-type strain, consistent with previous work [21], the strain still exhibited thermoregulation of fHbp (S6 Fig).
The fHbp transcript is not predicted to contain a ROSE element by RNA structure predictions and does not contain a U(U/C)GCU sequence close to the RBS, which is often present in this class of thermosensors [20]. Furthermore, the potential involvement of a temperature regulated sRNA is unlikely as fHbp thermoregulation occurs both in E. coli and in vitro assays. Instead, our experiments indicate that the mechanism responsible for fHbp thermoregulation depends on two α-RBS sequences located in the fHbp ORF. Modification of each α-RBS independently reduced thermoregulation, which was virtually abolished when both α-RBSs were altered, indicating some redundancy in their function. In contrast, mutation of the fHbp 5´-UTR had little or no effect on thermoregulation. Therefore our results support the model in which the fHbp transcript forms a stem loop at lower temperatures with the RBS occluded by base pairing with one of two α-RBS sequences located in the ORF. Attempts were also made to introduce the α-RBS mutations into N. meningitidis. However, we could not detect fHbp in the meningococcus containing modified α-RBSs (S6 Fig), even though fHbp was detected in E. coli using equivalent constructs on a multi-copy plasmid. The nucleotide changes lead to an alteration in the amino acid sequence of fHbp which might affect lipoprotein localisation and processing, which is distinct in N. meningitidis and E. coli [34]. We considered other modifications to the αRBS sequences. However the wobble rule of RNA:RNA binding [35] means that G can bind several bases (e.g. U and A), so their impact on the secondary structure of the RNA would have been uncertain. The precise contribution of the individual α-RBS sequences to the fHbp RNA thermosensor will be determined in future structural studies.
We also found that the ORF of fHbp contributes to levels of the protein in the meningococcus. Inclusion of first nine codons of fHbp in GFP fusions resulted in significantly higher protein levels of reporter the compared with constructs containing less sequence. This was not mediated by changes in protein stability (S2A Fig), and could result from increased efficiency of translation although these findings need to be confirmed in N. meningitidis.
Of note, attempts were made to generate GFP fusions with 12 and 20 fHbp codons in E. coli to define sequences required for maximal fHbp expression. However, it was not possible to obtain the fHbp12C-gfp and fHbp20C-gfp constructs. fHbp is a surface lipoprotein bound to the outer membrane via an N-terminal lipid anchor [10,11]. Unlike in the meningococcus, we have been unable to detect fHbp on the surface of E. coli even when expressing the full length protein, demonstrating that there are differences in protein sorting in these two Gram negative bacteria. It is therefore possible that additional fHbp sequences in gfp fusions cause accumulation of GFP at aberrant cellular sites, impairing bacterial viability.
fHbp thermoregulation may have implications for the effect of vaccines that target fHbp. Levels of this vaccine antigen in the meningococcus in the upper airway (at 32–35°C) may be lower than in assays to assess immune responses within the laboratory, which are typically performed at 37°C [36,37]. Therefore, vaccines that include fHbp might not impose selective pressure on bacteria at the mucosal surface in the upper airway, and offer limited herd immunity [38]. This is consistent with a recent study demonstrating that immunisation with a vaccine containing fHbp has a limited impact on the acquisition of meningococcal carriage among university students [39].
Aside from factors involved in immune evasion, RNA thermometers may control other features in N. meningitidis and other bacteria that inhabit the nasopharynx. Previous studies have focused mainly on the effect of temperature on mRNA levels in the meningococcus [40]. The identification of further RNA thermosensors will require bio-informatic and proteomic approaches. Further understanding the mechanisms of thermoregulation could be informative about strategies of immune evasion employed by this important pathogen, its adhesive properties, and the acquisition of relevant nutrients at different sites in the upper airways [41].
N. meningitidis was grown in Brain Heart Infusion broth (BHI, Oxoid, 37 g dissolved in 1 L dH2O with 1 g starch) or on BHI agar (1.5% w/v) supplemented with 5% Levinthal’s base (500 ml defibrinated horse blood, autoclaved with 1 L BHI broth). Bacteria on solid media were incubated for 16–18 hours at 37˚C with 5% CO2. Liquid cultures (10 ml) were inoculated with 109 bacteria and grown at 37˚C with shaking (180 r.p.m.) to an optical density (O.D.) measured at 600 nm of ~0.5 unless otherwise stated.
E. coli was grown in Luria-Bertani (LB) broth (2% w/v in dH2O, Oxoid, UK) or on LB agar (1% w/v) plates. Liquid cultures of E. coli were grown in 4 ml of media inoculated from a single colony overnight at 37˚C with shaking (180 r.p.m.). Overnight grown bacteria were diluted 1 in 100 in media and grown to an Optical Density (OD) A600 of ~0.5. When necessary antibiotics were added to the following final concentrations: carbenicillin, 100 μg ml-1; kanamycin, 50 μg ml-1; rifampicin, 250 μg ml-1. The strains and plasmids used in this study are listed in Table 1.
N. meningitidis was grown in liquid culture to mid. log phase at 30°C, 37°C or 42°C, prior to fixation for two hours in 3% paraformaldehyde. Surface localisation of fHbp on N. meningitidis was detected using anti-fHbp V1.1 pAbs and goat anti-mouse IgG-Alexa Fluor 647 conjugate (Molecular Probes, LifeTechnologies). Samples were run on a FACSCalibur (BD Biosciences), and at least 104 events recorded before results were analysed by calculating the geometric mean fluorescence intensity in FlowJo vX software (Tree Star).
A N. meningitidis strain containing only the monocistronic fHbp transcript was generated by insertion of a kanamycin resistance cassette with a Rho-independent terminator upstream of PfHbp (S6 Fig). Upstream and downstream fragments, and the kanamycin resistance cassette were amplified by from gDNA using the primer pairs fHbp-KM-(1)-F/fHbp-KM-(1)-R, fHbp-KM-(2)-F/fHbp-KM-(2)-R and fHbp-KM-(3)-F/fHbp-KM-(3)-R, respectively. Details of primers used in this study are given in Table 2. PCR products were ligated into pGEM-T (Promega) following Gibson Assembly (NEB), then digested with NcoI and NotI (NEB). Transformation of N. meningitidis strain MC58 was performed as described previously [42].
N. meningitidis MC58 containing only the bicistronic fHbp transcript was generated by replacing the nmb1869 (the gene upstream of fHbp) with the kanamycin resistance gene only, leaving the nmb1869 promoter intact while PfHbp was disrupted by changing the -10 sequence (TACCATAA to TACCATCC). Upstream and downstream fragments, and the kanamycin resistance cassette together with the modified -10 region mutation were amplified using the primer pairs prom-10-mut1-F/prom-10-mut1-R, prom-10-mut3-F/prom-10-mut3-R, prom-10-mut4-F/prom-10-mut4-R, prom-10-mut5-F/prom-10-mut5-R with genomic DNA as the target, while primers prom-10-mut2-F/prom-10-mut2-R, were used to amplify the kanamycin resistance cassette. PCR products were ligated into pGEM-T (Promega) following Gibson Assembly (NEB), and then digested with NcoI and NotI before being used to transform N. meningitidis.
Site-directed mutagenesis was performed with the Quickchange kit (Stratagene) according to the manufacturer’s protocol. To modify fHbp9C-GFP (introducing mut1, mut2, mut3, mut4, mut-αRBS-1, mut-αRBS-2 and mut-αRBS-1&2), oligonucleotides pairs mut1-F/mut1-R, mut2-F/mut2-R, mut3-F/mut3-R, mut4-F/mut4-R, mut-αRBS-1-F/mut-αRBS-1-R, mut-αRBS-2-F/Mut-αRBS-2-R, mut-αRBS-1&2-F/mut-αRBS-1&2-R were used respectively (Table 2). The reaction products were transformed into E. coli DH5α, and constructs were confirmed by sequencing.
Full length fHbp was amplified from N. meningitidis MC58 using primers fHbp(c)-F and fHbp(c)-R and ligated into pGEMT to yield pfHbp. Deletions were generated by PCR and the products were then ligated into pGEM-T (Promega) and transformed into E. coli DH5α. The identity of all constructs was confirmed by nucleotide sequencing.
Plasmids containing GFP fusions were generated by amplifying the promoter of fHbp with different lengths of the ORF with fHbp(TTS)-U or fHbp-GFP-F with either fHbp-GFP1C-R, fHbp-GFP5C-R or fHbp-GFP9C-R, and ligating the products into pEGFP-N2 (Clontech).
N. meningitidis was grown in liquid culture to mid. log phase, and RNA was isolated using the RNAeasy Miniprep Kit (Qiagen, UK) following the manufacturer’s protocol. For E. coli, bacteria were grown in liquid media to mid log phase, and RNA isolated using the FastRNA Pro Blue kit (MP Biomedicals) according to the manufacturer’s protocol. The purity and integrity of RNA were determined by gel electrophoresis and spectrophotometry.
For Northern blotting, 20 μg of total RNA was separated on a formaldehyde agarose gel prior to blotting as described previously [43], then transferred to Hybond N. Membranes were hybridised with 32P-ATP ɣ-labelled DNA fragments Northern blots were developed using a Fuji phosphorImager scanner. Probes for detecting fHbp and tmRNA were amplified from gDNA with primer pairs fHbp-U/fHbp-D, and tmRNA-U/tmRNA-D, respectively.
To analyse mRNA stability, E. coli containing relevant plasmids was grown in liquid culture to an OD600nm of 0.5–0.6 then exposed to rifampicin (250 μg ml-1 final concentration) and incubated at 37°C with agitation. Bacteria were harvested for RNA isolation 0, 5, 10 and 20 minutes afterwards. Blots were probed with a labelled oligonucleotides GGTGCAGATGAACTTCAGGGTCAGCTTGCCGTAGGTGGCATCGCCCTCGC (to detect GFP mRNA), and a tmRNA product amplified from E. coli gDNA with primers Tm1 and Tm2 (Table 2). Band intensities of Northern blots were quantified using AIDA image analyzer software, standardised to the tmRNA loading control, and expressed as a ratio to the respective band intensities at t = 0 min.
Cell lysates were prepared by addition of SDS:PAGE loading buffer to an equal volume of bacteria obtained from liquid cultures by centrifugation. Total protein levels were measured using a Bradford assay and equal amounts of total protein loaded into each lane. Samples were boiled, then the proteins separated on polyacrylamide gels and transferred to immobolin P polyvinylidine fluoride (PVDF) membranes (Millipore, USA) using semi-dry transfer (Biorad, USA). For Western blot analysis, membranes were washed three times in 0.05% (w/v) dry milk/PBS with 0.05% (v/v) Tween-20 for 10 minutes, and then incubated with the primary antibody for one hour. Membranes were washed again three times and incubated for a further hour with a secondary, HRP-conjugated antibody. Binding was detected with an ECL Western Blotting Detection kit (Amersham, USA) and exposed to ECL Hyperfilm. An α-GFP mouse antibody (BD-living colors) was used at a final dilution of 1:8,000. α-fHbp polyclonal sera was used at a final dilution of 1:5000. α-RecA rabbit antibody (Abcam, UK) was used at a final dilution of 1:10,000. As secondary antibodies, goat α-rabbit or α -mouse IgG HRP-conjugated antibody (Dako, UK) was used at a final dilution of 1:10,000.
To determine protein stability, translation was prevented by adding spectinomycin (Sigma, final concentration 100 μg ml-1) to bacteria grown to an OD600 = 0.5 in liquid media. Samples were removed at times afterwards for Western blot analysis. Relative expression was calculated by measuring band intensities with ImageJ software, standardised to signals for loading controls (i.e. RecA or tmRNA), and shown as the ratio, relative to intensity of the control strain or condition.
For the template, DNA was isolated from derivatives of pEGFP-N2 containing one, five or nine codons of fHbp fused to GFP (GenElute Gel Extraction Kit, Sigma), linearised by digestion with NotI (New England Biolabs) and then purified using QIAquick (Qiagen). For RNA and DNA templates, 1 μg of nucleic acid was used as a template. The in vitro transcription/translation reaction was performed at 30°C, 37°C or 42°C for one hour using the E. coli S30 Extract system for Linear Templates in vitro Transcription/Translation Kit (Promega). The products were precipitated in acetone, then re-suspended in SDS-PAGE buffer prior to Western blot analysis.
RNA sequences were analysed using the RNAfold web server of the Vienna RNA package (http://rna.tbi.univie.ac.at/egi-bin/RNAfold.cgi). For each sequence, the minimum free energy in kcal mol−1 was predicted [44].
|
10.1371/journal.pgen.1007287 | Identification and characterization of wheat stem rust resistance gene Sr21 effective against the Ug99 race group at high temperature | Wheat stem rust, caused by Puccinia graminis f. sp. tritici (Pgt), is a devastating foliar disease. The Ug99 race group has combined virulence to most stem rust (Sr) resistance genes deployed in wheat and is a threat to global wheat production. Here we identified a coiled-coil, nucleotide-binding leucine-rich repeat protein (NLR) completely linked to the Ug99 resistance gene Sr21 from Triticum monococcum. Loss-of-function mutations and transgenic complementation confirmed that this gene is Sr21. Sr21 transcripts were significantly higher at high temperatures, and this was associated with significant upregulation of pathogenesis related (PR) genes and increased levels of resistance at those temperatures. Introgression of Sr21 into hexaploid wheat resulted in lower levels of resistance than in diploid wheat, but transgenic hexaploid wheat lines with high levels of Sr21 expression showed high levels of resistance. Sr21 can be a valuable component of transgenic cassettes or gene pyramids combining multiple resistance genes against Ug99.
| Wheat stem rust is a devastating disease that is threatening global wheat production. The emergence of new virulent races of this pathogen in Africa, including the Ug99 race group, has prompted global efforts to find effective resistance genes. We report here the identification of stem rust resistance gene Sr21 that is effective against the Ug99 race group. We developed a diagnostic marker to accelerate its deployment in wheat breeding programs and demonstrated that the introduction of two Sr21 copies in transgenic wheat results in high levels of resistance. An unusual characteristic of Sr21 is its increased resistance to stem rust at high temperatures. We show here that this is associated with the ability of Sr21 to coordinate the upregulation of multiple pathogenesis related genes at high temperatures. These genes slow down the growth of the pathogen and result in the characteristic Sr21 intermediate resistance reaction at high temperatures. A better understanding of this temperature dependent resistance mechanism will be useful for controlling the rust pathogens in our changing environments.
| More than 700 million tons of wheat are produced every year, but further increases are needed to feed a rapidly growing human population. Reducing yield losses caused by wheat pathogens can contribute to this goal. Wheat stem rust, caused by the fungal pathogen Puccinia graminis f. sp. tritici (henceforth Pgt), poses an urgent threat to global wheat production since new virulent races have overcome widely deployed resistance genes.
In 1998, a new Pgt race was detected in Uganda that was virulent to the widely deployed stem rust resistance genes Sr31 and Sr38 [1]. This race, also known as TTKSK using the North American system of nomenclature for Pgt races [2,3], was virulent to roughly 90% of the global wheat cultivars [4]. Since then, the Ug99 race group has spread to more than 13 countries in Africa and the Middle East [4–8] and acquired virulence to additional resistance genes (Sr24, Sr36, Sr9h and SrTmp [3,9–11]), prompting efforts to identify and clone effective resistance genes.
The large and complex nature of the wheat genome has limited the number of cloned Ug99 resistance genes to six, including Sr35 [12], Sr33 [13], Sr50 [14], Sr22 [15], Sr45 [15], and Sr13 [16]. More resistance genes are necessary to diversify the combinations of Pgt resistance genes deployed as gene pyramids or in transgenic cassettes to provide durable resistance [15,17,18].
The stem rust resistance gene Sr21 was discovered in diploid wheat Triticum monococcum (genome Am), which is closely related to T. urartu, the progenitor of the A genome in polyploid wheat [19]. Sr21, which confers a more effective resistance to the Ug99 race group at high temperatures (20–24°C) than at low temperature (16°C) [20], was transferred to hexaploid wheat (T. aestivum) [21]. We previously mapped Sr21 to a 0.19 cM interval on the central region of chromosome arm 2AmL and showed that this region includes a cluster of NBS-LRR resistance genes in the colinear region in the A genome of T. aestivum [20].
Here, we report the identification, validation and characterization of the wheat stem rust resistant gene Sr21, which encodes a coiled-coil nucleotide-binding leucine-rich repeat protein (NLR). Six pathogenesis-related (PR) genes showed increased transcript levels in Sr21-resistant genotypes inoculated with Pgt, only when plants were grown at high temperatures. This result provides a tentative explanation for the more effective Pgt resistance conferred by Sr21 at high temperatures. Finally, we identified five different resistance haplotypes of Sr21 and developed a diagnostic molecular marker to accelerate its deployment in breeding programs.
The 0.19 cM region on T. monococcum chromosome arm 2AmL including Sr21 is colinear with an 88-kb region in Brachypodium distachyon chromosome 5 flanked by genes Bradi5g22090 and Bradi5g22200 [20]. In the present study, we identified the wheat orthologs of six B. distachyon genes present in this colinear region (Bradi5g22100, Bradi5g22117, Bradi5g22146, Bradi5g22162, Bradi5g22179 and Bradi5g22187, Fig 1A, S1 Table) and mapped them in ten wheat lines with recombination events between Sr21 flanking markers FD527726 and EX594406 (Fig 1B). Using these new markers, Sr21 was mapped 0.02 cM distal to CJ961291 and 0.04 cM proximal to a group of four linked NLR pseudogenes (Cscnl20, Cscnl21, Cscnl22 and Cscnl23, Fig 1B).
We screened a bacterial artificial chromosome (BAC) library of the resistant T. monococcum accession DV92 [22] using the closest markers, but sequencing of selected BACs showed no connection between proximal and distal groups (S1 Fig). We initiated a chromosome walk from the closest marker (CJ961291, 0.02 cM from the phenotype) in the proximal group (Fig 1C, BAC 205J7). Marker CJ961291 (RRM2) showed a recombination event with gene CNL1, which was completely linked to the phenotype, completing the proximal side of the physical map. After five cycles of library screening, BAC sequencing and marker development, we identified an additional recombination event between the phenotype and marker 179C18F7R7, which closed the distal end of the physical map.
Sequencing the 405-kb candidate region (Fig 1D, blue area, S1 Fig red colored BACs, GenBank accession MG582649) revealed four complete NLR genes (henceforth, CNL1, CNL3, CNL5, and CNL6) and three NLR truncated genes with premature stop codons (henceforth, cnl2, cnl4, and cnl7).
Multiple rearrangements were observed in this NLR cluster in T. aestivum cv. Chinese Spring (419 kb, Fig 1E, IWGSC RefSeq v1.0) and T. turgidum ssp. dicoccoides Zavitan (435 kb, Fig 1F, Zavitan WEWSeq v1.0) [23]. The two polyploid species included the complete gene CsCNL8 (ZaCNL8) that was absent in T. monococcum, whereas the T. monococcum region including cnl2, CNL3, cnl4 and CNL5 was missing in Chinese Spring and Zavitan (Fig 1E and 1F). NLR pseudogenes Cscnl10, Cscnl11 and Cscnl12 identified in CS were not detected in T. monococcum or Zavitan. The CS and Zavitan pseudogenes Cscnl1a (Zacnl1a) and Cscnl1b (Zacnl1b) were similar to T. monococcum CNL1 (~99% identity). Cscnl1a was interrupted by the insertion of a repetitive sequence in CS but not in Zavitan (Fig 1E and 1F).
We sequenced the four NLR genes completely linked to the Sr21 phenotype (CNL1, CNL3, CNL5 and CNL6) from diploid accessions G3116 (Sr21, resistant) and PI 272557 (no-Sr21, susceptible). Using seven different pairs of gene-specific primers, we amplified CNL3 and CNL5 from DV92 but not from G3116 and PI 272557 (similar to CS and Zavitan, Fig 1E and 1F). Since G3116 carries the Sr21 resistant allele, the absence of these two genes suggested that CNL3 and CNL5 were unlikely candidate genes for Sr21. CNL6 was PCR-amplified in all three accessions, but G3116 showed a premature stop codon at position 742 (Q742*), suggesting that this gene was not Sr21. By contrast, CNL1 was PCR-amplified from both resistant accessions (DV92 and G3116) but not from Pgt susceptible accession PI 272557.
Taken together, these results suggested that CNL1 was the best candidate gene for Sr21 among the NLR genes linked to Sr21. This hypothesis was further supported by expression data for these genes in transcriptome databases of DV92 and G3116 [24]. Among the four NLR genes completely linked to Sr21, we only detected transcripts for CNL1 in both DV92 and G3116 databases (Fig 1D).
To test if CNL1 was required for Sr21 resistance, we mutagenized the Sr21 introgression lines in Chinese Spring (henceforth CSSr21) with ethyl methane sulfonate (EMS) and generated 1,151 M1 mutant plants. Among the 1,151 M2 mutant families screened with race BCCBC, we identified seven showing susceptible plants (M9, M66, M71, M271, M279, M287 and M306, Fig 2A and 2B). M3 seeds from the susceptible plants were also susceptible to Pgt race MCCFC (Fig 2C).
Sequencing of CNL1 from the seven susceptible mutants (primers in S1 Table) revealed amino acid changes in five mutants and premature stop codons in two mutants (Fig 2A). Since the probability of a truncation or missense mutation in an annotated gene in a hexaploid wheat line mutagenized with 0.8% EMS is roughly 0.02 [25], the probability of detecting such changes in seven independent mutants by chance is < 1.3×10−12. These results demonstrated that CNL1 is required for Sr21 resistance to Pgt.
To test if CNL1 was also sufficient to confer resistance to Pgt, we generated thirteen independent transgenic events (T0Sr21) in the susceptible common wheat variety Fielder. Among the 13 T0 plants, we prioritized five that showed higher transcript levels of CNL1 than Fielder, which has a non-functional copy of CNL1 (S2 Fig). The T1 progenies from these five events showed segregation for resistance when challenged with Pgt race TTKSK (isolate 04KEN156/04). Some of the resistant transgenic plants showed even better levels of resistance than the CSSr21 positive control, which carries a Sr21 introgression from T. monococcum (Fig 3A). These results confirmed that CNL1 is sufficient to confer resistance to TTKSK. Taken together, the high-density map, the mutants and the transgenic results confirmed that CNL1 is Sr21.
The observed segregation of resistant and susceptible T1 transgenic plants (Fig 3A) suggested that four of these five transgenic lines have more than one independent functional CNL1 insertions. Genotypes from roughly 50 T1 plants from each event showed significant departures from the expected 3:1 segregation ratio (P < 0.001, S2 Table), which suggested the presence of three to six copies of CNL1. This was validated by TaqMan copy number assays [26] (S2 Table). The larger copy number estimates obtained from the genotypic data than from the phenotypic data is likely explained by the presence of non-functional CNL1 insertions and/or by the presence of linked functional copies.
On average, the five selected T1 transgenic events showed 2.6- to 5.7-fold higher CNL1 transcript levels than CSSr21 (S2 Table and Fig 3B), confirming the T0 results (S2 Fig). Copy number based on the CNL1 TaqMan assay was significantly correlated with average transcript levels in the five selected transgenics and CSSr21 (R = 0.83, P = 0.039). A similar correlation (R = 0.81, P = 0.0009) was observed between the T0 expression levels from all 13 transgenic lines and the copy number estimated by the TaqMan assay (S2 Fig). These results suggest that the differences in expression are driven, at least in part, by differences in CNL1 copy number. This was also reflected in the levels of resistance to TTKSK, where a negative correlation was observed between CNL1 copy number and average rust pustule size (R = -0.92, P < 0.0001, S3 Fig). In this experiment, we also compared the CNL1 transcript levels between the Sr21-resistant accessions G3116 (diploid) and CSSr21 (hexaploid) and found no significant differences (Fig 3B).
We compared the sequences of the CNL1 transcripts (from the DV92 and G3116 transcriptome database [24]) with the corresponding genomic sequences and determined that the 4,872 bp coding sequence of CNL1 is divided in three exons that encode 1,624 amino acids. The predicted protein includes an N-terminal coiled-coil (CC) domain, a central nucleotide-binding (NB) site, and a leucine-rich repeat (LRR) region, typical of many NLR proteins. A nuclear localization signal (NLS) was predicted in Sr21 between amino acids 141 and 152 (cNLS Mapper http://nls-mapper.iab.keio.ac.jp/cgi-bin/NLS_Mapper_form.cgi).
Comparing different transcripts and genomic sequences, we identified a 210-bp 5’ untranslated region (UTR) and a 2,159-bp 3’ UTR. The 5’ UTR does not include any introns, whereas the 3’ UTR shows 2 to 4 introns depending on the alternative splice forms (S4 Fig). These results were confirmed by 5’ and 3’ rapid amplification of cDNA ends (5’ and 3’ RACE). We identified 10 CNL1 alternative splicing forms, which differ only in their 3’ UTR regions (S4 Fig). We detected five alternative splicing forms (CNL1-1 to CNL1-5) in the transcriptomes of DV92 and G3116 [24] and an additional five (CNL1-6 to CNL1-10) in the 3' RACE reactions.
Since Sr21 resistance is modulated by temperature [20], we characterized the frequency of the different alternative splice forms in 3' RACE reactions from RNAs obtained from G3116 T. monococcum resistant plants grown at 24°C and 16°C. The plants at each temperature were further divided in mock-inoculated and inoculated with Pgt race BCCBC, and RNA samples were extracted six days after inoculation (80 clones per treatment were sequenced). χ2 tests showed no significant differences (P = 0.81) in the frequencies of alternative splice forms in inoculated vs. mock-inoculated plants (averaged across temperatures) but detected significant differences between temperatures (P = 0.002, averaged across inoculation treatments, S3 Table). The CNL1-1, CNL1-5 and CNL1-6 forms were more frequent at 24°C than at 16°C, and the opposite was observed for CNL1-2. We currently do not know the biological significance of these differences.
We analyzed CNL1 transcript levels in the four temperature / inoculation combinations described above in resistant accessions G3116 (diploid) and CSSr21 (hexaploid). We took samples immediately after moving the plants inoculated with Pgt from a greenhouse (~20°C) to growth chambers at 16°C and 24°C (time 0 h, Fig 4). As expected, we found no significant differences in CNL1 transcript levels between temperatures or inoculations at the control sampling point (S4A Table). The CNL1 basal transcript levels relative to ACTIN were 14.8% higher in CSSr21 than in G3116 (P = 0.0089, S4A Table).
For the samples collected at 1, 3 and 6 days post inoculation (dpi), we performed a four-way ANOVA for CNL1 transcript levels including genotype, temperature, inoculation treatment and day after inoculation as factors. We found no significant differences between genotypes (diploid G3116 and hexaploid CSSr21), but detected significantly lower transcript levels in plants grown at 16°C than at 24°C (36.5%, P < 0.0001), and in Pgt inoculated plants relative to mock-inoculated plants (45.8%, P < 0.0001) (Fig 4, S4B Table). A smaller effect was detected among days (P = 0.02), but no clear trend was observed.
We used the same RNA samples collected 6 dpi from CSSr21 and G3116 described above and primers described in a previous study [16] to quantify the transcript levels of six pathogenesis-related (PR) genes (PR1, PR2, PR3, PR4, PR5 and PR9 = TaPERO). For all six PR genes, transcript levels were significantly higher (P < 0.0001) in Pgt-inoculated plants than in mock-inoculated plants (S5 Table and S5 Fig). The overall differences in transcript levels between temperatures were not significant for some PR genes (PR1, PR4, PR9) but the interactions between temperature and inoculation were all highly significant in the combined ANOVA (P < 0.001, S5 Table). These interactions are clear in S5 Fig, which shows lower PR transcript levels at 24°C than at 16°C in the mock-inoculated plants but significantly higher levels in the Pgt-inoculated plants at 24°C than at 16°C.
To confirm that Sr21 was required for the coordinated upregulation of the six PR genes at 24°C, we compared the transcript levels of these six genes in the resistant hexaploid line CSSr21 and its derived susceptible sr21-mutant M9 (Fig 5). Plants from both genotypes were inoculated with race BCCBC or were mock inoculated. For all six PR genes, the differences between genotypes, inoculations and the interactions genotype x inoculation were highly significant (P < 0.001) in two-way factorial ANOVAs. In all cases, a strong upregulation of all six PR genes was observed after inoculation with race BCCBC relative to mock inoculation only in the resistant genotype CSSr21 (Fig 5). This result confirmed that the coordinated upregulation of these six PR genes was triggered by Sr21.
Sr21 showed higher levels of resistance to BCCBC when plants were grown at 24°C than when grown at 16°C (Figs 6A, 6B and S6). Similar differences were observed before for Sr21 resistance response to TTKSK at 20°C and 16°C [20]. These observations were confirmed in a three-way ANOVA for average pustule size 14 days post inoculation (dpi) (S6 Table). This analysis showed highly significant (P < 0.0001) effects for ploidy level and genotype (presence or absence of Sr21), and a very strong interaction between temperature and genotype (P < 0.0001). Plants having the Sr21 resistance gene showed smaller sporulation areas at high temperatures, whereas those without Sr21 showed significantly larger sporulation areas at high temperatures (Fig 6A and 6B). These opposite effects masked the main effect of temperature for race BCCBC (S6 Table).
We also quantified the differences in BCCBC growth at five dpi by measuring the ratio of Pgt DNA relative to wheat DNA (Fig 6C and 6D) and average infection areas using microscopy and a fluorescent dye that stains the pathogen (Figs 6E, 6F and S7). Both methods showed highly significant (P < 0.0001) differences between genotypes and between temperatures. As with the sporulation area, we also detected highly significant interactions between genotype and temperature for both methods (P < 0.0001, S6 Table), which reflected the larger differences between genotypes at high than at low temperature (Fig 6C–6F). Average areas for sporulation at 14 dpi (TTKSK and BCCBC) and for pathogen growth (BCCBC determined by fluorescence) were larger in the Sr21-resistant hexaploid than in the resistant diploid plants, which agrees with a previous report [20].
Fluorescent images of Pgt growth in diploid and hexaploid wheat plants lacking Sr21 at high temperature show a diffuse network of hyphae at the borders of the infected areas, which seems to be expanding without host resistance. By contrast, in the presence of Sr21, the infected areas are smaller and their borders are denser (S7 Fig), which suggests that the expansion of the hyphae is facing opposition from the host. At the macroscopic level, a chlorotic halo was observed around the Pgt pustules in the Sr21 resistant reactions, which in some cases resulted in cell death. However, Pgt resistance conferred by Sr21 is mostly associated with a delay in the progression of the disease (partial resistance) rather than with a rapid hypersensitive reaction.
We evaluated 114 T. monococcum and T. urartu accessions with Pgt races MCCFC and BCCBC and confirmed the presence of Sr21 in 44 of them and its absence in 70. We sequenced the complete CNL1 gene from these accessions using three pairs of primers (S1 Table and Fig 2A) and identified four susceptible haplotypes (S1 to S4) and five resistant haplotypes (R1 to R5, S8 Table and S8 Fig). The presence of Sr21 in the resistant diploid wheat accessions is based on inoculation with Pgt races TRTTF, TTKSK, TTTTF, QFCSC, and MCCFC (S7 Table).
Of the 44 resistant accessions, 28 were classified as haplotype R1 (MG582649, equal to DV92), six as R2 (MG601519), one as R3 (MG601520), six as R4 (MG601521) and three as R5 (MG601522, S8 Table). The R1 haplotype differed from the other four by one (R2), three (R3), two (R4) and three (R5) amino acid changes, respectively (S8 Fig).
Of the 30 T. monococcum susceptible accessions, 25 showed identical sequences and were classified as haplotype S1 (MG601523, 90.2% identical at the cDNA level to CNL1 haplotype R1). S1 is likely a non-functional gene since it carries a frame shift mutation. Surprisingly, polymorphisms between R1 and S1 were concentrated in the CC-NBS domains (87.2% identity) whereas the LRR was more similar to CNL1 (99.2% identity), which suggests a recombination or conversion event. A phylogenetic analysis including the DNA coding regions of the different NLR genes in the CNL1 region, and the closest paralogues in B. distachyon and H. vulgare showed that the CNL1 S1 haplotype is more closely related to cnl7 than to the cluster including the different CNL1 haplotypes (S9 Fig).
All 40 T. urartu accessions showed haplotype S2 (MG601524), which was similar to R1 (99.2% identical). This haplotype also includes three accessions classified as T. monococcum subsp. aegilopoides in the NSGC. However, analysis of 12 additional genes showed T. urartu haplotypes in these three accessions, suggesting that they are misclassified (we counted them as T. monococcum in the numbers reported above). The S2 haplotype from T. urartu, and the related haplotypes found in wild tetraploid Zavitan and hexaploid cultivars Chinese Spring and Fielder corresponds to the pseudogene Cscnl1b from Fig 1 and shares a 2-bp deletion at positions 3,255 and 3,256 (cDNA coordinates). When this shift mutation is manually corrected, these accessions share 20 amino acid polymorphisms compared to the R1 haplotype (S8 Fig). The cultivated tetraploid Kronos shares 14 of the 20 amino acid polymorphisms found in Chinese Spring, Fielder and Zavitan. Kronos does not have the 2-bp frame shift mutation, but has two different 1-bp deletions at positions 928 (a missing A) and 3,667 (a missing T). The susceptible haplotypes S3 (MG601525) and S4 (MG601526) were very similar to R1 (> 99.9% identical) and included a single accession each (S8 Table). Three accessions from the Balkans, classified as S1 based on their CNL1 sequence (PI 355538, PI 362610 and PI 377668), showed a resistance reaction when inoculated with race MCCFC but were susceptible to BCCBC. Since Sr21 is resistant to both BCCBC and MCCFC, this result suggested that these three accessions carry a Pgt resistance gene different from Sr21.
To develop a diagnostic marker for the Sr21 resistant haplotypes introgressed into polyploid wheat, we designed primers based on the C1228W diagnostic polymorphism and the T1500R and H1501S polymorphisms that separate Sr21 resistant and susceptible haplotypes. PCR amplification with primers Sr21TRYF5R5 (S1 Table) at an annealing temperature of 56°C generates a 951-bp fragment when the T. monococcum haplotypes are present and no amplification in T. urartu, tetraploid or hexaploid wheat (S10 Fig). Treatment of the amplified PCR products with restriction enzyme NsiI generated two bands of 836 and 115 bp for the T. monococcum accessions carrying the Sr21 susceptible haplotypes and a single 951-bp band for the accessions carrying the resistant haplotypes (S10 Fig).
The T. monococcum genomic region encompassing the Sr21 resistance gene includes four NLR genes and three pseudogenes (Fig 1). The clustering of NLR genes facilitates the generation of novel variants through recombination and conversion events, increasing genetic variability in resistance genes [27]. Ectopic recombination events can generate deletions and duplications, and these were frequent in the Sr21 region. We detected several rearrangements between T. monococcum, hexaploid wheat Chinese Spring and wild tetraploid wheat Zavitan [23] (S1 Fig) that diverged less than one million years ago [28]. A T. monococcum region of more than 150-kb including cnl2, CNL3, cnl4 and CNL5 was not detected in any of the polyploid species. Similarly, a large region in Chinese Spring between CsCNL8 and Cscnl12 was deleted in T. monococcum and part of it in Zavitan. Finally, a region including CNL6 and cnl7 was detected in diploid and tetraploid wheat but was absent in hexaploid wheat. These large differences suggest that this NLR cluster has experienced rapid evolutionary changes. The presence of four related NLR genes and pseudogenes in the colinear region of Brachypodium distachyon chromosome 5 (Figs 1 and S9) suggests that this NLR cluster has a long evolutionary history that extends beyond the divergence between the Triticum and Brachypodium lineages more than 30 million years ago [29].
When we characterized Sr21 expression, we detected multiple alternative splicing forms of CNL1, which differed in the intron structure of the 3’ UTR region (S4 Fig). Alternative splicing forms have been identified in several CC-NBS-LRR genes, including Pi-ta in rice [30], Mla in barley [31], and Lr10 [32] in wheat. Complex UTR regions with multiple introns have been also described for other wheat NLR genes involved in resistance to Pgt including Sr35 [12] and Sr13 [16]. In Sr21, we observed differences in the frequencies of the main alternative splicing forms with temperature, but the role of these differences is currently unknown.
Results presented here and in a previous study [20] indicate that Sr21 is less effective when present in a hexaploid background than in a diploid background. Since the hexaploid Sr21 gene was introgressed directly from T. monococcum [21], the encoded proteins are expected to be identical. In addition, we found no significant differences between diploid and hexaploid wheat in Sr21 transcript levels. Therefore, differences downstream of Sr21 transcription are likely responsible for the reduced resistance conferred by Sr21 in hexaploid than in diploid wheat. Reduced stability of the Sr21 protein in hexaploid wheat or reduced compatibility between the T. monococcum Sr21 protein and some of its downstream T. aestivum protein interactors are possible explanations. Both hypotheses can explain the stronger upregulation of downstream PR genes observed in diploid than in hexaploid Sr21-resistant wheat accessions (1.5 fold for PR2 to 14.6 fold for PR9, S5 Table). In both species, the coordinated upregulation of PR genes was observed only in the presence of the pathogen and the resistance gene, suggesting that the Sr21 protein needs to be activated through interactions with a Pgt effector or a wheat protein modified by Pgt.
The stronger resistance response observed at 24°C suggests that temperature modulates some of the involved processes. Elevated growth temperatures have been reported to affect plant resistance to diseases by reducing steady-state levels of resistance protein at high temperatures [33], affecting temperature-sensing NB-LRR proteins [34], or affecting salicylic acid (SA) regulation [35,36]. A similar inhibition of the resistance response by high temperatures has been reported for wheat Pgt resistance genes Sr6, Sr10, Sr15, and Sr17. By contrast, Sr13- [16] and Sr21-mediated Pgt resistances are more effective at higher temperatures. Both genes also show a coordinated upregulation of the same PR genes at high temperatures, which suggests that they may share some common mechanisms.
Although the mechanisms by which Sr21 and Sr13 [16] coordinate the upregulation of PR genes are currently unknown, information from other species suggests that the NPR1 pathway or WRKY transcription factors may be involved. Previous studies in Arabidopsis have shown that NPR1 interactions with TGA transcription factors play an important role in the regulation of several PR genes [37,38]. This was also observed in barley, in which overexpression of a conserved protein from the stripe rust pathogen that competes with TGA transcription factors for the binding with NPR1, reduced the induction of several PR genes in a leaf region adjacent to a bacterial infection [39]. WRKY transcription factors are also interesting candidates because the promoter of several PR genes contain W-box elements recognized by these proteins [40,41]. In addition, NLR proteins from barley (MLA) and rice (Pb1) have been shown to interact with WRKY transcription factors to regulate their defense responses [42,43]. It would be interesting to determine if Sr21 or Sr13 can interact with NPR1, WRKY or other transcription factors to coordinate the upregulation of wheat PR genes.
The results presented here provide useful information for the utilization of Sr21 in agriculture. Since Sr21 is susceptible to some Pgt races, it needs to be deployed in combination with other Pgt resistance genes or in transgenic cassettes including multiple resistance genes. Given the better Ug99 resistance levels observed in transgenic plants carrying more than one active copy of Sr21, it might be valuable to include at least two copies of Sr21 in the transgenic cassettes. It might be also advisable to avoid combining Sr21 and Sr13, since both genes seem to operate by a similar mechanism involving the activation of multiple PR genes at high temperature. Even if Sr21 and Sr13 recognize different effectors, the pathogen could bypass both resistance genes simultaneously by attacking a single target if the two genes share a common downstream signaling pathway. Combining genes that operate by different mechanisms may reduce the probability that a single change in the pathogen can defeat multiple pyramided genes [44].
Sr21 confers only partial resistance to Ug99, but this might be useful in programs that aim to combine multiple partial resistance genes and avoid major all-stage resistance genes, a strategy that has been proposed to increase the durability of wheat resistance to rusts [45]. The deployment of Sr21 in commercial tetraploid or hexaploid commercial varieties can be accelerated by the diagnostic marker developed in this study (S10 Fig).
A total of 7,168 recombinant gametes from two segregating populations were used to construct a high-resolution genetic map of Sr21. These populations included 734 F2 plants from population PI 272557 × DV92 and 2,850 from population PI 272557 × G3116. Plants with informative recombination events were challenged with races MCCFC (isolate 59KS19) and TTKSK (isolate 04KEN156/04) at the USDA-ARS Cereal Disease Laboratory and with race BCCBC (isolate 09CA115-2) at the University of California, Davis (UCD). Assays of response to race TTKSK were performed at 25°C during the day and 22°C during the night with a 16 h photoperiod (Fig 3) and those for BCCBC at 16°C (low temperature) and 24°C (high temperature). Procedures for inoculation and statistical analyses of infection types were reported previously [20].
A Bacterial Artificial Chromosome (BAC) library from the resistant parent DV92 [22] was used to generate the physical map by chromosome walking. DNAs from the selected BACs were extracted using QIAGEN Large-Construct Kit. BACs were fingerprinted using restriction enzyme HindIII, were sequenced using a combination of Illumina Hi Seq2500 at the Beijing Genomic Institute (Sacramento, CA, USA) and WideSeq at Purdue Genomics Core Facility (https://www.purdue.edu/hla/sites/genomics/wideseq-2/), and were assembled using Galaxy [46,47]. We identified and annotated the repetitive elements in the Sr21 region using the Triticeae Repeat Sequence Database (http://wheat.pw.usda.gov/ITMI/Repeats/blastrepeats3.html) and the genes using BLASTN / BLASTX searches in GenBank (http://www.ncbi.nlm.nih.gov/). These three websites were last accessed February 26, 2018.
Mutant lines were generated by treating 10,000 seeds from the hexaploid wheat line CSSr21 (Sr21 resistance haplotype R1) with 0.8% ethyl methane sulphonate (EMS). Seeds from 1,151 independent M1 mutants were harvested, and 25 M2 seeds per family were planted and inoculated with Pgt race BCCBC (isolate 09CA115-2) at UCD. Twenty-five M3 seeds from susceptible M2 plants were retested with race MCCFC (isolate 59KS19) at the USDA-ARS Cereal Disease Laboratory.
Rapid amplification of cDNA ends (RACE) was performed using total RNA extracted from leaves of resistant parent DV92. Both 5' RACE and 3' RACE were performed using the FirstChoice RLM-RACE Kit (Invitrogen) following the manufacturer’s instruction. The PCR products from Nested PCR amplifications were cloned using the TA cloning kit (Invitrogen).
A 10,463-bp genomic DNA fragment including Sr21 was amplified from DV92 BAC clone 205J7 using Phusion High-Fidelity DNA Polymerase (New England BioLabs Inc.). This fragment, including the complete Sr21 (5,208 bp) coding region and introns, 2,772 bp upstream from the start codon, and 2,483 bp downstream from the stop codon, was cloned into the binary vector pLC41Hm. The resulting plasmid pLC41HmSr21 was transformed into Agrobacterium strain EHA105, and was transformed into the Ug99-susceptible wheat variety Fielder (CNL1 haplotype S2) at the UC Davis transformation facility (http://ucdptf.ucdavis.edu/). Primers HptmikiF/R developed from the hygromycin resistance gene and primers S21CNL1F5R5 and Sr21TRYF5R5 developed from CNL1 were used to confirm the presence of transgene (S1 Table). A TaqMan Copy Number Assay was used to estimate the number of copies inserted in every transgenic event as described before [16].
Sequences from the 3' RACE described above and from the transcriptome databases of DV92 and G3116 [24] revealed the presence of several alternative splicing forms at the 3’UTR region of Sr21. To determine the number of alternative splicing variants, we first extracted total RNAs from G3116 T. monococcum plants grown at 16°C and 24°C six days after inoculation. We used these RNAs for 3’RACE reactions and cloned the PCR products using the TA cloning kit (Invitrogen). Eighty colonies from every 3' RACE reaction were PCR-amplified and sequenced using the Sanger method.
Total RNA was extracted from leaves of Sr21-resistant diploid T. monococcum ssp. aegilopoides accession G3116 and hexaploid CSSr21 using Spectrum Plant Total RNA Kit (Sigma-Aldrich). First strand cDNA was synthesized from 1 µg of total RNA using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Samples were collected from plants grown under the four treatments resulting from the combination of two temperatures (16°C or 24°C) and two inoculation treatments (Pgt race BCCBC or mock inoculation). Samples were collected immediately after inoculation (0 h) and 1, 3 and 6 days post inoculation (dpi). We quantified Sr21 expression using primers Sr21qRTF1R1 (S1 Table). The same cDNA samples were used to quantify the transcript levels of six pathogenesis-related (PR) genes (PR1, PR2, PR3, PR4, PR5 and PR9 = TaPERO) using primers described before [16].
We performed the qRT-PCR reactions on an ABI 7500 Fast Real-Time PCR System (Applied Biosystems) using Fast SYBR GREEN Master Mix. Transcript levels were expressed as fold-ACTIN levels (the number of molecules in the target / the number of ACTIN molecules) using the 2ΔCT method as described before [48]. We calculated the significance of the differences in expression levels using factorial ANOVAs and the SAS program version 9.4.
From a previous study [49], we selected 44 T. monococcum accessions (28 cultivated and 16 wild type) resistant to races MCCFC and TTKSK and susceptible to races TTTTF, TRTTF and QFCSC, which were previously postulated to carry only Sr21 (S7 Table). We also selected 30 accessions of T. monococcum and 40 from T. urartu that were susceptible to all five races (S7 Table). Seeds were obtained from the U.S. Department of Agriculture National Small Grains Collection (NSGC).
All T. monococcum and T. urartu accessions were re-evaluated with races MCCFC (isolate 59KS19) and BCCBC (isolate 09CA115-2) (at 24°C) for this study. We also confirmed the absence of Sr35 with the diagnostic marker developed from the cloned Sr35 gene [12]. Seeds from three T. monococcum accessions showed heterogeneous resistance reactions to MCCFC (PI 427971, PI 119422 and PI 225164) and were listed twice in S8 Table (as–S and–R) resulting in a total of 74 T. monococcum accessions. The sequenced genomes of diploid wheat T. urartu (http://plants.ensembl.org/Triticum_urartu/Info/Index), tetraploid wheat Zavitan (https://wheat.pw.usda.gov/GG3/wildemmer), and hexaploid wheat Chinese Spring RefSeq v1 (https://urgi.versailles.inra.fr/blast_iwgsc/blast.php) were used to detect the corresponding CNL1 alleles.
|
10.1371/journal.ppat.1004075 | Cytosolic Peroxidases Protect the Lysosome of Bloodstream African Trypanosomes from Iron-Mediated Membrane Damage | African trypanosomes express three virtually identical non-selenium glutathione peroxidase (Px)-type enzymes which preferably detoxify lipid-derived hydroperoxides. As shown previously, bloodstream Trypanosoma brucei lacking the mitochondrial Px III display only a weak and transient proliferation defect whereas parasites that lack the cytosolic Px I and Px II undergo extremely fast lipid peroxidation and cell lysis. The phenotype can completely be rescued by supplementing the medium with the α-tocopherol derivative Trolox. The mechanism underlying the rapid cell death remained however elusive. Here we show that the lysosome is the origin of the cellular injury. Feeding the px I–II knockout parasites with Alexa Fluor-conjugated dextran or LysoTracker in the presence of Trolox yielded a discrete lysosomal staining. Yet upon withdrawal of the antioxidant, the signal became progressively spread over the whole cell body and was completely lost, respectively. T. brucei acquire iron by endocytosis of host transferrin. Supplementing the medium with iron or transferrin induced, whereas the iron chelator deferoxamine and apo-transferrin attenuated lysis of the px I–II knockout cells. Immunofluorescence microscopy with MitoTracker and antibodies against the lysosomal marker protein p67 revealed that disintegration of the lysosome precedes mitochondrial damage. In vivo experiments confirmed the negligible role of the mitochondrial peroxidase: Mice infected with px III knockout cells displayed only a slightly delayed disease development compared to wild-type parasites. Our data demonstrate that in bloodstream African trypanosomes, the lysosome, not the mitochondrion, is the primary site of oxidative damage and cytosolic trypanothione/tryparedoxin-dependent peroxidases protect the lysosome from iron-induced membrane peroxidation. This process appears to be closely linked to the high endocytic rate and distinct iron acquisition mechanisms of the infective stage of T. brucei. The respective knockout of the cytosolic px I–II in the procyclic insect form resulted in cells that were fully viable in Trolox-free medium.
| In many cell types, mitochondria are the main source of intracellular reactive oxygen species but iron-induced oxidative lysosomal damage has been described as well. African trypanosomes are the causative agents of human sleeping sickness and the cattle disease Nagana. The parasites are obligate extracellular pathogens that multiply in the bloodstream and body fluids of their mammalian hosts and as procyclic forms in their insect vector, the tsetse fly. Bloodstream Trypanosoma brucei in which the genes for cytosolic lipid hydroperoxide-detoxifying peroxidases have been knocked out undergo an extremely rapid membrane peroxidation and lyse within less than two hours when they are cultured without an exogenous antioxidant. Here we show that the primary site of intracellular damage is the single terminal lysosome of the parasites. Disintegration of the lysosome clearly precedes damage of the mitochondrion and parasite death. Iron, acquired by the endocytosis of iron-loaded host transferrin, induces cell lysis. Contrary to the cytosolic enzymes, the respective mitochondrial peroxidase is dispensable for both in vitro proliferation and mouse infectivity. This is the first report demonstrating that cytosolic thiol peroxidases are responsible for protecting the lysosome of a cell.
| In many tissues, the mitochondrial electron transport chain constitutes the primary source of endogenously produced superoxide anion, the precursor molecule of most reactive oxygen species [1], [2]. Hydrogen peroxide and lipid hydroperoxides formed as products are primarily removed by glutathione peroxidases (GPxs) [3]. Among the eight GPxs described in mammals, GPx4 is the only one that accepts phospholipid hydroperoxides as substrates even within intact biomembranes [4]. Another organelle that plays a critical role in oxidant-induced cell damage is the lysosome [5]. Intralysosomal iron, which probably represents the major fraction of cellular redox-active iron, can catalyze the peroxidation of membrane lipids. Once lysosomal rupture has occurred, the cell is irreversibly committed to death [6].
African trypanosomes, the causative agents of human sleeping sickness and Nagana cattle disease, are extracellular parasitic protozoa with a digenetic life cycle. Trypanosoma brucei multiply as infective bloodstream (BS) forms in the blood and body fluids of their mammalian hosts and as procyclic insect form in the midgut of the tsetse fly vector. Trypanosomes possess mitochondria and lysosomes as single copy organelles. The mitochondrion of the BS parasites is functionally repressed and the cells rely exclusively on glycolysis for ATP production [7]. Nevertheless, the organelle plays a crucial role by harbouring the alternative oxidase, the final acceptor of reducing equivalents generated during glycolysis, as well as the machinery for iron sulfur cluster biogenesis [8], [9]. BS T. brucei have one of the highest endocytic rates ever measured [10]. All vesicular trafficking of macromolecules into or out of the parasites takes place at the flagellar pocket which is the only area of the cell surface with endocytic activity. Both fluid-phase and receptor-mediated cargo enter the parasite via an early endosomal compartment, pass on to a recycling endosome, and are ultimately delivered to the lysosome [10], [11]. The lysosome is the final repository of cargo taken up from the host serum for nutritional and immune evasion purposes. Trypanosomes are heme auxotroph. BS T. brucei obtain the cofactor by receptor-mediated endocytosis of the host haptoglobin/hemoglobin (Hb) complex and release of the heme in the lysosome [12], [13]. Also iron is acquired by endocytosis of a plasma protein. After internalisation of the host holo-transferrin [14], [15], iron is set free in the lysosome and apo-transferrin is proteolytically degraded [16]. Iron is required for DNA synthesis, antioxidant defence, mitochondrial FeS cluster biosynthesis, and, depending on the developmental stage, for mitochondrial respiration and alternative oxidase [17].
T. brucei lack catalase and selenocysteine-containing glutathione peroxidases. Hydroperoxide detoxification is achieved by 2-Cys-peroxiredoxins and non-selenium glutathione peroxidase (Px)-type enzymes which both obtain their reducing equivalents from the parasite-specific trypanothione/tryparedoxin system (for a recent review see [18]). Whereas the 2-Cys-peroxiredoxins use hydrogen peroxide as main substrate, the Px-type enzymes preferably detoxify lipid-derived hydroperoxides [19]. RNA interference studies targeting the cytosolic 2-Cys-peroxiredoxin or the three isoforms of the Px-type enzymes revealed that both types of peroxidases are essential [20], [21]. Proliferation of the Px-depleted BS parasites, however, can be fully restored by supplementing the culture medium with the vitamin E analogue Trolox as it is the case for GPx4-deficient mammalian cells [22]. This allowed us to clone cell lines that lack the individual genes [19]. The selective knockout of the mitochondrial px III gene results in parasites that display only a minor and transient growth retardation in vitro when compared to wild-type (WT) cells. In contrast, parasites lacking the cytosolic peroxidases (px I–II−/− cells) die within less than two hours after transfer into Trolox-free medium [19]. The relevance of Px III for parasite survival in vivo as well as the endogenous source of oxidants and the cell death mechanism of the px I–II−/− trypanosomes remained however elusive.
We here show that the mitochondrial Px III is fully dispensable for parasitism and identify a master role for the cytosolic trypanothione/tryparedoxin-dependent peroxidases in protecting the terminal lysosome. Our data revealed that, in the absence of Trolox, BS px I–II−/− T. brucei undergo lysosomal disintegration, followed by damage of the mitochondrion - and likely other cellular membranes - and total cell lysis. The process is closely linked to the endocytic uptake of iron ions. The discovery of this unprecedented biological role of cytosolic thiol peroxidases is expected to have strong impact on respective studies in other human pathogens as well as the mammalian host.
As a first step to elucidate the cellular processes responsible for the lethal phenotype of the px I–II−/− cells in the absence of Trolox, we followed the viability of the cells at different temperatures. After 90 min incubation at 37°C, the mutant parasites were completely lysed, whereas 60% and >90% of the cells were still viable when the parasites were kept at 21°C and 9°C, respectively (Figure 1). In contrast to the mammalian BS form, procyclic px I–II−/− cell lines were fully viable in the absence of Trolox (Figure S1). This may at least partially be due to the lower endocytosis rate in the insect form which is down-regulated approximately 10-fold compared to that of the BS form [23]. Because of the extremely rapid cell lysis observed at the normal culture temperature of 37°C, the uptake studies described in the following sections were conducted at room temperature (RT) to allow acquisition of reliable data. Endocytosis by BS T. brucei is highly sensitive to temperature [24]. To verify that under our conditions, cargo is still delivered to the lysosome, we followed the uptake of Alexa Fluor 488-conjugated dextran by living WT and px I–II−/− cells (the latter ones in the presence of Trolox) at different temperatures. Whereas after 10 min incubation at 37°C, all parasites displayed a discrete lysosomal staining, practically none of the cells was stained at 19°C and 9°C. After 2 h at 19°C, but not at 9°C, labeling of the cells was identical to that at 37°C in accordance with the dextran reaching the lysosomal compartment albeit at a much slower rate compared to 37°C (Figure S2).
Px I–II−/− and WT cells were treated with Alexa Fluor 488-conjugated dextran and subjected to fluorescence microscopy. In the mutants kept with Trolox as in WT cells, the fluid-phase marker was discretely located in the post-nuclear region of the parasite consistent with lysosomal delivery (1 in Figure 2A). As expected, in the absence of Trolox, the px I–II−/− cells progressively lysed. Of the remaining fluorescent parasites, 40 to 60% displayed a signal that was spread over the whole cell body (3 in Figure 2A). The enlarged but still confined fluorescence observed in 5–10% of the cells (2 in Figure 2A) suggests that swelling of the organelle can occur as an intermediate step. This has been observed in parasites that were treated with protease inhibitors or human serum or upon ablation of p67, a lysosomal transmembrane glycoprotein [11], [25], [26]. Trypanolysis caused by apoL-I, crucial component of the trypanolytic factor present in human serum, involves the formation of anion-selective pores in the lysosomal membrane of the parasite, a process which is abolished by addition of 1 mM 4,4-diisothiocyanatostilbene-2,2-disulfonic acid (DIDS) to the culture medium [26], [27]. In the case of the px I–II−/− cells, 0.1 mM or 0.5 mM DIDS had no protective effect and 1 mM DIDS even proved to be lethal for both WT and mutant parasites (Figure S3). Thus, the loss of lysosomal integrity in the px I–II−/− cells does not appear to involve membrane pore formation.
To further dissect the role of the lysosome, the px I–II−/− cells were treated with LysoTracker, a fluorescent acidotropic reagent that traces acidic organelles in living cells and has previously been used to stain the parasite lysosome [28]. Cells that were kept at 37°C for up to 30 min without Trolox or for 60 min in the presence of the antioxidant showed a discrete lysosomal staining. In contrast, after 45 and 60 min in Trolox-free medium, many and virtually all, respectively, parasites had lost the fluorescent signal (Figure 2B). This finding confirmed that disintegration of the lysosomal compartment preceded cell lysis.
Antibodies against p67 typically stain a prominent vesicular compartment between the nuclear and mitochondrial (kinetoplast) DNA, although sometimes also multiple discrete vesicles in the same region are visualized [11], [29]. Px I–II−/− cells that were harvested up to 15 min after transfer into Trolox-free medium or kept in the presence of the antioxidant displayed a lysosomal p67 staining (Figure 2C). In contrast, when kept in Trolox-free medium, an increasing percentage of parasites lacked an intense and well defined signal. Many cells showed a dispersed staining which was reminiscent of parasites depleted of the Rab4 protein, a regulator of lysosomal trafficking [30]. These signals however were hardly to distinguish from the background fluorescence of cells treated only with the secondary antibody. Therefore, the quantitative analysis was based on cells with a vesicular p67 staining between the two DAPI signals. Taken together, the three different approaches strongly suggested that damage of the lysosomal membrane is a primary event caused by the absence of the cytosolic peroxidases.
Supplementation of the medium with 100 µM iron chloride had no effect on the viability of the px I–II−/− cells provided the presence of Trolox. In the absence of the antioxidant, iron strongly accelerated cell lysis (Figure 3A). To further evaluate the role of iron, the parasites were treated with deferoxamine, a potent iron chelator that in mammalian cells is preferentially taken up by fluid-phase endocytosis [31]. Deferoxamine primarily prevents iron incorporation into newly synthesized proteins [32]. In Trolox-free medium, indeed the chelator slowed down cell lysis and thus protected the px I–II−/− cells (Figure 3B). In the presence of Trolox, deferoxamine had no effect on the short-term viability of the mutant cells but inhibited the long-term cell proliferation, as it has previously been shown for WT parasites [32], [33]. Neither iron nor deferoxamine affected the short term viability of WT cells independent of the presence or absence of Trolox (not shown).
BS T. brucei are cultured in HMI-9 medium which contains 10% of FCS. In accordance with our previous observations, px I–II−/− cells kept in this standard medium rapidly died, and after 2 h, the culture displayed only 50% of the starting cell density (Figure 3C). In medium lacking FCS, however, the mutant parasites remained viable and, remarkably, were insensitive towards exogenous iron, even in Trolox-free medium. Overnight cultivation in medium without FCS resulted in complete cell death due to the lack of essential nutrients and growth factors. In conclusion, the iron-induced cell lysis clearly required the presence of (a component of) FCS. BS T. brucei acquire heme by receptor mediated endocytosis of the haptoglobin-Hb complex [12]. In the absence of Trolox, supplementing the medium with 1 mg/ml of Hb resulted in a minor, but detectable, acceleration of cell lysis (Figure S4). This only moderate effect may at least partially be due to the fact that in the serum, haptoglobin is already essentially saturated with Hb.
FCS contains about 25 µM transferrin and the iron saturation of transferrin ranges from 55 to 92% [34]. To mimic these conditions, the medium was supplemented with 25 µM holo-transferrin. In the absence of Trolox, this treatment stimulated lysis of the px I–II−/− cells (Figure 4A). In contrast, apo-transferrin slowed down cell lysis in the absence of Trolox (Figure 4B). Parasites incubated overnight with apo-transferrin in medium containing Trolox remained viable but did not proliferate. Competition between holo- and apo-transferrin for the parasite transferrin receptor results in reduced iron uptake [14]. This should directly affect the synthesis of DNA precursors by the iron-dependent ribonucleotide reductase. Incubation of the mutant parasites with holo-transferrin in medium lacking both FCS and Trolox induced the cell lysis (Figure 4C). However, a 10-fold higher concentration of holo-transferrin (25 µM) was required for an effect comparable to that observed in the presence of 10% FCS. This suggests that in the absence of FCS, the overall metabolism of the parasite is affected and/or another serum component contributes to the lethal phenotype. To get a deeper insight in the mechanism, we prepared transferrin-depleted medium. Purified antibodies against bovine transferrin were covalently linked to sepharose and HMI-9 medium was chromatographed on this matrix. Western blot analysis confirmed the successful removal of transferrin from the medium (Figure S5). As expected, in the presence of Trolox, the px I–II−/− cells did not show any lysis in the transferrin-depleted medium with or without supplementation by 25 µM transferrin (Figure 4D). However, in the absence of Trolox, the px I–II−/− cells displayed lysis. This may be due to the uptake of Hb, which occurs via endocytosis of the haptoglobin/Hb complex, again resulting in lysosomal iron [26]. In addition, we cannot rule out that the medium still contained residual transferrin that was not detected in the Western blot. Even when assuming 99% depletion, the remaining transferrin may still be efficiently internalized due to the high affinity of the parasite transferrin receptor [14]. The contribution of transferrin to cell lysis could be clearly demonstrated: Supplementing the transferrin-depleted medium with transferrin accelerated cell lysis. Upon overnight cultivation in Trolox-supplemented transferrin-depleted medium, the mutant parasites proliferated, again indicating that the medium contained an iron source. This is in accordance with previous work showing that WT T. brucei can grow in transferrin-depleted medium [35]. Taken together, the data suggest that transferrin as well as Hb contribute to the lethal phenotype of the px I–II−/− cells in the absence of the exogenous antioxidant, both resulting in the generation of free lysosomal iron.
The px I–II−/− cells were transferred from Trolox-supplemented into medium ± Trolox and after different time points subjected to immunofluorescence microscopy with MitoTracker, p67 antibodies, and DAPI (Figure 5A). The parasites were subdivided into four groups that displayed a) both MitoTracker and discrete p67 staining; b) MitoTracker staining, but no discrete p67 signal; c) no MitoTracker staining, but discrete p67 signal; and d) neither MitoTracker nor p67 staining (Figure 5B). Px I–II−/− cells kept for up to 15 min in Trolox-free medium and those in the presence of the antioxidant displayed perfect mitochondrial and lysosomal signals. In contrast, prolonged incubation of the px I–II−/− parasites in Trolox-free medium resulted in the progressive formation of cells that lacked both signals. About 12% of the cells showed MitoTracker staining but no discrete p67 signal. The reverse phenotype, namely parasites that lacked the MitoTracker staining but had a discrete p67 signal, was practically not observed. Thus, lysosomal disintegration precedes the damage of the mitochondrion.
BS T. brucei in which the gene encoding the mitochondrial Px III has been knocked out display an only minor and transient proliferation defect in vitro [19]. It remained however elusive whether the px III−/− parasites adapted to in vitro growth with or without Trolox are also able to cope with the more hostile environment in the mammalian host.
In the first series of in vivo experiments, mice were infected with px III−/− cells that had been cultured in the presence of 100 µM Trolox. As depicted in Figure 6A, the animals showed a 5-days extended medium survival time compared to mice infected with WT parasites. These differences could not be ascribed to an impaired initial infectivity, since at day 4 post-infection, all animals from both cohorts were infected and displayed a very similar average parasite burden (Figure 6B). The delayed disease progression of the px III−/− group (p-value of 0.017 for a log rank test [36]) was however in agreement with the overall slower in vivo proliferation of the mutant parasites compared to the WT strain.
To elucidate if growth in the presence of the antioxidant was responsible for the delayed phenotype, in the second series of experiments mice were infected with px III−/− cells that had been cultured in the absence of Trolox. Although the mean survival time of the animals infected with WT and mutant parasites was not significantly different, a log rank test [36] gave a p-value of 0.046 again suggesting a slightly delayed lethality for the animals infected with the px III−/− T. brucei (Figure 6C). The survival profiles of mice infected with px III−/− cells were comparable independently if the parasites had been cultured with or without Trolox prior to infection (Figures 6A and C). Only a slightly faster overall development of parasitemia was observed for the px III−/− -Trolox group compared to the +Trolox group (Figures 6B and D). Whether this is indicative of an adaptive mechanism to overcome the lack of the mitochondrial peroxidase in the absence of Trolox complementation remains to be investigated.
Passage through animal has been shown to increase the virulence of parasites maintained in vitro [37]. To assess whether px III−/− trypanosomes are capable to recover the virulence of the parental cell line, mice were infected with px III−/− and WT parasites isolated from infected animals. In both groups, the disease developed very rapidly resulting in an identical profile of animal death (Figure 6E). As observed in the experiments described above, parasitemia appeared to develop more slowly in mice infected with the px III−/− cells compared to the WT parasites (Figure 6F). Taken together, and irrespective of the reasons that cause their initially attenuated in vivo phenotype, the px III−/− parasites fully restored their virulence. Thus, the mitochondrial peroxidase is clearly not essential for T. brucei to infect the mammalian host.
African trypanosomes express three virtually identical Px-type proteins in the cytosol (Px I–II) and mitochondrion (Px III). The closest related enzyme in higher organisms is GPx4 [3], [38]. Both types of enzymes have in common that they are monomeric proteins, prefer lipid-derived hydroperoxides as substrates, and their physiological functions can be replaced by α-tocopherol or Trolox but not by water-soluble antioxidants [19], [22], [39]. In contrast to GPx4, the parasite Px I–III do not contain a selenocysteine but a cysteine residue in the active site and obtain their reducing equivalents from the kinetoplast-specific trypanothione/tryparedoxin system, not from glutathione [21], [40]. In mammals, the cytosolic GPx4 is the only known glutathione peroxidase that is essential. The inducible inactivation of the cytosolic GPx4 in mice revealed that the enzyme counteracts the activity of 12/15 lipoxygenase which catalyzes hydroperoxide formation in membranes and triggers an apoptosis-inducing factor-mediated cell death [22]. Intriguingly, the mechanism of cell death remained unclear since the canonical markers of programmed cell death (e.g. caspase 3 activation, phosphatidylserine exposure) were not detected. Only disruption of the mitochondrial membrane potential was observed as a late event upon induction of the GPx4 knockout [22].
Here we show that the lethal phenotype of BS px I–II−/− trypanosomes originates from the damage of their lysosome. Px I–II−/− cells deprived of Trolox and treated with Alexa Fluor-conjugated dextran became completely fluorescent with time; and cells fed with LysoTracker lost the fluorescent signal totally. The enlarged but still confined fluorescent signal observed in a fraction of the dextran-fed px I–II−/− parasites indicates that lysosomal swelling may precede rupture of the organelle. Our data did not support a specific pore formation in the lysosomal membrane as it is the case in parasites treated with the trypanolytic factor component apoL-I [26], [27]. Probably a reduced rate of export and/or recycling of (the damaged) membrane and content contributes to the enlargement of the lysosome, a mechanism discussed in the context of p67-ablated parasites [11]. Accordingly, the immunofluorescence analysis of px I–II−/− cells with p67 as lysosomal marker showed a progressive loss of the discrete organelle signal. All these findings strongly suggest that cell death starts with disintegration of the lysosome and likely evolves to massive membrane damages, as shown here for the mitochondrion. The absence of these cytosolic peroxidases probably affects the integrity of all (sub)cellular membranes. The Px-type enzymes use trypanothione as substrate suggesting that a decrease of the low molecular weight thiol should cause a similar phenotype. Indeed, the ultrastructural analysis of T. brucei depleted of trypanothione synthetase indicates membrane damage at several organelles [41]. However, lowering the trypanothione level will affect also the peroxiredoxin-type peroxidase and many other redox pathways and therefore the overall cellular redox state.
Lysosomes are oxidizing rather than reducing compartments [42] and therefore should require the presence of effective antioxidant systems. Dietary vitamin E has been reported to result in increased α-tocopherol levels in the lysosomal fractions and to prevent lysosomal release [43]. The full rescue of the lethal phenotype of the px I–II−/− parasites by Trolox strongly suggests that one physiological role of the cytosolic peroxidases is protection of the lysosomal membrane from peroxidation.
Both free iron and heme are known to react with hydrogen peroxide to generate highly oxidizing species that are capable of initiating lipid peroxidation [13], [44]. Indeed, supplementing Trolox-free medium with iron or holo-transferrin accelerated lysis of the px I–II−/− cells whereas the removal of FCS or addition of apo-transferrin slowed down trypanolysis. In addition, the use of transferrin-depleted medium suggests that also iron derived from the uptake and degradation of Hb contributes to the lethal phenotype of the px I–II−/− cells in the absence of Trolox. The lysosomal iron probably induces damage of the organelle membrane by inducing a Fenton-like reaction as described previously [13]. In line with a key role of iron in lysosomal damage, treatment of px I–II−/− cells with the iron chelator deferoxamine [32], [33] increased their short-term viability in Trolox-free medium. In human epithelial cells and lysosome-rich murine macrophage-like J774 cells, deferoxamine localizes almost exclusively within these organelles [31], [45]. Also for T. brucei, the lysosome has been suggested as an iron storage organelle [46] and a mucolipin 1 orthologue has been implicated in iron transport into the cytosol [47]. Intralysosomal iron can powerfully synergize oxidant-induced cellular damage. The steps associated with cell collapse in different mammalian cell lines were not fully understood [31], [48] although it was clear that lysosomal disruption entailed cell death [6].
In BS px I–II−/− T. brucei, the cellular damage is apparently linked to endocytosis. This is supported by several observations. In the absence of FCS, the px I–II−/− cells do not require Trolox for short-term viability and are even insensitive towards exogenous iron. This would not be expected if the cellular damage started at other membranes such as e. g. that of the glycosomes or mitochondrion or the plasma membrane. Disintegration of the cell membrane as primary event could be ruled out also by the fact that after feeding with fluorescent dextran and transfer into Trolox-free medium, a large proportion of the px I–II−/− cells became completely stained. Since in a variety of other cell types, the mitochondrion is the origin of endogenous oxidative stress we studied the damage of the lysosome and mitochondrion in more detail. In the px I–II−/− parasites, mitochondrial damage occurs but follows the disintegration of the lysosome. The order of events is thus opposite to that in cells with injured mitochondria where externalization of cardiolipin to the outer mitochondrial membrane acts as an elimination signal for the mitochondrion by autophagy [49].
Notably, any treatment, such as with exogenous iron, Hb or transferrin affected the viability of the px I–II−/− cells only in the absence of Trolox. All of these compounds enter the parasite via the endocytic pathway suggesting that the extremely fast lethal phenotype of BS trypanosomes that lack the cytosolic peroxidases is linked to their high endocytic rate. Taken together, disintegration of other intracellular membranes and the plasma membrane appears to be a secondary event or is at least much slower than lysosomal damage.
Parasites inhabiting the insect vector harbor a fully developed mitochondrion rich in cytochromes and Krebs cycle enzymes, many of which require iron or iron/sulfur complexes as cofactors [50]. Strikingly, in the insect stage, the cytosolic Px I–II proved to be entirely dispensable under culture conditions. One reason may be the lower endocytic activity of the procyclic cells compared to BS parasites [23]. In addition, procyclic T. brucei do not express a transferrin receptor [51] but can take up iron via specific transporters [52] or extract it from internalized hemin as it is the case in Leishmania infantum [53]. L. amazonensis has been shown to express a heme transporter (LHR1) in both the plasma membrane and acidic endocytic compartments; and highly syntenic, close homologs are present in T. brucei [54]. The direct delivery to the cytosol and/or the rapid transport from the lysosome into the mitochondrion probably precludes lysosomal iron accumulation in the procyclic cells and thus renders Px I–II not necessary for protection of this organelle towards iron-mediated lipid peroxidation. On the other hand, these stage-specific differences highlight the biological role of the cytosolic peroxidases at the interface between iron homeostasis and protection against iron-induced and lipid-derived oxidative stress in the pathogenic stage of the parasites.
BS T. brucei lack an active respiratory chain [7] and, although obligate, mitochondrial iron-sulfur cluster biosynthesis is much lower compared to the insect stage [9], [17]. These may be main reasons why BS parasites in which the mitochondrial px III gene has been knocked out do not display any strong proliferation phenotype [19]. The in vivo data presented here revealed that the Px III is not essential for infectivity and survival although mutant parasites grown in vitro prior to infection displayed a slightly delayed proliferation and disease development compared to WT pathogens. The cytosolic isoenzymes are probably sufficient for protecting the membranes of the metabolically repressed mitochondrion of BS parasites. Despite the fact that the virulence of the px III−/− cells was fully restored after a single passage through mouse, again the parasitemia developed more slowly than in animals infected with the parental cell line. This may suggest a permanent, albeit minor, impairment by the lack of a Px III-dependent function. Remarkably, one week after infection with the px III−/− parasites cultured in the presence or absence of Trolox, in four and two, respectively, of the six mice, the blood parasitemia had dropped to undetectable levels. This was not the case in any of the 12 animals infected with WT parasites and may be a consequence of the host innate immune defense. In the early stage of African trypanosomiasis, one of the hallmarks is the activation of the macrophage/monocyte system (for reviews see [55], [56]). As a result, reactive nitrogen and oxygen species are released and have cytotoxic and cytostatic effects on trypanosomes. Notably, IFN-γ and TNF-α, two cytokines participating actively in parasite control during the early stage of the infection [56], peak around the first week post-infection [57]. One may thus speculate that the mitochondrial peroxidase could contribute to the parasite defense towards exogenous oxidative stresses.
As shown here for African trypanosomes, cytosolic glutathione peroxidase-type enzymes are responsible for protecting the lysosome from oxidative membrane damage. To our knowledge, this is the first report demonstrating such function for cytosolic thiol peroxidases. The underlying molecular mechanism is not yet known and we can only speculate. Peroxidized phospholipids formed in the inner leaflet of the organelle membrane may be exchanged with lipids in the outer leaflet to allow the subsequent repair by the cytosolic peroxidases. In addition, it remains elusive if this lipid exchange would be a spontaneous or catalyzed process. At least for the plasma membrane of different Leishmania species, a phospholipid scramblase activity has been described [58]. In addition, cardiolipin externalization to the outer mitochondrial membrane has been demonstrated recently [49]. Although not yet investigated in detail, it is worth to note that other human pathogens such as Trichomonas, Toxoplasma, and Entamoeba can acquire iron via endocytic uptake of host Fe-binding proteins such as transferrin, lactoferrin, and ferritin (for reviews see [59], [60]) and the genomes of T. vaginalis and T. gondii encode Px homologues. Future work by others may reveal if our findings are extensible to these pathogens and the mammalian GPx4.
Px I–II−/− BS T. brucei and polyclonal rabbit antibodies against Px were generated previously [19]. Bovine holo-transferrin, 10,000 Da dextran conjugated to Alexa Fluor 488, LysoTracker Green DND-26, DIDS, and MitoTracker Red CMXRos were purchased from Life Technologies. FeCl3, deferoxamine mesylate, bovine Hb, bovine apo-transferrin, (±)-6-hydroxy-2,5,7,8-tetramethylchromane-2-carboxylic acid (Trolox), DAPI, and phleomycin were from Sigma. FCS was from Biochrome. Affinity-purified bovine transferrin antibodies were purchased from Bethyl Laboratories Inc. The monoclonal mouse anti-p67 antibody was a kind gift of Dr. James D. Bangs, Buffalo and the polyclonal rabbit anti-aldolase antibody was kindly provided by Dr. Christine Clayton, Heidelberg.
WT and mutant BS T. brucei (449 cells from Lister strain 427) were cultivated in HMI-9 medium [61] without serum plus, supplemented with 36 mM NaHCO3, 50 U/ml penicillin, 50 µg/ml streptomycin, and 0.2 µg/ml phleomycin. This medium contains 10% FCS ( = standard medium). Thus, unless otherwise stated, all experiments were performed in the presence of 10% FCS. The px I–II−/− cells were grown in the presence of 100 µM Trolox. Parasites were harvested and studied at a density of 3–9×105 cells/ml in medium ± Trolox, ± FCS and supplemented with FeCl3, deferoxamine, Hb, holo-transferrin, apo-transferrin, and DIDS, respectively. Viable cells with normal morphology were counted in a Neubauer chamber and the mean ± SD of three independent experiments was calculated. Statistical analyses were performed with Prism (GraphPad) and Microsoft (Excel) software and evaluated by paired two-tailed Student's t-test. Differences were considered to be significant when the p-value was ≤0.05.
The procedure for preparing the transferrin-depleted HMI-9 medium was adapted from those described by Schell et al. [62] and Ekblom et al. [63]. Affinity-purified bovine transferrin antibodies (12 mg) were coupled to 8 ml of CNBr-activated Sepharose 4B (GE Healthcare Life Sciences) according to the manufacturer's instructions yielding 11 mg covalently bound protein. The column was equilibrated with FCS-free HMI-9 medium. Standard medium was applied at a flow rate of 0.2 ml/min and 5 ml fractions were collected. After each pass, the column was washed with PBS, 8 M urea in PBS to remove bound transferrin, again with PBS, and re-equilibrated in FCS-free medium. The successful removal of transferrin was confirmed by Western blot analysis. The first pass yielded three, all subsequent passes, two transferrin-free fractions. Fractions devoid of transferrin were pooled yielding 45 ml of transferrin-depleted HMI-9 medium.
To replace the px I and II genes in the insect stage, procyclic T. brucei 449 cells were transfected with the vectors pHD1747-KOpxI(I–II) and pHD1748-KOpxI(I–II) originally generated for the respective work in BS cells [19]. The successful replacement of both alleles by a puromycin and blasticidin resistance gene, respectively, was verified by PCR analyses (Figure S1A, Table S1). Procyclic cells were grown at 27°C in MEM-Pros medium as described previously [21].
2×106 Procyclic T. brucei cells were harvested, boiled for 10 min in 1× sample buffer containing 2-mercaptoethanol, and subjected to SDS-PAGE (12% gel). In the case of the transferrin-depleted medium, 0.5 µl of the standard medium (corresponding to 100 ng transferrin) and of the fractions collected from the immuno column were applied. After blotting, the PVDF membranes were reacted with the primary antibody against Px (1∶2000), aldolase (1∶40000), and transferrin (1∶1000), respectively, followed by the goat anti-rabbit and rabbit anti-sheep immunoglobulin G conjugated to horseradish peroxidase antibodies (1∶10000, Santa Cruz Biotechnology), respectively, and developed by the SuperSignal West Pico Chemiluminescent Substrate kit (Pierce).
The parasites were kept for 30 min at 37°C in medium ± Trolox, harvested, resuspended in 50 µl of the respective medium containing 2.5 mg/ml Alexa Fluor 488-conjugated dextran, and incubated for 10 min. The cells were washed twice with medium and life cell imaging was performed as previously described [19]. For LysoTracker staining, the cells were incubated for various times at 37°C in medium ± Trolox. Thirty minutes prior to analysis, 5 µM LysoTracker was added. After twice washing with medium, the slides were examined using a Carl Zeiss LSM 510 confocal microscope and the LSM 510 software (Zeiss, Jena). The data presented are the mean ± SD of three independent experiments.
About 2×106 cells from logarithmic growth phase were harvested and incubated in medium ± Trolox at 37°C for different times. For MitoTracker staining, cells were incubated in medium + Trolox containing 0.12 µM MitoTracker for 15 min at 37°C, washed with PBS, and incubated in medium + Trolox for 30 min at 37°C. After washing with PBS, cells were fixed in 4% paraformaldehyde in PBS for 20 min at RT, transferred to 8-well poly-L-lysine slides (BD Falcon), and allowed to settle down for 1 h at RT or to untreated slides and incubated overnight at 4°C. The cells were permeabilized with 0.2% Triton X-100 (v/v in PBS) for 20 min at RT and washed twice with PBS. Subsequently, the slides were treated with 0.5% gelatine in PBS for 20 min, incubated for 1 h at RT with the p67 antibody (1∶800 in 0.5% gelatine in PBS), and washed with PBS. After 1 h incubation with goat anti-mouse antibodies coupled to Alexa Fluor 488 (Molecular Probes, 1∶10,000) and PBS washing, the DNA was stained with 500 ng/ml DAPI in PBS for 15 min at RT. The cells were mounted and examined under a Carl Zeiss Axiovert 200 M microscope equipped with an AxioCam MRm digital camera using the AxioVision program (Zeiss, Jena). The quantitative analysis was based on cells that displayed a single DAPI signal for each nuclear and kinetoplast DNA. The data are presented as mean ± SD of three independent experiments.
The animal protocols used in this work were evaluated and approved by the Animal Use and Ethic Committee (CEUA) of the Institut Pasteur Montevideo (Protocol 2009_1_3284). They are in accordance with FELASA guidelines and the National law for Laboratory Animal Experimentation (Law no. 18.611).
Thirty 6–8 weeks old Balb/cJ female mice were bred at the SPF animal facility of the Transgenic and Experimental Animal Unit of the Institut Pasteur de Montevideo (IPMon). They were housed in individual ventilated cages with negative pressure (Sealsafe rack, Tecniplast, Milano, Italy) in controlled environment at 20±1°C with a relative humidity of 40–60% in a 14/10 light-dark cycle. Food and water were administered ad libitum. The animals received a single intraperitoneal injection of 104 parasites harvested in exponential growth phase and suspended in 0.3 ml fresh medium. The following groups were studied: animals infected with WT parasites grown in vitro or isolated from an infected mouse, animals infected with px III−/− parasites cultured for one week in medium ±100 µM Trolox or isolated from a mouse which had been infected with parasites grown in vitro in the absence of Trolox. The parasites isolated from an infected animal were enriched. Briefly, 5–10 µl of anti-coagulated blood subjected to hypotonic lysis of red blood cells was added to one well in a 24-well culture plate containing 1 ml medium and incubated at 37°C and 5% CO2. Proliferating parasites were transferred to culture flasks containing 5–10 ml medium and cultivated for 10 to 16 days followed by cryo-preservation. At least four days prior to infection, the parasites were thawed and grown to exponential phase. The health status and survival of infected animals were monitored daily. Parasitemia levels in mice were determined regularly in blood samples (≤50 µl) taken from the submandibular vein. Blood was collected in a tube containing tri-potassium ethylenediamine tetra-acetic acid (K3EDTA) anticoagulant at a blood∶K3EDTA ratio of 20∶1. After thorough homogenization, an aliquot was diluted 1∶20 in a hypotonic solution (BD Pharm Lyse) to lyse red blood cells, incubated for 2 min at RT and diluted in PBS-1% (w/v) glucose for cell counting. The minimum parasite density detectable by this method is about 2.5×104 cells/ml. Mice showing an impaired health status and/or a parasite load of ≥108 cells/ml blood were euthanized. For the data of the Kaplan-Meier survival plots, log rank tests [36] were performed and the mean ± SD was calculated for the average parasite loads.
The genes studied in this work are:
px I: Tb427.07.1120
px II: Tb427.07.1130
px III: Tb427.07.1140
(TriTrypDB: http://www.tritrypdb.org/tritrypdb/)
|
10.1371/journal.pgen.1001203 | Epigenetic Silencing of Spermatocyte-Specific and Neuronal Genes by SUMO Modification of the Transcription Factor Sp3 | SUMO modification of transcription factors is linked to repression of transcription. The physiological significance of SUMO attachment to a particular transcriptional regulator, however, is largely unknown. We have employed the ubiquitously expressed murine transcription factor Sp3 to analyze the role of SUMOylation in vivo. We generated mice and mouse embryonic fibroblasts (MEFs) carrying a subtle point mutation in the SUMO attachment sequence of Sp3 (IKEE553D mutation). The E553D mutation impedes SUMOylation of Sp3 at K551 in vivo, without affecting Sp3 protein levels. Expression profiling revealed that spermatocyte-specific genes, such as Dmc1 and Dnahc8, and neuronal genes, including Paqr6, Rims3, and Robo3, are de-repressed in non-testicular and extra-neuronal mouse tissues and in mouse embryonic fibroblasts expressing the SUMOylation-deficient Sp3E553D mutant protein. Chromatin immunoprecipitation experiments show that transcriptional de-repression of these genes is accompanied by the loss of repressive heterochromatic marks such as H3K9 and H4K20 tri-methylation and impaired recruitment of repressive chromatin-modifying enzymes. Finally, analysis of the DNA methylation state of the Dmc1, Paqr6, and Rims3 promoters by bisulfite sequencing revealed that these genes are highly methylated in Sp3wt MEFs but are unmethylated in Sp3E553D MEFs linking SUMOylation of Sp3 to tissue-specific CpG methylation. Our results establish SUMO conjugation to Sp3 as a molecular beacon for the assembly of repression machineries to maintain tissue-specific transcriptional gene silencing.
| Cell type–specific gene expression patterns are largely regulated by positively or negatively acting transcription factors binding to promoter and enhancer elements. The ubiquitous transcription factor Sp3 represents a paradigm for a dual function transcription factor as it can activate and repress transcription. The repression function of Sp3 is mediated by attachment of a small protein designated SUMO to a single lysine residue. SUMOylation of Sp3 thus acts as a molecular switch that determines whether Sp3 acts as an activator or repressor. In this study, we have generated mice with a subtle mutation in the SUMO attachment site of Sp3. We found that several spermatocyte- and brain-specific genes that are silenced in non-testicular and extra-neuronal tissues of wild-type animals become aberrantly de-repressed in mice in which the SUMO attachment site of Sp3 is mutated. De-repression of these genes is accompanied with dramatic epigenetic changes including the loss of repressive histone methylation marks and, most significantly, loss of DNA methylation. Our findings suggest that SUMO modification of a transcription factor can act as a molecular beacon for the assembly of repression machineries to maintain tissue-specific transcriptional gene silencing in vivo.
| A plethora of proteins involved in regulating gene expression such as promoter-specific transcription factors, cofactors and chromatin-modifying enzymes are reversibly modified by the Small Ubiquitin-like MOdifier SUMO (reviewed in [1], [2]). With few exceptions, SUMO modification of transcriptional regulators correlates with repression of transcription [3]–[5].
The ubiquitously expressed transcription factor Sp3 represents a well-studied paradigm for regulation of activity by SUMOylation [6]–[8]. Sp3 belongs to the Sp (specificity protein) family of transcription factors that is implicated in the expression of a wide variety of genes including housekeeping, tissue-specific, developmentally and cell-cycle regulated genes [9]–[12]. A major feature of Sp3 is that, depending on promoter context, it can either activate or repress transcription in reporter gene assays [6], [7], [13]. Two glutamine-rich domains are known to exercise the activation function of Sp3 [13]; whereas the repressive activity of Sp3 is mediated by attachment of SUMO to lysine 551. K551 lies within the SUMO consensus motif IKEE located between the second Q-rich activation domain and the DNA-binding domain [6], [7] (Figure 1A and 1B). The functional complexity of Sp3 is further increased in vivo by the expression of four different isoforms that differ in their N-terminal extension [14]. All of these isoforms are SUMO-modified at K551, giving rise to a composite pattern of at least eight distinct protein species [14].
Previous investigations of the molecular events associated with Sp3-SUMO-dependent repression have provided mechanistic clues underlying SUMO-dependent gene silencing [15], [16]. SUMO-modification can act as a signal for the recruitment of various chromatin-associated repression components including the chromatin remodeler Mi-2, the MBT-domain proteins L3MBTL1 and L3MBTL2, heterochromatin protein 1 (HP1) and the histone methyltransferases (HMTs) SETDB1/ESET and SUV4-20H, concomitant with the establishment of repressive histone modifications such as H3K9 and H4K20 tri-methylation [16].
Despite extensive studies on the repression function of SUMOylated Sp3 and other transcription factors, the significance of SUMO attachment for the expression of endogenous genes in vivo is still largely unknown. Here, we report the generation of mice and mouse embryonic fibroblasts (MEFs) with a point mutation in the SUMO attachment sequence of Sp3. Expression profiling revealed that SUMOylation of Sp3 is required for silencing of spermatocyte-specific genes such as Dmc1 and Dnahc8 in somatic cells, and neuronal genes including Paqr6, Rims3 and Robo3 in non-neuronal cells. Transcriptional de-repression of these genes in MEFs expressing the Sp3E553D mutant protein is accompanied by the loss of repressive heterochromatic marks such as H3K9 and H4K20 tri-methylation, impaired recruitment of repressive chromatin-modifying enzymes and loss of DNA methylation. Our results establish that SUMO-modification of Sp3 acts as a platform for the assembly of repression machineries to maintain tissue-specific transcriptional gene silencing.
To investigate the in vivo function of Sp3 SUMOylation, we generated mice in which the SUMO attachment site IK551EE is mutated to IK551ED (Figure 1A). We chose glutamic acid residue E553 for mutation because K551 might also be a target for other posttranslational modifications such as methylation or acetylation. A vector carrying the Sp3E553D mutation and a floxed neomycin-resistance cassette was used for targeting of ES cells to generate heterozygous mutant mice (Figure 1B; Figure S1). The neomycin-resistance gene was subsequently removed by mating with appropriate Cre recombinase-expressing mice [17]. Mice carrying the Sp3E553D mutation will from hereon be referred to as Sp3 knockin (Sp3ki) mutant. Heterozygous Sp3wt/ki mice and homozygous Sp3ki/ki mice were fertile, born at the expected Mendelian frequency and exhibited no obvious phenotype (Table S1).
To ensure that the E553D mutation impaired SUMOylation of Sp3, we performed Western blotting of adult mouse tissues and MEFs derived from E13.5 embryos. SUMO-modification of Sp3 was readily detectable in Sp3wt and heterozygous Sp3wt/ki but not in homozygous Sp3ki/ki tissues and MEFs (Figure 1C; Figure S1). This result demonstrates that the glutamic acid residue within the SUMOylation consensus motif ΨKXE is absolutely essential for the attachment of SUMO to endogenous Sp3. We conclude that the E553D mutation carried by Sp3ki/ki mice impedes SUMOylation of Sp3 at K551 in vivo, without affecting Sp3 protein levels.
To identify genes that are regulated by SUMO-modified Sp3, we performed gene expression profiling with RNA extracted from primary Sp3wt and Sp3ki/ki MEFs derived from E13.5 littermates. This identified 68 genes that were upregulated and 7 genes that were downregulated by more than 2-fold in Sp3ki/ki MEFs (Table S2). Notably, top candidate genes that were upregulated in Sp3ki/ki MEFs encode developmentally-regulated meiotic and neuronal proteins. Dmc1 and Dnahc8 are expressed in meiotic spermatocytes and encode a RecA-like recombinase and a flagellar protein, respectively [18], [19]. Paqr6, Rims3 and Robo3 are expressed in the central nervous system [20]–[22]. The expression pattern of another upregulated gene (Villin-like, Vill) is largely unknown, although a low level of expression in early embryogenesis was reported [23].
To validate aberrant de-repression of these genes, we analyzed their expression in MEF cultures by quantitative RT-PCR. Dmc1, Dnahc8, Paqr6, Rims3, Robo3 and Vill mRNA levels were elevated in primary Sp3ki/ki MEFs as well as in immortalized Sp3ki/ki MEFs (Figure 2). We also analyzed expression of Dmc1, Dnahc8, Paqr6, Rims3, Robo3 and Vill in Sp3-deficient (Sp3-/-) MEFs obtained from E13.5 Sp3 knockout embryos [24]. Consistent with an Sp3-SUMO-dependent silencing function, all six genes were upregulated also in Sp3-/- MEFs as compared to corresponding Sp3wt MEFs derived from littermates (Figure 2). We also performed immunoblot analysis for Dmc1 and Dnahc8 but failed to detect these proteins probably due to their low expression levels. However, de-repression of the Dmc1 gene in Sp3ki/ki and Sp3-/- MEFs was verified with different amplimers spanning different exons (data not shown), thereby precluding the possibility that the qRT-PCR analyses detected an aberrant Dmc1 transcript.
De-repression of Dmc1, Dnahc8, Paqr6, Rims3, Robo3 and Vill in Sp3ki/ki and Sp3-/- MEF cultures suggested that SUMO modification of Sp3 might be essential for silencing these genes in tissues other than testis and brain, respectively. We analyzed RNA from various tissues of adult Sp3wt and Sp3ki/ki mice. As expected, Dmc1 and Dnahc8 were strongly expressed in testis of Sp3wt and Sp3ki/ki mice at similar levels but were not or only marginally expressed in other tissues such as brain, heart, intestine, kidney, liver, lung and spleen. In Sp3ki/ki mice, Dmc1 and Dnahc8 mRNA levels were significantly higher in all tissues (Figure 3). However, the amount of RNA in testis was still one to three orders of magnitude higher indicating that testis-specific activators further enhance expression of Dmc1 and Dnahc8 in spermatocytes.
Previous reports have attributed silencing of several meiotic and male germ-line-specific genes in somatic cells such as Smc1ß and Stag3 to E2F6 [25], [26], a repressive member of the E2F family of transcription factors. We analyzed expression of Smc1ß and Stag3 in tissues of Sp3ki/ki mice as well. Both genes were only detectable in testis RNA preparations and were not de-repressed in non-testicular Sp3ki/ki tissues or Sp3ki/ki MEFs (data not shown). Vice versa, the Dmc1 gene is not de-repressed in E2F6-/- MEFs [27] although the Dmc1 promoter region is bound by E2F6 in vivo at a conserved binding site [27]. This observation suggests that different transcription factors and mechanisms are responsible for repressing spermatocyte-specific genes in somatic cells.
Consistent with published data, Paqr6, Rims3 and Robo3 mRNA levels were highest in mouse brain. Strikingly, all three genes were also highly expressed in testis and in the case of Robo3 also in kidney. In all other organs these mRNAs were either not detectable or expressed only at a very low level (Figure 3). Nevertheless, expression of Paqr6, Rims3 and Robo3 as well as Vill mRNA was significantly elevated in several organs of Sp3ki/ki mice (Figure 3). Taken together, these results demonstrate that SUMOylation of Sp3 is essential for silencing of a subset of spermatocyte-specific and neuronal genes in somatic and non-neuronal tissues, respectively, implying an important role of the SUMO moiety attached to Sp3 in establishing tissue-specific gene expression patterns.
To substantiate the notion that SUMOylation of Sp3 is directly responsible for silencing testis- and neuronal-specific genes in MEFs, we re-expressed the short and long isoforms of Sp3 (Sp3si-wt and Sp3li-wt) in Sp3-/- MEFs by retroviral transduction (Figure 4A). As controls, we used the corresponding SUMOylation-deficient Sp3 mutants (Sp3si-K551D and Sp3li-K551R). Particularly, re-expression of the long isoform of Sp3 resulted in significantly reduced expression of the Dmc1, Dnahc8, Paqr6, Rims3, Robo3 and Vill genes. Repression of these genes by re-expression of the small isoforms of Sp3 was less pronounced (Figure 4B–4G). The weaker effects observed with the small isoforms of Sp3 could be due to their lower expression level (see Figure 4A). Potentially, simultaneous expression of all four wild type Sp3 isoforms would be necessary to restore repression completely. Nevertheless, in contrast to the wild type Sp3 isoforms, introduction of the SUMOylation-deficient Sp3 mutants failed to rescue gene silencing but instead further enhanced expression of these genes (Figure 4B–4G). These results show that reintroduction of Sp3 partially reverses de-repression of these genes in a SUMOylation-dependent manner.
To analyze whether Sp3 is bound to the promoters of genes that are repressed by Sp3-SUMO, we performed ChIP analyses. Because of the lack of precise promoter information for Dnahc8, Robo3 and Vill we focused on Dmc1, Paqr6 and Rims3. All three promoters contain several potential binding sites for Sp3. Antibodies to Sp3 precipitated all three promoters from Sp3wt and Sp3ki/ki MEF chromatin but not from Sp3-/- MEF chromatin (Figure 5), demonstrating that both wild type Sp3 and the SUMOylation-deficient Sp3E553D mutant were bound to the Dmc1, Paqr6 and Rims3 promoters.
Next, we analyzed the Dmc1, Paqr6 and Rims3 promoters for the presence of repressive histone modifications (Figure 5). In Sp3wt MEFs, the H3K27me3 mark is abundantly present at the Dmc1 and Rims3 promoters but not at the Paqr6 promoter. Moreover, this mark is not or only marginally reduced in the absence of SUMOylated Sp3 suggesting that H3K27me3 does not contribute to SUMO-dependent gene silencing of these three genes. In contrast, H3K9me3 and H4K20me3 marks are present at all three promoters in Sp3wt MEFs but are strongly reduced in Sp3ki/ki and in Sp3-/- MEFs. Consistently, HP1α, which binds H3K9me3, is present at the Dmc1, Paqr6 and Rims3 promoters in Sp3wt MEFs but not in Sp3ki/ki and Sp3-/- MEFs. Thus, the presence or absence of H3K9me3, H4K20me3 and HP1α at the Dmc1, Paqr6 and Rims3 promoters correlates strictly with the repressed or de-repressed state of these genes. We also analyzed for the presence of H3K4 trimethylation, an epigenetic mark characteristic for promoter-proximal nucleosomes of most active as well as inactive genes [28]. The co-occurrence of H3K9me3, H3K27me3 and H3K4me3 marks is a characteristic property of “bivalent” promoters of euchromatic genes in ES cells, believed to reflect a repressed but poised transcriptional state [29]. The H3K4me3 mark was abundantly present on the Dmc1, Paqr6 and Rims3 promoters in Sp3wt MEFs. In Sp3ki/ki MEFs we found an approximately 2-fold and 3-fold increase of H3K4 trimethylation on the Dmc1 promoter and on the Paqr6 promoter, respectively, but not on the Rims3 promoter (Figure 5). In Sp3-/- cells, higher H3K4me3 levels were detected on the Dmc1 promoter but not on the Rims3 and Paqr6 promoters. Thus, there is no strict correlation between the changes of the H3K4me3 mark and the expression state of the different target genes.
Our previous investigations revealed that the establishment of repressive nucleosomal signatures on a chromatinized Gal4-driven reporter gene by Gal4-Sp3-SUMO involves the recruitment of the histone methyltransferase SETDB1, the chromatin remodeler Mi-2, and the chromatin-compacting MBT-domain proteins L3MBTL1 and L3MBTL2 [16]. Therefore, we analyzed the Dmc1, Paqr6 and Rims3 promoters for the presence of these proteins. ChIP analysis revealed that all four proteins are abundantly present on the promoters in Sp3wt MEFs but strongly reduced in Sp3ki/ki and Sp3-/- MEFs (Figure 5). The recruitment of these chromatin-modifying proteins to the endogenous Dmc1, Paqr6 and Rims3 promoters is thus dependent on the presence of the Sp3 SUMO moiety. Taken together, these results support the conclusion that the posttranslational modification of Sp3 at K551 provokes a local repressive chromatin structure on a subset of spermatocyte- and neuronal-specific genes in somatic and non-neuronal cells, respectively.
Repression of many nervous system-specific genes in unrelated tissues has been attributed to the corepressor CoREST1. CoREST1 is recruited to promoters by the transcriptional repressor REST/NRSF [30] and, alternatively, by a REST/NRSF-independent mechanism that involves direct interaction between CoREST1 and a thus far unknown SUMO2/3-modified transcription factor [31]. To examine whether CoREST1 is also involved in silencing the Dmc1, Paqr6 and Rims3 promoters, we performed ChIP analysis for the presence of CoREST1. CoREST1 was to some extent detectable on the Rims3 promoter but not on the Dmc1 and Paqr6 promoters irrespectively of the Sp3 SUMOylation status (Figure S2). This finding indicates that SUMOylation of Sp3 represents an alternative, CoREST1-independent pathway mediating extra-neuronal repression.
An EMBOS CpGPlot analysis (http://www.ebi.ac.uk/Tools/emboss/cpgplot/index.html) revealed that the Dmc1, Paqr6 and Rims3 promoters are embedded in CpG islands. CpG island methylation may contribute to silencing of these genes in MEFs and may be reversed in Sp3ki/ki and Sp3-/- MEFs. Therefore, we analyzed the methylation states of these promoters in Sp3wt, Sp3ki/ki and Sp3-/- MEFs. Bisulfite sequencing revealed that the proximal promoters and the first exons of the Dmc1, Paqr6 and Rims3 genes are highly methylated in Sp3wt MEFs. In contrast, CpG methylation is strongly reduced in Sp3ki/ki and in Sp3-/- MEFs (Figure 6). In summary, there is a very tight correlation between SUMO modification of Sp3, transcriptional repression, repressive histone modifications and DNA methylation of these promoters. Lack of SUMOylation of Sp3 results in their de-repression accompanied by the absence of repressive histone modifications and the absence of CpG methylation.
The generation and analysis of mice with a subtle point mutation in the SUMO attachment site of the transcription factor Sp3 has revealed the relevance of SUMO modification of Sp3 for gene silencing in vivo. SUMO attachment to Sp3 serves as a molecular beacon for the recruitment of chromatin-modifying machineries that impose epigenetic silencing on a subset of spermatocyte-specific and neuronal genes. Our data are consistent with the bidirectional crosstalk between repressive histone modification and DNA methylation, that was established by demonstrating direct interactions between SETDB1 and the de novo methylase DNMT3A [32]. However, SETDB1 may also be recruited indirectly through interaction with the methyl-CpG binding protein MBD1 that forms a stable complex with SETDB1 [33]. We note that a functional SUMO interaction motif (SIM) is present in the histone methyltransferase SETDB1 [34] and that potential SIMs are also present in DNMTs. Thus, SUMO-modified Sp3 might recruit SETDB1 and DNMTs independently. Sp3-SUMO-mediated repression might involve a timely coordinated recruitment of chromatin remodelers, nucleosome compactors such as MBT domain proteins, HMTs and DNMTs. Alternatively, these different types of epigenetic players might be recruited simultaneously. The Sp3ki/ki and Sp3-/- MEFs described here provide essential tools for future experiments addressing these questions.
The lack of Sp3 SUMOylation in Sp3ki/ki mice causes aberrant expression of several spermatocyte-specific and neuronal genes in various somatic tissues. Although de-repression of these genes is best described by the loss of the repressive function of SUMOylated Sp3 in Sp3ki/ki mice, it is conceivable that the activation function of the Sp3E553D mutant protein may contribute directly or indirectly to the aberrant expression of these genes as well. However, expression of the spermatocyte-specific and neuronal genes in testis and brain, respectively, is not affected. This tissue-selectivity could be due to low-level Sp3 SUMOylation in spermatocytes and neurons. To investigate this, we performed Western blot analysis of testis and brain extracts. Consistent with the lack of repression in testis and brain, SUMOylated Sp3 species were barely detectable in these tissues (Figure S3). Future studies using purified spermatocytes and neurons might provide further insight on the cell type-specific SUMOylation state of Sp3. Interestingly, it has been reported that Sp3 expression in germ cells declines during the leptotene to pachytene transition whereas the related transcription factor Sp1 did not decline until the mid-pachytene phase of meiosis [35]. Dmc1 is expressed in leptotene to zygotene spermatocytes [19], and Dnahc8 from mid-pachytene to diplotene spermatocytes [18]. Thus, activation of the Dmc1 and Dnahc8 genes at these meiotic stages correlates with down-regulation of Sp3. Accordingly, one could image a scenario in which down-regulation of the Sp3 protein level facilitates Sp1-mediated activation of these two Sp3-SUMO target genes during spermatocyte development.
The promoters of the three Sp3-SUMO target genes that we analyzed in detail share several features. They contain multiple GC-boxes, lack a TATA box and are embedded in CpG islands. Such a promoter arrangement is reminiscent of housekeeping genes. In contrast to the three Sp3 target genes, housekeeping genes are ubiquitously expressed and remain unmethylated in all tissues. It is currently unclear why the three Sp3 target genes are expressed in such a highly tissue-specific manner. We have not detected obvious features in the spacing or orientation of Sp-binding sites that can account for this interesting difference. Further investigations comparing these two types of promoters are required to address this enigmatic point.
The aberrant expression of several testicular and neuronal genes in Sp3ki/ki mice apparently does not lead to obvious histological or behavioral abnormalities under standard mouse housing conditions indicating that essential functions of differentiated cell types are not grossly impaired. We note that the overall amounts of aberrantly produced mRNA transcripts in somatic extra-neuronal cells are low. This provides an explanation for the absence of clear anatomical and physiological anomalies in the Sp3ki/ki mice.
The phenotype of the Sp3ki/ki mice differs significantly from mice lacking the entire Sp3 protein. Sp3-deficient mice display skeletal, tooth, hematopoietic and heart defects at late embryonic development, and die immediately after birth due to respiratory failure [24], [36]–[38]. Given that SUMOylation-deficient Sp3 proteins are strong activators [13], [14] the defects observed in Sp3-/- mice have to be attributed largely to the activation function of Sp3.
Research involving mice have been conducted according to the German Animal Protection Law (Tierschutzgesetz). The application for the experiments was reviewed and approved by the responsible local authorities (Regierungspräsidium Giessen, reference number V 54–19 c 20/15 cMR20/27).
A targeting vector containing the Sp3E553D mutation was constructed and transfected into ES cells (Text S1). For selection of ES cells, we used a floxed IRES-LacZ-neo-polyA cassette that integrates into intron 5 of the Sp3 gene by homologous recombination. A single clone out of >200 G418-resistant colonies showed the homologous recombination event. After karyotyping, the ES clone was injected into C57BL/6 blastocysts. Breeding of the chimeras revealed germ-line transmission of the targeted Sp3 allele. The IRES-lacZ-neo cassette was removed by mating heterozygous mice with mice expressing the Cre recombinase under control of the cytomegalovirus-immediate early enhancer-chicken beta-actin hybrid (CAG) promoter [17]. Offspring were genotyped by Southern blotting and PCR (Figure S1). The Sp3wt/ki heterozygous offspring were intercrossed and homozygous Sp3ki/ki mice were obtained.
Retroviral vectors for expression of the long and short isoforms of Sp3 and corresponding SUMOylation-deficient mutants were generated by cloning of appropriate wild type and mutant Sp3 cDNA fragments [14] into the pBABE-puro plasmid. Retroviral packaging in Phoenix cells and infection of immortalized Sp3-/- MEFs [24] were performed according to standard procedures. Transduced cells were selected for uptake of retrovirus with 2 µg/mL of puromycin.
Whole cell extract from MEFs and mouse tissues were prepared as described [7], separated on SDS-polyacrylamide gels, blotted on PVDF membranes and probed with anti Sp3 antibodies (Santa Cruz Biotechnology, sc-644). Secondary antibodies were visualized using the Immobilon Western HRP substrate (Millipore).
Total RNA was prepared from freshly isolated Sp3wt and Sp3ki/ki mouse embryonic fibroblasts of E13.5 siblings using the RNeasy kit (Qiagen). Purified RNA was labeled with the two-color Quick-Amp Labeling kit (Agilent) and hybridized to a whole genome microarray 4x44K 60mer slide (G4122F) according to the manufacturer's instructions (Agilent). Microarray data were analyzed using Bioconductor [39]. The loess method implemented in the Bioconductor package marray was applied for normalization. Two biological replicates (male and female MEFs) were performed. Genes were considered as regulated when they had a fold change of ≥2, a logarithmic intensity value (base 2) of ≥5 in Sp3ki/ki, and when the expression level of replicates were similar. Similarity for two log2 transformed expression levels was determined ad hoc by the constraint max(1, |e1, e2| ×0.75,) > |e1 – e2|.
Microarray data were deposited at ArrayExpress (www.ebi.ac.uk/arrayexpress) under accession number E-MEXP-2755.
One microgram of total RNA prepared from MEFs and mouse organs was used for cDNA synthesis along with 0.5 µg of oligo(dT) primer and 200 U of M-MLV reverse transcriptase (Invitrogen). Quantitative RT-PCR was performed with 1 µL of 1∶20 diluted cDNA using gene-specific primers (Text S1). qPCRs were performed in quadruplicate using the Absolute SYBRGreen qPCR Mix (Abgene) on the Mx3000P real-time PCR system (Stratagene). Values were normalized to Gapdh and/or Sp1 mRNA content.
Chromatin immunoprecipitation was performed using the One Day ChIP kit (Diagenode) in accordance to the manufacturer's instructions. Primer sequences specific for Dmc1, Paqr6 and Rims3 promoter regions can be found in Text S1. Antibodies used for ChIP analysis are described in [16].
For DNA methylation analysis, 2 µg of genomic DNA derived from immortalized Sp3wt, Sp3-/- and Sp3ki/ki MEFs were subjected to sodium bisulfite conversion of unmethylated cytosines using the EpiTect Bisulfite Kit (Qiagen) in accordance to the manufacturer's instructions. Converted DNA was subjected to PCR amplification using promoter-specific BamHI- and KpnI-tailed primers (Text S1) and the ImmoMix PCR reagent (Bioline). PCR products were cloned into the pcDNA3 vector and 10 clones were sequenced using the BGHrev primer.
|
10.1371/journal.pbio.1000542 | A Polarised Population of Dynamic Microtubules Mediates Homeostatic Length Control in Animal Cells | Because physical form and function are intimately linked, mechanisms that maintain cell shape and size within strict limits are likely to be important for a wide variety of biological processes. However, while intrinsic controls have been found to contribute to the relatively well-defined shape of bacteria and yeast cells, the extent to which individual cells from a multicellular animal control their plastic form remains unclear. Here, using micropatterned lines to limit cell extension to one dimension, we show that cells spread to a characteristic steady-state length that is independent of cell size, pattern width, and cortical actin. Instead, homeostatic length control on lines depends on a population of dynamic microtubules that lead during cell extension, and that are aligned along the long cell axis as the result of interactions of microtubule plus ends with the lateral cell cortex. Similarly, during the development of the zebrafish neural tube, elongated neuroepithelial cells maintain a relatively well-defined length that is independent of cell size but dependent upon oriented microtubules. A simple, quantitative model of cellular extension driven by microtubules recapitulates cell elongation on lines, the steady-state distribution of microtubules, and cell length homeostasis, and predicts the effects of microtubule inhibitors on cell length. Together this experimental and theoretical analysis suggests that microtubule dynamics impose unexpected limits on cell geometry that enable cells to regulate their length. Since cells are the building blocks and architects of tissue morphogenesis, such intrinsically defined limits may be important for development and homeostasis in multicellular organisms.
| Because many physical processes change with scale, size control is a fundamental problem for living systems. While in some instances the size of a structure is directly determined by the dimensions of its individual constituents, many biological structures are dynamic, self-organising assemblies of relatively small component parts. How such assemblies are maintained within defined size limits remains poorly understood. Here, by confining cells to spread on lines, we show that animal cells reach a defined length that is independent of their volume and width. In searching for a “ruler” that might determine this axial limit to cell spreading, we identified a population of dynamic microtubule polymers that become oriented along the long axis of cells. This growing population of oriented microtubules drives extension of the spreading cell margin while, conversely, interactions with the cell margin promote microtubule depolymerisation, leading to cell shortening. Using a mathematical model we show that this coupling of dynamic microtubule polymerisation and depolymerisation with directed cell elongation is sufficient to explain the limit to cell spreading and cell length homeostasis. Because microtubules appear to regulate cell length in a similar way in the developing zebrafish neural tube, we suggest that this microtubule-dependent mechanism is likely to be of widespread importance for the regulation of cell and tissue geometry.
| The physical properties of a system depend to a large extent upon its scale. Therefore, it is not surprising to find that many biological structures are maintained within relatively tightly constrained size limits [1],[2]. In some cases, the dimensions of macromolecular assemblies are enforced by “molecular rulers” like titin, which helps to govern the length of the sarcomeric repeats in muscle [3]. However, many seemingly stable structures, such as metaphase spindles [4] and cilia [1], exist in a state of dynamic equilibrium in which a stable form arises from the collective action of a large number of molecular machines functioning in concert. Although mechanisms have been proposed for the control of the length of such polymers [1], through for example length-dependent microtubule depolymerisation [5], little is known about this fundamental and widespread biological phenomenon.
For unicellular organisms, intrinsic mechanisms have been identified that regulate cell shape [2],[6], maintain a steady-state cell size, and couple cell length and size [7]. However, it remains unclear whether similar controls regulate the dimensions of cells from multicellular animals, which, by virtue of not having a cell wall, assume a form that is plastic and a variable size, both of which depend to a large degree upon the extracellular tissue environment in which cells find themselves [8],[9]. Nevertheless, since form and function are intimately linked and vary from cell type to cell type, it seems likely that the shape of many animal cells will be maintained within intrinsically defined limits. Such behaviour has been observed in assays of cell spreading [10] and cell migration on planar adhesive substrates [11],[12]. Moreover, studies of cells on grooved, scratched, or patterned substrates have in some cases [13],[14] revealed limits to cell extension. In addition, regulated changes in cell geometry have long been known to drive a variety of morphogenesis movements in developing animals. During Drosophila development, for example, changes in epithelial cell shape and height are thought to drive internalisation of the ventral furrow [15]. Similarly, during neural tube development in zebrafish, individual neuroepithelial cells act together to form a double-layered epithelium with a defined width, even under conditions in which the entire structure is mis-positioned or duplicated [16].
Here, to systematically investigate the mechanisms that control animal cell geometry, we employed micro-contact printing [17] to generate adhesive lines of extracellular matrix that limit cell width but leave cells free to regulate their geometry in the other two dimensions (length and height). This analysis of cells spreading on adhesive lines reveals that animal cells can spread to a characteristic steady-state length that is independent of cell size, pattern width, and cortical actin, but is dependent on a population of microtubules that aligns along the long axis of cells as the result of interactions between microtubule plus ends and the cell cortex. Similarly, a population of oriented microtubules mediates length control in epithelial cells of the developing zebrafish neural tube. A mathematical model shows that cell length homeostasis in these cases can be quantitatively explained by the collective action of dynamic, orientated microtubules as they drive cell extension and undergo cortex-dependent catastrophe. Together, this experimental and theoretical analysis reveals a role for microtubules in homeostatic cell length control in HeLa cells, Drosophila S2R+ cells, and zebrafish neuroepithelial cells, suggesting that it may be an important general feature of animal cell biology.
To explore the intrinsic regulation of cell shape, we began by seeding a population of freshly harvested exponentially growing HeLa cells onto micropatterned fibronectin lines (Figure 1A) ranging in width from 3 to 35 µm, separated by non-adhesive polyethylene glycol, and onto equivalent non-patterned areas of the substrate (Figure 1B). Other researchers previously performed similar experiments by plating cells on grooved or scratched substrates [18]–[20] or on adhesive strips [13]. In our case, after allowing 2 h for cells to adhere to and spread on the micropatterned substrate, cell length was monitored using semi-automated software designed to remove user bias and to facilitate the analysis of large datasets (Figure S1). Cell length for this analysis was defined as the maximum distance, parallel to the patterned line, separating extensions at distal cell tips (Figure 1A). Unexpectedly, cells from an exponentially growing population (with a wide range of masses) spread to a relatively well-defined average length of 44±10 µm, which proved largely independent of line width and similar to that of cells on non-patterned substrates (∞) (Figure 1C). Cell spreading in this assay was accompanied by a corresponding change in cell height, as expected if cell volume is conserved (Figure 1E and data not shown). These observations were similar to those previously reported for fibroblasts on scratched substrates [13]. Because a fixed-time-point assay was used for this analysis, however, the independence of cell length and pattern width could be the result of either a constant rate of cell spreading or the action of a cell length control mechanism. When live-cell imaging was used to examine the kinetics of cell spreading, we observed HeLa cells spreading monotonically over a period of approximately 60 min, reaching a steady-state length that was independent of pattern width (Figure 1D and 1E). These data suggest that HeLa cells have an intrinsically defined length.
To determine whether cell length control is a peculiarity of HeLa cells spreading on a fibronectin substrate or a more general feature of animal cell biology, we repeated these experiments using the adherent Drosophila hemocyte-derived S2R+ cells [21], which are significantly smaller than HeLa cells (with a mean volume, measured using an automated cell counter, of 1,177±64 µm3 for Drosophila S2R+ cells compared to 2,121±1,116 µm3 for HeLa cells; Figure 2A). These cells do not adhere well to fibronection-coated substrates, but could be induced to spread on glass dishes coated with Concanavalin A (ConA) [15]. Once spreading was complete (5 h after plating), the length of these cells was measured. Like HeLa cells, patterned S2R+ cells were found to achieve a reproducible steady-state length that was relatively independent of their width (Figure 2B and data not shown). Moreover, despite their very different average volumes, S2R+ and HeLa cells had comparable resting lengths (55±9 µm and 44±10 µm, respectively). This observation suggested the possibility that the limit of an animal cell's long axis might be regulated independently of its volume. As a direct test of this hypothesis we altered the culturing conditions to obtain a population of HeLa cells grown to confluence that had an average volume of 1,233±835 µm3 (about half that of HeLa cells grown to 50% confluence; Figure 2A). When the two populations of differently sized HeLa cells were plated onto micro-contact-printed fibronectin lines of varying widths for 2 h, they were found to spread to statistically similar lengths (Figure 2B) despite having different heights (data not shown), confirming the independence of cell length and volume.
It seemed unlikely that cell length control depends on plasma membrane tension, since dramatic changes in cell volume and line width had little impact on the length of HeLa or S2R+ cells, or on the spreading kinetics of HeLa cells on narrow and thick patterned lines (Figure 1D). In searching for mechanisms underlying cell length control we therefore turned to examine a possible role for the cytoskeleton [22], since actin filaments and microtubules have been implicated in the control of fixed-length structures [1],[3],[23] and in the regulation of cell length [6],[13] and cell spreading [18],[24]–[27]. We began by analysing the cytoskeletal changes that accompany cell spreading on micropatterned lines. Following attachment to a patterned surface and a brief period of blebbing (Figure 1E), cells developed spreading lamellipodia that quickly reached the edges of the patterned lines. While lamellipodia reaching laterally outside of the pattern into regions passified with polyethylene glycol underwent periodic cycles of extension and retraction [28], lamellipodia extending in the direction of cell elongation remained tightly bound to the adhesive substrate (data not shown). To test whether these actin-based lamellipodia play a role in cell elongation or cell length control, we took advantage of the efficacy of RNA interference (RNAi)–mediated gene silencing in fly cell culture to target SCAR/WAVE (hereafter SCAR) to selectively remove these structures from spreading cells [29]. Five days after treatment with control (LacZ) or SCAR double-stranded RNA (dsRNA) [30],[31], Drosophila S2R+ cells were seeded onto non-patterned substrates (Figure 3A and 3B) and onto micropatterned ConA lines (Figure 3C). Cell length was then assayed 5 h later (Figure 3D). As Figure 3C shows, SCAR RNAi cells assumed their typical RNAi phenotype, in which lamellipodia are replaced by long, radial, microtubule-rich processes [31]. Despite this, the majority of microtubule-based protrusions became oriented along the ConA lines. When we measured the distance parallel to the pattern between distal protrusion tips in these cells, we found that the length of SCAR RNAi cells was statistically indistinguishable from that of control cells (Figure 3D). Similarly, control RNAi and SCAR RNAi cells reached similar steady-state spread diameters on non-patterned substrates [31] (Figure 3D). Thus, although SCAR is required for the formation of lamellipodial-based protrusions, lamellipodial actin is not required for cell elongation and does not alter the steady-state length of microtubule-rich extensions. We then repeated these experiments in HeLa cells, using a Rac inhibitor [32] to compromise lamellipodial formation (Figure 3F). As with S2R+ cells, this did not alter cell length (compare Figure 3F and 3G). To test a possible role for actin-myosin-mediated cortical tension in the regulation of cell length, we also carried out a similar analysis using blebbistatin to inhibit Myosin II [33]. Once again, this did not affect the rate of cell spreading (Figure 3H and 3I) or resting cell length (54.3 [−8, +17] µm for HeLa cells [n = 345] on fibronectin lines, and 50.9 [−4, +11] µm for HeLa cells [n = 52] on non-patterned fibronectin) (Figure 3E). Although surprising, these data concur with previously reported work showing that the final spread area of fibroblasts and HeLa cells is relatively independent of membrane tension [34] and cortical actin [10].
Having failed to identify a role for the actin cytoskeleton, these data prompted us to examine the function of microtubules, which have previously been implicated in cell spreading [20],[24]–[26],[35] and in the generation of an elongated cell shape [36],[37]. Strikingly, in the presence of the microtubule inhibitor colcemid [38], HeLa cells were unable to elongate (Figure 4A and 4D). This effect was reversible since cells re-spread to their characteristic length after the drug was washed out and the microtubule cytoskeleton re-established (Figure 4C, 4E, and 4F). Cell spreading in this system therefore requires microtubules. To gain mechanistic insight into their precise role, we then analysed microtubule organisation at intervals during cell spreading (Figure 5). Cell elongation was accompanied by a progressive polarisation of the microtubule cytoskeleton, as microtubules concentrated on the basal part of the cell (Figure S3) became aligned along the long cell axis (Figure 5B and 5D; quantified in Figure 5F). Strikingly, a similar re-alignment was observed over shorter time scales as individual growing microtubules in cells at steady state became oriented to lie along the long cell axis (Figure 6A), as the result of contacts between growing microtubule plus tips (marked with EB3-GFP) and the ruffling cell margin at the interface between an adhesive and non-adhesive substrate (Figure 6B and 6C). This cortex-induced change in microtubule direction was similar to that previously described in yeast [6], animal cells [20],[39], and plant cells [40].
Interestingly, the fate of each growing microtubule meeting the cortex depended strongly on the angle of contact (Figure 6B, 6C, and 6E), such that microtubule plus ends contacting the cortex at a steep angle (32°–90° from the long line axis) underwent catastrophe, while others contacting the cortex at less of an angle (0°–25°) changed their direction of growth to run parallel to the cell edge. This explains the low probability of finding microtubules aligned at an angle of between 30° and 90° with respect to the long cell axis in Figure 5 (or −30° and −90° according to the notation in Figure 6C). As a result of this angular dependency, the microtubule cytoskeleton reached a highly polarised equilibrium state, with the majority of microtubules running along the cell edge (Figure 6A; quantified in Figure 6F; explained in Figure 6G). Importantly, this angular dependency of microtubule catastrophe/bending resulted in overt polarisation of the microtubule cytoskeleton. We also observed oriented microtubules leading during cell extension (Figure 6D and 6F; Video S1, S2, S3). Taken together these data suggest that microtubules oriented parallel to the adhesive boundary are strong candidates for drivers of cell spreading and the source of intrinsic cell length control.
To test whether the dynamic behaviour of this oriented array of microtubules could define cell length in this system, we generated a simple mathematical model of microtubule-based cell length control (Figures 7 and 8). In contrast with previous models of cell spreading [10],[41],[42], our model was constructed to assess the effects of a polarised array of dynamic microtubules on cell elongation and homeostasis. Based upon our analysis of the cellular distribution of microtubules (Figures 5 and 6), EB3-GFP comets (Figure 6D), and γ-tubulin (Figure S2), microtubules in the model were assumed to grow out from the cell centre towards cell tips (Figure 7A). Cycles of dynamic instability were then implemented using values of growth and catastrophe rates taken from the experimental literature [43]. The interaction between microtubules and the cell cortex was then modelled by assuming that contact (i) increases the microtubule catastrophe rate (by a factor of 16 [43]) and (ii) drives the extension of the cell margin [44]. (Although the mechanism by which this occurs is not specified in the model, we think it likely that it is through the delivery of new material required for local growth [24] rather than through force generation [45],[46].) A slow fixed rate of margin retraction was then implemented, based upon measurements of the rate of retraction of cells on lines in the absence of microtubules (0.4±0.3 µm/min), which may reflect the turnover of material from the cell periphery. Simulations using this simple scheme were found to recapitulate the path of cell elongation and cell length homeostasis (Figure 7A–7C). Furthermore, the model predicts a linear decrease in microtubule density from the cell centre to the cell edge, which was verified experimentally (Figure 7D). Interestingly, the model also revealed that, irrespective of the actual cell length, it is the small number of dynamic microtubules (approximately two) that reach the cell cortex that maintain cell length homeostasis, by countering the tendency of the cell margin to retract (Figures 7C and S4). This is in line with previous data suggesting that a small population of pioneer microtubules is sufficient in some systems to drive forward movement of the cell edge [47].
When we examined the effects of varying the remaining free parameters in the model (Figure 8A–8C), we found that cell length homeostasis (the coefficient of variation in cell length) was marginally sensitive to changes in the value of α, representing the level of microtubule cooperation in the system (Figure 8A–8C), and to changes in Nm, the number of cooperating microtubules. Based on measurements of cell length and microtubule numbers in HeLa cells on lines, we were able to estimate the value of α as 8, implying that individual microtubules cooperate to drive cell spreading. More significantly, an analysis of the effects of varying the other experimentally determined parameters in the model revealed that cell length control critically requires a high value of cB/cI, the ratio of the microtubule catastrophe rate at the cortex (cB) to that in the cell interior (cI). Thus, cell length becomes progressively more variable as the ratio of cB/cI is reduced, e.g., as cB tends to 0 (Figures 8D, 8E, and S4B). By contrast, changes in the microtubule polymerisation rate (vg) induce corresponding changes in cell length in the model without inducing a loss of homeostasis (Figures S4A, 9A, and 9B), causing the system to stabilise at a new steady-state length when the number of microtubule plus ends interacting with the cell cortex returned to the equilibrium value of ∼2 (Figure 9B). This serves as a good test of the likely effects of the addition of a “microtubule inhibitor” on cell length control (Figure 9A and 9B). To test whether this prediction is borne out in experiment, we used an inhibitor of microtubule dynamics [38] to reduce the rate of microtubule polymerisation in HeLa cells on lines of varying width at steady state. After allowing cells 2 h to spread, 40 nM colcemid or an equivalent amount of the carrier DMSO was added to the medium for 30 min (Figure 9C–9E). In the case of colcemid, but not DMSO, this was sufficient to disturb microtubule organisation without causing a complete loss of microtubules (data not shown), and induced active cell shortening (Figure 9C and 9D). As predicted, cells settled down to a new shorter length following this treatment, irrespective of line width (Figure 9C and 9D). We conclude that the dynamic behaviour of the population of longitudinally polarised microtubules plays a key role in homeostatic cell length control.
Previous work has suggested roles for microtubules in the regulation of cell shape in 3-D environments [48]. Therefore, we were prompted to test whether cells exhibit a similar type of cell length homeostasis in a tissue and developmental context. We used the zebrafish neural tube as a simple model system for this analysis for several reasons. First, cells in this tissue are bipolar in form and of similar length to S2R+ and HeLa cells on lines. Second, once formed [16], this tissue is maintained as a stable structure consisting of two parallel columns of highly elongated neuroepithelial cells [49], making reliable measurements of cell length relatively easy. Third, the tissue is amenable to imaging and perturbation experiments using morpholinos. To begin, we tested whether cell length depends upon cell volume in the zebrafish neural tube by arresting cells in the G2 phase of the cell cycle using an established protocol in which a morpholino against the translational start site of the G2/M regulator Emi1 (Emi1-MO) or a control morpholino (Con-MO) is injected into the one-cell embryo [50],[51]. As previously reported, this treatment does not affect cell division until the neural plate stage because of a maternal effect and has very limited cytotoxicity [50],[51]. For consistency across animals, measurements of cell lengths were then made using the neuroepithelium of the hindbrain close to the developing otic vesicle in 19 somite (19s)–stage embryos. At this stage in development the neural tube has not yet inflated its ventricle, and neuroepithelial cells from the left and right sides meet in the middle of the tube, as confirmed by the expression of the polarity protein Par3-GFP, which reveals the apical ends of all cells lining up along the tube midline in both control and emi1 morphant embryos (Figure 10A). This block in cell cycle progression (evident in the loss of pH3 staining in Figure 10A) led to a significant 2.6-fold increase in neuroepithelial cell volume (Figure 10B; Emi1-MO, 9,034.2±3,839.3 µm3, n = 7 cells from three embryos; Con-MO, 3,516.4±608.2 µm3, n = 8 cells from four embryos; t test, p = 0.007) when compared to control-morpholino-injected embryos. The variability in the extent of volume increase observed likely reflects the slight variability of the timing with which the Emi1-MO-induced cell cycle arrest kicks in in different embryos.
Since the vast majority of individual cells in the tissue span the entire width of the neural tube (Figure 10A), we were able to use the width of the tissue at three locations close to the otic vesicle to estimate average cell length. Despite the large difference in cell volumes between control and emi1 morphant embryos (readily visible in embryos labelled with Par3-GFP and H2B-RFP; Figure 10A), the length of neuroepithelial cells remained unaltered (Figure 10C; Con-MO, 44.97±3.66 µm, n = 10; Emi1-MO, 48.11±6.83, n = 10; t test, p = 0.188). This shows that cell length is independent of cell volume in this tissue context as it is in our cell culture models.
To test whether microtubule dynamics are required to define cell length in this system, we visualised microtubule cytoskeletal organisation within the neuroepithelium. Stochastic labelling of cells in the neural tube with GFP fused to the microtubule-associated protein doublecortin (DCX) [52] revealed bundles of parallel microtubules running the entire length of each neuroepithelial cell (Figure 10D), from the apical to the basal limit of the epithelium. To determine whether these microtubules function in cell length control, as was shown for microtubules in cells on micropatterned lines, we added low doses of the microtubule inhibitor nocodazole to these embryos for a period of 30 min. While high doses of microtubule inhibitors lead to a reversible loss of neuroepithelial form as all the cells in the tissue round up (data not shown), this treatment leaves overall tissue architecture intact (Figure 10E). The result was a significant reduction in the width of nocodazole-treated neural tubes when compared to those in DMSO control embryos (Figure 10F), without affecting differences in width that characterise different parts of the tissue. Moreover, as the concentration of nocodazole was increased, the neural tube became progressively narrower (DMSO, 49.37±3.11 µm, n = 8; 5 µg/ml nocodazole, 48.99±2.73 µm, n = 9; 10 µg/ml nocodazole, 46.93±3.29 µm, n = 8; 20 µg/ml nocodazole, 43.77±1.99 µm, n = 7). This analysis suggests that the parallel bundles of microtubules seen spanning the entire width of the epithelium within individual neuroepithelial cells play a critical role in the ability of cells to maintain their length and proper tissue architecture, as they do in homeostatic length control in cells in culture.
Taken together our analysis of hemocyte-derived Drosophila cells and epithelial-derived HeLa cells on micro-contact-printed lines in culture, and of neuroepithelial cells in the developing zebrafish neural tube, identifies a capacity for animal cells to maintain an intrinsically defined length. Cell length control in these systems appears to act independently of cell width and volume. Instead, it relies on the continuing presence of a polarised population of dynamic microtubules that, as a result of interactions with the cell cortex, come to lie parallel to the long axis of cells, running from the centre to either cell tip. Because cell extension depends on microtubules, this oriented array of dynamic microtubules is then in a position to regulate cell length. Somewhat surprisingly, we observed no role for actin-based cortical tension in opposing microtubule-driven cell elongation in our system. This is in line with data from previous studies showing that the final cell spread area is independent of the actin cytoskeleton [10],[24]. Because of this, cell length control is unlikely to reflect a balance of forces between contractile actin filaments and extending microtubule rods—as previously hypothesized for the regulation of animal cell form [22]. In fact, a recent study showed that microtubules do not bear a significant mechanical load in fully spread cells [53]. Based on these observations, it seems likely that microtubules drive cell elongation through the promotion of directional traffic of material from the Golgi to the cell surface [24], rather than through the direct generation of mechanical force itself. Interestingly, microtubule dynamics have also been shown to contribute to (i) the regulation of cell length and cell shape in fission yeast via the regulation of cell transport [54], (ii) axial elongation in plant cells [55], and (iii) the maintenance of spindle length within defined limits [56]. As such, limits to the length of dynamic microtubule-based structures may be a relatively widespread phenomenon in biology.
Recently, a mechanism for length-dependent microtubule depolymerisation was discovered [5], which could help to explain the regulation of processes such as cell length. While this mechanism may be involved in cell length control, our analysis suggests that it may not be necessary to invoke this type of control to understand the maintenance of microtubule-dependent structures within relatively well-defined size limits in all cases, since cell length control can arise as a relatively simple by-product of microtubule-based cell extension if a few simple propositions hold. These propositions are the following: (i) that dynamic microtubules are polarised so that they polymerise along the long cell axis towards cell tips, (ii) that the rate of microtubule catastrophe increases when microtubules reach the cell's ends, and (iii) that microtubules act together to prevent retraction of the cell margin and promote cell elongation, probably through the delivery of material to counter turnover of material at the cell tips. Significantly, a model based on these experimentally well-established assumptions predicts the spreading dynamics we observe on patterned lines, the steady-state microtubule distribution (Figure 7D), the path of cell elongation (Figures 9A, 9B, and S4B), and the effects of a microtubule poison on cell length (Figure 9C and 9D).
Intrinsic regulation of cell length in a tissue and developmental context is inherently difficult to demonstrate unambiguously because of the potentially confounding effects of forces and signals from neighbouring tissues and extracellular matrix. For example, tissue architecture and the dimensions of neuroepithelial cells in the newly formed zebrafish neural tube are likely to be influenced by both extrinsic and cell-intrinsic factors, as cells undergo interdigitation and intercalation across the midline [16],[57]. Moreover, zebrafish neuroepithelial cells achieve their final length through a complex mechanism involving initial overextension and then retraction (data not shown) as they establish a nascent apical domain at the tissue midline [16]. Because tissue architecture in this case is generated via a complex process of self organisation, an intrinsic mechanism that biases cell length may be indispensable to ensure robust organ size and form in the face of variations in the volume of component cells and variations in the movements and differentiation of surrounding tissues [58]. Indeed, here we show that neuroepithelial cell length in vivo is independent of cell volume but, consistent with our findings in vitro, dependent on a population of axially oriented microtubules. As a result, increasing levels of microtubule inhibitors administered over a relatively short period of time cause systematic reductions in epithelial height (Figure 10C). These data help to explain how it is that neuroepithelial height can remain relatively unchanged in ectopic neural tubes that are not situated at the embryonic midline in morphogenetic mutants (compare cells in Figure 4C and 4F in [16]; compare cells in Figure 1G, 1H, and, 1K in [57]).
Similar conclusions can be drawn from earlier studies on the effects of cell cycle arrest on development [59],[60], where it was noted that many ultrastructural features and functional properties of cells were conserved despite dramatic changes induced in cell size. Our study shows that the regulation of neuroepithelial height is one way by which embryos are able to do this. More generally, one could hypothesize that cell length homeostasis is likely to be required in all growing epithelia that need to maintain apical–basal structure despite continual changes in the volumes of their constituent cells.
While the emphasis of this study is cell length homeostasis, it is clear that changes in animal cell length or height are likely to play a critical role during tissue morphogenesis in vivo. During Drosophila gastrulation, for example, epithelial cells are thought to actively contribute to ventral furrow formation by undergoing apical constriction and cell shortening [15], a process that would seem to require orchestrated changes in cell length. Although the mechanism by which this occurs is not understood, it is plausible that microtubules, which are highly polarised along the apical–basal cell axis of Drosophila epithelial cells, play a role, as has been shown in other epithelia [61],[62]. Similarly, the changes in cell length that accompany neuron and myoblast differentiation are brought about by changes in microtubule dynamics [63]–[65]. Thus, many animal cells are likely to need to be able to regulate their optimal steady-state length, e.g., by changing the rate of microtubule polymerisation to alter their resting length whilst preserving length control (Figures 9A and S4A). Conversely, decreasing the susceptibility of microtubules to undergo catastrophe at the cell cortex, e.g., by crosslinking microtubules with Map1A [63], could induce dramatic but relatively unregulated cell elongation (Figure 8E). Because of this, long cells like neurons may rely on additional control systems, like length-dependent regulators of microtubule catastrophe [5] or environmental cues, to reach specific locations during the process of axon pathfinding. An important goal of future research will be to identify how intrinsic length constraints and additional layers of control are altered during animal development to give specific cells and tissues their characteristic forms, and to determine whether the deregulation of cell length control contributes to the loss of tissue homeostasis seen in diseases such as cancer.
A mathematical model was formulated to examine the likely role of microtubules in the control of cell length. The model was based on experimental observations that microtubules aligned along the long axis of the cell, with plus ends towards the cell tips (Videos S1, S2, S3), appear to drive the elongation of cell edges on adhesive lines (Figure 4A–4F) and in cells lacking lamellipodia (Figure 3). Using this as a framework, we constructed a simple stochastic half-cell model of cell elongation driven by a population of parallel dynamic microtubules. Where possible, parameters used to model microtubule dynamics were taken directly from experimental data (see Table 1). We then made a number of simplifying assumptions. First, we assumed that a fixed number of microtubules, Nm, nucleate at the cell centre and grow towards the cell ends at a rate determined by the known rate of microtubule polymerisation, vg, with an experimentally defined cytoplasmic catastrophe rate cI [43], and we made the simplifying assumption that there is no rescue of microtubule growth following catastrophe. Upon reaching the cortex (defined as the region within 3 µm of the cell boundary), microtubule plus ends then act together to promote extension of the cell boundary. We modelled the cooperative effects of n microtubules touching the cortex in driving cell elongation at each time step using the function e−α/n. Because microtubules can drive cell elongation through forces generated by the addition of tubulin subunits [70], through the delivery of new material required for local growth, and/or through local modification of the cell cortex [45],[46], in this study α is assumed to be a free, dimensionless parameter. At the same time, in line with cell biological data, contact with the cortex induces an increase in the rate of microtubule catastrophe cB [43] (See Table 1). Since catastrophe events free up a microtubule nucleation site in the model, the number of growing microtubules remains constant over time. Finally, the term vB was added, based upon experimental data (data not shown), to represent the slow retraction of the cell margin in the absence of microtubules (0.4±0.3 µm/min). After applying known rates of microtubule growth and catastrophe to the model (see Table 1), two free variables remain: α, which governs the cooperative effect of microtubules on the movement of the cell boundary, and Nm, the number of microtubule nucleation sites. In order to compare the results of simulations with experimental data from Figures 1–3, simulations were run in Matlab by using parameter values shown in Table 1, and the position of the boundary was used as a read out of half cell length.
Based on this scheme, given a random number r in the interval [0,1], the stochastic equations that describe the microtubule growth in time are given by(1)Where Lm is the microtubule length, r is a random number in the interval 0≤r≤1, Δt (0.001 min) is the simulation time step, and vm and cm are the microtubule velocity and catastrophe rates, respectively. The stochastic catastrophe event is defined by the random number r and Δt cm. In order to model cell extension, a cell length boundary equation LB is added to the model:(2)where vg is the microtubule growing rate internal to the cell and vB is the cell boundary velocity retraction rate when there are no microtubules crossing the boundary. α governs the cooperative effect of microtubules in promoting cell boundary extension, as defined by the function e−α/|mB(t)|. mI and mB describe internal and boundary microtubules as follows:(3)(4)Finally the general microtubule velocity and catastrophe rate equations are given by(5)(6)where cI and cB are the internal and boundary catastrophe microtubules rates, respectively.
The aim of our model was to quantitatively explore the relationship between microtubule behaviour and cell length control. Significantly, for a wide range of Nm and α values, this simple scheme recapitulated the path of cell elongation and length control seen in observations of cells on lines (Figure 1). As seen in experiments (Figure 1D), the rate of cell elongation in the model diminished over the course of 60 min, leading to a steady-state cell length within a few hours (Figure 7B). To understand the source of cell length homeostasis in these simulations we plotted the number of microtubules contacting the cell cortex over time (Figure 7C). This revealed a steady decrease in the number of microtubules reaching the cell cortex as cells elongate. As cells extend, this number plateaus, reaching a steady equilibrium between cell elongation and cell retraction that maintains cell length over time, which is typically approximately two microtubules—a number that is independent of the number of microtubule nucleation sites and cell length itself (Figure 7C). The model also predicted a linear decrease in microtubule density with distance from the cell centre similar to that measured in cells on lines (Figure 7D).
Significantly, the number of microtubule nucleation sites, Nm, had little impact on the ability of cells to achieve length homeostasis (Figure 8A and 8B; cell length variance is used as quantitative measure of homeostasis), while a moderate level of microtubule cooperation (α>3) was required for a reproducible cell length (Figure 8B and 8C). Above this threshold, while cells were able to maintain a homeostatic cell length irrespective of the specific values of Nm and α, cell length increased with increasing values of Nm and decreased with increasing values of α (Figure 8C).
Although catastrophe rates used in the model were based on experimentally well-defined parameters, we also determined the effects of changing the cortical cB and internal cI catastrophe rates on cell length homeostasis. This revealed that cell length control is gradually lost when cB/cI tends to zero, i.e., as cB values were reduced or cI increased. This is seen by the increase in cell length variance—a quantitative measure of homeostasis (Figures 8D and S4B). It should also be noted that at values of cB close to zero, the number of microtubules at the cell boundary increases to a high steady-state value, driving continuous cell elongation (Figure 8E). Finally, the model was used to test the likely effects of colcemid in this system [38] by altering vg. A reduction in vg leads to a linear reduction in cell length (Figure S4A), as cells re-establish equilibrium with an average of approximately two microtubules contacting the cell cortex per unit time (Figure 9A and 9B).
|
10.1371/journal.ppat.1005018 | Human Immunodeficiency Virus Type 1 Nef Inhibits Autophagy through Transcription Factor EB Sequestration | HIV Nef acts as an anti-autophagic maturation factor through interaction with beclin-1 (BECN1). We report that exposure of macrophages to infectious or non-infectious purified HIV induces toll-like receptor 8 (TLR8) and BECN1 dependent dephosphorylation and nuclear translocation of TFEB and that this correlates with an increase in autophagy markers. RNA interference for ATG13, TFEB, TLR8, or BECN1 inhibits this HIV-induced autophagy. However, once HIV establishes a productive infection, TFEB phosphorylation and cytoplasmic sequestration are increased resulting in decreased autophagy markers. Moreover, by 7 d post-infection, autophagy levels are similar to mock infected controls. Conversely, although Nef deleted HIV similarly induces TFEB dephosphorylation and nuclear localization, and increases autophagy, these levels remain elevated during continued productive infection. Thus, the interaction between HIV and TLR8 serves as a signal for autophagy induction that is dependent upon the dephosphorylation and nuclear translocation of TFEB. During permissive infection, Nef binds BECN1 resulting in mammalian target of rapamycin (MTOR) activation, TFEB phosphorylation and cytosolic sequestration, and the inhibition of autophagy. To our knowledge, this is the first report of a virus modulating TFEB localization and helps to explain how HIV modulates autophagy to promote its own replication and cell survival.
| Under basal conditions, the mammalian target of rapamycin (MTOR) phosphorylates transcription factor EB (TFEB) resulting in its cytoplasmic retention. When MTOR is inhibited, TFEB is dephosphorylated and translocated to the nucleus where it increases autophagy and lysosomal gene expression. As human immunodeficiency virus type 1 (HIV) Nef acts as an anti-autophagic maturation factor through interaction with beclin-1 (BECN1), we investigated the role of Nef and TFEB in the modulation of autophagy during HIV infection of human macrophages. We found that upon exposure to HIV, macrophages elicited an autophagic response through a toll-like receptor 8 (TLR8) and BECN1 dependent dephosphorylation and nuclear translocation of TFEB. However, once HIV infection is established, phosphorylation and cytoplasmic sequestration of TFEB as well as autophagy revert to pre-infection levels. Moreover, this reversion is dependent upon the presence of HIV Nef. Collectively, the data suggests that the interaction between HIV and TLR8 serves as a signal for autophagy induction that is dependent upon the dephosphorylation and nuclear translocation of TFEB. Once HIV establishes a productive infection, Nef binds BECN1 resulting in MTOR activation, TFEB phosphorylation and cytosolic sequestration and the inhibition of autophagy.
| As an obligate intracellular parasite, human immunodeficiency virus type 1 (HIV) survival is dependent upon its ability to exploit host cell machinery for replication and dissemination, and to evade intrinsic cellular processes and defenses that may limit viral replication and pathogenesis including macroautophagy (hereafter referred to as autophagy) [1]. Autophagy is a degradation pathway whereby cytosolic double membrane-bound compartments termed autophagosomes engulf and sequester cytoplasmic constituents such as sub-cellular organelles and microbial pathogens. These autophagosomes then fuse with lysosomes (organelles that contain an array of hydrolytic enzymes capable of degrading almost any biomolecule) forming autophagolysosomes (autolysosomes), resulting in the degradation of the engulfed components. Evidence of integrated and co-regulated roles of lysosomes and autophagosomes has emerged from the discovery of an overarching lysosomal regulatory gene network (CLEAR, Coordinated Lysosomal Expression and Regulation) and its master regulator, the basic helix-loop-helix leucine zipper transcription factor EB (TFEB). During starvation, cells activate a transcriptional program coordinated by TFEB that controls all major steps of the autophagic pathway, including autophagosome formation, autophagosome-lysosome fusion, and substrate degradation [2]. In resting cells, mammalian target of rapamycin (MTOR) complex 1 (MTORC1) is active and phosphorylates Ser142 and Ser211 of TFEB that results in retention of the transcription factor in the cytoplasm through binding of 14-3-3 proteins that occlude a nuclear localization sequence thereby promoting the cytoplasmic sequestration of TFEB [3–5]. When MTORC1 is inhibited or inactivated, the balance shifts towards dephosphorylation of Ser142 and Ser211 resulting in diminished interactions between TFEB and 14-3-3 proteins that reveals the nuclear localization sequence leading to nuclear accumulation of TFEB and the expression of autophagosomal and lysosomal proteins [3, 5]. Thus, TFEB is a regulator of autophagic clearance and is at the crossroads of the regulatory mechanisms that coordinate both the autophagy and lysosomal pathways.
Of the more than 35 human autophagy-associated genes currently known to be involved in autophagy, ten are now known to be essential for HIV replication [reviewed in 1]. However, although HIV may require the early stages of autophagy, it must control the antiviral proteolytic and degradative late stages of autophagy to avoid its degradation. The current data suggest that HIV has developed mechanisms to inhibit autophagic degradation involving the HIV negative regulatory factor (Nef) [6]. In HIV-infected macrophages, Nef inhibits the proteolytic stages of autophagy by binding to amino acids 267–284 in the beclin-1 (BECN1) evolutionarily conserved domain [7]. This is the same region that is necessary and sufficient for BECN1 to bind glioma-associated oncogene pathogenesis-related 2 (GLIPR2) [7], a protein that associates with lipid rafts at the cytosolic leaflet of the Golgi membrane [8] and that negatively regulates autophagy by sequestering BECN1 to the Golgi complex [7]. Despite the down-modulation of autophagy by HIV, inducers of autophagy including 1α,25-dihydroxycholecalciferol [9, 10], amino acid starvation [11], hydroxamate histone deacetylase inhibitors [12], sirolimus [7, 9], toll-like receptor (TLR) 8 ligands [13], romidepsin [12], and a cell-permeable autophagy-inducing peptide termed Tat–beclin (derived from the region of BECN1 that interacts with HIV Nef and conjugated to the basic region of HIV Tat) [7], overcome the imposed phagosome maturation block leading to inhibition of viral replication.
As HIV Nef acts as an anti-autophagic maturation factor through interaction with BECN1, we investigated the role of Nef and TFEB in the modulation of autophagy during HIV infection of macrophages. The present data suggest that the interaction between HIV and TLR8 serves as a signal for autophagy induction that is dependent upon the dephosphorylation and nuclear translocation of TFEB. Once HIV establishes a productive infection, Nef inhibits autophagy by binding BECN1 resulting in TFEB phosphorylation and cytosolic sequestration. These findings, to our knowledge, are the first that report a virus modulating TFEB localization and help to explain how HIV modulates autophagy to promote its own replication and cell survival.
Whereas exposure of primary human macrophages to HIV envelope proteins has no autophagy inducing effect, exposure to MOLT-4 cells chronically infected with HIV induces autophagy by day 3 post-co-culture [14]. However, the autophagic status of macrophages after exposure to purified HIV virions is unknown. Therefore, the effect of HIV on autophagy induction in human macrophages was determined using RNase/DNase I treated virus purified through an iodixanol velocity gradient (purified HIV), which effectively separates extracellular proteins and microvesicles from the virus (S1 Fig) [15]. During autophagy, cytosolic microtubule-associated protein 1 light chain 3 beta (LC3B)-I is converted to LC3B-II by a ubiquitin-like system that involves autophagy related (ATG) 7, ATG3 and the ATG12–ATG5 complex. The ATG12–ATG5 complex ligates LC3B-II to the nascent autophagosome membrane through phosphatidylethanolamine with the LC3B-II associated with the inner membrane degraded after fusion of the autophagosome with lysosomes. Therefore, the conversion of LC3B-I to LC3B-II and its turnover is an indicator of autophagy induction and flux [16]. Exposure of macrophages to purified HIV led to a significant dose-dependent increase in LC3B-II after 24 h (Fig 1A) in the absence of significant cytotoxic effects (P > 0.05; S2A Fig).
To verify that the increase in LC3 lipidation represented increased autophagic flux rather than an accumulation of LC3B-II, the degradation of the polyubiquitin-binding protein sequestosome 1 (SQSTM1) was also quantified. Inhibition of autophagy leads to an increase in SQSTM1 protein levels while autolysosomes degrade SQSTM1- and LC3-positive bodies during autophagic flux [17]. Purified HIV induced a significant dose-dependent decrease in SQSTM1 protein levels corresponding to the stimulation of autophagic flux at 24 h post-infection (Fig 1A). To confirm that the decreased SQSTM1 levels result from enhanced degradation via autophagy and not through diminished transcription, autophagosome degradation was inhibited with the lysosomal protease inhibitor pepstatin A. Both SQSTM1 and LC3B-II were significantly increased in the presence of pepstatin A indicative of autophagic flux (Fig 1B) in the absence of significant cytotoxic effects (P > 0.05; S2B Fig).
In order to infect productively a target cell, HIV envelope protein gp120 binds to CD4, triggering conformational changes in gp120 that ultimately leads to the fusion of the viral and target cell membranes allowing entry of the viral capsid. Within uninfected CD4+ T cells, the fusogenic activity of gp41 induces autophagy [18] leading to the induction of apoptosis [19]. In contrast, uninfected or infected macrophages do not undergo Env-mediated autophagy or apoptosis [14]. In addition to this route of entry, HIV can also enter macrophages through CD4-independent macropinocytosis [20] or a macropinocytosis-like mechanism (the pathway of HIV endocytic entry in macrophages [PHEEM]) [21]. Following entry through a CD4-independent pathway, the uridine rich HIV long terminal repeat (LTR) single-stranded RNA, which contains multiple pathogen-associated molecular patterns (PAMPs), can be recognized by the pattern recognition receptor (PRR) TLR8 expressed in macrophage endosomes [22, 23]. Previously, we demonstrated that ssRNA40, a GU-rich ssRNA derived from the HIV LTR, induces autophagy in human macrophages through a TLR8-dependent mechanism involving vitamin D, and the expression of both the vitamin D (1,25D3) receptor, and cytochrome P450, family 27, subfamily B, polypeptide 1 (CYP27B1), which 1α-hydroxylates the inactive form of vitamin D3, 25-hydroxycholecalciferol (25D3), into the biologically active metabolite 1,25D3, and (VDR) [13]. Therefore, we investigated whether productive infection was required for the induction of autophagy using 2,2′-dithiodipyridine (AT-2)-treated HIV. AT-2 inactivates the infectivity of retroviruses by covalently modifying the nucleocapsid zinc finger motifs (S3 Fig). Exposure of macrophages to AT-2-inactivated HIV for 24 h led to a significant increase in LC3B-II (P = 0.0045) and significant degradation of SQSTM1 (P = 0.006). Interestingly, there was no significant difference in either LC3B lipidation or SQSTM1 between AT-2-inactivated HIV and exposure to infectious HIV indicating that productive infection is not required for the HIV-mediated induction of autophagy (P > 0.05; Fig 1C). To determine the role of TLR8, RNA interference (RNAi) of TLR8 was employed. TLR8 silencing (Fig 2A) significantly inhibited HIV-mediated LC3B lipidation and degradation of SQSTM1 (Fig 2B) in the absence of significant cytotoxic effects (P > 0.05; S2D Fig) suggesting that TLR8 is the mediator of HIV-induced autophagy in macrophages. Although we observed an increase in LC3B lipidation and an increase in SQSTM1 degradation, ingestion of pathogens through TLR1/2, TLR2/6, and TLR4 can trigger the recruitment of LC3B-II to single-membrane phagosomes in a process termed LC3-associated phagocytosis (LAP) [24]. As opposed to canonical autophagy, LAP is the receptor-mediated internalization of extracellular cargo that occurs without the formation of a double membrane. The receptor triggers the ligation of LC3 to the phagosome. These LC3-positive SQSTM1-negative phagosomes then fuse with lysosomes and rapidly mature into a phagolysosomes. The pre-initiation complex that is required for autophagy is dispensable for LAP, as LC3B-II deposition at the phagosome proceeds normally in the absence of RB1-inducible coiled-coil 1, ATG13 and unc-51 like autophagy activating kinase 1 (ULK1) proteins [25]. The ability of HIV and TLR8 ligands to initiate LAP is unknown. Therefore, we analyzed whether HIV exposed macrophages contain more SQSTM1 and LC3B dual positive autophagosomes or LC3B-positive SQSTM1-negative phagosomes harboring HIV particles using confocal immunofluorescence microscopy (Fig 2C). We observed SQSTM1 and LC3B dual positive puncta that were also positive for HIV (cream pixels), but failed to observe the presence of LC3B-positive SQSTM1-negative puncta harboring HIV (orange pixels). We then investigated whether HIV or ssRNA40 induces TLR8-mediated LAP using RNAi for ATG13. Silencing of ATG13 (Fig 2D) abrogated LC3B lipidation and SQSTM1 degradation following ssRNA40 (P < 0.0005) and HIV (P < 0.005) exposure for 24 h suggesting that the autophagy pre-initiation complex is required for LC3B lipidation in response to TLR8 triggering (Fig 2E).
TLR8 is both phylogenetically and structurally similar to TLR7, activates similar signaling pathways, and is located within endosomes. The signaling pathway of TLR7 is dependent upon endosomal acidification (maturation) [26, 27]. Bafilomycin A1 is an inhibitor of the vacuolar H+ ATPase, effectively inhibiting endosomal acidification (maturation) and thus the signaling pathways of endosomal TLR3, TLR7 and TLR9 [26–28]. Importantly, although bafilomycin A1 blocks the fusion of autophagosomes with lysosomes, leading to an accumulation of autophagosomal structures it has no direct effect on the conversion of LC3B-I to LC3B-II. For instance, sirolimus, an inhibitor of MTOR that initiates autophagy independently of endosome acidification, induces significant LC3B-II accumulation in the presence of bafilomycin A1 (P = 0.015; Fig 2F) indicative of induced autophagy but arrested flux [29], in the absence of significant cytotoxic effects (P > 0.05; S2F Fig). However, pretreatment of macrophages with bafilomycin A1 resulted in the inhibition of HIV and ssRNA40 induced LC3B lipidation (P > 0.2; Fig 2F). Moreover, bafilomycin A1 also inhibited the increase in TLR8 mediated CYP27B1 (P > 0.28) and VDR (P > 0.37) expression, downstream effectors of TLR8 signaling that are required for TLR8 mediated autophagy induction [13]. Collectively, these results demonstrate that the induction of autophagy in macrophages by HIV does not require productive infection, and is mediated through a TLR8 signaling pathway that requires endosomal maturation.
We next assessed the autophagic status of primary macrophages after long-term infection with replication competent virus. There was a significant increase in LC3B lipidation at both 24 h and 72 h post-infection (P < 0.05; Fig 3A). By 5 d post-infection, although still significantly increased, LC3B lipidation had appreciably decreased and by 7 d was the same as the mock-infected controls (P > 0.05). Similarly, HIV infection induced a significant decrease in SQSTM1 protein levels (P < 0.001) corresponding to the stimulation of autophagic flux at 24 h post-infection. SQSTM1 protein levels became progressively greater from 3 d to 5 d post-infection and by 7 d post-infection were the same as the mock-infected controls (P > 0.05; Fig 3A). When autophagosomes are formed, LC3B redistributes from a soluble diffuse cytosolic pattern to an insoluble autophagosome-associated vacuolar pattern [30, 31] allowing the quantification of autophagosome-associated LC3B-II in human macrophages using saponin resistance and flow cytometry [30]. Staining for endogenous LC3B in saponin washed macrophages revealed that the percentage of cells containing a saponin resistant fraction was significantly increased at both 24 and 72 h post-infection (P < 0.001; Fig 3B). By 5 d post-infection, the number of cells expressing saponin resistant fractions had decreased, but was still significant (P = 0.0002) and by 7 d was the same as the mock-infected controls. Despite the downregulation of autophagy markers observed by 7 d post-infection, CYP27B1 and VDR were still significantly upregulated indicating that TLR8 is still sufficiently engaged at these late time posts post-infection (Fig 3C) in the absence of significant cytotoxic effects (P > 0.05; S2G Fig).
To investigate the role of TFEB in regulating autophagy activation upon exposure to HIV, we initially examined TFEB localization using immunoblotting (Fig 4A). The results show that TFEB localizes predominantly in the cytoplasm of uninfected macrophages. Moreover, almost all TFEB ran at a higher molecular size than in the HIV-exposed samples suggesting that a substantial fraction of TFEB is phosphorylated under basal conditions. Exposure to purified HIV for 24 h led to a significant and dose-dependent dephosphorylation and activation of TFEB as monitored by its more rapid mobility in sodium dodecyl sulfate polyacrylamide gel electrophoresis, and increased nuclear accumulation (Fig 4A). We then assessed the effect of long-term productive HIV infection on TFEB localization. The dephosphorylation and activation of TFEB lasted until at least 72 h post-infection and by 5 d post-infection, although levels were elevated, TFEB was localized predominantly to the cytoplasmic fraction. By 7 d TFEB localization was similar to the mock-infected controls (Fig 4B). To confirm further the nuclear translocation of TFEB, we monitored TFEB sub-cellular localization using confocal immunofluorescence microscopy. Macrophages were cultured in the presence of HIV and the sub-cellular distribution of TFEB was evaluated using 4',6-diamidino-2-phenylindole nuclear staining and an anti-TFEB antibody (Fig 4C). In untreated macrophages, TFEB localized predominantly to the cytoplasm whereas in HIV exposed macrophages TFEB translocated to the nucleus by 72 h. However, by 6 d post-infection, this effect had dissipated and TFEB localization was again similar to the mock-infected controls (Fig 4C).
As permissive HIV infection is not required for the induction of autophagy (Fig 1C), we investigated whether productive infection was required for the dephosphorylation and nuclear translocation of TFEB at the early time points. Treatment with AT-2-inactivated purified HIV led to the dephosphorylation and nuclear localization of TFEB by 24 h indicating that productive infection was not necessary (Fig 4D). Furthermore, silencing of TLR8 (Fig 4E) abrogated the HIV-mediated dephosphorylation and nuclear translocation of TFEB (Fig 4F). To assess the role of TFEB in HIV-mediated autophagy, macrophages were transduced with short hairpin RNA (shRNA) specific to TFEB, followed by exposure to HIV. TFEB silencing (Fig 5A) significantly reduced both the lipidation of LC3B and the degradation of SQSTM1 in macrophages post-HIV infection (Fig 5B) in the absence of significant cytotoxic effects (P > 0.05; S2I Fig). We then tested whether TFEB overexpression also regulated autophagy in macrophages using a lentivirus overexpressing TFEB. Transient TFEB overexpression (Fig 5C) significantly increased LC3B lipidation and SQSTM1 degradation post HIV exposure indicating that TFEB enhances HIV induced autophagic flux in macrophages (Fig 5D).
Finally, we analyzed the transcriptional activity of TFEB post-HIV exposure. For this, we exposed macrophages to both infectious HIV and AT-2-inactivated HIV and assessed the expression of the known TFEB targets ATG9B, UV radiation resistance associated gene (UVRAG) (both autophagy genes), and mucolipin 1 (MCOLN1) (a lysosomal gene) after 24 h using qRT-PCR. Both infectious HIV and AT-2-inactivated HIV increased the transcription of ATG9B, UVRAG, and MCOLN1 suggesting activation of TFEB by HIV does not require productive infection (Fig 6A). Moreover, when TLR8 was silenced, expression of these genes post-HIV exposure was similar to the mock-infected controls (Fig 6C). Collectively, these data suggest that exposure of macrophages to HIV induces TLR8-dependent TFEB dephosphorylation and nuclear translocation that induces autophagy, and that this is not dependent upon a productive infection.
Autophagy is well integrated into the innate immune system with PAMP induced PRR signaling activating autophagy [32]. For example, TLR4 signaling leads to ubiquitination of BECN1 by E3 ubiquitin protein ligase tumor necrosis factor receptor-associated factor 6 (TRAF6) which releases it from its inhibitor, B cell lymphoma 2 (BCL2) [33]. In the context of HIV, TLR8 signaling by the HIV LTR RNA stimulates enhanced binding of BECN1 to phosphoinositide-3-kinase (PIK3C3) forming the PIK3C3 kinase complex, which is essential for the induction of autophagosome formation at the vesicle elongation step [13]. In both cases, BECN1 is essential for the induction of autophagy. Therefore, we investigated whether the dephosphorylation and nuclear translocation of TFEB post-HIV infection was dependent upon BECN1. Macrophages transduced with shRNA specific to BECN1 were exposed to HIV. As expected, BECN1 silencing significantly inhibited autophagic flux as measured by significant reductions in both the lipidation of LC3B and the degradation of SQSTM1 in macrophages post-HIV infection (Fig 7B). BECN1 silencing also abrogated TFEB dephosphorylation and nuclear localization at all time points (Fig 7C).
HIV, in addition to using basal autophagy for its own replication [1], utilizes the Nef protein to protect itself against autophagic degradation [6]. Nef acts as an anti-autophagic maturation factor through interaction with BECN1 [6], and is required for efficient viral replication and HIV pathogenicity. Therefore, we investigated whether Nef was responsible for the down regulation of autophagy observed during permissive HIV infection. Both complete HIV (HIVNL(AD8)) and Nef deleted HIV (HIVNL(AD8)ΔNef), significantly increased LC3B lipidation and SQSTM1 degradation at 24 h and 72 h post-HIV-exposure (Fig 8A). As was the case for HIVBa-L, the levels of LC3B-II and SQSTM1 in HIVNL(AD8) treated cells were similar to that found in the mock infected controls by 5 d post-exposure and this trend continued through 10 d post-exposure. In contrast, macrophages infected with the Nef deleted HIVNL(AD8)ΔNef maintained significantly greater LC3B lipidation and SQSTM1 degradation at all time points (Fig 8A). Importantly, both HIVNL(AD8) and HIVNL(AD8)ΔNef demonstrated extracellular p24 release over the 10 d infection protocol indicating replication competent virus (S4A Fig). These data suggest that once HIV establishes a productive infection, it inhibits autophagy through a Nef-dependent mechanism. Based on our observations that TLR8 agonists inhibit HIV replication through the induction of autophagy, and that silencing TLR8 inhibits HIV-mediated autophagy, we sought to determine whether the inhibition of HIV-induced autophagy would rescue viral replication of Nef deficient HIV at later time points. Silencing of TLR8 resulted in a marked increase in the release of HIV p24 antigen at later time points by both HIVNL(AD8) and HIVNL(AD8)ΔNef (P < 0.05; S4B Fig).
As the dephosphorylation, activation and nuclear translocation of TFEB was required for the induction of autophagy post-HIV exposure, we examined the effect of HIVNL(AD8)ΔNef on TFEB localization using immunoblotting. Unlike exposure to wild-type HIV, which resulted in the transient dephosphorylation and nuclear translocation of TFEB, infection of macrophages with HIVNL(AD8)ΔNef resulted in the dephosphorylation and increased nuclear translocation of TFEB by 24 h post-infection that lasted throughout the experiment until cells were terminated at 10 d post-infection (Fig 8B). These findings indicate that Nef is required to inhibit the TLR8-mediated induction of autophagy that occurs through subsequent rounds of HIV infection at later time points. This is of key significance as the Nef-dependent actions of inhibiting autophagosome formation and preventing their maturation to autolysosomes [6] may spare the virus from early degradation.
The interaction between viruses and autophagy can be bi-directional and may function to both degrade viruses and/or promote viral replication. The known antiviral effects of autophagy include virophagy (the degradation of cytoplasmic viral constituents), the activation of innate and adaptive immunity (through the delivery of viral antigens to endosomal TLRs or major histocompatibility complex class I and II, respectively), and promotion of cell survival. The positive-strand RNA viruses dengue virus, poliovirus, and hepatitis C virus are good examples of viruses for which autophagy promotes and facilitates viral replication, all of which utilize the maturation of autophagosomes for their replication [34]. In the present study, we investigated the mechanisms by which HIV infection influences autophagy during primary infection of human primary macrophages. It was previously demonstrated that the fusogenic function of HIV gp41 was sufficient to trigger autophagy in CD4+ T cells, but not in macrophages [14, 18], and that ssRNA40 derived from the HIV LTR induces autophagy in macrophages through TLR8 in the absence of any other viral antigen [13, 32]. In the present study, we identify that HIV induces autophagy in macrophages through a mechanism that is dependent upon endosomal acidification, TLR8, ATG13, BECN1, and the dephosphorylation and nuclear translocation of TFEB. Moreover, we demonstrate that HIV induces autophagy in macrophages independently of viral replication. HIV is not alone in this respect. Vesicular stomatitis Indiana virus protein G induces virophagy independently of viral replication through interactions with TLR7 on Drosophilia cells [35], and nonvirulent measles virus hemagglutinin induces autophagy through the golgi-associated PDZ and coiled-coil motif containing protein (GOPC) and BECN1 through CD46 on the surface of HeLa cells [36]. TLR signaling reduces the binding of BECN1 to BCL2 via the TLR adaptor proteins myeloid differentiation primary response 88 (MYD88) and toll-like receptor adaptor molecule 1 (TICAM1) by recruiting BECN1 into the TLR-signaling complex leading to the inhibition of MTOR and the induction of autophagy [37]. Importantly, exposure of macrophages to HIV resulted in TFEB-dependent autophagy activation in the absence of apoptosis or cell death, suggesting that macrophages respond to HIV by activating virophagy.
During HIV infection, the expression of TLR8 within peripheral blood monocytes decreases with disease progression. Moreover, monocytes from HIV-infected individuals produce less tumor necrosis factor following TLR8 activation than those from uninfected individuals while successfully inhibiting HIV infection [38]. These monocyte responses are negatively correlated with CD4+ T cell numbers and positively associated with HIV viral load [39]. The ability of cells to respond strongly to a TLR8 agonist in the presence of high HIV viremia suggests that ongoing chronic immune activation may be continuously driven by HIV-encoded PAMPs. However, tolerance is not induced towards TLR8 agonists [39, 40]. Persistent immune activation during HIV infection contributes to the pathogenesis of disease by disturbing the functional organization of the immune system with induction of high levels of cytokines and chemokines. Therefore, chronic stimulation of the innate immune system by TLR ligands may result in the chronic production of proinflammatory cytokines which drive disease progression through generalized immune activation [41]. Indeed, HIV induces both pro-interleukin-1 beta (IL1B) expression and its subsequent cleavage into bioactive IL1B through NLRP3 inflammasome activation in monocytes and macrophages in an infection-independent process that requires clathrin-mediated endocytosis and recognition of the viral ssRNA by TLR8 [42, 43]. Supporting this model is the association of a single-nucleotide polymorphism in TLR8 (TLR8 A1G; rs3764880) which confers a significant protective effect against HIV disease progression [44].
Cytoplasmic TFEB is located both in the cytosol and on the lysosomal surface, where it interacts with MTORC1 and the lysosomal nutrient sensing (LYNUS) machinery [4]. Our findings are consistent with the model in which inhibition, rather than activation, of MTORC1 induces TFEB nuclear translocation. Indeed, pharmacological inhibition of MTOR using Torin 1, chloroquine, Sa1A, and transfection with mutant Rag proteins all result in the nuclear accumulation of TFEB [4]. Interestingly, other members of the basic-helix–loop–helix family of transcription factors, such as microphthalmia-associated transcription factor (MITF) and transcription factor binding to immunoglobulin heavy constant mu enhancer 3 (TFE3), the sequences of which are closely related to TFEB, seem to be regulated by a similar mechanism [3]. It will be interesting to investigate whether HIV or TLR signaling affects these proteins.
We found that in the absence of Nef the HIV-induced TFEB nuclear translocation and induction of autophagy was present at 10 d post-infection. We also found that Nef-deficient HIV replicated less efficiently. It is likely that Nef-deficient HIV replicates less efficiently due to its inability to overcome autophagic degradation. Nef plays a major role in the inhibition of the proteolytic stages of autophagy in macrophages by binding and sequestering BECN1 through its 174DD motif [6]. Although this motif is required for CD4 downregulation and interactions with the V1 domain of the vacuolar H+-ATPase [45], it is unlikely that it influences H+-ATPase assembly or activity thereby inhibiting autophagosome acidification or autophago-lysosome fusion as endosome acidification is independent of Nef [46]. However, as Nef binds the evolutionarily conserved domain of BECN1 (the same region that allows GLIPR2 to bind and sequester BECN1, thereby suppressing autophagy [7]), it is possible that Nef negatively regulates autophagy by sequestering BECN1. Autophagy initiation is thought to be strictly dependent upon phosphatidylinositol 3-phosphate (PtdIns(3)P) synthesis by PIK3C3, in complex with BECN1 at the trans-Golgi network. However, whereas all cellular BECN1 is associated with PIK3C3, 50% of PIK3C3 is not associated with BECN1, and is localized to endosomes [47]. Moreover, PtdIns(3)P has been implicated in the regulation of autophagy upstream of MTOR via PIK3C3 [48]. Therefore, by sequestrating BECN1, Nef may promote the localization of free PIK3C3 to endosomes, and in our amino acid rich environment, activate MTOR [48]. This phosphorylates TFEB and inhibits autophagy while simultaneously stimulating mRNA translation through the phosphorylation of RPS6KB1 and EIF4E1B. Although additional evidence is needed to support this model, consistent with this hypothesis, BECN1 silencing also inhibited the dephosphorylation and nuclear translocation of TFEB upon HIV exposure, indicating that MTOR had not been inactivated. HIV is not alone in its ability to suppress autophagy. As viruses have evolved under pressure from autophagic degradation within their eukaryotic hosts, it is not surprising that many have evolved strategies to circumvent autophagy, and a striking number of them target BECN1. Human cytomegalovirus TRS1 protein [49], African swine fever virus A179L protein [50], herpes simplex virus type 1 ICP34.5 protein [51], human herpesvirus 8 orf16 protein [52], and murid herpesvirus 68 M11 protein [53] all bind BECN1 and block autophagosome biogenesis. In contrast to the DNA viruses that inhibit autophagosome generation, RNA viruses seem to stabilize autophagosomes by preventing their degradation. For instance, the influenza A virus M2 protein binds BECN1 and inhibits autophagosome maturation [54]. The current data suggest that HIV, through Nef, falls into both categories, inhibiting both autophagosome biogenesis and maturation.
The potent inhibitory effects of autophagy on HIV replication [1, 7, 9, 12, 13] combined with the ability of HIV to inhibit autophagy serve to illustrate its importance in the cellular antiviral response [1, 6, 11]. To our knowledge, this is the first report demonstrating viral regulation of the autophagy master regulator TFEB. Understanding how HIV and other viruses such as influenza and herpes viruses inhibit autophagy may lead to the development of broad-spectrum antiviral drugs that restore autophagy through pharmacological means during viral infection with the aim of eliminating the virus. In the case of HIV, this is both attractive and novel as autophagy works at the host cellular level to improve intracellular killing of both replicating and non-replicating HIV while resistance is unlikely to develop. Dissecting the molecular mechanisms by which HIV utilizes autophagy has the potential to lead to the identification of novel drug candidates to treat HIV infection.
Venous blood was drawn from HIV seronegative subjects using a protocol that was reviewed and approved by the Human Research Protections Program of the University of California, San Diego (Project 09–0660) in accordance with the requirements of the Code of Federal Regulations on the Protection of Human Subjects (45 CFR 46 and 21 CFR 50 and 56). Written informed consent was obtained from all blood donors prior to their participation.
Monocyte derived macrophages were generated from whole blood of HIV seronegative donors as previously described [13]. All experiments were performed in RPMI 1640 supplemented with 10% (v/v) charcoal/dextran treated, heat-inactivated fetal bovine serum (FBS; Gemini Bio-Products), 10 ng/mL macrophage colony stimulating factor (Peprotech), and 40 ng/mL 25-hydroxycholecalciferol (Sigma) (growth media). Sirolimus and bafilomycin A1 were obtained from Sigma and LC Laboratories respectively. LyoVec, ssRNA40, and ssRNA41 were obtained from Invivogen. Cell death was estimated using the lactate dehydrogenase (LDH) Cytotoxicity Detection KitPLUS (Roche).
Lentiviral transduction of macrophages with MISSION pLKO.1-puro lentiviral vectors containing shRNAs targeting ATG13 (SHCLNV-NM_ 014741/TRCN0000172507), BECN1 (SHCLNV-NM_003766/TRCN0000033551 and TRCN0000299864), TFEB (SHCLNV-NM_007162/TRCN0000013108), TLR8 (SHCLNV-NM_138636/TRCN0000359320), or scrambled non-target negative control (SHC002V) was performed according to the manufacturer's protocol (Sigma). Macrophages were transduced with non-specific scrambled shRNA (shNS) or target shRNA and selected for using puromycin (Gibco). Five days later, cells were analyzed for target gene silencing and used in experiments.
HIVBa-L (HIV) was obtained through the NIH AIDS Research and Reference Reagent Program from Dr. Suzanne Gartner and Dr. Robert Gallo [55, 56] and was expanded and concentrated as previously described [57]. Virus was then subjected to a 6 to 18% iodixanol velocity gradient in 1.2% increments using OptiPrep (60% [w/v] iodixanol; Sigma) diluted in DPBS essentially as previously described [15]. Briefly, supernatants were laid over the gradient and centrifuged for 1.5 h at 37,500 rpm (250,000 × g at rmax) in an SW41 Ti rotor using an L8-70M ultracentrifuge (both Beckman Coulter). Fourteen gradient fractions were collected and analyzed for both total protein and HIV p24 content by SDS-PAGE and immunoblot analysis. HIV titers were determined on phytohemagglutinin-P-stimulated peripheral blood mononuclear cells (PBMC) as described previously using the Alliance HIV p24 antigen ELISA (Perkin Elmer) [57] and multiplicity of infection confirmed using TZM-bl cells obtained through the NIH AIDS Research and Reference Reagent Program, from Drs John C. Kappes and Xiaoyun Wu and Tranzyme Inc. [58]. R5-tropic, replication-competent HIV-1 strain HIVNL(AD8) and its derivative Nef deleted mutant HIVNL(AD8)ΔNef were generated by transient transfection of HEK293T cells separately with pNL(AD8) [59] or pNL(AD8)ΔNef [60] (both kind gifts from Olivier Schwartz, Pasteur Institute, France) using the calcium phosphate method [61]. Virus was harvested at 48 and 72 h post-transfection, filtered through a 0.22 μm polyethersulfone filter (Millipore), and purified as described above. For inactivation with 100 μmol/L AT-2 (Sigma), HIV was treated for 1 h at 37°C. For RNase treatment virus stock was resuspended in 1 mL 10 mmol/L 2-amino-2-(hydroxymethyl)-1,3-propanediol hydrochloride (1:1; pH 7.4), 100 mmol/L NaCl and treated with 80 U RQ1 DNase I (Promega) and 10 U RNase I (Ambion) for 1 h at 37°C. At the conclusion of AT-2 and/or RNase/DNase I treatments, agents were removed by ultrafiltration at 4°C using a Vivaspin 2 with a 300 kDa cutoff (Sartorius Stedim) followed by purification as described above. Control virus preparations were sham treated and processed in parallel with inactivated samples. Mock infection preparations were prepared from uninfected IL2 treated PBMC supernatants and processed in parallel with HIV stocks. For all procedures, frozen virus stocks were quickly thawed at 37°C in a water bath and cells exposed to 90 μL of a 106.2 TCID50/mL HIV per 105 cells for 3 h, washed then incubated in growth media for the times indicated.
The following antibodies were used: β-actin (ACTB; AC-74), TLR8 (4C6), VDR (N-terminal) (all Sigma), BECN1 (#3738), Histone H3 (H3; D1H2) (both Cell Signaling), LC3B (NB100-2220; Novus Biologicals), TFEB (Bethyl Laboratories), SQSTM1 (ab56416; Abcam), CYP27B1 (H-90; Santa Cruz Biotechnology), and HIV p24 (P131; Abcam). Whole cell lysates were prepared using 20 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, 1 mmol/L ethylenediaminetetraacetic acid (both Gibco), 150 mmol/L NaCl, 1% (v/v) 4-(1,1,3,3-tetramethylbutyl)phenyl-polyethylene glycol (both Sigma) and 1% (v/v) Halt protease and phosphatase inhibitor cocktail (Thermo Scientific). The NE-PER nuclear and cytoplasmic extraction reagent kit supplemented with 1% (v/v) Halt protease and phosphatase inhibitor cocktail (both Thermo Scientific) was used for cell lysis and extraction of separate cytoplasmic and nuclear protein fractions. For immunoblot analyses, cell lysates were resolved using 2-[bis(2-hydroxyethyl)amino]-2-(hydroxymethyl)propane-1,3-diol buffered 12% polyacrylamide gel (Novex) and transferred to low fluorescence 0.2 μm pore-size polyvinylidene difluoride membranes (Thermo Scientific), followed by detection with alkaline phosphatase tagged secondary antibodies (Invitrogen) and 0.25 mmol/L disodium 2-chloro-5-(4-methoxyspiro[1,2-dioxetane-3,2′-(5-chlorotricyclo[3.3.1.13.7]decan])-4-yl]-1-phenyl phosphate supplemented with 5% (v/v) Nitro-Block II (both Applied Biosystems). Relative densities of the target bands compared to the reference bands (ACTB for total lysates and cytoplasmic fractions and H3 for nuclear fractions) were analyzed using ImageJ (NIH). Each sample was normalized to the vehicle then log2 transformed.
mRNA quantification was measured by qRT-PCR using the LightCycler 1.5 Instrument and the FastStart RNA Master SYBR Green I kit (both Roche Applied Science). PCR reactions were carried out in a 20 μL mixture composed of 3.25 mmol/L Mn(CH3COO)2, 0.5 μmol/L of each primer, 1 μL sample and 1-fold LightCycler RNA Master SYBR Green I. Primers were synthesized by Integrated DNA Technologies: MCOLN1 sense 5’- AGGGGCTCTGGGCTACC-3’, antisense 5’- GCCCGCCGCTGTCACTG-3’; ATG9B sense 5’-TGTGCTCACCGTCTACGAC-3’, antisense 5’-GGGAGGTAGTGCATGTGGG-3’; UVRAG sense 5’-ATGCCAGACCGTCTTGATACA-3’, antisense 5’-TGACCCAAGTATTTCAGCCCA-3’; polymerase (RNA) II (DNA directed) polypeptide A (POLR2A) sense 5’-GCACCACGTCCAATGACAT-3’, antisense 5’-GTGCGGCTGCTTCCATAA-3’. Reaction parameters were as follows: 61°C at 20 min followed by 95°C at 30 s followed by 45 cycles of 10 s, 95°C; 10 s, 60°C; 15 s, 72°C. Data were analyzed using the Pfaffl method [62]. The ratio between the target gene mRNA and POLR2A (the reference gene) was then calculated and normalized so that mRNA expression in mock infected cells equals 1.00. Data were then log2 transformed.
Intracellular staining and analysis of endogenous saponin resistant LC3B was performed as previously described [9, 10, 30] using rabbit anti-LC3B (D11; Cell Signaling) followed by phycoerythrin (PE) conjugated goat anti-rabbit IgG (Santa Cruz Biotechnology).
The following primary antibodies were used: TFEB (ab2636), SQSTM1 (ab31545) HIV-1 p24 (ab155836), HIV-1 p55/p17 (ab2581; all Abcam), LC3B (2775; Cell Signaling). The following secondary antibodies were used: Alexa Fluor 647-conjugated donkey anti-goat, Alexa Fluor 647-conjugated donkey anti-mouse, Alexa Fluor 568-conjugated donkey anti-sheep, Alexa Fluor 488-conjugated donkey anti-rabbit (all Molecular Probes). Cells were fixed in Dulbecco's phosphate-buffered saline supplemented with 4% (w/v) paraformaldehyde for 10 min, permeabilized with 0.2% (v/v) 4-(1,1,3,3-tetramethylbutyl)phenyl-polyethylene glycol for 10 min, probed with primary antibodies for 30 min, washed, then probed with secondary antibodies for 30 min, washed, and counterstained with 4',6-diamidino-2-phenylindole (Molecular Probes). Labeled cells were visualized using an Olympus Fluoview FV-1000 confocal imaging system on an IX81 platform equipped with a U Plan Fluorite 40×/1.3 NA oil differential interference objective (Olympus).
Data were assessed for symmetry, or skewness, using Pearson’s skewness coefficient. Fold change data were log2 transformed to convert the ratio to a difference that better approximates the normal distribution on a log scale. Comparisons between groups were performed using the paired, two-tailed, Student's t test. Differences were considered to be statistically significant when P < 0.05. * P < 0.05; ** P < 0.01; *** P < 0.001.
|
10.1371/journal.pgen.1005415 | BMP Inhibition in Seminomas Initiates Acquisition of Pluripotency via NODAL Signaling Resulting in Reprogramming to an Embryonal Carcinoma | Type II germ cell cancers (GCC) can be subdivided into seminomas and non-seminomas. Seminomas are similar to carcinoma in situ (CIS) cells, the common precursor of type II GCCs, with regard to epigenetics and expression, while embryonal carcinomas (EC) are totipotent and differentiate into teratomas, yolk-sac tumors and choriocarcinomas. GCCs can present as seminomas with a non-seminoma component, raising the question if a CIS gives rise to seminomas and ECs at the same time or whether seminomas can be reprogrammed to ECs. In this study, we utilized the seminoma cell line TCam-2 that acquires an EC-like status after xenografting into the murine flank as a model for a seminoma to EC transition and screened for factors initiating and driving this process. Analysis of expression and DNA methylation dynamics during transition of TCam-2 revealed that many pluripotency- and reprogramming-associated genes were upregulated while seminoma-markers were downregulated. Changes in expression level of 53 genes inversely correlated to changes in DNA methylation. Interestingly, after xenotransplantation 6 genes (GDF3, NODAL, DNMT3B, DPPA3, GAL, AK3L1) were rapidly induced, followed by demethylation of their genomic loci, suggesting that these 6 genes are poised for expression driving the reprogramming. We demonstrate that inhibition of BMP signaling is the initial event in reprogramming, resulting in activation of the pluripotency-associated genes and NODAL signaling. We propose that reprogramming of seminomas to ECs is a multi-step process. Initially, the microenvironment causes inhibition of BMP signaling, leading to induction of NODAL signaling. During a maturation phase, a fast acting NODAL loop stimulates its own activity and temporarily inhibits BMP signaling. During the stabilization phase, a slow acting NODAL loop, involving WNTs re-establishes BMP signaling and the pluripotency circuitry. In parallel, DNMT3B-driven de novo methylation silences seminoma-associated genes and epigenetically fixes the EC state.
| The understanding of germ cell cancer pathogenesis is based on a linear model, where seminomas and non-seminomas represent distinct entities, although originating from a common precursor lesion, the carcinoma in situ. We demonstrate that germ cell cancer development is a microenvironment-dependent plastic process that allows latent pluripotent seminomas /TCam-2 to acquire primed pluripotency and transit into an EC. For the first time, we show that this plasticity is initiated after interference with BMP signaling and driven by NODAL signaling, which is accompanied by considerable remodeling of the methylome. In conclusion, our data strongly suggest that ECs might also be able to transit into a seminoma upon interference with the drivers of reprogramming identified in this study.
| Type II germ cell cancers (GCC) arise from a precursor lesion termed carcinoma in situ (CIS) [1]. CIS cells are thought to be the result of a defective germ cell development and progress into seminomatous and non-seminomatous GCCs [1] [2]. Seminomas grow as a uniform tumor mass and are similar to CIS and PGCs with respect to gene expression. They express PGC- and pluripotency markers like PRDM1 (BLIMP1), TFAP2C, cKIT, SOX17, NANOG and OCT3/4 [1] [3] [4] [5]. Like CIS and PGCs, seminomas display DNA hypomethlyation compared to other GCC entities [6] [7]. Embryonal carcinomas (EC) are totipotent and differentiate into teratomas (cells of all three germ layers), yolk-sac tumors and choriocarcinomas (extra-embryonic tissues) [1]. Further, the DNA of ECs is highly methylated compared to CIS and seminomas [6]. GCCs are termed seminomas, when they consist to 100% of seminoma cells (60.6% of all GCC cases), but GCCs can also present as mixed non-seminomas with or without a seminomatous component (38.8% of all GCC cases), raising the question, if a CIS gives rise to seminomas and ECs at the same time or whether seminomas can be reprogrammed to ECs or vice versa [8] [9] [10]. Generally, seminomas are highly sensitive towards irradiation as well as cisplatin-based chemotherapy [11], while non-seminomas require a more aggressive treatment strategy and are resistant to DNA damage therapies [11] [12]. Thus, a reprogramming of a seminoma to an EC increases the risk of a poor outcome and would make it necessary to adjust the treatment strategy during a patient’s therapy.
Both, seminomas and EC express the pluripotency markers OCT3/4 and NANOG, but expression of the pluripotency factor SOX2 is restricted to ECs, while seminomas express SOX17 instead. Recently, SOX17 has been shown to be a key specifier of the human PGC cell fate by acting upstream of PRDM1 [3] [5] [13] [14]. PRDM1, the transcription factor TFAP2C and SOX17 render the PGCs and seminomas in a state of dormant pluripotency, meaning that they express pluripotency markers, but are not able to induce differentiation into somatic tissues. In contrast, ECs display naïve or primed pluripotency, enabling the cells to differentiate in response to appropriate signals into cells of all germ layers. Furthermore, ECs express several other pluripotency and epigenetic reprogramming factors like REX1 (ZFP42), DPPA3 (STELLA), GDF3, SALL4, PRDM14, DNMT3B /L or ZIC3 [15] [16] [5]. RNAi-mediated knock down of the pluripotency factor ZIC3 in murine and human ESCs induced SOX17, demonstrating that SOX17 is normally repressed by ZIC3 [17]. Further, it is known that SOX17 antagonizes WNT signaling, which has been suggested to demarcate seminomas from ECs [18] [19] [20].
The members of the TGF-beta superfamily play an important role in regulation of proliferation, differentiation and cell death in a broad variety of cell types and processes, including during PGC formation and GCC pathogenesis [21]. The TGF-beta signaling pathway is activated by binding of its ligands (TGF-betas, Activin /Nodal, GDFs, AMH or BMPs) to a type II receptor (ACVR2A /B, BMPR2, TGF-betaR2) that phosphorylates and activates a type I receptor (ALK3–7: ACVR1B /C, BMPR1A /B, TGF-betaR1) [22]. In turn, type I receptors activate effector moleculs of the SMAD family, which can be subdivided into receptor-SMADS (SMAD1 /2 /3 /5 /9) and co-SMAD (SMAD4) [23]. A complex of R-SMADs and co-SMAD4 acts in the nucleus as transcription factors and regulates target gene expression. A third SMAD class, termed inhibitory-SMADs (SMAD6 /7) is able to counteract these processes. In general, TGF-beta and Activin /Nodal signal via R-SMAD2 /3, while BMPs utilize the R-SMADs 1 /5 /8.
Active NODAL signaling depends on the co-receptor CRIPTO /CRYPTIC and stimulates expression of NODAL as well as LEFTY1 /2, leading to establishment of a signaling loop that stimulates and limits (LEFTY1 /2) itself simultaneously, to prevent an overshooting of mitogenic NODAL signals during embryogenesis or cell differentiation [24] [25] [26] [27]. Active endogenous NODAL signaling has been shown to regulate germ cell potency during mammalian testis development, where NODAL signaling is activated by signals (including FGF9) from somatic cells that lead to upregulation of the NODAL-co-receptor CRIPTO in germ cells [28] [29]. Furthermore, NODAL signaling regulates entry into meiosis [28] [30] [29]. Additionally, Spiller et al. found expression of NODAL and its cofactor CRIPTO as well as LEFTY1 in CIS and ECs and NODAL signaling might also provide a mechanism regulating potency in GCCs [28] [31]. In human ESCs and in murine epiblast cells, NODAL signaling has been shown to contribute to maintenance of pluripotency and is a hallmark of the primed state of pluripotency [32] [32].
BMP family members transduce their signals via their downstream effectors ID1–3, thereby regulating embryonic developmental and differentiation processes [33] [34] [35]. Bmp signals (Bmp4 /8B) specify murine PGCs from early proximal epiblast cells by suppressing Wnt signaling response genes and promoting Prdm1 /14 expression via T [36] [37] [38]. Furthermore, Bmp signaling is important for murine PGC migration and survival, since reduced Bmp signaling within the genital ridge leads to reduced numbers of PGCs and disrupted migration [39]. A Zebrafish model carrying a mutation in an ortholog of the human BMPR1B develops a seminoma-like tumor [40] [41] [42]. Furthermore, BMP signaling activity distinguishes histological subsets of paediatric germ cell tumors [43] and expression of BMP effectors ID1–3 has been demonstrated in seminomas [44]. Thus, BMP signaling might also play an important role in GCC pathogenesis. In the murine system, Pereira et al. found that Bmp /Smad5 signaling contributes to negative regulation of Nodal, since Smad5-deficient amnion cells showed ectoptic activation of Nodal and its feedback loops [45]. In turn, Nodal was shown to act as a Bmp inhibitor by heterodimerizing with Bmps [46]. Thus, a reciprocal interaction between Nodal and Bmp signaling might be an important mechanism in germ cell development and GCC development.
In previous studies, we demonstrated that the seminoma cell line TCam-2 differentiates into a mixed non-seminoma, when being cultivated in murine embryonic fibroblast conditioned medium supplemented with FGF4 /Heparin or in a combination of FGF4 /TGF-B1 /EGF, which mimics a somatic microenvironment, [47]. During this process the morphology changes considerably from polygonal to very big, flat and round cells with a big nucleus. Furthermore, a network-like structure is build up, the amount of multinucleated giant cells increases strongly and the proliferation rate drops significantly. Pluripotency markers (NANOG, OCT3/4, Alkaline Phosphatase) are downregulated, while markers for somatic differentiation are upregulated (AFP, PAX6, HAND1, T, HOXB1). Interestingly, an EC-intermediate, indicated by upregulation of SOX2 or SOX17 downregulation is not detected. Additionally, the BMP /SMAD signaling is reduced, putatively leading to downregualtion of PRDM1, allowing for differentiation into a mixed non-seminoma.
In a further study, we demonstrated that TCam-2 cells presented as pure, undifferentiated ECs 6–8 weeks after xenografting into the murine flank or brain [47]. In these somatic microenvironments, TCam-2 cells upregulate EC-markers SOX2, CD30, DNMT3B/L and downregulate seminoma markers SOX17, cKIT and PRDM1. Furthermore, DNA methylation levels increased strongly [47] [48]. Using these experimental settings, development of teratomas has never been observed. In contrast, orthotopic injection of TCam-2 into the testis leads to CIS /seminoma-like growth within the seminiferous tubules, indicated by a CIS /seminoma-like morphology (uniformly growing big round cells with a big nucleus, weakly eosinophilic) and expression of typical markers like SOX17, BLIMP1, VASA, TFAP2C and cKIT. These previous studies suggest that the microenvironment affects the cell fate of seminomatous TCam-2 cells [48].
In this study, we took advantage of the xenotransplantation model to analyze the molecular mechanisms during the reprogramming of TCam-2 to an EC in the somatic microenvironment of the murine flank. We utilized the cell line 2102EP as an EC model, which has been widely used in different studies, ranging from analyzing differentiation abilities, DNA methylation and retinoic acid response to studying chemoresistance and pluripotency [49] [50] [51] [48] [52] [53] [54] [55] [56]. 2102EP cells were derived from a patient suffering from an EC /teratocarcinoma and show an EC-like morphology (small polygonal and flat cells). Furthermore, 2102EP cells express EC- and pluripotency markers like SOX2, CD30, DNMT3B /L, NANOG, OCT3/4, but lack expression of PGC /semimona markers like SOX17. Similar to ECs, 2102EP cells show cytoplasmic localization of PRDM1 [57] [48]. The DNA of ECs is hypermethylated compared to seminomas [6]. In line to this finding, the DNA of 2102EP cells is highly methylated compared to TCam-2 cells. A common feature of GCCs, the gain of chromosome 12p can also be found in 2102EP [53] [56]. After xenotransplantation into the murine testis, flank or brain, 2102EP cells show a typical morphology and gene expression profile of EC cells. Importantly, 2102EP cells are nullipotent, thus they do not tend to differentiate in vivo into teratoma-, yolk-sac tumor- or choriocarcinoma-like cells [58] [52]. So, 2102EP cells resemble an undifferentiated EC in vitro and in vivo, highlighting 2102EP as a valuable EC model.
In this study, we deciphered the molecular mechanisms involved in adaptation of seminomatous TCam-2 cells to an EC-like cell fate. We demonstrate that interference with the BMP signaling pathway leads to upregulation of NODAL signaling as well as pluripotency- and epigenetic reprogramming factors, which drive the reprogramming and epigenetic remodeling of TCam-2 cells during growth in the somatic microenvironment of the murine flank. Our data strongly suggest that seminomas can be reprogrammed to an EC upon interaction with the microenvironment /tumor stroma.
The seminoma-like cell line TCam-2 is able to develop into an EC-like state after being xenografted into the flank or brain of nude mice [7] [48]. In this study, we analyzed the kinetics of gene expression (Gex) and DNA methylation (5mC) during this seminoma to EC transition (SET) to gain insight into the mechanisms driving this transition. Our previous experiments demonstrated that 6 weeks after transplantation TCam-2 cells had adapted an EC-like state [48]. Thus, we xenografted TCam-2 and 2102EP into the flank of nude mice and analyzed 5mC and Gex levels using microarrays after 1, 2, 4 and 6 weeks to follow early and late events during the transition.
Unsupervised hierarchical clustering (UHC) analysis of Gex and 5mC data revealed the differences between TCam-2 and 2102EP—after transplantation, the UHC demonstrated that up to 2 weeks after transplantation the cells still clustered to the parental TCam-2 cells. Gene expression seemed to gradually adjust to the 2102EP sample pattern, while DNA methylation seemed to increase and reorganized to the 2102EP pattern later (Fig 1A and 1B). After 4 weeks, xenografted TCam-2 cells clustered to the 2102EP cells, indicating an adaptation to an EC-like state with regard to Gex and 5mC (Fig 1A and 1B).
To gain a detailed insight into 5mC dynamics during SET, we plotted the averaged CpG-methylation at various regulatory regions (transcription start site (TSS)1500, TSS200, 5’-UTR, 1st exon, gene body and 3’-UTR) across all genes found to be differentially methylated in TCam-2 cells in vitro, after 1, 2, 4 and 6 weeks of in vivo growth as well as in 2102EP (S1 Fig). Parental and xenografted TCam-2 cells display 5mC levels of about 40–50% at the TSS1500, TSS200 and 1st exon. In contrast, 5mC levels at the 5’-UTR (≤ 40%), the gene body (≤ 35%) and the 3’-UTR (≤ 35%) are low in in vitro cultivated TCam-2 and xenografted cells for 1 week, while 5mC levels steadily increase at these regions with progressive in vivo growth to an profile highly comparable to 2102EP cells (S1 Fig).
Next, we distinguished CpG-island-associated DNA methylation events from DNA methylation at open sea context (i. e. non-CpG-island context) (Fig 2A and 2B). In TCam-2, the vast majority of CpG-island-associated CpGs show low levels of DNA methylation in regions 1500 and 200 bp upstream of the TSS, the 5’-UTR and the 1st exon (orange circle in S2A Fig), while CpG-islands within the gene body and the 3’-UTR appear medium to hypermethylated (green circle in S2A Fig). Six weeks after xenografting the CpGs in the gene body display distinct changes in methylation (red circle in S2A Fig) and demethylation (yellow circle in S2A Fig), while probes within the TSS200 /1500, 5’-UTR and 1st exon remain hypomethylated or become demethylated (black circle in S2A Fig).
CpG-probes associated with the open sea show a higher methylation compared to CpG-islands (S2B Fig). The 5mC levels of these regions are dramatically altered after six weeks in vitro and in 2102EP (S2B Fig). Thus, during in vivo growth of TCam-2 the change of methylation mainly occurs in within gene bodies of CpG-islands and in the open sea, while methylation of CpG-island-associated TSSs remains nearly unchanged.
Following xenografting of TCam-2, numbers of medium (41–80%) and highly (> 81%) methylated CpGs showed a strong increase on all chromosomes, except chromosomes Y and 19 (S1A Data). Strongly reduced methylation on the Y chromosome can be explained by the fact that complete arms of this chromosome are deleted TCam-2 (purple arrow in S1A Data) [56]. On chromosome 19, however, high numbers of hypomethylated CpGs (0–40%) are maintained 6 weeks after xenografting (green arrow in S1A Data). Thus, chromosome 19 seems to escape the de novo DNA methylation process during the SET. 6 weeks after xenografting 5mC distribution across all chromosomes is more comparable to 2102EP cells, while parental TCam-2 and TCam-2 xenografted for 1 week show considerable differences in 5mC distribution compared to 2102EP (S3A Fig).
To define initiating events of this reprogramming, we analyzed early Gex and 5mC dynamics using a volcano plot and found an (almost linear) increase in the number of genes being deregulated in expression during in vivo growth over time (Fig 2A). Next, a violin plot was used to visualize 5mC level distribution across all differentially methylated CpGs during SET (Fig 2B). TCam-2 cells cultivated in vitro and for 1 week in vivo display a high number of hypomethylated CpGs (≤ 30%). Interestingly, 2 weeks after xenografting the majority of CpGs displays 5mC levels around 50%, indicative for intermediate methylation. 4–6 weeks after xenografting, TCam-2 cells peak at approximately 60 and 70% 5mC levels respectively, indicating that the majority of CpGs are hypermethylated. These data demonstrate that 5mC levels shift from hypomethylation at one week via intermediate methylation at two weeks gradually towards high levels seen at 4–6 weeks. This strongly suggests that the remodeling is a gradual and constant process.
Now, we wanted to understand whether the changes in 5mC correlate to changes in Gex. A Pearson’s correlation of the microarray data identified 601 genes, showing inverse correlation between 5mC and Gex (S1B Data). A BDPC methylation cluster analysis of these 601 genes demonstrates that the transplanted cells cluster to the parental TCam-2 cell line up to 2 weeks after transplantation. Thereafter, they cluster more to the 2102EP to become highly similar after 6 weeks after transplantation (Fig 3B).
We reasoned that the genes from the differentially methylated group displaying the highest change in expression during the SET might be the candidates for driving this process. Hence, from the 601 differentially methylated genes we excluded all genes with an expression fold change of <log21.5 versus parental TCam-2. We found 53 genes which passed this criteria and called them 5mC /Gex-group (S1C Data). Genes that are weakly expressed /hypermethylated in TCam-2 cultivated in vitro (Fig 2C, (upper panel: T i.v.)) and for 1 week in vivo (1w) become demethylated and upregulated 6 weeks after xenografting (6w) and are expressed and hypomethylated in 2102EP cells (2102EP). Vice versa, genes hypomethylated and expressed in TCam-2 in vitro (Fig 2C, lower panel: T i.v.) and grown in vivo for 1 week (1w) become /are hypermethylated and downregulated in TCam-2 xenografted for 6 weeks (6w) and 2102EP cells (2102EP), respectively. Thus, 6 weeks after xenografting, the 5mC and Gex status of these genes is more comparable to the 2102EP profile than to parental TCam-2 or TCam-2 xenografted for 1 week (Fig 2C).
Next, we compared the genes of the 5mC /Gex-group to a set of genes deregulated in expression after 1 (143 genes) and 6 weeks (503 genes) (D and E in S1 Data). This revealed that 6 genes of the 5mC /Gex-group were upregulated after 1 week, despite the fact that 5mC-levels had dropped only marginally (Fig 2D and S1F Data). After 6 weeks, expression of these genes had increased further and the genomic loci became hypomethylated (Fig 2D). These 6 genes are GDF3, NODAL, DPPA3, DNMT3B, GAL and AK3L1, which are pluripotency-associated genes, except AK3L1, which encodes an enzyme of the adenylate kinase family (Fig 2E) [59] [60]. After 6 weeks, the remaining 47 genes of the 5mC /Gex group were deregulated and showed inverse correlation to 5mC (S1C Data). From them, EC-, pluripotency- and reprogramming-associated genes REX1 (ZFP42), DND1, JARID2 and PRDM14 were hypomethylated and upregulated, while seminoma-related genes PRDM1, PROM1 and IGF1 became hypermethylated and were downregulated [61] [62].
Furthermore, additional EC and pluripotency genes were upregulated (SOX2, LEFTY1 /2, DNMT3L, SALL4, DPPA5, BCAT1, FZD7, LIN28, ZIC3), while seminoma-associated genes where downregulated (SOX17, TFAP2C, cKIT, PRAME), without changing 5mC-levels (D and E in S1 Data) [3] [17] [5].
We verified selected alterations in Gex 1–2 weeks after xenografting by qRT-PCR and immunohistochemical staining (IHC) (A and B in S4 Fig). Additionally, we confirmed demethylation of the GDF3 locus in TCam-2 4 weeks after xenografting by sodium-bisfulfite-sequencing (S4C Fig).
The heatmap of Gex data (Fig 1A) demonstrated that aside from the similarities between xenografted TCam-2 and the 2102EP samples, there are also differences in Gex between the analyzed cell types during SET. We normalized Gex data of all 2102EP samples (in vitro, 4w, 8w) and TCam-2 cells xenografted for 6w versus TCam-2 in vitro (S1G Data). Next, we excluded all genes deregulated in both, the TCam-2 6w and the 2102EP samples to produce datasets containing genes exclusively expressed in TCam-2 6w, but not in 2102EP samples and vice versa (S1G Data). We performed a STRING-based protein-protein interaction as well as a GeneTrail-based Gene Ontology (GO) analysis of these data sets to show in which molecular processes as well as interactive networks these genes are involved and summarized the results in (S1G Data). Genes exclusively expressed in the TCam-2 6w samples are mainly linked to developmental and regulatory processes as well as signaling, while genes expressed only in 2102EP samples are related to GO categories linked to cellular compartments, like cytoplasm, nucleus, membrane and intracellular organelles.
To further analyze the regulatory mechanisms underlying the SET, we performed a STRING-protein-interaction-analysis of all genes upregulated after 1 week. An interaction network between 4 of the 6 5mC /Gex-group genes (GDF3, GAL, DPPA3, DNMT3B) and SOX2 as well as DNMT3L was predicted (S5A Fig). Further, a regulatory link between NODAL and LEFTY2 was proposed (S5A Fig). Inclusion of all genes upregulated lately after 6 weeks led to extension of this network to many pluripotency- and reprogramming-related factors, like REX1, JARID2, FGF2, WNT3, ZIC3 and PRDM14 (S5B Fig).
We asked, whether the changes in Gex reflect differences between seminomas and ECs in vivo. We performed a meta-analysis of our data and a cDNA microarray of GCCs [15] and filtered genes that are informative to discriminate seminomas from ECs (S1H Data). With progressive in vivo growth, TCam-2 cells express more genes found in ECs than seminomas (Fig 3A and 3B). Among them, GDF3, NODAL, LEFTY1 /2, GAL, DPPA3, SOX2, DNMT3B /L, DPPA5, BCAT1, FGF2, PRDM14, ZIC3 and FZD7, while seminoma markers SOX17, cKIT and PRAME were downregulated (Fig 3C). Furthermore, a qRT-PCR analysis verified that pluripotency and epigenetic reprogramming factors (SOX2, ZIC3, GDF3, NODAL, DPPA3, DNMT3B, GAL, JARID2, REX1, WNT3, PRDM14) are expressed higher on average in EC cell lines (2102EP, NCCIT, NT2/D1, 833KE, GCT27, H12) than in parental TCam-2, while expression of seminoma markers SOX17 and TFAP2C is considerably higher in TCam-2 (S5C Fig). In line with previous publications, expression of pluripotency factors NANOG and OCT3/4 is high is all cell lines analyzed [4] [59] [63]. Thus, the genes upregulated during the SET represent EC core genes.
During reprogramming, we found decreasing 5mC-levels in 6 5mC /Gex-group genes (Fig 2E). We compared 450k array data of these genes in parental and xenografted TCam-2 /2102EP cells to seminoma tissues and three additional EC cell lines (NEC8, 833KE, SuSa [64] [65] [66]) (Fig 3D). We found that DPPA3, AK3L1, DNMT3B and NODAL are hypermethylated at analyzed loci in seminomas and parental TCam-2 compared to TCam-2 in vivo 6w and the EC samples (Fig 3D). Parental TCam-2 and all EC cell lines show GAL hypermethylation, which is strongly reduced after xenografting of TCam-2 and 2102EP, suggesting that GAL hypermethylation is established and maintained during in vitro cultivation of EC cells. GDF3 hypermethylation is restricted to parental TCam-2, but not seen in seminomas or EC cells (Fig 3D). The GDF3 locus became demethylated during xenografting of TCam-2 (Fig 3D). Thus, high 5mC-levels of GDF3 in parental TCam-2 point at a cell line-specific effect, but correlate inversely to Gex (S1D Data). A BDCP analysis demonstrates that parental TCam-2 cells cluster closely to seminoma tissues and align to the EC samples 6 weeks after xenografting with regard to 5mC status of analyzed genes (Fig 3E).
We interrogated our data with regard to expression of signaling pathway-related genes to further elucidate the initial trigger of the SET. We found deregulation of genes involved in BMP, NODAL, Retinoic acid (RA), FGF, HIPPO, STAT, IGF, NOTCH and WNT signaling (S6A Fig).
Since BMP signaling is central for germ cell specification, we concentrated on this pathway first. In parental TCam-2, moderate cytoplasmic and strong nuclear staining of phosphorylated SMAD (pSMAD)1 /5 was detected by immunofluorescence staining (IF) (Fig 4A). Also, expression of BMPR1A /R2, BMP4 /7 and SMAD4 suggested that BMP signaling is active (S1I Data). 1 week after xenografting, a rapid and strong decrease of the BMP pathway effectors ID1 /3 is detected (S6A Fig). Additionally, BMP receptors BMPR1A /R2 show a trend of downregulation (S1I Data). IHC and western blotting detected loss of SMAD1 /5 phosphorylation from 1–4 weeks after xenografting (Fig 4A and S6B Fig). Interestingly, after 6 weeks, cytoplasmic pSMAD1 /5 is detectable again, pointing at recovery of BMP signaling (Fig 4A). Also, ID1 /3 and BMPR1A expression recovers from 2 weeks on (S6A Fig and S1I Data).
To test the role of BMP signaling for initiation of SET-reprogramming, TCam-2 cells were treated with the BMP inhibitor NOGGIN for 8 days [67]. After application of NOGGIN, by western blotting and qRT-PCR analysis, we observed a reduction of pSMAD1 /5 levels (S6C Fig) and downregulation of ID1 /3 (Fig 4B), indicating inhibition of the BMP-pathway. Further, upregulation of 4 of the 5mC /Gex-group genes (NODAL, GDF3, GAL, DNMT3B) as well as the EC markers CRIPTO, CRYPTIC, LEFTY1, SALL4, LIN28, JARID2, PRDM14, DNMT3L and SOX2 was detected, while SOX17 was downregulated after 8 days (Fig 4B). Additionally, BMP- and WNT signaling-associated molecules BMP4, WNT3 and WNT5B as well as pluripoteny-related gene ZIC3 were upregulated 8 days after NOGGIN treatment (Fig 4C). Upregulation of GDF3, NODAL, LIN28, SOX2 and DPPA3 was shown by western blotting (Fig 4D). To confirm the results, we treated TCam-2 cells with the BMP inhibitor LDN193189. Again, we observed a reduction in SMAD1 /5-phosphorylation, downregulation of ID1 /3 as well as NODAL, SOX2, LIN28 and DNMT3B /L upregulation (D and E in S6 Fig). These findings suggest that BMP inhibition is an initial event in the reprogramming of seminomas to ECs. Inhibition of BMP signaling leads to derepression of NODAL signaling as well as upregulation of pluripotency- and reprogramming-associated factors.
To analyze if activation of NODAL signaling alone is sufficient to induce deregulation of pluripotency- and SET-associated genes, we treated TCam-2 cells with recombinant NODAL. Expression of endogenous NODAL and corresponding signaling keyplayers or pluripotency- and reprogramming-associated factors did not change, although increasing SMAD2 /3-phosphorylation verified an efficient treatment (F and G in S6 Fig). This suggests that inhibition of BMP signaling is a prerequisite for the establishment of NODAL signaling.
Next, we screened GCC tissues for expression of BMP and NODAL signaling keyplayers by re-analyzing cDNA microarray data and performing IHC as well as western blots (Fig 5A–5C) [15]. For IHC, only TFAP2C positive and SOX2 negative CIS and semiomas as well as SOX2 positive ECs were analyzed (S7A Fig). In CIS, seminomas and ECs, expression of BMP8B, BMPR1A /2, SMAD1 /4 and ID1 /2 was detected, while ID3 expression was restricted to ECs (Fig 5A). In line with this expression profile, ID1 was detectable in the vast majority of CIS, seminomas and ECs by IHC of GCC tissue microarrays (GCC-TMA), showing that in these GCC entities BMP signaling is active (Fig 5B and S7A Fig).
NODAL signaling induces its downstream effectors CRIPTO /CRYPTIC and LEFTY1 /2. Furthermore, NODAL signaling activity is maintained by ZIC3. We detected considerably higher levels of these NODAL signaling keyplayers in ECs compared to CIS /seminomas (Fig 5A). Additionally, a western blot analysis demonstrates the EC cell lines 2102EP and NT2/D1 display high levels of ZIC3 while TCam-2 cells, choriocarcinoma-like JAR cells and human adult fibroblasts show low levels (S7B Fig). Furthermore, expression of ZIC3 is higher in EC tissues than in seminomas, CIS or normal testis tissue (Fig 5C and S7C Fig). Taken together, in vivo ZIC3 mRNA and ZIC3 protein levels correlate to SOX2 as well as NODAL expression (ECs) and correlate inversely to SOX17 (CIS, semiomas) (S7C Fig).
NODAL and ACTIVIN signaling are closely related to each other and key components of ACTIVIN signaling are heterogeneously expressed in GCCs [68] [69]. Additionally, CRIPTO is able to inhibit ACTIVIN signaling [70]. Thus, during SET activation of NODAL signaling might also influence ACTIVIN signaling. We screened for expression of ACTIVIN signaling keyplayers in TCam-2 in vitro and in vivo, but could not detect any changes in expression of the ACTIVIN /INHIBINS, the ACTIVIN receptors (ACVRs), TGFBR3, MAN1 or the ACTIVIN inhibitor Follistatin (FST) (S1J Data) [68] [69]. Thus, ACTIVIN signaling seems not to contribute to reprogramming of TCam-2 cells.
In vitro, TCam-2 cells display negligible expression of WNT molecules and only expression of WNT receptors FZD3 /6 was detected (S1K Data). In contrast, in 2102EP cells WNT3/5B and FZD7 /9 are expressed. During SET, WNT3/5B and FZD7 /9 are induced, while FZD3 /6 tend to be downregulated (S7A Fig and S1K Data). Thus, TCam-2 cells change expression of WNT signaling associated genes to a profile comparable to 2102EP. Accordingly, WNT3 /5B and FZD7 expression is higher in EC tissues than in CIS /seminomas (Fig 5A). Using IHC, we demonstrate that CIS display only membraneous staining of the canonical WNT effector beta-CATENIN, while seminomas and ECs presented in two states, i. e. showing membraneous staining or positive at both, the membrane and the cytoplasm (Fig 5D and S6A Fig). 72% of seminomas stained positive at the membrane only, while 97% of ECs displayed both, strong membraneous and cytoplasmic beta-CATENIN, verifying results of Korkola et al. [71] (Fig 5D and S7A Fig). In line with these data, IF /IHC demonstrates that in parental TCam-2 beta-CATENIN is localized to the membrane, while increasing cytoplasmic staining is detectable 1–6 weeks after xenografting (Fig 5E). In conclusion, similar to EC tissues, beta-CATENIN accumulates in the cytoplasm of TCam-2 cells following xenografting.
In this study, we analyzed the epigenetic and molecular mechanisms underlying the seminoma to EC reprogramming process. After transplantation, expression of 6 genes was rapidly induced, with 5mC levels unchanged initially. Thus, in seminomas these genes seem to be poised for expression. Early induction of DNMT3B initiates a wave of de novo DNA methylation causing a gradual remodeling of the methylome two weeks after xenografting, leading to a genome-wide high 5mC levels similar to an EC. During SET, remodeling of the methylome affects mainly gene bodies (but not regulatory regions, like TSS) in the CpG-island and non-CpG-island context and follows deregulation in Gex, suggesting that DNA methylation rather reinforces than initiates the EC-like state of TCam-2.
A strong downregulation of the BMP signaling downstream effectors ID1 and ID3 during the SET prompted us to investigate BMP signaling in more detail. We show that inhibition of BMP signaling leads to induction of NODAL signaling and pluripotency- as well as epigenetic reprogramming factors comparable to the reprogramming of TCam-2 in vivo. Previously, we were able to show that during in vitro differentiation of TCam-2 cells into a mixed non-seminoma the activity of BMP signaling-related SMAD1 /5 /8 molecules was reduced [47]. This further demonstrates that high BMP signaling activity is associated with a CIS /seminoma-like character, while low levels are linked to a non-seminomatous cell fate. Hence, we propose that inhibition of BMP signaling is the initial event triggering SET-reprogramming. In contrast to the results reported here, upregulation of the EC-marker SOX2 was not observed during the in vitro differentiation [47]. We speculate that the particular experimental settings in vitro (supplementation with FGF4, TGF-B1, EGF) resulted in a persistent suppression of SOX2, leading to continuation of SOX17 expression. However, with downregulation of PGC- (PRDM1, TFAP2C, cKIT) and pluripotency (NANOG, OCT3/4, LIN28) marker genes, persisting SOX17 expression together with activation of the Hippo pathway resulted in differentiation into a mixed non-seminoma with predominant choriocarcinoma-like components [47].
Spiller et al. found expression of NODAL and its cofactor CRIPTO as well as LEFTY1 in CIS and ECs [39]. The authors utilized qRT-PCR to analyze expression of NODAL signaling keyplayers in testis containing up to 90% CIS cells and non-seminomas, while seminomas were not included [28]. We detected low expression levels of NODAL signaling factors in CIS /seminomas and high levels in ECs (Fig 5A) [16]. In our study, the cDNA microarray analysis of GCC tissues was performed on RNA isolated from pure micro-dissected CIS cells, without any normal testicular tubules or invasive tumors, pure classical seminomas and ECs [35]. In our case, RNA expression levels and protein detection via IHC of various markers in GCC tissues is also observed in our SET model system. Hence, we argue that the discrepancies with Spiller et al. might be of technical nature, i. e. residual somatic components, which eventually skews analyses by having active NODAL signaling.
Spiller et al. state further that active NODAL signaling provides a mechanism regulating potency in GCCs [28] [31]. In human ESCs and in murine epiblast cells, NODAL signaling has been shown to contribute to maintenance of pluripotency and is a hallmark of the primed state of pluripotency [32]. Thus, activation of NODAL signaling might trigger the shift from latent pluripotency (observed in seminomas) to primed pluripotency displayed by ECs.
During vertebrate development, expression of the pluripotency-related factor ZIC3 is repressed by BMP signaling and can be restored by NOGGIN-mediated inhibition of BMP signaling [72] [73] (Fig 4C). ZIC3, which is necessary for maintenance of NODAL signaling is highly expressed in ECs /xenografted TCam-2 and low in seminomas /TCam-2 in vitro (Fig 5A and S7B and S7C Fig; S1H Data) [17] [74]. The STRING analyses suggested that ZIC3 interacts with NODAL and LEFTY1 /2 (S5B Fig) and ZIC3 is activated by NANOG, OCT3/4 and SOX2 [17]. So, during SET, inhibition of BMP signaling leads to derepression of SOX2, restoring the classical pluripotency circuitry found in ECs and ESCs, subsequently leading to upregulation of ZIC3, which in turn helps to maintain NODAL signaling [17] [74].
What is the crossregulation between BMP- and NODAL signaling? Pereira et al. found that in mice Bmp /Smad5 signaling represses Nodal, since amnion cells deficient for Smad5 showed ectoptic activation of Nodal and its feedback loops [45]. In turn, NODAL inhibits BMP by heterodimerizing with BMPs [46]. Thus in our case, signals form the tumor stroma inhibit BMP, which leads to derepression of NODAL. Upregulation of NODAL leads to establishment of an autoregulatory loop, including LEFTY1 /2, CRIPTO /CRYPTIC and ZIC3. This results in a cell intrinsic repression BMP signaling.
Why does BMP signaling recover during the reprogramming of TCam-2? As described above, Nodal activates its autoregulatory loop, which has been denominated the fast acting loop [24]. In addition, over time, the so called slow feedback loop activates Bmp4, which re-establishes BMP signaling and results in upregulation of Wnt3 and Fgf4 /Fgf8 [45] [75]. This is in agreement with the data from our transplantation studies, where we detected increased WNT3 /5B, BMP4 /BMP7 and FGF2 /19 expression from 2–6 weeks after xenografting (S6A Fig; I and K in S1 Data). Additionally, 8 days after NOGGIN-treatment of TCam-2 cells BMP4, WNT3 and WNT5B were upregulated (Fig 4C), while ID1 /3 levels recovered like during in vivo growth (Fig 4B and S6A Fig).
During SET and after NOGGIN treatment, we detected downregulation of SOX17 and upregulation of SOX2. In fact, SOX17 expression is restricted to CIS and seminomas, while SOX2 is highly expressed in ECs [2]. SOX17 has been identified as a key factor for specification of human PGCs and regulator of PRDM1 [14]. Thus, downregulation of SOX17 during the SET indicates loss of a PGC-like character. In mice, Sox2 complexes with Oct3/4 and binds to a canonical motif, thereby driving the expression of pluripotency genes [76]. Overexpression of Sox17 is able to replace Sox2 in the complex with Oct3/4, leading to a change in target site selection to a compressed binding motif [76]. So, we speculate that during SET the strong increase in SOX2 protein levels force partnering with OCT3/4, which leads to a switch to promoters encoding for the canonical motif found in pluripotency genes. Further, it is known that SOX17 antagonizes WNT signaling activity, which has been suggested to be low in seminomas and high in ECs [18] [19] [20]. So, downregulation of SOX17 could explain the de-repression of WNT3 /5B during SET. The upregulated WNT3 results in cytoplasmic beta-CATENIN accumulation, but nuclear exclusion of beta-CATENIN suggests that the canonical WNT-pathway is not activated [77] [78] [79]. Thus, WNT3 /WNT5B most likely act in a non-canonical manner during SET [79].
Based on our findings, we propose a model in which the SET-reprogramming of xenografted TCam-2 is divided in three stages (initiation, maturation, stabilization) (Fig 6A) [80] [81]. The reprogramming is initiated by exogenous inhibition of BMP signaling causing rapid activation of NODAL. NODAL signaling establishes a fast acting autoregulatory loop (Fig 6B), leading to stimulation (CRIPTO /CRYPTIC) and limitation (LEFTY1 /2) of NODAL signaling and cell intrinsic suppression of BMP signaling. During this time, markers of pluripotency and reprogramming become upregulated and induction of DNMT3B initiates epigenetic remodeling. This phase we name the maturation phase. Thereafter, the slow acting NODAL feedback loop re-establishes BMP signaling to a level lower than in parental TCam-2, resulting in a balance between BMP and NODAL signaling and reinforcement of the acquired EC-like cell fate (the stabilization phase).
In summary, we demonstrated that in seminomas a set of 6 genes is rapidly induced after transplantation. These factors induce epigenetic remodelling of the genome and establish expression of the pluripotency network, leading to reprogramming into an EC. Further analysis revealed that interference with BMP is sufficient to induce these genes. We propose that BMP inhibition initiates the SET. The inhibition of BMP signaling, so we speculate, is initiated by factors like NOGGIN, which are expressed abundantly by the somatic microenvironment. So, upon transplantation into the flank, TCam-2 cells become exposed to BMP-inhibitors leading to initiation of SET. Corollary to this, a CIS or seminoma, which is exposed to BMP inhibitors by penetrating the testis confines during progressive growth, could be reprogrammed to an EC. Our data strongly suggest that GCC development is a plastic process that allows seminomas to progress into EC and maybe vice versa, depending on the signals from the tumor stroma. Therefore, seminoma patients might also develop an EC component during invasive tumor growth. ECs grow more aggressive than seminomas and need alternative treatment strategies, which requires adjustment of the therapy concept. The question remains, whether ECs might transit into a seminoma upon interference with the DNA methylation machinery or reprogramming key molecules identified in this study.
The ethics committee of the Rheinische Friedrich-Wilhelms-Universität Bonn approved the analyses of formalin fixed, paraffin-embedded type II GCC tissues in context of this study. No personal patient data will be collected or stored. Written permission to use the tissue for scientific purposes was obtained from the patients and was approved by the, Ethik-Kommission für klinische Versuche am Menschen und epidemiologische Forschung mit personenbezogenen Daten der Medizinischen Fakultät der Rheinischen Friedrich-Wilhelms-Universität Bonn’ (The ethics committee for clinical trials on humans and epidemiological research with patient-related data of the medical faculty of the Rheinischen-Friedrich-Wilhelms-University Bonn).
All animal experiments were conducted according to the German law of animal protection and in agreement with the approval of the local institutional animal care committees (Landesamt für Natur, Umwelt und Verbraucherschutz, North Rhine-Westphalia (approval ID: AZ-84-02.04.2013-A430). The experiments were conducted in accordance with the International Guiding Principles for Biomedical Research Involving Animals as announced by the Society for the Study of Reproduction.
GCC cell lines utilized in this study were cultivated as described previously [7]. Briefly, TCam-2 and NCCIT cells were grown in RPMI. The cell lines 2102EP, NT2/D1, 833KE, H12, GCT27, JEG-3 and JAR were grown in DMEM. Both media were supplemented with 10% fetal calf serum (FCS) (PAA, Pasching, Austria), 1% Penicillin /streptomycin (P /S) (PAN, Aidenbach, Germany), 200 mM L-Glutamine (PAN, Aidenbach, Germany). MPAF and ARZ were grown in DMEM (10% FCS, 1% P /S, 200 mM L-Glutamine, 1x non-essential amino acids (PAA, Pasching, Austria), 100 nM ß-Mercaptoethanol (Sigma-Aldrich, Taufkirchen, Germany). TCam-2 [82] cells were kindly provided by Dr. Janet Shipley (Institute of Cancer Research, Sutton, United Kingdom). 2102EP [83], NT2/D1 [84] and NCCIT [85] cells were provided by Prof. Dr. Leendert Looijenga (Erasmus MC, Daniel den Hoed Cancer Center, Josephine Nefkens Institute, Rotterdam, Netherlands). 833KE [86] cells were provided by PD Dr. Beate Köberle (KIT, Karlsruhe, Germany). H12.1 [87] and GCT27 [88] were kindly provided by Dr. Peter Andrews (University of Sheffield, United Kingdom) and obtained from Dr. Thomas Müller (Department of Internal Medicine IV, Oncology and Hematology, Martin-Luther-University of Halle Wittenberg, Halle, Germany). JAR (HTB-144) and JEG-3 (HTB-36) cells were purchased from ATCC. MPAF and ARZ were provided by Dr. Michael Peitz (Life & Brain Center, University of Bonn, Germany).
Tissue microarrays were assembled and prepared in house after approval by the internal review board. Further information is given in [7].
TCam-2 cells were seeded 24h before treatment (1 x 105 cells /9,5 cm2). 500 ng /ml NOGGIN (diluted in 10 mM HAc) (Abcam, Cambridge, UK), 500 ng /ml LDN193189 (diluted in H2O) (Sigma-Aldrich, Taufkirchen, Germany) and 500 ng /ml recombinant NODAL (diluted in 4 mM HCl, 0.1% BSA) (R&D Systems, Wiesbaden, Germany) were added in 2 ml fresh culture medium every second day.
DNA, RNA and proteins were isolated as described previously [47] [89]. DNA was isolated by phenol /chloroform /isoamylalcohol, RNA by TRIzol and proteins by RIPA buffer. DNA and RNA concentrations as well as 260 /280 nm, 260 /230 nm purity ratios were determined by NanoDrop measurement (Peqlab, Erlangen, Germany).
Western blots analyses were performed as described previously [47] [7]. Briefly, the Mini-PROTEAN Electrophoresis Cell and Trans-Blot Turbo system were used (BioRad, Munich, Germany). Gels were blotted onto PVDF membranes. Chemiluminescent signals were detected using ChemiDoc MP Imaging System (BioRad) and band intensities were calculated by Image Lab software (BioRad). Beta-ACTIN was used as housekeeper and for normalization. See S1 Table for antibody details.
Quantitative RT-PCR (qRT-PCR) was performed as described previously [7]. For first strand synthesis, the RevertAid First Strand cDNA Synthesis Kit manual (Fermentas, St. Leon-Rot, Germany) was used. For PCR, the Maxima SYBR Green qPCR Master Mix (Fermentas, St. Leon-Rot, Germany) was used. PCR was performed using the ViiA 7 Real Time PCR System (Applied Biosystems, distributed by Life Technologies, Carlsbad, CA, USA). At the end of each PCR run, a melting point analysis was performed. GAPDH was used as housekeeping gene and for data normalization. Variation of GAPDH expression between different experimental setups is very low (S1L Data). See S2 Table for primer sequences.
Immunohistochemistry (IHC) was performed as published previously [47] [7]. Tumor tissues were dissected, fixed in 4% formalin overnight and processed in paraffin wax. Signal detection was performed semiautomatically in the Autostainer 480 S (Medac, Hamburg, Germany). Nuclei were stained by hematoxylin. Immunofluorescence staining (IF) was performed as published [15] [47]. Nuclei were counterstained by Hoechst 33342. See S1 Table for antibody details and dilution ratios.
Sodium bisulfite sequencing was performed as described previously [89]. Briefly, 500 ng of DNA were sodium bisulfite converted using the ‘EZ DNA-Methylation Gold kit’ (Zymo Research, Freiburg, Germany). See S2 Table for primer details.
Xenotransplantation was performed as described previously [48]. 1 x 107 cells in 500 μl of 4°C cold Matrigel (BD, Heidelberg, Germany) were injected into the flank of CD1 nude mice.
RNA quality was checked for degradation via gel electrophoresis in a BioAnalyzer 2100 (Agilent Technologies, Waldbronn, Germany) using RNA 6000 nano lab chips. DNA was sodium-bisulfite converted using the EZ DNA Methylation kit (Zymo Research, Freiburg, Germany). Samples were processed on Illuminas' (San Diego, California, USA) human, HT-12v4’ and human, Infinium Methylation 450k Bead Chips’.
A subset quantile normalization approach developed by N. Touleimat & J. Tost was applied [90]. This approach includes signal correction for the adjustment of the color balance and background level correction as well as the Infinium I/Infinium II shift correction between sample normalization. Technical quality parameters such as hybridization, extension, bisulfite conversion and specificity were evaluated using the, Genome Studio’ software. Beta-value signal distributions were inspected by density plots. Data was analyzed using, Bioconductor R’ (www.bioconductor.org). To increase performance in terms of detection and true positive rate of highly methylated and unmethylated CpG-sites, beta-values were transformed to M-values [91]. Differentially methylated loci were identified using a t-test. p-values were corrected for multiple testing using the Benjamini-Hochberg correction. The expression values were quantile normalized using the, limma’-software-package (‘Linear Models for Microarray Data’, www.bioconductor.org). For inverse correlation analysis of methylation and gene expression data, methylation at CpGs and gene expression transcripts were mapped to the same gene identifiers. Inverse correlation was calculated using the Pearson correlation coefficient and p-values for association were corrected for multiple testing using the Benjamini-Hochberg correction. Microarray data sets are publically available via GEO (ncbi.nlm.nih.gov/geo/) (GSE60698, GSE60787).
The whole procedure has already been published [15]. The array was reanalyzed in context of this study. Normalized gene expression intensities of averaged seminomas were substracted from averaged intensities of EC tissues (Seminoma group) and normalized gene expression intensities of averaged ECs were substracted from averaged intensities of seminoma tissues (EC group).
BDPC analysis and STRING protein-protein-interaction prediction were performed online using default settings (services.ibc.uni-stuttgart.de/BDPC) (string-db.org) [92] [93]. GeneTrail-based GO analysis was also performed online using default settings (genetrail.bioinf.uni-sb.de) [94]. Circos diagrams were generated using ‘Circos Table Viewer’ (mkweb.bcgsc.ca/tableviewer) [95] and Venn diagrams were generated using ‘Venny’ (bioinfogp.cnb.csic.es/tools/venny).
|
10.1371/journal.pntd.0002484 | Ongoing Spillover of Hantaan and Gou Hantaviruses from Rodents Is Associated with Hemorrhagic Fever with Renal Syndrome (HFRS) in China | Longquan City, Zhejiang province, China, has been seriously affected by hemorrhagic fever with renal syndrome (HFRS) since the first cases were registered in 1974. To understand the epidemiology and emergence of HFRS in Longquan, which may be indicative of large parts of rural China, we studied long-term incidence patterns and performed a molecular epidemiological investigation of the causative hantaviruses in human and rodent populations.
During 1974–2011, 1866 cases of HFRS were recorded in Longquan, including 20 deaths. In 2011, the incidence of HFRS remained high, with 19.61 cases/100,000 population, despite the onset of vaccination in 1997. During 1974–1998, HFRS cases in Longquan occurred mainly in winter, while in the past decade the peak of HFRS has shifted to the spring. Notably, the concurrent prevalence of rodent-borne hantaviruses in the region was also high. Phylogenetic analyses of viral sequences recovered from rodents in Longquan revealed the presence of novel genetic variants of Gou virus (GOUV) in Rattus sp. rats and Hantaan virus (HTNV) in the stripe field mice, respectively. Strikingly, viral sequences sampled from infected humans were very closely related to those from rodents.
HFRS represents an important public health problem in Longquan even after years of preventive measures. Our data suggest that continual spillover of the novel genetic variant of GOUV and the new genetic lineage of HTNV are responsible for the high prevalence of HFRS in humans. In addition, this is the first report of GOUV associated with human HFRS cases, and our data suggest that GOUV is now the major cause of HFRS in this region.
| Hemorrhagic fever with renal syndrome (HFRS) is a major public health problem in China despite human vaccination. We investigated the epidemiology and emergence of HFRS in Longquan (Zhejiang Province), a rural area with a high incidence of HFRS. During 1974–2011, a total of 1866 cases of HFRS were recorded in Longquan, including 20 deaths. Strikingly, phylogenetic analyses of viral sequences sampled from local rodents in Longquan revealed the presence of novel variants of Gou virus (GOUV) in Rattus sp. rats and Hantaan virus (HTNV) in the stripe field mice, respectively. Moreover, viral sequences sampled from infected humans in Longquan were very closely related to those from rodents. Overall, these data indicate that there is a continual spillover GOUV and HTNV from rodents to humans in Longquan, and this might be responsible for the high prevalence of HFRS. As well as highlighting the importance of the human-animal interface, these data also suggest that GOUV is now the major cause of HFRS in this region.
| Hantaviruses are important zoonotic pathogens. Although they can establish a persistent and asymptomatic infection in their natural rodent reservoirs [1], in humans hantaviruses can cause two severe diseases: hemorrhagic fever with renal syndrome (HFRS) and hantavirus (cardio) pulmonary syndrome (HPS) [2]. In Eurasia HFRS is associated with Hantaan virus (HTNV), Seoul virus (SEOV), Amur/Soochong virus (ASV), Dobrava-Belgrade virus (DOBV), Saaremaa virus (SAAV), Sochi virus, and Puumala virus (PUUV), whereas HPS is due to the infection of Sin Nombre virus (SNV), Andes virus (ANDV), and other viruses in the Americas [2], [3]. The clinical severity of HFRS is related to the etiologic agents involved [4]–[8], with DOBV and HTNV being the most dangerous representatives, with fatality rates of up to 15% [4]–[7]. In contrast, SEOV usually causes a milder form of HFRS with a mortality rate of approximately 1% [6], [7]. PUUV causes a mild disease referred to as nephropathia epidemica (NE) with a mortality rate ranging from 0.1% to 0.3% in Europe 5,8. HFRS cases caused by HTNV mainly occur in the winter, while the HFRS cases caused by SEOV peak in the spring and summer [9], and which likely reflects occupation-connected differences in exposure to rodents in different seasons.
Following the implementation of comprehensive preventive measures and socioeconomic development, the numbers of HFRS cases and fatalities in China have decreased dramatically, although remain the highest globally [7]. In China, the most prevalent hantaviruses are HTNV and SEOV carried, respectively, by the striped field mouse (Apodemus agrarius) and Norway (or brown) rat (Rattus norvegicus) [6], [7], [9], [10]. To date, only these two viruses have been identified to cause HFRS in China. However, hantaviruses from bats, insectivores, and rodents (e.g. Dabieshan virus (DBSV), Gou virus (GOUV), Longquan virus (LQUV), Thottapalayam virus (TPMV)) have also been documented [7], [11]–[15], although whether they are associated with human disease is unclear.
Longquan is a county-level city located in the southwestern part of Zhejiang Province. It includes both urban and rural areas, with a population of approximately 280,000. More than 90% of the Longquan's total area is mountainous. In 1974, the first HFRS case was recorded in Longquan. Since that time, Longquan has been one of the most severely affected regions in Zhejiang and in China as a whole. However, little is known about the epidemiology and etiologic agents of HFRS in this region. Our recent surveys in Longquan revealed at least nine species of rodents and insectivores, with A. agrarius and R. norvegicus dominant in rural and residential areas, respectively [16]. Herein we report the changing incidence of HFRS in Longquan, the genetic characterization of the etiologic agents (hantaviruses) circulating in local rodents, and their connection to the human population.
This study was reviewed and approved by the ethics committee of National Institute for Communicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention (Chinese CDC). All animals were treated in strict according to the guidelines for Laboratory Animal Use and Care from the Chinese CDC and the Rules for the Medical Laboratory Animal (1998) from the Ministry of Health, China. These protocols were approved by the National Institute for Communicable Disease Control and Prevention of the China CDC. All surgery was performed under ether anesthesia, and all efforts were made to minimize suffering. Collecting human serum samples from HFRS patients was also approved by the ethics committee of National Institute of Communicable Disease Control and Prevention of the China CDC, according to the medical research regulations of Ministry of Health, China. A signed individual written informed consent was obtained from each of five patients when their blood samples were collected.
Records for HFRS cases occurring during 1974–2011 were obtained from the Longquan Center for Disease Control and Prevention. Until 1982, HFRS cases were defined according to the national standard of clinical criteria, and confirmed by detection of hantavirus-specific IgM and IgG antibodies against HTNV or SEOV. From 1982 clinical cases were confirmed by a four-fold or greater titer increase of IgG antibodies in paired sera, as well as a IgM antibody titer >1∶20 in single serum as scored positive by an indirect immunofluorescent assay (IFA) (see below) [9]. The reaction pattern of positive serum was characterized as scattered and green granular cytoplasmic fluorescence in hantavirus-infected Vero E6 cells. The incidence rates of HFRS during 1974–2011 were calculated according to the population census number for each year.
Small mammals were trapped in fields and residential areas in Longquan during 2008–2011. Cages with a treadle release mechanism were used for live trapping according to the protocols described previously [17]. Traps were set in the same fields or residential areas during each season. Lung and kidney samples were collected from the trapped animals and stored in liquid nitrogen. All surgery was performed under ether anesthesia to reduce suffering. Ethanol-cleaned instruments were used for each animal.
Serum samples collected from five patients who suffered from acute HFRS during 2009–2011 were also studied. These serum samples were tested by IFA using HTNV-infected or GOUV-infected Vero-E6 cells as antigens [18]. The secondary antibody used was fluorescei-isothiocyanate-conjugated goat anti-human IgG or IgM (Southern Biotech, Birmingham, Alabama, USA).
Hantavirus antigen in lung or kidney tissues from rodents and insectivores was detected by IFA as described previously [18], with rabbit antibodies against the mixed antigens of HTNV/76-118 and SEOV/L99 prepared in this laboratory as the primary antibodies and FITC-labeled goat anti-rabbit IgG antibodies used as the secondary antibodies (Sigma, St. Louis, MO, US). Generally, lung tissues were tested first, and kidney tissues were tested if lung tissues were found to be negative.
Total RNA was extracted from hantavirus antigen-positive lung and kidney tissues, and human serum samples, using the TRIzol reagent (Invitrogen, San Diego, CA) according to the manufacturer's instructions. cDNA of the Small (S) and Medium (M) segments of the hantavirus genome was prepared with AMV transcriptase (Promega, Beijing, China) in the presence of primer P14 [19]. Partial or complete sequences of the S and the M segments were amplified as described previously [10], [20], [21]. All voles and insectivores were also screened for hantaviruses using RT-PCR as described previously [22].
DNA products were purified by a QIAquick Gel Extraction kit (QIAGEN, Beijing, China) and subjected to direct sequencing using the ABI-PRISM Dye Termination Cycle Sequencing ready reaction kit and a ABI-PRISM3730 genetic analyzer (Applied Biosystems, Carlsbad, CA, USA).
The genome sequences of hantaviruses were aligned using the ClustalW method implemented in the Lasergene program, version 5 (DNASTAR, Inc., Madison, WI). Nucleotide (nt) and amino acid (aa) sequence similarities were calculated using DNAStar. Phylogenetic trees for each segment were inferred using the Bayesian method implemented in MrBayes 3.1 [23] and the Maximum likelihood (ML) method available in the RAxML Blackbox webserver [24], employing the best-fit GTR+I+Γ model of nucleotide substitution as determined using jModeltest [25]. Trees were visualized with the TreeView software [26].
The GenBank accession numbers for the sequences obtained here are JQ912697 to JQ912907, and KC344236 to KC344269 (Table S2).
The first clinical HFRS case in Longquan was reported in 1974 (Figure 1). During the 38-year period between 1974 and 2011, a total of 1,866 HFRS cases were registered in this city. Only nine cases were recorded in the 1970s, such that the annual incidence of HFRS increased dramatically during 1980s and 1990s. A peak of 138 cases (51.2 cases/100,000 population) was reached in 1998, after which it decreased, likely in part due to the onset of hantavirus vaccination in 1997 and the intense rodent control efforts undertaken in China [7]. In total, more than 63,000 people have been vaccinated either by inactivated vaccines (Youerjian, Tianyuan Bio-Pharma, Hangzhou, China) for HTNV (during 1997–2000) or purified bivalent vaccine for HTNV and SEOV cultured in sand rat renal cells (Youerjian, Tianyuan Bio-Pharma, Hangzhou, China) or Vero cells (Royal, Royal (Wuxi) Bio- Pharmaceutical, Wuxi, China) (during 2001–2011). However, despite this vaccination the incidence of HFRS remained relatively high during 1999–2011, with between 11.15 and 23.6 cases/100,000 population.
During 1974–2011, a total of 20 patients died of HFRS in Longquan, with an average fatality rate of 1.07%. The highest fatality rates were observed during the first 10 year period (1974–1983), and reached 11% (10 fatal cases of 91 cases). Notably, all fatal cases occurred in autumn and winter. Additional fatalities were recorded in 1985 (1), 1986 (2), 1989 (1), 1992 (1), 1997 (1), 1998 (1), 2002 (1), and 2006 (2): these cases occurred in autumn and winter, with the exception of one death in March 2002. No patients have died of HFRS since 2007, likely reflecting improvements in disease treatment.
The seasonality of HFRS noted above may provide important clues to its cause [9]. We therefore analyzed the seasonality of HFRS in Longquan for different time periods during 1974–2011. HFRS cases occurred in winter (November to January) and in spring/summer (May to July) at respective frequencies of 49.66% and 14.51% during 1974–1990, 38.18% and 24.18% in 1991–2000, and 36.20% and 31.31% in 2001–2011 (Figure 2). As the peak of HFRS associated with rats occurred in the spring, whereas HFRS associated with mice occurred mainly in the winter [9], the recent increase in cases in spring/summer suggests a rising disease toll due to rat-associated hantavirus(es) in Longquan. A similar seasonal shift, from mice-dominated to rat-dominated transmission, has been reported in other HFRS endemic regions [27].
To analyze genetic diversity in the natural hantavirus reservoir and its relationship to those viruses found in humans, a total of 2,652 small mammals, representing 10 species of rodents and 3 species of insectivores (Table 1), were captured in Longquan during 2008–2011. A. agrarius mice and M. fortis voles were the dominant field species, accounting for 41.82% (1109) and 18.17% (482) of all small mammals collected, respectively. However, in residential areas the dominant species were rats of the family Rattus including 425 R. losea (16.02%), 372 R. norvegicus (14.03%), and 201 R. flavipectus (7.58%). Using IFA and RT-PCR, hantavirus antigens were detected in a total of 118 rodents including 78 A. agrarius (7.03%), 5 M. fortis (1.04%), 32 R. norvegicus (8.60%), and 3 R. flavipectus (1.49%). No hantaviruses were found in insectivores. Thus, the etiologic agents of HFRS cases in Longquan were likely hantaviruses carried by A. agrarius mice and R. norvegicus rats.
Serum samples from five human patients were collected on day 1 of hospitalization. Samples were tested for IgM and IgG antibodies by IFA using GOUV- or HTNV-infected cells (Table S1). Three serum samples showed higher IgM and IgG titers in HTNV- specific IFA, and one in GOUV-specific IFA. One sample showed higher IgM titers in HTNV-specific IFA, but with the same titers in HTNV- or GOUV specific IFA, suggesting cross-reactivity of HTNV with GOUV.
To further characterize the etiologic agents of human infection in Longquan, complete or partial hantavirus M segment sequences were recovered from 118 hantavirus antigen-positive rodent samples (Table S2). In addition, complete S segment sequences were amplified from all 118 hantavirus antigen-positive rodent lung tissues, and partial S segment sequences were recovered from the five human serum samples collected from patients with acute HFRS.
Notably, the sequences recovered from 78 A. agrarius, 5 M. fortis, and 4 human samples were very closely related to each other, with 98.2–100% nt and 98.4–100% aa sequence identities in the M segment and 98.6–100%/99.3–100% identities in the S segment. This similarity is indicative of direct viral transmission from rodents to humans. To determine the phylogenetic relationships among the viruses described here and to known hantaviruses, phylogenetic trees were estimating using the M and S segment sequences (in which Bayesian and ML methods produced similar topologies). Most notably, the sequences sampled from A. agrarius mice, M. fortis voles, and humans clustered together and formed a distinct and well-supported lineage in both trees (Figures 3–4). Interestingly, these strains also exhibited a close evolutionary relationship to strains HTNV and Z5 previously isolated from A. agrarius mice in Zhejiang Province [28].
All hantavirus sequences recovered from rats were very closely related to each other, with 97.5–100% nt and 98.5–100% aa sequence identities in the M segment and 98.5–100%/98.6–100% identities in the S segment. The partial S segment sequence recovered from the serum of a human patient in which high titers of IgG antibodies against GOUV had been detected (Table S1) was very closely related to sequences recovered from rats (98.9–99.5%/99.2–100%). Remarkably, these hantavirus sequences were closely related to previously described variants of GOUV – Gou3, ZJ5, YongjiaRf45 and YongjiaRn14 – but more distant from variants of SEOV. In phylogenetic trees of the M or S segments those strains from Rattus rats (R. flavipectus and R. norvegicus) and human clustered together, forming a distinct and strongly supported cluster (posterior probabilities of 1.0 for both M and S sequences) within the broader group of GOUV sequences (Figures 3–4). This phylogenetic pattern is indicative of a new genetic variant of GOUV in Longquan.
HFRS was a serious problem in China during the 1980 and 1990s [7], [9]. As a result of comprehensive preventive measures and improved living conditions, the incidence of HFRS in China has declined dramatically during the last decade [7]. Because of favorable ecological conditions and low socioeconomic status in rural areas, farmers have frequently been the major victims of HFRS, both inside and outside of China [5], [7], [9], [29]–[31]. The annual number of registered HFRS cases in Longquan has decreased, from 138 in 1998 to 46 in 2011, with a similar pattern observed in other parts of China [7], [32]. However, the incidence rate (>10 cases/100,000 population) in this region is still the highest in China despite ongoing vaccination. Considering the dramatic decrease in the rural population of Longquan in recent years (at least 30% of the rural population had moved into cities or towns by the end of 1990s), the real incidence rate of HFRS in rural areas may be much higher than reported here. In addition, the prevalence of hantavirus infection is high in rodents in both rural and residential areas in Longquan, especially GOUV in Norway rats (>8%). Thus, hantavirus infection will likely remain a major public health problem for the foreseeable future in Longquan city.
GOUV was first isolated from R. rattus captured in Zhejiang Province in 2000 [11], and initially considered as a variant of SEOV [10], [11]. However, GOUV is distinct from SEOV both serologically and genetically [11] and is found in a different rat species (R. rattus, R. flavipectus). Such distinction means that GOUV is currently defined as a tentative hantavirus species by the International Committee on Taxomony of Viruses (ICTV) [14]. In this study hantavirus variants originating from rats (R. flavipectus and R. norvegicus) from Longquan were most closely related to GOUV, forming a distinct and strongly supported lineage in both the M and S segment trees. Hence, these data suggest that the hantavirus variants carried by Rattus rats in Longquan represent a new genetic variant of GOUV. As no SEOV or other hantaviruses have been found in Rattus rats from Longquan and the sequences recovered from one patient in Longquan belonged to GOUV, our data clearly indicate that GOUV carried by Rattus rats (R. flavipectus and R. norvegicus) can cause human disease, and that there is ongoing spillover from the rodent reservoir to the human population. This is the first report of GOUV being associated with human HFRS cases since its discovery in 2000 [11]. Accordingly, further studies are needed to determine the pathogenicity and severity of GOUV in humans, as well as the possibility of human-to-human transmission.
Similar to other hantaviruses [2], [33], HTNV exhibits considerable genetic diversity and displays a geographic clustering of genetic variants, especially in mountainous regions [34]. To date, at least nine genetic lineages of HTNV have been found in Apodemus mice in Eastern Asia [34]. In this study, the virus sequences recovered from Apodemus mice and Microtus voles in Longquan formed a distinct lineage within HTNV in both the M and S segment trees, suggesting that a new genetic variant of HTNV is circulating in Longquan. Our earlier studies in the northeast and central parts of China documented Yuanjiang virus (YUJV) and Vladivostok virus (VLAV) in M. fortis voles, respectively [35]. However, these viruses were not detected in the Microtus voles from Longquan. Additional study is needed to determine if these viruses are present and whether they are pathogenic to humans in the HFRS-affected region.
Previous investigations revealed that HFRS caused by HTNV transmitted by Apodemus mice occurred mainly in winter, while the peak of HFRS caused by hantaviruse(s) transmitted by Rattus rats was in spring [9], [36], [37]. The seasonal analyses of HFRS cases performed here indicated that most of the HFRS cases registered in Longquan during 1974–1998 occurred in winter, in turn suggesting that human infections were due to HTNV. However, during the last decade the peak of HFRS has shifted to the spring in Longquan, with a similar pattern observed in other HFRS endemic regions [27]. In addition, GOUV was highly prevalent in R. norvegicus in residential areas, and more so than HNTV in Apodemus mice (prevalences of 8.60% and 7.03%, respectively). In sum, these data suggest that hantavirus(es) carried by rats may have become the major cause of HFRS in Longquan city over the past decade.
In conclusion, we have shown that hantavirus infection is endemic in both humans and rodents in Longquan, with the latter acting as a major reservoir for the former. Epidemiological and phylogenetic analyses indicate that GOUV and HTNV are circulating in local rodents and have a direct connection to the human population. As rats (Rattus species) are more mobile than the hosts of other hantaviruses [10], this study strongly reinforces the need for vigilance in preventing the spillover of GOUV from rats in China.
|
10.1371/journal.pcbi.1004065 | A Biologically Constrained, Mathematical Model of Cortical Wave Propagation Preceding Seizure Termination | Epilepsy—the condition of recurrent, unprovoked seizures—manifests in brain voltage activity with characteristic spatiotemporal patterns. These patterns include stereotyped semi-rhythmic activity produced by aggregate neuronal populations, and organized spatiotemporal phenomena, including waves. To assess these spatiotemporal patterns, we develop a mathematical model consistent with the observed neuronal population activity and determine analytically the parameter configurations that support traveling wave solutions. We then utilize high-density local field potential data recorded in vivo from human cortex preceding seizure termination from three patients to constrain the model parameters, and propose basic mechanisms that contribute to the observed traveling waves. We conclude that a relatively simple and abstract mathematical model consisting of localized interactions between excitatory cells with slow adaptation captures the quantitative features of wave propagation observed in the human local field potential preceding seizure termination.
| Nearly 50 million people worldwide suffer from epilepsy, a chronic neurological condition characterized by recurrent, unprovoked seizures. Although some clinical and biological principles of seizures are known, many aspects of spontaneous human seizures remain poorly understood. Recordings from electrodes placed directly on and within the brain provide a unique view of seizure activity, and have revealed specific brain voltage patterns associated with this pathological state. In particular, there is evidence that organized waves of activity propagate over the brain during a seizure. However, quantitatively characterizing and understanding the mechanisms that support these waves remains an open challenge. The goal of this work is to address this challenge through a combination of mathematical modeling and clinical recordings. Through this interdisciplinary approach, we seek to understand general features that support the spatiotemporal patterns of seizure termination. We propose that a relatively simple and abstract mathematical model consisting of localized interactions of closely neighboring excitatory cells with slow adaptation can support the propagation of the waves found in clinical recordings. Improved understanding of the mechanisms supporting seizure activity promises novel developments in treatment strategies tailored to the observed activity of individual patients.
| Epilepsy is a dynamical disease [1] that manifests in many ways, including as organized patterns of brain voltage activity during a seizure. In general, a patient’s epilepsy may be classified through established clinical and imaging procedures and, based on the classification, a treatment strategy may be developed [2]. Although pharmacological and surgical treatment of epilepsy often succeeds, the exact mechanisms that lead to different kinds of epilepsy and produce a seizure are still largely unknown; common proposed biological mechanisms include altered interactions between excitatory and inhibitory neurons [3, 4] and hyperexcitation [5]. Although the underlying mechanisms that initiate and support the seizure may widely vary [6], some manifestations of the seizure remain stereotyped, including clinical symptoms and voltage dynamics [2]. For human patients, one of the most common observations of brain activity during seizure consists of chronic voltage recordings. These invasive or noninvasive observations provide detailed spatiotemporal information about the in vivo voltage dynamics of spontaneous seizures. Invasive local field potential (LFP) recordings provide fine spatial resolution of brain voltage activity during seizure, and have recently led to new insights [7–9].
LFP recordings are thought to represent the active ionic and synaptic currents within a volume of cortical tissue; in this way, the LFP captures the aggregate activity of large neuronal populations [10–12]. In healthy and diseased brain tissue, wave-like spatiotemporal activity has been observed in the field activity of many systems including the olfactory system of invertebrates [13] and vertebrates [14, 15], turtle visual cortex [16–20], rat visual cortex [21–23], rat hippocampus [24], rat somatosensory cortex [25], monkey motor cortex [26], human motor cortex [27], and human retina [28].
Coordinated spatiotemporal activity is thought to serve a functional role in computation and communication between subsystems of the brain. For example, waves are thought to support synaptic modification during development, as observed in the visual system (e.g., [28, 29]). Although seizure activity is characterized by stereotyped voltage rhythms [30, 31] and coupling between rhythms across space [7, 32], the role of spatiotemporal patterns (e.g., waves [21]) remains an active research area [33, 34]. Moreover, the biological mechanisms that support these manifestations of seizure remain incompletely understood; further understanding these features promises improved therapies for epilepsy, in addition to a deeper understanding of organized neuronal population activity in brain function and dysfunction.
In addition to clinical and experimental recordings, computational models provide an alternative, powerful approach to investigate the biological mechanisms that support observed brain voltage activity. In general, the combination of experimental data and mathematical modeling has proved useful in understanding propagation dynamics in the brain. For example, experimental observations made in a cultured one-dimensional slice agree with a theoretical framework based on an integrate-and-fire model [35, 36], and the compression and reflection of visually evoked cortical waves [37] has been modeled in [38]. Both animal models (e.g., [39]) and computational models (e.g., [6, 40]) permit controlled, detailed observations of a given seizure process, and the ability to accurately manipulate this process. Importantly, unlike typical observations from clinical recordings, models permit a detailed accounting of the biological mechanisms that support the observed activity. However, the starting assumptions of a model oversimplify the biological processes of the in vivo brain (e.g., removal of a brain region from the surrounding network, or omission of some cell types). An exact relationship between these models and human epilepsy is often difficult to determine. Clinical observations and models therefore provide different insights into seizure activity. Clinical recordings provide accurate in vivo observations of spontaneous seizures from human patients, yet the biological mechanisms that support this activity remain predominantly unknown. Models provide detailed control and manipulations of candidate biological mechanisms, but the relationship to spontaneous seizures in humans remains unknown. Ideally, a unified procedure would exploit the advantages of each approach and mitigate the disadvantages. Implementing this type of procedure linking human clinical recordings to mechanisms in an abstract and simple mathematical model is one goal of this paper.
We propose to characterize invasive clinical voltage recordings from small regions of human cortex preceding seizure termination through comparison with a mathematical model. To do so, we simulate the cooperative synaptic transmembrane current found in clinical LFP recordings using a relatively simple and abstract mathematical mean field model. Mean field neural models, or neural fields, are used to represent coarse-grained variables in space, consisting of thousands of interconnected neurons (i.e., spanning approximately a few hundred micrometers) [41, 42]. Models of neural fields have a long history in computational neuroscience [21, 43–45], and have been successfully employed in many areas, including the study of spatiotemporal dynamics [46–51], with features such as periodic patterns [52], bumps and multi-bumps [53, 54], and waves [37, 44, 55–57]. Because these models are expressed as differential-integral equations, mathematical theory exists to rigorously analyze the model behavior. Here we undertake a mathematical analysis of a mean field model consistent with the observed LFP data to obtain the exact solution for traveling wave dynamics, and deduce parameter relationships that support wave propagation. We then constrain the model solutions using features of LFP recordings of traveling wave dynamics preceding seizure termination observed in a population of human subjects during seizure. In particular, by using the observed width and speed of the LFP waves we obtain parameter estimates consistent with known biological features of cortex, namely timescales and the synaptic connectivity profile. We show that a relatively simple mathematical model consisting of a population of excitatory neurons with localized interactions and an adaptation term is sufficient to mimic the observed LFP waves preceding seizure termination. In this way, the proposed framework links clinical recordings with mathematical models to propose candidate mechanisms supporting a poorly understood aspect of seizure activity: the spatiotemporal dynamics preceding seizure termination in a small patch of human cortex.
Our goal is to isolate and characterize in a relatively abstract mathematical model the mechanisms that support the emergence of traveling wave dynamics preceding seizure termination. To do so, we first characterize these dynamics as observed in invasive brain voltage recordings from a population of human subjects during seizure. We show that stereotypical traveling wave patterns emerge in the LFP with consistent quantitative features. Then, we implement an activity-based mathematical model of neural population dynamics. We obtain explicit traveling wave solutions for the model together with conditions that ensure the existence of a wave of given speed and width. We then further constrain the model parameters using the wave features observed in the in vivo LFP data. Finally, we use these model results to propose candidate mechanisms that support the observed traveling wave activity preceding seizure termination.
The mechanisms that produce organized neuronal population activity are extremely complex [58]. In an effort to characterize and understand the neuronal population activity observed in the clinical recordings preceding seizure termination, we implement here a relatively simple neural field model [59]. The biophysical basis for these types of models are understood by considering the interaction of a finite number of synaptically coupled neurons. Many different formulations for neural fields exist [60], with implications for the interpretation of the model variables and parameters. These different mathematical formulations of neural field models can be broadly separated into two categories: voltage-based formulations, and activity-based formulations [59]. In a voltage-based model, the time scale of the dynamics is related to the membrane properties of the post-synaptic cells, while in an activity-based-model, the time scale of the dynamics is related to the synaptic decay [59]. We choose the latter formulation here. In its simplest form, the activity-based model is one of the most basic models to arise in mathematical neuroscience [61]. Beyond this simple form, activity-based models have been extended to include additional features (e.g., absolute refractoriness [41, 62]). In addition, the activity-based model is consistent with the notion that the LFP dynamics are dominated by the time scale of synaptic effects [10, 63], and activity-based models have been proposed as more realistic than voltage-based models [64, 65]. We note that most mathematical analysis of neural field models utilizes the voltage-based formulation [44, 65, 66]. In particular, in [67] the authors performed a complete analysis of the existence and stability of traveling wave solutions in the voltage-based formulation. To the best of our knowledge, a mathematical analysis of traveling wave existence and stability in an activity-based model with adaption has not been performed.
We now develop a one-dimensional model to describe important features of the neuronal population activity observed in vivo. The choice of a one-dimensional model is motivated by the observation that a majority of traveling waves observed in the LFP recordings travel in approximately one-dimension, with features as described in the previous section. To simplify the model, we consider only a single population of excitatory neurons. In doing so, we will show that—in the mathematical model—inhibitory neurons are not required to mimic features of the observed LFP data immediately preceding seizure termination. To prevent the activity from remaining in a permanent excited state, which will give rise to a front solution (see Methods), we include an adaptation term that directly regulates the activity. This adaptation accounts for a natural process that will drive the population activity back to a rest state. From the mathematical point of view, adding this adaptation term permits traveling pulse solutions in the model consistent with key features of the clinical recordings. As we describe, using this relatively abstract and simple activity-based model with an adaptation term, we are able to replicate the reverberation observed in the LFP recordings.
The specific neural field model we employ is
ut(x,t)=−αu(x,t)+αH(12σ∫−∞+∞e−∣x−y∣σu(y,t)dy+P(x,t)−k) −αβ0q(x,t)qt(x,t)=δu(x,t)−δq(x,t),(1)
where u(x, t) is the mean synaptic activity, q(x, t) is the adaptation, and P(x, t) is an external input, all evaluated at position x and time t. In particular, we consider that u(x, t) represents the activity of a cortical column with extent less than 20 μm situated at position x and time t. We interpret u(x, t), a dimensionless quantity, as the deviation from a baseline of activity. Therefore, u(x, t) = 0 represents a resting state of activity, and negative values represent a depression of resting activity [41]. We note that “negative activity” (i.e., a reduction in activity below the baseline rate) in one region reduces the input received in neighboring regions. In this formulation, we interpret the adaption term, q(x, t), as representing a local homeostatic regulation mechanism that evolves on a slower timescale than u(x, t) and acts to maintain the activity near a target baseline. When the activity u(x, t) falls below the baseline value (i.e., u(x, t) < 0), the adaption q(x, t) decreases which acts to increase u(x, t). Conversely, when the activity increases above baseline (i.e., u(x, t) > 0), the adaption q(x, t) increases and acts to decrease u(x, t). We note that homeostatic regulation mechanisms act on a variety of timescales, including relatively short timescales (on the order of seconds) [68]. H is the Heaviside function, which becomes non-zero when the synaptic input exceeds a synaptic threshold k:
H ( x − k ) = { 1 if x ≥ k 0 if x < k .
We note that the adaptation term in (1) is located outside of the Heaviside function. In this phenomenological model with a simple adaptive scheme, the adaptation term acts as a local feedback mechanism to depress the synaptic drive. This model is motivated by the linear negative feedback proposed in [44]. We note that, in voltage-based models, different formulations for adaption exist; these include negative feedback both inside the threshold function [44, 51, 69] and outside of the threshold function [49, 53]. We show in S1 Text of Supporting Information that the model (1) updated to include the adaption term inside of the Heaviside function does not produce damped oscillations; instead, the traveling wave solution returns monotonically to rest after excitation. This monotonic evolution is inconsistent with the reverberation observed in the LFP data of interest here (examples in Fig. 2).
There are 5 parameters in the model (1). Each possesses a biological interpretation: α is the decay rate parameter for the synaptic activity term, δ is the decay rate parameter for the adaptation term, σ is the spatial rate of decay of connectivity, k is the synaptic input threshold, and β0 accounts for the strength of the adaptation term on the synaptic dynamics. For simplicity we set β = α β0. Both time and space units were scaled to represent milliseconds and microns, respectively (see Methods). There are two additional parameters that we employ in the subsequent analysis: c is the wave speed, and w is the wave width. These parameters are not directly specified in the model, but instead are features of the traveling wave dynamics.
Our goal is to identify the parameter configurations that support traveling waves in this model consistent with the observed LFP activity. In particular, we are interested in solutions that support only one extremum of high amplitude activity, so called pulses, as these have been characterized using the LFP data. To that end, we first determine under what parameter configurations traveling waves of high amplitude activity exist in the model. To do so, we rewrite the equations in a moving coordinate frame z = x-ct; this frame is moving with a constant speed c. By identifying the stationary solutions of this system, we determine solutions that move with a constant speed c, and a constant width w, without changing their shape: so called traveling waves. Depending on the model parameters, we find that the linearization of the associated system in the moving coordinate frame consists of either purely real or complex eigenvalues. The explicit traveling wave solutions for both the real and imaginary case are now considered. We state the solutions here; detailed analysis may be found in Methods.
The mathematical model (1) contains five free parameters: α, δ, σ, β and k. In the previous section, we began restricting these parameters by establishing relationships between parameters that support traveling wave solutions. In particular, by fixing the time scales α and δ, together with a choice of speed c and width w deduced directly from the LFP data and hence constrained by the clinical observations, we may solve for the remaining parameters β, σ, and k. The matching conditions establish a relationship between σ and β (example in Fig. 7), and by choosing β and σ we can solve for the corresponding k, as described in the previous section. We now proceed to use the “reverberation” observed in the clinical data (examples in Fig. 2) to estimate the parameter β for each wave. In doing so, we will have used the clinical data and biophysical intuition to constrain further the model parameters.
Visual analysis of the in vivo LFP data shows that high amplitude pulses are followed by a reverberation, i.e., a secondary, smaller amplitude increase in activity (for more details, see Methods). Due to the nature of the traveling wave solutions, this feature is only present in the complex eigenvalue solution, i.e., when damped oscillations follow the pulse of high amplitude activity (example in Fig. 5b); we propose that the damped oscillations following the main pulse of the traveling wave mimic the reverberations observed in the LFP recordings. Hence, we restrict the following analysis to the complex eigenvalue case. We use the reverberation times estimated from the LFP data to fix the parameter β for each wave; we label these estimates βempirical. To do so, we set the periodic portion of the complex eigenvalue solution to possess a period consistent with the observed reverberation: given a reverberation time τ (example in Fig. 8), then β e m p i r i c a l = ( δ − α ) 2 4 δ + 4 π 2 δ τ 2 (see Methods). In this way we constrain the model to replicate the period of the secondary bump (i.e., reverberation) present in the data (Fig. 8). Having done so, the model parameters β, σ, and k are now directly determined for each observed LFP wave.
As a final illustration of the suitability of the model, we consider an example numerical simulation of the model (1) (see Methods). To do so, we choose a particular wave from the LFP data of Seizure 1, estimate c and w directly from the data, and fix α = 7.5/s, as for this value of α non-trivial parameters from both the real and complex eigenvalue solutions can be obtained from the matching conditions. Following an initial stimulus (5 ms initial input at position 0 μm) the model produces a traveling pulse that is followed by a smaller amplitude reverberation. A comparison of a wave from the clinical recordings with the real and complex eigenvalues case is shown in Fig. 11. We note that both simulations accurately replicate features of the observed LFP wave (namely, the speed and width), but that the complex eigenvalue case solution also produces a secondary bump of activity consistent with the reverberation in the observed LFP wave. We also note that, in the model, the activity decreases below 0 between the mean crest of the traveling wave and the subsequent reverberation of activity in Fig. 11(c). A decrease in activity also appears in the in vivo data between the crest of the traveling wave and the reverberation (example in Fig. 11(a)); however, this decrease is smaller in magnitude than that produced in the model. An updated model that includes inhibition helps reduce this discrepancy, as illustrated in the next subsection.
The original model formulation (1) is analytically tractable and capable of reproducing important features of the observed traveling wave dynamics. However, as expected, this relatively simple model exhibits some inconsistencies with the in vivo data, for example the large negativity following the traveling wave crest.
Increasing the complexity of the model through the addition of more biological features may help reduce these inconsistencies. To that end, we consider an updated model that includes an inhibitory population. In particular, we implement the following system:
u t ( x , t ) = − α e u ( x , t ) + α e H ( g e e ⊗ u ( x ) − g i e ⊗ v ( x ) + P ( x , t ) − k e ) − α β 0 q ( x , t ) q t ( x , t ) = δ u ( x , t ) − δ q ( x , t ) v t ( x , t ) = − α i v ( x , t ) + α i H ( g e i ⊗ u ( x ) − g i i ⊗ v ( x ) + Q ( x , t ) − k i ) , (2)
where u(x, t) is the mean synaptic activity of the excitatory population, v(x, t) is the mean synaptic activity of the inhibitory population, q(x, t) is the adaptation term in the excitatory population, and P(x, t) and Q(x, t) are external inputs to the excitatory and inhibitory populations, respectively. The convolutions account for the spatial extent of the synaptic connectivities,
g j k ⊗ w ( x ) = g ¯ j k 1 2 σ j k ∫ − ∞ + ∞ e − ∣ x − y ∣ σ j k w ( y , t ) d y ,
where j = {e, i}, k = {e, i}, and g ¯ j k = { 0 , 1 }. H is the Heaviside function, which becomes non-zero when the total input exceeds the threshold kj.
To characterize the behavior of this model, we perform numerical simulations. We set the parameters to match the wave speed and width used for the original model (1) in Fig. 5b, and fix αi = 2.5/s, ki = 1, σei = 20 μm, σie = 20 μm, and σii = 0. We first consider the case g ¯ e i = 0, g ¯ i e = 0 and g ¯ i i = 0 so that the excitatory and inhibitory populations do not interact. The resulting wave profile (Fig. 12a) reveals a large amplitude pulse, followed by a deep depression of activity, and then a smaller amplitude reverberation, as expected for the original model formulation (1). Then, using the same parameter settings, we activate interactions between the excitatory and inhibitory populations (g ¯ e i = 1, g ¯ i e = 1, g ¯ i i = 1). The resulting wave profile (Fig. 12b,c) exhibits qualitative differences from those in the original model; by including inhibition, the wave profile becomes smoother and thinner, and the depression of activity following the large amplitude pulse is shallower. These results suggest that a neural field model with adaptation and inhibition produces wave profiles with additional features consistent with the in vivo data, including a smoother wave profile and a shallower depression of activity following the main pulse. We conclude that the original model (1), even in the absence of inhibition, supports wave propagation as observed in the clinical recordings. However, incorporating additional biological features in the model - such as inhibition - may improve fidelity with the clinical data.
In this paper, we considered invasive local field potential (LFP) recordings from a population of human patients during seizures. We showed that, in the late stages of seizures, spatiotemporal patterns of activity propagate across a small patch of cortex. These patterns can be well approximated as one-dimensional plane waves, and we characterized important features of these waves (i.e., the speeds and widths). We found traveling wave speeds of ≈ 80 380 μm/ms, consistent with the propagating velocity of a pulse when GABAergic local inhibition is blocked (e.g., 60–90 μm/ms in [70], 70 μm/ms in [71], 130–190 μm/ms in [25], and 120–150 μm/ms in [72]). In addition, we examined the features of small amplitude “reverberations” in the voltage activity following each wave.
To further characterize the observed LFP waves, we implemented a relatively simple neural field model consisting of an excitatory population of cells with adaptation. This abstract mathematical model is flexible enough to replicate important features of wave propagation near seizure termination for the population of patients and seizures. Moreover, the relative simplicity of the model permits analytic solutions; we showed here, for the first time, that traveling wave solutions exist and are stable in this activity-based model formulation with adaptation. In addition, the model parameters permit biophysical interpretation (e.g., as the extent of synaptic connectivity). By combining analytic model solutions with features of the observed waves - such as the speed and width - we estimated parameters in the model. The estimated parameters included the timescales of activity and adaptation, and the spatial extent of the connectivity. We find that the timescale of the model consistent with the observed LFP data is biologically reasonable: the adaption is an order of magnitude slower than the activity. Measures of synaptic connectivity in a local neighborhood of cortical tissue have been reported to range from 40 μm to 2 mm [12, 41, 63, 73–75]. For the deduced range of parameters obtained in this study, we find that the extent of connectivity, σ, for Patients 1 and 2 coincides with this established range. For Patient 3, we obtain connectivities between 60 μm to 4 mm, which is larger, but not wholly inconsistent with existing estimates. We find for all three patients that the parameter β0, which is the strength of the adaptation, is between 2 and 4; and the parameter k, which accounts for the synaptic threshold, is between 0.12 and 0.2. The variability in the estimates of σ, β0 and k may reflect changing biophysical features during seizure (e.g., progressive changes in synaptic efficacy or changes in the extracellular environment) as well as the variability inherent in measuring a noisy biological system. We also note that for the three patients, as the timescale of the activity increases, the extent of the connectivity decreases (Fig. 10) suggesting that faster activities (large α) require less distant connectivity. Finally, we note that the parameter estimates are consistent both within individual patients, and across the population of patients and seizures. We conclude from these results the following hypothesis: plane waves observed in vivo late in human seizure can be supported in a relatively simple mathematical model without inhibition, consistent with in vitro slice and theoretical work (e.g., [25, 36, 70–72, 76–78]). However, we note that inclusion of inhibition may improve features of the model (e.g., may better mimic aspects of the wave profile, see Fig. 12 and S2 Text in Supporting Information for additional illustrations).
The analysis and modeling focused on an interval preceding seizure termination, in which the data have transitioned to large amplitude spike-and-wave (or spike-and-polywave) oscillations. A goal of this modeling study was to simulate some of the spatiotemporal aspects of this spike-and-wave activity. Animal studies suggest the mechanisms that support this spike-and-wave activity are complex. Some studies have suggested that the “wave” component of the spike-and-wave oscillation reflects inhibitory GABAergic processes [79–81]. However, other animal studies instead propose that slow intrinsic currents (e.g., a calcium-activated potassium current) support the “wave” component of the spike-and-wave oscillation [82–87], and in vitro slice experiments indicate that features of wave propagation (i.e., wave velocity and wave amplitude) during epileptiform activity do not depend on inhibition [88]. In addition, during seizures with spike-and-wave oscillations, neural populations are (at least transiently) highly active and thereby drive large changes in intra- and extracellular ion concentrations (e.g., intracellular chloride accumulation and extracellular potassium accumulation) [89]. This may result in pathological changes in brain dynamics, for example the reversal potential of GABA-receptor-mediated inhibitory postsynaptic potentials may shift to positive values [85], and inhibitory mechanisms may engage in the generation of the depolarizing component of spike-and-wave oscillation.
Here we have implemented a mathematical model with a tight focus on one aspect of the late seizure interval: the (approximately) one-dimensional traveling waves that appear in spike-and-wave oscillations near seizure termination. In doing so, we have presented a modeling formulation more consistent with the proposed intrinsic current mechanisms of spike-and-wave oscillations. Nevertheless, we suspect that inhibition plays a fundamental role in seizure, for example at seizure onset [90, 91] when fast-spiking interneurons are highly active. We expect that the addition of more biophysical features to the model (including inhibition) will permit a better match to the in vivo LFP data (see Fig. 12 and S2 Text of Supporting Information), at the cost of increased model complexity and reduced analytic tractability.
In this work we implemented a relatively simple one-dimensional neural population model, consisting of a synaptic activity variable and an adaptation variable. The simplicity of the model allows rigorous mathematical analysis, although the biophysical mechanisms remain relatively abstract. The validity of the model is based on the reproduction of wave features present near seizure termination, and parameter estimates consistent with known physiology (i.e., estimates of synaptic connectivity and difference in timescales). The purpose of this model is not to capture the detailed biophysical mechanisms of seizure, as in more realistic computational models [92, 93]. However, we may use the mathematical model to make the following prediction: the traveling waves near seizure termination represent relatively “simple” brain phenomena. Consistent with this notion, we hypothesize that the diversity of complex components that support normal cortical function (e.g., the diversity of inhibitory neuronal populations [94, 95]) have shut down, and allowed these simple dynamics to dominate. Restoration of this diversity and complexity (e.g., activation of silenced inhibitory neuronal populations) would then help disrupt these pathologically organized and simple traveling waves.
To further validate the model results, in vitro experiments that reproduce important features of the human in vivo data (e.g., the spectrographic properties [90, 96]) would allow detailed pharmacological exploration of the proposed biophysical mechanism of this model. In particular, the more abstract model parameters (like β0, the strength of the adaptation) may be better understood in terms of specific neuronal mechanisms through experiments in controlled biological systems. These experiments may in turn motivate future work developing more biologically detailed models to provide additional insight into the spatiotemporal dynamics of seizure activity. One important future modeling direction is the further analysis and inclusion of inhibitory populations in this activity-based formulation. Such inclusions may further illuminate the mechanisms of wave propagation, and might help to explain differences in waves seen during the initial and terminal stages of human seizure.
We have focused here on the analysis of the observed LFP plane waves near seizure termination. Rich spatiotemporal patterns also emerged in the clinical LFP data throughout the seizure (and perhaps in other functional states, such as sleep) and will require an expanded two-dimensional model for characterization. For example, we note that near seizure onset complex spatiotemporal patterns emerge, without obvious traveling wave dynamics. The mechanisms that govern the transition from these disorganized spatiotemporal dynamics to more organized traveling waves remain unknown. The analysis of seizures from more patients may help to develop more sophisticated - and biologically detailed models - to explain these complex phenomena. The combination of quantitative data analysis and mathematical modeling of seizure activity across space remains an active research area with important implications for improved treatment of epilepsy.
All patients were enrolled after informed consent was obtained and approval was granted for these studies by local Institutional Review Boards.
For each patient and seizure, we analyzed a subset of the diverse spatiotemporal patterns observed approaching seizure termination. We focus here on the analysis of one-dimensional plane waves of activity, which were the most common type of wave we observed in Patients 1 and 2 (Seizure 1, 36 out of 40 waves; Seizure 2, 36 out of 41; Seizure 3, 39 out of 59; Seizure 4, 26 out of 33; Seizure 5, 35 out of 52). Upon visual inspection, the excluded waves exhibited different spatiotemporal patterns, including disorganized waves of high activity, and two-dimensional patterns, such as waves that initiated at the center of the microelectrode array, and spiral waves. Again, we focus here only on the one-dimensional plane waves and estimates of the model parameters from these waves. For Patient 3, we focused on a contiguous half (2 mm by 4 mm) subsection of the entire (4 mm by 4 mm) microelectrode array. For this patient, we were able to detect waves moving closer to the horizontal direction (from −45° to 45° and from 135° to 225°). Having selected these one-dimensional waves from the three patients, all waves were analyzed using the same set of data analysis algorithms described below. Components of these data may be made available by request to the corresponding author.
The purposes of the data analysis were: i) To obtain a time interval for the propagation of each planar wave; ii) To obtain the direction of wave propagation; iii) To obtain the different one-dimensional paths through the two-dimensional microelectrode array for a given direction; iv) To obtain the speed, width, and reverberation time along each one-dimensional path; and v) To obtain the mean speed, mean width and mean reverberation time for each wave across different paths. To determine the time interval for the propagation of each planar wave, we computed the gradient of the LFP activity at each moment in time. The gradient assigns to each spatial location a vector specifying the direction and magnitude of maximal increase in activity (Fig. 13a). To compute the gradient, we analyzed voltage differences between adjacent electrodes. A histogram of the angles of the gradient at each position, weighted by the magnitude of the gradient, was then constructed for each moment in time (Fig. 13b). We label t0 the time at which the LFP z-scored signal at the center of the microelectrode array exceeded a threshold of 2.5. We then determined the peak of the unimodal angle distribution at time t0, which we labeled θ0. We considered angles between θ0-20 and θ0+20 degrees and analyzed the proportion of angles within the interval (θ0-20, θ0+20), forward and backwards in time starting at t0. The time tinitial denotes the first time at which the number of counts in the angular interval becomes non-zero. The time tfinal is the last time at which counts appear in the angular interval. In this way, each wave is assigned a time interval (tinitial, tfinal) for which angles appear in the interval (θ0-20, θ0+20). In this time interval, the weighted histograms of the angles showed a clear organization of the gradient directions and appearance of two peaks in the histogram distributions (Fig. 13b). These two peaks account for the preferred angle before the wave enters the microelectrode array and after the wave exits the microelectrode array. To determine the direction of each wave we focused on the first peak (Fig. 13b). This peak typically occurs in the time interval (tinitial, t0). In addition, we visually inspected each peak and verified that the associated angle accurately described the direction of propagation for each wave. The notions of t0, tinitial, tfinal and θ0 are illustrated in Fig. 13c.
Having determined the angle at which LFP activity propagated, we then constructed one-dimensional paths spanning the microelectrode array. Each path consisted of 10 adjacent electrodes and ran parallel to the direction of the observed wave. Along each such path we determined the speed and width of the wave. For each path, we determined the time at which the activity at each electrode exceeded a threshold of one standard deviation above the mean LFP computed for the entire duration of seizure termination investigated. In this way, every electrode along a path was assigned a time of wave onset, which was used to compute the speed. We used all possible combinations of the 10 electrodes along each one-dimensional path to compute the speed, resulting in a total of 45 estimates of speed. To mitigate the impact of outliers, the speed for each one-dimensional path was then calculated as the median of the 45 speed estimates. We then estimated the speed for each wave as the mean speed among the different one-dimensional paths. Depending on the direction of the wave, from the 10 electrodes that form a one-dimensional path, there is one electrode at which the large amplitude activity of the wave reaches last, and we label this the “last electrode” (example in Fig. 14). To measure the wave width, for each one-dimensional path we computed the time at which the activity at the last electrode exceeded a threshold of 2.5 of the LFP z-scored signal. At that instant in time, the activity of the other electrodes along the path was also determined. The location at which the activity transitioned from above the threshold (of 2.5 of the LFP z-scored signal) to below the threshold was determined. The spatial extent from the last electrode to this transition point on the one-dimensional path defined the width of the wave. An illustration of the wave width determination is shown in Fig. 14. We note that if no electrode along the one-dimensional path transitioned to below the threshold, then the wave covered the entire spatial extent of the path, and the width of the wave indicates a lower bound. For each wave, the width refers to the mean widths obtained from all one-dimensional paths. To obtain the reverberation time we first determined the time at which the large amplitude wave of activity fell below a threshold of 0.5 of the LFP z-scored signal; we consider this time as the “end” of the primary traveling wave. Starting from this time point, we then determined the time for the activity to first exceed a reverberation threshold, defined as 0.5 of the LFP z-scored signal, and then for the activity to decrease again below this threshold. This decrease below the reverberation threshold defined the reverberation time. For an illustration of the reverberation time, see Fig. 15. We computed the reverberation time for each electrode along the one-dimensional path. The mean among the different one-dimensional paths gave the reverberation time of each wave. Using a t-test for small samples we computed a 90% confidence interval for the mean speed and mean width of each wave (Fig. 3), where the number of samples was given by the number of one-dimensional paths existent for each wave.
In the section, we describe in detail the mathematical analysis of the model (1). We note that the model (1) supports traveling front solutions when the adaptation term is removed. However, these front solutions are not consistent with observed LFP activity, and therefore not examined here.
As mentioned in Results, we use the moving frame z = x-ct and identify stationary solutions in this frame. These solutions will be of the form u(x, t) = u(x-ct, t) = u(z, t) and q(x, t) = q(x-ct, t) = q(z, t), such that ut(z, t) = 0 and qt(z, t) = 0. We use the connectivity function w ( z ) = 1 2 σ e − | z | σ. By making this change of variables, we obtain the system of differential-integral equations
− c u ′ ( z ) = − α u ( z ) + α H ( ∫ − ∞ ∞ w ( z ¯ − z ) u ( z ¯ ) d z ¯ − k ) − β q ( z ) − c q ′ ( z ) = δ u ( z ) − δ q ( z ) ,
which can be rewritten in the form
( u ′ ( z ) q ′ ( z ) ) = ( α / c β / c − δ / c δ / c ) ( u ( z ) q ( z ) ) + ( − α c H ( ∫ − ∞ ∞ w ( z ¯ − z ) u ( z ¯ ) d z ¯ − k ) 0 ) . (3)
We assume c > 0 which corresponds to a rightward moving wave. An analogous consideration holds for leftward moving waves (c < 0). We note that the nonlinear part of system (3) will be either zero or nonzero depending on the Heaviside function. For that reason the system can be analyzed by considering when the Heaviside function is zero (Case 1), and when the Heaviside function is non-zero (Case 2). We consider both cases below.
In order to ensure the continuity of the solutions, we look at the change points from Case 1 to Case 2. In particular, k = 1 2 σ ∫ − ∞ + ∞ e − ∣ x − y ∣ σ u ( y , t ) d y at the points x = 0 and x = w. This assumption gives rise to the matching conditions. Once the explicit traveling wave solutions are obtained, it is possible to solve for the exact value of the threshold k given by the matching conditions. We list below the solutions for the matching conditions in the case of real eigenvalues and complex eigenvalues.
The linear stability of the traveling wave solutions was analyzed in detail in [97]; here, we summarize these results. To study the linear stability of the traveling wave solutions we construct a complex-valued Evans functions whose zeros determine the eigenvalues associated with the stability of the wave [98]. By obtaining the eigenvalues it is possible to determine stability (or instability) of the linearized wave. Using the Evans functions, we explore the stability of wave solutions for parameter choices restricted by the LFP data. We have shown that for some parameter settings two wave solutions exist (e.g., Fig. 6). We note that one of these wave solutions is slow and narrow, whereas the other solution is fast and wide. Moreover, the fast and wide wave has speed and width consistent with the LFP data (as illustrated in Fig. 11). Using the Evans function we find that, in the case of the fast and wide wave, the associated eigenvalues consist of eigenvalues with negative real part and the trivial zero eigenvalue (due to the translation invariance of the wave solution); this implies linear stability of the fast and wide wave. In the slow and narrow wave case, we find a positive eigenvalue (purely real) in addition to the zero eigenvalue, implying linear instability of the wave solution. For more details, please see S3 Text of Supporting Information.
Space was discretized using 2000 points, to represent the length of a one-dimensional path. To each of these points the differential equation system (1) was associated. Numerical simulations were written to solve these systems using a Runge-Kutta method of order four with Δt = 0.005 ms. Convolutions integrals were approximated by assuming the activity was fixed within a Δx interval, where Δx represented a distance of 40 μm. Smaller grids were also examined of Δx = 20 μm, and Δx = 10 μm, and similar results found (not shown). The waves were created by applying a 5 ms input to points in space representing 10 μm. Both time and space were rescaled in order to have units of distance x in microns and time t in milliseconds.
|
10.1371/journal.pcbi.1002724 | MOSAIC: A Multiscale Model of Osteogenesis and Sprouting Angiogenesis with Lateral Inhibition of Endothelial Cells | The healing of a fracture depends largely on the development of a new blood vessel network (angiogenesis) in the callus. During angiogenesis tip cells lead the developing sprout in response to extracellular signals, amongst which vascular endothelial growth factor (VEGF) is critical. In order to ensure a correct development of the vasculature, the balance between stalk and tip cell phenotypes must be tightly controlled, which is primarily achieved by the Dll4-Notch1 signaling pathway. This study presents a novel multiscale model of osteogenesis and sprouting angiogenesis, incorporating lateral inhibition of endothelial cells (further denoted MOSAIC model) through Dll4-Notch1 signaling, and applies it to fracture healing. The MOSAIC model correctly predicted the bone regeneration process and recapitulated many experimentally observed aspects of tip cell selection: the salt and pepper pattern seen for cell fates, an increased tip cell density due to the loss of Dll4 and an excessive number of tip cells in high VEGF environments. When VEGF concentration was even further increased, the MOSAIC model predicted the absence of a vascular network and fracture healing, thereby leading to a non-union, which is a direct consequence of the mutual inhibition of neighboring cells through Dll4-Notch1 signaling. This result was not retrieved for a more phenomenological model that only considers extracellular signals for tip cell migration, which illustrates the importance of implementing the actual signaling pathway rather than phenomenological rules. Finally, the MOSAIC model demonstrated the importance of a proper criterion for tip cell selection and the need for experimental data to further explore this. In conclusion, this study demonstrates that the MOSAIC model creates enhanced capabilities for investigating the influence of molecular mechanisms on angiogenesis and its relation to bone formation in a more mechanistic way and across different time and spatial scales.
| The healing of a fracture largely depends on the development of a new blood vessel network (angiogenesis), which can be investigated and simulated with mathematical models. The current mathematical models of angiogenesis during fracture healing do not, however, implement all relevant biological scales (e.g. a tissue, cellular and intracellular level) rigorously in a multiscale framework. This study established a novel multiscale platform of angiogenesis during fracture healing (called MOSAIC) which allowed us to investigate the interactions of several influential factors across the different biological scales. We focused on the biological process of tip cell selection, during which a specific cell of a blood vessel, the “tip cell”, is selected to migrate away from the original vessel and lead the new branch. After showing that the MOSAIC model is able to correctly predict the bone regeneration process as well as many experimentally observed aspects of tip cell selection, we have used the model to investigate the influence of stimulating signals on the development of the vasculature and the progression of healing. These results raised an important biological question concerning the criterion for tip cell selection. This study demonstrates the potential of multiscale modeling to contribute to the understanding of biological processes like angiogenesis.
| The biological process of fracture healing comprises three main stages: (i) the “inflammation phase”, where the trauma site becomes hypoxic and is invaded by inflammatory cells, fibroblasts, endothelial cells and mesenchymal stem cells [1]; (ii) the “reparative phase”, which starts with the production of cartilaginous and fibrous tissue resulting in a soft callus, later replaced by a hard callus, through the process of endochondral ossification; (iii) in the final “remodeling phase” the woven bone is replaced by lamellar bone and the vasculature is reorganized.
The healing of a fracture depends largely on the development of a new blood vessel network (angiogenesis) in the callus. Sprouting angiogenesis involves the following steps: first a “tip cell” is selected; this cell extends filopodia sensing the haptotactic and chemotactic cues in the environment and leads the newly formed “sprout” comprised of following, proliferating “stalk cells”; the newly formed sprout, or “branch” then connects with another branch in a process called anastomosis, which results in the formation of a closed loop allowing the initiation of blood flow; finally the newly formed vascular network is stabilized by pericytes [2].
In order to ensure a correct development of the vasculature, the balance between stalk and tip cell phenotypes must be tightly controlled. The process of tip cell selection consists of the following main steps. Firstly a gradient of vascular endothelial growth factor (VEGF) is formed by the up-regulation of VEGF-expression and secretion, triggered by hypoxia (low oxygen concentration). The VEGF-mediated activation of the VEGFR-2 receptors induces the up-regulation of Dll4 which activates the Notch1-receptors on the neighboring cells, thereby down-regulating their expression of VEGFR-2. This process of lateral inhibition, with cells battling to inhibit each other leads eventually to a “salt and pepper” alternating pattern, where cells with high Dll4 levels remain with high VEGFR-2 receptor levels, allowing them to migrate (and becoming tip cells) whilst their neighbors become inhibited, making them less susceptible to VEGF, and thus these adopt the non-migratory stalk cell phenotype. In this manner the adequate amount of tip cells, required for a correct sprouting pattern, is established [2]–[5].
Both fracture healing as well as angiogenesis are very complex biological processes involving the coordinated action of many different cell types, biochemical and mechanical factors across multiple temporal and spatial scales. The time scales of individual events that underlie biological processes range from seconds for phosphorylation events to hours for mRNA transcription to weeks for tissue formation and remodeling processes [6]. The spatial scales vary from nanometers at the molecular level to millimeters at the tissue level and meters at the level of the organism [6], [7]. Thus, it can be concluded that most biological processes have an intrinsic multiscale nature and must be studied and modeled accordingly.
Depending on the biological spatial scale of interest a variety of experimental and modeling approaches can be used, which are nicely summarized by Meier-Schellersheim et al. [7]. The modeling approaches can be arranged in two broad categories: continuum and discrete modeling techniques. Continuum models use ordinary differential equations (ODEs) or partial differential equations (PDEs) to describe the evolution of cell and tissue densities and protein concentrations. The model variables are averages, which makes it difficult to represent individual cell-cell and cell-matrix interactions [8], [9]. Moreover, since the cells are not individually represented, it is challenging to model the individual intracellular processes. Also, continuum models fail to correctly capture the process of angiogenesis due to the inherent discreteness of vascular networks [10]. Discrete approaches are often used to study small-scale phenomena, e.g. biological processes at the cellular and subcellular levels [11]. However, these techniques often become computationally expensive when used for predictions of larger cell population sizes at the tissue scale [11].
Here we briefly review hybrid, multiscale models of angiogenesis, i.e. models that combine different modeling techniques for various scales mentioned above into one framework, as this is the approach we have adopted here. For comprehensive reviews on (multiscale) mathematical models of angiogenesis the reader is referred to Mantzaris et al., Qutub et al. and Peirce [12]–[14]. Notably none of the hybrid models to date include Dll4-Notch1 tip cell selection.
Milde et al. presented a deterministic hybrid model of sprouting angiogenesis where a continuum description of VEGF, MMPs, fibronectin and endothelial stalk cell density is combined with a discrete, agent-based particle representation for the tip cells [15]. The hybrid model describes the biological process of sprouting as the division of tip cells depending on chemo- and haptotactic cues in the environment and a phenomenological “sprout threshold age”. In the model of Lemon et al. the formation of a new branch also occurs at the tip cell position, but is modeled as a random process with an average number of branches per unit length of the capillary [16]. Checa et al. model sprout formation stochastically by making the probability of sprouting from a vessel segment proportional to the segment length. The capillary growth rate is also regulated by the local mechanical stimulus [17]. Peiffer et al. proposed a hybrid bioregulatory model of angiogenesis during bone fracture healing [18], based on the deterministic hybrid model of Sun et al. [19]. The process of angiogenesis is modeled discretely, including sprouting and anastomosis. The selection of tip cells in the growing vascular network is, however, modeled with phenomenological rules. A three dimensional model of cellular sprouting at the onset of angiogenesis was developed by Qutub et al. [20]. Although this framework describes sprout formation as a function of the local VEGF concentration and the presence of Dll4, it is only a first approximation of the complex tip-to-stalk cell communication by Dll4/Notch signaling.
A more detailed model of the lateral inhibition that underlies the tip cell selection process in angiogenic sprout initialization was presented by Bentley et al. [3], [21]. They use an agent-based framework to accurately simulate a small capillary comprising 10 endothelial cells that can change shape and sense the local VEGF concentration by extending very thin filopodia. Moreover, every endothelial cell is characterized by its individual protein levels of VEGFR-2, Dll4 and Notch – and their distribution on the cell membrane, by further subdividing the membrane into separate agents – which does not only allow assessment of the effects of the VEGF environment on tip/stalk cell patterning but also those of Dll4 over- and under-expression and cell shape change.
Sprouting angiogenesis involves multiple biological scales: the intracellular scale where gene expression is altered so that different phenotypes (e.g. tip and stalk cells) can arise, the cellular scale that involves proliferation and migration and the tissue scale that encompasses the concentration fields of soluble and insoluble biochemical factors. As these scales are highly coupled, multiscale models are needed to study the mechanisms of sprouting angiogenesis. To the best of the authors' knowledge, there is only one model of sprouting angiogenesis with Dll4-Notch1 rigorously implemented [3] but this model simplified the extracellular environment to a uniform or linearly varying field of VEGF concentration, which is constant in time. While this simplification is justified for a detailed study of the short term phenotypic changes of a few neighboring endothelial cells, it is not for more complex, multicellular systems that involve cell-matrix interaction and highly dynamic, extracellular environments. This is certainly true for fracture healing, in which matrix densities and (gradients of) extracellular concentrations of soluble signals, like VEGF, are spatially and temporally changing as a result of cellular activity. While efforts have been done to model the interplay of VEGF diffusion and sprouting angiogenesis in the context of skeletal muscle tissue [6], these multiscale models did not incorporate Dll4-Notch1 signaling. Moreover, in the context of fracture repair multiscale models that consider angiogenesis and that relate tissue, cell and intracellular scales have not been established yet [22].
In this study, we present a multiscale model of osteogenesis and sprouting angiogenesis with lateral inhibition of endothelial cells (MOSAIC) which extends the bioregulatory framework of Peiffer et al. [18] with an intracellular model based on the work of Bentley et al. on tip cell selection [3]. We hypothesize that the MOSAIC model creates enhanced capabilities for investigating the influence of molecular mechanisms on angiogenesis and its relation to bone formation. Simulation results will illustrate the interplay between molecular signals, in particular VEGF, Dll4 and Notch1, endothelial cell phenotypic behavior and bone formation. They will demonstrate the advantages of multiscale modeling in the context of fracture healing, thereby exploring the importance of the model of Bentley et al. [3] for a much more complex and dynamic extracellular environment. At the same time, by comparison to the more phenomenological model of Peiffer et al. [18] the potential of a more mechanistic treatment of tip cell selection will become clear.
The MOSAIC model presented in this work integrates an intracellular module based on the work of Bentley et al. [3] into the model of Peiffer et al. [18]. Figure 1 gives a schematic overview of the MOSAIC model which consists of (1) a tissue level describing the various key processes of bone regeneration with continuous variables, (2) a cellular level representing the developing vasculature with discrete endothelial cells and (3) the intracellular level that defines the internal dynamics of every endothelial cell. The combination of continuous and discrete modeling techniques results in a hybrid, multiscale model.
The discrete variable cv represents the blood vessel network, which is implemented on a lattice. When a grid volume contains an endothelial cell this variable is set to 1, otherwise cv = 0. The vessel diameter is defined by the grid resolution and is always one endothelial cell wide, although the movement of the tip cell is grid independent as explained below. Every endothelial cell (cv = 1) has unique intracellular protein levels, which control the behavior of that specific cell. The intracellular module is adapted from the agent-based model of Bentley et al. [3] and consists of the following variables: the level of VEGFR-2 (V), Notch1 (N), Dll4 (D), active VEGFR-2 (V′), active Notch1 (N′), effective active VEGFR-2 (V″), effective active Notch1 (N″) and the amount of actin (A). The effective active levels (V″ and N″) include the time delay of translocation to the nucleus, thereby representing the levels at the nucleus, influencing gene expression. The amount of actin (A) refers to the polymerized actin levels (F-actin) inside the cell. In particular, it is associated to actin used for filopodia formation, owing to its importance for tip cell migration. As such, an increase in actin levels can be interpreted as filopodia extension, while a decrease as filopodia retraction. Other intracellular signaling pathways that involve actin, such as energy metabolism [23], [24], are not considered.
The following equations describe the intracellular dynamics. An overview of all the parameters of the intracellular module can be found in Table 1.
The activation of the VEGFR-2 receptor, described by V′, is given by: (1)where the constant Vsink represents the amount of VEGFR-1 receptors that act as a sink (decoy receptor) by removing VEGF from the system, t represents the time and δt the time step of the inner loop (more information on these parameters can be found in the section “implementation details” below), Vmax represents the maximal amount of VEGFR-2 receptors, gv is the local VEGF concentration (at the tissue level) and Mtot is the total number of membrane agents (constant for all ECs). Equation 1 is adapted from Bentley et al. [3] where every EC is composed of a varying amount of membrane agents, representing small sections of the cellular membrane. In the current framework, however, every EC is represented by one agent so that Mtot was chosen to be constant for all ECs and equal to an intermediate value between the initial and maximal amount of membrane agents in the agent-based framework of Bentley et al. [3]. The level of activated VEGFR-2 remains in a range going from 0 to V. When the VEGFR-2 receptors are activated above a certain threshold (V′*), the actin levels of the endothelial cell are incremented in a constant manner (ΔA). As mentioned earlier, this represents the extension of filopodia by the endothelial cell, which is shown to be regulated downstream of VEGFR-2 [4]. If the endothelial cell fails to extend its filopodia for a certain amount of time D3, the filopodia retract which is mathematically translated into a reduction of the actin levels in a constant manner (−10.ΔA). The actin level remains in a range between 0 and Amax. The amount of Notch1 is considered to be constant in every EC, representing a balance between the rate of degradation and expression. At the same time, initial Notch activity levels are assumed to be zero and in the model Notch activity can only be increased through binding with Dll4 from neighboring ECs. The model therefore neglects the role of Notch in EC quiescence and the fact that high Notch activity levels have been measured in quiescent ECs [25]–[27]. Instead, it only focuses on the role of Dll4-Notch in tip cell selection. The number of activated Notch receptors (N′) will be equal to the total amount of Dll4 available (with an upper bound, given by the total number of Notch receptors N). The amount of Dll4 in the environment of an EC is the sum of the amount of Dll4 at the junctions with its neighboring ECs, whereby every cell is assumed to distribute Dll4 uniformly across its cell-cell junctions (see Figure 2).
After a delay of D1 for V′ and D2 for N′ the active VEGFR-2 and Notch levels become the effective active levels (V″ and N″) representing the levels at the nucleus, influencing gene expression. The delays were taken from Bentley et al. which were fitted to a somite clock Delta-Notch system [3], [28]. Note that there is a delay between receptor activation and gene expression (transcription) but not between gene expression and protein synthesis (translation), which is a simplification of the model.
The amount of Dll4 is modeled in the following way:(2) represents the previous amount of Dll4, the change in Dll4 expression due to the activation of the VEGFR-2 receptor [4], [29] and is the amount of Dll4 that is removed from the environment due to the activation of Notch-receptors on neighboring ECs. If-conditions are used to ensure that the Dll4 level remains in a range between 0 and Dmax. When Notch signaling is activated in a cell, the amount of VEGFR-2 receptors is down-regulated, suppressing the tip cell phenotype as follows [4], [5]:(3)Vmax represents the maximal amount of VEGFR-2 receptors and models the VEGFR-2 expression change due to Notch1 activation. If-conditions are used to ensure that the VEGFR-2 level remains in a range going from Vmin to Vmax. Since the amount of VEGFR-2 (V) at the previous timestep () is not present in Equation 3, the amount of VEGFR-2 is continuously in equilibrium with the amount of effective active Notch1 (). Equation 3 implies that in quiescent cells the number of VEGFR-2 receptors will be maximal, owing to the absence of any Notch activity. As mentioned earlier, the model neglects the role of Notch activity in quiescence and the fact that it will lead to reduced VEGFR-2 levels in quiescent ECs [25]–[27].
Note that Bentley et al. [3] represent every EC by a varying number of agents (to account for changes in cell shape and cell growth), whereas in this study every EC is represented by one agent. However, in order to use the parameter values and equations (in an adapted form) of Bentley et al. [3], Mtot was fixed at a constant value for all ECs. Consequently, the values of V, N, V′, N′, V″, N″, D and A are evaluated at the cellular level, not at the level of individual membrane agents. This also implies that here cellular polarity is not captured explicitly as receptor and ligand concentrations are uniformly distributed across the membrane junctions. In the current model cell directional behavior follows from gradients of extracellular signals alone.
The evolution of the vascular network is determined by tip cell movement, sprouting and anastomosis [18], [19], outlined below.
At the tissue level, the fracture healing process is described as a spatiotemporal variation of eleven continuous variables: mesenchymal stem cell density (cm), fibroblast density (cf), chondrocyte density (cc), osteoblast density (cb), fibrous extracellular matrix density (mf), cartilaginous matrix density (mc), bone extracellular matrix density (mb), generic osteogenic growth factor concentration (gb), chondrogenic growth factor concentration (gc), vascular growth factor concentration (gv) and concentration of oxygen (n). The set of partial differential equations (PDEs) accounts for various key processes of bone regeneration. Initially the callus is filled with granulation tissue and the mesenchymal stem cells and growth factors will quickly occupy the regeneration zone. Subsequently the mesenchymal stem cells differentiate into osteoblasts (intramembranous ossification – close to the cortex away from the fracture site) and chondrocytes (central callus region). This is followed by VEGF expression by (hypertrophic) chondrocytes, which attracts blood vessels and osteoblasts and which is accompanied by cartilage degradation and bone formation (endochondral ossification). The model does not include bone remodeling. The general structure of the set of continuous equations is given by: (7)(8)where t represents time, the space and the density of a migrating cell type (mesenchymal stem cells and fibroblasts). represents the vector of the other nine concentrations, ECM densities, growth factor concentrations and oxygen concentrations for which no directed migration is modeled. and are the diffusion coefficients. represents the taxis coefficients for both chemotaxis and haptotaxis. and are reaction terms describing cell differentiation, proliferation and decay and matrix and growth factor production and decay. Detailed information, including the complete set of equations, boundary and initial conditions, parameter values and implementation details can be found in Peiffer et al. [18] and Geris et al. [31] and are provided here as online supplement.
The partial differential equations are solved on a 2D grid with a grid cell size of 25 µm. The width of the discrete ECs is determined by the size of a grid cell (25 µm). Since the ECs in the model of Bentley et al. [3] have a width of 10 µm, the parameter values taken from Bentley et al. are multiplied with a factor of 2.5 (see Table 1). The partial differential equations are spatially discretized using a finite volume method assuring the mass conservation and nonnegativity of the continous variables [32]. The resulting ODEs are solved using ROWMAP, a ROW-code of order 4 with Krylov techniques for large stiff ODEs [33]. The MOSAIC model is deterministic and implemented in Matlab (The MathWorks, Natick, MA).
The flowchart in Figure 1B gives a schematic overview of the computational algorithm used in this study. Firstly the continuous variables are calculated. Then the inner loop is iterated which consists of four subroutines: (1) the current tip cells are evaluated by the tip cell selection criterion and, if necessary, they lose their tip cell phenotype; (2) the new position of every tip cell is calculated using a central difference scheme in space in combination with explicit Euler time integration; (3) the code checks whether sprouting occurs, meaning that other endothelial cells also satisfy the criterion for tip cell selection; (4) the intracellular levels of every endothelial cell are updated. Finally, the inner and outer loops are iterated until the end time of the simulation is reached.
The outer loop has a maximal time step size of 8.57 hours (row). Since the tip cells do not move more than one grid cell (25 µm) in this time interval ( = 35 µm/day [19]), this maximal time step size (row) implies that the 11 PDEs can be solved for a constant vasculature. The inner loop has a maximal step size of 1.2 hours (ee), similar to Peiffer et al. [18], and was chosen so that the movement of the tip cells within one grid cell could be accurately followed (ee≪row). To reduce implementation difficulties, the time step of the inner loop (δt) is determined by calculating how many maximal inner loop time steps (ee) can fit in one outer loop time step (ΔT) and dividing the outer loop time step by this number. Consequently, the time step of the inner loop is not constant, which means that D1, D2 and D3 vary slightly, but this is a minor trade-off for the computational efficiency. Numerical convergence tests have shown that the average inner time step δt is equal to 155 s. Consequently, D1, D2, D3 approximate the delays chosen by Bentley et al. [3]. Since the time step δt is approximately 10 times the time step of Bentley et al. [3], the parameter values of σ and δ have been altered to match the dynamics of the Dll4-Notch system. Numerical tests have shown that similar behavior is retrieved when both σ and δ are multiplied with 3.16 (see Table 1).
Simulations were conducted using a quad-core Intel® Xeon® CPU with 12 GB RAM memory. Initially the callus domain is filled with granulation tissue only (mf,init = 10 mg/ml), all other continuous variables are initialized to zero. Boundary conditions are presented in Figure 3. Further information on the choice of appropriate boundary and initial conditions of the continuous variables can be found in Peiffer et al. [18] and Geris et al. [31].
The MOSAIC model predicts the evolution of the continuous variables as well as the evolution of the intracellular variables during normal fracture healing. The osteoprogenitor cells enter the callus from the surrounding tissues and differentiate into osteoblasts under the influence of osteogenic growth factors. This leads to rapid intramembranous ossification near the cortex and distant from the fracture line. In the endosteal and intercortical callus the bone is formed through the endochondral pathway, starting from a cartilage template that is mineralized as the blood vessel network is formed to supply the complete fracture zone with oxygen. Figure 4 compares the predictions of the Peiffer-model [18] and the MOSAIC model with the experimentally measured tissue fractions of Harrison et al. in a rodent standardized fracture model [35]. Both models capture the general trends in the experimental data equally well: the bone tissue fraction gradually increases throughout the healing process; the fibrous tissue fraction disappears; the cartilage template is first produced and later replaced by bone.
After one, two and three weeks of simulated healing time the surface fraction of the blood vessels in the callus is respectively 2.34%, 18.20% and 46.25%. Experimental results also show that the vascular plexus is very dense in the fracture callus, although quantitative results are lacking [36], [37]. Images, illustrating the angiogenic and osteogenic process in the fracture callus can be found in Maes et al. and Lu et al. [38], [39]. These experimental studies report that at the progressing front, there is a tree-like structure of tip cells extending filopodia to sense their environment and to guide the developing sprout. At the back, the vasculature is being remodeled into a more structured network of larger vessels with more quiescent endothelial cells. At present this remodeling phase of the vasculature, which will remove some blunt ends as well as redundant vessels, is not included in the MOSAIC model.
Figure 5 shows that the tip cells have high VEGFR-2 levels. The stalk cells are inhibited and have low VEGFR-2 and actin levels. The Dll4-Notch signaling stops when the VEGF-concentration in the callus drops (the VEGFR-2 levels stay constant) (Equations 1–3). The VEGF concentration goes down since the vasculature brings enough oxygen to the fracture site. The endothelial cells far away from the vascular front all have maximal VEGFR-2 levels.
The average VEGFR-2 concentration, predicted across all ECs present in the fracture callus, drops at day 7 in the standard condition (Figure 6, standard). Indeed, after 7 days the ECs start to inhibit each other in gaining the tip cell phenotype, resulting in a prediction of enhanced Notch1-signaling and reduction of the average VEGFR-2 levels at the vascular front. At the back, VEGFR-2 levels are predicted to return to their maximal value, which is a direct consequence of Equation 3 (effect of Notch activity on VEFGR-2), and the fact that in the model Notch activity levels of an EC are only governed by VEGF-induced Dll4 expression (in its neighboring cells). As mentioned before, the model only focuses on the lateral inhibition between tip cells and stalk cells through Dll4-Notch. It does not address EC quiescence and the fact that Notch activity in quiescent ECs will be associated with reduced VEGFR-2 receptor levels [25]–[27]. Despite this anomaly in terms of the number of VEGFR-2 receptors, the model correctly predicts highly reduced VEGFR-2 activity levels in quiescent cells (i.e. cells at the back of the vasculature), because of the low VEGF concentrations encountered here. This trend in the average VEGFR-2 concentration (Figure 6, standard) was also measured by Reumann et al. [40]. Reumann et al. characterized the time course of VEGFR-2 mRNA expression during endochondral bone formation in a mouse rib fracture model by quantitative RT-PCR [40]. They observed a small drop (although statistically insignificant) in the median value of VEGFR-2 mRNA expression at three days post fracture [40]. The simulated average VEGFR-2 concentration (Figure 6, standard) follows a similar trend but drops at later time points (day 7 versus day 3). The experimental results [40] were, however, determined in a mouse rib fracture model whereas the parameter values of the model presented in this study were derived from a rat femur fracture model [35]. At the same time, it should be mentioned that Figure 6 represents protein values where Reumann et al. [40] measured mRNA levels. Moreover, Reumann et al. [40] measured the total mRNA content, of all cells present in the callus whereas Figure 6 shows the average VEGFR-2 concentration on the cell membranes of the ECs in the callus. Not only ECs but also osteoblasts and other osteogenic cells express VEGFR-2 [41], which might also explain the temporal difference seen between the experimental and simulated data.
If pharmacological blocking of VEGFR-2 receptors is simulated, the vasculature does not develop since the actin production is inhibited, meaning that the ECs cannot extend filopodia and gain the tip cell phenotype. Due to this impaired vascularization only a small amount of bone is predicted intramembranously, resulting in a non-union between the fractured bone ends.
If the VEGF concentration is increased (Figure 7B), the vascular density is initially increased since the VEGFR-2 receptors are being more activated in this stimulating environment (Equation 1). This simulation result is confirmed by many experimental studies that reported an increase in vascularity at the site of VEGF application in a murine femoral fracture healing model [34], a lapine mandibular defect model [42], a murine ectopic model [43] and a rat femoral bone drilling defect model [44]. In the simulations the increase in vascular density leads to faster healing, which is also found experimentally [34], [42] (Figure 8). The MOSAIC model predicts earlier bone formation, less cartilage formation in both the periosteal, intercortical as well as the endosteal callus. Moreover, the cartilage resorption is predicted to be accelerated, another trend which has been reported in experimental studies [43], [44]. It is striking, however, that after 35 days, the MOSAIC model predicts that the VEGF-treated callus contains slightly less bone and more remnants of fibrous tissue than the normal condition (Figure 8).
A further increase of the VEGF concentration (+2%, Figure 7B(iii)) reduces however the vascular density and bone tissue fraction in the MOSAIC model. This is consistent with the trend seen by Street et al., where an optimal dose of VEGF (250 µg) leads to a maximal amount of callus volume (both total and calcified) in a critical rabbit radius segmental gap model [34].
In higher VEGF environments (e.g. Figure 7B(iii)) the ECs strongly inhibit each other; creating a salt and pepper pattern of high and low VEGFR-2 levels (Figure 9). In addition, the development of the vasculature ceases after a certain period, as can be seen in Figure 9. The ECs that fulfill the tip cell criteria (Equation 6) will sprout and will initially move perpendicular to the vessel from which it is originating. In case this “mother” vessel is a growing vessel as well that extends towards the source of the chemotactic and haptotactic signals, this implies that the new sprout will initially move perpendicular to the gradients. In this particular configuration (Figure 7B(iii)), the vascular front was progressing in a more “sheet-like” fashion. Consequently, the tip cells persistently want to sprout towards already occupied grid cells, an action that is not allowed in the computational framework. The period after which the vascular development ceases, is shorter in higher VEGF environments. Note that the average VEGFR-2 concentration of all ECs in the callus reaches a constant level which is reduced in high VEGF environments (Figure 6).
When a reduction of VEGF concentration by means of the addition of VEGF-antibodies is simulated, results demonstrate that the VEGFR-2 receptor is not sufficiently stimulated. This leads to an impaired vasculature and a non-union between the fractured bone ends (see Table 3, Equation 1). The simulated reduction in vascular density is consistent with the experimental findings of Street et al. who incorporated a fracture hematoma supernatant with neutralizing monoclonal antibody to human VEGF in a Matrigel vehicle, which was then implanted in a murine dorsal wound model [45]. There was a significant decrease in the number of blood vessels formed in the Matrigel vehicle with neutralizing monoclonal antibody when compared to the fracture hematoma supernatant alone [45].
The sensitivity analysis indicates that the model results are greatly influenced by the parameters σ, δ and Vsink and that the MOSAIC model is insensitive to the initial conditions of Dll4 (D0) and actin (A0). An overview of the results of the sensitivity analysis can be found in Table 3.
Simulations of the heterozygous knockout genotypes (δ = 50%, σ = 33%) show an increased sprouting due to a clearly reduced inhibition of the tip cell phenotype (Equations 2–3). Figure 5 demonstrates that the endothelial cells behind the brown (tip) cells with high VEGFR-2 levels are strongly inhibited in the normal case but are weakly inhibited in the simulated knockout. The overexpression of Dll4 (δ = 200%) increases the inhibition of the tip cell phenotype resulting in a decrease of the vascular density (Table 3) and a delay in the endochondral bone formation process, particularly in the periosteal callus (Figure 8). The increase in σ also causes a more potent suppression of the tip cell phenotype in the stalk cells, due to an increased down-regulation of VEGFR-2 by Notch1 (Equation 3). In turn, this leads to a decrease of the vascular density (see Figure 5B, Table 3). Thus, if the tip cell phenotype is more inhibited (by increasing δ, which defines the enhancement of Dll4 expression due to VEGFR-2 activation (Equation 2) or increasing σ, which represents the inhibition of the VEGFR-2 expression due to Notch1 activation (Equation 3)), the vascular density is reduced. This simulation result corresponds to experimental observations [5], [46], [47]. Hellström et al. [5] showed that the inhibition of Notch signaling (by inhibiting Notch receptor cleavage and signaling with γ-secretase inhibitors, by heterozygous inactivation of the Notch ligand Dll4 or by endothelial cell specific deletion of Notch1) promotes an increase in the number of tip cells in the retina of newborn mice. Conversely, a 35% decrease in filopodia density and a 45% decrease in vessel density were found in a direct gain of function experiment of the Notch1 receptor.
Simulating an increase of the decoy receptor VEGFR-1 (by decreasing Vsink) results in a reduction of the vascular density since less VEGF remains available for VEGFR-2 activation (see Figure 7A, Equation 1). This is consistent with Flt-1 (VEGFR-1) loss- and gain-of-function data in zebrafish embryos [34], [48]. Moreover, Street et al. showed that Flt-IgG treatment decreased the vascularity by 18% and impaired cortical bone defect repair in a murine femoral fracture healing model [34]. Similarly, the in silico results show a delayed or even impaired healing of the fracture due to the reduced vascular density. In Figure 7A(iv) the VEGF concentration is too low to activate the VEGFR-2 receptors which stops the angiogenic process. Since only a small amount of bone will be formed intramembranously in the fracture callus of Figure 7A(iv), the impaired vascularization will result in a non-union.
If the threshold of VEGFR-2 activation V′*, below which the actin production and tip cell movement is inhibited, is increased, the vascular density decreases (see Table 3). The initial amount of actin (A0) only slightly influences the final vascular density (see Table 3). The other variables related to the fracture healing, were not influenced. Similarly, the final vascular density is insensitive to the initial intracellular amount of Dll4 (D0) (see Table 3).
This study established a novel multiscale model of angiogenesis in the context of fracture healing, by integrating an agent-based model of tip cell selection [3] into a previously developed hybrid model of fracture healing [18]. The bone regeneration process was predicted by the MOSAIC model in accordance with experimental reports and previously validated in silico results [18]. The MOSAIC model was also able to capture many experimentally observed aspects of tip cell selection: the salt and pepper pattern seen in developing vascular structures under normal angiogenic conditions, i.e. a tip cell with high VEGFR-2 and actin levels followed by a stalk cell characterized by strong Notch1 signaling and therefore reduced VEGFR-2 and actin levels [30], an increased tip cell density and a higher vascular density in case of Dll4 heterozygous knockouts [5] and an excessive number of tip cells (leading to a very high vascular density) in high VEGF concentrations [3], [34]. The sensitivity analysis also indicated the most influential parameters of the MOSAIC model (δ, σ and Vsink).
This study has addressed some, but not all of the limitations of the Peiffer-model [18]. In the MOSAIC model the tip cell selection is based on Dll4/Notch1 signaling whereas the Peiffer-model [18] implemented sprouting with phenomenological rules such that(9)i.e. in the Peiffer-model the VEGF concentration needs to be high enough (10 ng/ml), there needs to be a minimal separation of 100 µm between two tip cells and the movement direction of the new tip cell should make an angle of >24° with the orientation of its mother vessel. Moreover, the Peiffer-model foresees three healing days between subsequent sprouting events, which has no experimental foundation and has been removed in the MOSAIC model. Consequently, the MOSAIC model is more mechanistic, allowing investigation of different mutant and druggable cases in the signaling pathways, leading to real predictions for experimentation, which was not possible in the Peiffer-model. Since the incorporation of the lateral inhibition mechanism leads to a denser plexus in the MOSAIC model than in the Peiffer-model, we have reduced the oxygen production rate by a factor of two so that the final oxygen concentrations are the same in both models. Experimental results show that the vascular plexus is indeed very dense in the fracture callus [36], [37].
In the MOSAIC model the tip cell velocity increases with the active VEGFR-2 levels, indicating that both the level of VEGFR-2 and the external VEGF-concentration influence the tip cell speed. This is consistent with the experimental data of Arima et al. [49]. They used time-lapse imaging in a murine aortic ring assay (with and without VEGF) to quantify the behavior of the endothelial cells during angiogenic morphogenesis [49]. Arima et al. reported that VEGF-induced vessel elongation was only due to greater displacement per tip cell [49]. Moreover, treatment with Dll4-antibodies also resulted in a greater displacement per tip cell [49]. This is due to the reduced inhibitory actions of Dll4, causing a greater number of cells to have high VEGFR-2 levels. These results are however contested by Jakobsson et al. who quantified the average migration speed of wild-type (DsRed and YFP) and heterozygous Vegfr2+/egfp endothelial cells in different chimaeric embryoid bodies [50]. They observed no difference in migration speed, indicating that VEGFR-2 levels do not determine EC migration velocity [50]. Clearly, more research is necessary to elucidate the above observations and improve the current implementation of tip cell migration in future versions of the model.
The MOSAIC model indicates a key role of the decoy receptor VEGFR-1 (modeled via Vsink) (Equation 1). Increasing the amount of VEGFR-1, results in a decrease of the vascular density (Figure 7A) which is also seen in loss- and gain-of-function data [34], [48]. Both the MOSAIC model and the model of Bentley et al. [3], use a constant value to represent the decoy-effect of the VEGFR-1 receptor. There is however experimental evidence that both VEGFR-1 and its soluble form are up-regulated in Notch-activated stalk cells [4], [51]. Hence, the stalk cells phenotype is not only consolidated by a decrease in VEGFR-2 but also by an increase in the competing VEGFR-1 receptor. The results of Krueger et al. also suggest that VEGFR-1 regulates tip cell formation in a Notch-dependent manner [48].
The MOSAIC model displays interesting behavior in high VEGF environments (see Figure 7). Initially, the increase in VEGF has a positive effect, resulting in a very dense vasculature since the VEGFR-2 receptors are being more activated in this stimulating environment (+0.1%; see Equation 1). This leads to a faster healing, which is also found experimentally [34], [42]. A further increase (+2%), however, reduces the vascular density in the MOSAIC model. This is consistent with the trend seen by Street et al. [34]. Note that a salt and pepper pattern of high and low VEGFR-2 levels is created and maintained in high VEGF environments (+2%; Figure 7B(iii) and Figure 9). To explain this observation, one needs to look at the beginning of the angiogenic process in the fracture callus. Initially, some endothelial cells gain the tip cell phenotype and start to migrate. Gradually sprouts arise in the developing vasculature which increases the network size and alters the local VEGF-levels due to the influence of the oxygen tension on VEGF-production. The original tip cells maintain their advantage (e.g. located in a higher VEGF environment) by strongly inhibiting their neighboring ECs, creating a salt and pepper pattern of VEGFR-2 levels. In standard conditions the vasculature would start to mature, leading to quiescent ECs. In high VEGF environments (+2%), however, this salt and pepper pattern of VEGFR-2 is maintained (Figure 9), illustrating that some ECs have (very) high VEGFR-2 levels leading to a persistent inhibition of their neighboring ECs (characterized by low VEGFR-2 levels). Figure 6 shows that these high VEGFR-2 levels are cancelled out by the low VEGFR-2 levels resulting in a “steady state” level of the average VEGFR-2 concentration. In high VEGF environments (+2%, +10%) this “steady state” level is gradually reduced (Figure 6), implying the dominance of the lower VEGFR-2 levels. Mathematically, this result follows from Equations (1) and (3), indicating that in high VEGF concentrations both the active VEGFR-2 (V′) and Notch (N′) (and with a delay the effective active VEGFR-2 (V″) and Notch (N″)) are high, resulting in a reduction of the VEGFR-2 receptor (Figure 10). Consequently, the average VEGFR-2 concentration is reduced below the threshold for tip cell formation (Figure 6). In other words, the majority of the ECs have too little VEGFR-2 receptors to assume the tip cell phenotype. In the extreme case, this finally results in the inhibition of the development of the vasculature since there are no tip cells to lead the sprouts towards the VEGF source (Figure 7B(iv)).
Interestingly, Figure 7D shows that similar results cannot be obtained with the Peiffer-model [18], i.e. the vascular density is not reduced in high VEGF environments (+2%, +10%). This is due to the phenomenological rules that determine the tip cell selection in the Peiffer-model (Equation 9). A similar “non-linear” EC response to VEGF concentrations would only be possible with the Peiffer-model if another phenomenological rule would be implemented that e.g. down-regulates tip cell selection at high VEGF responses. In contrast, the “non-linear” response follows naturally from the mechanistic rules of tip cell selection that were implemented in the MOSAIC model. That is, the down-regulation of the tip cell selection in high VEGF environments (+2%, +10%) arises from the negative feedback loop in the Notch-Dll4 signaling pathway. Moreover, in high VEGF-environments (+10%) and at the back of the developing vasculature, we see an indication that patches of endothelial cells oscillate between cell fates (switching between high and low VEGFR-2 levels). These patches are also predicted by the model of Bentley et al. and are observed during pathological angiogenesis [3]. In the future, we will further investigate the conditions that give rise to these oscillations and their implications on the development of the vasculature.
The results of the MOSAIC model are based on the assumption that a tip cell phenotype can only be acquired if the levels of VEGFR-2 and actin are sufficiently high (Equation 6). If this criterion was changed by replacing the requirement on VEGFR-2 by a similar requirement for the level of active VEGFR-2, some ECs could become tip cells, since V′ is high in high VEGF environments (+2%, +10%), and the vasculature would fully develop (Figure 7C). These results show the added value of the MOSAIC model: the intracellular module and its related state variables and rules decide on the EC response to the extracellular VEGF environment, in turn determining the healing response at the tissue level (Figure 8). In case the criterion for tip cell selection is specified in terms of VEGFR-2 (and actin) the absence of blood vessel formation will result in a non-union or a delayed union of the fracture. However, when this criterion is replaced by one that relies on the levels of active VEGFR-2 (and actin), a vascular and healing response is retrieved, similar to the Peiffer-model.
Clearly, these findings give rise to some interesting biological questions on a proper criterion for the tip cell phenotype. Since the VEGFR-2 levels are strongly reduced in high VEGF environments (+10%), the tip cells lose their tip cell phenotype and stop migrating although there is a strong angiogenic signal present. Can tip cells move in high VEGF environments although they do not have enough VEGFR-2 receptors? If so, should the tip cell criterion (Equation 6) be based on the active VEGFR-2 levels (V′), since these remain high in high VEGF concentrations? Or is the down-regulation of VEGFR-2 receptors in high VEGF environments compensated by other signaling cascades that have VEGFR-2 as one of their downstream targets (leading to an increase of VEGFR-2)?
In this study, the model of Peiffer et al. [18] was combined with a detailed model of Dll4-Notch1 signaling [3]. Some simplifications were however made to the model of Bentley et al. due to computational reasons, i.e. the size and shape of the ECs are fixed in the PDE framework of the MOSAIC model. Consequently, every EC is represented by one agent whereas Bentley et al. use a varying amount of membrane agents for every EC [3]. This does not only allow Bentley et al. to model the change in membrane and cell shape in great detail, but also to include cellular polarity (non-uniform distribution of receptors and ligands across the cell membrane and cell-cell junctions). In the MOSAIC model, filopodia extension is modeled implicitly by an increase in the level of the “actin” variable upon VEGFR-2 activation. This is consistent with current knowledge that activation of Cdc42 by VEGF triggers filopodia formation [4]. Bentley et al. modelled filopodia extension in more detail by adding membrane agents to the cellular membrane. As a result, the number of VEGFR-2 receptors will alter due to filopodia extension, which is proposed to be a mechanism to consolidate the tip cell fate [3]. In the MOSAIC model the accumulation of actin does not lead to an increase in the amount of VEGFR-2 levels or to a change in the microenvironmental range that can be probed by the tip cell. However, if the molecular mechanisms of filopodia extension and its implications on probing the environment and the directionality of tip cell movement are clearer, these can be readily incorporated in the multiscale framework.
The mechanism of lateral inhibition is based on Dll4/Notch1 signaling between the endothelial cells of the developing sprout. Delta-Notch signaling is however an evolutionary conserved pathway that is also involved in cell fate specification, tissue patterning and morphogenesis [2], . In angiogenesis specifically, Notch signaling influences endothelial cell specification [4], [5], [30], [54], [55], endothelial proliferation [29], [30], cell migration [2], [30], filopodia formation [30], cell adhesion [30], and post-angiogenic vessel remodeling and endothelial cell quiescence [56]. These effects are not only dependent on Dll4 and Notch1 but also on the other ligands (Delta-like 1, Delta-like 3, Jagged-1 and Jagged-2) and receptors (Notch2, Notch3 and Notch4) [57]. Due to the complexity and interdependency of these pathways, only the influence of Dll4-Notch1 signaling on tip cell selection was modeled. Consequently, in the model once the VEGF levels are reduced due to the restoration of the blood flow and tissue oxygenation, the Dll4-Notch1 signaling pathway is not active anymore. This is predicted to occur in the ECs that are located at the back of the vasculature, returning their VEGFR-2 levels to the maximal value. As mentioned before this contradicts the fact that in quiescent cells VEGFR-2 levels will be minimal and smaller than those of migrating cells, which is consistent with high Notch activity in quiescence [25]–[27]. Since the MOSAIC model does not include the role of Notch in quiescence, the simulation results are only accurate for the initial formation of the vasculature and not for the maturation and stabilization of the vascular plexus. This does not, however, alter the main findings of this work concerning sprouting angiogenesis.
Besides VEGFR-2, also other VEGF receptors, such as VEGFR-1 and VEGFR-3 play a role in angiogenesis. Although the VEGFR-3 receptor is mainly active in lymphangiogenesis, recent experimental evidence indicates that VEGFR-3 is up-regulated in tip cells during pathological angiogenesis [4], [58]. Blocking this receptor reduces the amount of sprouting and EC proliferation. It appears that VEGFR-2 induces VEGFR-3 expression in tip cells, whereas it is down-regulated in stalk cells by Notch [4], [59]. However, when more quantitative experimental data become available on the role of VEGFR-1 and VEGFR-3 in sprouting angiogenesis, this can be incorporated in the MOSAIC model.
The MOSAIC model only focuses on soluble VEGF, whereas VEGF-isoforms that bind to the extracellular matrix are essential to establish the VEGF gradients required for guided tip cell migration [60]. Some modeling work has already been done in this area [61], [62], e.g. Vempati et al. used a detailed molecular model of VEGF ligand-receptor kinetics and transport to investigate the VEGF-isoform specific spatial distributions observed experimentally [61]. Many other factors, such as neuropilin 1 (NRP-1), fibroblast growth factor (FGF), and platelet-derived growth factor (PDGF) regulate the angiogenic response as well [2]. Nevertheless, it has been stated repeatedly that VEGF is “the principal dancer” during angiogenesis [29],[30].
The proposed MOSAIC model incorporates biological processes at various temporal and spatial scales: an intracellular module that includes Dll4/Notch1 signaling to determine tip cell selection, a discrete representation of the ECs allowing an accurate representation of the developing vascular network and a continuum description of oxygen, growth factors and tissues that finally result in the healing of the fracture by the formation of bone. Our simulation results demonstrate the advantages of such a multiscale approach. Firstly, the interplay between molecular signals, in particular VEGF, Dll4 and Notch1, endothelial cell phenotypic behavior and bone formation was explored. In this way, the MOSAIC model could be used to verify to what extent gene knockouts, injection of VEGF-antibodies or blockage of VEGF-receptors leads to a “bone phenotype” in terms of rate and amount of bone formation (see e.g. Figure 8). While some of these simulation results could be (qualitatively) compared to experimental data, it is clear that future research efforts must be focused on a more comprehensive quantitative validation. Again, the multiscale nature of the simulation results presents an advantage here, as it allows for a validation at different scales (molecular, cellular and tissue scale). Secondly, the proposed multiscale model is more mechanistic since tip cell selection is based on intracellular dynamics (Dll4-Notch1 signaling), rather than the phenomenological rules that were used in Peiffer et al. [18]. As such, the MOSAIC model enabled to extend the model of Bentley et al. [3] to the context of fracture healing, leading to interesting emergent behavior at the macro-scale. More specifically, whereas the Peiffer-model predicts the presence of a vascular network in high VEGF environments (+10%) the MOSAIC model (depending on the tip cell criterion) predicts the absence of a vascular network (see Figure 7), which was a direct consequence of the Dll4-Notch feedback mechanism (see explanation related to Figure 10). In conclusion, the proposed multiscale method was found to be a useful tool to investigate possible biological mechanisms across different time and spatial scales, thereby contributing to the fundamental knowledge of sprouting angiogenesis and its relation to fracture healing.
|
10.1371/journal.pgen.1001005 | Dual Roles for DNA Polymerase Theta in Alternative End-Joining Repair of Double-Strand Breaks in Drosophila | DNA double-strand breaks are repaired by multiple mechanisms that are roughly grouped into the categories of homology-directed repair and non-homologous end joining. End-joining repair can be further classified as either classical non-homologous end joining, which requires DNA ligase 4, or “alternative” end joining, which does not. Alternative end joining has been associated with genomic deletions and translocations, but its molecular mechanism(s) are largely uncharacterized. Here, we report that Drosophila melanogaster DNA polymerase theta (pol theta), encoded by the mus308 gene and previously implicated in DNA interstrand crosslink repair, plays a crucial role in DNA ligase 4-independent alternative end joining. In the absence of pol theta, end joining is impaired and residual repair often creates large deletions flanking the break site. Analysis of break repair junctions from flies with mus308 separation-of-function alleles suggests that pol theta promotes the use of long microhomologies during alternative end joining and increases the likelihood of complex insertion events. Our results establish pol theta as a key protein in alternative end joining in Drosophila and suggest a potential mechanistic link between alternative end joining and interstrand crosslink repair.
| DNA double-strand breaks, in which both strands of the DNA double helix are cut, must be recognized and accurately repaired in order to promote cell survival and prevent the accumulation of mutations. However, error-prone repair occasionally occurs, even when accurate repair is possible. We have investigated the genetic requirements of an error-prone break-repair mechanism called alternative end joining. We have previously shown that alternative end joining is frequently used in the fruit fly, Drosophila melanogaster. Here, we demonstrate that a fruit fly protein named DNA polymerase theta is a key player in this inaccurate repair mechanism. Genetic analysis suggests that polymerase theta may be important for two processes associated with alternative end joining: (1) annealing at short, complementary DNA sequences, and (2) DNA synthesis that creates small insertions at break-repair sites. In the absence of polymerase theta, a backup repair mechanism that frequently results in large chromosome deletions is revealed. Because DNA polymerase theta is highly expressed in many types of human cancers, our findings lay the groundwork for further investigations into how polymerase theta is involved in repair processes that may promote the development of cancer.
| DNA double-strand breaks (DSBs) and interstrand crosslinks pose serious threats to cell survival and genome stability. Because these lesions compromise both strands of the double helix, they impede DNA replication and transcription and therefore must be removed in a timely and coordinated manner. Interstrand crosslink repair has been shown to involve a DSB intermediate in some cases (reviewed in [1]). Therefore, there may be substantial mechanistic overlap in the processes used during repair of these two lesions.
Error-free repair of DSBs can be accomplished through homologous recombination (HR) with an undamaged homologous template (reviewed in [2]). However, in contexts where suitable templates for HR do not exist, error-prone repair mechanisms are also used. For example, non-homologous end joining (NHEJ) frequently creates small insertions and deletions during DSB repair, particularly in cases where the broken ends cannot be readily ligated (reviewed in [3]). Analogously, the use of translesion DNA polymerases during interstrand crosslink repair can result in mutations, due to the reduced fidelity of these polymerases [4], [5].
Accumulating evidence suggests that NHEJ is composed of at least two genetically distinct mechanisms. Classical NHEJ (C-NHEJ) involves the sequential recruitment of two highly conserved core complexes (reviewed in [6]). First, the Ku70/80 heterodimer recognizes and binds to DNA ends in a sequence-independent manner, thereby protecting them from degradation. In many eukaryotes, Ku70/80 also recruits DNA-PKcs, forming a synaptic complex that can recruit additional processing enzymes such as the Artemis nuclease and the X family DNA polymerases mu and lambda. These proteins expand the spectrum of broken ends that can be rejoined. The second core complex, composed of DNA ligase 4, XRCC4, and XLF/Cernunnos, catalyzes ligation of the processed ends. Depending on the substrate, C-NHEJ can result in perfect repair of broken DNA, or it can result in small deletions of 1–10 nucleotides and/or insertions of 1–3 nucleotides [7]. Although C-NHEJ can repair blunt-ended substrates, a subset of C-NHEJ products appear to involve annealing at 1–4 nucleotide microhomologous sequences on either side of the break.
Alternative end-joining (alt-EJ) is defined as end-joining repair that is observed in cells or organisms lacking one or more C-NHEJ components (reviewed in [8]). Alt-EJ in yeast is associated with deletions larger than those typically created by C-NHEJ, together with an increased tendency to repair by annealing at microhomologous sequences. Ku and ligase 4-independent end joining observed in mammalian cells also displays an increased tendency towards use of short microhomologies compared to C-NHEJ [9], [10]. Therefore, alt-EJ is sometimes called microhomology-mediated end joining (MMEJ) [11]. However, the relationship between MMEJ and alt-EJ is still unclear, and alt-EJ may comprise one or more C-NHEJ-independent repair mechanisms [8].
The importance of alt-EJ repair is highlighted by multiple studies that suggest it may promote chromosome instability and carcinogenesis. Alt-EJ produces chromosome translocations in mouse embryonic stem cells lacking Ku70 [12] and the use of alt-EJ during V(D)J recombination in C-NHEJ-deficient murine lymphocytes causes complex chromosome translocations and progenitor B cell lymphomas [13]. Furthermore, alt-EJ has been implicated in various translocations associated with chronic myeloid leukemia and human bladder cancer [14], [15]. Importantly, alt-EJ also operates during V(D)J rejoining in C-NHEJ-proficient B lymphocytes [16], suggesting that its role in DSB repair is not limited to situations where C-NHEJ is defective. However, alt-EJ is frequently masked by more dominant repair processes that are essential for vertebrate development, making it difficult to study. Therefore, its molecular mechanisms and the proteins involved remain largely unknown.
Several lines of evidence demonstrate that Drosophila is an excellent model system in which to study alt-EJ in a metazoan. The Drosophila genome lacks several mammalian C-NHEJ components, including DNA-PKcs and Artemis. This may predispose flies towards non-C-NHEJ repair. Consistent with this, we have previously shown that a DSB caused by excision of a P element transposon in flies is readily repaired by a DNA ligase 4-independent end-joining process [17]. Interestingly, although Drosophila orthologs for the Pol X family DNA polymerases mu and lambda have not been identified [18], we and others have found evidence for polymerase activity in Drosophila end-joining repair [17], [19], [20]. Specifically, end joining in flies is often associated with the insertion of nucleotides at repair junctions, frequently involving imperfect repeats of 5–8 nucleotides. Full or partial templates for the insertions, occasionally possessing mismatches, can often be identified in adjacent sequences, suggesting the action of an error-prone polymerase. Similar templated nucleotides (T-nucleotides) have previously been identified at translocation breakpoints in human lymphomas [21]–[23]. Therefore, T-nucleotides could represent a signature of alt-EJ and may be informative regarding its molecular mechanisms.
Additional insight into alt-EJ is provided by recent reports suggesting a mechanistic link between alt-EJ and interstrand crosslink repair. For example, a study of two Chinese hamster ovary cell lines sensitive to the crosslinking agent mitomycin C found that they were also deficient in alt-EJ [24]. Furthermore, certain interstrand crosslink-sensitive cell lines from Fanconi Anemia patients are also impaired in DNA-PKcs-independent rejoining of linearized plasmids [25]. Based on these reports, we hypothesized that additional mechanistic insight into both interstrand crosslink repair and alt-EJ could be gained by searching for mutants defective in both processes. To test this, we have screened Drosophila mutants that are sensitive to DNA crosslinking agents for additional defects in alt-EJ repair. In this work, we describe our studies with one such mutant, mus308.
The mus308 (mutagen sensitive 308) mutant was originally identified by its extreme sensitivity to interstrand crosslinking agents but normal resistance to alkylating agents [26]. Subsequently, mus308 was found to code for DNA polymerase theta, which is most similar to A family DNA polymerases such as Escherichia coli Pol I [27]. Orthologs of polymerase theta (hereafter referred to as pol θ) are found in many metazoans, including Caenorhabditis elegans, Arabidopsis thaliana, and Homo sapiens, but not in unicellular eukaryotes, including the yeasts [28]–[30]. Several lines of evidence suggest that pol θ may play an important role in maintaining genome stability. Similar to flies, C. elegans with mutations in POLQ-1 are defective in repair of interstrand crosslinks [28]. Mice lacking pol θ (chaos1 mutants) have a high frequency of spontaneous and mitomycin C-induced micronuclei in erythrocytes, consistent with genomic instability [31]. In addition, vertebrate pol θ orthologs have been implicated in a wide range of repair processes, including base excision repair, bypass of abasic sites, and somatic hypermutation of immunoglobulin genes [32]–[36]. Finally, upregulation of pol θ is observed in a variety of human tumors and is associated with a poor clinical outcome, suggesting that its overexpression may contribute to cancer progression [37].
Pol θ is unusual in possessing an N-terminal helicase-like domain and a C-terminal polymerase domain. Although pol θ purified from human cell lines and Drosophila has error-prone polymerase activity and single-stranded DNA-dependent ATPase activity, helicase activity has not been demonstrated in vitro [30], [38], [39]. Therefore, it remains unclear exactly how the structure of pol θ relates to its multiple functions in DNA repair in different organisms.
We report here that in addition to its role in DNA interstrand crosslink repair, Drosophila pol θ is involved in end-joining repair of DSBs. This alt-EJ mechanism operates independently of both Rad51-mediated HR and ligase 4-dependent C-NHEJ. Genetic analysis using separation-of-function alleles provides support for distinct roles of both the N- and C-terminal domains of pol θ in alt-EJ. Collectively, our data support a model in which helicase and polymerase activities of Drosophila pol θ cooperate to generate single-stranded microhomologous sequences that are utilized during end alignment in alt-EJ.
Drosophila mus308 mutants were initially identified based on their sensitivity to low doses of chemicals that induce DNA interstrand crosslinks [26]. To confirm this phenotype, we assembled a collection of previously identified mus308 mutant alleles [40], [41] and measured the ability of hemizygous mutant larvae to survive exposure to the crosslinking agent mechlorethamine (nitrogen mustard). Of the mutants that we tested, four were unable to survive exposure to 0.005% mechlorethamine: D2, D5, 2003, and 3294 (data not shown), consistent with their inability to repair interstrand crosslinks.
To determine the molecular lesions responsible for mechlorethamine sensitivity, we sequenced the entire mus308 coding region of flies hemizygous for each mutant allele. Pol θ possesses both a conserved N-terminal helicase-like domain and a C-terminal pol I-like polymerase domain (Figure 1, Figure S1) [30]. Three of the four alleles contain unique sequence changes that are predicted to affect pol θ primary structure (Figure 1, Figure S2, and Figure S3). The 2003 allele is a nonsense mutation upstream of the polymerase domain, while the D5 allele is a missense mutation that alters a highly conserved proline in the conserved N-terminus. The 3294 allele changes an invariant glycine in the helicase domain to serine. Interestingly, this residue is conserved in the related mus301 helicase, but not in other DNA helicases (data not shown). No mutations were found in the coding sequence of the D2 allele. Because homozygous D2 flies have undetectable levels of pol θ protein [38], the D2 mutation may affect a regulatory region of mus308.
One explanation for the extreme sensitivity of mus308 mutants to mechlorethamine could be a defect in the repair of certain types of DSB intermediates that are created during crosslink repair. To test this, we exposed flies hemizygous for each mutant allele to increasing doses of ionizing radiation (IR). Although IR creates many different types of lesions, unrepaired DSBs are thought to be the main cause of cell death following irradiation. All four mus308 mutants survived IR exposures as high as 4000 rads (Figure 2A), although bristle and wing defects characteristic of apoptotic cell death were frequently observed at high doses. Drosophila lig4 mutants, which are completely defective in C-NHEJ, also survive IR doses in excess of 4000 rads [17]. However, spn-A mutants, which lack the Rad51 recombinase required for strand invasion during homologous recombination initiation [42], are highly sensitive to IR [17]. Thus, in Drosophila, HR is the dominant mechanism used to repair IR-induced DSBs.
To test whether pol θ acts to repair IR damage in the absence of HR, we created mus308 spn-A double mutants and exposed them to doses of 125–1000 rads. Strikingly, doses as low as 125 rads resulted in almost complete killing of mus308 spn-A mutants (Figure 2B). In contrast, lig4 spn-A double mutants are only slightly more sensitive than spn-A single mutants to IR [17]. Thus, in the absence of HR, pol θ participates in a process crucial for repair of damage caused by ionizing radiation.
Because interstrand crosslink repair and alternative end joining have been shown to have partially overlapping genetic requirements in mammals [24], [25], we hypothesized that the extreme sensitivity of mus308 spn-A mutants to IR might relate to a role of pol θ in an alternative end-joining mechanism. To explore this hypothesis, we tested each mus308 mutant allele using a site-specific double-strand break repair assay that can distinguish between synthesis-dependent strand annealing (SDSA, a specific type of HR) and end joining (EJ) (Figure 3A) [43]. We have previously shown that the majority of end joining observed in this assay occurs independently of DNA ligase 4, and is therefore a form of alt-EJ [17]. In this system, excision of a P element (P{wa}) located on the X chromosome is catalyzed by P transposase, resulting in a 14-kilobase gap relative to an undamaged sister chromatid. The DNA ends remaining after excision each have 17-nucleotide non-complementary 3′ single-stranded overhangs [44]. These ends are highly recombinogenic and repair by SDSA is initially favored. However, because repair synthesis in this system is not highly processive, most repair products that are recovered from wild-type flies result from incomplete repair synthesis from one or both sides of the break, followed by end joining of the nascent DNA (SDSA+EJ events) [45]. To quantitate the percentage of repair events that derive from each mechanism, repair events are recovered from male pre-meiotic germline cells by mating individual males to females homozygous for the P{wa} element. Each of the resulting female progeny represents a single repair event that can be classified by eye color. Red eyed-females inherit a repair event involving homology-dependent synthesis that generated complementary single-stranded regions that subsequently anneal (repair by SDSA). Yellow-eyed females inherit a chromosome that was repaired by EJ or SDSA+EJ mechanisms (these repair events are hereafter referred to as (SDSA)+EJ; for further details, see Materials and Methods).
Overall, the results from the P{wa} assay indicated that mus308 mutants are defective in end-joining repair of DSBs. We observed no decrease in the percentage of red-eyed progeny recovered from mus308 mutant males (Figure 3B), suggesting that SDSA repair is not impaired when pol θ is missing or defective. In contrast, all four mus308 mutant alleles resulted in a significantly decreased percentage of yellow-eyed progeny relative to wild type (p<0.001, Kruskal-Wallis test). Because yellow-eyed progeny can only result from a repair mechanism involving end joining, these data suggest that pol θ is involved in an end-joining process.
To further demonstrate that pol θ is not involved in DNA synthesis during SDSA, we recovered independent (SDSA)+EJ events in males, isolated genomic DNA, and used PCR to estimate the approximate amount of DNA repair synthesis that occurred prior to end joining. The amount of repair synthesis in (SDSA)+EJ repair products did not differ significantly between wild-type and mus308 mutant flies (Figure 3C). We conclude that pol θ is not required for DNA synthesis during SDSA, but plays an important role in end-joining repair following aborted SDSA.
Mutations that abolish end joining in flies cause an increased frequency of genomic deletions during repair of site-specific DSBs [46], [47]. To determine whether mutation of mus308 also results in repair-associated deletions, we took advantage of the fact that deletions can be easily scored in the P element excision assay. Because P{wa} is inserted in the essential scalloped (sd) gene, repair events that delete into sd exons cause a scalloped-wing phenotype when recovered in heterozygous females and lethality in hemizygous males [48], [49]. We observed a substantial increase in the percentage of deletion-associated repair events isolated from mus308 mutant males relative to wild type (Figure 3D). Overall, the total percentage of end-joining repair events involving deletions recovered from mus308 mutants was elevated from 3- to 26-fold over wild type, depending on the mus308 allele tested.
Previously, we observed a similar deletion-prone phenotype in flies lacking the DmBlm protein, which is involved in repair of DSBs by SDSA [48]. Because our data did not support a role for pol θ in homologous recombination, we expected the deletion-prone phenotype of mus308 mutants to persist even in SDSA-deficient flies. To confirm this, we assayed repair following P{wa} excision in mus308 mutants lacking the Rad51 protein, which renders them unable to carry out HR repair [42], [45].
As expected, PCR analysis of repair products showed that SDSA was abolished in both spn-A and spn-A mus308 mutants (data not shown). Approximately 17% of P{wa} chromosomes recovered from spn-A mutant males showed evidence for end joining at the 17-nucleotide overhangs that are created by P transposase (Figure 4A and Table 1); the other 83% of P{wa} chromosomes recovered were presumably uncut. We observed a 30–50% decrease in end-joining repair products in spn-A mus308 double mutants compared to spn-A mutants (p<0.001, Kruskal-Wallis test), confirming a unique role for polθ in end joining when HR is absent. Importantly, mutation of mus308 still caused an increased percentage of deletions in the absence of Rad51 (Figure 4B). From these data, we conclude that the deletions formed during break repair in mus308 mutants are not the result of aborted SDSA. Rather, they are a consequence of a deletion-prone repair mechanism that operates in the absence of both SDSA and pol θ-dependent end joining.
During the course of these experiments, we made a number of observations suggesting that Rad51 and pol θ act in parallel and distinct DSB repair mechanisms. First, we recovered fewer spn-A mus308 double mutant males than would be predicted from Mendelian ratios. For example, in crosses between mus308D2 and mus308D5 heterozygotes, 38% of the progeny were mus308D2/mus308D5 compound heterozygotes. In contrast, only 16% of progeny recovered from parallel matings between spn-A057mus308D2 and spn-A093mus308D5 mutants were spn-A mus308 compound heterozygotes (P<0.05, Fisher's exact test; Figure 5A). This difference in viability between mus308 and spn-A mus308 mutants was even more extreme in flies in which excision of P{wa} was occurring (P<0.01; Figure 5A). In addition, we observed heightened male sterility in various combinations of spn-A mus308 mutants undergoing P{wa} transposition, with 51% of the double mutant males unable to produce viable progeny in the most severe allele combination (Figure 5B). Finally, we observed morphological abnormalities, specifically abdominal closure defects and aberrant cuticle banding patterns, in 100% of spn-A093mus308D5/spn-A057mus308D2 double mutants (Figure 5C). These defects were more severe in the double mutants experiencing P{wa} transposition, but were not apparent in either mus308 or spn-A single mutants. From these data, we conclude that Rad51 and pol θ participate in independent pathways required for repair of DSBs that arise during both endogenous developmental processes and during P element transposition.
P element ends are good substrates for DNA ligase 4-independent end joining [17]. Based on the results presented above, it seemed likely that pol θ is involved in an end-joining process different from C-NHEJ. To formally test this, we repeated the P{wa} assay in lig4 mus308 double mutants that lack DNA ligase 4 and are unable to repair DSBs by C-NHEJ. Unlike spn-A mus308 mutants, we observed no viability, fertility, or morphological defects in lig4 mus308 mutants. We also observed no defect in HR repair in the double mutants (Figure 4C), consistent with results obtained using mus308 single mutants. In contrast, we observed a further decrease in the percentage of end joining repair products recovered from lig4 mus308 double mutants relative to mus308 mutants, from 3.0% to 1.3% (P<0.01, Kruskal-Wallis test). Previously, we have shown that end-joining repair of DSBs induced by P{wa} excision is unaffected in lig4 mutants [17]. Therefore, the removal of pol θ-mediated end joining reveals a previously hidden role for DNA ligase 4 in the repair of DSBs created by P transposase. Strikingly, although only 50% of end-joining products isolated from mus308 mutants involved large, male-lethal deletions, 100% of end-joining products recovered from lig4 mus308 mutant males were associated with large deletions (Figure 4D).
From these results, we conclude that at least three distinct mechanisms for end-joining repair exist in Drosophila. One, which corresponds to C-NHEJ, requires DNA ligase 4 and other canonical NHEJ proteins, including XRCC4, Ku70, and Ku80 [46], [47], [50]. Another mechanism, which is at least partially independent of DNA ligase 4, is defined by a requirement for pol θ and corresponds to alt-EJ. Interestingly, alt-EJ appears to be used more frequently than C-NHEJ, at least for the repair of P element-induced breaks. In the absence of these two repair processes, a Rad51-independent backup mechanism characterized by extensive genomic deletions operates at low efficiency.
Alt-EJ repair in Drosophila is frequently associated with annealing at microhomologous sequences of more than four nucleotides and with long DNA insertions at repair junctions [8]. To determine whether pol θ-dependent end joining involves either of these types of repair, we sequenced repair junctions obtained from spn-A and spn-A mus308 double mutants following P{wa} excision. Because we sequenced only one junction per male germline, each junction analyzed represents an independent repair event. Five distinct junction types were identified. Three of these types are characteristic of junctions arising from C-NHEJ in mammalian systems [7]: junctions involving small, 1–3 base pair insertions, junctions involving annealing at 1–3 nucleotide microhomologies, and junctions for which no microhomologies can be identified (apparent blunt end junctions). The other two types of junctions, characteristic of alt-EJ [8], involve annealing at 5–10 nucleotide microhomologous sequences or insertions of more than three base pairs.
Approximately 58% of junctions from spn-A mutants showed structures considered typical of C-NHEJ repair, while 29% involved annealing at 5–10 nucleotide microhomologies and 13% had insertions of greater than three base pairs (Figure 6A and Table 1). Potential templates for the larger insertions could almost always be identified in flanking sequences. These insertions may be analogous to T-nucleotides that have been observed at translocation breakpoint junctions isolated from certain human cancers [21]–[23].
When we sequenced repair junctions from spn-A mus308 mutants, we observed two distinct patterns, depending on the mus308 alleles used. For both the D2/2003 and D5/2003 allele combinations, the percentage of junctions involving annealing at long microhomologies was significantly decreased (P<0.01, Fisher's exact test; Figure 6A, Table 2, and Table 3). Only 12% of D2/2003 junctions possessed an insertion greater than three base pairs, compared to 44% of junctions recovered from males with the D5 and 2003 alleles. In addition, most insertions isolated from D5/2003 males were highly complex and had multiple copies of imperfect repeats of T-nucleotides. Similar results were obtained with the D5/3294 allele combination (data not shown). An overall comparison of insert length showed that flies with wild-type mus308 alleles had an average insert length of 5.5 nucleotides, compared to 3.8 nucleotides for D2/2003 mutants and 13.3 nucleotides for D5/2003 mutants.
In summary, both mus308 mutant combinations significantly abrogated annealing at long microhomologies during alt-EJ repair. However, we observed a distinct difference in repair junctions recovered from males harboring the D2 allele, which greatly reduces overall pol θ protein levels [38], compared to flies with the D5 allele, which alters a conserved residue near the helicase-like domain. These results suggest that pol θ has two distinct functions in alt-EJ: one that promotes the annealing of long microhomologous sequences during end alignment, and another that is responsible for complex T-nucleotide insertions. Flies with the D2 allele are impaired in their ability to carry out both the annealing and insertion functions, whereas flies possessing the D5 separation-of-function allele cannot perform the microhomology annealing function but can still produce complex insertions.
P element-induced breaks are unique in that they possess 17-nucleotide non-complementary ends that are poor substrates for C-NHEJ. To test whether the results obtained with P elements can be generalized to other types of breaks, we used the I-SceI endonuclease and the previously characterized [Iw]7 reporter construct [50] to create site-specific DSBs in wild-type flies and flies lacking either DNA ligase 4 or pol θ. I-SceI produces a DSB with 4-nucleotide complementary overhangs that can be directly ligated through a C-NHEJ mechanism [50], [51]. Accurate repair regenerates the original I-SceI recognition sequence, which can then be cut again, while inaccurate end-joining repair abolishes further cutting. We utilized an hsp70 or ubiquitin-driven I-SceI construct integrated on chromosome 2 to drive high levels of I-SceI expression [50], [52]. Nearly 100% of repair events that we recovered involved gene conversion (HR repair from the homologous chromosome) or inaccurate end-joining (data not shown). In the [Iw]7 system, both gene conversion events and large deletions that remove the white marker gene are phenotypically indistinguishable. PCR analysis confirmed that many repair events recovered from mus308 mutants involved large deletions (>700 base pairs, data not shown). Our subsequent analysis focused on the characterization of repair events involving smaller deletions.
Twenty-three percent of I-SceI repair junctions isolated from wild-type flies possessed insertions of more than 3 base pairs (Figure 6B). This percentage was significantly increased to 46% in lig4 mutants (P<0.01, Fisher's exact test), consistent with increased use of alt-EJ in the absence of C-NHEJ. If pol θ plays a general role in insertional mutagenesis during alt-EJ repair, one would predict that the frequency and length of insertions following I-SceI cutting should decrease in mus308 mutants. Indeed, the percentage of large insertions decreased to 9% in mus308 mutants (P = 0.03, Fisher's exact test). Wild-type flies had an average insertion length of 7.6 base pairs, compared to 4.2 base pairs for mus308 mutants. Strikingly, no mus308 insertion was longer than twelve base pairs, while insertions of more than twenty base pairs occurred in both wild type and lig4 mutants. Because microhomologies of greater than four base pairs are not present near the I-SceI cut site in this construct, repair involving annealing at long microhomologies was not observed.
Surprisingly, the total percentage of repair junctions with short, 1–3 base pair insertions was not decreased in lig4 mutants relative to wild type (17% vs. 13%, respectively). Furthermore, the percentage of junctions involving annealing at 1–3 nucleotide microhomologies was also similar between the two genotypes (25% for lig4 mutants vs. 34% for wild type). These two types of junctions have historically been associated with ligase 4-dependent C-NHEJ repair. Our results suggest that this may not be the case. Indeed, a fine-level sequence analysis of I-SceI repair junctions that we have recently conducted suggests that alt-EJ may produce C-NHEJ-like junctions in certain sequence contexts [53]. Nevertheless, our data obtained using two independent site-specific DSB repair assays strongly suggest that C-NHEJ and alt-EJ represent at least partially independent mechanisms for the repair of DSBs and that pol θ plays an important role in the generation of T-nucleotide insertions during alt-EJ repair of both P element and I-SceI-induced breaks.
Several studies have identified proteins important for end-joining repair of DSBs in the absence of C-NHEJ factors in yeasts [11], [54], [55]. More recently, the Mre11 protein has been identified as an important alt-EJ component in vertebrate systems [10], [56]–[59]. However, much remains uncertain about the genetics or mechanisms of alt-EJ in metazoans. Our previous work demonstrated that end-joining repair of P element-induced breaks can occur independently of DNA ligase 4, suggesting the presence of a highly active alternative end-joining mechanism in Drosophila [17]. We have now identified the mus308 gene, encoding DNA polymerase theta, as a critical component of alt-EJ. Pol θ-dependent alt-EJ operates in parallel to C-NHEJ to promote repair of both P element and I-SceI-induced breaks. Because we observed similar alt-EJ defects with four different mus308 mutant alleles, several of which were studied in trans to at least two independently-derived deficiencies, we consider it highly unlikely that the phenotypes we observed are due to second-site mutations or other differences in genetic background.
Importantly, pol θ does not appear to participate in homology-directed repair. HR repair of DSBs following P{wa} excision is thought to proceed largely through synthesis-dependent strand annealing (SDSA) [60]. We observed that SDSA frequencies in the P{wa} assay were similar in wild type and mus308 mutants. Although we did not formally test for a role of pol θ in single-strand annealing (SSA), a non-conservative HR pathway that involves annealing at direct repeats larger than 25 nucleotides, a separate study demonstrated that pol θ has no effect on SSA repair [47]. In addition, comparison of the DNA sequences located 3 kilobases to either side of P{wa} by BLAST does not reveal any significant similarities of more than 20 nucleotides. Therefore, it seems unlikely that the repair observed in mus308 mutants arose through an SSA mechanism.
Three findings suggest that mus308-dependent alt-EJ is an important repair option for both cell and organism survival in flies, particularly in the absence of homologous recombination. First, spn-A mus308 double mutants are sub-viable and have severe defects in adult abdominal cuticle closure, consistent with a high level of apoptosis in rapidly proliferating histoblasts during pupariation. Second, spn-A mus308 double mutant males undergoing P{wa} excision have up to 30-fold increased sterility relative to spn-A mutants. Third, mus308 mutant males undergoing I-SceI cutting show premature sterility and produce few progeny. The few germline repair events that are recovered from each male are frequently clonal, suggesting extensive germline apoptosis (A. Yu, unpublished data).
Pol θ orthologs characterized from a variety of metazoans possess both helicase-like and DNA polymerase domains [27]–[31], [35]. Pol θ purified from both Drosophila and human cells has a Pol I-like polymerase activity and single-stranded DNA-dependent ATPase activity [30], [38]. However, DNA helicase activity of the purified protein remains to be demonstrated. Although our experiments did not formally test for helicase activity of pol θ, our results are consistent with pol θ having a DNA unwinding or strand displacement function. Flies with the D5 and 3294 mutations (located in or near the conserved helicase domain) produce repair products with complex T-nucleotide insertions but not products involving annealing at long microhomologies. The D5 and 3294 alleles may therefore encode proteins that retain polymerase activity but lack unwinding activity, resulting in an inability to expose internal microhomologous sequences. Because the microhomologies used in repair following P element excision are often located in the 17-nucleotide 3′ single-stranded tails, pol θ may also be important for the unwinding of secondary structures that form in single-stranded DNA. Alternatively, the DNA-dependent ATPase activity demonstrated by pol θ might represent an annealing function of the protein that is required during alt-EJ. Such an annealing activity was recently described for the human HARP protein, which is able to displace stably bound replication protein A and rewind single-stranded DNA bubbles [61].
One notable aspect of alt-EJ in Drosophila is the large percentage of repair junctions with templated insertions. These insertions may be “synthesis footprints” that are formed during the cell's attempt to create microhomologous sequences that can be used during the annealing stage of alt-EJ when suitable endogenous microhomologies are not present or are not long enough to allow for stable end alignment. Indeed, analysis of the insertions from I-SceI repair junctions suggests a model involving local unwinding of double-stranded DNA and iterative synthesis of 3–8 nucleotide runs [53]. The P{wa} repair junctions isolated from mus308D2/mus3082003 mutants are consistent with an important (but not exclusive) role for the polymerase domain of pol θ in the synthesis of T-nucleotides.
We speculate that pol θ may be involved in both DNA unwinding and repair synthesis during alt-EJ (Figure 7). Linking these two activities in one protein would provide a convenient mechanism for creating longer microhomologies that could increase the thermodynamic stability of aligned ends prior to the action of a DNA ligase. Studies based on the crystal structure of a dual function NHEJ polymerase-ligase protein found in Mycobacterium tuberculosis suggest that a synaptic function for an NHEJ polymerase is plausible [62]. Because ligase 4 is not involved in alt-EJ in Drosophila, another ligase must be involved in the ligation step. Studies from mammalian systems have identified DNA ligase 3 as a likely candidate [63], [64].
Pol θ was originally identified in Drosophila based on the inability of mus308 mutants to survive exposure to chemicals that induce DNA interstrand crosslinks. A crucial question posed by our findings is whether pol θ performs a common function during the repair of both DSBs and interstrand crosslinks. The C. elegans pol θ ortholog, POLQ-1, is also required for resistance to interstrand crosslinks and acts in a pathway that is distinct from HR but depends on CeBRCA1 [28]. In S. cerevisiae, several repair mechanisms are utilized during interstrand crosslink repair, including nucleotide excision repair (NER), HR, and translesion synthesis [65], [66]. Given our results and the findings from C. elegans, it seems unlikely that the role of pol θ in interstrand crosslink repair involves a function in HR.
In human cells, exposure to agents that induce interstrand crosslinks causes a shift in repair mechanisms that leads to increased use of non-conservative pathways associated with complex insertions and deletions [67]. Furthermore, interstrand crosslinks can cause frequent recombination between direct repeats [68], [69], suggesting that single-strand annealing may provide a viable mechanism for interstrand crosslink repair. The single-strand annealing model of interstrand crosslink repair posits that NER-independent recognition and processing of the crosslinked DNA is followed by generation of single-stranded regions flanking the crosslink and annealing at repeated sequences. Because alt-EJ frequently proceeds through annealing at short direct repeats, it is tempting to speculate that the role of pol θ in interstrand crosslink repair might be to expose and/or promote the annealing of microhomologous single-stranded regions that flank the crosslinked DNA. Consistent with this model, the initial incision step made after recognition of the interstrand crosslink remains normal in mus308 mutants [26]. Alternatively, pol θ might utilize its polymerase activity and nearby flanking sequences as a template to synthesize short stretches of DNA that could be used to span a single-stranded gap opposite of a partially excised crosslink. Such a model has been proposed to explain the formation of microindels in human cancers [70]. We are currently testing these two models using helicase- and polymerase-specific mus308 mutant alleles.
Although it seems counterintuitive, alt-EJ likely functions in some situations to promote genome stability. As evidence of this, we found that DSB repair following P element excision in mus308 mutant flies frequently results in genomic deletions of multiple kilobases. A similar deletion-prone phenotype was previously observed in mus309 mutants, which lack the Drosophila BLM ortholog [43], [71]. Epistasis analysis demonstrated that the mus309 deletion phenotype depends on Rad51, implying that DmBlm acts after strand invasion during HR and that the deletions observed in mus309 mutants are likely a result of failed SDSA [48]. In contrast, the deletions observed in mus308 mutants do not depend on Rad51, demonstrating that the function of pol θ in DSB repair is independent of HR. The deletion phenotype is exacerbated in lig4 mus308 double mutants, suggesting that C-NHEJ and alt-EJ represent two parallel mechanisms to prevent deletions. In the absence of these two end-joining options, resection at the broken ends may continue unchecked, resulting in extensive genomic deletions that are generated by an unknown Rad51-independent repair mechanism. Therefore, both C-NHEJ and alt-EJ function to prevent overprocessing of broken DNA ends and extreme degradation of the genome. Microhomology-mediated end joining, which shares many features with alt-EJ, has been proposed to perform a similar function in urothelial cells [72].
Nonetheless, alternative end-joining repair can also be genome destabilizing, as demonstrated by an increasing number of reports linking it to cancer. We have shown that complex insertions observed in alternative end-joining products are more frequent in flies possessing pol θ. These insertions, which are often combinations of nucleotides derived from several templates inserted in both direct and reverse-complement orientations, are remarkably similar to T-nucleotide insertions found in translocation breakpoints reconstructed from follicular and mantle cell lymphomas (reviewed in [73]). Therefore, if pol θ also functions in alternative end joining and T-nucleotide generation in mammals, it might be an important factor involved in translocation formation.
A recent study suggests that pol θ levels are tightly regulated in humans and that loss of this regulation may promote cancer progression [37]. The protein is primarily found in lymphoid tissues but is upregulated in lung, stomach, and colon cancers. Furthermore, high levels of pol θ expression correlate with poorer clinical outcomes. Intriguingly, pol θ is regulated by endogenous siRNAs in Drosophila [74], [75], although the significance of this regulation is currently unclear. We suggest that polθ-mediated alt-EJ serves as a medium-fidelity repair option used by cells when precise repair cannot be carried out for any number of reasons. As such, it prevents extreme loss of genetic information. However, its error-prone nature requires tight regulation, which, when lost, may lead to excessive inaccurate repair and ultimately, carcinogenesis.
The results described here establish that Drosophila pol θ plays two distinct roles in an alternative end-joining mechanism operating in parallel to canonical DNA ligase 4-mediated C-NHEJ. This novel finding lays the groundwork for future studies focusing on the specific roles of the pol θ helicase-like and polymerase domains in alt-EJ and DNA interstrand crosslink repair. Whether pol θ plays a similar role in alt-EJ in other organisms, including mammals, remains to be determined. Regardless, these studies reveal an unexpected role for DNA polymerase θ that is required for genomic integrity in Drosophila and possibly other metazoans.
All flies were maintained on standard cornmeal-based agar food and reared at 25°C. The mus308 D2 and D5 stocks were obtained from the Bloomington Stock Center and the 2003 and 3294 stocks were from the Zuker collection [76]. To identify mutations in these stocks, genomic DNA was isolated from flies harboring each allele in trans to Df(3R)Exel6166 and PCR was performed with primers specific to overlapping regions of the entire coding sequence. PCR products were sequenced and the sequence was compared to the Drosophila reference sequence release 5.10. Sequence changes unique to each allele were verified by sequencing in both orientations. The lig4169a [17], spn-A093 and spn-A057 [42] stocks harbor null alleles of DNA ligase 4 and Rad51, respectively.
For mechlorethamine sensitivity assays, balanced, heterozygous parents were crossed to Df(3R)Exel6166 and allowed to lay eggs in vials containing 10mL of food for three days, after which they were moved to new vials for two additional days. The first vials were treated with 250µL of 0.005% mechlorethamine dissolved in ddH2O, while the second vials were treated only with ddH2O. Survival was calculated as the number of homozygous mutant adults divided by the total number of adults that eclosed within 10 days of treatment. Ratios were normalized to untreated controls for each set of vials (five to eight sets of vials were counted for each experiment). For ionizing radiation sensitivity assays, heterozygous parents laid eggs on grape-juice agar plates for 12 hr. Embryos developed at 25°C until larvae reached third-instar stage, at which point they were irradiated in a Gammator 1000 irradiator at a dose rate of 800 rads/min and larvae were transferred to food-containing bottles. Relative survival rates were calculated as above.
Repair of DNA double-strand breaks was monitored after excision of the P{wa} transposon as described previously [43], [77]. P{wa} was excised in males using a second chromosome transposase source (CyO, H{w+,Δ2–3}) and individual repair events were recovered in female progeny over an intact copy of P{wa}. Females with two copies of P{wa} have apricot eyes [78]. Progeny with red eyes possess a repair event involving HR with annealing of the copia LTRs. A fraction of apricot-eyed females also possess HR repair events, but these cannot be distinguished from chromosomes in which no excision event occurred (using the CyO, H{w+,Δ2–3} transposase source, ∼80% of apricot-eyed female progeny inherit a non-excised P{wa} element). Yellow-eyed females harbor a repair event in which repair is completed by end joining.
For each genotype, at least 50 individual male crosses were scored for eye color of female progeny lacking transposase. The percentage of progeny from each repair class was calculated on a per vial basis, with each vial representing a separate experiment. Statistical comparisons were done with a Kruskal-Wallis non-parametric ANOVA followed by Dunn's multiple comparisons test using InStat3 (GraphPad).
For analysis of HR synthesis tract lengths, genomic DNA was purified and PCR reactions were performed as in [43], using primer pairs with the internal primer located 250, 2420, and 4674 base pairs from the cut site at the 5′ end of P{wa}.
For deletion analysis, the percentage of females with scalloped wings was calculated relative to all yellow-eyed females counted. The percentage of male lethal and small (0.1–3.6 kb) deletions was calculated based on a subset of yellow-eyed females (one from each original male parent) that were individually crossed to males bearing the FM7w balancer. Vials for which no white-eyed male progeny were recovered were scored as male lethal. Some of the male lethal events also caused a scalloped-wing phenotype in heterozygous females. For those that did not, testing to ensure that the male lethality was due to deletion of scalloped coding sequence was performed by recovering the repaired chromosomes in trans to the hypomorphic sd1 mutation [79] and scoring for a scalloped-wing phenotype. Repair events which could be recovered in males were subjected to PCR analysis, using primers internal to P{wa} [43], to detect small deletions into one or both introns of sd.
Repair of I-SceI mediated DNA double strand breaks was studied in the context of the chromosomally integrated [Iw]7 construct [52], which contains a single target site for the I-SceI endonuclease. DSBs were induced in the male pre-meiotic germline by crossing females harboring [Iw]7 to males expressing the I-SceI endonuclease from a second chromosomal location under the control of either the hsp70 promoter (70[I-SceI]1A) [52] or the ubiquitin promoter (UIE[I-SceI]2R) [50]. Independent inaccurate end-joining repair events from the male pre-meiotic germline were recovered in male progeny and DNA was isolated for analysis [80]. PCR was performed using primers PE5′ (GATAGCCGAAGCTTACCGAAGT) and jn3′b (GGACATTGACGCTATCGACCTA) to amplify a 1.3 kb fragment of the [Iw]7 construct including the I-SceI target site. Products were gel purified (GenScript) and sequencing of PCR products was performed using the PE5′ primer. Sequences were aligned using ClustalW or by manual inspection against sequence obtained from an uncut [Iw]7 construct. Statistical comparisons were done using Excel and SPSS.
|
10.1371/journal.pgen.1004612 | Arabidopsis RRP6L1 and RRP6L2 Function in FLOWERING LOCUS C Silencing via Regulation of Antisense RNA Synthesis | The exosome complex functions in RNA metabolism and transcriptional gene silencing. Here, we report that mutations of two Arabidopsis genes encoding nuclear exosome components AtRRP6L1 and AtRRP6L2, cause de-repression of the main flowering repressor FLOWERING LOCUS C (FLC) and thus delay flowering in early-flowering Arabidopsis ecotypes. AtRRP6L mutations affect the expression of known FLC regulatory antisense (AS) RNAs AS I and II, and cause an increase in Histone3 K4 trimethylation (H3K4me3) at FLC. AtRRP6L1 and AtRRP6L2 function redundantly in regulation of FLC and also act independently of the exosome core complex. Moreover, we discovered a novel, long non-coding, non-polyadenylated antisense transcript (ASL, for Antisense Long) originating from the FLC locus in wild type plants. The AtRRP6L proteins function as the main regulators of ASL synthesis, as these mutants show little or no ASL transcript. Unlike ASI/II, ASL associates with H3K27me3 regions of FLC, suggesting that it could function in the maintenance of H3K27 trimethylation during vegetative growth. AtRRP6L mutations also affect H3K27me3 levels and nucleosome density at the FLC locus. Furthermore, AtRRP6L1 physically associates with the ASL transcript and directly interacts with the FLC locus. We propose that AtRRP6L proteins participate in the maintenance of H3K27me3 at FLC via regulating ASL. Furthermore, AtRRP6Ls might participate in multiple FLC silencing pathways by regulating diverse antisense RNAs derived from the FLC locus.
| Arabidopsis FLOWERING LOCUS C (FLC) delays flowering; therefore, repressing expression of FLC provides a critical mechanism to regulate flowering. This mechanism involves multiple levels of regulation, including genetic regulation by transcription factors, and epigenetic regulation by modifications of genomic DNA and histones at the FLC locus. This work examines the role of non-coding RNAs in the epigenetic regulation of FLC, finding that the different RNAs produced from the FLC locus may have different functions in altering the epigenetic landscape at the FLC locus, and revealing that AtRRP6L1 and AtRRP6L2, two components of the exosome, an RNA-processing complex, play roles in regulating these non-coding RNAs. Therefore, this work explores the complex ties between RNA processing, non-coding RNAs, and epigenetic regulation of FLC, a key repressor of flowering.
| The regulation of gene silencing occurs at multiple levels and non-coding RNAs (ncRNAs) have emerged as important regulators of genome silencing at the transcriptional and posttranscriptional levels [1]. At the chromatin structure level, chromatin remodeling factors and DNA- and histone-modifying enzymes alter chromatin to control its structure and compaction, thus affecting silencing [2]. ncRNAs significantly contribute to the regulation of chromatin structure and play important roles in eukaryotic genomes by affecting the epigenetic architecture, including both establishment and maintenance of epigenetic marks. ncRNAs can initiate heterochromatin formation through the RNA interference (RNAi) pathway, or independently of RNAi, and through the RNA processing machinery [3], [4].
The exosome is an evolutionarily conserved complex of RNase-like and RNA binding proteins involved in 3′ to 5′ decay and processing of various RNA substrates [5]–[10]. The exosome complex plays an important role in regulating both coding and ncRNAs [11]. The nuclear and cytoplasmic forms of the eukaryotic exosome complex share ten common subunits [12]. In most organisms, all nine subunits of the exosome core complex are inactive, and enzymatic activities are provided by the tenth catalytic RRP44 subunit, a 3′-5′ hydrolytic exoribinuclease, which also has endonucleolytic activity [13].
The nuclear form of eukaryotic exosome also associates with the second catalytic RRP6 subunit, a substoichiometric, nuclear-specific, 3′ to 5′ exoribonuclease [14]–[16]. The RRP6 subunit has a number of unique functions in the exosome [14], and also additional functions not associated with the exosome core [17]–[19]. Arabidopsis has three possible functional homologs of RRP6: the nuclear proteins AtRRP6L1 and AtRRP6L2, and the cytoplasmic protein AtRRP6L3 [20]. When we purified the exosome complex from Arabidopsis, AtRRP6L proteins were not detected in our experiments, which is likely due to the fact that, as a substoichiometric subunit restricted to a nuclear form, it was underrepresented in our preparations [8]. Therefore, whether these Arabidopsis RRP6Ls physically associate with the exosome core remains to be tested.
The exosome complex broadly affects epigenetic silencing of heterochromatic and euchromatic loci by regulating a variety of ncRNAs [11], [21]–[24]. In fission yeast (Schizosaccharomyces pombe), the exosome acts in several different small RNA (smRNA) pathways to affect constitutive and facultative heterochromatin silencing, in either RNAi-dependent or RNAi-independent manners [4], [25]–[27]. The exosome also acts in gene silencing through RNA quantity and quality surveillance, and in collaboration with the 3′ termination machinery [21], [22], [28]–[30].
Our previous genome-wide survey revealed that many exosome targets in Arabidopsis correspond to ncRNAs, many originating from heterochromatic loci, suggesting that the exosome participates in various silencing pathways in Arabidopsis [8]. Our recent analysis of exosome functions in smRNA-mediated silencing of genes in Arabidopsis showed that the exosome has little effect on the smRNAs that function in the main silencing mechanisms, siRNA-dependent methylation of DNA (RdDM). Rather, we showed that the exosome associates physically with long, polyadenylated RNAs transcribed from the scaffold regions of several heterochromatic loci, and exosome defects affected the level of histone H3K9me2, an epigenetic mark that alters chromatin structure [31]. We also found that the Arabidopsis rrp6 homologues AtRRP6L1 and AtRRP6L2 participate in these epigenetic mechanisms and may function redundantly [31]. With the exception of AtCSL4, the genes encoding the Arabidopsis exosome core complex subunits are essential for viability [8]. By contrast, unlike the core exosome subunits, the exosome nuclear catalytic subunit RRP6 and the Arabidopsis RRP6-Like proteins are not essential for viability [14], [20]; thus, the rrp6 and rrp6l mutants provide tools to study the role of the exosome during development.
Epigenetic regulation by long non-coding RNAs (lncRNAs) and histone modifications plays a key role in controlling the expression of Arabidopsis FLC (FLOWERING LOCUS C), which encodes a MADS-box transcription factor that suppresses flowering [32]–[35].
FLC regulation through chromatin modifications has been well studied [36], [37]. FLC expression requires H3K4 methylation, a permissive chromatin modification, on FLC chromatin, and H3K4 demethylation leads to FLC repression [37]–[39]. FLC silencing requires Polycomb complex 2 (PRC2) -deposited H3K27me3, a repressive chromatin modification [36], [40], [41]. Repressing the expression of FLC provides a central mechanism for both the vernalization pathway, which regulates flowering time in response to periods of prolonged cold, and the autonomous pathway, which regulates flowering independently of environmental signals [37].
The processing and metabolism of lncRNAs play a crucial role in FLC silencing, and different lncRNAs produced from the FLC locus may have distinct functions. For example, lncRNAs COOLAIR and COLDAIR participate in the epigenetic silencing of the FLC locus [40], [42]. The COOLAIR antisense transcript is produced from the FLC locus as two alternatively polyadenylated isoforms, AS I and AS II, and it was shown that the processing of AS I and II function in FLC epigenetic silencing by affecting H3K4 demethylation [37], [38], [42]. COOLAIR transcription does not appear to be required for vernalization, but it has been implicated in FLC repression early during cold treatment, possibly mediated by direct effects on the FLC promoter [37]. By contrast, the COLDAIR sense transcript is produced from within FLC intron 1, and plays a role in FLC silencing via recruitment of Polycomb repressing complex 2 (PRC2) to FLC during vernalization [40]. The establishment of H3K27 trimethylation to silence FLC in vernalization requires COLDAIR [40], but not AS I and AS II [41], [43]. Also, in the autonomous pathway, the proteins involved in 3′-end RNA processing act as the main factors in FLC silencing [38], [39], [44]–[46].
Here, we report that mutations of AtRRP6L1 and AtRRP6L2 result in delayed flowering in early-flowering Arabidopsis ecotypes that do not require vernalization for flowering. We found that AtRRP6L1 and AtRRP6L2 epigenetically regulate FLC silencing by regulating different antisense transcripts and modulating H3K4me3 and H3K27me3 histone modifications at the FLC locus. Moreover, we discovered a novel antisense transcript, termed Antisense Long (ASL), which originates from the FLC locus in wild type plants and is regulated by AtRRP6L1 and AtRRP6L2. Our study demonstrates that Arabidopsis RRP6L proteins play an important role in the regulation of genes expressed in specific developmental phases via participating in lncRNA-mediated epigenetic silencing.
We previously examined T-DNA mutations in Arabidopsis RRP6L1 and RRP6L2 and found that these mutations lead to de-repression of known heterochromatic loci and that RRP6L1 and 2 likely function redundantly in this process [31]. The T-DNA insertion allele of AtRRP6L1 was isolated from the Wisconsin population of T-DNA mutants [31], in the Wasilevskaya (Ws) ecotype, and the T-DNA insertion allele of AtRRP6L2 comes from the SALK collection, in the Columbia (Col-0) ecotype [31]. We previously used RT-PCR analysis to demonstrate that rrp6l1-3 mutant is a null allele and rrp6l2-3 is nearly null [31].
To control for the ecotype, we examined the phenotypes of rrp6l1-3, rrp6l2-3, and rrp6l1-3 rrp6l2-3 double mutants (hereafter rrp6l1/2) and compared them to wild-type plants of Col-0 and Ws ecotypes. When we examined the phenotype of different rrp6l1 and rrp6l2 alleles, we found that these single mutants did not show significant phenotypic alterations, although they exhibited a mild delay in flowering, as measured by leaf number at bolting (Fig. 1A). By contrast, the late-flowering phenotype becomes pronounced in rrp6l1/2 double mutants grown under long day conditions (Fig. 1A) and becomes very severe in plants grown under short day conditions (Fig. 1B). These findings indicate that AtRRP6L1 and AtRRP6L2 likely have redundant functions in the pathways activating flowering. The flowering defects in plants grown under short and long day conditions suggest that AtRRP6L1 and AtRRP6L2 function in the regulation of flowering through a pathway different from the photoperiod pathway, which promotes flowering in response to day length. The vernalization and autonomous pathways promote flowering through repression of FLC expression. During vernalization, cold temperatures repress FLC expression, and FRIGIDA (FRI) activates FLC expression [35]. Ecotypes that lack functional alleles of FRI, such as the early-flowering accessions Col-0 and Ws ecotypes used in this study, do not require vernalization for flowering. Thus, our data suggest that AtRRP6L1and AtRRP6L2 could be involved in regulation of FLC expression through the autonomous flowering pathway.
To examine the roles of AtRRP6L1and AtRRP6L2 in regulation of FLC expression, we examined whether mutations of AtRRP6Ls affect the expression of FLC. We observed no change in FLC expression in the rrp6l1-3 and rrp6l2-3 single mutants, consistent with the degree of the observed phenotypic alterations, and we confirmed this observation with additional AtRRP6L1 and AtRRP6L2 T-DNA alleles (Fig. S1B). By contrast, we observed an increase in the levels of FLC transcript in the rrp6l1/2 double mutant relative to wild-type plants (Fig. 1C). These results suggest that AtRRP6L1 and AtRRP6L2 affect FLC expression and function redundantly in this process. To make sure that this phenotype does not result from transgressive segregation, we constructed AtRRP6Ls mutants using different AtRRP6L alleles isolated from the same Col-0 background and confirmed that this mutant combination causes a similar delay in flowering and derepression of FLC (Fig. S1A).
FLC acts as a dosage-dependent floral repressor [47]. The exosome complex and RRP6 subunits affect the processing and turnover of various RNAs, and thus regulate RNA quality and quantity [5], [8], [11]. To find out whether the FLC transcript, which increased in abundance in rrp6l1/2 mutants, corresponds to a functional transcript rather than nonfunctional byproduct, we examined the expression of the flowering genes SUPPRESSOR OF CONSTANS OVEREXPRESSION 1 (SOC1) and FLOWERING LOCUS T (FT), which act downstream of FLC. We found that the rrp6l1/2 mutants showed lower levels of both SOC1 and FT transcripts (Fig. 1C), indicating that the increased level of FLC transcript in rrp6l1/2 mutants corresponds to a functional FLC transcript, and the increase in FLC expression enhances the repression of the downstream genes. In addition to increased levels of FLC transcript, we also detected increased amounts of the unspliced FLC RNA in rrp6l1/2 mutants (Fig. 2B).
The autonomous pathway constitutively represses FLC, independent of environmental inputs [35]. To find out whether AtRRP6L1 and AtRRP6L2 affect FLC expression directly or by regulating the expression of the upstream genes that silence FLC in the autonomous pathway, we examined the expression of upstream genes in the AtRRP6L1 and AtRRP6L2 mutants. We found that the rrp6l1/2 plants showed no changes in expression of the genes that act upstream of FLC (Fig. 1D). These data imply that the de-repression of FLC observed in AtRRP6L mutants results from a direct effect of AtRRP6Ls on the expression of FLC, not from regulation of the genes acting upstream of FLC.
To test whether the effect of rrp6l1/2 on FLC expression requires the core exosome complex, we next examined AtCSL4-2 and AtRRP41, which encode core exosome complex subunits. We did not observe de-repression of FLC transcription in an AtRRP41 inducible RNAi line or in AtCSL4-2 T-DNA insertion mutants (Fig. 1E), suggesting that AtRRP6L1 and AtRRP6L2 likely function independently of the exosome in the regulation of FLC expression.
Antisense transcripts regulate FLC silencing and antisense expression appears to independently intersect with both the vernalization and autonomous pathways to repress FLC expression [37]. During the vegetative phase, the FLC locus produces two alternatively spliced, polyadenylated regulatory antisense (AS) transcripts, AS I and AS II [37] (Fig. 2A). Targeted 3′ end processing of these antisense transcripts affects the recruitment of histone chromatin remodelers to the locus, which results in reduced FLC transcription [38], [39], [46]. Therefore, we investigated whether the rrp6l1/2 mutants showed changes in the ratio of 3′ end processing and polyadenylation of these antisense transcripts. We found that, compared to wild type plants, the rrp6l1 or rrp6l2 single mutants, and the rrp6l1/2 double mutants had lower levels of processed AS I and II transcripts (Fig. 2C and Fig. S1C); also, the rrp6l1/2 plants showed reduced levels of AS I and II, consistent with the stronger phenotype of the rrp6l1/2 mutants (Fig. 1A and S1C). Interestingly, the pattern of down-regulation of AS I and II transcripts in rrp6l1/2 mutants was similar to the pattern observed in the mutants of cleavage stimulation factors CstF64 and CstF77, components of the cleavage polyadenylation machinery required for the 3′-end processing of AS I and II transcripts [38]. RRP6 plays an important role in formation of the 3′ ends of many RNAs in yeast and humans [14], [48], [49], and in budding yeast, also participates in the regulation of antisense transcripts derived from the PHO84 locus [21], [50]. Therefore, it is possible that AtRRP6L1 and AtRRP6L2 proteins, along with CstF, could participate in the 3′ end processing of both antisense transcripts. To further investigate the relationship between CstF64 and AtRRP6Ls, we attempted to construct a triple rrp6l1-3 rrp6l2-3 cstf64-2 mutant; however, since cstf64-2 mutants are sterile, we could not obtain the triple homozygous mutant. Taken together, our data suggest that AtRRP6L proteins could negatively regulate FLC expression by affecting the expression of the regulatory antisense transcripts.
Methylated H3K4 marks active chromatin states and the antisense transcripts synthesized from FLC may function in FLC silencing by recruiting chromatin remodeling factors that drive H3K4 demethylation [38], [39]. To investigate whether the decrease in the antisense transcripts in the rrp6l1/2 mutants leads to changes in histone modifications at the FLC locus, we used chromatin immunoprecipitation (ChIP) to analyze the levels of H3K4me3 at various regions of the FLC locus (Fig. 2A). We found that the rrp6l1/2 mutant had significantly increased levels of H3K4me3 along the entire length of FLC, compared with wild type (Fig. 2D). These data suggest that the decrease in the level of antisense transcripts in rrp6l1/2 mutants might lead to the decreased recruitment of chromatin remodeling factors required for H3K4 demethylation, thereby regulating the accessibility of the transcription machinery to the locus, and in turn leading to FLC de-silencing, similar to previously reported observations [38].
The expression of FLC sense and antisense transcripts together with the increased level of H3K4 trimethylation in rrp6l1/2 mutants suggested that AtRRP6s could participate in FLC transcriptional silencing by affecting the chromatin structure at the FLC locus. Therefore, we asked whether AtRRP6L proteins can directly interact with the FLC locus to participate in the silencing pathway. To address this question, we constructed transgenic rrp6l1-3 lines that were complemented by a wild type copy of AtRRP6L1 fused with the TAP-tag for affinity purification, AtRRP6L1-TAP (see Methods).
We then used ChIP on these lines to examine the association of AtRRP6L1 protein with several regions of FLC (Fig. 2A). ChIP showed a modest enrichment of AtRRP6L1 protein in the regions corresponding to the 3′-UTR and intron 1 of FLC (Fig. 2E). The AtRRP6L1 binding region within intron 1 appears to be further downstream of the 3′-end of the AS I transcript (with respect to the direction of AS I and II transcription), implying that AtRRP6L1 may bind in this region to process a longer antisense precursor transcript. The AtRRP6L1 binding regions do not overlap with the regions reported to be bound by FPA and FCA proteins, RNA-binding 3′-end processing factors required for the processing of the AS I transcript [39], [46]. However, FPA binds to the region between exon 4 and 5 (Fig. 2A), which is also downstream of the 3′-end region of AS I [46]; thus, AtRRP6L1 and FPA have somewhat similar, but not identical, binding patterns (Fig. 2A and E). The association of AtRRP6L1 protein with a larger region of the FLC locus (downstream of the 3′end of AS I) also suggests that AtRRP6L proteins might participate in the processing of different types of antisense RNAs, in addition to the known antisense transcripts derived from the FLC locus. Alternatively, AtRRP6L could also participate in co-transcriptional regulation of antisense transcription by binding to nascent antisense transcripts. Also, the level of AtRRP6L1 enrichment at the FLC locus was relatively modest (Fig. 2E). Thus, we cannot rule out the possibility that AtRRP6L proteins could associate with the locus by binding other protein and RNA complexes that physically interact with the locus.
When we were examining the pattern of known antisense transcripts produced from the FLC locus, we observed the presence of a different antisense transcript in wild type plants, but not in rrp6l1/2 plants. Therefore, we set out to investigate the features of this antisense transcript by tiling RT-PCR using a set of primers that cover the entire FLC locus (Fig. S2A). We found that the transcript is a novel antisense RNA of over 2000 nucleotides in length. The sequence of this antisense RNA, which we termed ASL (Antisense Long), corresponds to intron 1 and the 3′-UTR region of the sense FLC transcript (Fig. 3A). Interestingly, 5′ region of the ASL transcript overlaps with the 5′ region of the AS I and II transcripts. Sequencing of ASL revealed that it has two different isoforms, 2,236 nt and 2,536 nt (ASLa and b, respectively), produced by alternative splicing (Fig. 3A). Moreover, ASL spans intron 1, an important region for maintenance of FLC silencing [51]. Notably, ASL also overlaps with the COLDAIR lncRNA, which is transcribed within intron 1 in the sense direction during vernalization [40] (Fig. 3A).
To determine whether the ASL RNA has a 5′ cap, we used Terminator 5′-Phosphate-Dependent Exonuclease (TPE), which degrades uncapped RNA. We found that TPE treatment did not affect ASL levels, indicating that ASL has a 5′ cap (Fig. 3B). Next, we examined the 3′ end of ASL by performing cDNA synthesis primed by either sequence specific primer or oligo-dT primers. To our surprise, we found that ASL is not polyadenylated, as we detected ASL only from the cDNA primed by specific primers, not by oligo-dT (Fig. 3C).
In plants, the RNA polymerases RNA Pol IV and Pol V [52] participate in gene silencing through smRNA-mediated mechanisms [53]–[55]. To investigate which RNA polymerase synthesizes ASL, we treated plants with α-amanitin, an inhibitor of RNA Pol II, and then used RT-PCR to examine ASL levels. We did not detect ASL in plants treated with α-amanitin (Fig. 3D), implying that RNA Pol II synthesizes ASL. To confirm this, we also examined the presence of ASL in nrpd1 (Pol IV) and nrpe1 (Pol V) mutants; we detected ASL in these mutants, indicating that Pol IV and V do not affect ASL, although nrpd1 mutants showed a minor change in ASL levels (Fig. S2B).
Taken together, our data indicate that ASL is capped, synthesized by RNA Pol II and non-polyadenylated, and also suggest that it is distinct from the known antisense transcripts originating from the FLC locus. The tiling RT-PCR analysis indicates that the same promoter produces ASL, AS I, and AS II. Thus, it is possible that the ASL transcript could function differently from AS I and II in the silencing of FLC. Indeed, different antisense RNAs transcribed from same promoter of the human pseudogene PTENpg1 have different functions in transcriptional and post-transcriptional silencing of the tumor suppressor gene PTEN [56].
Next, we investigated how the AtRRP6Ls participate in the regulation of ASL expression. We found that the level of ASL transcript decreased in rrp6l1-3 and rrp6l2-3 single mutants (Fig. 4A and Fig. S2C). Moreover, we detected little or no ASL transcript in rrp6l1/2 double mutants (Fig. 4A), indicating that AtRRP6L proteins function as the main factors regulating the levels of the ASL transcript. Consistent with AtRRP6L functions in regulation of ASL expression, we observed that the level of the ASL transcript recovered to wild type levels in rrp6l1-3 mutant complemented by the wild type copy of AtRRP6L1-TAP (Fig. 4B). These data suggest that both AtRRP6L1 and AtRRP6L2 directly regulate the expression of ASL and are the main factors in this process.
RRP6 is a 3′ - 5′ exoribonuclease and RRP6 defects usually result in abnormal accumulation of various RNAs due to failure to degrade or process them [21], [49], [50], [57]. We next asked how AtRRP6Ls could regulate ASL levels. The rrp6l1/2 mutants showed nearly undetectably low levels of ASL. We reasoned that, if AtRRP6Ls regulate the stability of the ASL transcript, then we would observe a difference in ASL decay rate in AtRRP6L single mutants. Therefore, we used the rrp6l1-1 single mutant and conducted an α-amanitin chase to compare the rates of ASL transcript decay in rrp6l1-1 and wild type plants. We found that the rrp6l1-1 mutant and wild type had similar rates of ASL transcript decay (Fig. 4C), suggesting that AtRRP6L proteins do not directly participate in the degradation of the ASL transcript but rather affect its production or biogenesis.
To find out whether AtRRP6L1 could play a direct role in the expression of ASL, we examined if AtRRP6L1 protein physically associates with the ASL transcript. To this end, we conducted RNA immunoprecipitation (RNA-IP) in wild type plants using antibodies against AtRRP6L1 protein (Fig. 5A). The RNA-IP showed that AtRRP6L1 protein physically associates with the ASL transcript (Fig. 5A). We also obtained identical results from RNA-IP in the rrp6l1-3 mutant complemented with AtRRP6L1-TAP (Fig. 5B). Together these results indicate that AtRRP6L1 protein likely participates directly in the regulation of ASL transcript levels.
We then hypothesized that the ASL RNA may play a role distinct from that of AS I and II. Histone remodeling factors affect FLC silencing and several lncRNAs affect H3K4 demethylation and H3K27 trimethylation [41]. The sense lncRNA COLDAIR participates in recruiting PRC2 and is necessary for the establishment of H3K27 trimethylation during vernalization [40], but this does not require AS I and II [41], [43]. However, H3K4 demethylation in the autonomous pathway does require AS I and II [37]–[39]. COLDAIR may also contribute to the maintenance of H3K27me3 during vegetative growth. In addition, the H3K27me3-binding protein LHP1 functions in the maintenance of H3K27me3 during the vegetative phase in actively dividing cells, suggesting that H3K27me3 maintenance could be important for FLC silencing during the vegetative phase after H3K27me3 has been established [58].
To examine whether ASL has a role distinct from that of AS I and II, we examined the level of the repressive histone mark, H3K27me3 at the FLC locus in rrp6l1/2 mutants. To our surprise, we observed that the rrp6l1/2 mutants showed significantly decreased levels of H3K27me3 along the entire FLC locus (Fig. 5C). The rrp6l1-3 single mutant also showed a mild decrease in the level of H3K27me3 (Fig. S1D). This observation is consistent with the very mild phenotype and the decreased level of the ASL transcript observed in rrp6l1 single mutants (Fig. 1A, 4A and S2C). Together, our data indicate that the knock-out of both AtRRP6L1 and AtRRP6L2 affected the levels of both H3K4me3 and H3K27me3 in the FLC locus (Fig. 3B). This is in contrast to AS I and II, which affect only the levels of H3K4me3 at FLC, at least in vernalization pathway [37], [41], [43].
The level of H3K27me3 correlates with the nucleosomal density [59], [60]. To examine whether the reduction in the level of H3K27me3 in rrp6l/2 mutants affects nucleosome positioning, we performed Micrococcal Nuclease (MNase)-ChIP assays using anti-H3 antibodies. MNase degrades nucleosome-free regions, allowing the estimation of nucleosome density. We found a lower nucleosomal density at the FLC locus in rrp6l1/2 mutants (Fig. 5D). These data indicate that the reduction of H3K27me3 levels observed in the rrp6l1/2 mutants could also result in relaxation of the chromatin state, which then allows factors involved in FLC transcription to gain easier access to the locus. Taken together, our results suggest that the regulation of the ASL transcript by AtRRP6L proteins may contribute to the maintenance of H3K27me3 at the FLC locus, which in turn contributes to the compact chromatin structure of the locus.
The rrp6l1/2 mutations lead to a decrease of H3K27me3 levels and also affect the nucleosome density at the FLC locus (Fig. 5C and D). Therefore, we asked whether the ASL transcript could be directly involved in H3K27 trimethylation, a role similar to that played by the COLDAIR lncRNA. To answer this question, we performed RNA-IP using antibodies against H3K27me3. We found that the ASL transcript physically associates with H3K27me3 regions (Fig. 5A). Taken together, our data suggest that the ASL transcript could function in the maintenance of H3K27 trimethylation during the vegetative phase.
Here, we report that AtRRP6L1 and AtRRP6L2, the homologues of rrp6 subunits of the exosome in other organisms, negatively regulate the expression of FLC to promote flowering in Arabidopsis, and act redundantly in this process. This finding provides the first evidence that plant RRP6 proteins participate in the regulation of a specific developmental process. Our data indicate that AtRRP6L proteins play a role in the regulation of antisense RNA production and the chromatin landscape of the FLC locus. Moreover, we identified a novel antisense transcript produced from the FLC locus in wild type plants and found that AtRRP6L proteins regulate the expression of this transcript. ASL appears to be distinct from the previously described FLC antisense transcripts, AS I and II, and our data suggest that it could function in the regulation of levels of H3K27me3 at the FLC locus.
For most exosome complex subunits, mutations cause a lethal phenotype, which indicates that, for most organisms, development requires exosome-mediated regulation of diverse RNAs [8], [61]. Unlike exosome core subunits, RRP6 and AtRRP6-Like proteins are not essential for viability [14], [20]. Most studies of exosome-dependent and exosome-independent RRP6 functions in developmental processes have been performed in fission and budding yeast [18], [21], [24], [49], [57], [62]. In these systems, RRP6 participates in facultative gene silencing and also regulates the transition from mitosis to meiosis through RNA-mediated epigenetic mechanisms [24], [57]. Similar to these findings, we observed that AtRRP6L1 and AtRRP6L2 function in regulation of flowering time by repressing FLC expression. The defect in the AtRRP6L proteins results in mis-regulation of antisense RNA production from the FLC locus and affects the level of histone modification of the locus.
AS I and II transcripts function in FLC silencing in both vernalization and autonomous flowering pathways [37], possibly by affecting the level of H3K4 demethylation [38], [39], although the exact mechanism remains unknown. A decrease in the levels of processed AS I and II in mutants of CstF64 and CstF77, proteins involved in 3′-end processing, leads to increased levels of H3K4 trimethylation and subsequent de-repression of FLC [38]. Thus, AtRRP6Ls may contribute to 3′-end processing of the antisense RNAs similarly to the CstFs; indeed, 3′-end processing of various RNAs is one of the well-known functions of the exosome complex. The exosome processes a number of structural RNAs including rRNA, snRNA and snoRNA via trimming the 3′-ends of their precursors [63], [64]. Furthermore, RRP6 acts together with the Nrd1-Nab3 termination complex in budding yeast in non-canonical 3′-end processing and termination of the antisense RNA derived from PHO84, as well as processing of several other mRNAs [49], [50]. Disruption of the Nrd1-exosome pathway leads to de-repression of reporter genes integrated into heterochromatic regions and results in alteration of chromatin structure at specific loci and heterochromatic regions [11], [21], [22]. The exosome function in processing of mRNA and antisense lncRNA is likely to be conserved in plants, suggesting that AtRRP6L proteins may participate in regulating synthesis of the antisense RNAs derived from the FLC locus and this regulation could be important to maintain a repressive chromatin state for silencing of FLC.
Surprisingly, we found that the defect of AtRRP6Ls caused a reduction of the level of H3K27 trimethylation, which has not been reported in studies of AS I and II in the vernalization and autonomous flowering pathways [38], [43]. A different intronic sense lncRNA, COLDAIR, derived from intron 1 of FLC, physically associates with the PHD-PRC2 complex to establish H3K27 trimethylation during vernalization [40]. Moreover, the AS I and II transcripts are not required for PcG-mediated silencing via regulation of H3K27me3 trimethylation in the vernalization pathway [41], [43]. Together, these data suggest that other lncRNAs, not AS I and II, function in regulating H3K27 trimethylation.
AtRRP6Ls might affect the level of H3K27me3 indirectly, by regulating H3K4 demethylation through regulating either AS I and II, or ASL transcripts, via a mechanism similar to the interplay between Trithorax and Polycomb groups, which antagonistically regulate the levels of H3K4me3 and H3K27me3 at the FLC locus during vernalization [40], [41]. Alternatively, AtRRP6Ls may regulate the level of H3K27me3 directly via an unknown mechanism that functions independently of the silencing pathway involving AS I and II. Indeed, ARABIDOPSIS TRITHORAX 1 (ATX1), an Arabidopsis homolog of Trithorax 1, dynamically regulates activation of FLC through trimethylation but not dimethylation of H3K4 and atx1 mutations led to the loss of H3K4me3 and gain of H3K27me2 during the vegetative phase, but did not affect H3K4me2 and H3K27me3 [65]. This indicates that the regulation of the levels of H3K4me3 and H3K27me3 at the FLC locus could be independent of each other, at least during the vegetative phase. Together, the previous reports and our data suggest that the decrease of H3K27me3 in rrp6l1/2 mutants could be caused by the decrease in ASL RNA expression, and AtRRP6Ls may participate in the respective pathways for FLC silencing through regulating the expression of AS I, AS II, and ASL RNAs.
We identified ASL, a novel, long antisense RNA that is distinct from the previously-described antisense RNAs. We observed that the ASL transcript physically associates with H3K27me3, suggesting that it could play a role in H3K27 trimethylation and function differently than AS I and II RNAs. The PRC2 complex binds ncRNAs with high affinity but does not recognize specific sequences, while its binding affinity correlates with the length of the RNA [66], [67]. Thus, it is possible that ASL transcript could also participate in recruitment of the PRC2 complex to the FLC locus, which leads to maintenance of H3K27 trimethylation during vegetative growth. Taken together, the previous reports and our data suggest that the regulation of chromatin structure via various lncRNAs is a central mechanism in FLC silencing, and different lncRNAs may function in different chromatin modification pathways. The AtRRP6L proteins may play a role in silencing pathways by regulating antisense transcription.
ASL has several features in common with AS I and AS II. First, ASL is transcribed by RNA Pol II; second, it is alternatively spliced, existing in 2 isoforms; third, its transcription is driven by the same promoter that drives AS I and AS II; fourth, the 5′ part of ASL overlaps with the 5′ region of AS I and II. However, in contrast to AS I and II, the ASL transcript is long (over 2,000 nucleotides long), is non-polyadenylated, and extends into intron 1 of FLC. These differences suggest that ASL may have functions distinct from the functions of AS I and II in FLC silencing, and the mechanism of FLC silencing could be more complicated than previously thought.
RNA Pol II transcribes ASL. Various non-polyadenylated Pol II RNAs, such as snRNA, snoRNAs, and some mRNAs, are processed by the exosome, which is recruited by the Nrd1-Nab3-Sen1 termination complex in the noncanonical 3′ end-processing pathway [22], [49], [50], [63], [68]. Thus, the noncanonical 3′ end-processing pathway may also participate in processing of ASL, if this pathway is conserved in plants. However, Arabidopsis homologs of Nrd1, Nab3 and Sen1 have not yet been characterized.
The rrp6l1-3 and rrp6l2-3 single mutants showed decreased levels of ASL, and ASL was almost undetectable in the rrp6l1/2 double mutant. Based on the results of the α-amanitin chase experiments, the rrp6l1-1 mutation does not affect the stability of ASL, suggesting that AtRRP6L1 and AtRRP6L2 could be the main regulators of ASL synthesis. This finding is very intriguing, since defects in the exosome and RRP6 usually lead to abnormal accumulation of various RNAs due to failures of RNA degradation or processing [21], [22], [49], [50], [57]. This result may be caused by an unknown function of AtRRP6L proteins, which participate in either the synthesis or biogenesis of ASL, rather than in its degradation. Indeed, we previously reported that the expression of a number of loci decreased in AtRRP4 and AtRRP41 inducible RNAi plants and the AtCSL4-2 mutant [8], and many of these loci are located within euchromatic regions as well as in regions harboring H3K27me3 (unpublished data). Therefore, the exosome and AtRRP6Ls may function in regulation of RNA synthesis, different from their conventional functions in RNA degradation. Similarly, inactivation of the human homologue of RRP6 leads to dramatically reduced levels of Xist ncRNA involved in X-chromosome inactivation, although it remains to be seen whether this effect is direct [69].
In our study, we found that AtRRP6L1 protein physically associates with the ASL transcript, suggesting that AtRRP6L1 plays a direct role in regulation of ASL levels, likely through ASL synthesis rather than degradation. In addition, recent work demonstrated that another 5′-3′ exoribonuclease, Xrn1, also directly contributes to RNA synthesis of several mRNAs in budding yeast, by physically associating with chromatin and contributing to transcription elongation [70]. Alternatively, the decrease in ASL levels in the AtRRP6L mutants may indicate that different RNA decay proteins participate in degradation of these RNAs.
We found that the level of FLC transcript was unaffected in the exosome core subunit mutants, AtRRP4 and AtRRP41 inducible RNAi lines and AtCSL4-2 T-DNA mutants, suggesting that the function of AtRRP6L1 and AtRRP6L2 in regulation of FLC expression could be independent of the exosome complex. This could also indicate that the exosome core complex is not necessary for metabolism of RNAs produced from the FLC locus and different RNA decay factors could participate in their degradation. For example, in yeast, the XRN family of 5′ to 3′ exoribonucleases works in both the nucleus and cytoplasm, and has diverse functions in RNA metabolism [71], including in the degradation of XRN1-sensitive unstable antisense RNAs [72].
Silencing of FLC is regulated mainly through histone modifications rather than DNA methylation [73]. We previously reported that the exosome complex and AtRRP6L proteins function in DNA methylation-independent silencing and affect the histone modification pathway in some heterochromatic loci in Arabidopsis [31]. Along with our previous findings, regulation of FLC silencing mainly by histone modifications suggested that the FLC locus could be one of the targets of the AtRRP6L proteins. In accord, we observed that defects in AtRRP6L proteins caused a decrease in antisense RNAs, resulting in the alteration of histone modifications and de-repression of FLC.
It is intriguing to speculate that AtRRP6L proteins may have dual functions in FLC silencing via regulation of antisense transcription, which means that AtRRP6L proteins could participate in 2 different pathways, one involved in H3K4 demethylation and the other involved in H3K27 trimethylation. First, AtRRP6Ls could participate in the H3K4 demethylation pathway via regulating synthesis of AS I and AS II. Second, AtRRP6L proteins could function in the H3K27 trimethylation pathway via regulating the synthesis of ASL. However, more work will be needed to untangle the interrelationships of the different lncRNAs and the roles they play in the epigenetic architecture at FLC. How AtRRP6L proteins and the ncRNAs controlled by them help recruit chromatin modifiers to modulate silencing by affecting histone modifications that repress transcription remains an intriguing topic for future work.
The atrrp6l1-3 allele was isolated from the BASTA population from the University of Wisconsin [31]; the atrrp6l1-1, atrrp6l2-2, atrrp6l2-3, atrrp6l2-4 and atrrp6l3-1 alleles correspond to SALK_004432, SALK_113786, SALK_011429, and SALK_149898, and SALK_122492, respectively. iRNAi lines of RRP41, csl4-2, RNA Pol IV (SALK_128428, nrpd1a-3, nrpd1-3), RNA Pol V (SALK_029919, nrpd1b-11, nrpe1-11) mutants were described previously [8], [74], [75]. All Salk alleles are in the Col-0 ecotype and the University of Wisconsin alleles are in the Ws ecotype. The RNAi-mediated knockdown of RRP41 was induced by germinating and growing seedlings on ½× MS plates containing 8 mM 17β-estradiol, following a previously published method [8].
Long day and short day conditions for plant growth were 16 hours light/8 hours dark and 8 hours light/16 hours dark, respectively. Flowering time was measured by counting rosette leaf number at the time of flowering [76].
For chromatin immunoprecipitation (see below), we used the Tandem Affinity Purification (TAP) affinity tag to selectively precipitate RRP6L1 by expressing a RRP6L1- TAP fusion protein. For construction of plant lines with affinity-tagged RRP6L1 for RNA-IP and ChIP, we complemented atrrp6l1-3 with the RRP6L1-TAP transgene. For RRP6L1-TAP complementation, the entire genomic region of RRP6L1 including 1.5 kb upstream from the ATG codon was amplified by PCR using LA taq polymerase (Takara) and cloned into TAP-tag carrying destination vector pDB1008 [8]. For complementation with the TAP tagged transgene, the homozygous atrrp6l1-3 mutant was transformed using Agrobacterium-mediated transformation and the progeny plants containing both the T-DNA insertion allele and the transgene were identified by PCR.
Trizol reagent (Invitrogen) was used to isolate total RNA from seedlings. 10-, 14-, 18-, and 21-day-old seedlings were used for examining the expression of ASL. 11-day-old seedlings were used for investigating expression of genes and antisense RNAs examined in our study. For RT-qPCR, 2–4 µg of total RNA was digested with DNase I (Fermentas) and reverse transcribed for one hour at 42°C (oligo-dT primers) or at 50°C (gene-specific primers), with 100 units of PrimeScript reverse transcriptase (Takara). RT-qPCR (MyiQ-iCycler; Bio-Rad) was used to quantify transcripts using the comparative threshold cycle method (ΔΔCt, Table S1 shows primer sequences), with ACTIN 7 (At5g09810) as an internal reference.
For tiling RT-PCR, sets of serial primers were designed in intervals of 100–200 nt. After obtaining the full-length ASL transcript, another set of overlapping primers was designed to make sure the 3′ and 5′-ends of the RNA have been isolated. The PCR products amplified by tiling RT-PCR were cloned into the pBluescript KS vector and sequenced using T7 and T3 primers.
ChIP was conducted following a previously-described method [77]. Each experiment used 1.5 grams of tissue from 11-day-old seedlings. All ChIP experiments used at least two biological replicates and at least two technical replicates. Anti-H3K4me3 (07-473) and anti-H3K27me3 (ab6002) were purchased from Millipore and Abcam, respectively. IgG Sepharose 6 Fast Flow (GE Healthcare) was used for ChIP using RRP6L1-TAP tagged line. The mock antibody control used an equal amount of chromatin that was not treated with antibody. The ChIPed DNA was purified using PCR purification kit (Fermentas) and qPCR was performed. Supplemental Table S1 lists the primers used for PCR.
MNase-ChIP was performed following a previously-described method [78]. Two grams of tissue from 11-day-old seedlings was fixed using 1% formaldehyde solution for 10 min and washed with distilled water several times. The fixed samples were homogenized with HONDA buffer (20 mM HEPES-KOH pH 7.4, 0.44 M sucrose, 1.25% Ficoll, 2.5% Dextran T40, 10 mM MgCl2, 0.5% Triton X-100, 5 mM DTT, 1 mM PMSF, 1% plant protease inhibitors) and then filtered through miracloth. After isolation of the nucleus-containing fraction by centrifugation, the fraction was treated with MNase (NEB) at 37°C for 10 min. Anti-histone H3 (ab1791) was used for the ChIP. The purification of ChIPed DNA and qPCR was performed as described in ChIP assay.
RNA-IP assays were performed as described previously [31], [79]. Two grams of tissue from seedlings at 11-days-old was fixed with 1% formaldehyde. For purification of RRP6L1-TAP or RRP6L1 RNA complexes, the chromatin was incubated with prewashed IgG Sepharose 6 Fast Flow (GE Healthcare) or with polyclonal anti-AtRRP6L1 antibodies, respectively, at 4°C overnight. H3K27me3-RNA complex purification was performed using anti-H3K27me3 (ab6002) overnight following by incubation with protein A agarose beads. Immunoprecipitated RNA purification used phenol∶chloroform and PrimeScript reverse transcriptase (Takara) and sequence specific primers were used for cDNA synthesis. Supplemental Table S1 lists the primers used for PCR.
Eleven-day-old seedlings were treated with 5 µM α-amanitin (Sigma) for 0, 6 and 9 h or 17 h. After RNA extraction, cDNA was synthesized using ASL-specific and ACTIN 7 primers, followed by qRT-PCR. The level of ASL transcript was normalized relative to the level of ACTIN 7 transcript.
Total RNA extracted from 11-day-old seedlings was treated with TPE (Epicentre) at 42°C, and purified with phenol∶chloroform. Complementary cDNA was synthesized using RNA sequence specific primers followed by RT-qPCR.
|
10.1371/journal.ppat.1003749 | A Viral Genome Landscape of RNA Polyadenylation from KSHV Latent to Lytic Infection | RNA polyadenylation (pA) is one of the major steps in regulation of gene expression at the posttranscriptional level. In this report, a genome landscape of pA sites of viral transcripts in B lymphocytes with Kaposi sarcoma-associated herpesvirus (KSHV) infection was constructed using a modified PA-seq strategy. We identified 67 unique pA sites, of which 55 could be assigned for expression of annotated ∼90 KSHV genes. Among the assigned pA sites, twenty are for expression of individual single genes and the rest for multiple genes (average 2.7 genes per pA site) in cluster-gene loci of the genome. A few novel viral pA sites that could not be assigned to any known KSHV genes are often positioned in the antisense strand to ORF8, ORF21, ORF34, K8 and ORF50, and their associated antisense mRNAs to ORF21, ORF34 and K8 could be verified by 3′RACE. The usage of each mapped pA site correlates to its peak size, the larger (broad and wide) peak size, the more usage and thus, the higher expression of the pA site-associated gene(s). Similar to mammalian transcripts, KSHV RNA polyadenylation employs two major poly(A) signals, AAUAAA and AUUAAA, and is regulated by conservation of cis-elements flanking the mapped pA sites. Moreover, we found two or more alternative pA sites downstream of ORF54, K2 (vIL6), K9 (vIRF1), K10.5 (vIRF3), K11 (vIRF2), K12 (Kaposin A), T1.5, and PAN genes and experimentally validated the alternative polyadenylation for the expression of KSHV ORF54, K11, and T1.5 transcripts. Together, our data provide not only a comprehensive pA site landscape for understanding KSHV genome structure and gene expression, but also the first evidence of alternative polyadenylation as another layer of posttranscriptional regulation in viral gene expression.
| A genome-wide polyadenylation landscape in the expression of human herpesviruses has not been reported. In this study, we provide the first genome landscape of viral RNA polyadenylation sites in B cells from KSHV latent to lytic infection by using a modified PA-seq protocol and selectively validated by 3′ RACE. We found that KSHV genome contains 67 active pA sites for the expression of its ∼90 genes and a few antisense transcripts. Among the mapped pA sites, a large fraction of them are for the expression of cluster genes and the production of bicistronic or polycistronic transcripts from KSHV genome and only one-third are used for the expression of single genes. We found that the size of individual PA peaks is positively correlated with the usage of corresponding pA site, which is determined by the number of reads within the PA peak from latent to lytic KSHV infection, and the strength of cis-elements surrounding KSHV pA site determines the expression level of viral genes. Lastly, we identified and experimentally validated alternative polyadenylation of KSHV ORF54, T1.5, and K11 during viral lytic infection. To our knowledge, this is the first report on alternative polyadenylation events in KSHV infection.
| Kaposi sarcoma-associated herpesvirus (KSHV), also referred as human herpesvirus 8 (HHV-8), is a member of gammaherpervirus subfamily [1]. KSHV infection in healthy individuals is well-controlled by host immune system and other host factors, and hence, it is usually asymptomatic. However prolonged immunosuppression may lead to occurrence of KSHV-induced malignancies. KSHV has been linked to three malignancies including all forms of Kaposi sarcoma, a complex solid tumor of endothelial origin, and two rare B-cell lymphomas, primary effusion lymphoma (PEL or body cavity-based large cell lymphoma) and multicentric Castleman disease [2]–[4]. KSHV, like other herpesviruses, exhibits two distinguishable states of infection, latent and lytic infection. During latency only a small fraction of viral genes are expressed to facilitate the maintenance of viral genome, drive cell proliferation, and mediate immune invasion. Various external and internal stimuli cause a disruption of KSHV latency and induction of KSHV lytic infection with the expression of all viral lytic genes and replication of viral progeny [5]–[7]. KSHV has a large DNA genome (∼168 kb) encoding more than 90 genes for production of viral structural and non-structural proteins, small peptides, long non-coding RNAs (lncRNAs) and small regulatory miRNAs [1], [8]–[10]. Like many DNA viruses, KSHV has a complex gene organization and depends on host cell machinery for its gene expression. However, a full compendium of viral genome annotation is still unknown and the true nature of viral gene expression and its regulation remains to be fully understood.
RNA polyadenylation (pA) of nascent transcripts is a critical posttranscriptional step in maturation of eukaryotic transcripts [11]. A primary role of RNA polyadenylation is to release newly synthesized RNA from DNA template through endonuclease cleavage and protect it from degradation by addition of a poly(A) tail to the RNA 3′ end. The presence of a poly(A) tail also promotes nucleocytoplasmic export and efficient protein translation of mRNAs [12], [13]. RNA polyadenylation is carried out by a large protein complex composed of at least 85 protein factors and binds to specific sequences within nascent transcripts surrounding the cleavage site [14]. An A/U-rich element upstream, recognized by cleavage and polyadenylation specificity factor (CPSF), and an U/GU-rich element downstream, recognized by cleavage stimulatory factor (CstF) [15], [16], of the cleavage site are two major determinants of RNA polyadenylation, although other auxiliary cis-elements may be also involved in pA site definition [17], [18]. After assembly of polyadenylation complex the pre-mRNA is generally cleaved at “CA” dinucleotide followed by addition of a poly(A) tail [19]. While the process of polyadenylation itself is well characterized, the selection of pA site remains a puzzle. Recent genome-wide studies on polyadenylation of host transcripts in various organisms revealed highly promiscuous polyadenylation of large population of RNAs from multiple pA sites [20]–[22]. As a result, genes affected by alternative polyadenylation produce a subset of transcripts with different coding potentials or 3′ untranslated regions (3′ UTRs) [23].
A polyadenylation landscape of human herpesviruses has not been reported at the genome-wide level. In this study, we performed a genome-wide analysis on RNA polyadenylation of KSHV transcripts from B cells with latent or lytic viral infection by using a modified polyadenylation- sequencing (PA-seq) technology [24]. We identified that KSHV utilizes 67 active pA sites for the expression of its latent and lytic genes and a few novel or unannotated genes. We also found alternative RNA polyadenylation of several known KSHV genes and revealed pA site cis-elements in regulation of KSHV gene expression.
To elucidate a role of RNA polyadenylation in KHSV gene regulation, we performed a genome-wide analysis of viral pA sites to monitor their usage during KSHV infection. Three KSHV-positive B-cell lines (JSC-1, BCBL-1 and TREx BCBL-1), which support both latent and lytic virus infection, were chosen in this study. For each cell line, polyadenylation events were compared between latent and lytic infection (Figure S1A). The cells with lytic infection were harvested at 48 h after virus reactivation by chemical induction to allow full viral replication cycle and sufficient expression of viral late transcripts. We observed dramatic reduction of cell viability associated with virus reactivation (31% vs 88% for JSC-1, 50% vs 89% for BCBL-1 and 20% vs 82% for TREx BCBL-1 cells [further referred as TREx]) by trypan blue exclusion analysis. Poly (A)+ RNA fraction from each sample was used for preparing 3′-end cDNA libraries with a modified PA-seq method followed by Illumina paired-end sequencing [24], [25]. In total, we obtained more than 119 million of paired reads from all samples (Figure S1B). KSHV- and human-specific reads were extracted by alignment of obtained sequence reads to the reference KSHV (GenBank acc no U75698.1) and human (UCSC version hg19) genomes. More than 100 million (∼84%) of all reads were uniquely mapped, with about 35 million (∼29%) to KSHV and approximately 65 million (∼55%) to human genome. The remaining 19 million (16%) are unmapped reads. As expected, a remarkable correlation was noticed between KSHV-specific reads and the stat of KSHV infection in all three cell lines, with less KSHV reads (0.10–0.47%) in the cells with viral latent infection and much more KSHV reads (20–77%) in the cells with KSHV lytic infection (Figure S1B and S1C).
For KSHV pA site analysis we focused only on the sequence reads uniquely mapped to KSHV genome and further clustered to identify PA-peaks. We pooled KSHV sequence reads from all samples and performed a peak calling analysis using F-seq algorithm with significant enrichment over a background model [26] and a threshold of >50 read counts per peak (Figure S2). As expected, only a handful of PA-peaks were found in the cells with latent KSHV infection, but significantly more peaks were detected in the cells with lytic KSHV infection (Figure S3A). To further analyze the PA-peak distributions in the context of selected viral genes and to ensure the PA-peaks obtained from our PA-seq representative of authentic pA site regions, we looked into a PA-peak distributed in a well-characterized ORF50 (RTA)/K8/K8.1 locus which encodes three collinear KSHV genes. Although each gene in the locus has its own promoter, their transcripts are all polyadenylated at a single pA site downstream of K8.1 (see diagram in Figure S3B). We found a prominent PA-peak in all lytic samples, but less so in the latent samples, overlapping with the mapped pA site (Figure S3B) reported in previous studies [27], [28]. No other PA-peaks were seen either upstream or downstream of this pA site. These data indicate that PA-seq libraries were in high quality and suitable for comprehensive analysis of viral pA sites.
Subsequently, we determined the nucleotide (nt) position with the highest number of reads within individual peaks in the peak calling analysis as a PA mode (Figure S2) and designated such a PA mode as an unique pA site (Table S1). We also determined a PA peak from the peak start to the peak end, and the total number of the reads within a PA peak was used to approximate the usage of a pA site (Figure S2). With this approach we identified 67 pA sites on both viral DNA strands of the KSHV genome (Figure 1A). The pA sites mapped by PA-seq in this study were remarkably close to several known pA sites previously mapped by traditional methods both in terms of mapped nucleotide position and strand specificity (Table S2).
A higher prevalence of pA sites shows strand bias, with 43 pA sites in the minus strand and 24 in the plus strand of the KSHV genome. The majority of the mapped pA sites are positioned in the intergenic regions of KSHV genome, outside of annotated ORFs, with exception of the pA sites in the coding regions of ORF7 at nt 7032 and ORF61 at nt 98274 and of K12 at nt 118012, 118032 and 118087.
Our genome-wide pA site analysis allowed us to correlate each mapped pA site to annotated KSHV genes and to identify novel KSHV gene(s). We assigned each pA site to a known viral gene or gene cluster region based on the following criteria: (1) both of the gene(s) and the corresponding pA site must be on the same strand of viral genome, (2) the pA site must be positioned outside of the coding region of the viral gene(s), and (3) the gene(s) assigned to a mapped pA site must be positioned upstream of the pA site. These criteria assume that viral transcripts originated from a promoter(s) upstream of the gene will be polyadenylated from the first available pA site downstream. Accordingly, we assigned 55 pA sites to all known KSHV genes (Figure 1B, Table S3). The remaining 12 pA sites unable to assign would indicate the presence of transcripts from unknown KSHV genes for further validation. Interestingly, the majority of unassigned pA sites are positioned antisense to known KSHV genes, suggesting the existence of putative antisense transcripts to these viral genes [29]. Among 55 pA sites assigned to known KSHV gene transcripts, 20 are positioned immediately downstream of a single KSHV gene for polyadenylation of a monocistronic mRNA, and the remaining 35 have multiple upstream KSHV genes ranging from 2 to 5 for polyadenylation of bicistronic or polycistronic transcripts (Figure 1B). Interestingly, we found two or more pA sites mapped to a region downstream of the same gene. These include two alternative pA sites downstream of ORF54, K2 (vIL6), K9 (vIRF1), K10.5 (vIRF3), K11 (vIRF2), and K12 (Kaposin A), three downstream of T1.5 RNA and PAN (nut-1) RNA or in an internal K12 region, and five downstream of vnct internal repeats (Figure 1A). From protein-coding genes, ORF54 showed the highest usage of alternative pA sites (∼24%) followed by K10.5 (∼17%) and K11 (∼11%), but K2, K9, and K12 did so much less frequently (Table S4). Thus, our analyses provide not only the first comprehensive landscape of functional pA sites in the context of KSHV genome, but for the first time the alternative polyadenylation of KSHV transcripts during virus infection.
We next aimed to determine the length and composition of 3′ UTR for each KSHV protein-coding gene. The unassigned pA sites and the pA site for viral non-coding RNA genes were excluded from this analysis. A total of 50 pA sites were used to calculate the 3′UTR length from a pA site to the adjacent termination codon of the closest upstream ORF. We found that the calculated 3′ UTR length of KSHV genes varies greatly in size from 2 nts (ORF38) to 1925 nts (ORF62) (Table S3). The distribution of KSHV 3′UTR is shown in Figure 1C, with a median size of the 3′UTR in ∼80 nts which is significantly shorter than human 3′ UTR with a median size of ∼300 nts [20].
Based on the number of sequence reads obtained at each pA site, one can infer the relative steady-state level (pA site usage) of the pA site-associated transcripts. The limitation of this approach is cluster genes utilizing a single pA site, in which the number of sequence reads reflects a combined level of all gene transcripts. The pA site usage was compared from latent to lytic infection. First, we determined each pA site usage in individual samples to obtain a sample-specific pA site usage and then normalized the number of sequence reads within each viral pA peak to the total sequence reads mapped to KSHV and host genome in each sample (Figure S4, Table S5). Combination of the normalized sequence reads from all latent samples was compared with that from all lytic samples (Figure 2A, 2B, Table S6). The pA site 122069 (+) of latent polycistronic RNA ORF73 (LANA), ORF72 (vCyclin), and K13 (vFLICE) was served as a reference (red bar in Figure 2A–C). Surprisingly in the samples with latent infection, the top 5 sites based on the pA site sequence counts were PAN (nut-1), ORF2/K2 cluster, K12, ORF50/K8/K8.1 cluster, and T1.5 (Figure 2A) which supposed to be KSHV lytic genes, but spontaneously reactivated in a small fraction of cells with latent infection (Figure S3, Table S6). The usage of pA site for the expression of ORF73/ORF72/K13 ranked the 6th during the latency. In the samples with lytic infection, the top 5 pA site usage was the pA sites for abundant expression of PAN, K12, ORF62-58 cluster, T1.5, and K4.2/4.1/4 cluster and usage of the reference latent pA site for expression of ORF73/ORF72/K13 dropped to the 41st. When the changes in utilization of each identified pA site from lytic to latent infection were calculated, however, the usage of mapped pA sites for virus lytic gene expression became remarkable, with more than 500-fold increase from latent to lytic infection for ORF62, ORF24/23, ORF44, PAN, and K12 (Figure 2C, Table S6). The smallest usage change (<10 fold) during virus lytic infection was the pA sites for the expression of ORF2/K2, K10.6/10.5, and ORF73/ORF72/K13, and the transcripts antisense to ORF50 (RTA) and K15/ORF75 (Figure 2C, inset). The smallest change in the pA site usage was the reference pA site of ORF73/ORF72/K13, with only 2.1-fold increase. Thus, these pA sites are truly used to express viral latent genes. Notably, the peak sizes of pA sites vary considerably ranging from 3 (pA site at nt 17227) to 98 nts (pA site at 29740) (Table S7), and represents heterogeneity of the cleavage sites within each mapped pA site [30]. This heterogeneity of a given pA site, such as the pA site at nt 29740, nt 76738 or nt 122069, remained invariable either from JSC-1 to BCBL-1 cells or from latent to lytic infection (data not shown). Based on their bimodal distribution (Figure 3A), we grouped 38 pA sites with a narrow peak (≤30 nts, with a median size of 17 nts), 24 pA sites with a broad peak (>30, ≤45 nts, with a median size of 36.5 nts), and 5 pA sites with a wide peak (>45 nts, with a median size 61 nts) (Figure 3A). Interestingly, we found a strong positive correlation of pA site usage by sequence reads in the order from narrow (4,013 reads), broad (62,764 reads), to wide (425,475 reads) peaks with Spearman correlation coefficient rs = 0.86 (Figure 3B, Table S8). Because each transcript could produce only one read in PA-seq, whereas RNA-seq relies on the read coverage on the entire region of a transcript, the read count in a given PA peak of a pA site simply reflects the abundance of the corresponding transcript.
To investigate the regulatory elements responsible for polyadenylation of KSHV viral transcripts, we analyzed flanking sequences (±50 nts) of all 67 pA sites identified. Prevalence of each nucleotide at individual position was calculated and followed by motif analysis using WebLogo software (Figure 4A). A high prevalence of “A” residues between 10 to 30 nts upstream of the cleavage site was identified, representing the upstream A/U-rich polyadenylation signal. The cleavage site itself was also enriched in A residues, followed by a ∼30 nt long, mostly U-rich element. This distribution of RNA cis-elements around viral pA sites is in agreement with what has been found in human transcripts [16]. To better understand the role of cis-elements in regulation of KSHV polyadenylation, we performed similar analyses separately for three groups of pA sites with a narrow, broad, or wide peak (Figure 4A). The profiles of pA sites with a narrow and broad peak showed the highest similarity to the canonical pA site, with a defined upstream A-rich and a downstream U-rich element. The pA sites with a broad peak also exhibit a U-rich region further upstream. However, there is no significant U-rich element downstream of the pA site with a wide peak, nor other sequence motifs could be seen. These differences in sequence context surrounding the pA sites with different peaks could devote to their notable abundance of the associated transcripts, and was further reiterated by analysis of top 10 pA sites with the highest numbers of sequence reads and bottom 10 pA sites with the lowest number of sequence reads among all 67 pA sites. As shown in Figure 4B, the top 10 pA sites show highly conserved polyadenylation signals upstream and an U-rich region downstream. In contrast, the bottom 10 pA sites only exhibit less conserved polyadenylation signals and lack an U-rich region downstream.
Analysis of upstream poly(A) signal (PAS) strength of KSHV pA sites further reaffirmed this conclusion. The canonical (AAUAAA) and non-canonical (NNAUNA) PAS were identified within 50 nts upstream of the mapped 59 pA sites (Figure 5A, Table S9). Two most common PAS in KSHV as seen in human polyadenylation are canonical AAUAAA (69%) followed by AUUAAA (9%). The usage of other non-canonical PAS for viral RNA polyadenylation ranges from 1% to 3% (Figure 5A, Table S10). Similar to human transcripts [20], about 12% of pA sites mapped in this study have no PAS. Surprisingly, we found that most of the non-canonical PAS were associated with a narrow peak and low level of expression. In contrast, the broad and wide peaks use predominantly canonical AAUAAA and AUUAAA PAS. This became even more obvious with PAS in the top and bottom used 10 pA sites. We found that all top 10 pA sites, but only 60% of the bottom 10 pA sites, contain the canonical AAUAAA (Figure 5B, Table S10).
Given that the pA sites obtained by PA-seq, in general, showed a high correlation with previously mapped KSHV pA sites, we carried out a series of experiments to reconfirm several novel pA sites discovered in this study by 3′ RACE. These include the pA site downstream of ORF27, the pA site mapped within the coding region of ORF61, the alternative pA sites downstream of ORF54 and T1.5, and a cluster of 5 unassigned pA sites downstream of the vnct internal repeats, in addition to the known pA sites and unknown alternative pA sites of K11, K2/vIL6, and K12. Most of the selected pA sites determined by PA-seq were verified by sequencing the expected 3′ RACE products in the predicted size(s) (Figure 6, Table S11). The alternative pA site at nt 25192 within T1.5 lncRNA was not experimentally confirmed because of its <1% usage among T1.5 transcripts and lack of a searchable PAS upstream, nor the alternative pA sites at nt 17227 for K2/vIL6 and at nt 117868 for K12 because of their lower level usage. We were also unable to detect any 3′RACE product in the predicted sizes from five pA sites downstream of vnct internal repeats, despite their moderate usage based on the number of associated read counts. These pA sites identified by PA-seq are in proximity to the short internal 13-bp repeats region “vnct” between nt 29775 and nt 29942 of the KSHV genome. It is worth noting that four of the five pA sites have no detectable PAS upstream and a pA site at the nt 29615 has a non-canonical AAUAUA PAS. A reported pA site at nt 18200 for an RNA antisense to K2 (vIL6) [29], [31] which was not revealed by PA-seq was also not detectable by 3′ RACE in this study (Figure 6).
We further verified the PA-seq-identified pA sites from the mRNAs antisense to ORF21, ORF34, and ORF K8 by 3′ RACE and confirmed the production of the antisense RNAs in B cells during viral lytic infection (Figure 7A). The read abundance of these novel pA sites associated with each antisense RNA was correlated, as predicted, to the amount (measured by band intensity) of the 3′ RACE products derived from its corresponding RNA transcript (Figure 7B).
KSHV T1.5 RNA is a long non-coding RNA, which is transcribed from nt 24243 in the KSHV genome, next to the left lytic origin of replication (oriL)(Figure 8A). The expression of T1.5 RNA is strongly inducible by viral transactivator RTA [32]. While the expression of T1.5 is required for viral DNA replication, its functional characteristics remain unknown [33]. Our study showed that T1.5 RNA is one of the most abundant transcript expressed during KSHV infection. The T1.5 RNA 3′ end was mapped to nt 25440 [34] and we mapped it to nt 25441 by PA-seq and by 3′RACE (Figure 6). In addition, we found that about 10% of T1.5 transcripts were also polyadenylated from two additional pA sites upstream of the nt 22541 pA site (Figure 8A), leading to the production of ∼300 nts shorter transcripts as verified by Northern blot analysis of BCBL-1 total RNA (Figure 8B). Because the antisense probe used in the assay was derived from a upstream region of the mapped pA sites, this probe could detect all transcripts running over this region: a strong band corresponding to the reported size of the inducible T1.5 RNA, a smaller size band (∼1.2 kb) with weaker intensity representing the alternatively polyadenylated T1.5 transcripts, and a much larger T6.1 transcript. The T6.1 RNA does not use T1.5 pA sites [34], but rather a PAN pA site for its expression (Figure 1A).
T1.5 RNA contains a few short ORFs and has potential to encode small peptides [34]. We thus assumed that T1.5 might be exportable to the cytoplasm. As expected, we demonstrated by Northern blot analysis its partial presence in the cytoplasm (Figure 8B). RNA FISH assays further showed T1.5 RNA distribution both in the cytoplasm and nucleus of KSHV infected PEL cells using an antisense RNA probe to the 3′ end of T1.5 (Figure 8A, Figure S5). In these two assays, nuclear PAN RNA served as a control (Figure 8B–C, Figure S5) and displayed, as expected, predominantly in the nucleus overlapping with Hoechst DNA staining [8], [35]. Interestingly, the nuclear coexpression of T1.5 and PAN RNA appears mutually exclusive. We found that the cells expressing high level of nuclear T1.5 RNA display much less nuclear PAN RNA or vice versa (Figure 8C). Compared with the subcellular distribution profile of PAN RNA, we saw more B cells with both cytoplasmic and nuclear distribution of T1.5 RNA during virus lytic infection (Figure 8D).
The usefulness of PA-seq was further extended to examine the expression of a few host genes for its possible application to unveil a pA site landscape of the host genome before and after KSHV lytic infection. Human IL6 (hIL6) and GAPDH were initially chosen because B cells with lytic KSHV infection exhibit increased expression of human IL6 [36], [37], but decreased expression of GAPDH (Figure 8B). As shown in Figure 9, the results from PA-seq on GAPDH and hIL6 were comparable with that from RT-qPCR. The decreased expression of GAPDH RNA could be found by both methods in all three tested B cell lines with KHSV lytic infection and a significant increase of hIL6 expression in TREx BCBL-1 cells with lytic KSHV infection. However, we did not see in either method an increased hIL6 expression in JSC-1 cells with butyrate (a very potent inducer)-induced KSHV and EBV lytic coinfections, but observed the increased hIL6 expression in BCBL-1 cells with valproate (a weak inducer)-induced lytic KSHV infection by RT-qPCR. Human IL6 is a cytokine highly sensitive to (vulnerable for) RNA degradation and PA-seq detects the transcripts carrying an intact 3′-end poly (A) tail, while RT-qPCR detects only a small region of the IL6 RNA. Thus, multiple factors could contribute to the variations in detection of hIL6 gene expression from one cell line to another.
In this report we present the first viral genome landscape of polyadenylation sites from three PEL cell lines with KSHV latent or lytic infection. The comprehensive pA site landscape for the entire KSHV genome was revealed by using a modified PA-seq strategy which conveys single nucleotide resolution and strand specificity [24], [25]. The mapped pA sites have been annotated to all known KSHV genes and four putative novel genes in the KSHV genome. The steady-state expression level of every gene in the KSHV genome from viral latent to lytic infection was quantified by PA-seq reads associated with each mapped pA site and was used to distinguish viral latent genes from lytic genes. By analyzing the flanking sequences of each mapped pA site, we determined the regulatory elements governing viral RNA polyadenylation and gene expression. More importantly, we identified several viral genes utilizing alternative polyadenylation as a mechanism for their expression during KSHV infection. In general, the mapped viral pA sites in this study have high accuracy both in terms of nucleotide position and strand orientation, when compared with the known viral pA sites identified by the conventional methods (Table S3) [8], [34], [38]–[42]. However, we were unable to verify a few pA sites previously reported in other studies, including a pA site at nt 124061 for a C-terminal truncated LANA [43] and a pA site at nt 18200 (+) for the expression of a 0.7 kb transcript antisense to K2 (vIL-6) [29], [31]. The 0.7 kb transcript was discovered using custom-made tiling arrays covering the entire KSHV genome [29], [31] and a T7-Oligo(dT) primer for sample cDNA synthesis. The likelihood internal priming of the oligo primer used in the study might create aberrant synthesis of cDNA probes hybridizing to the tiling arrays. In our study the detection of any pseudo pA sites resulting from internal priming was largely avoided by exclusion of the sequence reads upstream of an A-stretch in the KSHV genome. It is worth noting that the expression of the 0.7 kb transcript antisense to K2 was originally discovered only in viral lytically-infected endothelial iSLK.219 cells derived from Kaposi sarcoma, but not detected in PEL-derived B cells [29]. In addition to assigning the known pA sites and many novel viral pA sites from this study to the KSHV genes being previously annotated, we also identified a few novel viral pA sites (Table S3) that could not be assigned to any known KSHV genes. These unassigned pA sites are often found in the opposite strand to known KSHV genes, including ORF8, ORF21, ORF34, K8 and ORF50. Some of those antisense transcripts were described in other reports [29] and the existence of these RNAs antisense to ORF21, ORF34, and ORF K8 transcripts could be confirmed by 3′ RACE in this study (Figure 7). Their potential roles in KSHV biology are now under active investigation.
KSHV has been evolved to use one pA site for the expression of multiple genes in many regions of the genome. Supporting this notion, our PA-seq analysis identified numerous regions of the KSHV genome with several viral genes (up to 5 genes) sharing a common pA site (Figure 1B). As a consequence, many KSHV genes are expressed as bicistronic or polycistronic transcripts with a long 3′ UTR covering the coding region (s) of downstream gene (s). These RNA structures are vulnerable to viral and cellular miRNAs [44]–[46] and all transcripts from the gene cluster regions could be regulated even by a single miRNA. Others could avoid this regulation by RNA splicing of the downstream ORF(s) as shown in ORF50/K8/K8.1 and K1 transcript [28], [47], [48]. Thus, understanding the gene organization and pA site position is critical for knocking-out or knocking-down studies of various virus genes from the KSHV genome in order to make appropriate interpretation on the function of individual viral genes in a cluster region.
The usage of each mapped pA site in this study was determined by counting the sequence reads associated with each pA site to approximate the steady-state expression level of the associated gene(s). When the sequence-reads of a given pA site in viral lytic infection were compared with that in viral latent infection, we could distinguish pA site usage from viral lytic genes to viral latent genes. The pA site for lytic gene expression could be used 100-fold more in lytic infection than in latent infection, whereas the pA site usage for latent gene expression displays only little increase (less than 10-folds) in lytic infection. Two pA sites downstream of K12, a classical viral latent gene, could be an exception because both showed an increased usage in viral lytic infection. The increased usage of two K12 pA sites is consistent with the finding that a lytic inducible promoter could be activated for K12 expression [49]. Analysis of pA site usage in lytic viral infection also confirmed PAN RNA being an extremely abundant RNA species, with sequence-read counts in the mapped pA site at nt 29740 (+) from viral lytic infection alone representing more than 80% of the total sequence-reads for all pA sites.
Moreover, the efficient expression of viral RNA transcripts was found being related to the peak size of a pA site in this study. It should note that each transcript could give rise to only one read in PA-seq. Thus, the read count in a PA peak simply reflects the abundance of the corresponding transcript. In fact, more sequence reads are not expected to inflate the size of a PA peak, especially when the pA cleavage events are precise. Therefore, the positive correlation we observed between the sizes of PA peaks and the expression levels of corresponding transcripts may suggest some degree of “slippage” in polyadenylation of viral transcripts to ensure high-level expression at the lytic stage. When compared to the pA sites falling into a broad or wide peak, an RNA transcript carrying a pA site with a narrow peak was less expressed, with fewer PA-seq sequence reads. Although this difference in the pA sites with a narrow peak might be attributable partially to their frequent usage of non-canonical PAS, there must be other unknown mechanisms governing the utilization of a pA site with a narrow peak, other than canonical vs non-canonical PAS per se. Previous reports showed that the PAS strength directly affects the overall level of mature transcripts [50], [51] and is determined by conservation of RNA cis-elements UGUAN upstream and an run of U/G downstream of the PAS AAUAAA. For example, the presence of a weaker early SV40 PAS leads to lower expression of a reporter gene than the construct containing a stronger SV40 late PAS when both were driven by the same promoter [52]. Therefore, the PAS strength governing polyadenylation of individual viral transcripts may provide additional level of regulation to fine tune their proper expression during viral infection. In addition, the length of the 3′ UTR could be another factor to affect RNA expression level. A shorter 3′ UTR in KSHV transcripts would provide expression advantage of viral genes in escaping from miRNA-mediated RNA degradation [53], [54].
Recent studies unveiled highly prevalent alternative RNA polyadenylation in various organisms and its profound role in regulation of gene expression [23]. We identified several KSHV genes, including both non-coding and protein-coding genes, exhibit alternative RNA polyadenylation (Table S4). These alternative pA sites were previously ignored because of their relatively lower prevalence and the conceptual bias toward the longest detectable transcripts. All alternative pA sites identified in our study were located in the 3′ UTR of the respective transcripts and thus, their utilization does not affect coding potential of these variant transcripts. Notably, alternative polyadenylation was identified in two most abundant viral lncRNAs PAN and T1.5, each of which harbors three alternative pA sites. We experimentally verified the two alternative pA sites for the expression of corresponding T1.5 transcripts in B cells with viral lytic infection (Figures 6, 8). In addition, alternative polyadenylation of PAN RNA expression had been reported in our earlier study [35]. Therefore, the role of alternative polyadenylation in PAN and T1.5 expression will become an attractive subject for better understanding the function of PAN and T1.5 lncRNAs.
Two unusual clusters of pA sites located downstream of the internal repeat regions were identified by PA-seq, but could not be validated by 3′ RACE in this study. The first cluster is located in the minus strand of the KSHV genome, downstream of “vnct” 13-bp repeats and composed of 5 individual pA sites within a ∼250-bp region from nt 29376 (-) to 29615 (-). The second cluster of three pA sites from 118012 (-) to 118087 (-) is also located in the minus strand downstream of “zppa” repeat region containing two 23-bp repeats (Figure 1A). These pA sites are located within the coding region of K12, but no transcripts associated with these mapped pA sites were detected in previous studies [40]. None of them has a canonical PAS upstream. The sequence reads detected by PA-seq are more likely associated with cryptic transcription from the internal regions [55]. However, these transcripts are unstable and their degradation by cellular exosome is initiated by addition of a short pA tail, which is mediated by a non-canonical pA polymerase and is therefore is not dependent on PAS [56]. These transcripts with the rapid turnover may not be detectable by 3′RACE, but could be picked up by our high sensitive PA-seq.
Primary effusion lymphoma cells lines (JSC-1 [KSHV+, EBV+], BCBL-1 [KSHV+ only] and TREx BCBL-1-vector and –RTA [BCBL-1 derived]) [57], [58] were used in this study The viral lytic replication was induced for 48 h by 3 mM sodium butyrate (Bu) for JSC-1 cells, 0.6 mM sodium valproate (VA) for BCBL-1 cells, or 1 µg/ml doxycycline (DOX) for both TREx BCBL-1-vector and –RTA cells. Total RNA was isolated by TRIzol (Invitrogen) and genomic DNA contamination was removed by RNeasy Mini kit (Qiagen) using on-column DNase I digestion step.
The 3′end library for each sample was constructed using a modified PA-seq strategy [25], [24]. Briefly, 10 µg of DNA-free total RNA from each sample described above was sheared into 200–300 nt fragments by heating (94°C for 3 minutes) with magnesium. After precipitation a reverse transcription was carried out using a modified oligo(dT) primer (5′-bio-T16dUTTTVN-3′, ‘bio’ denotes duo biotin group, ‘dU’ stands for deoxyuridine, ‘V’ represents any nucleotide except T and ‘N’ denotes any nucleotide). After second strand synthesis, resulted dsDNA was pulled down by Dynabeads MyOne C1 (Invitrogen) and dephosphorylated with APex Heat-Labile Alkaline Phosphatase (Epicentre) enabling PCR strand specificity for selective adaptor ligation. Dephosphorylated dsDNA was released from beads by USER enzyme digestion (NEB) and end-repaired, followed by an “A” base addition at the ends. Notably, only the first-stand cDNA contains a 5′ phosphate, and thus can be ligated to bar-coded Illumina paired-end Y linker without a nick. The usage of a dUTP in the oligo(dT) primer and the de-phosphorylation step reinforce strand-specificity, and allow precisely mapping of pA cleavage site at singe-base resolution. Ligation products between 250 bp and 450 bp were gel purified and PA-seq libraries were generated by 16-cycle PCR with Phusion Hot Start High-Fidelity DNA Polymerase (Finnzymes). The obtained libraries were subjected to two technical replicate sequencing by an Illumina HiSeq2000 sequencer.
Obtained raw reads were first aligned to KSHV genome (GenBank acc no U75698.1), EBV B95-8 strain genome (GenBank acc no V01555.2) and human genome (UCSC version hg19) by Burrows-Wheeler Alignment tool (BWA) [59] allowing two mismatches and processed by SAMtools [60]. All uniquely mapped KSHV-specific sequence pairs were used for downstream analyses. First the distribution of obtained reads along KSHV genome was visualized using IGV genome browser (www.broadinstitute.org/igv/) to assure their suitability for pA site analysis. Individual KSHV pA sites were then designated by peak calling using F-Seq program [26] on combined libraries. The PA-seq peaks above the threshold of 50 reads were considered as true peaks. The peaks were further refined by removing pseudo pA sites resulting from “internal priming” due to continuous “A-stretch” in the template. After the peak calling the sequence reads were assigned back to individual samples to obtain the reads-counts for both latent and lytic infection. To obtain a relative expression level the total reads-counts were normalized per million to overall reads mapped to both KSHV and human [61].
The sequence surrounding the mapped pA sites was covered from 50 nts upstream and 50 nts downstream of each identified pA site for the motif analysis. The percentage of occurrence for each nucleotide was calculated, plotted and smoothed with the loess function in R software (R version 2.12.1). Polyadenylation signals (PAS) occurred within 50 nts upstream of pA site were assigned manually. Graphical representation of sequence conservation was generated by Weblogo v3 (http://weblogo.berkeley.edu/) [62], [63].
Transcript 3′ end was identified by SMARTer RACE cDNA Amplification Kit (Clontech). The primer sequences used in 3′RACE are listed in Table S11. The obtained 3′RACE products were sequenced directly or after cloning in pCR2.1-TOPO vector (Invitrogen).
Total RNA was isolated using TRIzol reagent. The cytoplasmic and nuclear fractions of RNA were isolated as described [64]. Obtained RNA (5 µg) was separated on agarose gel and analyzed by Northern blot analysis with 32P labeled oligo probes: oVM 208 (5′-CGTGGCTGTGCTTCTCATCAT-3′) for T1.5 lncRNA, oJM7 (5′-GTTACACAACGCTTTCACCTACA-3′) for PAN lncRNA, oZMZ270 (5′-TGAGTCCTTCCACGATACCAAA-3′) for GAPDH and oST197 (5′-AAAATATGGAACGCTTCACGA-3′) for U6 snRNA.
The single stranded sense and antisense RNA probes were prepared by FISH Tag RNA Multicolour Kit (Invitrogen) by in vitro transcription using DNA fragment of KSHV genome (nt 24906–25375 for T1.5 and nt 29018–29481 for PAN lncRNAs) as templates. The hybridization was performed as previously described [65]. After immobilization the cells were fixed with 2% paraformaldehyde, permeabilized with 0.5% Triton X-100 and blocked with hybridization buffer (50% formamid, 5×SSC, 0.1% Tween-20, 50 µg/ml heparin, 100 µg/ml salmon DNA). The hybridization was carried out overnight at 55°C. The nuclei were counterstained with Hoechst dye. The pictures were collected using a Zeiss LSM510 META laser-scanning microscope (Zeiss).
Total cell RNA isolated by TRIzol (Invitrogen) was treated with Turbo DNA-free DNase to remove DNA. Five micrograms of total cell RNA was used to synthesize cDNA using SuperScript First-Stand Synthesis System (Invitrogen). The GAPDH and human IL6 (hIL6) transcript levels were determined by RT-qPCR using ΔCt method [44], [66], [67].
|
10.1371/journal.ppat.1003879 | Inflammatory Stimuli Reprogram Macrophage Phagocytosis to Macropinocytosis for the Rapid Elimination of Pathogens | Following an infectious challenge, macrophages have to be activated in order to allow efficient clearance of infectious pathogens, but how macrophage activation is coupled to increased clearance remains largely unknown. We here describe that inflammatory stimuli induced the reprogramming of the macrophage endocytic machinery from receptor-mediated phagocytosis to macropinocytosis, allowing the rapid transfer of internalized cargo to lysosomes in a receptor-independent manner. Reprogramming occurred through protein kinase C-mediated phosphorylation of the macrophage protein coronin 1, thereby activating phosphoinositol (PI)-3-kinase activity necessary for macropinocytic uptake. Expression of a phosphomimetic form of coronin 1 was sufficient to induce PI3-kinase activation and macropinocytosis even in the absence of inflammatory stimuli. Together these results suggest a hitherto unknown mechanism to regulate the internalization and degradation of infectious material during inflammation.
| The main cells that are involved in cleaning up microbial pathogens are macrophages. Upon an infection, macrophages are being recruited to the site of infection by a number of different stimuli. In addition, during an infection, macrophages are also activated by cytokines such as interferon-γ and tumor necrosis factor-α that is released from other immune cells. Such macrophage activation is important to achieve a rapid and efficient clearance of microbial pathogens. In this study, we found that macrophage activation induces uptake through macropinocytosis rather than receptor-mediate phagocytosis. As a consequence, microbial material as well as particles can be internalized far more efficiently; In addition, the internalized cargo is rapidly destroyed within lysosomes. We also provide the mechanisms for the switch from phagocytosis to macropinocytosis, which turned out to be the cytokine-induced phosphorylation of the host protein coronin 1. Phosphorylated coronin 1 activated the lipid kinase phosphoinositide 3-kinase, which is known to be responsible for the entry of cargo through macropinocytosis. Together these results provide evidence for a hitherto unrecognized mechanisms to regulate the internalization and degradation of infectious material during an infection.
| Macrophages are the main scavengers responsible for clearance of solutes and particulate material as well as to act as defense cells against invading microbes [1]. The main mechanisms via which macrophages can internalize and clear microbial material occurs through receptor-mediated phagocytosis. This process, making use of different cell surface receptors, including Fc receptors, complement receptors, scavenging receptors as well as several lectin receptors, ensure the uptake of particulate material into phagosomes followed by delivery of the cargo to lysosomes [1].
Under certain conditions phagocytosis is not sufficient for an effective elimination of microbial pathogens. For example, pathogenic mycobacteria, which include the causative agent of tuberculosis, can be internalized via phagocytosis using different receptors, including complement receptor, scavenging receptors as well as lectin receptors such as the mannose receptor and DC-SIGN [2] [1]. Once internalized into phagosomes, pathogenic mycobacteria have evolved to withstand lysosomal degradation by effectively blocking phagosome-lysosome fusion thereby surviving within macrophage phagosomes instead of being degraded in lysosomes prior to cytosolic escape [3]–[10]. Also, during an acute infection, the phagocytic capacity of macrophages may become limiting in being able to internalize and destroy sufficient numbers of bacilli in order to curb the infection [11]. Furthermore, the particular receptor involved may modulate the macrophage killing capacity by silencing certain macrophage responses such as the respiratory burst [12].
As an alternative to receptor-mediated phagocytosis, macrophages can also internalize material via macropinocytosis, a non-saturable mode of uptake that allows the internalization of large amounts of cargo independent of any receptor usage [11], [13]–[15]. In several cell types, macropinocytosis can be transiently induced by growth factors as well as certain pathogens such as Salmonella, Shigella or viruses [16], [17]. In macrophages, as well as dendritic cells, where macropinocytosis also occurs constitutively, macropinocytosis allows to efficiently process infectious material as well as activate immune responses [14].
Here, we show that inflammatory stimuli reprogram the macrophage endocytic pathway from receptor-mediated phagocytosis to macropinocytosis, enabling macrophages to internalize large amounts of cargo for direct transfer to lysosomes. We found that upon macrophage activation, serine phosphorylation of the macrophage protein coronin 1 is the key molecular switch that reprograms the macrophage from a phagocytic uptake mode to macropinocytosis. Coronin 1 (also known as P57 or TACO, for Tryptophan aspartate containing Coat protein), was originally identified as a survival factor for intracellular residing mycobacteria by blocking the delivery of pathogenic mycobacteria to lysosomes via the activation of the Ca2+/calcineurin pathway [18]–[23]. In resting, non-activated macrophages, coronin 1 is associated with the cell cortex via an interaction (either direct or indirect) with plasma membrane cholesterol [24]. We found that upon cytokine-mediated macrophage activation, coronin 1 was phosphorylated on multiple serine residues by protein kinase C, which induced the relocation of coronin 1 from the cortex to cytoplasmic puncta. Serine phosphorylation of coronin 1 was sufficient to induce phosphoinositol-3-kinase activity thereby switching the internalization mode from receptor-mediated phagocytosis to macropinocytosis. Together these results not only provide a molecular explanation for the mycobactericidal effect of macrophage activating cytokines, but furthermore suggest that macrophage activation reprograms the endocytic machinery through coronin 1 phosphorylation in order to efficiently eliminate infectious cargo.
Macrophage activation by either interferon-γ (IFN-γ) or tumor necrosis factor-α results in the rapid delivery of the internalized mycobacteria to lysosomes followed by mycobacterial killing (see Fig. 1A, Fig. S1A,B and [25]). Interestingly, close inspection of the mycobacterial internalization process in activated macrophages by light microscopy showed that mycobacteria entered macrophages in large spacious vacuoles (Fig. 1A, arrows), which is an indication that the bacilli entered cells via macropinocytosis, rather than phagocytosis [14], [26]–[29]. Upon addition of the phorbol ester phorbol 12-myristate 13-acetate PMA), that is known to induce macropinocytosis [30], [31], similar spacious vacuoles were observed (Fig. 1A).
To further analyze entry of mycobacteria in activated macrophages, resting, i.e., non-activated, or activated macrophages were infected with mycobacteria, fixed, and analyzed by immunofluorescence microscopy. Consistent with macropinocytic uptake, both early macropinosomal markers sorting nexin 1 and sorting nexin 5 [13] strongly colocalized with mycobacterial vacuoles in activated, but not in resting macrophages (Fig. 1B and Fig. S1C–E). Furthermore, internalization of GFP-expressing mycobacteria in IFN-γ-activated macrophages was prevented by the macropinocytic inhibitors amiloride and 3-methyladenine [32], [33] only in activated macrophages (Fig. S1F) as judged by the presence of GFP immunoreactivity in macrophage lysates following mycobacterial uptake. In contrast, the actin depolymerizing agent cytochalasin D prevented bacterial entry in both resting as well as activated macrophages (Fig. S1F). An inhibitor of clathrin-mediated endocytosis, monodansyl cadaverine (MDC; [34]) on the other hand could not inhibit internalization of mycobacteria in resting or activated macrophages (Fig. S1F).
Macropinocytic entry is associated with the formation of large membrane ruffles and blebs at the site of uptake [14], [26], [35]. To qualitatively assess phagocytosis versus macropinocytosis, scanning electron microscopy was used. As shown in Fig. 1C, when macrophages that had either been untreated or activated by interferon-γ were allowed to internalize mycobacteria, this resulted in the presence of extensive membrane ruffles in activated, but not resting macrophages, characteristic of macropinocytic uptake (Fig. 1C). Similarly, macrophages infected in the presence of the macropinocytosis inhibitors blebbistatin [36] or 5-(N-Ethyl-N-isopropyl) amiloride (EIPA) exhibited no membrane ruffles, and macropinocytic entry was not observed. Treatment of cells with the cytoskeletal inhibitor cytochalasin D prevented mycobacterial entry in both resting and activated cells (Fig. 1C).
To further analyze entry through phagocytosis versus macropinocytosis, we established a fluorescence activated cell sorting (FACS) assay, that allowed to distinguish phagocytosis from macropinocytosis in a quantitative manner. To that end, macrophages were incubated either with IgG-coated AlexaFluor568-conjugated beads (to asses phagocytosis) or rhodamine-conjugated dextran 70000 (to asses macropinocytosis) and analyzed by flow cytometry. Internalization of IgG-coated beads resulted in the appearance of several peaks, with the highest peak corresponding to cells having internalized a single bead, with a gradual decrease in the number of cells internalizing multiple beads which is consistent with phagocytic uptake (Fig. S1G, left panel). In contrast, internalization of rhodamine-coupled dextran resulted in a broad peak of high fluorescence, consistent with macropinocytic uptake (Fig. S1H, left panel). While cytochalasin D blocked bead internalization as well as dextran entry (Fig. S1G,H, right panels), the macropinocytosis inhibitor blebbistatin specifically prevented dextran uptake, while not affecting the internalization of IgG-coated beads (Fig. S1G,H middle panels).
Using this assay, we analyzed the uptake of mycobacteria in both resting as well as activated macrophages. As shown in Fig. 1D, incubation of non activated macrophages with fluorescently labeled mycobacteria resulted in a FACS profile reminiscent of phagocytosis, while the incubation of interferon-γ-activated macrophages with mycobacteria revealed a broad peak of fluorescence, consistent with macropinocytic uptake. Moreover, while the actin poisoning agent cytochalasin D blocked uptake of mycobacteria in both resting and activated macrophages, the macropinocytosis inhibitors blebbistatin and EIPA selectively blocked uptake in activated, but not resting macrophages (Fig. 1D,E). Together these results suggest that upon macrophage activation with interferon-γ mycobacteria are engulfed by macropinocytosis instead of phagocytosis.
To analyze whether the switch from phagocytosis to macropinocytosis upon macrophage activation is specific for mycobacterial uptake or represents a general mechanism, we analyzed entry of both E. coli as well as M. marinum by scanning electron microscopy as well as the aforementioned FACS-based assay. To that end, bacilli were labeled with the fluorescent dye PKH-26. The analysis of entry of M. marinum as well as E. coli as shown in Fig. S2 shows that the bacteria were internalized into activated macrophages through macropinocytosis as judged by the ability of both blebbistatin as well as EIPA to prevent bacterial entry into activated, but not resting macrophages (Fig. S2).
To further asses the capacity of activated macrophages to internalize material via macropinocytosis, we analyzed uptake of fluorescently-labeled beads coated with either complement 3 (Fig. 2 A,B), IgG (Fig. 2 C,D) or mannan (Fig. 2 E,F), that in resting macrophages is internalized through complement type 3, Fc gamma or mannose receptors, respectively. While all cargo entered resting, non-activated macrophages through phagocytosis, upon macrophage activation by interferon-γ, entry occurred through macropinocytosis as judged by the FACS profiles (Fig. 2A,C,E). Furthermore, while in all cases incubation with cytochalasin D prevented entry in both resting and activated macrophages, the macropinocytosis inhibitors blebbistatin and EIPA, did not affect entry into resting macrophages, but blocked the internalization process in activated macrophages. These results therefore strongly suggest that macrophage activation causes a general reprogramming of the entry machinery from phagocytosis to macropinocytosis.
In the course of analyzing IFN-γ-mediated macropinocytic uptake, we noticed the relocation of the macrophage protein coronin 1 [19], [21], [23] from the cell cortex to cytoplasmic puncta (Fig. 3A,B and the Movies S1 and S2). Similarly, both tumor necrosis factor (TNF) α, as well as phorbol 12-myristate 13-acetate, (PMA) a direct activator of macropinocytosis [30], [31] caused coronin 1 delocalization from the cell cortex to the cytoplasm (Fig. 3C and Fig. S3A as well as Movie S3). The coronin 1-containing cytoplasmic puncta were positive for the cholesterol labeling dye filipin as well as for the endocytic vesicle maker FM4-64 (Fig S3B,C). Since PMA is a direct activator of protein kinase C (PKC) [37], we analyzed whether coronin 1 delocalization resulted from cytokine-mediated PKC activation [38]. Indeed, preincubation of macrophages with the PKC inhibitor chelerythrine prior to activation prevented coronin 1 relocalization (Fig. 3D). Furthermore, IFN-γ and tumor necrosis factor-α as well as PMA stimulation resulted in the activation of PKC (Fig. 3E). PKC activation was a direct result of IFN-γ triggering, since stimulation of macrophages isolated from IFN-γ receptor–deficient mice failed to result in PKC activation (Fig. 3F). Furthermore, when coronin 1-deficient macrophages were stimulated with IFN-γ, PKC was readily activated (Fig. 3EF), indicating that PKC activation precedes coronin 1 delocalization.
To understand the contribution of PKC activation to the lysosomal delivery of internalized cargo, macrophages were either left untreated or stimulated for different time periods in the absence or presence of chelerythrine followed by infection with mycobacteria. Activation of macrophages by IFN-γ was sufficient to result in lysosomal transfer of mycobacteria, whereas the presence of the PKC inhibitor chelerytrine prevented mycobacterial delivery to lysosomes (Fig. S4A). Also, direct induction of PKC activity by PMA readily resulted in lysosomal delivery, which was prevented by the inclusion of chelerythrine (Fig. S4B). Thus, macrophage activation by IFN-γ activates PKC which in turn causes the redistribution of cortical coronin 1. As a result, mycobacterial cargo, that is normally retained in non-lysosomal phagosomes is efficiently delivered to lysosomes via macropinocytosis.
We next analyzed whether direct phosphorylation of coronin 1 was responsible for IFN-γ-mediated coronin 1 relocation from the cell cortex to cytoplasmic puncta. The primary sequence of coronin 1 contains several potential PKC consensus sites [19], [39], and analysis of coronin 1 purified from resting and activated macrophages revealed serine phosphorylation but not tyrosine phosphorylation of coronin 1 upon activation while threonines were phosphorylated in both resting and activated cells (Fig. S4C). Subsequent two-dimensional IEF-PAGE revealed the appearance of 4 additional spots in activated, but not resting macrophages, suggesting that interferon-γ activation induces phosphorylation of 4 serines on coronin 1 (Fig. S4D). As expected, the protein kinase C inhibitor chelerythrine blocked activation-induced serine phosphorylation on coronin 1 (Fig. 3G and Fig. S4C). In activated coronin 1-deficient as well as IFN-γ receptor deficient macrophages, none to background levels of coronin 1 or serine phosphorylation was observed (Fig. S4D).
Bioinformatic analysis suggested a high probability for serines 9, 311, 356 and 412 for phosphorylation by PKC (See Material and Methods). To directly analyze the involvement of these serines in coronin 1 relocation, these four residues were mutated to alanine or phosphomimetic glutamate and expressed in coronin 1-deficient macrophages as C-terminal EGFP fusion proteins (Fig. 3H,I). While wild type coronin 1-EGFP localized to the cell cortex in non-activated macrophages, as observed for non-tagged coronin 1 [19], [21], [40] macrophage activation resulted in the relocation of coronin 1-EGFP to cytoplasmic puncta (Fig. 3I). Furthermore, while upon macrophage activation the serine to alanine mutant of coronin 1-EGFP, coronin 1S-A-EGFP, failed to relocate to cytoplasmic puncta, the phosphomimetic glutamate mutant of coronin 1-EGFP (coronin 1S-E-EGFP) did not localize at the cell cortex but instead localized within cytoplasmic puncta (Fig. 3I). Similarly, upon subcellular fractionation wild type coronin 1-EGFP was predominantly recovered in the pellet fraction, suggesting membrane association, and relocalized to the cytoplasmic fraction (‘supernatant’) upon macrophage activation (see Fig. S4E,F). However, the alanine mutant of coronin 1 remained associated with the pellet fraction in both resting and activated cells, while in contrast, the phosphomimetic mutant was localized to the soluble fraction even in non-activated cells (Fig. S4E,F). Together these data indicate that cytokine-induced macrophage activation results in PKC-mediated coronin 1 phosphorylation on serine 9, 311, 356 and 412 that is a pre-requisite for coronin 1 relocation from the plasma membrane to cytoplasmic puncta.
The above results suggest that macrophage activation resulted in PKC-mediated coronin 1 phosphorylation concomitant with the induction of macropinocytosis. However, whether or not coronin 1 is directly involved in the induction of macropinocytosis remained unclear. We therefore initiated a series of experiments to analyze whether or not coronin 1 phosphorylation was sufficient for the induction of macropinocytosis, even in the absence of macrophage activation.
First, to analyze the consequences of coronin 1 phosphorylation for mycobacterial internalization within macrophages, coronin 1-deficient macrophages expressing the EGFP tagged wild type (Cor1-EGFP), the serine-to-alanine (coronin 1S-A-EGFP) mutant or the phosphomimetic mutant (coronin 1S-E-EGFP, see Fig. S5A) were infected with mycobacteria for 1 hour, followed by a 3-hour chase and analyzed by confocal microscopy analysis as well as for mycobacterial survival. While in coronin 1-deficient macrophages, as expected [21] [41], mycobacteria were rapidly transferred to lysosomes and killed even without activation (Fig. 4D,E,F), upon expression of wild type coronin 1-EGFP macrophage activation was required to induce lysosomal transfer and killing of internalized mycobacteria (Fig. 4A and panels D–F). However, when IFN-γ-activated coronin 1-deficient macrophages expressing the alanine mutant (coronin 1S-A-EGFP) were infected with mycobacteria, lysosomal transfer did not occur (Fig. 4B,D) and the mycobacteria proliferated within macrophages (Fig. 4E,F). This suggest that serine phosphorylation of coronin 1 is essential to relocate mycobacteria to lysosomes upon activation. Conversely, expression of the phosphomimetic coronin 1 mutant (coronin 1S-E-EGFP) resulted in all mycobacteria being transferred to lysosomes followed by their elimination, regardless of the macrophage activation state (Fig. 4C,D and panels E,F). These data show that even in the absence of inflammatory stimuli, coronin 1 phosphorylation is sufficient to redirect phagocytic cargo via macropinocytosis to lysosomes.
To analyze whether coronin 1 phosphorylation on serines is a general switch from phagocytosis to macropinocytosis, wild type or coronin 1-deficient macrophages expressing either wild type coronin 1, the serine – alanine coronin 1 mutant or the phosphomimetic coronin 1 mutant as EGFP fusion proteins were incubated with IgG-coated fluorescent beads. As shown in Fig. S5,BC, constitutive macropinocytosis in either resting or activated macrophages was unaltered by transfection of coronin 1 mutants. However, while in wild type cells, as well as in coronin 1-deficient macrophages transfected with wild type coronin 1, macropinocytic uptake of IgG-coated beads was only seen following macrophage activation with interferon-γ (Fig. 5A–C), coronin 1-deficient macrophages alone did not show macropinocytic uptake of beads even upon activation. Notably, in cells expressing the serine-to-alanine (coronin 1S-A-EGFP) mutant macropinocytosis did not occur, even upon macrophage activation with interferon-γ (Figure 5D). Moreover, when macrophages expressed the phosphomimetic glutamic acid mutant of coronin 1 (coronin 1S-E-EGFP), the IgG-coated beads were internalized through macropinocytosis even in the absence of interferon-γ (Fig. 5E). These results suggest that serine phosphorylation of coronin 1 is the crucial switch from a phagocytic to a macropinocytic uptake mode upon macrophage activation.
Macropinocytosis depends on the expression and recruitment of the phosphoinositol-binding protein sorting nexin 5 (SNX5) [42], [43]. Immunoprecipitation of the different sorting nexins from activated and infected macrophages followed by immunoblotting for coronin 1 revealed the specific association of coronin 1 with sorting nexin 5, but not with other sorting nexins (Fig. S6A,B). Coronin 1 was not associated with sorting nexin 5 in resting and infected macrophages (Fig. 6A, left panels and Fig. S6C) and the association between coronin 1 and sorting nexin 5 decreased with increased chase times following mycobacterial infection as analyzed by co-immunoprecipitation and immunofluorescence analysis (Fig. 6A, right panels and Fig. S6D, right panels), suggesting that the association was transient. Furthermore, inclusion of the macropinocytosis inhibitors amiloride or 3-MA as well as the protein kinase C inhibitor chelerythrine prevented association between coronin 1 and sorting nexin 5 (Fig. 6B); Finally, consistent with the importance of serine phosphorylation of coronin 1 for the induction of macropinocytosis upon activation, sorting nexin 5 was not associated nor colocalized with coronin 1S-A-EGFP, whereas sorting nexin 5 was associated and colocalized with the phosphomimetic form of coronin 1 (coronin 1S-E-EGFP) even in the absence of IFN-γ-mediated activation (Fig. 6C and Fig. S6E,F). These results suggest that during infection, following delocalization from the cell cortex into cytoplasmic puncta upon macrophage activation, serine-phosphorylated coronin 1 is relocalized to nascent macropinosomes in a complex with sorting nexin 5.
Macropinocytosis is crucially dependent on the activation of the lipid kinase phosphoinositide 3-kinase (PI-3 kinase [17], [33]). Given the relocation of serine-phosphorylated coronin 1 to macropinosomes in association with sorting nexin 5 and the possible association of coronin 1 with PI-3 kinase [44], we asked whether coronin 1 was involved in phosphoinositide 3-kinase activation by monitoring the phosphorylation of Akt/protein kinase B on Ser-473 [45]. Although PI-3 kinase is activated both following phagocytosis as well as macropinocytosis [33], the ruffle formation involved in macropinocytosis is associated with immediate phosphatidylinositol (3,4,5)-trisphosphate (PIP3) generation through rapid PI-3 kinase activity upon addition of cargo, which precedes requirement for PI-3 kinase for cup-closure in both phagocytosis and macropinocytosis [17], [46]. To monitor rapid PI-3 kinase activation, phosphorylation of the downstream substrate AKT on serine 473 was analyzed by immunoblotting of lysates from either resting or activated macrophages to which bacilli or IgG-coated beads had been added. As shown in Figure 7A, incubation of interferon-γ-activated, but not resting macrophages with mycobacteria resulted in substantial AKT phosphorylation at early time points, similar to the activation of AKT by PMA (Fig. 7A). Strikingly, in coronin 1-deficient macrophages, no AKT phosphorylation was detected upon incubation of activated macrophages with mycobacteria, despite similar phosphorylation when incubated with PMA (Fig. 7B). As expected, incubation for longer time points (∼90 mins) resulted in AKT phosphorylation both in resting and IFN-γ-activated macrophages (data not shown). Similar to the coronin 1-dependent PI-3 kinase activation by mycobacteria in interferon-γ-activated macrophages, incubation of IgG-coated beads induced AKT phosphorylation only in wild type, but not coronin 1-deficient macrophages (Fig. 7C,D). These data suggest that coronin 1 is essential for the activation of PI-3 kinase during macropinocytic uptake.
To analyze the importance of serine phosphorylation on coronin 1 for the activation of PI-3 kinase, J774 macrophages or J774 cells in which coronin 1 expression was knocked down by RNAi ([20], see also Fig. S7) were transfected with either RNAi-resistant wild type, serine-to-alanine or the phosphomimetic form of coronin 1 fused to EGFP, and the resulting cells were incubated with mycobacteria or IgG coated beads in resting and IFN-γ-activated macrophages (Fig. 8 and S7). Expression of wild type coronin 1 in these knock-down cells restored AKT phosphorylation following incubation with either mycobacteria or IgG-coated beads (Fig. 8A,D). Importantly, while expression of the serine-alanine mutant failed to show AKT phosphorylation in either resting or activated cells (Fig. 8B,E), in cells expressing the phosphomimetic coronin 1 mutant, AKT was phosphorylated rapidly even in the absence of interferon-γ (Fig. 8C,F).
Together these results suggest an essential role for serine-phosphorylated coronin 1 in the activation of PI-3 kinase to induce macropinocytosis upon macrophage activation.
During inflammation, immune defense mechanisms must be upregulated to ensure a coordinated response towards the invaded infectious microbes. We here demonstrate that inflammatory stimuli reprogram the macrophage endocytic pathway from phagocytosis to macropinocytosis in a coronin 1-dependent manner. Reprogramming receptor-mediated phagocytosis to macropinocytosis allows macrophages to internalize cargo by bulk flow, rather then being restricted by specific receptor interactions; furthermore, internalizing material through macropinocytosis allows macrophages to efficiently target all incoming microbes to lysosomes for degradation. This may be especially important in the case of pathogens that can survive within non-activated macrophages by resisting phagosome-lysosome fusion, such as Mycobacterium spp. Also, although some bacteria can induce macropinocytic entry into non-phagocytes [17], most bacteria enter macrophages via phagocytosis and do not co-opt the macropinocytic pathway. Therefore, the ability of macrophages to switch from phagocytosis to macropinocytosis endows these cells with a mechanism to rapidly eliminate infectious material in lysosomes during an inflammatory challenge.
The molecular switch inducing macropinocytosis of cargo upon macrophage activation was revealed to be serine phosphorylation of coronin 1, that was found to directly activate of the lipid kinase phosphatidylinositol 3-kinase, which is required for macropinosome formation, (see also Figure 9). Phosphorylation on coronin 1 occurred on residues 9, 311, 356 and 412 and induced coronin 1 relocation from its location at the cell cortex to cytoplasmic puncta. Upon addition of cargo, coronin 1 assembled with sorting nexin 5 and relocalized to the cell cortex to activate phosphatidylinositol 3-kinase and macropinocytosis.
Importantly, the here demonstrated switch in phagocytic to macropinocytic uptake upon macrophage activation is based on several independent lines of evidence; first, morphological analysis by light and electron microscopy revealed that upon macrophage activation cargo was internalized into spacious vacuoles involving large membrane ruffles, which are hallmarks of macropinocytic as opposed to pseudopod-mediated phagocytic uptake [14], [26]–[29]. Second, upon macrophage activation, internalized cargo strongly colocalized with the early macropinocytic markers SNX1 and SNX 5. Third, we made use of three inhibitors that block macropinocytosis via distinct modes of action; while amiloride blocks macropinocytosis via inhibition of the Na+/H+ exchanger [32], 3-methyl adenosine inhibits macropinocytosis via blocking a specific class of PI-3-kinase and blebbistatin blocks macropinocytosis via the inhibition of nonmuscle myosin II [36], [47] all of which have a crucial role in macropinocytosis. Fourth, analysis by flow cytometry showed that upon macrophage activation, internalization of IgG-coated beads occurred in bulk as opposed to single uptake events in non-activated macrophages.
Induction of macropinocytosis through cytokine-mediated macrophage activation is an exquisite strategy from the host immune defense point-of-view to efficiently eradicate pathogenic material; First, this pathway provides a way for the immune system to clear large amounts of extracellular material [11], [48]; second, since some receptor-mediated phagocytic entry pathways result in silencing of inflammatory responses [12], avoiding phagocytosis altogether and instead taking up cargo via macropinocytosis ensures a complete microbial clearance. Third, there is little to none communication between macropinosomes and conventional endosomes [49], and it is conceivable that this strict separation between macropinosomes and the phagocytic/endocytic pathway may help to ensure a rapid and efficient clearance of the pathogens through lysosomal degradation in activated macrophages, thereby preventing extensive exchange of microbes to more hospitable subcellular organelles such as used by several intracellular pathogens including Listeria spp., Brucella spp. and Mycobacterium spp.
An intriguing finding in this study is that the macrophage protein that functions as a switch from phagocytosis to macropinocytosis is coronin 1. Coronin 1 was identified in a search for molecules that allow intracellular survival of mycobacteria that are being internalized through phagocytosis via one of the macrophage phagocytic receptors [1], [18], [19], and subsequent work revealed a role for coronin 1 in promoting Ca2+/calcineurin signaling upon mycobacterial infection, but any other role for coronin 1 in macrophages has not been defined [21], [50], [51]. Although other members of the coronin protein family are expressed in macrophages [20], the here described role for coronin 1 in acting as a molecular switch to induce macropinocytosis is clearly a non-redundant function, since in coronin 1-deficient macrophages macropinocytosis was not induced. Why coronin 1 is unable to prevent the delivery of mycobacteria from macropinosomes to lysosomes may lie within the phosphorylation-induced monomerization of coronin 1, which is not anymore capable of protecting the macropinosome residing mycobacteria from lysosomal destruction (SBDG and JP, unpublished).
A function for coronin 1 in the activation of macropinocytosis is furthermore consistent with the association of coronin 1 with cholesterol [24], [52]. Cholesterol is an essential component of the macropinocytic pathway in the absence of which macropinocytosis cannot occur [48], [53]. Interestingly, the cholesterol-staining agent filipin colocalized with coronin 1 within the cytoplasmic puncta; whether cholesterol is required for an efficient phosphorylation of coronin 1, or rather needed for coronin 1-mediated activation of PI-3 kinase during macropinocytic uptake remains to be established.
Whereas the process of phagocytosis is well characterized, the mechanisms involved in macropinocytic entry are only beginning to become elucidated [13], [15], [17], and which proteins exactly are involved in macropinosome formation is still largely unclear [13]. Importantly, constitutive macropinocytosis is unaltered regardless of the activation state of the macrophages and proceeds even in the absence of coronin 1 [21]. Although recent work shows the importance of sorting nexin 5 in macropinocytosis, the precise spatio-temporal pattern of signaling events leading to the induction of macropinocytosis as well as the precise role for sorting nexin 5 in macropinosome formation remains unclear [17], [43]. SNX5 is recruited to the plasma membrane via its phosphoinositide-(PX) binding domain that binds to PI(3,4)P2 [54]. Therefore, the here reported finding that PI-3 kinase activation was dependent on IFN-γ-mediated coronin 1 phosphorylation is consistent with a model in which serine phosphorylated coronin 1 associates with sorting nexin 5 and is targeted to PI(3,4)P2-containing plasma membrane microdomains, after which PI-3 kinase activity is induced to initiate macropinosome formation [55], see also Fig. 9.
Importantly, the coronin 1-mediated reorganization of the endocytic pathway occurred independently from IFN-γ-mediated activation of gene expression [56], since even in the absence of macrophage activation, expression of a serine phosphomimetic coronin 1 mutant was sufficient to induce macropinocytosis. Thus, inflammatory stimuli, besides inducing the expression of a cohort of genes that are directly involved in microbial killing [56]–[61], can also modulate entry pathways in order to efficiently transfer infectious cargo to lysosomal organelles. Many of the IFN-γ-induced genes contribute to an effective immune response, including up regulation of genes important for the induction of autophagy [56]–[62]. Notably, the late macropinosome/autophagosomal marker LC3b was recruited at late times following cargo internalization, at a time point when early macropinocytic markers were not anymore associated with bacteria-containing vacuoles (data not shown), clearly indicating that coronin 1-mediated macropinocytosis precedes autophagy upon macrophage activation [56], [58], [62].
Coronin 1 is emerging as a leukocyte-specific regulator of intracellular signaling processes, and has been shown to promote both the viability of intracellular mycobacteria as well as T lymphocytes via the activation of Ca2+-dependent signaling [23]. Interestingly, a rise in intracellular calcium was shown to be required for macropinocytosis to proceed in dendritic cells [63], [64], and it is possible that relocalized coronin 1 is responsible for the Ca2+ rise upon induction of macropinocytosis.
In conclusion, the work described here defines coronin 1 phosphorylation as a master switch inducing macropinocytic uptake of cargo upon cytokine activation, thereby coordinating induction of an entry pathway that allows for the macropinocytic engulfment of large amounts of cargo with an up regulation of genes involved in the antibacterial response. It will be interesting to establish whether or not the manipulation of this pathway may be useful in the development of therapies to induce cargo transfer to lysosomes, including the shuttling of pathogenic mycobacteria to lysosomes for rapid elimination.
Macrophages were derived from the bone marrow of wild type, coronin 1-deficient, or IFN-γ receptor deficient mice as described [21] unless stated otherwise. All animal experimentation was approved by the veterinary office of the Canton of Basel-Stadt (approved license number 1893) and performed according to local guidelines (Tierschutz-Verordnung, Basel-Stadt) and the Swiss animal protection law (Tierschutz-Gesetz). Macrophages were immortalized with the J2 virus obtained from culture supernatants of NIH-J2-leuk cell line (kind gift from Prof. U. Rapp; [65]), and confirmed to be of the macrophage lineage by staining with F4/80 and CD11b. When stated, J774 wild type or coronin 1 knock down cells as described before [20] were used. All macrophages were grown in DMEM (Sigma; 4.5 g/l glucose), supplemented with 10% heat inactivated FBS (PAA; low endotoxin) and 2 mM L-glutamine (Sigma). E. coli (DH5α) was grown in LB. For mycobacterial infections M. bovis BCG (Pasteur strain), M. bovis BCG-GFP (Montreal strain) [19], [66] or M. marinum (strain ZF214Cs, a kind gift from Wilbert Bitter, Amsterdam, the Netherlands) was used which were cultered in 7H9 including 10% OADC enrichment and including kanamycin 50 µg/ml in case of M. bovis BCG-GFP.
Interferon-γ and tumor necrosis factor-α were from R & D, stocks of 100 µg/ml were prepared in sterile PBS and for activation 1000 U/ml was used. Phorbol myristate acetate (PMA) was from Sigma, stocks of 1 mM were prepared in DMSO and 100 nM was used for activation. Monodansyl cadaverine (MDC) was from Sigma and 100 mM stocks were prepared in DMSO and used at 200 µM, Cytochalasin D (Sigma) stocks were 5 mg/ml in DMSO and 10 µg/ml was used as final concentration. 5-(N-Ethyl-N-isopropyl)amiloride (EIPA; Sigma) stocks were 100 mM in water and 50 µM was used as final concentration. 3-Methyladenine (Sigma) stocks were 1 M in water and 50 mM was used as final concentration. Blebbistatin (Sigma) stocks were 10 mM in DMSO and 150 µM was used as final concentration. Chelerythrine (Sigma) stocks were 10 mM and 10 µM was used as final concentration, while Herbimycin A (Calbiochem) stocks were 100 ug/ml prepared in DMSO and 100 ng/ml was used as final concentration. Amikacin (Sigma) stocks were 100 mg/ml prepared in water and used at 100 µg/ml final concentration. Filipin (Fluka) stocks were 5 mg/ml prepared in methanol and 50 µg/ml was used as final concentration, FM4-64 (Molecular Probes) stocks of 2 mg/ml were prepared in DMSO and 5 µg/ml was used as final concentration.
Coronin 1 antibodies were either polyclonal rabbit antisera as described before [40] or monoclonal mouse anti-coronin 1 (Abcam). Other antibodies used were: mouse monoclonal anti-actin (Millipore); goat polyclonal (R&D); goat polyclonal (Santa Cruz Biotech) anti-SNX1, anti-SNX2, anti-SNX3, anti-SNX4 and anti-SNX5 as well as rabbit polyclonal anti-SNX5 (Abcam) and rabbit polylonal anti-rab5a (Santa Cruz Biotech); rat monoclonal anti-LAMP-1 clone 1D4B (IgG2a; developed by T. August and obtained from the Developmental Studies Hybridoma Bank at the University of Iowa); rabbit polyclonal anti-GFP (SantaCruz Biotech), mouse monoclonal anti-phosphoserine, anti-phosphothreonine and anti-phosphotyrosine (Cell Signalling) and rabbit polyclonal anti-Mycobacterium tuberculosis (SeroTech); rabbit polyclonal anti-panAKT antibody (Abcam) and rabbit polyclonal phosphoAKT (Ser473) antibody (Cell Signalling). All secondary antibodies (Southern Biotech) for western blotting were horse radish peroxide (HRP)-conjugated goat and donkey anti-rabbit, goat anti-mouse or donkey anti-goat. All secondary antibodies for immunofluorescence (Molecular Probes) were AlexaFluor488, 568 or 647-conjugated anti-rabbit, anti-mouse, anti-rat or anti-goat raised in goat or donkey.
PKH26 labeling was performed according to the manufacturer's protocol. Briefly, a total of 5×108 number of each of M. bovis BCG, M. marinum and E. coli were transferred to an 1.5 ml Eppendorf tube and washed 3 times with DMEM (without phenol red). PKH26 solution (Sigma) was prepared by diluting 10 µl of PKH26 dye (Sigma) in 1.5 ml of PKH 26 diluent (Sigma) to a final concentration of 4×10−6 M. Thereafter 500 µl of diluted PKH26 solution was added to each bacterial suspension in 500 µ l of DMEM and rotated at RT for 30 min. Thereafter the bacteria were washed once with 1 ml of FBS to stop the labeling and to remove excess PKH26, followed by 3 washes in DMEM (without phenol red). The bacteria were resuspended in DMEM (without phenol red) containing 2% FBS. The OD600 was measured and all the bacterial suspensions were brought to a OD600 of 0.1 before adding them to the macrophages.
Macrophages were seeded on Teflon-coated 10 well slides (BD Falcon) and either kept non-activated or activated. For FM4-64 staining, the dye at final concentration (5 ug/ml in DMEM) along with AlexaFluor 647 conjugate dextran 70,000 was added to the cells, incubated for 30 min followed by fixation in 4% formaldehyde in Hank's Balanced Salt Solution (HBSS) for 10 min at 4°C. Thereafter slides were blocked and stained for coronin 1 with rabbit anti-coronin 1 (1∶1000, 45 min at room temperature) followed by staining with secondary antibodies (anti-rabbit AlexaFluor 488, 30 min at room temperature). Slides were embedded using Pro-Long antifade (Molecular Probes), mounted with coverslips and analyzed using a Zeiss LSM510 Meta confocal laser-scanning microscope. For Filipin staining, incubation with rhodamine-conjugated dextran 70000 was first carried out as indicated above, followed by coronin 1 staining as stated above. Filipin at 250 µg/ml final concentration was prepared in the secondary antibody solution and incubated in the dark along with the secondary antibody solution, followed by embedding and analysis as above.
Coronin 1 cloned in pEGFP-N1 served as the wild type control for the mutants. pEGFP-N1 was the vector control while site-directed mutagenesis was carried out to mutate serines 9, 311, 356, 412 to alanine and glutamic acid using primers given in Table S1. RNAi-resistant coronin 1 constructs were generated by mutating the region targeted by RNAi (ACTGGACGAGTAGACAAG to ACTGGACGTGTGGACAAG with the mutated residues in italics) to nucleotides present in the same region of human coronin 1 by site directed mutagenesis using primers indicated in Table S1. The RNAi mutant Cor1-EGFP constructs were denoted with an (*) at the end. Transfection was carried out initially using Amaxa Nucleofector kit V (Lonza; program T-20) or the Neon Transfection system, 100 µl kit (Invitrogen) using the program: 1720 V, 25 sec and 1 pulse. Fluorescent cells were sorted using a FACS Aria III (Becton Dickinson) and either used directly for localization studies or expanded for immunoprecipitation, immunoblotting and flow cytometry studies.
Mycobacterial infection was carried out as described previously [19]. The mycobacterial inoculum was prepared by centrifuging the initial culture in 7H9 at 445×g for 5 min to remove all the clumped mycobacteria. Thereafter mycobacteria were pelleted at 2650×g for at 30°C in a swing bucket rotor (Eppendorf 5417R), followed by 3 washes in DMEM and finally diluting it to a O.D of 0.1 prior to addition to the cells. Both non-activated or differentially activated macrophages with or without different pre-treatments were seeded on 10-well glass slides (10000 cells for immunofluorescence) or 48 well plates (for colony forming unit enumeration, CFU) and incubated with mycobacteria at OD 0.1 for 1 hr, treated with amikacin (Sigma, 100 µg/ml) in DMEM, washed with DMEM followed by a chase of the times indicated.
For quantitation of lysosomal transfer, the number of LAMP1 positive mycobacteria containing cells were divided by the total number of cells analyzed and multiplied by 100 to obtain the percentage of lysosome-transferred mycobacteria. For CFU analysis the samples were diluted 1∶10 and plated onto 7H11 agar plates. Thereafter the colonies formed were counted, multiplied by 10, averaged for 3 independent experiments and plotted for each time point. For co-localization with macropinocytic markers, the number of cells exhibiting macropinocytic markers (SNX1 or SNX5) that co-localized with mycobacteria was divided by the total number of infected cells analyzed for a given time point and multiplied with 100 to result in the percentage of macropinosome-localized mycobacteria. Reorganization of coronin 1 following macrophage activation was carried out using the fluorescence images of non-activated and activated macrophages using Fiji [67]. In brief, the entire cell outline was marked and total cellular fluorescence was obtained (F-total). Thereafter, cell-internal cellular fluorescence was measured by outlining the intracellular region apart from the cell cortex (F-internal). F-total divided by F-internal multiplied by 100 allowed to determine the percentage of reorganized coronin 1 while F-internal subtracted from F-total and then divided by F-total followed by multiplication with 100 resulted in the percentage of cortical Coronin 1.
Macrophages were grown on glass coverslips in 12 well plates (5×104 cells per well). Cells were incubated with IFN-γ or TNFα for 20 hrs. or PMA for 4 hrs, prior to incubation with mycobacteria. Specific wells were pre-incubated with blebbistatin, amiloride, cytochalasin D or anti-CR3 antibody for 1 hr prior to incubation with bacteria. M. bovis BCG-GFP, M. smegmatis or E. coli at an MOI of 40 was added to the macrophages and incubated at 37°C for 90 min. Cells were immediately washed three times with ice cold PBS followed by fixation with 2.5% glutaraldehyde (EM grade). After fixing, the cells were processed for scanning electron microscopic analysis using the critical point drying technique [68] followed by analysis using a Phillips XL 30 ESEM. In Fig. 1D, mycobacteria were false-colored using Adobe Photoshop CS (version 5.1).
Bone marrow-derived macrophages were seeded in 6-well plates (1×106 cells/well) and either non-activated or IFN-γ activated in the absence or presence of different inhibitors. Thereafter complement type 3, mouse IgG or mannan -coated AlexaFluor568 conjugated 1 µm beads or rhodamine-coated Dextran 70000 or PKH26 labeled M. bovis BCG, M. marinum or E. coli in DMEM without phenol red containing 2% FBS was added to the cells and incubated for 60 min at 37°C. Subsequently, cells were washed three times with DMEM+2%FBS and collected by flushing in 300 µl DMEM+2%FBS. After incubation, cells were stained with anti-F4/80-FITC and anti-1-A/1-E-Pacific Blue for 20 min on ice followed by washing three times with DMEM+2%FBS. Just before analysis 5 µl of 7AAD-PerCP (Life technologies) was added to the cells. For compensation, unstained or single stained cells were taken. As a control for apoptotic cells control cells incubated with staurosporine for 3 hrs were taken. Cells were analyzed using a Becton-Dickinson FACS Canto II. The cells expressing the EGFP fusion constructs were sorted prior to the experiment and gated in the GFP channel during analysis and non-transfected cells served as controls. Mean fluorescence intensity of bead or bacteria uptake was obtained by multiplying the average fluorescence intensity (Mean) of the internalized cargo with the total number of cells that had internalized cargo (counts) in a fixed time by a fixed number of cells
Macrophages, either wild type, coronin 1-deficient or interferon-γ receptor-deficient, were either non-activated or activated with interferon-γ (20 hrs.), TNFα (20 hrs.) or PMA (4 hrs.) in the absence and presence of chelerythrine (1 µM). Cells were lysed with buffer P (20 mM HEPES-NaOH pH 7.4, 25 mM KCl, 1 mM MgCl2, 1% NP-40, 0.25% sodium deoxyclolate along with Halt Protease and phosphatase inhibitor (Thermo Scientific)) for 30 min on ice. Subsequently lysates were centrifuged at 16128× g for 5 min, at 4°C and the supernatant was diluted 1∶1 with buffer P without detergents and used for activated PKC analysis using a PKC assay kit (Calbiochem). In brief, equal protein amount of cell lysates were incubated in 96 well plates and mixed with radioactive [α32P]ATP and the non-phosphorylated PKC substrate RFARKGSLRQKNV. After an incubation for 30 min at 30°C, the phosphorylated substrate was separated from the residual [α32P]ATP using P81 phosphocellulose paper and quantitated by using a scintillation counter. As a positive control, activated PKCα (10 ng, Calbiochem) was mixed in the dilution buffer P (see above) and used in the assay.
The mouse coronin 1 protein sequence was analyzed using the MotifScan program (http://myhits.isb-sib.ch/cgi-bin/motif_scan). In parallel the sequence was also analyzed using ProtScale (http://web.expasy.org/protscale) for residues with lowest hydrophobicity (Kyte and Doolittle) and highest accessibility. Thereafter the residues were confirmed using NetPhosK. Finally residues 9, 311, 356 and 412 were identified as residues putatively phosphorylated by protein kinase C.
Macrophages either non-activated or activated with IFN-γ or PMA and in the absence and presence of Chelerythrine were grown in 15 cm dishes (4 plates for each sample). Thereafter the cells were washed twice with ice-cold PBS, lysed for 15 min on ice with 5 ml of T-X100 lysis buffer per dish (50 mM Tris-HCl, pH 7.5, 137 mM NaCl, 2 mM EDTA, 1 mM PMSF, 10% glycerol, 1% Triton X-100, 0.05% digitonin along with HALT protease and phosphatase inhibitor (GE healthcare)). Lysates were pooled and centrifuged at 1800×g for 5 min. at 4°C. The lysate was passed through a 0.45 µm filter and loaded onto an anti-coronin 1 column prepared by crosslinking anti-coronin 1 rabbit antiserum to NHS-coupled sepharose beads (GE healthcare). The column was washed with 100 mM glycine pH 8 followed by elution of the bound coronin 1 with 100 mM glycine pH 3. Fractions were collected (0.5 mL) and immediately neutralized using 1/10 volume of 1 M Tris-HCl, pH 8. Protein concentration was determined using BCA with bovine plasma gamma globulin (BioRad) as a standard and the coronin 1-containing fractions were concentrated followed by buffer exchange using an Amicon centrifuge column (0.5 ml, 10 kDa cutoff). The fractions (100 µl) were mixed with two-dimensional PAGE buffer (GE healthcare) and traces of Bromophenol blue) and separated on 18 cm pH 4–7 immobilized pH gradient (IPG strips GE healthcare) according to the manufacturers protocol and electrophoresed using a Multiphor system II at step increments up to 3500 V in 30 min followed by a run time of 7 hrs. Subsequently, strips were loaded on top of 10% SDS-PAGE gels of 20 cm length and electrophoresed. Immunoblotting was carried out by semi-dry transfer to nitrocellulose membrane (GE healthcare) as described before [19], [69].
Immunoprecipitation was carried out upon lysing cells in the following buffer: 20 mM HEPES-NaOH pH 7.4, 50 mM NaCl, 1 mM MgCl2, 1 mM EGTA, 0.5 mM PMSF, 0.4% Igepal CA630 (Sigma), 0.3% Na-β-D maltoside (Sigma), 0.2% digitonin (Sigma), 0.1% NP-40 including a protease and phosphatase inhibitor cocktail from Thermo Scientific. The cells were incubated in lysis buffer on ice for 20 min, followed by centrifugation at 4°C for 10 min at 20,000× g. Antibodies were coupled to Dynabeads Protein G using dimethyl pimelidate (DMP) according to the manufacturers protocol (Abcam). Antibody-coupled beads were added to these lysates and incubated overnight at 4°C. Thereafter the beads were washed 4–5 times with lysis buffer using a DynaMag (Invitrogen). Bound antigens were either eluted with 100 µl of 100 mM glycine pH 3 followed by neutralization using 1/10th volume of 1 M Tris-HCl pH 8 and were solubilized by boiling 10 min) in sample buffer and loaded on 10% SDS-PAGE, followed by immunoblotting as described above.
Cultured cells were resuspended in ice cold homogenization buffer (20 mM HEPES, pH 7.9, 10 mM NaCl, 0.5 mM EDTA, 200 mM sucrose, 0.5 mM PMSF and protease and phosphatase inhibitor cocktail (Aprotinin, Bestatin, E-64, Leupeptin, Sodium fluoride, Sodium orthovanadate, Sodium pyrophosphate, b-glycerophosphate, Thermo Scientific) kept on ice for 10 min and then homogenized on ice in a Dounce homogenizer (10–15 strokes). Subsequently, homogenates were centrifuged for 10 min at 4°C at 400×g. The pellet was discarded and the supernatant was centrifuged at 18000×g for 15 min at 4°C. The resulting pellet served as the plasma membrane fraction, while the supernatant includes the non-plasma membrane fraction [70]. The pellet was solubilized with 1% sodium-β-D-maltoside in homogenization buffer on ice and then suspended in the same buffer, kept on ice again for 15 min followed by centrifugation at 20000× g for 10 min at 4°C. Proteins were quantitated from the pellet and the supernatant fractions, separated by SDS-PAGE (10%) followed by immunoblotting as described above.
Macrophages were seeded onto Teflon-coated 10 well slides (BD Falcon) followed by the treatments as indicated. Cells were fixed with 4% paraformaldehyde in phosphate buffered saline (PBS) and permeabilized using 0.2% saponin. After blocking with 5% FBS/BSA in phosphate buffered saline, cells were stained with the primary antibodies as indicated (diluted in Dulbecco's PBS containing 5% FBS) followed by incubation with AlexaFluor-conjugated secondary antibodies (diluted in D-PBS containing 5% FBS). Slides were embedded using Pro-Long antifade (Molecular Probes), mounted with coverslips and analyzed using a Zeiss LSM510 Meta confocal laser-scanning microscope. For quantitation, 25 cells were analyzed in three separate experiments and the mean +/− SD is displayed.
For the preparation of SDS-PAGE samples of mycobacteria-containing cell fractions, macrophages either non-activated or activated with IFN-γ in the absence and presence of different inhibitors were incubated with M. bovis BCG-GFP for 1 hr followed by a chase period of 1 hr. Subsequently, cells were lysed using a Triton X-100 buffer (50 mM Tris-HCl, pH 7.5, 137 mM NaCl, 2 mM EDTA, 1 mM PMSF, 10% glycerol, 1% Triton X-100, 0.05% digitonin along with HALT protease and phosphatase inhibitor (GE healthcare)) followed by addition of glass beads equivalent to 200 µl for 500 µl sample and disrupting mycobacteria in the lysates using a mixer mill (type MM 300; Retsch, Germany) as described before [71]. Cell debris and non-lysed cells were removed by centrifugation (10 min at 10,000×g) followed by electrophoresis in 10% SDS-PAGE gels. Immunoblotting was carried out using anti-GFP antibody to specifically monitor internalized mycobacteria.
Bone-marrow derived macrophages and J774 macrophages depleted for coronin 1 by siRNA [20] were transfected with RNAi mutants of Cor1-EGFP, namely Cor1-EGFP*, Cor1S-A-EGFP* and Cor1S-E-EGFP* were either non-activated or activated with IFN-γ for 20 hrs. or PMA for 4 hrs. prior to infection with M. bovis BCG for 5 min or with IgG coated beads (1 µm) for 30 mins. Cells were washed with ice-cold HBSS followed by lysis using Triton X-100 buffer containing protease and phosphatase inhibitors. Proteins from the lysates were electrophoresed in 10% SDS-PAGE and immunoblotted using anti-phosphoAKT (Ser473), anti-panAKT, anti-coronin 1 and anti-actin.
|
10.1371/journal.pntd.0001334 | Histopathological Changes and Clinical Responses of Buruli Ulcer Plaque Lesions during Chemotherapy: A Role for Surgical Removal of Necrotic Tissue? | Buruli ulcer (BU) caused by Mycobacterium ulcerans is a necrotizing skin disease usually starting with a subcutaneous nodule or plaque, which may ulcerate and progress, if untreated, over months and years. During the currently recommended antibiotic treatment with rifampicin/streptomycin plaque lesions tend to ulcerate, often associated with retarded wound healing and prolonged hospital stays.
Included in this study were twelve laboratory reconfirmed, HIV negative BU patients presenting with plaque lesions at the CDTUB in Allada, Benin. Punch biopsies for histopathological and immunohistochemical analysis were taken before start of treatment and after four to five weeks of treatment. Where excision or wound debridement was clinically indicated, the removed tissue was also analyzed. Based on clinical judgment, nine of the twelve patients enrolled in this study received limited surgical excision seven to 39 days after completion of chemotherapy, followed by skin grafting. Lesions of three patients healed without further intervention. Before treatment, plaque lesions were characterized by a destroyed subcutis with extensive necrosis without major signs of infiltration. After completion of antibiotic treatment partial infiltration of the affected tissue was observed, but large necrotic areas remained unchanged.
Our histopathological analyses show that ulceration of plaque lesions during antibiotic treatment do not represent a failure to respond to antimycobacterial treatment. Based on our results we suggest formal testing in a controlled clinical trial setting whether limited surgical excision of necrotic tissue favours wound healing and can reduce the duration of hospital stays.
| The tropical necrotizing skin disease Buruli ulcer (BU) caused by Mycobacterium ulcerans is associated with extensive tissue destruction and local immunosuppression caused by the macrolide exotoxin mycolactone. Chemotherapy with a combination of rifampicin and streptomycin for 8 weeks is the currently recommended treatment for all types of BU lesions, including both ulcerative and non-ulcerative stages (plaques, nodules and edema). Our histopathological analysis of twelve BU plaque lesions revealed extensive destruction of sub-cutaneous tissue. This frequently led to ulceration during antibiotic treatment. This should not be mistaken as a failure of the antimycobacterial chemotherapy, since we found no evidence for the persistence of active infection foci. Large necrotic areas were found to persist even after completion of antibiotic treatment. These may disturb wound healing and the role of wound debridement should therefore be formally tested in a clinical trial setting.
| Buruli ulcer (BU), the third most common human mycobacterial disease, is caused by M. ulcerans [1], [2]. While the disease is present in around 30 countries worldwide the main focus with the highest prevalence is found in West African countries like Benin, Ghana, Cameroon and the Ivory Coast [3], [4]. Three categories of pre-ulcerative lesions, painless movable subcutaneous nodules or papules, oedema and plaques are distinguished. All three forms of pre-ulcerative lesions may progress to ulcerative lesions, when destruction of the subcutis is leading to the collapse of the overlying epidermis and dermis [1], [5], [6].
In 2004 WHO treatment recommendations for BU changed from a purely surgical treatment to a dual antibiotic therapy with rifampicin and streptomycin for eight weeks [7]. Recurrence rates after antibiotic treatment are low, but a proportion of antibiotic treated patients, in particular those with extensively ulcerated wounds, requires excisions and skin grafting [8]–[13]. During and after completion of antibiotic treatment paradoxical reactions associated with the enhancement of local immune responses and increases in size of lesions may be mistaken as disease progression [14], [15]. Observational studies have shown that while nodules usually heal after antibiotic treatment without further intervention, ulceration and an increase in the size of the lesion is often observed in the case of plaque lesions. These paradoxical reactions may occur after initial improvement and often require extensive medical care, causing long hospital stays. To elucidate the underlying mechanisms and to gain a better insight into the histopathological features of plaque lesions we conducted detailed histopathological and immunohistochemical analyses of tissue specimen from 12 plaque patients treated with rifampicin and streptomycin.
Ethical approval (clearance N° 011, 12/10/2010) for analyzing patient specimens was obtained from the provisional national ethical review board of the Ministry of Health Benin, registered under the N° IRB00006860. Written informed consent from the patients or from the guardians of the patients was obtained before surgical specimens were used for reconfirmation of clinical diagnosis and detailed immunohistological analysis.
12 patients aged between five and 70 years (median age 12 years) reporting between April 16 and August 15 2009 with laboratory reconfirmed BU plaque lesions at the Centre de Depistage et de Traitement de l'Ulcere de Buruli d'Allada in Benin were included in the study (Table 1). Most (9/12) lesions were located at the upper (4/12) or lower (5/12) extremities. The diameter of the lesions was between four and 15 cm. All patients were coming from the highly BU endemic Ze commune in the Department Atlantique of Benin. The gender distribution was nine males and three females. Clinical diagnosis was reconfirmed by positive results in at least two of the three laboratory tests (IS2404 PCR, detection of acid fast bacilli (AFBs) on microscopy and histopathology) performed. All patients completed the WHO recommended combination dual chemotherapy with oral rifampicin (10 mg/kg/day) and i. m. streptomycin (15 mg/kg/day) for 56 days. All patients were tested negative for HIV.
Punch biopsies were taken for histopathological analyses prior to start of chemotherapy on day −2 to day 0 (T1). Of these 12 T1 samples, one was not suitable for immunohistochemical analysis. A second punch biopsy taken 26 to 48 days after start of chemotherapy (T2) became available from 11/12 patients. According to the judgment of the responsible clinician, based on the evolution of the lesions including remaining induration and increasing lesion surface area, nine patients received adjunct surgical treatment seven to 39 days after completion of chemotherapy and had skin grafting five to 14 days after excision. Samples from seven of the nine excised lesions became available for histopathological analysis. In the case of the two other patients that received surgery, a third punch biopsy was taken and analyzed prior to surgical excision. Tissue samples were fixed in 4% neutral-buffered paraformaldehyde for 24 h and subsequently transferred to 70% ethanol for transport. Biopsies were dehydrated, embedded in paraffin, cut into 5 mm thin sections and retrieved on glass slides. After dewaxing and rehydration, sections were stained with haematoxylin/eosin (HE) and Ziehl-Neelsen (ZN) staining of AFBs was performed. Immunohistochemistry was performed with antibodies against Elastase (polymorphonuclear neutrophils [PMNs]; Dako clone NP57), CD3 (T lymphocytes; Dako clone F7.2.38), CD8 (cytotoxic T lymphocytes; Serotec clone 4B11), CD4 (helper T lymphocytes; Dako clone 4B12), CD68 (macrophages/monocytes; Dako clone KP1), Ki67 (proliferation marker; Dako polyclonal rabbit serum) and CD20 (B lymphocytes; Novocastra clone7D1). Staining was performed using Vector NovaRED and haematoxylin as a counterstain.
Included in this study were twelve BU patients (Tab. 1) with single new laboratory-reconfirmed plaque lesions reporting between April 16 and August 15, 2009 at the Centre de Depistage et de Traitement de l'Ulcere de Buruli d'Allada in Benin. All patients received the WHO recommended dual chemotherapy with rifampicin and streptomycin for 56 days [7]. Seven out of 12 lesions ulcerated during chemotherapy (Fig. 1C, E, F). Six of these patients received surgical treatment to remove necrotic tissue (Fig. 1E, F). Based on clinical judgment, tissue was also excised from the lesions of another three patients, which had at this stage not yet developed ulceration, but an induration (Fig 1D). Three patients healed without surgical intervention (Fig. 1A–C), one of them had developed a small ulceration during chemotherapy (Fig. 1C) which needed no further intervention.
For patients who received surgical treatment, excisions were performed 7 to 39 days (average 19 days) after completion of chemotherapy. Skin grafting followed 5 to 14 days (average 8 days) after excision and patients were discharged from hospital 16 to 100 days (average 36 days) after skin grafting. The time interval between start of antibiotic treatment till discharge from hospital was between 55 to 179 days (average 103 days) for the 12 patients (Tab.1). Depending on the location of the lesion this period was prolonged by a phase of physiotherapy (Tab. 1).
For histopathological characterization of the untreated plaque lesions, punch biopsies were taken before start of antibiotic treatment. Certain features, depicted in Fig. 2, were found in all samples analyzed. The dermis presented with relatively intact collagen with minor infiltrations around vessels and glands (Fig. 2A, C), reflecting the pre-ulcerative nature of the plaque lesions. Most strikingly, the subcutis was in all patients extensively necrotic and oedematous (Fig. 2A). Additional features typical for an untreated BU lesion, like fat cell ghosts (Fig. 2D) and minimal infiltration limited to the surrounding of a few remaining partially intact blood vessels (Fig. 2E), were always present. Immunohistochemical staining revealed N-elastase positive neutrophilic debris (Fig. 2F) reflecting an early wave of neutrophilic infiltration. In addition, only few intact neutrophils (Fig. 2G) and CD68 positive macrophages (Fig. 2H) were found. Tissue of only two patients contained also a few intact CD3 positive T-cells in the dermal tissue (data not shown).
Acid fast bacilli (AFB) were found in only 7/31 (23%) of the tissue samples analyzed altogether. This reflects the focal distribution of the mycobacteria and the extension of tissue destruction, attributable to the diffusion of mycolactone, into tissue areas with low mycobacterial burden [16]. Fig. 2B depicts an example, where a mycobacterial focus was sampled. Here a band of extracellular AFBs was found in a deep layer of the necrotic subcutis.
Punch biopsies taken between 26 and 48 days after start of antibiotic treatment typically consisted still primarily of large oedematous necrotic areas with fat cell ghosts (Fig. 3A/B). Overall, infiltration was much less pronounced than typically found in ulcerative lesions at this time point of antibiotic treatment [17]. 9/11 patients presented with a mild infiltration consisting of very few N-elastase positive neutrophils (Fig. 3D), more CD68 positive macrophages (Fig. 3E) and CD3 positive T-cells (Fig. 3F). These infiltrates were scattered throughout the dermis and extended only in 3/11 samples into the large necrotic areas. Structured infiltrates typically found in healing BU lesions [14], were rare: granuloma formation (Fig. 3H) was found in 2/11 samples, CD20 positive B-cell clusters (Fig. 3G) and giant cells (Fig. 3I) in 3/11 biopsies. AFBs were found in 4/11 samples; they were primarily intracellular or had a ‘beaded’ appearance (Fig. 3C).
While lesions of three of the enrolled patients healed without adjunct treatment, the responsible clinician decided to support wound healing by surgical excision of affected tissue in 9/12 patients. All nine excisions were performed after completion of antibiotic treatment, 56 to 94 days after start of chemotherapy. While six of the nine tissue samples were excised from lesions, which had spontaneously ulcerated during antibiotic treatment, the other three were excised from non-ulcerated lesions showing no adequate clinical improvement.
All nine excisions consisted to a large extent of necrotic and oedematous tissue with fat cell ghosts (Fig. 4A). In the case of the three patients which had still non-ulcerative lesions at the time of excision, the dermis presented with necrosis and infiltration, indicative for progression towards ulceration (Fig. 4C). Most (7/9) samples showed massive infiltration of the subcutis (Fig. 4B) often with a clear border between intact leucocytes, mainly CD14 positive macrophages/monocytes (Fig. 4D1), and the still necrotic areas containing N-elastase positive neutrophilic debris (Fig. 4D2). Infiltrates were mainly composed of CD68 positive macrophages (Fig. 4E) and CD3 positive T cells (Fig. 4F) with a higher proportion of CD8 positive (Fig. 4G) than CD4 positive (Fig. 4H) T-cells. In addition CD68 positive langhans and foreign body giant cells (Fig. 4I), granulomas and small CD20 positive B-cell clusters (Fig. 4J) were found. Some areas were strongly infiltrated with N-elastase positive neutrophils (Fig. 4K). Angiogenesis inside the necrotic and hypoxic tissue was observed in 5/9 patients (data not shown). Small numbers of intra- and extracellular AFB with a beaded appearance were found in the specimens of 2/9 patients (Fig. 4L).
BU plaque lesions are defined as firm, painless, elevated and well demarcated lesion with more than 2 cm in diameter [18]. Our histopathological analysis of the tissue specimen from plaque lesions reconfirmed earlier quantitative RT-PCR analyses [16], demonstrating a focal distribution of the mycobacteria and a mycolactone-mediated extension of tissue destruction to tissue areas with low mycobacterial burden. Our studies provided no evidence for survival of mycobacterial clusters after chemotherapy. The few AFBs found after chemotherapy had beaded appearance and were largely phagocytosed [17] and all tissues turned culture negative during treatment. Interestingly, in those punch biopsies of untreated BU lesions, where a mycobacterial focus was sampled, mycobacteria were typically found as a band of extracellular AFBs in a deep layer - several mm below the epidermis- of the necrotic subcutis (Fig. 2B). This might explain why in cases, with clinical features consistent with BU, microscopic results are frequently and PCR results are occasionally negative. Especially when collecting diagnostic specimens from non-ulcerative lesions by fine needle aspiration, also deep layers should be sampled to increase chances to reach these bacteria. Focal distribution of M. ulcerans and related lack of AFB in some of the tissue samples reflects a major limitation of histopathological analyses using punch biopsies. While findings at a particular location of the lesion may not be representative for the entire lesion, many major histopathological features described here were very consistently found in all samples analyzed.
Also earlier studies [14] demonstrating a reversal of local immune suppression during chemotherapy were confirmed. This process starts with a diffuse chronic infiltration, primarily by macrophages and T cells. While neutrophils play only a minor role in this process, N-elastase positive debris in the necrotic areas are indicative for a wave of neutrophil invasion during the early phase of BU pathogenesis. It has been shown [14] that after an initial phase of diffuse infiltration development of structured leukocyte aggregates, such as B cell clusters and granulomas usually are observed. While the development of such highly organized ectopic lymphoid tissue was also observed in the case of plaque lesions, this was confined to the margins of the necrotic areas. Large regions showing massive coagulative necrosis without significant infiltration were still found in the surgical specimens excised 7–39 days after completion of chemotherapy.
Infiltration and angiogenesis in the affected tissue during and after chemotherapy promotes the resorption of tissue debris. Initial inflammatory responses may be associated with paradoxical reactions, before converging into a phase of wound healing. In two of the 12 patients enrolled, this process of resorption of necrotic tissue was efficient enough to permit healing without ulceration. However, in most cases the necrotic areas of plaque lesions seemed to be too extensive to permit complete resorption without ulceration. In general, our histopathological analysis of plaque lesions revealed a much larger and deeper destruction of the subcutaneous tissue than expected. Spontaneous ulceration during antibiotic treatment was observed in 7/12 patients and tissue samples from three other non-ulcerated patients showed gradual degeneration of the dermis, indicative for a progression to ulceration. While ulceration result in the loss of necrotic tissue, our analysis of tissues surgically excised 7–39 days after completion of chemotherapy from lesions that spontaneously ulcerated during chemotherapy, revealed incomplete loss of necrotic tissue. These findings support the decision of the responsible clinician to support wound healing by debridement of the margins of the ulcers. It is well documented that wound debridement i.e. the removal of materials incompatible with healing, can substantially accelerate the complex wound healing process [19]–[21]. Even if superinfections are controlled with antibiotics, chronic wounds can be caught in a chronic inflammatory phase and debridment is then required to convert the chronic wound bed into an acute wound and mediate healing through the stages of inflammation, proliferation and maturation [20]. This may also apply for BU lesions which show massive infiltration during antibiotic treatment and may subsequently be arrested in a chronic stage without the chance of proper healing. In patients with severe and extensive lesions an early decision for wound debridement may therefore reduce hospital stays to less than 100 days.
In the present study the time span between start of treatment and discharge from the hospital was 55–61 days for the three patients, which did not require an excision. For those three patients, which showed no spontaneous ulceration, but were surgically treated to remove necrotic tissue this time span was 90–95 days. Degeneration of the dermis of the excised lesions harbouring large areas of necrotic tissue indicated that these patients would have developed ulceration at a later stage, leading to a severely delayed subsequent start of the healing process. Those six patients who showed spontaneous ulceration during chemotherapy and received later wound debridement, stayed for 108–179 days in the hospital. Surgical treatment was performed 7–39 days after completion of chemotherapy and skin grafting 5–7 days after excision. Five of these six patients were discharged from hospital 27–38 days after skin grafting, only in one case this time period was much longer (99d) due to secondary infection and delayed wound healing. While bacteria are present on basically every open wound, secondary infections above a critical bacterial load (>105 organisms per gram of tissue) may lead to an arrest of the wound healing process [22]–[24].
Taken together our analysis indicates that due to massive coagulative necrosis only a minority of BU plaque lesions may heal without spontaneous ulceration. It appears advisable to consider debridement of ulcers that have developed in the course of chemotherapy in order to remove necrotic tissue and to favour wound healing. The effect of adjunct surgical treatment should therefore be formally tested in a clinical trial setting to support development of differentiated guidelines for BU wound management.
|
10.1371/journal.pcbi.0030167 | Functional Representation of Enzymes by Specific Peptides | Predicting the function of a protein from its sequence is a long-standing goal of bioinformatic research. While sequence similarity is the most popular tool used for this purpose, sequence motifs may also subserve this goal. Here we develop a motif-based method consisting of applying an unsupervised motif extraction algorithm (MEX) to all enzyme sequences, and filtering the results by the four-level classification hierarchy of the Enzyme Commission (EC). The resulting motifs serve as specific peptides (SPs), appearing on single branches of the EC. In contrast to previous motif-based methods, the new method does not require any preprocessing by multiple sequence alignment, nor does it rely on over-representation of motifs within EC branches. The SPs obtained comprise on average 8.4 ± 4.5 amino acids, and specify the functions of 93% of all enzymes, which is much higher than the coverage of 63% provided by ProSite motifs. The SP classification thus compares favorably with previous function annotation methods and successfully demonstrates an added value in extreme cases where sequence similarity fails. Interestingly, SPs cover most of the annotated active and binding site amino acids, and occur in active-site neighboring 3-D pockets in a highly statistically significant manner. The latter are assumed to have strong biological relevance to the activity of the enzyme. Further filtering of SPs by biological functional annotations results in reduced small subsets of SPs that possess very large enzyme coverage. Overall, SPs both form a very useful tool for enzyme functional classification and bear responsibility for the catalytic biological function carried out by enzymes.
| Sequence motifs are known to provide information about functional properties of proteins. In the past, many approaches have looked for deterministic motifs in protein sequences, by searching for functionally over-represented k-mers, with moderate levels of success. Here we revisit and renew the utility of deterministic motifs, by searching for them in a partially unsupervised and context-dependent manner. Using a novel motif extraction algorithm, MEX, deterministic sequence motifs are extracted from Swiss Prot data containing more than 50,000 enzymes. They are then filtered by the Enzyme Commission classification hierarchy to produce sets of specific peptides (SPs). The latter specify enzyme function for 93% of the data, comparing well with existing approaches for enzyme classification. Importantly, SPs are found to have biological significance. A majority of all known active and binding sites of enzymes are covered by SPs, and many SPs are found to lie within spatial pockets in the neighborhood of the active sites. Both these results have extremely high statistical significance. A user-friendly tool that displays the hits of SPs for any protein sequence that is presented as a query, together with the EC assignments due to these SPs, is available at http://adios.tau.ac.il/SPSearch.
| One of the major efforts of computational research in molecular biology is to predict the function and spatial structure of proteins from the protein sequence of amino acids [1,2]. Conventional approaches to function prediction rely on sequence [3] or structure [4] similarity with proteins whose functions are known. This is sometimes misleading [4–6]. Alternatively, one may use motif approaches [7–12], trying to extract from the data subsequences that are responsible for particular functions. Motifs can be deterministic sequences of amino acids, regular expressions that allow various alternatives for specific locations within the motif, or stochastic structures specifying the probability of an amino acid at every location. This work aims to uncover deterministic sequence motifs, and considers their relationships with protein functionality. We focus on enzymes, whose functions are classified by the Enzyme Commission (EC) four-level hierarchy which is represented by four integers, n1.n2.n3.n4, corresponding to the different levels of classification. For example, the oxidoreductases class corresponds to n1 = 1, one of the six main divisions. For this class, n2 (subclass) specifies electron donors, n3 (sub-subclass) specifies electron acceptor, and n4 indicates the exact enzymatic activity.
Conventional sequence motif searches in enzymes are performed in a supervised fashion, using sequences of proteins that are known to have the same function and looking for (deterministic, regular-expression, or stochastic) motifs that are over-represented in this group of proteins. The motifs in question should then subserve such functions as [9] phosphorylation of protein kinases; metal binding sites for calcium, zinc, copper, and iron; enzyme active sites, etc. With the advent of studies of protein–protein interactions, interest grew in finding sequence motifs that are responsible for them, and span an “interaction space” [13,14].
Here we perform a large-scale search for deterministic sequence motifs without specifying a priori their exact functional roles, using the unsupervised motif extraction (MEX) algorithm [15]. We have used one functional guidance: MEX was separately applied to each one of the six major EC classes. The same motifs may also appear in other classes, yet many of them turn out to occur in only one class, and belong to a specific EC branch. The latter (see Figure 1A) are termed specific peptides (SPs). By representing some 50,000 enzymes (of average length of 380 amino acids) in terms of about the same number of SPs (of average length 8.4), we obtain a largely compressed functional representation and an EC classification with 93% accuracy.
This may be compared with other methods based on e-motifs [16], sequence similarity [17], or physicochemical properties of the amino acids contained in the sequence [18,19]. Our results compare favorably with such methods, as will be shown below, yet our approach differs in several respects: we use a largely unsupervised motif extraction method, we perform a comprehensive study of all enzymes, and we put major emphasis on the biological relevance of the SPs themselves.
Importantly, in comparison with the large-scale and popular motif database ProSite [8], our approach displays a wide-margin advantage, their motifs coverage extending only to 63% of all enzymes in the database.
SPs, as defined above, are MEX motifs that are specific to a single branch of the EC hierarchical classification. Most belong to single branches of the fourth level of the hierarchy, to be denoted as SPs of level 4 (SP4) (see Figure 1A). SPs of higher hierarchy, SP3, SP2, and SP1, appear in more than one lower EC level. Thus, if a peptide is shared by two or more level 4 groups that belong to the same third EC level, and appears nowhere else, it is assigned to SP3. The SPs were further screened to eliminate any peptide that includes within it another peptide carrying the same SPN (N = 1,2,3,4) label.
The majority of SPs found at level 4 of the EC hierarchy (Table 1) are probably due to the high homology within this level, that often includes many orthologous genes. Thousands of SPs occur at higher levels of hierarchy, reflecting functional similarity among enzymes with lower sequence similarity. The occurrence of any one SP on the sequence of an enzyme specifies its EC functionality according to the specific branch N of its SPN. For example, enzyme P45048 (see Figure 1B) contains SSAATYG, an SP3 specific to 5.1.3, and LNVYGYSK, an SP4 specific to 5.1.3.20. The relationship of these SPs to the EC hierarchy of SP families is shown in Figure 1A.
Table 1 shows that the SPs cover (i.e., appear on the sequence of) most enzymes in the dataset. The coverage columns display the cumulative coverage of all SPs to their left. Coverage is a measure of the success of the SP approach. Thus, from the sixth column one can deduce that functional classification at the third level of EC is specified by 45,819 peptides of SP3 and SP4, covering 89.8% of the data.
Information about the separate coverage of each SPN group is provided in Table S1. The length distribution of SPs is displayed in Figure S1 for all enzyme classes. No SP exists with a length shorter than four amino acids. The average SP length is 8.4 (s.d. 4.5). The distribution of the number of SPs occurring on enzymes is given in Figure S2. It is very flat. On average, 15.6 SPs appear on each enzyme and the standard deviation is 16. Enzyme sequences that share long SPs are highly similar, while sharing short SPs indicates smaller sequence similarity. This is displayed for short (smaller than nine amino acids) and medium length (between nine and 12 amino acids) SPs in Figures S3 and S4: most enzyme pairs that share SPs of length larger than 12 amino acids possess sequence identity of over 90%.
The SwissProt 48.3 dataset contains 260 enzymes that have more than one annotation, and, therefore, have been excluded from the training set (see Methods). Using them as a test set, we find 849 hits of SPs on 157 of these enzymes. 711 of the 849 hits agree with one of the given annotations and 138 do not, thus obtaining an accuracy of 84%. The results are displayed in Table S2, comparing the Swiss-Prot EC annotations with SP predictions. For example, the first protein on the list has Swiss-Prot EC annotations of 2.7.2.4 and 1.1.1.3. Its sequence matches two SPs, one SP1 of class 1 and one SP4 of 2.7.2.4. This is counted as two correct matches. An analysis of Table S2 shows that predictions based on a single SP hit may be erroneous, while those based on more than two SPs whose EC assignments are consistent with one another are correct.
We have tested the generalization quality of our SP-based enzyme classification by running MEX on the Swiss-Prot 45 release (October 2004) and testing its predictions on 10,000 novel enzymes that are listed in the Swiss-Prot 48.3 release (for the relation between these two sets see Figure S5 and Table S3). Generalization quality is assessed in Table 2 by recall (matching SPs extracted from the 45 data on novel enzymes) and precision (correctness of the “45” EC assignment according to “48.3” annotations). Precision can be defined at the SP level, i.e., to what extent did the EC of this SP match the true EC of the enzyme that it hits. Precision can also be defined at the enzyme level: how many enzymes are correctly identified by all SPs that hit them. In other words, demanding the EC assignments of all SPs to be consistent with one another as well as with the “48.3” annotation of the enzyme. Overall recall is 84%. Precision at the SP level is almost perfect, 98.7%; nonetheless, at the enzyme level it reduces to 81.7%. The reason is that usually there are many SPs hitting each enzyme, and the small error at the SP level is magnified by the requirement that the EC labels of all SPs on the same enzyme should be consistent with each other.
This generalization test suffers from bias, i.e., there exist enzymes in the test set that have high sequence similarity to some enzymes in the training sets. In conventional machine-learning analysis of sequence to function classification [2], one often tries to eliminate bias by avoiding high sequence similarity between proteins in the test set and proteins in the training set. In our case this is problematic, because it effectively calls for eliminating from the test set all enzymes that have four-digit EC numbers appearing in the training set. Alternatively, one could produce for each enzyme in the test set a new training set that does not contain sequences with the same EC number, which is both unconventional and computationally very complex.
To overcome this predicament, we have used the following procedure: a) start with the test set consisting of all sequences of SwissProt release 48.3 that do not appear in release 45; b) blast each one of these (test set) sequences against the sequences of the training set (SwissProt release 45) that do not have the same four-digit EC number; c) include in the non-redundant test set only sequences whose BLAST score [20] with all other training sequences (including those with the same first three EC digits) is larger than 10−3; d) test generalization on this non-redundant set only for peptides in SP1, SP2, and SP3, thus avoiding the SP4 peptides that were extracted from the same fourth-level EC sequences as those of the non-redundant test set. It should be noted that removing the SP4 peptides makes the functional annotation task much more difficult because the coverage of enzymes by SPs is strongly reduced. Only 440 enzymes obey the BLAST > 10−3 condition, and less than 40% of them carry SP1, SP2, and SP3 matches.
The results are displayed in Table 3. We obtain correct classification with an accuracy of 88%. The test is that of precision of SP assignments, i.e., to what extent do the EC labels of the SPs, observed to exist on the enzyme sequences, correspond to “48.3” EC classifications.
Whereas even the unbiased tests have high precision, we should emphasize that many successes of the SP approach are due to SP4 peptides, whose existence stems from high homology among different sequences that belong to the same EC number. These successes include the high coverage of enzymes (see Table 1) and the coverage of active and binding sites to be discussed below. The fact that these SPs have been extracted by MEX may be viewed as the essence of homology, as illustrated in Figure 1B, where the existence of SPs is displayed on various enzymes aligned according to their matching SPs.
We provide a Web tool, available at http://adios.tau.ac.il/SPMatch, which displays the hits of SPs for any protein sequence that is presented as a query, together with the EC assignments due to these SPs.
We have tested the usefulness of the SP approach by comparing it with conventional functional prediction methods. For this purpose we have used all oxidoreductases in the 48.3 data and divided them into training data and test data with a 75%:25% ratio. MEX was run on all data and SPs were selected from the MEX motifs according to the training data. Only this subset of motifs was then employed to classify the test data. This procedure has been repeated 45 times to gain statistics, and has been subjected to a support vector machine (SVM) analysis. It has been compared with a state-of-the-art method [17] based on an analogous SVM procedure, applied to the same data using the same divisions and relying on classification of (train and test) data according to a matrix of Smith-Waterman distances from all oxidoreductases. The results are displayed in Tables S4 and S5 and show a clear advantage to SP classification. For comparison, we use the Jaccard score defined as J = TP / (TP + FP + FN) where TP, FP, and FN denote true positives, false positives, and false negatives, accordingly. Whereas sequence similarity leads to an average Jaccard score of 0.86 on the second EC level and 0.82 on the third level, SP classification has average Jaccard scores of 0.93 and 0.92, accordingly. Comparing with yet another method, SVM-Prot [18,19], which classifies enzymes on the basis of physical and chemical features of their amino acids, we note that the latter achieves a Jaccard score of only 0.74 on all oxidoreductases data at the second EC level.
The common lore, that large sequence identity between two proteins implies that the two have the same function, has its exceptions. Motifs, although often extracted from homology, may serve as better measures for functional specification of proteins [21] than overall sequence similarity. Table 4 demonstrates this point, by contrasting SP predictions with Smith-Waterman similarity results for pairs of enzymes. These extreme cases have been posed as a problem by Ross [5] (see Table 1 there). All displayed EC assignments correspond to those of SPs located on the enzyme sequences, and match the correct EC numbers. As a more detailed example, we point out that the enzymes of the sixth pair in Table 4, GTFB_STRMU and AMY3B_ORYSA, have 42% sequence identity along an alignment of 105 amino acids. Nonetheless, the sequences are not identical at the SP locations. AMY3B_ORYSA contains 24 SPs, none of which have an exact match on GTFB_STRMU, and a single SP4 (GGAFLE) found on the latter matches correctly its EC number.
It is of interest to compare our SPs with ProSite motifs [8], which are listed in the Swiss-Prot database as standard motif annotations on 63% of the enzymes. ProSite motifs are either regular expressions (of average length 18.3 amino acids) or weight matrices, while SPs are deterministic motifs (with average length of 8.4). We search for all appearances of ProSite regular expression motifs on enzymes. Each such appearance is noted on the enzyme sequence and checked whether it is also (partially) covered by an SP. Figure S6 compares the appearance of SPs and ProSite motifs on the data, and Figure S7 displays the relative coverage of ProSite motifs by SPs as function of the minimal percentage of amino acids belonging to the ProSite motif that are also located on SPs. Thus we find that if at least 40% of the amino acids of the ProSite motif also belong to SPs, which would be appropriate for an average SP to be located within an average ProSite motif, then SPs cover 48% of all ProSite motif occurrences. This may be compared with a random model (see Methods) which covers on average only 24% of ProSite motif occurrences, with a standard deviation of 0.06%. This extremely significant result (400 s.d.) demonstrates that SPs carry information that is highly correlated with that of ProSite motifs.
We started our study with 50,698 enzymes from which 52,365 SPs were extracted. These SPs provided coverage of about 93% of all enzymes. By introducing further screening of SPs according to biological findings, a much reduced number of SPs may suffice for the purpose of classification. 21,228 enzymes carry active or binding site annotations in the 48.3 data. The number of SPs hitting these enzymes is 26,931; however, only 2,337 cover the active or binding sites. These 2,337 are found to occur on 79% of the 21,228 enzymes. Thus, instead of the approximately 1:1 ratio between the number of SPs and the number of enzymes they cover as found previously, we now obtain an order of magnitude parsimonious ratio, of about 1:8, while maintaining a similar level of classification accuracy.
The same SPs cover 36% of all original enzymes of our dataset. Performing a similar analysis on the 45 data, one finds that the 2,014 SPs that cover the annotated enzymes in it hit 75% of the relevant set of enzymes. Moreover, using the same SPs to classify the 10,585 novel enzymes contained in the 48.3 release and absent from the 45 release, one obtains coverage of 28% of them. This last fact demonstrates that the relatively large coverage reached by the small fraction of SPs that hit active sites is not limited to the dataset (training set) used to define the SPs. All these results are summarized in Table 7. It seems therefore quite reasonable to conclude that, adding information of biological markers, one can reduce the ratio of the number of SPs deduced from a certain number of enzymes and needed to label their EC classification from 1:1 to about 1:8.
This, however, does not mean that all other SPs should be disregarded. First, there exist good chances that they are of biological importance for various structural and functional reasons that may warrant further investigation. Second, when extreme classification issues come up, as in the cases displayed in Table 4, every single SP may count.
Conventional wisdom attributes protein functions to large domains, as well as to specific amino acids at strategic structural points on the protein. Large-scale studies often make use of multiple sequence alignment (MSA), phylogenetic information, and sophisticated mathematical models, thus leading to the plethora of algorithms and Web tools that permeate bioinformatics. While all that may be necessary to obtain a thorough understanding of the way proteins develop and perform, much can be gained by shifting attention to deterministic linear motifs on proteins. In doing so, we return to a way that has been often tried in the past. Thus, in the 1990s, many investigations looked for k-mers that are over-represented in sequences of proteins that have common functional properties. Some examples are ProSite [8,12], with which we have compared our results, and papers such as [28–30], where major emphasis has been put on finding a complete dictionary of motifs that cover all strings of amino acids that are of any importance. In the case of [30], the search has been an unsupervised one leading eventually to a coverage of 98% of all amino acids on the protein strings. Some reviews of the motif approaches of the 1990s are [7,9]. More recently, interests have shifted to automated prediction tools that may make use of motifs but are not limited to them. Examples are the GOtcha method [31] that uses sequence-identity searches of various genomes to predict functional annotation, and [32] who pursue the same goal using PSI-BLAST searches with varying resolution.
Our goal is more moderate, restricting ourselves to the functional classification of enzymes. By doing so, and by applying the MEX algorithm together with limiting ourselves to SPs within the EC hierarchy, we are able to classify all enzymes by SPs occurring on them with coverage between 87% to 93%, depending on the EC level that is being looked for (Table 1). Classification success of novel sequences that belong to the same type of data has coverage of 84% and precision of 99% at the SP level and 82% at the enzyme level (Table 2). Restricting ourselves to low bias (Table 3), we still have a large precision of 88% at the SP level. We have demonstrated that our results surpass the classification accuracy of sequence similarity (using Smith-Waterman [33]), and our SPs have a higher coverage than ProSite motifs. As such, they become a powerful tool that may be added to existing automated searches.
It should be noted that the SPs were extracted by an unsupervised motif search algorithm, applied to each one of the six EC classes. This is quite different from conventional supervised approaches. Our method may disregard motifs that obey some over-representation criterion, and choose others that do not satisfy such a global statistics measure. Another major difference from other approaches is that we do not make use of MSA. MEX finds significant motifs without requiring alignment as a preprocessing stage. In fact, MEX can serve as a source for MSA by employing its motifs for alignment (see Figure 1B).
SPs were selected from all MEX motifs by imposing the condition that they should be specific to particular levels of the EC hierarchy. This has led to a large number of SPs, as numerous as the set of all enzymes (but, obviously, providing a much more concise description). Imposing further biological conditions, one may find much smaller sets that suffice for classification. In an analysis of enzymes for which the active sites are known, we have shown that the set of SPs bearing these active sites, which comprises just 8.6% of all relevant SPs (i.e., those occurring anywhere on these enzymes), suffices to cover (and therefore label) all enzymes.
Conventional classification methods rely on homology. While large homology is also at the root of our success for most SPs of level 4 (see some examples in Figure 1B), we have demonstrated (in Table 4) that SPs can also be of importance in extreme cases, where straightforward comparison of an enzyme to another one with large sequence similarity may be misleading.
In conclusion, we have established a comprehensive and accurate classification scheme for enzymes based on the occurrence of short peptides on their sequences. The SPs contain, on average, just 8.4 amino acids, yet they suffice to correctly classify an overwhelming majority of known enzymes. Moreover, we have found indications for some of the biological roles of SPs, e.g., covering a majority of active sites. This study has laid the foundations for the further experimental investigation of these intriguing sets of SPs.
MEX is a motif extraction algorithm that serves as the basic unit of ADIOS [15], an unsupervised method for extraction of syntax from linguistic corpora. We apply it to the problem of finding sequence motifs in enzymes.
Each enzyme sequence is represented as a path over a graph containing 20 vertices, each vertex representing one amino acid. After uploading all enzyme sequences onto the graph, one counts the number of paths connecting vertices in order to define probabilities such as
p(ej|ei) = (number of paths proceeding from ei to ej) / (total number of paths leaving ei)
p(ek|ej,ei) = ( number of paths proceeding from ei to ej to ek) / (number of paths proceeding from ei to ej)
for all vertices ei of the graph. These data-driven probabilities allow for the definition of a position-dependent variable-order Markov model describing the data.
A motif that is extracted by MEX is a subpath along the graph defined by probability-based criteria that account for convergence of many paths into the beginning point of a motif, and divergence of many paths from the endpoint of the motif. Motifs are not constrained by length, and may overlap with one another (see, e.g., the two SPs that overlap at the active site D in Figure 2B). The only two parameters of MEX are η, specifying a decrease in probability measures that determine convergence and divergence, and α specifying their statistical significance. For more details, see [15] and http://adios.tau.ac.il. Throughout this paper, we use η = 0.9 and α = 0.01.
Protein sequences annotated with EC numbers were extracted from the Swiss-Prot database (Release 48.3, 25 October 2005). To obtain a high-quality, well-defined training dataset, the data were strictly screened and the following sequences were removed: sequences shorter than 100 amino acids or longer than 1,200 amino acids, sequences with uncertain annotation, and enzymes that catalyze more than one reaction (e.g., have more than one EC number).
Enzyme sequences are searched for matches with regular expressions of ProSite motifs. The resulting strings of amino acids are checked for matches with SPs. The latter are compared with matches of a random model where, for each given enzyme, random peptides are selected with the same lengths as those of the SPs that hit this enzyme. The random model provides a probability distribution which serves as a zero model for calculating the significance of the SP hit on the ProSite motif. This comparison is being made for each enzyme and for varying fractions of amino acids that are shared by the SP with the ProSite motif.
In analyzing the significance of SP coverage of active (and binding) sites, we compare this coverage with that of randomly chosen residues on enzyme sequences. This is carried out on all data (i.e., annotated enzymes with SP hits) and on a non-redundant set composed of only one enzyme from each EC number (i.e., EC classification at level 4). The deviations of the measurements from random distributions are very high, and are quoted in numbers of standard deviations. The corresponding p-values are zero according to Matlab accuracy, i.e., are well bellow 10−308.
Let us define an event as the occurrence of a given SP within an active pocket in a given enzyme. For each such event, we evaluate the probability that at least one of randomly selected sequences from this enzyme, which coincide in length with the various SPs that occur on this enzyme, lies (with at least four amino acids) within the active pocket. This defines the p-value that we assign to the event. We then select the significant events according to an FDR limit [33] of 0.05. |
10.1371/journal.ppat.1003682 | Bacterial Pathogens Activate a Common Inflammatory Pathway through IFNλ Regulation of PDCD4 | The type III interferon (IFNλ) receptor IL-28R is abundantly expressed in the respiratory tract and has been shown essential for host defense against some viral pathogens, however no data are available concerning its role in the innate immune response to bacterial pathogens. Staphylococcus aureus and Pseudomonas aeruginosa induced significant production of IFNλ in the lung, and clearance of these bacteria from the lung was significantly increased in IL-28R null mice compared to controls. Improved bacterial clearance correlated with reduced lung pathology and a reduced ratio of pro- vs anti-inflammatory cytokines in the airway. In human epithelial cells IFNλ inhibited miR-21 via STAT3 resulting in upregulation of PDCD4, a protein known to promote inflammatory signaling. In vivo 18 hours following infection with either pathogen, miR-21 was significantly reduced and PDCD4 increased in the lungs of wild type compared to IL-28R null mice. Infection of PDCD4 null mice with USA300 resulted in improved clearance, reduced pathology, and reduced inflammatory cytokine production. These data suggest that during bacterial pneumonia IFNλ promotes inflammation by inhibiting miR-21 regulation of PDCD4.
| The role of interferons (types I, II, and III) in viral and bacterial infections has been a topic of intense research over the last decade. The contribution of the type I interferons during bacterial pneumonias particularly has been shown to be highly variable depending on the specific pathogen. Our data for the first time demonstrate that type III interferon plays a significant role in the pathogenesis of bacterial pneumonia, and its contribution is similar in both Gram positive and Gram negative infections. We show in epithelial cells that miR-21 and PDCD4 are downstream effectors of type III interferon that prolong production of inflammatory cytokines. Utilizing mice that lack the receptor for type III interferon or PDCD4, we show that inhibiting this pathway improves bacterial clearance from the airways and lung tissue. These data suggest novel targets for therapy in a variety of bacterial pneumonias.
| The interferon (IFN) family is composed of three subgroups (types I, II, and III IFN), and through their distinct receptors, IFNs signal through STAT transcription factors to upregulate expression of over 300 IFN dependent genes. In the lung bacterial pathogen associated molecular patterns (PAMPs) can be internalized and thus gain access to the intracellular receptors involved in type I IFN signaling [1]–[3]. Activation of type I IFN signaling can be either protective or detrimental to the host depending on the specific pathogen [4]–[9]. Type I IFNs promote the pathogenesis of Staphylococcus aureus pulmonary infection through upregulation of CXCR3 chemokines and T-cell recruitment while improving eradication of Pseudomonas aeruginosa by reducing inflammasome signaling [1], [10]–[16].
Interferons induce expression of downstream genes through distinct receptors. Type I IFN signals through the ubiquitously expressed interferon-α/β receptor (IFNAR), while the type III IFN (IFNλ) family, composed of IL-28A/B and IL-29, signals through the more cell specific receptor complex of IL-10R2 and IL-28R [17]–[21]. Following activation of either IFN pathway, an autocrine signaling network mediates the cellular response, primarily through JAK/STAT signaling and the induction of IFN dependent gene expression. It would seem that the two IFNs activate redundant downstream signaling pathways. However altered signaling kinetics and the limited distribution of IL-28R, restricted primarily to mucosal tissues, suggest distinct roles for type I and type III IFN depending on the infection site [22]–[28]. For example in the lung type III IFN is the primary IFN produced by respiratory epithelial cells in response to viral stimulation and is required for clearance of influenza from the airway [29]–[32]. Several of the immunological effects of IFNλ such as upregulation of MHC I and II, induction of NF-κB dependent cytokine production and effects on DC maturation and differentiation, could be highly relevant to the pathogenesis of bacterial infection [33]–[36].
Interferon signaling is linked to the regulation of micro RNAs, small non-coding RNA inhibitors of mRNA translation capable of directly influencing innate immune signaling [4]–[9]. In tumor cells, miR-21 targets the tumor suppressor programmed cell death protein 4 (PDCD4) promoting tumor growth and contributing to the inflammatory tumor microenvironment [1], [10]–[16]. PDCD4 represses translation of cellular mRNA through its binding to eukaryotic initiation factor 4A (eIF4A) [17]–[21]. Phosphorylation of PDCD4 by the ribosomal S6 kinase (p70S6K) results in release of eIF4A from PDCD4, ubiquitin dependent degredation of PDCD4, and enhanced mRNA translation [22]–[28]. Inflammatory cytokine production in response to toll like receptor (TLR) 2 or 4 activation by decorin or lipopolysaccharide (LPS) is significantly influenced by expression of miR-21 and PDCD4 [29]–[32]. Therefore the pro-inflammatory contribution of PDCD4 to host signaling during bacterial pneumonia could significantly contribute to lung pathology [33]–[36].
In the experiments detailed in this report, we examined the importance of type III IFNs in the innate immune response to two major airway pathogens, USA300 MRSA and P. aeruginosa, currently the most common causes of ventilator associated pneumonia [37], [38]. We found that IFNλ is induced in the course of bacterial airway infection and results in downregulation of the microRNA miR-21, sustained expression of PDCD4, and increased proinflammatory cytokine production. Mice lacking the IFNλ receptor, IL-28R, or PDCD4 had significantly improved clearance of both airway pathogens and less pulmonary pathology associated with reduced levels of inflammatory cytokines.
In vitro studies were performed to establish the kinetics of IFNλ induction in response to the extracellular bacterial pathogen S. aureus. A significant increase in IFNλ transcript (p < 0.001) was observed in wild type (WT) bone marrow derived dendritic cells (BMDCs) following 4 hours of stimulation with heat killed USA300, which returned to baseline by 8 hours (Fig. 1A). Similar induction was observed when BMDCs were stimulated with live bacteria (MOI 100, 4 hours), and induction was significantly reduced (p = 0.0062) in IL-28R−/− BMDCs (Fig. S1A).
To evaluate the role of type III IFN in MRSA pneumonia we infected WT C57BL/6 and IL-28R−/− mice intranasally with 1×107 CFU USA300. IFNλ mRNA levels increased significantly by 4 hours in the lung of WT mice (p = 0.0047) (Fig. S1B), and protein levels were significantly increased in the airway of WT mice by 4 (p = 0.0051) and 18 (p < 0.0001) hours post infection (Fig. 1B). By 4 hours following intranasal infection the numbers of bacteria recovered from the airway were significantly reduced in IL-28R−/− mice compared with control (p = 0.0178) (Fig. 1C), a difference not observed in the lung tissue (Fig. 1D). By 18 hours significantly fewer bacteria were recovered from both the airway (p = 0.0002) and tissue (p = 0.0043) of knockout mice compared to control. Lung pathology was reduced in IL-28R−/− compared to WT controls as determined by trichrome stained lung sections (Fig. 1E). The numbers of dendritic cells, macrophages, or neutrophils recruited to the airway (Fig. 1F) or lung tissue (Fig. 1G) were not affected by lack of IL-28R, suggesting that differences in bacterial clearance and pathology were not due to increased numbers of phagocytotic cells.
Type III IFN activates a family of over 300 genes through a Jak-Stat signaling cascade [39], [40]. We analyzed expression of cytokines in the bronchial alveolar lavage (BAL) of wild type and IL-28R−/− mice following 4 and 18 hour MRSA infection by ELISA. Significant differences in cytokine expression were not observed following a short (4 hour) infection suggesting that initial activation of cytokine production does not depend on signaling through IL-28R. By 18 hours significant reductions in KC (p < 0.0001), GM-CSF (p = 0.0009) and IL-1β (p = 0.0002) but not TNF or IL-10 were observed in IL-28R−/− as compared with WT mice (Fig. 2A). Levels of the interferon response gene MX1 were reduced in IL-28R−/− mice compared to WT at the 18 hour time point, confirming the inhibition of interferon signaling (Fig. 2B). Installation of recombinant IFNλ did not significantly increase cytokine expression in the BAL of uninfected mice (Fig. S2). Therefore it appears that IFNλ does not directly induce cytokine production, but is more likely involved in the regulation of inflammatory cytokines during acute S. aureus infection.
The pro-inflammatory cytokine IL-1β has shown to promote lung injury during P. aeruginosa pneumonia, and was significantly decreased in the airways of S. aureus infected IL-28R−/− mice [11]. To determine if decreased IL-1β was similarly associated with the improved outcome of the IL-28R−/− mice, we predicted that Anakinra, a synthetic IL-1R antagonist, would improve clearance of S. aureus from the lung and reduce lung pathology. Numbers of bacteria recovered 18 hours following infection with USA300 were significantly lower from the airway (p = 0.0336) (Fig. 2C) and lung (p = 0.0088) (Fig. 2D) of Anakinra treated compared to PBS pretreated controls. Lung pathology was reduced in Anakinra treated mice following an 18 hour infection compared to infected control mice (Fig. 2E). These data suggest that the reduced pathology observed in IL-28R null mice was due in part to reduction in the inflammatory cytokine IL-1β.
PDCD4 has been shown to promote inflammatory cytokine production in response to LPS, and reduction of PDCD4 expression in human epithelial cell monolayers (16HBE) significantly reduced IL-8 induction (p = 0.0103) in response to USA300 (Fig. 3A) [31]. Since PDCD4 expression is regulated by the micro RNA miR-21, we tested whether IFNλ promotes inflammatory signaling by inhibiting miR-21 expression. In vitro, miR-21 expression in 16HBEs was reduced following 1 hour treatment with recombinant IFNλ (Fig. 3B), which correlated with an increase in PDCD4 protein levels (Fig. 3C). Reduction of miR-21 (Fig. 3D) and increased PDCD4 expression (Fig. 3E) following IFNλ treatment was dependent on STAT3 phosphorylation.
In response to acute LPS exposure macrophages upregulate PDCD4 expression [31]. S. aureus, as a Gram positive pathogen, can induce inflammatory signaling in epithelial cells through interaction of its major surface component protein A (SpA) and host receptor TNFR1 [13], [41]–[43]. We hypothesized that SpA-TNFR1 interaction would upregulate PDCD4 expression and promote production of inflammatory cytokines. 16HBE epithelial cells were exposed to S. aureus protein A (SpA) and increases in the expression of PDCD4 (p = 0.0068) were observed (Fig. 3F). To confirm the relationship between TNFR1 and PDCD4, we pre-treated 16HBEs with antibody to the extracellular domain of TNFR1 and demonstrated significantly reduced induction of PDCD4 by SpA compared to IgG pretreatment (p = 0.029) (Fig. 3G). These data demonstrate that SpA, like LPS, is able to induce expression of PDCD4 in host cells.
In vivo, miR-21 expression in WT and IL-28R null mice was not different prior to infection. Expression of miR-21 increased by greater than 10-fold in both WT and IL-28R null mice following four hour infection with USA300 (Fig. 4A). By 18 hours following the initial infection miR-21 levels in WT mice returned to uninfected levels, while levels in IL-28R null mice remained significantly higher (p = 0.002). PDCD4 mRNA levels following 18 hour infection inversely correlated with miR-21 expression, and were significantly lower in IL-28R null mice compared to WT (p = 0.045) (Fig. 4B). PDCD4 protein expression at the 4 hour time point was reduced in both WT and IL-28R−/− mice compared to unstimulated levels, and reduced further at the 18 hour time point (Fig. 4C). PDCD4 protein expression tended to be lower at the 18 hour time point in IL-28R null mice than WT, correlating with elevated levels of miR-21. Thus it appears that following the initial pro-inflammatory innate immune response to S. aureus in WT mice, IFNλ signaling suppresses expression of miR-21 maintaining PDCD4 levels. Loss of IL-28R resulted in sustained miR-21 levels and a decrease in PDCD4 gene expression by 18 hours of S. aureus infection.
To confirm that PDCD4 expression negatively influences clearance of USA300 from the airway we infected WT and PDCD4−/− mice with 107 CFU USA300 for 18 hours. Significantly fewer organisms were recovered from both the airway (p = 0.007) (Fig. 4D) and lung tissue (p = 0.014) (Fig. 4E) of PDCD4−/− mice compared to WT. Bacterial clearance correlated with reduced lung pathology (Fig. 4F) but not immune cell recruitment into the lung (Fig. S3). Expression of the inflammatory cytokines KC and IL-1β, not TNF, were reduced and expression of anti-inflammatory IL-10 was increased in PDCD4−/− mice compared to WT. Although differences in cytokines were not statistically significant, the trends were similar to those observed in IL-28R−/− compared to WT suggesting a link between type III IFN and PDCD4.
Due to previous reports demonstrating improved survival of PDCD4−/− mice during LPS challenge, we hypothesized that IL-28R−/− mice would clear the Gram negative pathogen P. aeruginosa PAK more rapidly than WT mice [31]. In vitro, P. aeruginosa PAK stimulation (MOI 10, 4 hours) of BMDCs induced increased mRNA levels of IFNλ compared to unstimulated cells (Fig. S1C). As shown for MRSA in Figure S1A, loss of IL-28R resulted in reduced IFNλ mRNA levels following P. aeruginosa stimulation.
WT and IL-28R null mice were intranasally infected with 107 CFU PAK and bacterial clearance from the airway and lung tissue monitored at 4 and 18 hours of infection. In vivo, IFNλ mRNA levels (p = 0.0031) (Fig. S1D) and protein expression (p = 0.005) (Fig. 5A) were significantly increased in WT mice by 4 hours post infection, but were not different than baseline at the 18 hour time point. Significantly fewer bacteria were recovered from BAL (p = 0.0003) (Fig. 5B) and lung tissue (p = 0.0023) (Fig. 5C) of IL-28R−/− at the 18 hour time point compared to WT, and no differences were observed at the 4 hour time point. Bacterial clearance was not altered in mice lacking the type I IFN receptor IFNAR (Fig. 5D). Pathology in the lung was reduced in knockout mice following PAK infection (p = 0.0010) (Fig. 5E). Similar to S. aureus infection, there were no differences in the numbers of immune cells recruited to the BAL or lung (Figs. 5F and 5G).
BAL fluid was analyzed for the presence of pro- and anti-inflammatory cytokines to determine if IFNλ signaling increased expression of pro-inflammatory cytokines during PAK infection. Significant reductions in expression of pro-inflammatory cytokines KC (p = 0.0141) and TNF (p = 0.0006) were observed in BAL of IL-28R null mice at 18 hours compared to WT (Fig. 6A). The anti-inflammatory IL-10 was significantly increased (p = 0.0134) in the IL-28R−/− mice as compared to controls at this time point. No differences in IL-1β production were observed. MX1 levels were lower in IL-28R null mice at 18 hours compared to WT (Fig. 6B).
We then determined if the reduced pro-inflammatory cytokine expression in IL-28R−/− mice infected with PAK was also due to alterations in miR-21 and PDCD4. In vitro, PDCD4 siRNA significantly reduced IL-8 expression (p = 0.0069) in 16HBE in response to PAK (Fig. 6C). MiR-21 expression in lung tissue of IL-28R null mice was increased (p = 0.0122) compared to WT at 18 hours, demonstrating that loss of IFNλ signaling resulted in increased expression of this miRNA during PAK infection (Fig. 6D). PDCD4 mRNA levels were significantly decreased in IL-28R null mice compared to WT correlating with the changes in miR-21 level (p = 0.0323) (Fig. 6E). PDCD4 protein in the lung was almost undetectable at the 4 hour time point. PDCD4 levels increased slightly by 18 hours although no differences were observed between knockout and WT mice (Fig. 6F).
PDCD4 function can be regulated by p70S6K dependent phosphorylation at serine 475, which inhibits its interaction with AP-1 and eukaryotic translation initiation factor 4A (eIF4A) transcription factors and leads to ubiquitin dependent degradation [22], [33], [44]-[46]. In vitro stimulation of 16HBE monolayers with recombinant IFNλ for 1 hour reduced phosphorylation of p70S6K, suggesting that type III IFN inhibits this regulatory pathway (Fig. 6G). In vivo we observed phosphorylation of PDCD4 in uninfected IL-28R−/− mice compared to WT control. In both WT and knockout mice phosphorylation of p70S6K increased following 18 hour PAK infection (Fig. 6H), not observed during MRSA infection (Fig. 6I). Phosphorylation of PDCD4 in PAK infected WT mice also increased following 18 hours, correlating with phosphorylation of p70S6K. These data suggest that modifications to PDCD4, mediated by both miR-21 and p70S6K, regulate its ability to influence cytokine production during P. aeruginosa pneumonia.
Much like the other members of the IFN family, IFNλ is required for host defense against viral pathogens and is capable of inhibiting tumor growth [23]–[28], [30], [47], [48]. There is clearly a major role for IFNλ in the successful clearance of the hepatitis C virus consistent with expression of IL-28R on hepatocytes [49], [50]. Type III IFN has also been linked to killing of the intracellular bacterial pathogen Listeria monocytogenes, correlating with reduced colonization of the spleen and liver [51]. Fitting with the abundance of IL-28R in the respiratory tract, we found a major role for IFNλ signaling in the pathogenesis of both S. aureus and P. aeruginosa pneumonia [21], [28].
In contrast to the beneficial effects of IFNλ in the eradication of viral pathogens, our data suggest that IFNλ promotes a prolonged inflammatory cytokine response to bacterial pathogens that detracts from the efficiency of bacterial clearance and promotes pathology. Mice lacking the type III IFN receptor demonstrated significantly improved clearance of S. aureus and P. aeruginosa coincident with a reduced duration of the host inflammatory response. Particularly in the pathogenesis of pneumonia, the balance between pro and anti-inflammatory signaling is critical. While the mechanism through which the inflammatory milieu affects bacterial clearance is unclear, our data clearly demonstrate that reducing levels of inflammatory cytokines through modulation of IFNλ or inhibition of IL-1R improves clearance of S. aureus and P. aeruginosa [11]. Once a sufficient number of phagocytes are recruited to the airways to deal with the pathogens, excessive cytokine production, particularly toxic cytokines such as IL-1β, elicits tissue damage and impairs normal host defenses.
A major effector of IFNλ signaling is PDCD4, which appears to be regulated by multiple pathways in response to infection. A confirmed target of miR-21, PDCD4 influences production of pro-inflammatory signaling through its interaction with AP-1, NF-κB, and eIF4A, and has been linked to the inflammatory microenvironment surrounding tumor cells [10], [12], [29], [31], [33], [52]. Similar to previous findings with LPS, S. aureus SpA, through its interaction with TNFR1 upregulates PDCD4 in epithelial cells, promoting pro-inflammatory signaling while inhibiting anti-inflammatory cytokine production. It is likely that additional S. aureus pathogen associated molecular patterns (PAMPs) such as the lipoproteins that activate TLR2 also participate.
Following ligation of IL-28R by IFNλ, STAT3 suppresses expression of miR-21 resulting in increased expression of PDCD4. MiRNAs like miR-21 comprise a regulatory system which makes subtle changes to mRNA expression levels, including the genes involved in Toll-like receptor signaling, and therefore are capable of having significant affects on innate immunity [7]–[9], [31], [53]. MiR-21, specifically, is induced by NF-κB signaling and has been implicated in the regulation of TLR2 mediated signaling in both skin and lung inflammation [33], [54]–[57]. We similarly observe rapid increases in miR-21 during the initial 4 hour infection with either S. aureus or P. aeruginosa, and sustained expression of miR-21 in IL-28R−/− mice was associated with reduced levels of inflammatory cytokines in the airway. As predicted by our in vitro studies, in mice expressing IL-28R reduced levels of miR-21 at 18 hours post infection correlated with prolonged PDCD4 mRNA expression. Mice lacking PDCD4 were better able to clear USA300 from their lungs, similar to previous findings demonstrating PDCD4 null mice are resistant to lethal LPS challenge [31]. Therefore mice lacking IL-28R are less susceptible to S. aureus and P. aeruginosa induced lung damage due in part to reduced PDCD4 and a shortened inflammatory response.
A second pathway acting through mTOR and p70S6K regulates PDCD4 function by phosphorylating PDCD4, inhibiting its interaction with AP-1 and eIF4A and promoting its degradation [17], [19], [22], [29]. Our data suggest that type III IFN has an inhibitory affect on p70S6K and PDCD4 phosphorylation. In vitro IFNλ treatment of epithelial cells reduced p70S6K phosphorylation. In vivo, PDCD4 was strongly phosphorylated in uninfected IL-28R−/− mice while phosphorylation was not observed in uninfected control mice. Rapid turnover of PDCD4 due to phosphorylation dependent degradation, or an inability of PDCD4 to bind eIF4A in IL-18R null mice could further limit its affects on signaling.
Increases in p70S6K phosphorylation during bacterial infection were IL-28R independent and were only observed 18 hours following infection with P. aeruginosa. The ability to activate different regulatory pathways such as p70S6K signaling could contribute to the differences in cytokine induction observed during S. aureus and P. aeruginosa infections. For example levels of TNF and IL-10 were significantly different between WT and IL-28R−/− mice during PAK not USA300 infections, whereas IL-1β was significantly different during a S. aureus infection. Thus IFNλ and PDCD4 are conserved elements in the response to airway pathogens which clearly have other differential effects in inducing innate immunity.
The selective expression of IL-28R, substantially more abundant on murine and human epithelial cells than on immune cells as previously shown and demonstrated by the inability of macrophages to respond to IFNλ (Fig. S4), may also be important in the specific cytokine responses evoked by the different pathogens [28]. Epithelial cells are not the primary producers of IL-1β in response to PAK, therefore IL-1β levels in the airways of P. aeruginosa infected WT and IL-28R−/− mice were not different [11], [58]. It is unclear which cells in the lung produce IL-1β in response to S. aureus. Our data suggest that IL-1β producing cells express IL-28R, as cytokine levels were significant reduced in IL-28R null mice. While we do not propose that IL-1β is the sole cytokine responsible for lung damage in response to S. aureus, inhibition of this signaling pathway in WT mice does improve clearance of S. aureus and limits the extent of tissue damage. We conclude that while the specific cytokines modulated by type III IFN are different in the context of specific bacterial infections, the overall affect of inhibiting IL-28R dependent signaling is a reduction of lung inflammation and improved bacterial clearance.
The type III IFN pathway is recognized as an important host defense pathway for the eradication of viral infection and tumor growth and the distribution of the receptor at mucosal sites makes it an active participant in the innate immune response to bacterial pathogens as well. While differences in the expression of PAMPs and their receptors, mechanisms of p70S6K and PDCD4 phosphorylation, and ultimately levels of cytokine expression are evident in the models of S. aureus and P. aeruginosa pneumonia, IFNλ functions as a central regulator of airway inflammation in both types of bacterial pneumonia. Given the pathology associated with excess cytokine induction in the airways, targeting specific components of the type III IFN/PDCD4 regulatory pathway could be a potential therapy for limiting lung damage in the setting of acute airway infection.
Animal work in this study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, the Animal Welfare Act and US federal law. The protocol was approved by the Institutional Animal Care and Use Committee (IACUC) of Columbia University (protocol number AAAC3059).
S. aureus strain LAC USA300 and P. aeruginosa strain PAK were grown on Luria-Bertani agar at 37°C. For infection, LB broth was inoculated with single colonies and grown overnight at 37°C, diluted 1∶100 in the morning and grown to OD 1.000 (S. aureus) or 0.500 (PAK). S. aureus cultures grown to OD 1.000 were incubated at 55°C to inactivate the bacteria for heat killed experiments.
Seven week old C57BL/6 WT, IL-28R−/−, IFNAR−/−, and PDCD4−/− mice were intranasally inoculated with 50 µl S. aureus or P. aeruginosa (1*107 CFU/mouse) as previously described [2], [14]. WT and IFNAR−/− mice were previously described, IL-28R−/− mice were provided by Bristol Myers Squibb, and PDCD4−/− mice were from Jackson Laboratories [13]. Control mice received 50 µl of PBS. To determine the affect of IFNλ on cytokine production, mice were intranasally treated with 1 µg recombinant mIL-28 (PBL Interferon Source) in 50 µl PBS or PBS control. The role of IL-1β was determined in mice pretreated i.p. with Anakinra (10 mg/kg, Amgen) for four days and were infected on the fourth day as described previously [59]. Control mice were treated with PBS also given i.p. BAL fluid was harvested 18 hours after infection and used to quantify immune cell populations, cytokine expression, and bacteria CFU. Histology was performed by Columbia University Molecular Pathology core on tissues fixed in 4% paraformaldehyde.
Bone marrow- derived dendritic cells (BMDCs) were cultured from wild-type and knockout mice as described previously [3]. Human airway epithelial cells (16HBE) were grown as previously described [60]. Human monocytic THP-1 cells were grown in RPMI 1640 medium with 10% fetal bovine serum with penicillin (100 units/ml) and streptomycin (100 µg/ml) and treated with 100 nM phorbol 12-myristate 13-acetate (PMA) to induce terminal differentiation for IFN stimulation studies. On-target control or PDCD4 siRNA (Thermo Scientific) was transfected into 16HBE monolayers using Fugene (Promega) according to the manufacturers instructions. Cells were stimulated with 5 µM S. aureus protein A (Calbiochem) or 0.1 µg/ml human recombinant IL-28 or IFNβ (PBL Interferon). In select experiments antibody to the extracellular domain of TNFR1 or control IgG (4 µg/ml) (Santa Cruz Biotechnology Inc.) or STAT3 inhibitor V (50 µM) (Calbiochem) or DMSO control were applied to 16HBEs 1 hour prior to SpA.
Enumeration of neutrophil (MHCII−Ly6+), macrophages (CD11c+MHCIIlow), and DC (CD11c+MHCII+) populations was performed as previously described [1].
BAL was analyzed for cytokine and chemokine content by ELISA (R&Dbiosystems, PBL Interferon Source, or eBioscience) according to the manufacture’s instructions. Anti-phospho-p70S6K, anti-phospho-PDCD4, anti-PDCD4 (Rockland Immunochemicals), anti-MX1 (Santa Cruz Biotechnology Inc.) and anti-β-actin (Sigma) antibodies followed by secondary antibodies conjugated to horseradish peroxidase (Santa Cruz Biotechnology Inc.) were used to measure expression in human epithelial cells and mouse lung. Protein separation, transfer and immunnoblotting were performed as described [13]. Densitometry was done using Image J (NIH).
Total RNA was isolated using mirVana miRNA Isolation Kit (Life Technologies) according to the manufacturers instructions. For mRNA analysis cDNA was generated using the High Capacity cDNA reverse Transcriptase Kit (Applied Biosystems). Quantitative real-time RT-PCR (QRT-PCR) was performed using Power SYBR Green PCR Master Mix in a Step One Plus Thermal Cycler (Applied Biosystems). Mouse primers for PDCD4 were 5′ – ATGGATATAGAAAATGAGCAGAC -3′ and 5′ – CCAGATCTGGACCGCCTATC -3′ and actin were 5′ – CCTTTGAAAAGAAATTTGTCC – 3′ and 5′ – AGAAACCAGAACTGAAACTGG – 3′. Human primers for IL-8 were 5′-TACTCCAAACCTTTCCAACCC-3′ and 5′- AACTTCTCCACAACCCTCTG-3′ and actin were 5′- GTGGGCCGCTCTAGGCACCA-3′ and 5′-CGGTTGGCCTTAGGGTTCAGGGGGG- 3′. IL-8 and PDCD4 expression was normalized to actin. For miRNA analysis cDNA was generated and qRT-PCR was run using NCode miRNA First-Strand cDNA Synthesis and qRT-PCR Kit (Life Technologies) according to the manufacturers instructions. A universal reverse primer for small RNAs was supplied with the NCode kit. The forward primer for miR-21 was 5′-TAGCTTATCAGACTGATGTTGA-3′ and ΔCt values were normalized to the small non-coding RNA U6 5′-GGGCAGGAAGAGGGCCTAT-3′.
Significance of data was determined using a nonparametric Mann-Whitney test. For experiments with greater than one comparison we used a nonparametric Kruskal-Wallis test followed by post-hoc Dunn’s test to correct for multiple comparisons. Statistics were performed with GraphPad Prism software with significance defined as p<0.05. Cytokine, bacterial counts, and immune cell number data are presented as individual points with a bar representing the median value.
|
10.1371/journal.pcbi.1005086 | Likelihood-Based Inference of B Cell Clonal Families | The human immune system depends on a highly diverse collection of antibody-making B cells. B cell receptor sequence diversity is generated by a random recombination process called “rearrangement” forming progenitor B cells, then a Darwinian process of lineage diversification and selection called “affinity maturation.” The resulting receptors can be sequenced in high throughput for research and diagnostics. Such a collection of sequences contains a mixture of various lineages, each of which may be quite numerous, or may consist of only a single member. As a step to understanding the process and result of this diversification, one may wish to reconstruct lineage membership, i.e. to cluster sampled sequences according to which came from the same rearrangement events. We call this clustering problem “clonal family inference.” In this paper we describe and validate a likelihood-based framework for clonal family inference based on a multi-hidden Markov Model (multi-HMM) framework for B cell receptor sequences. We describe an agglomerative algorithm to find a maximum likelihood clustering, two approximate algorithms with various trade-offs of speed versus accuracy, and a third, fast algorithm for finding specific lineages. We show that under simulation these algorithms greatly improve upon existing clonal family inference methods, and that they also give significantly different clusters than previous methods when applied to two real data sets.
| Antibodies must recognize a great diversity of antigens to protect us from infectious disease. The binding properties of antibodies are determined by the DNA sequences of their corresponding B cell receptors (BCRs). These BCR sequences are created in naive form by VDJ recombination, which randomly selects and trims the ends of V, D, and J genes, then joins the resulting segments together with additional random nucleotides. If they pass initial screening and bind an antigen, these sequences then undergo an evolutionary process of reproduction, mutation, and selection, revising the BCR to improve binding to its cognate antigen. It has recently become possible to determine the BCR sequences resulting from this process in high throughput. Although these sequences implicitly contain a wealth of information about both antigen exposure and the process by which we learn to resist pathogens, this information can only be extracted using computer algorithms. In this paper we describe a likelihood-based statistical method to determine, given a collection of BCR sequences, which of them are derived from the same recombination events. It is based on a hidden Markov model (HMM) of VDJ rearrangement which is able to calculate likelihoods for many sequences at once.
| B cells effect the antibody-mediated component of the adaptive immune system. The antigen-binding properties of B cells are defined by their B cell receptor, or BCR. BCRs bind a wide variety of antigens, and this flexibility arises from their developmental pathway. B cells begin life as hematopoietic stem cells. After a number of differentiation steps the cells perform somatic recombination, or rearrangement. For the heavy chain locus, a V gene, D gene, and J gene are randomly selected, trimmed some random amount by an exonuclease, and then joined together with random nucleotides (forming so-called N-regions). The light chain process is slightly simpler, in that only a V and J recombine, but proceeds via similar trimming and joining processes. These processes form the third complementarity determining region (CDR3) in each of the heavy and the light chain, which are important determinants of antibody binding properties. Then a series of checkpoints on the BCRs ensure that the resulting immunoglobulin is functional and not self-reactive through negative selection (reviewed in [1]). This process results in naive B cells with fully functioning receptors. When stimulated by binding to antigen in a germinal center, naive cells reproduce and mutate by via the process of somatic hypermutation, and then are selected on the basis of antigen binding and presentation to T follicular helper cells [2]. This process is called affinity maturation. It is now possible to sequence B cell receptors in high throughput, which in principle describes not only the collections of antigens to which the immune system is ready to react, but also implicitly narrates how they came to be.
It is of great practical interest for researchers to be able to reconstruct events of this development process using BCR sequence data. Such reconstruction would shed light on the process of B cell receptor maturation, a subject of continual study since the landmark work of Eisen and Siskind in 1964 [3, 4]. Furthermore, there are specific maturation pathways of great importance, such as the B cell lineages leading to broadly neutralizing antibodies to HIV [5, 6]. Being able to reconstruct the structure and history of these lineages allows investigation of the binding properties of these intermediates, which could be helpful to design effective vaccination strategies to elicit high-affinity antibodies [7]. For example, recent studies have shown the promise of a sequential immunization program for eliciting these antibodies [8]; lineage reconstruction will aid in identifying desirable intermediate BCRs.
The clonal family inference problem is an intermediate step to such lineage reconstruction (Fig 1). Rather than trying to reconstruct the full lineage history of the set of sequences, the goal is only to reconstruct which sequences came from the same rearrangement event. Full lineage reconstruction would also require building phylogenetic trees for each of the clonal families. However, these clonal families can be an object of interest themselves [9].
The motivation behind our approach to the clonal family inference problem, like many before us, is to use the special structure of BCR sequences (which for simplicity we describe for the heavy chain; the same concepts and approaches can be applied to the light chain). This structure follows from VDJ recombination and affinity maturation: for example, by definition the identity of the germline genes cannot change through affinity maturation. Thus, if the per-read germline gene identity could be inferred without error, then any pair of sequences from a clonal family must have the same inferred germline gene identity. If one also assumes that sequences evolve only through point mutation, then sequences must have identical-length CDR3s if they are to be in the same clonal family.
Most current methods for B cell clonal family inference make these assumptions, and proceed by first stratifying sequences by inferred V and J germline genes and CDR3 length, then only consider pairs of sequences within a stratum as potential members of the same clonal family. If one assumes further that any clonal families with pairs of highly diverged sequences also contain intermediates between those sequences, one might assume that there is a path between any pair of sequences such that neighboring sequences in a path are similar. This suggests a strategy in which pairs of sequences that are similar at some level (such as 90% similar in terms of nucleotides) in the CDR3 are considered to be in the same clonal family, and where membership is transitive, which corresponds to an application of single-linkage clustering.
Instead of designing such an algorithm that works only when a set of rigid, predefined assumptions are satisfied, an alternative is to formalize a model of B cell affinity maturation into a generative probabilistic process with a corresponding likelihood function. Once this likelihood function is defined, one can infer clonal families by finding the clustering that maximizes the likelihood of generating the observed sequences.
Likelihood methods in the form of a hidden Markov models (HMM) have been applied to B cell receptor sequences for a decade [10–13]. This previous work has been to use HMMs to analyze individual sequences. For likelihood-based clustering we are only aware of the work of Laserson [14, 15], who uses Markov chain Monte Carlo to infer clusters via a Dirichlet mixture model (reviewed in [16]). Unfortunately the Laserson algorithm is only described in a PhD thesis and does not appear to be publicly available. In related work, Kepler [17, 18] uses a likelihood-based phylogenetics framework to perform joint reconstruction of annotated ancestor sequence and a phylogenetic tree.
In this paper we present a method for inferring clonal families in an HMM-based framework that comfortably scales to tens of thousands of sequences via parallel algorithms, with approximations that scale to hundreds of thousands of sequences. For situations in which specific lineages are of interest, users can specify “seed” sequences and find the clonal family containing that seed in repertoires with one million sequences. Our clustering algorithm is based on a “multi-HMM” framework for BCR sequences that we have previously applied to the annotation problem: to infer the origin of each nucleotide in a BCR (or TCR) sequence from the VDJ rearrangement process [19]. We use this framework to define a likelihood ratio comparing two models which differ by the collapse of two clonal families into one, and use it for agglomerative clustering. Because this likelihood ratio comes from an application of the forward algorithm for HMMs, it integrates out all possible VDJ annotations. We find that it outperforms previous algorithms on simulated data, and that it makes a significant difference when applied to real data.
In order to calculate a set of probabilities suitable for use in the clonal family inference problem, we begin with the HMM framework introduced in [19]. In that paper we focused on inferring parameters of an HMM and using it to obtain BCR annotated ancestor sequences, which was primarily based on the most likely path through each HMM, i.e. the Viterbi path. We also described Viterbi annotation with a multi-HMM, i.e. annotation using a collection of sequences that were assumed to form a clonal family.
In this application, we will use the forward algorithm for HMMs [20] to obtain the corresponding marginal probability, which is the sum of sequence generation probabilities over all possible paths through the HMM. This is a more appropriate tool for the clonal family inference problem because here we are interested in integrating over annotated ancestor sequences (that is, paths through the HMM) to decide whether sequences are related. By using a multi-HMM, we can use this total probability to calculate a likelihood ratio that two clusters derive from the same, or from different, rearrangement events. We perform agglomerative clustering using this likelihood ratio to group sequences for which the probability of a common ancestry is higher than that of separate ancestry (details in the Methods). This approach allows us to calculate the total probability of the partition (i.e. clustering) at each stage in the clustering process, which provides both an objective measure of partition quality, and easy access to not only the most likely partition but also to a range of likely partitions of varying degrees of refinement. As in our previous work, the parameters of the HMM can be inferred “on the fly” given a sufficiently large data set or be inferred on some other data set. Briefly, we do a cycle of Viterbi training, which is started with an application of Smith-Waterman alignment, in which the best annotation for each sequence with a current parameter set is used to infer parameters for the next cycle. As described in detail elsewhere [19], data is aggregated if there are insufficient observations for a given allele for training.
In addition to this principled method for full-repertoire reconstruction, we have implemented two more approximate versions which trade some accuracy for substantial increases in speed. In the first, which we call point partis, we forgo integration over all possible annotated ancestor sequences and instead find the most likely naive sequence point estimate for each cluster. Clusters are then compared based on the Hamming fraction (Hamming distance divided by sequence length) between their respective naive sequences, and are merged if the distance is smaller than some threshold. This threshold is set dynamically based on the observed mutation rate in the sample at hand.
In order to achieve further improvements in speed, we can also avoid both complete all-versus-all comparison of the sequences at each step, and calculation of the joint naive sequence for each merged cluster. For this we find the most likely naive sequence for each individual sequence, and then pass the results, together with a dynamically-set clustering threshold, to the clustering functionality of the vsearch program [21]. We call this vsearch partis.
We have also included a method which, using the full likelihood, reconstructs the clonal family containing a given “seed” sequence. Because clonal families are generally significantly smaller than the total repertoire, this option is much faster than the full-repertoire reconstruction methods. We see this option as being useful when specific sequences are identified as interesting through a binding assay or because they are shared between repertoire samples. This is labeled full partis (seed).
This clustering has been implemented as part of continued development of partis (http://github.com/psathyrella/partis). As before, the license is GPL v3, and we have made use of continuous integration and containerization via Docker for ease of use and reproducibility [22]. A Docker image with partis installed is available at https://registry.hub.docker.com/u/psathyrella/partis/.
In the absence of real data sets with many sequences for which the true annotations and lineage structures are known, we compare these new clustering methods against previous methods using simulated sequences generated as described in [19]. These simulations were done for the heavy chain locus only. We performed comparison both on samples, which we call 1×, which mimic mutation frequencies in data (overall mean frequency of about 10%) and on samples, which we call 4×, with quadrupled branch lengths (overall mean frequency of about 25%) to explore results in a more challenging regime. Per-sequence mutation frequencies are distributed according to the empirical distribution (see [19]). We compare the three partis methods to three methods from the literature. The first, labeled “VJ CDR3 0.9”, is representative of annotation- and distance-based methods which have been used in a number of papers [18, 23–26]. It begins by annotating each individual sequence, and proceeds to group sequences which share the same V and J gene and the same CDR3 length, and have CDR3 sequence similarity above some threshold, which is commonly 0.9 [24]. For this comparison we use partis annotation; for a comparison of annotation methods themselves see [19]. We also compare against Change-O’s clustering functionality [27] fed with annotations from IMGT, with IMGT failures (when it does not return an annotation) classified as singletons. We perform a partial comparison against MiXCR [28]. Since this method does not currently report which sequences go into which clusters, and instead only reports cluster summary statistics, we cannot perform a detailed evaluation. The authors of MiXCR note in personal communication, however, that they plan to report this information in future versions.
We use per-read averages of precision and sensitivity to quantify clustering accuracy. In this context, the precision for a given read is the fraction of sequences in its inferred cluster which are actually in its clonal family, while sensitivity for a given read is the fraction of sequences in its true clonal family that appear in its inferred cluster (details in Methods). We find that partis is much more sensitive than previous methods, at the cost of some loss of precision (Fig 2). The point partis approximate implementation is less specific than the full implementation, while the even faster vsearch approximation loses some precision and some sensitivity.
We investigate these differences in more detail for the first simulation replicate via an intersection matrix with entries equal to the size of the intersection between each of the 40 largest clusters returned by pairs of algorithms (Figs 3 and 4, S6, S7, S8 and S9 Figs). Full partis infers clonal families correctly the majority of the time at typical mutation levels, and in this experiment it incorrectly split a cluster of true size around 45. These results degraded somewhat with the point partis approximation, and somewhat more with the vsearch approximation. The VJ CDR3 0.9 method consistently under-clustered for the largest cluster sizes. The seeded full partis method correctly reconstructed the lineage of interest starting from a randomly sampled sequence, while ignoring all others.
In order to understand performance on the many smaller clusters and to get a simpler overall picture, we also compared cluster size distributions for the various methods with the simulated distribution (Figs 5 and 6, S1 and S2 Figs). Here we can see that partis is able to accurately infer the true cluster size in a variety of regimes, whereas other methods tend to under-merge clusters of all sizes.
In order to further understand the source of these differences, we also compare results against two methods of generating incorrect partitions starting from the true partition, which we call synthetic partitions (S3 and S4 Figs). The first, called synthetic 60% singleton is generated from the true partition by splitting 60% of the sequences into singleton clusters. The second, called synthetic neighbor 0.03, merges together true clonal families which have true naive sequences closer than 0.03 in Hamming distance divided by sequence length. We find that the performance of synthetic 60% singleton tracks that of the VJ CDR3 method, while the performance of synthetic neighbor 0.03 tracks that of partis.
Finally, to investigate the performance of the seeded full partis method, we calculate the precision and sensitivity of this method on a number of widely varying sample sizes (Fig 7). For these simulations we used a Zipf (power-law) distribution of cluster sizes with exponent 2.3, and randomly selected one seed sequence from a randomly selected large cluster. We find that seeded partis frequently obtains very high sensitivity, although precision decreases as sample size increases. This precision decrease is from incorrect merges of clusters. We have manually checked these incorrect merges, and found that the true (i.e. simulated) naive sequences of clusters which are incorrectly merged with the seeded cluster typically differ by one to six bases. Because these differences occur either within the bounds of the true eroded D segment, or within the true non-templated insertions, it is difficult to distinguish them from somatic hypermutation. This echoes the observation that partis precision is driven by the presence or absence of clusters which stem from different rearrangements, but which are very similar in naive sequence (compare partis and synthetic neighbor 0.03 in S5 Fig).
In order to handle insertion-deletion (indel) mutations which occur during somatic hypermutation, we have implemented a heuristic method in the preliminary Smith-Waterman alignment step in partis. In short, this works by “reversing” inferred indel mutations in germline-encoded regions and proceeding with the clustering algorithm. We find that partis performance is typically unaffected when indels occur in non-CDR3 germline-encoded regions, although performance suffers when indels occur in the CDR3 (Fig 8). This is because indel mutations in the CDR3 are quite difficult to distinguish from insertions and deletions stemming from the VDJ rearrangement process using indel-handling schemes (such as ours) that only take one sequence at a time into account.
In order to understand the difference this method makes on real data, we applied partis and the other algorithms to subjects in the Adaptive data set from [29] used in previous publications [19, 30, 31], as well as the data set from [24], which we will call the “Vollmers” data set. These data sets were Illumina sequenced via amplicons covering the heavy chain CDR3, and thus do not have complete V or J sequences. Especially in the case of the V region for the Vollmers data, it is not possible to confidently identify the germline V gene for each of the BCR sequences. Thus, these data sets make for an interesting comparison between methods (such as VJ CDR3) which require single germline gene identifications, to our method, which integrates over such identifications. Results are shown for Adaptive subject A (Fig 9), and for a subject from the Vollmers data set (Fig 10). The rest may be found on figshare at http://figshare.com/s/9b85e4ac54d011e5bd3e06ec4b8d1f61. Note that the identifiers shown for the Vollmers data are an obfuscated version of the original identifiers in the data; contact the authors for more details. These results are not presented to make any strong statement about the true cluster size distribution, the correctness of which cannot be be independently evaluated, but rather to show that the partis results are different from those of other methods on real data, as seen under simulation.
When we applied the various methods to a randomly chosen set of 20,000 sequences from two different sets, we found that the various methods agree that both samples are dominated by singletons, but there is substantial discord at the high end of the distribution, especially in Adaptive subject A (Figs 9 and 10). These differences in composition are examined in more detail using cluster intersection matrices. The cluster size distribution inferred by partis approximately follows a power-law, with exponent about 2.3.
Adaptive subject A (Fig 9) has mutation levels two and a half times higher than Vollmers subject 15-12 (Fig 10), making inference more challenging for A. Both of these data sets consist of shorter sequences than the simulated sequences, which contain the entire V and J regions. Reads in the Adaptive samples are 130 base pairs (losing about two thirds of the V and one half of the J), while those in the Vollmers data set vary in length, but typically span all of the J but only 20 to 30 bases in the V.
Likelihood-based clustering using partis is computationally demanding, though within a range applicable to real questions given appropriate computing power (Fig 11). On a computing cluster with about 25 8-core machines, full and point partis can cluster ten thousand sequences in 4 to 7 hours, while vsearch partis can cluster one hundred thousand sequences in 4 hours. Our implementation of “VJ CDR3 0.9” used partis annotation, but this approach could be made much faster by using a fast method for annotation [28, 32]. Time required can also vary by an order of magnitude depending on the structure of the sample (cluster size and mutation level).
We have developed an algorithm to infer clonal families using a likelihood-based framework. Although the framework does take annotation information into account by using a VDJ-based HMM, the algorithm is distinguished from other clustering methods in that it does not fix a single annotation first and then use that annotation for downstream steps. Instead we find that by integrating over annotated ancestor sequences using an HMM, we are able to obtain better clonal family inference than with the current common practice of rigidly inferring VJ annotation and then clustering on HCDR3 identity for heavy chain sequences. Our simulations show that existing algorithms frequently do not sufficiently cluster sequences which sit in the same clonal family. Our application to real data shows that the partis algorithms using our default clustering thresholds return more large clusters on two real data sets, indicating that this difference in clustering is not simply an artifact of our simulation setup.
The performance differences between our various approximate algorithms indicates the sources of the partis’ improved performance. The reasonably good performance of the point partis variant shows the importance of clustering on inferred naive sequences rather than observed sequences and inferring these naive sequences with an accurate probabilistic method. Furthermore, the difference between point and full partis is some measure of the importance of integrating out uncertainty in annotated ancestor sequences.
We find that partis’ main weakness is in separating out clusters with highly similar naive sequences. Indeed, its performance tracks a simulated method that merges clonal families with true (i.e. simulated) naive sequences that are closer than 3% in nucleotides, in simulations with about 10% divergence from the naive sequence. Although the VDJ rearrangement process generates a very diverse repertoire, biases in gene family use and other rearrangement parameters mean that pairs of highly similar naive sequences are frequently generated. This may indicate an inherent limitation in clonal family inference methods that only use data from heavy chain.
Our method builds on previous work for doing likelihood-based analysis of BCR sequences. In particular, we are indebted to Tom Kepler for initiating the use of HMMs in BCR sequence analysis [10] and for developing likelihood-based methods to infer unmutated common ancestor sequences while integrating over rearrangement uncertainty [17, 18, 33–35].
We did not compare to several related methods that have been described in the literature. ClonalRelate [36] is an extension to the “VJ CDR3” method that allows some flexibility in requiring V and J calls to be the same by combining various mismatch penalties into a distance that is used for agglomerative clustering. IMSEQ [32] is a recent method which is reported to be quite fast; however the current version appears mainly aimed toward T cell receptors, as it does not handle somatic hypermutation. As it clusters based on V and J genes and 100% CDR3 similarity, it is equivalent to the annotation-based method described above, except with a threshold inappropriate to B cells. Cloanalyst performs joint reconstruction of annotated ancestor sequence and a phylogenetic tree given a collection of sequences assumed to form a clonal family [17]. Immunitree apparently uses a Dirichlet process mixture model for clustering, however, the algorithm is only fully described in a PhD thesis [14], and does not appear to be publicly available (note that https://github.com/laserson/vdj performs straightforward single-linkage clustering and is in fact written by a sibling of the Immunitree author). IgSCUEAL [37] is a recent method that performs annotation and clustering using a phylogenetic approach. Its clustering algorithm, however, is not part of the public distribution and is apparently undergoing revision.
There are several opportunities to improve partis. First, our current approach requires likelihood ratios to exceed a value based on cluster size; these cluster sizes are based on observing distributions of likelihood ratios under simulation. A more principled approach would be preferable. Second, our approach to insertion-deletion mutations in affinity maturation only uses one sequence at a time. Thus it has an inherent difficulty differentiating between mutations in the course of affinity maturation versus insertion-deletion events that are part of VDJ rearrangement. Third, our current code is only for the heavy chain alone or the light chain chain alone. Extending the work to paired heavy and light chain BCR data is conceptually straightforward, although will require additional software engineering. Fourth, HMMs have certain inherent limitations, stemming from the central Markov assumption that the current state is ignorant of all states except for the previous one. As reviewed in [19], this limits the scope of events that can be modeled using partis, excluding correlation between different segments of the BCR [31, 38, 39], palindromic N-additions [40], complex strand interaction events [41, 42], or the appearance of tandem D segments [43]. Some of these limitations could be avoided by using Conditional Random Fields (reviewed in [44]), and although linear-chain conditional random fields enjoy many of the attractive computational properties of HMMs, this flexibility will come with a computational cost. Fifth, partis does not attempt to infer germline genotype, as do [45], and so treats genes and alleles on an equal footing. We will treat this as a model-based inference problem in future development. Sixth, we will continue to refine heuristics to provide the accuracy of the full likelihood-ratio calculation with minimal compute time. We note, for instance, that a small decrease in the lower naive Hamming fraction threshold substantially improves performance for the seed partis simulation compared to that shown here (in Fig 7).
In additional future work, we will explore opportunities to combine clonal family inference and phylogenetics to obtain inference of complete B cell lineages. This could potentially take the form of a phylo-HMM [46], although a more straightforward approach would be to take the product of a phylogenetic likelihood and a rearrangement likelihood [17]. For example, one might use HMM-based clustering as is described here with a high likelihood ratio cutoff to obtain a conservative collection of clusters, and then a phylogenetic criterion to direct further clustering.
In addition to these methodological improvements, we will also apply partis to a variety of data sets for validation and to learn about the structure of natural repertoire. For validation, there are some data sets, e.g. [47], which due to experimental setup have sequences known to make a clonal lineage. Also, new microfluidics technology applied to BCR sequencing also gives heavy and light chain data [48, 49]; although a single heavy chain clonal lineage can have light chains from independent rearrangement events, this type of data does provide further evidence of clonality for validation of clonal family inference procedures. In addition to this sort of validation, there are now an abundance of data sets that can be used to characterize the size distribution of the clonal families in various immune states, such as health, immunization, and disease.
As a final note, partis works to solve a challenging likelihood-based inference problem. We recognize that in contrast to existing heuristic approaches based on sequence identity, our software is quite computationally demanding. In this first paper we have developed the framework and overall approach, as well as many computational optimizations. This optimization work is ongoing, and there remain many avenues for improvement. As a comparison, likelihood-based phylogenetic inference has taken two decades of optimization to scale to tens of thousands of sequences at a time with approximate algorithms [50]. We are continually making improvements to the algorithm to make it scale to larger data sets and are committed to building algorithms that scale to the size of contemporary data sets. Although such algorithms may end up being rather different than this version of partis, we believe that likelihood-based algorithms will provide a solid foundation for large-scale molecular evolution studies of B cell maturation.
To introduce the way in which we use HMMs for BCR clustering, consider the canonical “dishonest casino” HMM [20]. In this introductory example, one imagines that a casino offers a game in which the casino alternates between a fair die and a die that is biased towards a given number, say 6. Assume the dice are switched with probability p each roll, corresponding to the HMM on two states, with a transition probability of p between the states. One favorite game of bioinformaticians is to infer the maximum likelihood identity of the die for each roll given a sampled sequence of roll outcomes, which is solved by the Viterbi algorithm. The so-called forward algorithm, on the other hand, infers the marginal probability of a sequence of outcomes.
The likelihood ratio used in this paper fits into the metaphor with a slight variant of the game (Fig 12). In this variant, a pair of outcomes (a.k.a. emissions, in this case integers in the range 1 to 6) are sampled at each step. The player knows that either the emissions came from rolling the same die twice and then switching out the die with probability p after each step, or they came from rolling two dice which are independently switched out with probability p. The new game, corresponding to the methods in this paper, is to figure out which of these scenarios is correct, and with what support.
The marginal probability of a sequence of emissions under the “double roll” scenario is that of a pair-HMM with transition probability p with identical emission probabilities, while the latter “two dice” scenario is that of two independent HMMs. The ratio of these two marginal probabilities is a likelihood ratio quantifying the strength of evidence for the “double roll” scenario.
Now, stepping back into the world of VDJ recombination, we will apply this logic to the HMM structure introduced in [19]. This HMM, building on prior work [10–12], has one state for each position in every V, D, and J gene, and a state for each of the joining N-regions for heavy chain sequences. Light chain sequences are simpler, in that they have only V, J, and one N-region, and so for the rest of this methods section we will only describe the heavy chain procedure.
Continuing with the metaphor, the identity of the die (of which there are now many) for each roll corresponds either to an annotation of that nucleotide as being from a given non-templated insertion base, or as being from a specific nucleotide in a specific V, D, or J gene. That is, a path through the HMM corresponds to an annotated ancestor sequence. Our previous paper [19] was focused on inferring these annotated ancestor sequences using the Viterbi algorithm. Here we focus on the question of whether a group of sequences came from the same rearrangement event rather than on the annotated ancestor sequences themselves. However, this distribution of annotated ancestor sequences is highly informative about the clonality of a group of sequences. We would like use these annotated ancestor sequence inferences but avoid putting too much trust in one specific and necessarily uncertain inference, and instead account for the diversity for possible annotations. We do so as follows.
Using σ to designate paths and x for a sequence, the marginal probability P ( x ) of generating x via any path is
P ( x ) = ∑ σ P ( x ; σ ) ,
where P ( x ; σ ) designates the probability of generating x with the path σ through the HMM. Now for a pair of sequences x and y,
P ( x , y ) = ∑ σ P ( x , y ; σ ) ,
is the probability of generating both x and y using emissions from a single pass through the HMM. Thus P ( x , y ) / ( P ( x ) P ( y ) ) is a likelihood ratio such that values above 1 support the hypothesis that x and y come from the same rearrangement event and values less than 1 support the hypothesis that they do not. Recall that all of these probabilities can be calculated efficiently via the forward algorithm.
More generally, if we would like to evaluate whether sequence sets A and B (each of which are assumed to descend from single rearrangement events) actually all came from a single rearrangement event. For that we can calculate
P ( A ∪ B ) P ( A ) P ( B ) (1)
where P ( X ) can be calculated by a (simple) HMM if X has one element, a pair-HMM if X has two elements, etc., so in general a multi-HMM. Note that this not a phylogenetic likelihood, but a rather strictly HMM-based likelihood, and so does not attempt to incorporate any tree structure into the computations.
We use this likelihood ratio for agglomerative clustering. Specifically, at each step we pick the pair A and B that have the largest likelihood ratio Eq (1) and merge them by replacing A and B from the list of clusters and adding A∪B (Fig 13). We stop agglomerating according to a likelihood ratio threshold, as described in the section after next. Agglomerative clustering has been applied before for clonal family inference [36], however in this case rather than averaging the distances for the merged clusters (as for average linkage clustering) we recompute likelihood ratios with the newly merged set of sequences.
The HMM architecture we use is the same as that of [19], which for the most part follows previous work [10–12] by representing each germline base in each V, D, and J allele as an HMM state. All of these states can be combined to create a single HMM for the entire VDJ rearrangement process. In order to allow likelihood contributions from the N-region, we replace the single insert state found in previous work with four states, corresponding to naive-sequence N-addition of A, C, G, and T. The emissions of these four states are then treated as for actual germline states: the A state, for example, has a large probability of emitting an A, and a complementary probability (equal to the observed mutation probability) of emitting one of the other three bases.
Our application of HMMs also differs from previous work using HMMs for B cell receptor sequence analysis in that we do inference under a model which simultaneously emits an arbitrary number of symbols k. When k = 2 this is typically called a pair HMM [20], and we call the generalized form a multi-HMM (k ≥ 2). One can also think of this as doing inference while constraining all of the sequences to come from the same path through the hidden states of the HMM. In our setting, the k sequences resulting from such a multi-HMM model are the various sequences deriving from a single rearrangement event (which differ only according to point substitution from somatic hypermutation). HMM inference is performed by an efficient new HMM compiler, called ham, which we wrote to inference on an arbitrary (multi-)HMM specified via a simple text file (https://github.com/psathyrella/ham/).
A straightforward application of hierarchical clustering in this setting, in which the likelihood ratio is computed for every cluster at every stage of the algorithm, would not scale to more than a few hundred sequences. Thus as described above, we also use Hamming fraction (Hamming distance divided by sequence length) between inferred naive sequences to avoid expensive likelihood ratio computation. In order to compare unequal-length sequences, we first align the conserved cysteine in every sequence, and then pad all sequences on both ends with ambiguous nucleotides until they are all the same length. In addition to point partis described as an approximate method above, we also use naive Hamming fraction in the full partis method in order to identify sequences that are either very likely or very unlikely to be clones. We assume that clusters which differ by more than 0.08 in naive Hamming fraction are not clonal, and therefore avoid calculating the full likelihood for these cases. This threshold is for repertoires with typical mutation levels (around 5%); we find that increasing the threshold as mutation increases (to 0.15 at 20% mutation) provides optimal performance. We interpolate and extrapolate linearly for other mutation levels. In addition, we assume that clusters that are closer than 0.015 (regardless of mutation levels) in naive Hamming fraction are clonal, and merge these without calculating the full likelihood. While the naive Hamming fraction only takes into account the Viterbi path (i.e. it does not sum over all potential annotated ancestor sequences), and it has no probabilistic interpretation, it has the not insignificant virtue of being much faster to calculate.
According to standard statistical theory, we should merge an a priori specified pair of clusters A and B when the likelihood ratio Eq (1) is greater than one. However, in the midst of a series of agglomerations, we are not in the setting of a single decision for clusters that have been presented to us. Instead, at every stage we are comparing a quadratic number of potential merges and asking if the pair of clusters with the largest likelihood ratio deserve to be merged. This effectively presents substantial multiple testing issues: even when no more clusters should be merged, the nonzero-width of the empirical likelihood ratio distribution will typically have points above one. Furthermore, the marginal probability P ( A ) of, say, the kth largest cluster after some number of merges is going to be biased by the fact that the sequences in that cluster were selected to merge. Such issues are not new in computational biology [51]. We also note that we are only calculating this likelihood ratio when pairs of sequences are similar enough in their inferred naive sequences to merit such a likelihood ratio calculation, further taking us from the statistically ideal setting.
We have found it useful to use a likelihood ratio threshold greater than 1, and use a threshold that decreases as the candidate cluster size, i.e. the size of a proposed cluster, increases (Table 1). These values were selected as a trade-off between accurate reconstruction of large clonal families on the one hand, and accuracy at the low end of the cluster size distribution on the other. Thus if we want to minimize the chance of missing highly-mutated members of a large clonal family we should choose lower thresholds, but if we instead want to avoid mistakenly merging unrelated singletons we should choose higher ones. In light of this, the thresholds can be set on the command line.
While it would be straightforward in principle to account for insertions and deletions (indels) during somatic hypermutation within the HMM by adding extra transitions for deletions and extra states for insertion, this approach would entail a very substantial computational cost. When restricting to substitution mutations, each germline state can either transition to the next germline state, or it can leave the region. If we allowed indels within the V, D, and J segments, however, each state would also need to investigate the probability to transition to a special insertion state as well as to any subsequent germline state. This would introduce a quadratic dependence on the number of states and the resulting algorithm would not be able to analyze realistically-sized data sets.
We thus instead adopt an approach to indel mutations based on the annotation from our preliminary Smith-Waterman step (implemented with ighutil [30]). In cases where ighutil detects an insertion with respect to a germline segment, we “reverse” the insertion by removing it from the query sequence. Similarly, candidate deletions are reversed by inserting the corresponding germline bases from the best germline match when the putative deletion happens in a germline segment. In both cases the original sequences are maintained, but the partis processing of the sequences is done on the modified sequences.
As with any Smith-Waterman implementation, this approach depends on several arbitrary parameters: the match and mismatch scores and the gap-opening penalty. In particular, a larger gap-opening penalty relative to the match/mismatch scores decreases sensitivity to indels. On all samples which we have encountered, a good initial set of match:mismatch scores is 5:1. Sequences with lower mutation rates, for which 5:1 is less optimal, are returned with no D segment match, and then re-run with match:mismatch scores of 5:2. Sequences which still have no D matches are then rerun with scores of 5:3. This procedure gives good results in all parameter regimes which we have encountered in the data. Similarly, we find that a gap-opening penalty of 30 provides good sensitivity to indel mutations in simulation. Each of these parameters may also be set with a command line flag.
In order to test the effectiveness of this method, we made simulated samples in which each sequence has a 50% chance of having an indel mutation after being generated on a tree. Each indel has equal probability of being an insertion or a deletion, and the indel’s position is chosen from the uniform distribution either on the bulk of the V segment (between position 10 and the conserved cysteine), or on the CDR3. The length of each indel is drawn from a geometric distribution with mean 5. These samples are not intended to mimic any particular data set, but are instead designed to provide an extremely stringent test of performance in the presence of indel mutations (Fig 8).
The accuracy of the full likelihood framework which we have described above does not come without some computational cost. As such we have also implemented two other algorithms which make some reasonable trade-offs in accuracy in order to gain some speed.
Point partis. One of the biggest contributors to both annotation and partitioning accuracy comes from our multi-HMM framework’s ability to run simultaneously on an arbitrary number of sequences. Since this ability is entirely separate to the summation over all possible rearrangements, it makes sense to decouple the two in order to optimize for speed. We can, in other words, cluster using the single best (Viterbi) annotated ancestor sequence for all sequences in a cluster (inferred simultaneously on the whole cluster with the multi-HMM), without summing over all germline genes and all rearrangement boundaries. We call this point partis, to emphasize that it uses the best point (i.e. single) annotation inference to do clustering. In order to cluster on these inferred naive sequences, we use the hierarchical agglomeration described above, but with Hamming fraction as the metric (instead of log likelihood ratio). As in the case of the likelihood ratio merging thresholds described above, we perform a simple optimization procedure on a wide variety of simulation samples which span the range of possible lineage structures and mutation levels that we observe in real data. For typical (low) mutation levels near 5%, we use a threshold of 0.035; the threshold then increases to 0.06 as the mutation frequency reaches 20%. Simple linear interpolation (extrapolation) is used inside (outside) of this range. Note that these thresholds are much tighter than those mentioned above for full partis optimization: while above we were trying to exclude cases where there was any doubt as to their clonality, here we are attempting to accurately divide clonal from non-clonal clusters in the naive Hamming distribution. Comparing to Fig 7 in [19], we note that this threshold is equivalent to the expected fractional error in the inferred naive sequence.
vsearch partis. The point method, however, still performs full all-vs-all comparisons on the entire data set, and recalculates the full Viterbi naive sequence on each cluster each time more sequences are added. While this is a good way to ensure the best accuracy, there exist clustering algorithms with many optimizations which trade some of this accuracy for improved speed. vsearch [21] is one such tool, and we have included a version of partis which infers the Viterbi naive sequence for each single query, and then passes these sequences to vsearch. This sacrifices some accuracy, particularly on larger clonal families, but is extremely fast. We use vsearch version 1.1.3 in cluster fast mode with the maximum accept and reject thresholds set to zero, and the id threshold set (again, based on coarse heuristic optimization) to one-half the threshold described above for point partis.
We have added an option to reconstruct the lineage of a user-specified sequence using full partis, for situations in which one is only interested in one specific clonal family. We call such a user-specified sequence a seed sequence. This is shown as “full partis (seed)” (Figs 3 and 4). Here we chose a seed sequence at random from a randomly-selected “large” cluster, where “large” means with size greater than or equal to the mean N leaves for the sample. It can be seen that this method accurately reconstructs the single lineage of interest while running much more quickly than the other methods (Fig 11).
To benchmark results, we simulate sequences using the procedure described in [19]. This provides a bountiful supply of sequences for which the correct lineage structures are known, and with any desired combination of tree topologies and mutation parameters, but with all other properties mimicking empirical values. Briefly, the simulation proceeds by sampling a set of parameters defining a single rearrangement (e.g. V exonuclease deletion length, V allele, etc.) from their empirical joint distribution observed in a data set. Then TreeSim [52] is used to simulate a tree and Bio++ [53] is used to simulate sequences. We emphasize that these sequences are not generated at any stage using partis’ HMM, and no information concerning the simulation is fed to the clustering code other than the simulated sequences. The number of leaves (BCR sequences per clonal family) is distributed geometrically with the indicated mean value in all figures except Figs 7 and S1. In Fig 7 we have used a Zipf (power law) distribution. In S1 Fig, on the other hand, we have used a box-shaped distribution to check that our methods do not depend on a monotonically decreasing distribution. In order to simulate a given number of sequences, we simply divide the desired number of sequences by the expected number of sequences per clone and simulate the resulting number of clones. For indel simulations, half of the simulated sequences have a single indel, whose length is drawn from a geometric distribution with mean 5. In order to emphasize the importance of the indel’s location, we show samples where they are distributed evenly either within the CDR3, or within the bulk of the V segment (specifically between position 10 and the conserved cysteine).
We use per-sequence averages of sensitivity and precision to quantify clustering accuracy. In this context, a true positive (TP) statement about a sequence x is the correct identification of another sequence in x’s clonal family, i.e. correctly clustering a sequence with x. A false postive (FP) statement is incorrectly clustering a sequence with x, while a false negative (FN) statement is not clustering a sequence with x that should be clustered.
Sensitivity x = | TP x | | TP x + FN x | Precision x = | TP x | | TP x + FP x |
Thus, as described above, the precision for a given read is the fraction of sequences in its inferred cluster to which it is truly clonally related. The sensitivity for a given read is the fraction of sequences in its true cluster that appear in its inferred cluster. We average these two quantities over all sequences (Figs 2, S3 and S4). These figures also show the average harmonic mean of this sensitivity and precision (a.k.a. F1 score), as an aggregate measure of the quality of the clustering.
We also show intersection matrices: the matrix of intersection sizes between pairs of large clusters in two partitions (examples in Figs 3, and 4, 9 and 10; the full set of plots is available at figshare at http://figshare.com/s/9b85e4ac54d011e5bd3e06ec4b8d1f61. To make these plots, we first take the 40 largest clusters from each of the two partitions. Each non-white square indicates that there was a non-empty intersection between the two clusters; the square is shaded by the size of the clusters’ intersection divided by their mean size. The position of the square shows the relative sizes of the two clusters. Thus a value of 1.0 implies identity, so very similar partitions will show many dark squares near the diagonal, and will also have similar cluster sizes marked on the x and y axes.
Performance versus sample size. Given the large size of modern deep sequencing data sets, we have also investigated performance as a function of sample size. This function depends on the clonal lineage structure. At one extreme, a sample with only a few sequences stemming from a few clonal families is generally trivial to partition even just by visual inspection. As the number of clonal families increases, however, each family becomes closer and closer to other families, and it becomes more and more difficult to distinguish between them. At the point where the naive sequences corresponding to each family are separated by only a few bases, accurate overall clustering becomes impossible even in principle, since a difference of only a few bases which stems from rearrangement cannot be distinguished from somatic hypermutation.
In order to evaluate this performance we show several performance metrics as a function of sample size (S5 Fig). Here we show the two complementary precision and sensitivity metrics in the top row, and their harmonic mean (F1 score) in the bottom row. It can be seen the behavior of the partis with sample size is similar to that of the synthetic partition which joins neighboring true clusters which are closer than some threshold. This is expected, and demonstrates that performance of the partis method decreases as the number of true naive rearrangements in the sample increases, and thus the clonal family inference problem is becoming inherently more difficult.
Non-independence of clustering steps poses a challenge for parallelization, and we approach this challenge with a combination of principled probability calculations and reasonable heuristics. The basic strategy is to begin with a large number of processes, each running on a small subset of the data sample. When each of these processes finishes clustering its allotted sequences, it reports back to the parent program, which collects the results from each subprocess and reapportions the resulting clusters among a new, smaller number of processes for the next step. The process then repeats until we arrive at a single process which is comparing all clusters against all other clusters. On the face of it, each step in this scheme would take much longer than the previous one since it is comparing more sequences. However, because each process caches all the likelihoods it calculates, and because both factors in the denominator for each likelihood ratio Eq (1) is guaranteed to have been calculated in a previous step, we can choose the process number reduction ratio such that each stage of paralellization takes roughly the same time.
An important part of this process is the allotment of sequences to processors. At present we apportion them randomly in order to achieve a (very) roughly equal number of computations per process. This is far from ideal, however, because we want to merge clonal sequences as soon as possible in order to avoid unnecessary comparisons to non-clonal sequences. This must be balanced, however, by the need to evenly distribute the workload across all processes. In the future we will study in more detail the optimal allotment scheme, and anticipate substantial speed increases.
|
10.1371/journal.ppat.1004020 | MAVS-MKK7-JNK2 Defines a Novel Apoptotic Signaling Pathway during Viral Infection | Viral infection induces innate immunity and apoptosis. Apoptosis is an effective means to sacrifice virus-infected host cells and therefore restrict the spread of pathogens. However, the underlying mechanisms of this process are still poorly understood. Here, we show that the mitochondrial antiviral signaling protein (MAVS/VISA/Cardif/IPS-1) is critical for SeV (Sendai virus)-induced apoptosis. MAVS specifically activates c-Jun N-terminal kinase 2 (JNK2) but not other MAP kinases. Jnk2−/− cells, but not Jnk1−/− cells, are unable to initiate virus-induced apoptosis and SeV further fails to trigger apoptosis in MAPK kinase 7 (MKK7) knockout (Mkk7−/−) cells. Mechanistically, MAVS recruits MKK7 onto mitochondria via its 3D domain, which subsequently phosphorylates JNK2 and thus activates the apoptosis pathway. Consistently, Jnk2−/− mice, but not Jnk1−/− mice, display marked inflammatory injury in lung and liver after viral challenge. Collectively, we have identified a novel signaling pathway, involving MAVS-MKK7-JNK2, which mediates virus-induced apoptosis and highlights the indispensable role of mitochondrial outer membrane in host defenses.
| The mitochondrial antiviral signaling protein (MAVS/VISA/Cardif/IPS-1) is critical for the innate immune response during viral infection, and its function has been well documented in mediating type I interferon production. In this study, we revealed the essential role of MAVS in virus-induced apoptosis, independent of Retinoic acid-Inducible Gene I (RIG-I) signaling. Upon viral infection, MAVS recruits MKK7 onto mitochondria, followed by MKK7 induced activation of JNK2, which subsequently initiates apoptosis. Importantly, we have clearly differentiated the roles of JNK2 versus JNK1, and MKK7 versus MKK4 in virus-induced apoptosis. Thus, we define a novel apoptotic signaling pathway, involving MAVS-MKK7-JNK2, which sheds a new perspective on the crosstalk between the antiviral and apoptotic signaling pathways in innate immunity.
| The induction of innate immunity upon viral infection represents the first line of host defense against microbe invasion. During infection with a RNA virus, the mitochondrial antiviral signaling protein (MAVS/VISA/Cardif/IPS-1) has been recently uncovered to seed a critical protein complex on the mitochondrial outer membrane [1]–[4]. This signalosome consists of TNFR-associated factors (TRAF2/3/6) [5], TNFR-associated death domain protein (TRADD) [6], translocase of outer mitochondrial membrane 70 (TOM70) [7], ubiquitously expressed transcript (UXT-V1) [8], Autophagy proteins (Atg5/Atg12) [9], Mitofusin-2 (Mfn2) [10] et.al. The MAVS signalosome activates TANK-binding kinase 1 (TBK1), which then phosphorylates interferon (IFN) regulatory factor 3(IRF3). In addition, the MAVS signalosome can also activate the IκB kinase (IKK) complex and nuclear factor κB (NF-κB). Synergistically, these transcription factors (IRF3 and NF-κB) induce the early production of type I interferons and inflammatory cytokines [11].
The induction of apoptosis following viral infection is another effective means by which the host attempts to restrict the spread of pathogens by sacrificing the virus-infected cells [12]. Several recent studies suggested that MAVS mediates virus-induced cell apoptosis and that this function is independent of the interferon pathway. MAVS, as well as UXT-V1/V2, have been reported to promote the clearance of infected cells [13]. MAVS overexpression causes cell apoptosis and several viral proteins, such as hepatitis C virus NS3/4A and the Severe Acute Respiratory Syndrome coronavirus (SARS-CoV) nonstructural protein (NSP15), are inhibitors of the MAVS-induced apoptosis [14]. Borna disease virus (BDV) X protein can also interact with MAVS to inhibit MAVS-mediated apoptosis [15]. In addition, MAVS decreases the K48-linked ubiquitination of voltage-dependent anion channel 1(VDAV1) [16]. Activation of MAVS by Bunyavirus infection upregulates the adaptor protein SARM1, leading to neuronal death [17]. However, the molecular mechanism of how MAVS regulates apoptosis remains poorly understood.
It is intriguing to understand the roles of mitochondria in virus-induced apoptosis, in particular, elucidation of specific apoptotic signaling pathways. Sendai virus (SeV), Newcastle disease virus (NDV) and Vesicular stomatitis virus (VSV) are enveloped negative-strand RNA viruses, which are able to trigger acute infection in rodents and induce robust cell apoptosis [18]. As such, these infections represent effective models for studying virus-induced apoptosis. Due to the central role of mitochondria in apoptosis and the mitochondrial localization of MAVS, it is important to explore the putative function of MAVS in the apoptotic signaling pathway. This work will be instrumental towards understanding the molecular mechanisms regarding host decisions on cell survival and apoptosis, and shedding further light on the cross-talk between innate immunity and apoptosis.
In this study, we report that MAVS plays an essential role in virus-induced apoptosis. Upon SeV infection, MAVS recruits MAPK kinase 7 (MKK7) onto mitochondria via its 3D domain, which specifically activates c-Jun N-terminal kinase 2 (JNK2) and triggers cell apoptosis. SeV was unable to trigger apoptosis in either Mavs−/− cells or Mkk7−/− cells. In addition, Jnk2−/− cells, but not Jnk1−/− cells, failed to initiate the virus-induced apoptosis. Interestingly, RIG-I and MDA5 (melanoma differentiation-associated protein 5) were not required for this apoptosis. Consistently, Jnk2−/− mice, but not Jnk1−/− mice, displayed marked inflammatory injury in both lung and liver after viral challenge. This study identifies a novel signaling pathway, MAVS-MKK7-JNK2, to mediate virus-induced apoptosis, revealing the indispensable role of the mitochondrial outer membrane in host defense.
Given that MAVS robustly activates TBK1 and IKK kinases, we wondered if MAVS could influence Mitogen-Activated Protein Kinase (MAPK) signaling. To explore this possibility, we overexpressed MAVS in HEK293 cells and then checked the activation of JNK, extracelluar signal-regulated kinase (ERK) and p38. Interestingly, the phosphorylation of JNK exhibited a dose-dependence on the expression of MAVS, whereas similar effects did not apply to either ERK or p38 (Figure 1A), suggesting that MAVS may specifically activate JNK. Using siRNA to knock down endogenous MAVS, we further explored the SeV-induced activation of JNK. As expected, SeV could induce the phosphorylation of all the MAP kinases (JNK, ERK and p38). Notably, knockdown of MAVS appeared to attenuate JNK activation, whereas this did not influence the activation of ERK or p38 (Figure 1B). In addition, the knock down of MAVS did not affect tumor necrosis factor alpha (TNFα)-triggered phosphorylation of JNK, ERK or p38 (Figure 1B). These observations suggest that virus-induced activation of JNK is dependent on MAVS.
Considering the critical function of MAVS in innate immunity, we tested if the relevant signaling proteins could also activate JNK. Ectopic-expression of RIG-I, TBK1, IKKα or IKKε did not influence JNK phosphorylation (Figure 1C). RIG-I and MDA5 are the viral nucleic acid sensors and are both upstream of MAVS signaling. We used Rig-i−/− MEF cells to determine whether RIG-I also mediated JNK phosphorylation. Surprisingly, knockout of RIG-I didn't influence the SeV-triggered JNK phosphorylation, although it did abrogate IRF3 activation (Figure 1D). We also prepared siRNA sets to specifically knock down RIG-I or MDA5. The results showed that a decrease in either RIG-I or MDA5 did not impact SeV-induced JNK phosphorylation (Figure S1A). The absence of TBK1 also had no effect on JNK phosphorylation (Figure 1E). In contrast, MAVS deficiency completely blocked SeV-induced JNK activation (Figure 1F). Taken together, the SeV-induced activation of JNK is dependent on MAVS, yet independent of RIG-I/MDA5 and TBK1/IKK. These results suggest that MAVS is the converging point for activating JNK, TBK1 and IKK during viral infection.
We went on to explore whether JNK could modulate type I interferon signaling. Interestingly, we observed no difference of SeV-induced Interferon Stimulated Gene 15/60(ISG15/ISG60) production amongst control, JNK1 deficiency or JNK2 deficiency, using either siRNA knock down in HEK293 cells (Figure 2A, left) or in knockout mouse embryonic fibroblast cells (MEFs) (Figure 2A, right), indicating that JNK1/2 are dispensable for virus-induced interferon β (IFN-β) signaling.
In order to test whether MAVS plays a role in virus-induced apoptosis, we measured cell apoptosis by monitoring the apoptosis marker poly ADP ribose polymerase (PARP) in Mavs+/+ and Mavs−/− MEFs. Notably, Mavs−/− cell had much less apoptosis than Mavs+/+ cells following SeV challenge. In contrast, TNFα/CHX (cycloheximide) induced comparable level of apoptosis in Mavs+/+ and Mavs−/− cells (Figure 2B). These results indicate that MAVS specifically modulates virus-induced apoptosis. In addition, ectopic-expression of MAVS potentiated JNK phosphorylation and apoptosis, whereas RIG-I (full length), RIG-I-CARDs (constitutively activated form of RIG-I), TBK1, IKKα and IKKε failed to do so (Figure 2C). We also measured SeV-induced apoptosis in Rig-i−/− MEFs. Consistently, there was no difference in the cleavage of PARP or caspase-3, between RIG-I knockout and wild type control (Figure S1B).
Based on these results, we hypothesized that the MAVS-dependent activation of JNK was linked to virus-induced apoptosis. It was observed that the general inhibitor for JNK1/2(SP600125) markedly attenuated the SeV-induced PARP/caspase-3 cleavages (Figure 2D). Consistently, the caspase inhibitor Z-VAD effectively blocked the PARP/caspase-3 cleavages, whereas the inhibitor did not affect the phosphorylation of JNKs upon SeV stimulation (Figure S2A and S2B), suggesting that JNK activation is primary, not secondary to cell apoptosis. Unexpectedly, knock down of endogenous JNK2 alone significantly attenuated the SeV-induced PARP/caspase-3 cleavages, whereas knockdown of JNK1 alone did not appear to influence apoptosis (Figure 2E). These observations were further substantiated by using Jnk1−/− and Jnk2−/− MEF cells after SeV infection (Figure 2F) or VSV infection (Figure S2C). Collectively, these data differentiate the functions of JNK1 and JNK2, revealing the specific role of JNK2 in SeV-induced apoptosis.
Since both TRADD and TRAF proteins are known components of the mitochondrial MAVS antiviral signalosome, we probed the role of TRADD and TRAFs in virus-induced JNK activation and subsequent apoptosis. Interestingly, knocking down TRADD or TRAFs had no observable influence on either JNK activation or apoptosis upon SeV infection (Figure S3), suggesting that TRADD and TRAFs are not involved in virus-induced MAVS/MKK7/JNK2 signaling.
MKK4 (MAPK kinase 4) and MKK7 are potential upstream kinases for JNK1 and JNK2 [19]. Given that both MKK4 and MKK7 could phosphorylate JNK1/2 in response to TNF-α, growth factors, DNA damage et.al, we tested whether MKK4/7 is essential for the SeV-induced JNK phosphorylation. To do this, we used Mkk4/7−/− or Mkk3/6−/− double knockout MEF cells. As expected, JNK1/2 phosphorylation was abolished in Mkk4/7−/− cells, whereas this was not the case in Mkk3/6−/− cells (Figure 3A). In addition, we individually knocked down MKK4 and MKK7 in HEK293 cells. MKK7, but not MKK4, was both necessary and sufficient for JNK phosphorylation (Figure 3B). Consistent with these results, SeV-induced apoptosis was significantly impaired in Mkk4/7−/− MEF cells, in contrast to Mkk3/6−/− MEF and wild type cells (Figure 3C). The specific knock down of MKK7, but not MMK4, further attenuated apoptosis in HEK293 cells (Figure 3D). Collectively, our results reveal MKK7 as the essential signal transducer for SeV-induced apoptosis.
To elucidate the mechanism of MAVS-dependent activation of JNK2, we tested the potential interactions between MAVS and JNK1, JNK2, MKK4, MKK7, respectively. It was found that only MKK7 could interact with MAVS, whereas JNK1, JNK2 or MKK4 failed to do so (Figure 4A). We also confirmed the endogenous interaction between MAVS and MKK7. Notably, this endogenous interaction was markedly enhanced upon SeV infection (Figure 4B). In addition, MKK7 could not bind RIG-I, TBK1 or IKKε (Figure 4C). MKK7 was also unable to bind to MAVS-ΔTM, which is deprived of the trans-membrane domain(TM) and is localized inside the cytoplasm (Figure S4), suggesting that the trans-membrane domain of MAVS is important for its interaction with MKK7.
To further investigate the MAVS-MKK7 interaction, we generated three truncated mutants of MKK7, which were MKK7-Δ3D (lack of 3D domain), MKK7-ΔDVD (lack of DVD domain) and MKK7-PK (protein kinase domain only; lack of 3D, DVD and C-terminal) (Figure S5). We mapped the interaction between MAVS and the mutants. Only MKK7-ΔDVD, neither MKK7-Δ3D nor MKK7-PK could bind MAVS (Figure 4D), indicating that 3D domain mediates the MKK7-MAVS interaction.
Since MAVS is anchored on the mitochondrial outer membrane, we reasoned that MKK7 could dynamically be recruited onto mitochondria upon virus infection. Subcellular fractionation and confocal immunofluorescence revealed that a sub-pool of MKK7 constitutively co-localized with mitochondria, whereas no co-localization was evident with MKK4 (Figure 4E and 4G). Importantly, the presence of MKK7 on mitochondria apparently increased upon SeV stimulation, while MKK4 could not do so (Figure 4E and 4G), suggesting that MKK7 dynamically translocates onto the mitochondria upon virus infection. In Mavs−/− cells, MKK7 lost the ability to localize to mitochondria (Figure 4F), indicating this translocation is MAVS-dependent. In addition, MKK7-Δ3D, which lacks the 3D domain and is unable to bind MAVS, could not translocate onto mitochondria (Figure 4H), suggesting that the recruitment of MKK7 onto mitochondria depends on its interaction with MAVS.
To delineate the topology of apoptosis signaling, we re-introduced MKK4 or MKK7 into the Mkk4/7−/− double knockout MEFs and then monitored for apoptosis in these cells. It was observed that MKK7, but not MKK4, could rescue the MAVS-induced PARP and caspase-3 cleavages (Figure 5A), indicating that MKK7 functions downstream of MAVS. In addition, MAVS could induce the same apoptosis in Jnk1−/− MEF cells, but not in Jnk2−/− MEF cells (Figure 5B).
Furthermore, we re-introduced MKK4 or MKK7 respectively into the Mkk4/7−/− double knockout MEF cells, and accordingly named them as Mkk7−/− or Mkk4−/− cells. Consistently, ectopic-expression of wild type MKK7 could rescue the SeV-induced apoptosis in Mkk7−/− cells. However, MKK7-Δ3D could not (Figure 5C), which is consistent with the inability of MKK7-Δ3D to bind MAVS. MKK7-ΔDVD could partially rescue cell apoptosis due to its interaction with MAVS (Figure 4D and Figure 5C), implicating the functional interaction of MAVS and MKK7.
Biochemically, MKK7 phosphorylates JNK2 at 183-threonine (T) and 185-tyrosine (Y). We generated JNK2(183/185A) with both 183-threonine (T) and 185-tyrosine(Y) mutated to alanine and found that JNK2(183/185A) could not rescue the apoptosis in Jnk2−/− MEF cells, in comparison to JNK2(WT) (Figure 5D). Likewise, MAVS-ΔTM could not rescue the SeV-induced apoptosis in Mavs−/− MEF cells, indicating that it is necessary to form the signaling complex on mitochondria (Figure 5E). Taken together, MAVS→MKK7→JNK2 defines a novel apoptotic signaling pathway during viral infection.
To determine the in vivo function of JNK2, we employed the vesicular stomatitis virus (VSV) infection model using wild type, Jnk1−/− and Jnk2−/− mice. The mortality of the Jnk2−/− mice was proportional to the amount of VSV administrated. Almost all of the Jnk2−/− mice died when high dose of VSV was used (Figure 6A). In contrast, only a small fraction of the Jnk1−/− mice died and no wild type mice died, under the same conditions (Figure 6A). Two days after VSV infection, we prepared serum from the mice and performed plaque assays in HEK293 cells (Figure 6B, upper). Consistently, VSV replicated much more robustly in Jnk2−/− mice (Figure 6B, lower). Quantification revealed that the viral titer of VSV in Jnk2−/− mice was 7–10 fold higher than that in Jnk1−/− or wild-type mice (Figure 6C). These results indicate that JNK2 could protect mice against the VSV infection, whereas JNK1 was dispensable.
As a second viral infection model to investigate the role of JNK2, GFP-labeled Newcastle Disease Virus (NDV-GFP) was used to challenge the mice intranasally. Two days after infection, the lungs of the wild-type, Jnk1−/− and Jnk2−/− mice were respectively imaged in situ by fluorescence microscope. Strikingly, NDV-GFP was markedly observed in the lung of the Jnk2−/− mice, whereas no or only marginal NDV-GFP were detected in the lungs of wild-type or Jnk1−/− mice (Figure 6D). Serum was also collected from the mice and the NDV-GFP titers were monitored by infecting HEK293 cells and using fluorescence microscopy (Figure 6E) and flow cytometry (Figure 6F). Both assays revealed that much more NDV was produced in vivo in Jnk2−/− mice, as compared with wild-type or Jnk1−/− mice. These data indicate that JNK2 is essential to protect mice against viral invasion.
Viral infection induces both innate immunity and cell apoptosis, of which macrophages represent one of the major innate immune cells in early infection. We harvested and cultured bone marrow-derived macrophages (BMDM) from wild type, Jnk1−/− and Jnk2−/− mice. These BMDM were then infected with SeV and their cytokine expression was measured at both the protein and mRNA level (Figure 7A and Figure S6). It was observed that JNK1 and JNK2 only marginally influenced the expression of cytokines induced by innate immune signaling, and there was no difference in IFN-β (Interferon β) protein expression between wild type and JNK2 knockout cells (Figure 7A). Such results suggest that JNK1 and JNK2 do not individually modulate innate immune signaling, but does not rule out the possibility that they could cross-talk with each other.
SeV-induced apoptosis in BMDM was quantitated by Annexin-V/propidium iodide (PI) staining. This apoptosis was markedly attenuated in Jnk2−/− cells compared to both Jnk1−/− cells and wild type cells (Figure 7B). In addition, we confirmed that JNK2 did not directly modulate the IRF3 or NF-κB signaling pathways during virus infection by using the corresponding reporter systems in both JNK1/JNK2 siRNA knock down cells and JNK1/JNK2 overexpressing cells (Figure 7C and Figure S7).
To investigate the in vivo function of JNK2 in apoptosis, the lung and liver of Jnk1−/− and Jnk2−/− mice were prepared for immunohistochemistry following SeV, VSV or NDV challenge. It was observed that the lung and liver of Jnk1−/− mice appeared similar to those of the wild type. Remarkably, the tissues of these organs from Jnk2−/− mice displayed severe inflammatory injury, particularly around the blood vessels in the liver (Figure 7D and 7F). Morphologically, those cells appeared necrotic, based on the absence of staining to detect caspase-3 activation in Jnk2−/− samples, which was evident in the wild type and Jnk1−/− samples (Figure 7E and 7G). Overall, this data clearly demonstrates that JNK2 plays an indispensable role in the initiation of virus-induced apoptosis, without which cells undergo necrosis and trigger inflammatory damages.
Viral infection induces innate immunity and apoptosis, which represent effective means for the host to restrict the spread of microbial pathogens. The essential function of MAVS has been well documented in mediating RIG-I/MDA5 innate immune signaling. In this work, we demonstrate that MAVS is also critical for cell apoptosis upon viral infection and highlight the dual function of MAVS in innate immunity and apoptosis.
Our current study defines a novel apoptotic signaling pathway (Figure 8): upon viral infection, MAVS recruits the MAPK kinase MKK7 onto mitochondria, MKK7 phosphorylates and activates JNK2, and JNK2 then initiates the corresponding cell apoptosis. Apoptosis is essential for sacrificing virus-infected cells and dampening the detrimental inflammation, as evidenced by the response of Jnk2−/− mice to viral infection. This study highlights the convergence of innate immunity and apoptosis on MAVS during viral infection, further substantiating the notion that the mitochondrial outer membrane is the critical signaling platform for cellular stress responses.
C-Jun N-terminal kinases (JNKs) play important roles in death receptor-initiated extrinsic as well as mitochondria-initiated intrinsic apoptotic pathways, in response to stress stimuli, such as cytokines, DNA damages, heat shock and osmotic stress et.al. They are also known to modulate cell proliferation and differentiation [19], [20]. JNKs can quickly induce the expression of pro-apoptotic genes via their activation of specific transcription factors. In addition, they can directly regulate the pro- and/or anti-apoptotic activities of mitochondria-related proteins via distinct post-translational phosphorylation [20]. Whether the activation of JNKs leads to cell proliferation or apoptosis is dependent on the stimuli and cell type involved [21]. Further investigation is needed to dissect the manner by which JNK2 triggers the relevant apoptosis effector mechanism, downstream of MAVS-mediated signaling. One possibility is that JNK2 is integrated into the well-characterized effector responses, such as the induction of pro-apoptotic genes, cleavage of Bid, BAD/Bim phosphorylation, or the Bax clustering on mitochondria et.al. It is also intriguing to probe the potential cross-talk among these various apoptotic pathways.
JNK1 and JNK2 are expressed in all cells and tissues, whereas JNK3 expression is predominantly localized in the brain [22]. The functions of JNK1 and JNK2 are largely redundant in canonical apoptosis signaling, with JNK1 as a major player [23], [24]. Using both in vitro and in vivo approaches, we clearly differentiate the role of JNK1 and JNK2 in virus-induced apoptosis, establishing the indispensable role of JNK2 in the MAVS-mediated apoptosis. In addition, we rule out the potential role of p38 and ERK in this apoptosis.
It remains to be addressed why MAVS selectively interacts with MKK7 instead of MKK4. Likewise, it is unknown why JNK2 was chosen for this pathway instead of JNK1. We speculate that there might be yet-to-discovered scaffold proteins responsible for the selectivity. For example, JNK-interacting protein 1 (JIP1) has been shown to enhance the activation of JNKs via MLK3 and MKK7, whereas JIP2 specifically interacts with MKK7 and not MKK4. JNK3 has been shown to have more affinity for JSAP1/JIP3 than JNK1 or JNK2 [25].
During our preparation of this manuscript, MAVS was reported to be activated by La Crosse virus (LACV) infection and to upregulate the adaptor protein SARM1, which is related with neuronal death [17]. Although the authors demonstrated that MAVS was associated with SARM1-mediated cell death in neurons, no specific molecular mechanism was presented in their study. Given the tissue-specific expression of JNK3 in neurons, it is reasonable to presume that JNK3 probably plays some role in MAVS-SARM1 mediated neuronal death. We believe that the MAVS-MKK7-JNK may represent a general mechanism for the host to quickly respond to viral infections. It is also unexpected that members of the JNK family display such delicate and clear-cut differences in terms of their biological functions.
With the exception of MAVS, there are no overlapping signaling proteins between the innate and apoptosis signaling pathways during RNA virus infection. The loss of JNK1/2 did not affect the RNA-virus-induced IRF3 or NF-κB activation. Unexpectedly, RIG-I/MDA5 was dispensable for the virus-induced apoptosis, suggesting the existence of other sensors upstream of MAVS. This is analogous to the cellular recognition of DNA viruses. Several DNA sensors have been reported recently, including Absent In Melanoma 2 (AIM2) [26], DNA-dependent activator of interferon regulatory factors (DAI) [27], DEAD box polypeptide 41 (DDX41) [28], interferon inducible protein 16 (IFI16) [29], and cGAMP synthase (cGAS) [30], [31]. Each of these sensors activate the endoplasmic reticulum protein called stimulator of interferon genes (STING). Like MAVS, STING activates both IKK and IRF3, thereby turning on the NF-κB and IRF3 signaling pathways [32], [33]. In the current study, we established the role of MAVS-MKK7-JNK2 signaling in mediating RNA virus-induced apoptosis. Future studies will address whether STING plays a role in JNK activation during DNA virus-induced apoptosis. In addition, it remains intriguing to examine potential sensor(s) for RNA virus-induced apoptosis.
C57BL/6 mice were purchased from the Shanghai SLAC Laboratory Animal Company. The mice were maintained under specific pathogen-free (SPF) conditions at the Shanghai Institute of Biochemistry and Cell Biology. All animals used in this study were 4∼8 weeks of age. Animal experiments were carried out in strict accordance with the regulations in the Guide for the Care and Use of Laboratory Animals issued by the Ministry of Science and Technology of the People's Republic of China. The protocol was approved by the Institutional Animal Care and Use Committee of the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences (Permit Number: IBCB0027 Rev).
Jnk1−/− and Jnk2−/− mice [34], [35] were kindly provided by Dr. Lijian Hui. These stains were maintained on a C57BL/6 background. For infection, 4∼8 weeks old mice were narcotized and then intranasally inoculated with SeV, VSV or NDV-GFP (volume≤20 µl). Negative controls included water and PBS. Clinical symptoms were observed 2∼7 days post inoculation, and the following analyses were performed.
After sacrifice, mice were perfused sufficiently with PBS, and then lung and liver tissues were fixed in 4% paraformaldehyde (in PBS) for 12 hours. Tissue sections were prepared at the Core Facility for Cell biology, Shanghai Institute of Biochemistry and Cell Biology and then hematoxylin and eosin (H&E) staining was performed. For immunohistochemistry of cleaved caspase-3, the EnVision FLEX Systems Kit (Dako) was used.
Two days post NDV-GFP infection (107 PFU/mouse), mice were sacrificed and the lungs perfused with PBS. In situ NDV-GFP was observed using Olympus SZX16 fluorescence stereomicroscopy.
After VSV or NDV-GFP infection for the indicated days, peripheral blood was collected to obtain serum. For VSV titer assay, sera was added into culture media of HEK293 cells and the viral titer was determined by a standard plaque assay [8]. For NDV-GFP title assay, equal amounts of sera were added into culture media of HEK293 cells and cultured for 24 hours. GFP positive cells (NDV-GFP infected cells) were analysis by both fluorescent microscopy and flow cytometry.
Mavs−/− MEFs were provided by Dr. Zhengfan Jiang. Mkk4/7−/− and Mkk3/6−/− MEFs were provided by Dr. Jiahuai Han and Rig-i−/− MEF cells were provided by Dr. Deyin Guo and Dr. Hong-Bing Shu. HEK293, HEK293T and L929 cell lines were obtained from the American Type Culture Collection (ATCC). For bone marrow-derived macrophages (BMDM), femurs were aseptically harvested from mice. Bone marrow cells were flushed from the bones and then cultured in RPMI 1640 media (20% FBS, 2 mM L-glutamine, 50 mM 2-mercaptoethanol) with 40 ng/ml macrophage colony stimulating factor (M-CSF). The medium was refreshed every two days and the cells differentiated to BMDMs in one week. All cells were maintained in a humidified 5% CO2 incubator at 37°C.
Recombinant Human TNF-α was purchased from R&D Systems. Cycloheximide (CHX) was purchased from Sigma. The following antibodies were used for western blot or immunoprecipitation: anti-β-actin (A5316, Sigma), normal mouse IgG (sc-2025, Santa Cruz Biotechnology), normal rabbit IgG (sc-2027, Santa Cruz Biotechnology), anti-HA (sc-7392, Santa Cruz Biotechnology), anti-Flag (F1804, Sigma), anti-Tom20 (11802-1-AP, Proteintech; sc-17764, Santa Cruz Biotechnology), anti-Caspase-3 (9662, Cell Signaling; 9661, Cell Signaling), anti-PARP (sc-7150, Santa Cruz Biotechnology; 9542, Cell Signaling), anti-MAVS (generated by this laboratory and also purchased from Cell Signaling–3993), anti-JNK (sc-571, Santa Cruz Biotechnology), anti-p-JNK (9255, Cell Signaling), anti-p38 (sc-7149, Santa Cruz Biotechnology), anti-p-p38 (9211, Cell Signaling), anti-ERK (9102, Cell Signaling), anti-p-ERK (9101, Cell Signaling), anti-TBK1 (sc-73115, Santa Cruz Biotechnology), anti-ISG15 (M24004, Abmart), anti-ISG60 (15201-1-AP, Proteintech), anti-MKK4 (sc964, Santa Cruz Biotechnology), anti-MKK7 (generated by this laboratory and also purchased from Abcam–ab52618), anti-p-IRF3 (4947, Cell Signaling), anti-TRAF2 (sc-876, Santa Cruz Biotechnology), anti-TRAF3 (sc-1828, Santa Cruz Biotechnology), anti-TRADD (sc-46653, Santa Cruz Biotechnology), anti-RIG-I (AB54008, Shanghai Sangon Biotech; 4520, Cell Signaling), anti-MDA5 (5321, Cell Signaling), anti-MKK3 (5674, Cell Signaling) and anti-MKK6 (9264, Cell Signaling). Mitotracker Red was obtained from Molecular Probes and DAPI was obtained from Life Technologies. Annexin-V-FLUOS Staining Kit was purchased from Roche (11858777001). Sendai virus (SeV), Vesicular stomatitis virus (VSV) and Newcastle disease virus-GFP (NDV-GFP) were kindly provided by Drs. Hong-Bing Shu (Wuhan University) and Zhigao Bu (Chinese Academy of Agricultural Sciences).
Human JNK1, JNK2, RIG-I, MKK4, MKK7, IKKα, IKKε, TBK1 and MAVS cDNAs were cloned from a human thymus plasmid cDNA library (Clontech) using standard PCR techniques and then sub-cloned into the indicated vectors. Mutants were generated by using a Quickchange XL (Stratagene). MAVS-ΔTM was truncated of its C-terminal transmembrane domain (514–540 aa). MKK7-Δ3D was truncated of its N-terminal docking domain (1–85 aa), MKK7-ΔDVD was truncated of its domain of versatile docking (378–400 aa), and MKK7-PK only maintained its protein kinase domain (80–380 aa). The siRNA oligos were synthesized by GenePharma:
JNK1 siRNA: 5′GCCCAGUAAUAUAGUAGUATT3′, 5′GAGCUAGUUCUUAUGAAAUTT3′;
JNK2 siRNA: 5′GUUGCAGUCAAGAAACUAATT3′, 5′GUGAACUUGUCCUCUUAAATT3′;
MKK4 siRNA: 5′ AAUGCGGAGUAGUGAUUGTT3′, 5′GAUUUCACUGCAGAGGACUUTT3′;
MKK7 siRNA: 5′ UAAGCUACUUGAACACAGCTT3′, 5′GAACAAGGAGGAGAACAATT3′;
RIG-I siRNA: 5′ ACGGAUUAGCGACAAAUUUAATT3′, 5′GAAUUUAAAACCAGAAUUAUCTT3′;
MDA5 siRNA: 5′AUCACGGAUUAGCGACAAATT3′, 5′GAAUAACCCAUCACUAAUATT3′
TRAF2 siRNA: 5′ CGACAUGAACAUCGCAAGCTT3′, 5′AGGAGCAUUGGCCUCAAGGATT3′;
TRAF3 siRNA: 5′ GGAGGUUACAAGGAAAAGUTT 3′, 5′ GAAGGUUUCCUUGUUGCAGAAUGAA3′;
TRADD siRNA: 5′ GGAGGAUGCGCUGCGAAAUUU3′, 5′ AACUGGCUGAGCUGGAGGAUGTT 3′;
MAVS siRNA: 5′CCACCUUGAUGCCUGUGAATT3′, 5′CAGAGGAGAAUGAGUAUAATT3′;
Negative control siRNA, 5′UUCUCCGAACGUGUCACGUTT3′.
Cells were transfected with siRNA oligos using Lipofectamine 2000 and then incubated for 48 hours before further analysis. The plasmids were introduced into cells using Lipofectamine 2000, and then cells were cultured for 24 hours before further analysis.
Total cellular RNA was isolated using Trizol (Invitrogen) according to the manufacturer's instructions. Reverse transcription of purified RNA was performed using oligo(dT) primers. The quantification of gene transcripts was performed by real-time PCR using SYBR Green PCR mix (Applied Biosystems). All values were normalized to the level of β-actin mRNA. The primers used are listed below:
β-actin, sense (AAAGACCTGTACGCCAACAC) and antisense (GTCAT ACTCCTGCTTGCTGAT);
IFN-β, sense (AGATCAACCTCACCTACAGG) and antisense (TCAGAAACACTGTCTGCTGG);
ISG15, sense (GGAACGAAAGGGGCCACAGCA) and antisense (CCTCCATGGGCCTTCCCTCGA);
ISG56, sense (AGTGCA GGCAGAAATTCACC) and antisense (AGCAGTCAGTAGTTTCCTCC);
IL-6, sense (GAGAGGAGA CTTCACAGAGG) and antisense (GTACTCCAGAAGACCAGAGG);
IL-12, sense (GCTTCTTCATCAGGGACATC) and antisense (GTCAGGGAGAAGTAGGAATG);
Both floating and adherent cells were collected at the designated time points and stained with Annexin-V (Roche, 11858777001) according to the manufacturer's instructions. Cell apoptosis was detected using a FACS Calibur (BD Biosciences) and the data was analyzed using Flowjo software (Tree Star).
Cell pellets were collected and resuspended in RIPA buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.5% NP40, 0.25% Na-deoxycholate, 1 mM Na3VO4, 0.1 mM PMSF, Roche complete protease inhibitor set) for immunoprecipitation, or in RIPA buffer plus 0.1% SDS for western blot analysis. The resuspended cell pellet was vortexed for 20 seconds and then incubated on ice for 20 minutes, followed by centrifugation at 20,000 g for 20 minutes. Supernatants were collected for subsequent immunoprecipitation or western blot analysis
For immunoprecipitation, cell lysates were pre-cleared with Protein A/G Plus-Agarose (Santa Cruz Biotech) at 4°C for 2 hours, then antibody or control IgG was added and incubated overnight. The next day, cell lysates were incubated for an additional 2 hours since the Protein A/G Plus-Agarose beads were added. The beads were washed with TBS buffer containing 0.5% NP40, boiled using 1× SDS loading buffer, and the supernatants were loaded for western blot analysis.
HEK293 or MEF cells were washed with cold PBS and lysed using a Dounce Homogenizer in homogenization buffer (210 mM sucrose, 70 mM mannitol, 1 mM EDTA, 1 mM EGTA, 1.5 mM MgCl2, 10 mM HEPES [pH 7.2]). The homogenate was centrifuged at 500 g for 10 minutes, and the pellet was discarded as crude nuclei. The supernatant was centrifuged at 5,000 g for 10 minutes to precipitate crude mitochondria; the supernatants were collected as the cytosolic fraction and the precipitate was lysed by RIPA buffer to obtain the mitochondrial fraction.
MEF cells were plated on coverslips in 12-well plates and transfected with the indicated plasmids. Twenty-four hours later, cells were treated with or without SeV (MOI = 1) for 6 hours. To label mitochondria in a specific experiment, cells were incubated with 250 nM Mitotracker Red for 30 min at 37°C. Coverslips with the cells were washed once with PBS and fixed in 3.7% formaldehyde in PBS for 15 minutes. After permeabilization with Triton X-100 (0.25%) in PBS for 15 minutes, cells were blocked with PBS containing BSA (5%) for 1 hour and then incubated with primary antibodies for 1 hour. After three separate washes, cells were incubated with secondary antibody for another hour and then stained with DAPI for 2 minutes. The coverslips were washed extensively and fixed on slides. Images were captured using a Leica laser scanning confocal microscope (Leica TCS SP2 AOBS).
BMDMs were isolated from wild-type, Jnk1−/− and Jnk2−/− mice and cultured as described previously. Cell culture supernatants were collected at 18 hours post virus infection and were analyzed for IFN-β production by ELISA (PBL Biomedical Laboratories).
Luciferase reporter assays were performed as described previously [36].
All experiments in this study were performed independently at least three times. For western blots, fluorescence images and histology sections, one representative result has been shown. For real-time PCR, ELISA, Luciferase, Annexin-V and viral titer assays, the data are represented by three independent experiments. The scale of animal experiments has been instructed in figure legends. Student's t-test was used for the comparison of three independent treatments. For all tests, a p value <0.05 was considered statistically significant.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession numbers for the genes and gene products discussed in this paper are: RIG-I (NM_014314.3→NP_055129.2; NM_172689.3→NP_766277.3), MDA5 (NM_022168.3→NP_071451.2), MAVS (NM_020746.4→NP_065797.2; NM_001206385.1→NP_001193314.1), TBK1 (NM_013254.3→NP_037386.1; NM_019786.4→NP_062760.3), IKKα (NM_001278.3→NP_001269.3), IKKε (NM_014002.3→NP_054721.1), JNK1 (NM_139046.2→NP_620634.1; NM_016700.4→NP_057909.1), JNK2 (NM_002752.4→NP_002743.3; NM_207692.2→NP_997575.2), MKK3 (NM_008928.4→NP_032954.1), MKK4 (NM_003010.3→NP_003001.1; NM_009157.4→NP_033183.1), MKK6 (NM_011943.2→NP_036073.1), MKK7 (NM_145185.2→NP_660186.1; NM_001042557.2→NP_001036022.1).
|
10.1371/journal.pgen.1000354 | LINE Retrotransposon RNA Is an Essential Structural and Functional Epigenetic Component of a Core Neocentromeric Chromatin | We have previously identified and characterized the phenomenon of ectopic human centromeres, known as neocentromeres. Human neocentromeres form epigenetically at euchromatic chromosomal sites and are structurally and functionally similar to normal human centromeres. Recent studies have indicated that neocentromere formation provides a major mechanism for centromere repositioning, karyotype evolution, and speciation. Using a marker chromosome mardel(10) containing a neocentromere formed at the normal chromosomal 10q25 region, we have previously mapped a 330-kb CENP-A–binding domain and described an increased prevalence of L1 retrotransposons in the underlying DNA sequences of the CENP-A–binding clusters. Here, we investigated the potential role of the L1 retrotransposons in the regulation of neocentromere activity. Determination of the transcriptional activity of a panel of full-length L1s (FL-L1s) across a 6-Mb region spanning the 10q25 neocentromere chromatin identified one of the FL-L1 retrotransposons, designated FL-L1b and residing centrally within the CENP-A–binding clusters, to be transcriptionally active. We demonstrated the direct incorporation of the FL-L1b RNA transcripts into the CENP-A–associated chromatin. RNAi-mediated knockdown of the FL-L1b RNA transcripts led to a reduction in CENP-A binding and an impaired mitotic function of the 10q25 neocentromere. These results indicate that LINE retrotransposon RNA is a previously undescribed essential structural and functional component of the neocentromeric chromatin and that retrotransposable elements may serve as a critical epigenetic determinant in the chromatin remodelling events leading to neocentromere formation.
| The centromere is an essential chromosomal structure for the correct segregation of chromosomes during cell division. Normal human centromeres comprise a 171-bp α-satellite DNA arranged into tandem and higher-order arrays. Neocentromeres are fully functional centromeres that form epigenetically on noncentromeric regions of the chromosomes, with recent evidence indicating an important role they play in centromere repositioning, karyotype evolution, and speciation. Neocentromeres contain fully definable DNA sequences and provide a tractable system for the molecular analysis of the centromere chromatin. Here, the authors investigate the role of epigenetic determinants in the regulation of neocentromere structure and function. They identify that a retrotransposable DNA element found within the neocentromere domain is actively transcribed and that the transcribed RNA is essential for the structural and functional integrity of the neocentromere. This study defines a previously undescribed epigenetic determinant that regulates the neocentromeric chromatin and provides insight into the mechanism of neocentromere formation and centromere repositioning.
| Despite the fact that the functional role of the centromere in mitotic and meiotic cell divisions is evolutionarily conserved, the underlying DNA sequences of the centromeres are highly variable across the phylogeny and show no obvious conservation [1],[2]. Thus, a conundrum remains as to whether there are any specific sequence requirements for the different types of, primarily tandemly repeated, DNA in providing the template for centromere formation. In recent years, accumulating evidence has pointed to epigenetic factors including DNA methylation and histone modifications as having important roles in the establishment of centromeric chromatin [3],[4]. In addition, the discovery of fully functional human neocentromeres that arise ectopically from non-tandemly repetitive chromosomal sites further supports the fundamental roles of epigenetic phenomena in the regulation of centromere activity [5]. This class of variant centromeres not only represents an apparently sequence-independent epigenetic model for centromerization but also serves as an excellent tool for the detailed mapping of centromeric chromatin domains – an undertaking that has previously been hampered by the repetitive nature of the mammalian centromeric DNA [6].
The core neocentromeric chromatin is fundamentally defined by the presence of specialized centromere-specific histone H3 variant CENP-A nucleosomes; however, the exact molecular mechanisms involved in the formation of a neocentromere have yet to be defined [7],[8],[9],[10]. To date, approaching one hundred cases of neocentromere emergence have been reported on all the human chromosomes except for chromosomes 7, 19, and 22 [6]. Interestingly, some genomic regions, such as the terminal chromosomal segments of 3q, 8p, 13q, and 15q, are more prevalent in neocentromere cases, with these ‘hotspots’ collectively accounting for approximately half of all the cases reported [5],[11]. Although the ectopic emergence of neocentromeres in hitherto non-centromeric genomic sites suggests the involvement of epigenetic mechanisms of formation, it remains possible that the underlying genomic DNA sequences exert a specific role in the establishment and/or maintenance of the functional integrity of the neocentromeric chromatin. For example, such a possibility is suggested by the universal observation of an elevated AT content, an increase in the density of LINEs (Long Interspersed Nuclear Elements), and a decrease in the density of SINEs (Short Interspersed Nuclear Elements) for the six different neocentromeric domains that have been mapped to date [7],[8],[9],[10].
The first human neocentromere was identified at position 10q25 on the derivative marker chromosome mardel(10) following a de novo interstitial pericentric deletion that has removed the presiding centromere of a normal chromosome 10 [12]. Despite the lack of detectable α-satellite DNA, the 10q25 neocentromere was able to form a mitotically stable kinetochore that binds over 40 of the known functionally important centromere-associated proteins tested [13],[14],[15],[16]. Using a combined BAC (Bacterial Artificial chromosome)-array/ChIP (Chromatin Immunoprecipitation) technique, the CENP-A-associated domain was mapped to a 330-kb genomic segment along the 10q25 neocentromeric chromatin [9]. Subsequently, other centromere protein-binding domains such as those of HP1 and CENP-H, and an increased scaffold/matrix attachment region (S/MAR), were mapped, defining an overall neocentromeric chromatin region of approximately 4.0 Mb in size [17].
To further define the finer structural organization of the core neocentromeic chromatin, we have recently performed high-resolution chromatin mapping using PCR fragment-array/ChIP analysis. The CENP-A domain was found to be assembled as multiple clusters (seven in total) along the 10q25 neocentromeric chromatin [18]. Interestingly, in silico sequence analysis indicated that these CENP-A-binding clusters contain a 2.5-fold increase in the prevalence of L1 retrotransposon sequences (which belong to the only active subfamily of LINEs) when compared to the surrounding non-CENP-A-binding regions or the genome average [18],[19],[20]. L1 retrotransposon is a major group of interspersed repetitive elements that comprise 17% of the human genome. Although the great majority of L1s are inactive due to 5′ end truncations, active transcription and translation of these retrotransposons has recently been detected in a variety of cell types and implicated to be a potential regulator for cellular processes [19],[20]. However, detailed investigations on the functional role of individual L1 retrotransposon in the human genome have been limited by technical difficulties associated with its repetitive nature. In this study, we present an in-depth bioinformatic analysis and the experimental investigation of the possible functional roles of the L1 retrotransposons in the regulation of neocentromere activity.
Our previous in silico analysis of the various types of DNA motifs and sequence properties revealed a significant, 2.5-fold, increase in the prevalence of L1 retrotransposons within the CENP-A-binding domain of the 10q25 neocentromere [18]. Here, we extended the analysis to the investigation of the genomic distribution and sequence characteristics of L1 retrotransposons across a 6-Mb genomic region spanning the 10q25 neocentromere using the RepeatMasker track of the UCSC genome browser. Besides an enrichment of L1 retrotransposons, the CENP-A-binding clusters of the 10q25 neocentromere were also associated with a higher number of intact L1 genomic segments (Figure 1A). These CENP-A-binding clusters contained 56 L1s per 100 kb DNA, whereas the flanking non-CENP-A-binding regions contained only 26 L1s per 100 kb DNA, with an overall 2.1-fold increase in L1 content in the CENP-A-binding regions (Table S1). In addition to the bioinformatics analysis, ChIP/quantitative PCR analysis using a specific antibody against CENP-A also showed a specific enrichment of L1 genomic sequences in the CENP-A-associated chromatin of 10q25 neocentromere ( Figure S1).
Although there was no significant difference in term of the rate of divergence, deletion, and insertion between the L1 retrotransposons within the CENP-A and non-CENP-A-associated regions across the 6-Mb region of the 10q25 neocentromere (Table S1), the average length of the L1 retrotransposons located within the CENP-A-binding regions (average length of 865 bp) was significantly longer (increased by 2 folds) compared with those found within the non-CENP-A-binding regions (average length of 440 bp) (Figure 1A; Table S1). Such a difference was attributed to an increase in the proportion of the primate-specific L1 subfamily, as shown by a higher L1P/L1M ratio (L1P, primate-specific; L1M, mammalian-wide), within the region. Given the L1P subfamily included active full-length L1 (FL-L1) retrotranposons, we next searched for the presence of FL-L1 at this region. Functional annotation of the FL-L1 retrotransposons spanning across the 6-Mb region of the 10q25 neocentromere using the online L1Base program (http://l1base.molgen.mpg.de/) identified six FL-L1s, four of which (L1a–d) residing within or close to the CENP-A-associated clusters, while the remaining two (L1e–f) were located >1.5 Mb away from the CENP-A-associated domain (Figure 1B and Figure 2).
Although the functional role of L1s in the regulation of genomic architecture is not well defined, it is of significant interest that L1s can be transcribed into RNA and subsequently translated into proteins for retrotransposition activity [20],[21],[22],[23]. Recent reports indicate that L1 RNAs are actively transcribed in a variety of cell types from full-length L1 elements (∼6 kb in size) that contain an internal promoter, two ORFs, and a poly-A tail at the 3′ UTR [20],[21],[22],[23]. To address if any of the six FL-L1s at the 10q25 neocentormere chromatin were transcriptionally active, RT-PCR primers were designed to specifically target each of the elements (L1a–f) in monochromosomal CHO-human hybrid lines, CHOK1-M10 and CHOK1-N10 (containing the human neocentromeric mardel(10) and the progenitor normal human chromosome 10, respectively) (Figure 3). The specificity of each primer was confirmed by direct sequencing of the PCR products, which established that only the desired target sites were amplified.
No transcripts from FL-L1a, FL-L1c, FL-L1d, FL-L1e and FL-L1f were detected. However, as shown in Figure 3A–C, based on the use of three independent primer sets that targeted to a combined genomic segment of 415 bp within the 5′ UTR, transcripts for FL-L1b were clearly detected in CHOK1-M10 and CHOK1-N10 cells. Further analysis of four additional monochromosomal hybrid cell lines – two human/hamster hybrids CHOK1#8 and GM10926 (each containing an unrelated normal human chromosome 10) and two human/mouse hybrids GM11688 (containing a unrelated normal human chromosome 10) and ES-M10 (containing the mardel(10) chromosome) – showed positive transcription activities of FL-L1b in three of the hybrid lines (GM10926, GM11688, and ES-M10) but not in CHOK1#8 (Figure 3B; Table S2). No detectable transcriptional activity was detected for FL-L1a, FL-L1c, FL-L1d, FL-L1e and FL-L1f in any of these cell lines. These results indicated that the FL-L1b locus was actively transcribed both before and following neocentromere formation. In addition, it was of interest to note that FL-L1b was located within the central and largest CENP-A cluster (Figure 2B), and belonged to the active L1PA2 subfamily [20],[24],[25],[26].
To investigate whether the FL-L1b locus is the only active L1 element within the 10q25 neocentromeric chromatin, additional primers were specifically designed to target those truncated L1s that contained intact promoter sequences and also others that were greater than 4 kb in size identified within the 6-Mb genomic region (Figure 2A). These targets included five long truncated L1s with or without the promoter sequence (L1g–k) and other short orphan L1 promoter sequences (L1l–m). The results of RT-PCR analysis indicated no detectable transcripts from any of these L1 targets in the three monochromosomal hybrid cell lines assayed - CHOK1#8, CHOK1-N10, and CHOK1-M10 (Table S3).
Given that antisense transcription has been detected from the 5′ UTR of L1 elements [26],[27], we performed RT-PCR analysis on all promoter-containing FL-L1s and truncated L1s (L1a–f, i, k, l, m; Figure 2) within the 6-Mb region using primer sets each targeted to the 5′ upstream flanking sequence at one end and to the 5′ UTR of L1 at the other end. No antisense transcript could be detected for all promoter-containing L1s across the 6-Mb genomic region (Table S3). These results showed that, across the 6-Mb neocentromeric domain, active transcription was found only at the FL-L1b locus, and that the resulting RNA products were predominantly long sense transcripts of at least 415 bp in size.
Next we investigated if the corresponding FL-L1b RNA transcripts were incorporated into the 10q25 neocentromeric chromatin. Chromatin immunoprecipitation was performed using a specific anti-CENP-A antibody. RNAs from both the input and immunoprecipitated fractions were isolated, reverse transcribed into cDNAs, and subjected to real-time quantitative PCR analysis using three independent primer sets targeted to the 5′ UTR of FL-L1b. A significant enrichment (P<0.001) of FL-L1b RNA in the CENP-A bound fractions was observed, as indicated by a 4 to 5 fold increase in the yield of PCR products (Figure 3D). In contrast, none of the negative control sequences, 18S, 5S, and β-actin, was enriched in the immunoprecipitated fractions. We have also performed similar RNA-ChIP experiments and analyzed the RNA-ChIP products using primers targeting to the other L1s (FL-L1a, -L1c, -L1d) as well as four genes (KIAA1600, TRUB1, GFRA1) that reside around the CENP-A-binding domain and detected no enrichment of any of these transcripts in the CENP-A chromatin of the CHOK1-M10 cells (Figure S2). Together, these results indicated the specific incorporation of the FL-L1b RNA into the CENP-A-associated chromatin of the 10q25 neocentromere.
To study the potential role of the FL-L1b RNA at the 10q25 neocentromere, we designed two sets of siRNA oligonucleotide duplexes (Figure S3) for the specific transcriptional knockdown of FL-L1b in the monochromosomal CHOK1-M10 hybrid line; the study of RNAi knockdown in a CHO background offered the advantage of minimizing any potential off-target RNAi knockdown effects because the CHO genome contained significantly diverged L1 elements. The transfection conditions for RNAi knockdown were optimized to achieve >80% reduction in the FL-L1b transcripts as compared to the transfection-reagent-only and Stealth siRNA negative controls (Figure 4A). Similar efficiency of FL-L1b transcriptional knockdown was also achieved in the other mouse/human and hamster/human somatic hybrids described above (data not shown).
To determine the cellular effects of the FL-L1b knockdown, a kill-curve analysis was performed on a CHOK1-M10 hybrid cell line containing a mardel(10) chromosome that had been tagged with a Zeocin resistance gene [13],[17]. At the optimal concentration of 200 µg/ml of Zeocin, the majority (>80%) of non-mardel(10)-containing CHOK1-N10 cells were killed 48 hours post Zeocin treatment, whereas the normal growth of CHOK1-M10 cells was not affected (Figure 4C-i). A significant loss of cell viability was observed in CHOK1-M10 following FL-L1b RNAi-knockdown, with the percentage of surviving CHOK1-M10 cells being reduced to approximately 50% compared to the transfection-reagent-only and Stealth siRNA negative controls 48 hours post Zeocin selection (Figure 4C). These results indicated a presumed FL-L1b-induced impairment of neocentromere function that has led to the loss of the Zeocin-resistant mardel(10) chromosome.
To further extend the Zeocin kill-curve results, a direct assessment of the loss of the mardel(10) chromosome following FL-L1b knockdown was determined by FISH (Fluorescence In Situ Hybridization) analysis using a mardel(10)-specific BAC probe. The stability of mardel(10) was greatly affected 48 hours post FL-L1b RNAi-knockdown, with a significant reduction from ∼95% to ∼55% in the CHOK1-M10 cell line, and from ∼100% to ∼60% in the mouse-human hybrid cell line ES-M10 (Figure 4D). Under similar conditions, the stability of the normal human chromosome 10 in control CHO-human (GM10926, CHOK1-N10) and mouse-human (GM11688) hybrid lines were not affected after FL-L1b transcriptional knockdown, suggesting that the loss of mardel(10) was directly linked to the effect of the FL-L1b knockdown on the neocentromere activity.
In order to further investigate the structural integrity of the neocentromere after FL-L1b transcriptional knockdown, a combined immunofluorescence and FISH analysis was performed on metaphase CHOK1-M10 cells using an anti-CENP-A antiserum (CREST6) and a BAC DNA probe (RP11-359H22) that hybridized to the 10q25 neocentromeric region of mardel(10). Cells were harvested at 24 hours following RNAi-knockdown in order to capture the early to intermediate stages of the disruption of neocentromere function prior to the complete loss of the mardel(10) chromosome. The mean fluorescence intensity of the CREST6 signals on the 10q25 neocentromere was reduced by 20 to 30% (P<0.001) after the FL-L1b transcriptional knockdown using either siRNA#1 or siRNA#2 (Figure 4B; examples of reduced CENP-A levels on 10q25 neocentromere post FL-L1b RNAi knockdown are shown in Figure S5). In some cases, the CREST signals on the 10q25 neocentromere were as low as 20% that of the control cells. In addition to the quantitative immunofluorescence data, ChIP and real-time PCR analysis was also performed using an anti-CENP-A antibody for analysis comparing the enrichments of CENP-A at the 10q25 neocentromere with and without FL-L1b RNAi knockdown in CHOK1-M10 cells (Figure S6). Consistently, the ChIP/PCR results showed a reduction of CENP-A protein at 10q25 neocentormere following RNAi knockdown of FL-L1b transcript, providing independent confirmation of the importance of FL-L1 transcript in regulating the structural integrity of 10q25 neocentromere.
We have previously reported that genes located across the 10q25 neocentromere region are transcriptionally competent [17]. Here, we used the transcription status of these genes as a measure to determine the effect of FL-L1b knockdown on the overall neocentromeric chromatin environment. The transcriptional levels of 13 actively transcribed genes within the 6-Mb 10q25 neocentromere region (see Figure 2) were determined by qRT-PCR analysis at 24 hours post FL-L1b RNAi-knockdown. While most of the genes were unaffected, the transcriptional activities of 2 genes, ATRNL1 (which spanned the CENP-A-binding domain) and TRUB1 (located outside the CENP-A domain, with its 5′-end CpG island being ∼410 kb away from the FL-L1b locus), were significantly reduced (by approximately 60–70%; P<0.05) after the FL-L1b transcriptional knockdown (Figure 4E).
To ensure that the FL-L1b RNAi knockdown-mediated mardel(10) chromosomal instability was not attributed to a reduction in the level of TRUB1 and/or ATRNL1 transcripts, siRNA oligonucleotide duplexes were designed to target these and two other immediately surrounding genes, KIAA1600 and GFRA1. Approximately 70–90% transcriptional knockdown was achieved for each of these genes in the CHOK1-M10 cells (Figure S4). No significant difference in the percentage cell survival was observed in the Zeocin kill-curve analysis, providing support for a specific role of FL-L1b rather than these genes in the maintenance of the mardel(10) stability (Figure S4).
Our earlier bioinformatic analysis revealed a >2.5-fold increase in the prevalence of L1 retrotransposons in the underlying DNA sequence of the 10q25 CENP-A-binding clusters [18]. In this study, we described the increased frequency of intact L1 segments and average length of L1 DNA within the 330-kb CENP-A domain. Across the 6-Mb region of the 10q25 neocentromeric chromatin, a concentrated cluster of four FL-L1s was found at the CENP-A-binding domain of the 10q25 neocentromere [18]. Furthermore, in silico analysis of other neocentromere sites (Figure S7) has revealed the presence of at least one FL-L1 element at the CENP-A-binding domain of five out of the six neocentromeres mapped to date [7],[8],[10]. The average FL-L1 density across these neocentromeres was also higher by 1.5 times compared to that of the human genome. These observations indicated a potential role of the L1 retrotransposon, particularly the full-length members (FL-L1s), in the regulation of neocentromeric chromatin.
In humans, active transcription and translation of L1 retrotransposons has been detected in a wide-range of cell types, including germ cells, tumours and transformed cell lines, and a smaller number of non-malignant somatic cells [21],[22],[28],[29],[30],[31],[32]. Importantly, multiple lines of evidence indicated that L1 RNAs are actively transcribed from full-length elements (∼6 kb in size) that contain an internal promoter, two ORFs, and a poly-A tail at the 3′ UTR [20],[21],[22],[23]. However, a detailed investigation of the transcriptional status of a single FL-L1 has not been described due to the technical difficulties associated with its repetitive nature. However, unlike tandemly-repeated satellite DNAs, which are highly homogeneous, L1 interspersed repeats are comparatively more diverged in sequence. Here, we took advantage of sequence divergence amongst the L1 repeats and designed oligonucleotide primers that targeted the diverged sites within a single FL-L1 retrotransposable element for RT-PCR and RNAi-knockdown analysis in monochromosomal somatic cell hybrids to determine its transcriptional activity and associated function – an undertaking that has not been previously described.
We determined the transcriptional status of all six FL-L1s and other non-full-length L1 targets within the 6-Mb genomic window spanning the core neocentromere. Interestingly, only one of them (i.e. FL-L1b) was actively transcribed from the mardel(10) in CHOK1-M10, although all six FL-L1s contained the internal promoter sequences (for sequence comparisons between transcriptionally active and silent FL-L1s assayed in this study, see Tables S4 and S5). Our previous study has described the active transcription of multiple genes within the broader 10q25 neocentromeric domain, including ATRNL1 that spanned the entire length of the CENP-A-associated chromatin [17]. However, it was uncertain if the core neocentromeric chromatin was permissive to active transcription given that the putative promoter of ATRNL1 was located outside the CENP-A domain. Here, based on the active transcription status of FL-L1b that is located within the central CENP-A-binding cluster at the 10q25 neocentromere, our study provided clear evidence for the permissibility of transcription within the core neocentromeric chromatin. More recently, this phenomenon of active transcription through the core centromere has also been demonstrated in α-satellite-containing human artificial chromosomes, where the CENP-A-associated domain was shown to spread into the adjacent transcriptionally active selectable marker gene [33],[34]. Furthermore, transcriptional competence of the core centromeric chromatin has also been described in Oryza sativa (rice) and Zea mays (maize) [35],[36]. These studies, including our current data, clearly show that CENP-A-associated chromatin is permissive to the transcription of genes and non-genic retrotransposable elements.
The pattern of FL-L1 transcription within the 6-Mb domain in the hamster-human hybrids CHOK1-N10 (containing the progenitor normal human chromosome 10) and GM10926 (containing an unrelated normal human chromosome 10) was identical to that found in CHOK1-M10. The formation of the 10q25 neocentromere did not significantly change the transcription level of FL-L1b, in consistent with our previous finding on the transcription competence of multiple genes located within this region [17]. Similar results were obtained from mouse-human hybrids GM11688 (containing an unrelated normal human chromosome 10) and ES-M10 (containing the neocentromeric mardel(10) chromosome), indicating that the active transcription of FL-L1b was not affected by differences in species background. Interestingly, FL-L1b transcription was not detected in one of the normal human chromosome 10 in the CHOK1#8 cell line – an observation that may be explained by differential epigenetic silencing or by mutations at the promoter or upstream regulatory sequences of the CHOK1#8 FL-L1b DNA.
Using RNA-ChIP-qPCR analysis, we showed that FL-L1b single-stranded RNA transcripts were incorporated as part of the ribonucleoprotein component of the CENP-A-associated domain. Interestingly, the presence of long single-stranded centromeric RNA transcripts including CentC satellite repeats and CRM retrotransposons in Zea mays [36], 160B/Athila2 retrotransposon in Arabidopsis thaliana [37], PRAT satellite repeats in Palorus ratzeburgi [38], and α-satellite repeats in humans [39] were also reported in recent studies. Furthermore, chromatin immunoprecipitation experiments in Zea mays and humans independently showed that these centromeric RNA transcripts were associated with the core centromeric chromatin [36],[39]. Together, these results indicated that a pool of single-stranded RNA could be directly transcribed from the satellite repeats (and centromere-specific retrotransposons) of the normal centromeres or the L1 retrotransposon of a neocentromere and subsequently incorporated into the core centromeric/neocentromeric chromatin.
The functional role of FL-L1b RNA at the 10q25 neocentromere was determined by RNAi knockdown of FL-L1b in human/mouse and human/hamster monochromosomal hybrid lines. FISH and/or Zeocin kill-curve analysis indicated that FL-L1b knockdown led to a significant reduction (by ∼40–50%) of the mitotic stability of mardel(10) and the compromised structural integrity of the 10q25 neocentromere. These FL-L1b knockdown-mediated mitotic effects at the 10q25 neocentromere were fast and similar to the rapid response previously described in RNAi knockdown or conditional knockout of core centromere proteins, such as CENP-A [40], CENP-H [41],[42],[43] and CENP-K [44]. Our results therefore demonstrate a functional significance of L1 RNA transcripts at the core neocentromere region which has not been fully defined in previous studies.
In addition to the two FL-L1b siRNA duplexes, we have included the analysis of siRNA duplexes that targeted four genes spanning and surrounding the CENP-A-associated region. None of these siRNAs exerted any effect on mardel(10) stability, as indicated by the cell viability assay (Figure S4). More specifically, RNAi knockdown of ATRNL1, a gene that spanned across the CENP-A-associated domain, did not result in any compromise in the functional integrity of the 10q5 neocentromere. These data indicate that the FL-L1b RNAi-induced mardel(10) instability is likely to be a result of the depletion of FL-L1b RNA transcripts rather than due to indirect effects arising from the recruitment of chromatin remodelling or modifying complexes to the 10q25 neocentromere via the RNAi pathway.
The precise functional role(s) of FL-L1 RNA transcripts at the core neocentromeric chromatin remains to be delineated. Transcription at the FL-L1 locus and/or the L1 transcript itself may act as an early-specification epigenetic signal for the recruitment of CENP-A nucleosomes. Interestingly, the transcriptional knockdown of FL-L1b leads to a more ‘closed’ local chromatin state, as indicated by the reduction in the transcription of two surrounding genes ATRNL1 and TRUB1. At the 10q25 neocentromere, the transcription activity may facilitate the process of histone replacement by partially disassembling the nucleosomes to provide a more ‘open’ chromatin structure [45] for subsequent deposition of CENP-A nucleosomes. The recent identification of GATA-type transcription factor Ams2, which promotes the centromere localization of CENP-A in Schizosaccharomyces pombe, also provides supports toward a role of transcription in defining a centromere state [46].
Rather than the transcriptional activity itself, it is also possible that the FL-L1b RNA transcript may serve as a specific epigenetic signal at the 10q25 neocentromere since by RNAi knockdown several neighbouring genes did not affect the mitotic stability of mardel(10). Although this hypothesis remains to be tested, the underlying process may be similar to the function of long Xist RNA in promoting the establishment of a specialized chromatin state such as the incorporation of macroH2A during X-inactivation [47],[48]. Alternatively, the chromatin-bound FL-L1b RNA at the 10q25 neocentromere may be involved in the formation of a flexible ribonucleoprotein complex that brings together and/or stabilizes the proteins of the core neocentromere, as suggested by the observed CENP-A delocalization after FL-L1b RNAi knockdown. Nonetheless, it is interesting to note that the FL-L1b locus is being actively transcribed from both the progenitor chromosome 10 and the neocentric mardel(10). The absence of active CENP-A recruitment to the FL-L1b locus on the progenitor chromosome suggests that the FL-L1b transcript is unlikely the sole epigenetic specification determinant for CENP-A recruitment. The transcribed FL-L1b locus and/or FL-L1b RNA-bound chromatin may require additional players (e.g. specific RNA-binding proteins) in recruiting CENP-A for the formation of a neocentromere.
The reduction in the transcriptional activity of the two genes surrounding CENP-A domain (ie. ATRNL1 and TRUB1) following FL-L1b knockdown indicated that the FL-L1b RNA could be regulating a larger genomic domain than that of the CENP-A-associated chromatin. It is unknown how these FL-L1 RNA transcripts mediate such long-range chromosomal effects, however, it is interesting that this extended genomic domain overlaps with a region of high L1 DNA content (using the human genome average as the baseline threshold) (Figure 2B). The incorporation of FL-L1b RNA into the neocentromeric chromatin may potentially involve a simple base pair recognition mechanism [49], similar to what has been described for the assembly of the telomerase complex by telomerase RNA or the formation of heterochromatin structure by short interfering siRNA [50],[51]. In future studies, the identification of potential chromatin remodelling proteins that interact with the centromeric or neocentromeric RNA transcripts should shed new light on the epigenetic mechanisms of regulation of centromere/neocentromere architecture and function.
Increasing evidence now point to neocentromere formation as the underlying mechanism for centromere repositioning that underpins karyotype evolution and speciation [6],[52]. The elucidation of the molecular mechanisms of neocentromere formation will not only provide important insights into the inherent epigenetic determinants that initiate de novo centromere assembly, but will also provide a better understanding of the operating mechanisms for centromere repositioning and karyotype evolution.
The somatic hybrid cell lines were cultured as previously described [17],[53]. These include (i) human/hamster monochromosomal hybrid CHOK1#8, CHOK1-N10 and CHOK1-M10, containing unrelated chromosome 10, progenitor chromosome 10, and mardel(10) respectively; (ii) human/mouse mardel(10)-containing monochromosomal hybrid ES-M10 [14],[17]. Two additional somatic hybrid cell lines, GM10926 (CHOK1 background) and GM11688 (mouse A9 background), each containing an unrelated human chromosome 10, were obtained from the Human Genetic Cell Repository of Coriell Institute of Medical Research and were cultured in Ham's Kao and Michayluk medium (KAO) supplemented with 10% dialyzed FCS (Gibco BRL) at 37°C and Ham's F12 Medium/DMEM (1∶1 mixture) 2 mM L-glutamine, 10% FCS with 500 µg/ml Geneticin (Gibco) at 37°C. 200 µg/ml Zeocin (Invitrogen) was added into the media for selection of the mardel(10) chromosome in CHOK1-M10 and ES-M10.
A time-course experiment was first performed to determine the time duration required to kill the non-resistant CHOK1 cells (CHOK1-N10) at 200 µg/ml of Zeocin. The transcription-knockdowns of FL-L1b and other genes of interest were performed by siRNA transfection of the relevant siRNA oligonucleotide duplexes for 48 hours at 25 nM in CHOK1-M10. Subsequently, cells were incubated with 200 µg/ml of Zeocin for an additional 48 hours following RNAi knockdown. The number of viable cells was determined by staining with Trypan Blue (0.8 mM Trypan Blue in 1× PBS) for 5 minutes at room temperature and counting with a hemocytometer under the microscope. The mitotic stability of the mardel(10) was calculated as the ratio of the percentage of viable cells under Zeocin selection to the number of viable cells without Zeocin selection.
The genomic location of each chromatin domain and the sequence characteristics were determined using the UCSC Genome Browser (http://genome.ucsc.edu.au) May 2004 builds and its in-build RepeatMasker track [54]. Full-length L1s were identified and annotated using the online L1Base software package (http://l1base.molgen.mpg.de/) [55]. Specifically, several key features were analyzed and these included (1) general characteristics, such as the GC content, target site duplications, intactness scores, the polyadenylation signal, and the presence of poly-A tails; (2) classifications of L1s; (3) 5′UTR promoter features and the conservations of transcription factor binding sites; (4) the conservation of amino acid residues in the two ORFs (Table S4).
In the L1Base program, the ‘intactness score’ was calculated for the query FL-L1 sequence. One point was awarded to every conserved sequence feature (according to the consensus L1 sequence) that was known to affect the transcriptional and/or translational activity [55]. The transcriptionally active FL-L1b had an intactness score of 25, being the highest of the six FL-L1s (Table S5A). As for the 100 bp internal promoter [56],[57] within the 5′ UTR, the nucleotide sequence conservation of the six FL-L1s (FL-L1a–g) ranges from 72.3 to 91.6% and FL-L1b ranked the equal highest of the six FL-L1s (Table S5B). FL-L1b also contained all of the known conserved transcription factor binding sites within the 5′ UTR, while more than one mutation was found within the 5′ UTR of the other FL-L1s. In addition, a CpG island that was potentially important for transcriptional regulation was present within FL-L1b. Other noted sequence features of FL-L1b that could contribute to its transcription and functional activities were listed following: (1) FL-L1b contained an intact polyadenylation signal and a relatively long poly-A tail, which were important for mRNA maturation and subsequent protein translation; (2) FL-L1b was the only FL-L1 of the 6 FL-L1s with no ORF frame shifts or mutations at the important amino acid residues analysed; (3) FL-L1b belonged to the retrotranspositionally competent Ta subfamily and was flanked by 15 bp target-site duplications (Table S5 C to F).
RNA chromatin immunoprecipitation was performed as described in [58] with slight modifications. RIPA Buffer (50 mM Tris-Cl, pH 7.5, 1% NP-40, 0.5% sodium deoxycholate, 0.05% SDS, 1 mM EDTA, 150 mM NaCl, 1 tablet of Roche Complete Protease Inhibitor per 10 ml of RIPA buffer) was used for cell lysis and immunoprecipitation was performed using a rabbit polyclonal anti-mouse CENP-A antibody at 1∶500 dilution [15]. Immunocomplex recovery was achieved following two washes with RIPA High Stringency Wash Buffer (50 mM Tris-HCl pH 7.5, 1% Nonidet P-40, 1% sodium deoxycholate, 0.1% SDS, 1 mM EDTA, 0.1 mM PMSF, 1 tablet of Roche Complete™ Protease Inhibitor per 10 ml buffer) containing 250 mM and 500 mM NaCl in stepwise manner. Elution of RNA was performed with RNA-ChIP Elution Buffer (50 mM Tris-HCl pH 7.5, 5 mM EDTA, 1% SDS, 10 mM dithiothreitol) and reverse cross-linked at 70°C for 45 minutes. Total RNA was then isolated and subsequently subjected to quantitative PCR analysis.
Total RNA was extracted using either the RNeasy Mini Kit (Qiagen) for transcription detection assays or TRIZOL reagent (Invitrogen) for RNA-ChIP. TURBO DNA-free Kit (Ambion) was used to remove possible contaminating DNA. cDNA synthesis was performed using Transcriptor First Strand cDNA Synthesis Kit (Roche). Quantitative RT-PCR was carried out using SYBR Green PCR Master Mix (Applied Biosystems) on 7300 or 7900HT Real-Time PCR System (Applied Biosystems) according to manufacturer's instructions. cDNA equivalent to 10 ng RNA was amplified with 150 nM forward and reverse primers in a 25 µL reaction (for primer sequences, see Table S6). Dissociation curves were performed to confirm specific amplifications without primer dimer formation. Samples were also subjected to gel electrophoresis analysis to confirm that the PCR products were of expected size. For the transcription assay of the FL-L1s, sequencing experiments were also performed to confirm the identity of each RT-PCR product. For calculations and statistics in the analysis, see below.
The comparative CT method was used for data analysis in transcription detection assay and quantitative ChIP-PCR analysis. The ΔCT value was calculated as [ΔCT = CT (test gene/genomic segment)−CT (control gene/genomic segment)]. The CT value of each test gene/segment was normalized against the CT value of control gene/segment, either 5S (for DNA-ChIP-qPCR analysis) or β-actin (for transcription assay and RNA-ChIP-qPCR analysis) to give the ΔCT value. The ΔΔCT value was calculated as [ΔΔCT = ΔCT(test cell line)−ΔCT(control cell line)] for transcription analysis, or [ΔΔCT = ΔCT(before siRNA knockdown)−ΔCT(after siRNA knockdown)] for transcription knockdown assay, and [ΔΔCT = ΔCT(input)−ΔCT(bound)] for ChIP-qPCR analysis, respectively. Relative fold-difference in transcription activity was expressed as in transcription analysis and transcription knockdown assays. Relative-binding value in ChIP-qPCR analysis was calculated by .
Two sets of Stealth siRNA oligonucleotide duplexes targeting FL-L1b were designed using the online BLOCK-iT RNAi Designer software (Invitrogen). In contrast, siRNA oligonucleotide duplexes targeting genes KIAA1600, TRUB1, ATRNL1, and GFRA1 were obtained as pre-designed Stealth Select siRNA (Invitrogen). Sequences of the siRNA oligonucleotide duplexes are listed in Table S6. CHOK1-M10 cells were seeded in 6-well culture plates without antibiotic selection at a density of 2×104 cells/well, 24 hours prior to siRNA transfection. Transcriptional knockdown was performed by transfecting cells with Stealth siRNA oligonucleotide duplexes (Invitrogen) at a final concentration of 25 nM in DMEM (Dulbecco's Modified Eagle's Medium) using 2.5 ng/µl Lipofectamine 2000 (Invitrogen) for a period of 24 to 48 hours according to the manufacturer's instructions. The effects of RNAi knockdown of FL-L1b and other target genes were assayed by quantitative RT-PCR. Stealth siRNA Negative Control Low GC Duplex (Invitrogen) was also included as control for sequence independent RNAi knockdown effects.
Immunofluorescence [59] and FISH [13] were performed as previously described. Anti-centromere autoimmune serum CREST6 (which predominantly recognize CENP-A protein) and RP11-359H22 BAC were used for the identification of 10q25 neocentromere on mardel(10) [13]. Metaphase spreads were visualized using an Imager M1 microscope (Zeiss) and the digital images were captured by the AxioCam MRm camera (Zeiss). CREST6 immunofluorescence signals on 10q25 neocentromere were quantified and normalized against CHO centromeres in CHOK1-M10 cells using AxioVision software version V4.6.1.0 (Zeiss).
The quantification of CREST6 immunofluorescence signals was performed following FL-L1b RNAi knockdown. A circular area of defined size (diameter of 2 µm) was selected around the centromere of interest. Total intensity (I) of each pixel within the delineated area was determined and defined as arbitrary fluorescence unit (a.f.u.). Digital images obtained from immunofluorescence analysis were nonsaturating and auto-corrected for background removal. Non-specific background signal (IBK) for each metaphase spread was calculated by the average arm intensity from five chromosomes and subsequently subtracted from the total intensity (I). Average signal intensity of 15–20 endogenous CHO centromeres (ICHO) from each spread was calculated and used as normalization control to correct for the variation in hybridization between spreads. The ratio of CREST6 fluorescence intensities (R) on 10q25 neocentromere to CHO centromeres was calculated using the following equation: R = (IM10−IBK)/(ICHO−IBK). The mean fluorescence intensity of CREST6 (M) on 10q25 neocentromere was calculated using the following equation: M = R×ICHOALL. ICHOALL represents the average intensity for all CHO centromeres (∼750) calculated for each treatment in the RNAi knockdown experiments.
|
10.1371/journal.ppat.1002769 | Linking the Transcriptional Profiles and the Physiological States of Mycobacterium tuberculosis during an Extended Intracellular Infection | Intracellular pathogens such as Mycobacterium tuberculosis have evolved strategies for coping with the pressures encountered inside host cells. The ability to coordinate global gene expression in response to environmental and internal cues is one key to their success. Prolonged survival and replication within macrophages, a key virulence trait of M. tuberculosis, requires dynamic adaptation to diverse and changing conditions within its phagosomal niche. However, the physiological adaptations during the different phases of this infection process remain poorly understood. To address this knowledge gap, we have developed a multi-tiered approach to define the temporal patterns of gene expression in M. tuberculosis in a macrophage infection model that extends from infection, through intracellular adaptation, to the establishment of a productive infection. Using a clock plasmid to measure intracellular replication and death rates over a 14-day infection and electron microscopy to define bacterial integrity, we observed an initial period of rapid replication coupled with a high death rate. This was followed by period of slowed growth and enhanced intracellular survival, leading finally to an extended period of net growth. The transcriptional profiles of M. tuberculosis reflect these physiological transitions as the bacterium adapts to conditions within its host cell. Finally, analysis with a Transcriptional Regulatory Network model revealed linked genetic networks whereby M. tuberculosis coordinates global gene expression during intracellular survival. The integration of molecular and cellular biology together with transcriptional profiling and systems analysis offers unique insights into the host-driven responses of intracellular pathogens such as M. tuberculosis.
| The impact of Mycobacterium tuberculosis on global health is undeniable, with ∼2 million deaths and ∼9 million new cases of tuberculosis each year. A key to the success of M. tuberculosis as a persistent, intracellular pathogen is its ability to survive for extended periods within professional phagocytes. Sustained growth within macrophage phagosomes requires avoiding or resisting antimicrobial mechanisms and adapting to replicate in a stressful, nutrient-restricted environment. Our understanding of the survival strategies, metabolism, and physiology of M. tuberculosis during intracellular growth remains incomplete. We employed multi-disciplinary approaches to gain new insights into adaptive responses that M. tuberculosis mobilizes to secure a productive infection. We simultaneously quantified replication and death rates, used electron microscopy to evaluate bacterial integrity, and determined the temporal changes in bacterial gene expression during a 14-day infection. By overlaying this temporal transcriptome dataset onto an extended Transcriptional Regulatory Network model, we identified regulatory pathways, stress responses, and metabolic adaptations activated during key physiological transitions over the 14 days of infection.
| Mycobacterium tuberculosis (Mtb) has evolved successful strategies to survive in the human host and thwart the attempts of the immune system to eradicate it. The astounding morbidity and mortality caused by Mtb – with roughly 9 million new cases and 2 million deaths annually – is a testament to the success of this enduring pathogen. Its remarkable resistance to killing by both innate and acquired immune effectors and successful adaptation to constantly changing environments allows Mtb to maintain infections for decades. The ability of Mtb to survive and replicate within macrophages (Mφ) is central to the pathogenesis of tuberculosis [1].
A considerable body of literature has been devoted to understanding how Mtb survives its interaction with the Mφ [1]–[5], which serves as its niche for much of the infection cycle in the human host. The phagosome into which the bacterium is internalized fails to acquire the full complement of lysosomal characteristics due to its restricted maturation; nonetheless, this environment still represents a considerable challenge to bacterial survival. It has been suggested that Mtb survives better through escape into the cytosol [6], however recent work has indicated that accessing the cytoplasm leads rapidly to host cell death [7], which suggests that survival within the phagosome remains the more stable environment to which Mtb must adapt to succeed.
A characteristic essential for Mtb's survival is its ability to sense its environment and modulate genetic pathways controlling resistance mechanisms, metabolism, and other processes in a timely manner. The importance of gene regulation during intracellular survival and in vivo infection is illustrated by the attenuated phenotypes of numerous mutants in which key transcriptional regulators have been inactivated [8]–[11]. In previous microarray studies focusing on early stages of infection, we have characterized the initial transcriptional changes in Mtb immediately following Mφ invasion [12]. In the current study we focus on the long-term survival of Mtb within resting Mφ over the course of a 14-day infection. During this extended timeframe the bacteria appear to experience diverse and changing environments and adopt differing physiological states. Shortly after infection, the bacterial numbers decline indicating a period of acute stress as the macrophage attempts to eradicate the infection. This is followed by a second period during which the bacterial numbers remain relatively constant, presumably while the bacterium adapts to, and tries to modulate its environment. Finally, approximately 5–6 days post-infection (p.i.) the bacterial population increases, indicating that conditions to support survival and replication have been established. This model affords us unique access to unraveling Mtb's adaptation strategies to the changing stresses and environmental cues experienced during the establishment of a productive infection. We have exploited a range of technical approaches to define the growth states of the bacterium through the course of this 14-day infection, and to correlate these physiological states with transcriptional profiles of the surviving bacteria to identify genetic networks consistent with survival and the establishment of a productive infection.
The establishment of a productive infection in the macrophage is not dependent merely on the short-term response following phagocytosis but the subsequent stages of adaptation that lead to a net increase in intracellular Mtb. To appreciate the adaptations required by Mtb to enter this replicative state one has to employ a temporal approach that accurately correlates bacterial numbers, including the assessment of death versus growth, with changing transcriptional profiles. We employed three different methods to probe the growth states of Mtb during a 14-day intracellular infection model.
First, resting primary Mφ isolated from C57BL/6 mice were infected with Mtb at a low multiplicity of infection (MOI) of 1∶1. At two-day intervals, infected monolayers were lysed and dilutions plated to determine viable colony forming units (CFU). Supernatants were also diluted and plated (prior to lysis of Mφ) to monitor extracellular bacilli that could contribute to reinfection. An initial 0.5 log decrease in cell-associated viable CFU at day 2 was followed by a period of minimal change (days 2–6) and then a more steady increase in CFU (days 6–14) (Figure 1A). This profile, consistent with our previous observations [13], [14], suggested an early killing phase followed by delayed adaptation of surviving organisms to survive and replicate within Mφ phagosomes. At every time point, extracellular Mtb totaled less than 5% of the intracellular burden (data not shown) suggesting that re-infection is minimal.
The intramacrophage growth of Mtb was also monitored by parallel transmission electron microscopy (TEM) analysis of fixed infected Mφ samples (Figure 1B). Enumeration of both morphologically normal and damaged Mtb in 100 randomly chosen macrophages revealed a rapid increase in the number of detectable bacteria per cell from 1.5 Mtb/cell at 2 hr post-infection (p.i.) to 10.3 Mtb/cell by day 4 (∼1.7-fold increase/day). From day 4 to day 14, the average bacterial burden increased at a notably slower rate (∼0.3-fold increase/day) (Figure 1B, Figure S1A). Interestingly, this early period of rapid Mtb replication following Mφ invasion coincided with the marked decrease in viable CFU (Figure 1A). This indicated that early in the infection bacterial replication is rapid, but is countered by effective bacterial killing by the Mφ.
To further validate this conclusion, we calculated Mtb growth and death rates during long-term infection of resting Mφ based on the loss of an unstable replication clock plasmid [15]. The proportion of intracellular Mtb retaining the clock plasmid pBP10 was determined by plating bacteria on media with or without 25 µg/ml kanamycin. The profile of pBP10 plasmid loss (Figure 1C, red) indicated an initial period of rapid replication (day 0–day 2) followed by an extended phase of much slower cell division (day 2–day 9). The cumulative bacterial burden (CBB) – the total number of Mtb live, dead, or degraded within Mφ during the infection – was calculated based on the mathematical model developed by Gill et al. [15]. The predicted CBB (Figure 1C, black) at day 2 p.i., closely mirrored the quantitative TEM results enumerating all detectable bacteria regardless of viability, at ∼12-fold higher than the number of viable CFU determined by plating. As shown in Figure 1D, this indicates substantial replication coinciding with even greater death rates during the early phase of infection. However, following the adoption of a slower growth rate from day 2 of infection, Mtb exhibited a steady increase in viable CFU.
These data indicate that, following invasion of Mφ, Mtb encounters a bottleneck during which the rate of bacterial killing outpaces its relatively rapid rate of replication. A similar early phase of pronounced bacterial killing was noted in vivo by Gill et al. during the first 14 days in an Mtb-mouse model of infection [15]. Following day 2, a period of apparent adaptation ensues extending to approximately day 6, during which both the rate of bacterial replication and the rate of killing decrease. The subsequent overall increase in viable CFU reflects further enhanced survival and the successful establishment of a productive infection. The shift in the balance between replication and death over time (Figure 1D) suggests that the successful adaptation of some Mtb cells to avoid killing by Mφ-derived effectors is at least as important as mechanisms for sustained replication within phagosomes. It is tempting to speculate that the slower growth rate of Mtb at later time points may contribute to the shift in growth∶death balance, perhaps by rendering Mtb more resistant to Mφ-derived pressures. Studies modeling the dynamics of Mtb-host interactions within human granulomas by Segovia-Juarez et al. support this idea, showing that slow intracellular growth rates are correlative with Mtb survival [16]. This novel insight into the life∶death equilibrium of Mtb during a sustained model of infection provides a physiological context for the interpretation of the global gene expression profiles discussed hereafter.
Descriptions of host-pathogen interactions based on population-level data, such as microarrays, are often interpreted without appreciation of the heterogeneity within that population. With this in mind, we performed detailed TEM image analysis of Mtb-infected Mφ at 2 day intervals over 14 days simultaneously with survival assays (Figure 1) and microarrays (see below).
Consistent with previous observations [17]–[20], at 2 hours p.i. most single bacteria resided in a vacuole surrounded by a phagosomal membrane tightly apposed to the Mtb cell wall (Figure 2A). While intracellular Mtb proved to be quite effective at resisting fusion with lysosomes, identified following endocytic uptake of colloidal gold (Figure 3A), ∼25% of single Mtb did traffic to phagolysosomes (P-L) and colocalize with colloidal gold as early as 2 hr p.i. (Figure 3B). At 2 days p.i., image analysis of Mtb-infected Mφ revealed an increased frequency of morphologically-damaged Mtb in large granular lysosomes (Figure 2C–D), consistent with the early phase of killing of replicating Mtb indicated by the clock plasmid experiments. Whereas 99% of bacteria in the inoculum cultures were scored as intact, by day 2 only 75% of phagocytosed Mtb appeared morphologically normal. We also noted heavily damaged Mtb that appeared as hollowed out “ghosts”, but these were not scored as damaged. The discrepancy between the magnitude of killing, calculated based on clock plasmid data (Figure 1D), and the small proportion of visibly damaged bacteria as well as the constant ratio of normal to damaged bacilli (Figure S1A) suggests a relatively efficient clearance of nonviable organisms. At later time points, the bacterial population continued to increase (Figure 2E–F, Figure S1B) with up to ∼150 Mtb/Mφ. In contrast to a recent report that a large subset of Mtb H37Rv translocated into the cytosol of Mφ within 48 hrs [6], phagosomal membranes were definitively detected surrounding ∼90% (88% at day 4 p.i., 86.5% at day 8 p.i.) of intracellular Mtb CDC1551 throughout the duration of the infection (25 Mφ and >100 Mtb examined per time point).
There was a surprising degree of heterogeneity of compartments in which visibly intact Mtb appeared to reside during long-term infections of Mφ, including “replicating” Mtb in typical tight niches, small fused P-L, and putative double membrane-bound autophagosomes that colocalized with gold (Figure 3, Figure S2). The diversity of intra-host environments undoubtedly contributes to heterogeneity within the bacterial population.
The transcriptional profile of Mtb during the course of the 14 day infection should mirror the physiological states and transitions defined in the previous section. Therefore, overlaying the temporal transcriptional changes with these physiological states will generate a correlative linkage between the two datasets. We conducted microarray analysis of RNA isolated from intracellular Mtb at 2 hr p.i. and at 2 day intervals over a 14 day period (GEO accession GSE35362). Fluorescent amplified RNA (aRNA) targets from each time-point were prepared from linearly amplified total Mtb RNA and hybridized against a 2 hr “no Mφ” control as previously described [12], [13]. This enabled us to determine dynamic expression ratios over time relative to a single common denominator mRNA sample, in this case an uninfected control time-matched with the earliest time-point (2 hr). Semi-quantitative RT-PCR of select target genes (aprAB, hspX, bfrB, icl, groEL2, katG, whiB7) was conducted to confirm temporal array profiles ([21] and data not shown). We identified 3626 genes with significant changes in expression during extended Mφ infection by combining both static statistical cutoffs (p<0.05 in at least one time-point) and EDGE (Extraction and analysis of Differential Gene Expression) methodology (Figure 4A) [22]. EDGE analysis identifies temporal changes in transcript levels that would not be deemed statistically significant at any single time-point by standard p-value measurements.
Our data indicate that Mtb encounters a bottleneck during the first two days p.i. during which a subpopulation is killed before surviving bacilli enter a sustained period of enhanced survival, albeit at a reduced replication rate. Temporal dynamics of the Mtb transcriptome during intracellular adaptation highlight genes responding, and thus likely active, during specific stages of infection. Given the short half-life of bacterial mRNA [23], [24] and steady clearance of nonviable Mtb, the transcriptional profiles observed are likely derived only from the surviving bacilli.
In addition to the temporal alterations in the transcriptional profile, we also examined transcription profiles of genesets linked to some of the more well-characterized environmental responses to probe how they matched the physiological transitions we have proposed occur during the 14-day infection. We have focused on the DosR regulon, regulation of carbon flux, and the response to pH, as some of the better characterized physiological themes.
In addition to analyzing the transcriptional profiles according to temporal dynamics and known functional themes we also conducted a systems-level analysis to characterize the behavior of the Mtb Transcriptional Regulatory Network (TRN) underlying pathogen survival during the 14-day infection. This network-based approach incorporates extensive a priori information on Mtb gene regulation and network topology, combined with expression data, to assess network responses in surviving bacteria elicited by the Mφ.
We started by expanding a large-scale Mtb TRN containing gene regulatory interactions extracted from both experimental and computational datasets [37]. The previous TRN comprised 738 genes (18% of the genome) and 937 regulatory links obtained from three sources: (i) literature mining; (ii) MtbRegList database [63]; (iii) inference from orthology with Escherichia coli [64]. To obtain the expanded network used in this study, we collected gene regulatory data from the following additional sources. (iv) We added orthology-based interactions inferred from the closely related Corynebacterium glutamicum available in the MycoRegNet database, which considerably expanded the TRN by adding 425 new interactions (Figure S9A). The extensive overlap with regulatory links supported by experimental data validated these interactions identified from orthologous Transcription Factor (TF)-Target Gene (TG) pairs in the two organisms (p = 10−5, Fisher's exact test, Figure S9A). (v) Further, we expanded the TRN by adding 114 protein-DNA interactions discovered by a new bacterial one-hybrid reporter system termed TB1Hybrid [65], being that 31 interactions were exclusively identified by this method. (vi) Finally, we performed operon-based network expansion, propagating a TF's regulatory effect to all members of the operon containing a given TG. The final expanded Mtb TRN contained 1133 genes (28% of the genome) and 1801 regulatory links, more than a half of which were experimentally determined (Figure S9B, C); the complete list of interactions is available in the Table S12.
The global TRN provides a static summary of all possible regulatory interactions that mycobacteria may use when facing a broad spectrum of environments, ranging from normal to stressful conditions inside the Mφ. However, previous work has suggested that only parts of the network are utilized in specific conditions [36]. Such parts (subnetworks) function as network modules regulated by a hierarchy of transcription factors in an environment-dependent fashion. To understand how the surviving subset of Mtb bacilli specifically utilizes the TRN modules during prolonged intracellular infection, we analyzed the temporal response of the TRN by overlaying the 14-day Mφ infection time-course array data on the extended Mtb regulatory network.
We improved the earlier method called NetReSFun (Network Response to Step Functions), which identifies responsive TF-regulated subnetworks from time course microarray data [36]. NetReSFun computes the Cov-score (Methods) to quantify the expression change within the module (subnetwork) that we define as the total genes directly regulated by a given TF. A significant Cov-score indicates subnetwork response, when the expression levels of the subnetwork's gene members are either down- or upregulated during consecutive time points (t, t+1). Alternatively, simultaneous change of a TF's direct target genes may also be a surrogate of the TF's activity. Under this assumption, NetReSFun may recognize TFs that are “turned on” through posttranslational modifications such as phosphorylation and metabolite binding, but may or may not show increased expression levels themselves.
The temporal map of network responses (Figure 8A) depicts specific TF-regulated subnetworks responsive during the time course, at a significance level of 0.05. The color scale indicates whether the overall trend of expression change within the subnetwork was positive (red) or negative (blue) at a given time interval. Intermediate colors denote subnetworks involving both up- and downregulated genes. The accompanying heatmap in Figure 8B indicates the source of regulatory links within the subnetwork (darker colors corresponding to higher fractions of links based on experimental evidence).
Strikingly, the map reveals that the dynamic utilization of the TRN occurs in a defined pattern that can be mapped back to distinctive phases of intracellular growth. In the first 2 days p.i. - which corresponds to the stress phase in the growth curve of Figure 1A – we observed a high number of responsive TF-subnetworks (20 out of 83), mostly exhibiting increased expression of involved genes. Among these were DosR, HspR, KstR, members of the WhiB family (WhiB3, WhiB4), two-component response regulators (Rv0260c, Rv0818, RegX3), and alternative sigma factors (SigE, SigK, SigM). The sharp induction of a large number of subnetworks is reminiscent of the general Environmental Stress Response (ESR) in yeast [66] and in Bacillus subtilis [67].
In contrast, after ∼6 days inside Mφ, we observed a reciprocal scenario where the TRN reflects a significant downmodulation of target genes, many of which had been induced immediately after invasion. This pattern is especially evident for a number of stress-responsive subnetworks that shift into downregulation during the slow growth phase, including RegX3, HspR and DosR. This “repressive” transcription phase indicates that the surviving bacteria have either adapted to stress, or they reside in a less hostile niche. For example, the SigH-controlled subnetwork displays a strong negative response only late in the time course (∼8 days and onwards). As a global regulator, SigH modulates the transcription of SigE and SigB, as well as its own promoter. Although SigH is not required for growth in Mφ, mutants lacking sigH caused reduced immunopathology and lethality in mice [68], [69]. Alternatively, it is possible that these regulatory changes are associated with the surviving bacilli reprogramming their physiology to assume the slow growth phenotype observed from day 4 onwards.
Importantly, we observed responsive subnetworks throughout the entire time-course, which indicates their importance for establishing productive infection. The presence of sustained responders such as HspR and DosR can have two possible implications. First, the opposite trends in the early phase of the infection (primarily upregulation) and later phases of infection (primarily downregulation) may indicate that the stress to which these modules respond initially is ameliorated at later time points. Alternatively, sustained responders may be necessary both to counteract initial phagosomal stress during the early phase of infection as well as for driving the persistor phenotype encountered in later phases of infection. This is the case of DosR, which is crucial for maintaining redox balance and energy levels during transitions into and out of dormancy-like conditions that perturb aerobic respiration, electron transport, or menaquinone pools [70]. HspR, which activates a subset of the heat-shock general stress response upon Mφ invasion [71], is also necessary in the persistent phase since ΔhspR strains exhibited attenuated growth in the chronic infection [72].
Finally, TRN analyses revealed novel sustained responders that might be critical for Mtb adaptation within the intracellular compartment. For example, the Rv2034-controlled subnetwork (inferred from C. glutamicum orthology) contains multiple fadE homologs implicated in β-oxidation of fatty acids and redox homeostasis. Notably, Rv2034 was recently characterized as an activator of the phoP virulence regulator in mycobacteria [73], which makes this regulator an interesting candidate for follow-up studies.
By overlaying the genome-scale temporal expression data onto the TRN, we revealed the activation of additional regulons during macrophage survival not readily apparent in our supervised analyses. Thus, by leveraging the behavior of multiple TG as a readout of TF activity, TRN analysis of time-course microarray data further enhances the ability to detect adaptive changes in Mtb gene expression during productive infection of macrophages.
The process of infection is extremely dynamic as both host and pathogen seek to respond to the stimuli that they sense at their interface. In the current study we applied multiple analytical tools to establish a link between the transcriptional responses and physiological states through which Mtb transitions on its way to the establishment of a productive infection in its host macrophage. This analysis revealed several unexpected findings. Firstly, the initial phase of infection is marked by rapid bacterial replication coupled with effective bacterial killing by the macrophage. This is a period of marked stress for Mtb, which is illustrated by the greatest transcriptional response with respect to both up-regulated and down-regulated genes. Subsequently, the rate of replication slows and the bacterial number appears constant or at equilibrium, during which period the expression of many genes returns closer to control levels, whilst the divergent level of expression of others is sustained. Finally, the bacterial numbers start to increase indicating that the rate of replication exceeds that of death. At this time there is a marked down-regulation of many of the genes linked to general stress supporting the contention that Mtb has entered into a productive phase of infection characterized by enhanced intracellular survival. This is an important functional framework on which to hang the transcriptional profiles to determine which responses and/or metabolic themes impact which phase of infection.
Our examination of dynamic alternations of gene expression across the Mtb TRN also highlights stress responses and survival mechanisms deployed during distinct phases of host interaction. Among other things, our results indicate adaptive changes in lipid and energy metabolism akin to those observed in various dormancy models. Based on this, it is tempting to speculate that early infection of resting macrophages may serve to prepare Mtb for conditions encountered within granulomas after the onset of the adaptive immune response. Detailed understanding of the sensory and regulatory pathways required for Mtb virulence remains rudimentary at best. This point is illustrated by the large number of “genes of unknown function” which are actively regulated during intracellular survival.
In the discussion, we have presented the observed transcriptomic changes as the result of all Mtb cells sensing and responding in concert to phagosomal cues in the intracellular environment. Clearly, the microarray profiles in this study capture an average behavior over time of a population of intracellular bacilli that, as we have shown, exist in distinct vacuolar niches and presumably metabolic states. It could be argued that the changes in gene expression represent the minority of Mtb that fail to block P-L fusion mounting dramatic stress responses. However, our data appear to be inconsistent with this view. For example, we know that Mtb in P-L encounter Fe-deplete conditions whereas the average behavior of Fe-responsive signature genes reflects an Fe-replete environment (data not shown). Alternatively, the apparent up- and down-regulation may be due to the selective enrichment of pre-existing phenotypic variant cells with randomly upregulated stress response subnetworks, similar to the cell-to-cell heterogeneity reported by Aldridge et al. [74]. The rapid killing observed immediately after macrophage entry supports this scenario, suggesting that a subpopulation of bacilli capable of surviving the bottleneck may be present prior to infection, due to random phenotypic variation within the microbial population. However, the ability to block the induction of an acid regulon following Mφ invasion by chemical manipulation of phagosomal pH [12] would suggest that Mtb are altering gene expression in response to host-derived cues.
Thus, our data suggest that the dominant transcriptional profiles highlighted here represent the adaptive responses of the majority population. Analysis at the single-cell level will be required to explore the strategies employed by Mtb to survive across diverse host environments experienced by each individual bacterium. To address this, we have begun to exploit fluorescent reporter strains responsive to specific environmental cues to gain a high-resolution view of Mtb intracellular adaptation and cell biology. Interestingly, the expression of an acid-inducible locus required for normal intracellular survival, aprABC, is expressed at distinct levels by individual bacilli within the same host cell [21]. The impact of this type of heterogeneity on Mtb-host interactions and pathogenesis has yet to be determined. Previous work has shown that stochastic variation (intrinsic to individual cells) can aid survival in stressful environmental conditions [75]. However, a similar role of extrinsic, environmental variation is yet to be established.
The identification of mutants attenuated for intracellular survival is a popular and powerful tool for defining in vivo survival mechanisms, however, there are some limitations to this methodology that can be addressed by transcriptional profiling. Array-based mutant screens such as Transposon Site Hybridization (TraSH) [76], [77] are in essence end-point assays that are not readily amenable to quantitative kinetic analyses. In addition, there are several well-characterized examples of genes known to be required for Mφ survival, such as pckA [42], [43], icl ([44], [78], prpCD [56], katG [79], [80], or phoPR [9], [81], that have not been identified by TraSH screens indicating that the method is not comprehensive [82]. The TraSH methodology may preferentially identify mutants with severe survival defects, while being less effective at isolating mutants with less extreme phenotypes, such as ΔphoPR, that resist killing but fail to grow within phagosomes [9]. Finally, mutant screens are limited by their inability to query the in vivo role of genes that are essential in vitro or identify genes whose phenotype upon inactivation is masked by compensatory changes in gene expression. Future studies would benefit from the coordinated application of these two distinct but complementary approaches to identify genes contributing to the pathogenesis of Mtb.
We feel that the significance of this current study is that it transforms transcriptional profiling from a purely descriptive analysis to the generation of a predictive discovery tool that can be used to identify genes, and therefore metabolic pathways and physiological states, that are required to support distinct phases in the intracellular life cycle of M. tuberculosis.
As previously described [83], the M. tuberculosis clinical isolate strain CDC1551 [84], [85] was cultured in Middlebrook 7H9-OADC medium in ventilated T-75 flasks without shaking. Bone marrow-derived macrophages (Mφ) were isolated from C57BL/6 mice and grown in DMEM supplemented with 20% L-cell conditioned medium, 10% fetal calf serum (FCS), and antibiotics (penicillin and streptomycin). Media lacking antibiotics was added to Mφ at least 24 hr prior to infection with M. tuberculosis.
To monitor the survival/growth of Mtb in Mφ, confluent Mφ monolayers in 24-well dishes were infected (MOI ∼1∶1) with Mtb CDC1551. After 2 hr (t = 0) and at two day intervals up to 14 days, intracellular Mtb released from monolayers by lysis with ddH2O+0.05% Tween-80 were serially diluted, and plated on 7H10+cycloheximide agar. CFU were enumerated after ∼3 weeks incubation at 37°C. The overall integrity of Mφ monolayer was verified by microscopy.
In order to measure the dynamics of Mtb replication and death during Mφ infection, a replication clock plasmid, pBP10, was used (generous gift of Dr. David R. Sherman). As previously described [15], in the absence of kanamycin selection pBP10 is lost at a rate proportional to the growth rate. Macrophage survival assays using Mtb CDC1551 harboring pBP10 (introduced by electroporation) were conducted as described above except that aliquots were plated on 7H10+cycloheximide agar both with and without 25 µg/ml kanamycin. Based on the loss of kanamycin resistance over time, rates of replication and death were quantified using the mathematical model of Gill et al. [15].
In parallel with Mφ-Mtb infections for survival assays and microarray analysis, confluent monolayers of Mφ were infected with Mtb CDC1551 at a low MOI (1∶1) as described above. At select times p.i. (2 hr and day 2-day 14, alternating days), samples were fixed in buffered glutaraldehyde solution (2.5% glutaraldehyde in 0.1 M sodium cacodylate, 5 mM CaCl2, 5 mM MgCl2, 0.1 M sucrose, pH 7.2), rinsed with 0.1M sodium cacodylate buffer (see above), post-fixed with 1% osmium tetroxide (4% stock osmium diluted in 0.1 M sodium cacodylate, 5 mM CaCl2, 5 mM MgCl2, pH 7.2), rinsed with buffer, soaked in 1% aqueous uranyl acetate, and then rinsed with water. Samples were then dehydrated in a graded ethanol and propylene oxide series, followed by gradual infiltration of Spurr's resin. Blocks were then polymerized and ultrathin sections (∼70 nm) were cut and contrasted with both lead citrate and uranyl acetate. In independent experiments to determine phagosome-lysosome fusion by the colocalization of Mtb with colloidal gold, at various time-points after infection (2 hr, day 2, 6, and 10) infected monolayers were pulsed with 15 nm colloidal gold (Aurion) for 2 hr, washed, and then chased for 45 min in infection media before fixation as above. In assessment of the morphology of intracellular Mtb, bacilli were considered intact if they maintained a rod shape (longitudinal sections) or circular shape (cross sections), if ultrastructural organization and electron opacity of the cytoplasm was preserved, and if no breaks in the cytoplasmic membrane or cell wall were detected. Otherwise, bacilli were counted as damaged.
The methodology for transcriptional profiling of intramacrophage Mtb including RNA isolation, linear amplification, and hybridization has been described previously [12], [83]. Briefly, C57BL/6 bone marrow-derived Mφ were infected (MOI ∼10∶1) with M. tuberculosis CDC1551 from 2 hr up to 14 days (2 day intervals). In controls, aliquots of the same bacterial samples were incubated in flasks without a Mφ monolayer for 2 hr. Addition of guanidine thiocyanate-based lysis buffer selectively lysed Mφ and stabilized bacterial RNA while leaving mycobacteria intact. Pelleted bacilli were lysed in 65°C Trizol using a BeadBeater and 0.1 mm silicon beads. Total RNA was isolated from Trizol lysates by chloroform extraction and Qiagen RNeasy column purification. To generate array targets, 250 ng of total RNA was amplified using the MessageAmp-II Bacteria RNA Amplification system (Ambion). Amino-allyl UTP was incorporated into aRNA during transcription to allow labeling with Alexa dyes.
Amino-allyl modified aRNA were labeled with Alexa Fluor 555 and Alexa Fluor 647 (Invitrogen) and purified using a MegaClear kit (Ambion). 10 µg of Alexa-labeled aRNA from paired samples was dried and resuspended in 75 µl of hybridization buffer (5× SSC, 25% formamide, 0.1% SDS, and 25 µg salmon sperm DNA). Slides were prehybridized for 1 hr in 25% formamide, 5× SSC, 0.1% SDS, 1% BSA and washed with H2O and isopropanol. Labeled targets were denatured at 95°C for 5 min, cooled to 60°C, and hybridized to microarrays at 45°C for 16–18 hr. Following hybridization, arrays were washed and processed for scanning as described previously [83]. The microarray platform used can be accessed via NCBI's Gene Expression Omnibus (GEO) database [86] under platform accession number GPL5754. This dataset has been deposited in GEO under series accession number GSE35362.
Microarrays were scanned with a GenePix 4000B instrument (Axon Instruments, Inc.) with preliminary image analysis, spot intensity determination, background measurements, spot quality assessment and flagging conducted using Imagene software (version 6.0, Biodiscovery). Poor quality spots with signal intensities less than three standard deviations above background were excluded from further analysis. Genes that were not flagged as Present in at least 14 of 16 slides were omitted from further analyses. Subsequent normalization, statistical analysis, and visualization of array data were performed with Genespring 7.3 (Agilent). We utilized the EDGE (Extraction and analysis of Differential Gene Expression) methodology of Storey et al. to identify time-dependent transcriptional changes [22], allowing detection of genes whose expression exhibited significant temporal trends but failed to meet static statistical cutoffs at any single time point. Genes with significant changes in expression levels relative to controls were identified based on both static (p<0.05 for at least one time point) and EDGE analysis (cubic spline of 4, q<0.03).
RNA amplification and microarray methodology used in this study have been previously validated by qRT-PCR [12]. Additional qRT-PCR validation of temporal expression patterns of select genes during long-term Mφ infection was conducted by two-step real-time RT-PCR using iScript and iTaq SYBR Green reagents (Biorad). Each sample was analyzed in triplicate on an ABI 7500 starting with 100 ng of total RNA (amplified and unamplified). CT values were normalized to values obtained for sigA, a constitutively expressed Mtb gene, and relative changes in gene expression were calculated using the 2−ΔΔCT method [87].
We modified the tool NetReSFun (89) in order to identify TF-regulated subnetworks sequentially responsive during TB lifespan in the macrophage environment. Consider a set of t step functions each of which “jumps” from 0 to 1 at time point τ:(1)NetReSFun compares gene expression profiles to these pre-defined step functions using scaled covariance to detect whether a gene's expression changed in the interval [τ-1, τ]. The method starts by computing the scaled covariance between the expression (log10-ratio) profile of gene i and a step function that jumps at time point τ:(2)where brackets denote averaging over genes, horizontal bar is averaging over time, and σ is the standard deviation. Thus, is the response of gene i during interval [τ−1, τ]. The combined response of subnetwork I, or Cov-score, is the mean of the absolute covariances of subnetwork genes(3)To assess statistical significance, we compare the subnetwork response with a reference cumulative density function (c.d.f) , constructed with scores drawn of 1,000 random subnetworks (same size of subnetwork I, but assembled using nodes randomly chosen from the network). The density estimation of the c.d.f is done in a standard fashion by assigning probability mass 1/1000 for each observed random score; linear interpolation is applied to transform the discrete into a continuous function in the interval [0,1]. Finally, a p-value of observing by chance can be readily estimated as below, and a p-value≤0.05 was considered significant.(4)To determine the direction of transcriptional change displayed in the temporal map (Figure 8), we first compute the deviation index of the subnetwork as the positive/negative ratio of individual covariances(5)The z-score of the deviation index, which relates the change of a particular subnetwork I to all other subnetworks, is given by(6)where A denotes all TF-regulated subnetworks, brackets denote averaging over all subnetwork deviations, and σ is the standard deviation. Thus, the deviation z-score is the quantity plotted in the temporal map of Figure 8.
|
10.1371/journal.pgen.1005671 | RNA Polymerase III Output Is Functionally Linked to tRNA Dimethyl-G26 Modification | Control of the differential abundance or activity of tRNAs can be important determinants of gene regulation. RNA polymerase (RNAP) III synthesizes all tRNAs in eukaryotes and it derepression is associated with cancer. Maf1 is a conserved general repressor of RNAP III under the control of the target of rapamycin (TOR) that acts to integrate transcriptional output and protein synthetic demand toward metabolic economy. Studies in budding yeast have indicated that the global tRNA gene activation that occurs with derepression of RNAP III via maf1-deletion is accompanied by a paradoxical loss of tRNA-mediated nonsense suppressor activity, manifested as an antisuppression phenotype, by an unknown mechanism. We show that maf1-antisuppression also occurs in the fission yeast S. pombe amidst general activation of RNAP III. We used tRNA-HydroSeq to document that little changes occurred in the relative levels of different tRNAs in maf1Δ cells. By contrast, the efficiency of N2,N2-dimethyl G26 (m22G26) modification on certain tRNAs was decreased in response to maf1-deletion and associated with antisuppression, and was validated by other methods. Over-expression of Trm1, which produces m22G26, reversed maf1-antisuppression. A model that emerges is that competition by increased tRNA levels in maf1Δ cells leads to m22G26 hypomodification due to limiting Trm1, reducing the activity of suppressor-tRNASerUCA and accounting for antisuppression. Consistent with this, we show that RNAP III mutations associated with hypomyelinating leukodystrophy decrease tRNA transcription, increase m22G26 efficiency and reverse antisuppression. Extending this more broadly, we show that a decrease in tRNA synthesis by treatment with rapamycin leads to increased m22G26 modification and that this response is conserved among highly divergent yeasts and human cells.
| Transfer RNAs (tRNAs) are molecular adapters necessary for translation of the genetic code from DNA to messenger RNA (mRNA) to synthesis of the proteins that constitute the energy producing enzymes and structural components of all cells and organisms on earth. In eukaryotic cells, tRNAs are synthesized by RNA polymerase III (RNAP III). Proteins are composed of different amino acids, sequentially arranged according to the triplet sequence code, each carried to the ribosome by a different tRNA which reads or decodes the triplet sequence of the mRNA. After their synthesis by RNAP III, tRNAs are chemically modified by enzymes on multiple of their nucleosides. Here we report that one of these modifications, dimethyl-guanine-26 (m22G26) varies in the efficiency by which it is added to its target tRNAs, in a manner that is dependent on the overall activity rate of RNAP III. We show that this is important because the m22G26 modification activates the tRNA for function to translate the code. This link between RNAP III activity rate and m22G26 modification efficiency was previously unknown, and we show that it is conserved from yeast to human cells.
| Apart from their role in translation, tRNAs can regulate gene expression [1,2] and serve as metabolic sensors [3], and their over-expression is associated with cell proliferation and transformation [4,5]. RNAP III is activated by oncogenes [6,7] whereas its repression reduces transformation and tumorigenesis [8].
Accumulating evidence indicate the importance of matching tRNA activity with mRNA codon demand [9]. Different cells and tissues show differences in tRNA abundances that vary with codon use [1,10]. tRNA specific activity for codon-specific decoding can be controlled by posttranscriptional modifications, most notably in the anticodon loop for nucleosides at wobble position 34 and position 37 [2,11–14].
Despite RNAP III ubiquity in eukaryotic cells, mutation in one of its catalytic subunits can manifest as a tissue-specific developmental defect in zebra fish [15,16]. In humans, mutations in either of the two catalytic subunits lead to a nervous system disorder, hypomyelinating leukodystrophy (HLD) and other tissue-specific defects ([17] and refs therein), although how these mutations affect RNAP III transcription and cause disease is unknown.
The highly conserved RNAP III repressor, Maf1, acts in response to stress including lack of nutrient, and in S. cerevisiae, mammals and other species, is under the control of the target of rapamycin (TOR) kinase [18,19], which integrates information from several environmental cues and stress states, and functions to sustain growth and homeostasis in various conditions [20]. When Maf1 is nonfunctional, cells produce much increased and unregulated transcription by RNAP III, the energy cost of which is wasted, highlighting a function for Maf1 as a key contributor to metabolic economy [21].
A striking phenotype of S. cerevisiae maf1-mutants is antisuppression [19,22] which reflects loss of suppressor-tRNA (sup-tRNA)TyrUUA mediated suppression of a nonsense codon in a mRNA encoding an adenine metabolic enzyme. Although described nearly 20 years ago and to date only for S. cerevisiae, maf1-antisuppression is paradoxical because it occurs amidst global increases in tRNA synthesis [19,22,23].
We deleted maf1+ from S. pombe and also observed antisuppression, in this case by sup-tRNASerUCA, amidst general increases in tRNA levels. We employed a tRNA-enriched limited hydrolysis sequencing method, termed tRNA-HydroSeq, on S. pombe maf1+, maf1Δ and other strains. While the levels of different tRNAs relative to each other varied little upon maf1+ deletion or over-expression, consistent with global regulation, a sup-tRNASerUCA modification, N2,N2-dimethylguanosine-26 (m22G26), was specifically decreased in maf1Δ and shown to be required for efficient suppression. Trm1 is a nuclear enzyme that produces m22G26 which likely contributes to proper tRNA folding [24,25] (see Discussion). Trm1 activity is limiting in the context of increased tRNA production in maf1Δ cells and we show that its over-expression reverses antisuppression. Treatment with rapamycin or over-expression of maf1+ reduces tRNA transcription with increase in the m22G26 content of sup-tRNASerUCA and its specific activity for suppression. We also introduced mutations in a catalytic subunit of RNAP III associated with hypomyelinating leukodystrophy (HLD) to show that a general decrease in tRNA transcription by another mechanism also increases m22G26 modification efficiency and reverses antisuppression in S. pombe. The results establish a link between RNAP III activity, tRNA production and Trm1 modification activity that impacts tRNA function. We show that this response is conserved, as deletion of MAF1 from S. cerevisiae is also accompanied by m22G26 hypomodification, and maf1-antisuppression is reversed by over expression of TRM1. Finally, we show that human cellular tRNA m22G26 modification efficiency increases with serum starvation or rapamycin treatment and decreases following serum stimulation.
Unlike for all other species examined, S. pombe is naturally resistant to the growth inhibitory effect of rapamycin [26]. Thus, it was important to determine if maf1+ regulates tRNA production in S. pombe and if it does so under TOR control. We created a maf1-deletion strain and showed that it lacked maf1+ mRNA relative to wild type (WT; Fig 1A, WT/vector vs. maf1Δ/vector). Ectopic over-expression of plasmid-borne maf1+ in maf1Δ increased maf1+ mRNA about 4-fold relative to endogenous maf1+ in WT cells (Fig 1A). Levels of tRNAAlaUGC and tRNASerGCU were increased in maf1Δ relative to WT but decreased relative to WT when maf1+ was over expressed (Fig 1B). Quantitation of these tRNAs relative to U5 snRNA, a RNAP II transcript on the same blot, from triplicate cultures on triplicate blots revealed that three levels of maf1+ expression led to three levels of tRNA expression (Fig 1C).
We analyzed the effect of the TOR inhibitor, rapamycin on maf1Δ cell growth. Deletion of tit1+, encoding the tRNA isopentenyltransferase (MOD5 homolog, see below), which forms i6A37 on some tRNAs, is known to cause sensitivity to rapamycin [14] and served as a control (Fig 1D). In media lacking rapamycin, these strains exhibited relatively similar growth (Fig 1D, upper panel). While the control tit1Δ was sensitive to rapamycin, maf1Δ was insensitive (Fig 1D, lower panel). By contrast to maf1Δ, the cells over-expressing maf1+ were sensitive to rapamycin (Fig 1D, lower).
To test whether S. pombe maf1+ regulates RNAP III in response to rapamycin, we analyzed tRNA from triplicate cultures of cells in logarithmic growth to which rapamycin was added and incubated for an additional hour. The northern blot in Fig 1E compares maf1Δ and maf1+ cells using probes to three tRNAs and control U5 RNA. In maf1+ cells, rapamycin decreased precursor and mature tRNA species relative to DMSO treatment while U5 was unchanged (Fig 1E, lanes 1–3 vs. 4–6). maf1Δ cells showed higher levels of tRNAs relative to maf1+ (Fig 1E, lanes 1–3 vs. 7–9). Rapamycin did not reduce tRNA levels in maf1Δ cells (Fig 1E, lanes 7–9 vs. 10–12). This established that the ability of S. pombe to repress tRNA production in response to rapamycin is dependent on maf1+, as in other species. In addition, we conclude that maf1+ over-expression causes slow growth and rapamycin sensitivity, likely due at least in part, to decreased tRNA production.
tRNA-mediated suppression (TMS) of adenine synthetic genes prevents accumulation of a red pigmented metabolic intermediate and is useful for studying biogenesis and activity of sup-tRNAs (reviewed in [27]). Because all yeast tRNA genes including sup-tRNAs share similar promoters, and maf1-mutants are expected to activate RNAP III globally, maf1-antisuppression has been an unexplained paradox [22,23]. S. cerevisiae Maf1 was noted to affect cellular localization of Mod5 which carries out i6A37 formation, leading to the possibility that sup-tRNA i6A37 hypomodification is responsible for maf1-antisuppression [19].
We examined TMS in S. pombe maf1Δ, tit1Δ and wild-type (WT; tit1+, maf1+) strains all of which contain the opal suppressor sup-tRNASerUCA and the opal suppressible allele, ade6-704. On limiting adenine, maf1Δ exhibited antisuppression relative to WT, similar to tit1Δ [28] (Fig 2A, Ade10). Significantly, over-expression of maf1+ in maf1Δ not only reversed antisuppression, it caused more suppression than in WT (Fig 2A, Ade10). Thus as RNAP III activity decreases from high to intermediate to low, sup-tRNASerUCA-mediated TMS activity goes in the opposite direction, from low to intermediate to high.
As noted, i6A37 hypomodification was suggested to cause maf1-antisuppression [29,30]. Availability of methods to monitor i6A37 on total tRNAs and on specific tRNAs allowed us to test whether maf1+ deletion led to decreased i6A37. Three tRNAsSer that decode serine UCN codons comprise the major i6A37-tRNA component in S. pombe while the shorter length tRNATyr and tRNATrp comprise a minor component [28]. Midwestern blotting using anti-i6A antibody [31] showed that the i6A37 content of tRNA differed only slightly among the strains (Fig 2B). No signal was observed with this antibody in tit1Δ as expected [28]. The blot was also hybridized with an oligo-DNA probe complementary to U5 RNA as a control (lower panel).
To examine i6A37 levels in sup-tRNASerUCA, we used a PHA6 assay (positive hybridization in the absence of i6A37 modification) in which the northern blot signal intensity increases as i6A37 content decreases because the isopentenyl modification interferes with annealing of a probe targeted to the ACL (anticodon loop), while a 'body' probe to the pseudo-U stem loop of the same tRNA serves as an internal control [14,28,32]. The EtBr stained gel in the upper panel of Fig 2C was blotted, probed, stripped and rehybridized with probes to the RNA species indicated to the right of the other panels. Specificities of the sup-tRNASerUCA ACL and body probes were revealed by paucity of signal in lanes 1 and 2 representing a strain that lacks the sup-tRNASerUCA allele. The ACL probe detected relatively high levels of sup-tRNASerUCA in tit1Δ as expected, reflecting lack of the i6A37 modification, but relatively low levels in maf1Δ and WT reflecting efficient i6A37 modification. A similar pattern of ACL vs. body probe signal was observed in WT, tit1Δ and maf1Δ, for endogenous tRNASerUGA which also carries i6A37 [28]. Quantification of sup-tRNASerUCA i6A37 modification efficiency for duplicate samples revealed that it was comparable in maf1Δ and maf1+ WT cells (Fig 2D). We also quantified steady state levels of the tRNAs as monitored by their body probes relative to the U5 control, which confirmed that both were elevated in maf1Δ relative to maf1+ cells (Fig 2E).
The cumulative data suggest that deletion of maf1+ leads to a decrease in the specific activity of sup-tRNASerUCA for TMS while over-expression of maf1+ increases its specific activity for TMS. However, the decrease in sup-tRNASerUCA activity in the maf1Δ strain is not due to i6A37 hypomodification.
Limitations to high throughput sequencing of tRNAs include inefficient adapter ligations due to secondary structure as well as the multiple modifications that cause reverse transcriptase to pause at each, processively diminishing generation of full length sequence reads. Recent advances have overcome some of the limitations by removing a subset of the blocking modifications by pre-treatment with specific demethylases and/or by use of highly processive thermostable reverse transcriptase [33,34]. Another approach used limited alkaline hydrolysis of total RNA and subsequent mining of reads corresponding to tRNAs [35]. Thus, by generating 19–35 nt hydrolysis products followed by adapter ligation, each fragment will have less potential to form secondary structures and significantly fewer modifications for reverse transcriptase to get past (S1 Fig). We introduced a modification to the Karaca et al. method [35]; namely, purification of tRNA prior to hydrolysis. This was followed by adapter ligation to the fragments, reverse transcription, PCR amplification, and sequencing using Illumina HiSeq technology (S1 Fig).
S. pombe contains 171 tRNA genes that produce 61 unique tRNA sequences comprising 45 anticodon identities (S1 Table). Although multicopy tRNA genes encode identical mature tRNAs they typically differ in the precursor sequences of 5’ leader, 3’ trailer and/or intron if present [36]. By this account, 150 of the 171 tRNA genes are unique. Sequence reads were first mapped to a reference list representing the 61 unique mature tRNA sequences. The remaining reads were then aligned to a reference list representing the 150 unique precursor-tRNA gene sequences. The total read counts listed in S1 Table provides evidence for expression of all of the tRNA genes in S. pombe.
Quantitation using DEseq [37] revealed good correlations of the tRNA expression profiles from WT, maf1Δ and maf1Δ+maf1+ cells (Fig 3A). This would be expected of Maf1 as a global regulator of RNAP III (Discussion).
tRNA-HydroSeq data included high levels of nucleoside misincorporations at specific positions in specific tRNAs. An example for tRNASerUGA using the IGV display tool is shown in Fig 3B; grey bars indicate match to genomic sequence and colored bars reveal positions at which mismatch was ≥15% as set by IGV. These positions were previously noted to undergo base modifications associated with misincorporation by reverse transcriptase [38]. Modifications of this type disrupt potential for hydrogen bonding and normal base pairing (S2A Fig). We saw no misincorporation at i6A37, as this modification would preserve the potential of adenine for hydrogen bonding (S2B Fig), consistent with prior observations (see [39]). Misincorporations observed at G9, G26, C32 and A58 occurred in tRNAs known to carry m1G9, m22G26, m3C32 and m1A58 in yeast and/or other species as well as at position 34 for the eleven tRNAs with encoded A34, consistent with their deamination to inosine (I) [40,41] (S3 Fig).
G26 is modified by N2,N2-dimethylation (m22G26) in many tRNAs by the Trm1 methyltransferase [42]. Of the 36 tRNAs in S. pombe that have G26, 27 showed significant misincorporation (≥10%) at G26 (Fig 3C). The extent of G26 misincorporation varied with tRNA identity from 10–80% (Fig 3C). Biochemical studies of S. cerevisiae Trm1 identified determinants of modification including length of the variable loop [43]. Consistent with those data, we found that G26 tRNAs with variable loops of <5nt (e.g., GlyGCC, GlnTTG, GlnCTG, Fig 3C) showed very low misincorporation. These and data described below suggest that the nine G26 tRNAs with misincorporations of ~1% are not Trm1 targets and reflect background (Fig 3C).
To confirm that G26 misincorporations are due to m22G26 modification, we deleted trm1+ from its genomic locus and examined misincorporation. The heat map in Fig 3D shows that G26 misincorporations were decreased to ≤1% in trm1Δ cells, demonstrating that they are due to m22G26. Notably, expression of trm1+ from a high copy plasmid in trm1Δ cells restored the misincorporations to higher levels than in WT cells (Fig 3D). This suggested that a significant amount of Trm1 substrates are not fully modified in WT cells because Trm1 activity is limiting.
Comparative analysis of all tRNAs at all positions in WT, maf1Δ and maf1Δ+maf1+ cells revealed that misincorporation levels at G26 were specifically altered in Trm1 target tRNAs in response to three levels of maf1+ expression in a manner that positively correlated with TMS (Fig 4A, G26 panel, compare with Fig 2A). Misincorporations detected at G9, A34 and A58, reflecting m1G9, I34 and m1A58, did not vary with maf1+ expression (Fig 4A). The specific correlation of G26 misincorporations with three levels of maf1+ expression and TMS also fits with the finding that Trm1, which appears limiting for G26 modification in WT cells (Fig 3D) may become more so as tRNA levels increase in maf1Δ cells, and as shown below, m22G26 is required for suppressor activity.
To more directly link m22G26 to Maf1 and its effects on suppression phenotype, we followed three tacks. First, we analyzed sequence reads that uniquely mapped to the sup-tRNASerUCA for G26 misincorporation in WT, maf1Δ and maf1Δ+maf1+ cells (Fig 4B). The Sup-tRNASerUCA G26 misincorporations in the three strains reflected the general pattern for the larger subset of Trm1 targets (compare Fig 4A, G26 panel and Fig 4B).
An antibody raised against a C-terminal peptide of S. pombe Trm1 was used to examine Trm1 levels. This revealed no significant difference in Trm1 levels in maf1Δ and WT cells (Fig 4C). The data support a model in which as Trm1 substrates increase with elevated RNAP III activity in maf1Δ cells, the efficiency of m22G26 modification decreases presumably because a limiting supply of Trm1 cannot keep up with the increase in substrate, and TMS decreases leading to the antisuppression phenotype.
A second tack was to monitor changes in G26 modification levels by an approach other than tRNA-HydroSeq. We devised a northern blot probing method to monitor G26 modification that we refer to as the PHA26 assay (positive hybridization in the absence of G26 modification). Since m22G26 modification debilitates normal base pairing [39], we expected it to inhibit annealing of a short probe, designated D-AC stem, to this region of a tRNA while a probe to the variable arm-T loop region of the same tRNA would serve as internal control (Fig 4D). As proof of the PHA26 assay method, the D-AC stem probe showed high signal in trm1Δ cells but was decreased upon over-expression of trm1+ (Fig 4E, tRNALeuCAA, D-AC stem; compare lanes 5 & 7). Although tRNALeuCAA is not as differentially modified in WT, maf1Δ, and maf1Δ+maf1+, as is the sup-tRNASerUCA, it does reflect some differential m22G26 modification (Fig 4E, D-AC stem vs. T-loop, lanes 1–4). This differential pattern was also seen for tRNALeuTAG (Fig 4E, lanes 1–4); quantification of T-loop/D-AC stem signal is expressed as a modification index, in this case relative to lane 1, below the lanes for each tRNA (Fig 4E). Significantly, over-expression of trm1+ in maf1Δ decreased the D-AC stem probe signal (Fig 4E, compare lanes 2 & 4) demonstrating that Trm1 activity is limiting in maf1Δ cells.
PHA26 confirmed that the m22G26 content of tRNALeuTAG is lower than for tRNALeuCAA (Fig 4E compare with Fig 3C). Moreover, PHA26 also suggested that tRNALeuTAG is not as ideal a substrate for trm1+ over-expression as compared to tRNALeuCAA (Fig 4E, compare both tRNAs, lanes 6 & 7, and see below).
A third and most key approach was to determine if Trm1 is limiting for suppression in maf1Δ by over-expressing it and assaying for TMS. Deletion of trm1+ from our WT strain caused antisuppression (Fig 4F). This demonstrated that m22G26 modification is required for the suppressor activity of sup-tRNASerUCA, confirming results in S. pombe with an ochre sup-tRNASerUUA [44]. Most relevantly, over-expression of trm1+ in maf1Δ led to substantial reversal of the antisuppression phenotype in the context of elevated tRNA levels in these cells (Fig 4F). From this we can conclude that m22G26 hypomodification due to limiting Trm1 in maf1Δ is a determinant of antisuppression in S. pombe. Thus, changes in RNAP III activity inversely impact m22G26 modification and the functional specific activity of sup-tRNASerUCA with consequent phenotype.
We wanted to ask if a decrease in RNAP III activity by a Maf1-independent mechanism would also affect functional G26 modification. RNAP III is a conserved enzyme whose two largest subunits, Rpc1 and Rpc2, form the catalytic center while other subunits serve supportive and regulatory functions (reviewed in [45]). Certain point mutations in Rpc1 and Rpc2 cause hypomyelinating leukodystrophy (HLD), a tissue-specific developmental disorder, although if they might affect global tRNA transcription has not been reported [46]. Since most of these mutations affect residues invariant from yeast to humans, we introduced them into S. pombe RNAP III and examined activity after over-expression in vivo. S. pombe RNAP III had previously been used to characterize molecular defects of a zebra fish rpc2/polr3b-mutant, slimjim that exhibits a tissue-specific phenotype [15,16]. We examined two HLD mutations in Rpc1, D366N in the catalytic center, and V891N at a critical interface with the Rpb5 subunit [45], along with unmutated WT Rpc1, in S. pombe. We assessed their effects on nascent precursor-tRNA levels, which are widely used to compare RNAP III transcription rates (reviewed in [47]). By this measure, Fig 5A showed for the three tRNA genes examined that the mutations reduced RNAP III transcription relative to WT Rpc1.
According to the model proposed here, as RNAP III activity decreases, reduced amounts of pre-tRNA substrates would better meet the limited supply of Trm1, and their modification efficiency, i.e., the mole fraction of a mature tRNA bearing m22G26, would increase. In agreement with this, tRNA-HydroSeq shows significantly more G26 misincorporation in both Rpc1 mutants relative to WT (Fig 5B). This was confirmed by the PHA26 assay, which showed lower D-AC stem probe signal in the Rpc1 mutants relative to Rpc1-WT (Fig 5C, lanes 1–3). The Rpc1 mutations also led to m22G26 hypermodification in maf1Δ (Fig 5C, lanes 4–6). Moreover, both Rpc1 mutants robustly reversed antisuppression in the maf1Δ strain relative to wild-type Rpc1 and also increased TMS in WT cells (Fig 5D). These data strengthen the model.
As noted above, rapamycin induces nutrient-related stress through the TOR pathway. We examined effects of differing nutrient on m22G26 modification and TMS. tRNA-HydroSeq was performed on wild-type (WT) S. pombe cells grown in minimal (EMM) and rich (YES) media, the latter known to support faster growth, and in EMM in which trm1+ was over-expressed. Total G26 misincorporations in Trm1 targets were higher in YES relative to EMM, comparable to cells over-expressing trm1+ in EMM (Fig 6A).
Examination of individual tRNA G26 misincorporations in minimal (EMM) and rich (YES) media yielded intriguing results. Although some of the greatest increases were in tRNAs that were largely hypomodified in EMM, the response was nonuniform (Fig 6B). Most strikingly was that a very similar pattern of increased misincorporation was observed for cells in YES and when trm1+ was over-expressed in EMM (Fig 6B). This nonuniform response would appear to reflect individualized substrate-specific responses to an increase in Trm1 activity (Discussion). The nine G26 tRNAs that showed no misincorporation in EMM (≤0.05) also showed no significant increases in misincorporation in YES or with trm1+ over-expression, providing more evidence that these are not substrates for m22G26 modification, non-targets of Trm1 (Fig 6B).
The increases in G26 misincorporation in WT cells grown in rich (YES) relative to minimal (EMM) media were greater for some tRNAs than others, and this was especially so for tRNAThrCGT (Fig 6B). The PHA26 assay was used to validate this by comparing tRNAThrCGT m22G26 content in YES and EMM. Fig 6C showed that the D-AC stem probe signal for tRNAThrCGT was much lower in YES vs. EMM, while the T-loop probe yielded similar signals, indicating a relatively high level of m22G26 modification in the rich (YES) media. By contrast, tRNAGlnCTG in YES and EMM showed near equal reactivity with its D-AC stem and T loop probes (Fig 6C).
To gain insight into a potential mechanism controlling the differential m22G26 modification levels in minimal and rich media, we examined Trm1 levels in extracts from the WT cells grown in YES and EMM and the trm1+ over-expressing cells in EMM by western blotting using tubulin on the same blot as a loading control (Fig 6D). Surprisingly, this showed similar levels of endogenous Trm1 in extracts from cells in grown in YES and EMM (Fig 7D, lanes 1, 2). The over-expressed 3X-FLAG-tagged Trm1 was observed as a slower migrating band in lanes 3 and 4 indicated to the right of Fig 6D. Quantification using Odyssey infrared imaging revealed that the 3X-FLAG-Trm1 accumulated to about 4-fold higher than endogenous Trm1 in the same cell extracts (lanes 3, 4). We conclude that while m22G26 modification efficiency differs dramatically in EMM and YES this is not reflected by Trm1 polypeptide levels detectable by western blotting, whereas over-expression of Trm1 is readily observed.
We next reasoned that rapamycin, which induces a nutrient-related stress response that includes RNAP III repression [48], would lead to increases in m22G26 modification efficiency and TMS. As alluded to above, the yYH1 (WT; maf1+ trm1+) strain is partially suppressed relative to a wild-type ade6+ allele strain and can therefore reveal an increase or decrease in TMS ([27] and refs therein). Fig 6E shows that while our WT strain is less suppressed relative to ade6+ but more suppressed than maf1Δ in the absence of rapamycin, its suppression increases relative to maf1Δ in the presence of rapamycin. Thus, rapamycin led to an increase in sup-tRNASerUCA-mediated suppression in maf1+ but not in maf1Δ or trm1Δ cells. Western blotting showed that Trm1 levels were comparable in the rapamycin and control (DMSO) treated WT (maf1+) cells (Fig 6F). The PHA26 assay showed a lower ratio of D-AC stem to T-Loop signal for tRNALeuCAA in WT (maf1+) cells treated with rapamycin relative to no rapamycin whereas no difference in D-AC stem signal was observed for the maf1Δ cells (Fig 6G). This reflected significant increase in m22G26 modification efficiency specific to rapamycin treated WT (maf1+) cells. As expected, maf1Δ cells showed no response to rapamycin in the assay for sup-tRNASerUCA activity, TMS (Fig 6E), or in m22G26 modification efficiency (Fig 6G).
S. cerevisiae MAF1 WT and maf1Δ cells were compared for tRNATyr m22G26 content by PHA26 (Fig 7A). The quantitative modification index revealed 2-fold hypomodification in maf1Δ relative to WT (MAF1) (Fig 7A, mod index under lanes). Ectopic expression of TRM1 in S. cerevisiae maf1Δ cells carrying the SUP11 ochre suppressor-tRNATyrUUA and the ochre-suppressible ade2-1 allele led to significant reversal of the maf1-antisuppression phenotype (Fig 7B).
We also examined m22G26 modification efficiency in human embryonic kidney (HEK) 293 cell tRNAs in response to serum starvation (Fig 7C, lanes 1–3) and treatment with rapamycin (Fig 7D), both of which repress RNAP III and cellular proliferation [49]. In both conditions the m22G26 content of tRNAAsnGTT increased as reflected by the modification index (Fig 7C lanes 1–3, D).
For the experiment in Fig 7C, after serum was added back to the serum-starved cells, their tRNAAsnGTT became less modified (Fig 7C, compare mod index, lanes 3 & 4). These results collectively indicate that the relationship between RNAP III activity and tRNA m22G26 modification efficiency has been conserved through evolution.
This work demonstrates that increases or decreases in global RNAP III activity lead to inverse changes in the efficiency of m22G26 modification of specific tRNAs and this impacts the functional activity of sup-tRNASerUCA in the expression of ade6-704 mRNA encoding an adenine synthetic enzyme, that produces a suppression phenotype. The collective results show that the link connecting RNAP III and m22G26 efficiency is due to a limiting amount of Trm1, the tRNA G26 dimethyltransferase. The data indicate that this link has been conserved through evolution, among two very highly diverged yeasts and human cells [50].
Maf1 is the conserved central regulator of RNAP III and is under the control of TOR which acts to coordinate growth and proliferation in response to multiple environmental cues including nutrient availability. Repression of RNAP III during starvation is fitting since the energy cost of excessive tRNA synthesis that occurs in the absence of Maf1 is wasted. Indeed, recent analyses indicate that Maf1 is a major mediator of metabolic efficiency [21]. We speculate that the increased efficiency of m22G26 modification that accompanies decreased RNAP III activity may be an economical way to enhance tRNA function under these conditions.
By quantifying tRNA-HydroSeq read counts representing mature tRNAs we found good correlations among the tRNA profiles in WT and maf1Δ cells (Fig 3A). An earlier study of MAF1 in S. cerevisiae that used microarrays containing tRNA gene sequences found that transcripts encoded by intron-containing tRNA genes were generally much more elevated in the maf1-mutant than were tRNAs from intron-less genes [51]. It was later found that the apparent accumulation of intron-containing precursor-tRNAs appeared to result from nonuniform posttranscriptional events such as precursor-tRNA intron processing due to saturation of the tRNA nuclear exportin, Los1 [52]. By using tRNA-HydroSeq to analyze mature tRNAs, good correlations were found among different tRNAs in the WT and maf1Δ cells (Fig 3A).
It is interesting that S. pombe cells in rich media harbor enough Trm1 activity for high efficiency m22G26 modification. However, although m22G26 modification efficiency was significantly higher in rich relative to minimal media (Fig 6B), the levels of Trm1 protein in rich and minimal media were similar (Fig 6D, lanes 1, 2), suggesting that Trm1 activity may be stimulated during growth in rich media by a posttranslational mechanism.
All tRNAs share features that allow recognition by RNase P, RNase Z, and certain other enzymes, but each also harbors features that must contribute to their unique identity. The wide range of m22G26 modification efficiency seen in Figs 3C and 6B is consistent with a hierarchical substrate preference of Trm1. Analysis in EMM and YES indicate increased m22G26 modification efficiency in the latter, more strikingly for some tRNAs than others, and this pattern was mimicked by over-expression of Trm1. These data would appear to reflect a relationship between Trm1 and the distinctive specificities of its many substrates, consistent with biochemical studies [43], but illustrated here for a wide range of unique cellular tRNAs on a tRNAomics-wide scale.
S. cerevisiae TRM1 exhibits genetic interactions with a number of genes including ones encoding factors involved in tRNA biogenesis and metabolism including several other modification enzymes, the tRNA export factor LOS1, the pre-tRNA chaperone and La protein homolog, LHP1, and MAF1 (see this at the Saccharomyces genome database, SGD, at http://www.yeastgenome.org/locus/S000002527/interaction).
In the absence of TRM1 and m22G26, some tRNAs are substrates for surveillance by rapid tRNA decay (RTD) in S. cerevisiae and the trm1Δ cells were shown to exhibit temperature-sensitive growth deficiency [53]. That m22G26 may affect tRNA structure is also consistent with findings that in its absence, tRNALysUUU and tRNATyrGUA become substrates for surveillance by 5'-3' exonuclease Xrn1-mediated RTD [54]. However, although m22G26 modification efficiencies of target tRNAs differed in S. pombe maf1Δ and maf1+ cells, their relative steady state levels remained similar (Fig 3A). Moreover, there was no deficiency of sup-tRNASerUCA or tRNASerUGA levels in maf1+ relative to maf1Δ cells (Fig 2C and 2E) despite differences in the percent content of their m22G26. Thus the data indicate that a difference in m22G26 content was a critical determinant to the function of sup-tRNASerUCA in maf1Δ and maf1+ cells. The cumulative results indicate that m22G26 increases the specific activity of the tRNA.
Apparently, m22G26 also increases specific activity of S. cerevisiae sup-tRNATyrUUA (Fig 7A and 7B), and we believe that it would be reasonable to expect that it may do so for some other tRNAs. We note that the tRNAs whose m22G26 modification efficiencies vary most upon changes in RNAP III activity may impact the translation of some mRNAs more than others, dependent on their cognate codon use bias, and that this may contribute to a phenotype or stress response (see [9,11]) although to attempt to determine if this is decipherable for the subset of m22G26-tRNAs would require substantial bioinformatics and experimental resources and is beyond the scope of this study.
As noted above, evidence that Trm1 acts redundantly with the pre-tRNA chaperone, La protein, suggests that by modifying tRNA with m22G26 it may contribute to proper tRNA folding [24]. G26 resides at the junction between the D-stem and the anticodon stem, and its N2-dimethylation, which interferes with normal Watson-Crick base pairing may contribute to prevention of tRNA misfolding. It is also notable that treatment of cells with 5-flurouracil (5FU), which is incorporated into RNA, sensitizes S. cerevisiae to loss of genes that encode tRNA modification enzymes whose nucleoside targets localize at or near the stems junction, and include TRM1 [55]. These observations together with evidence that m22G26 can stabilize correctly folded anticodon stems [56], suggest that it may enhance tRNA specific activity by improving fit in the ribosome.
It has been known that inactivation of MAF1 leads to increased translation fidelity in S. cerevisiae although the mechanism has been unclear [57]. As a greater percentage of the tRNAs acquire m22G26 in maf1-mutants this may be a mechanism that contributes to their increased fidelity. Future studies that compare translational fidelity in maf1Δ single mutants with maf1Δ trm1Δ double mutants may address this.
S. cerevisiae Trm1 is tethered to the inner nuclear membrane via a specific amino acid sequence tract [25,58]. A genome-wide global ORF analysis of S. pombe found Trm1-GFP as nucleoplasmic and mitochondrial with no noted observation of perinuclear localization [59]. Although nuclear residence may limit the time during which a nascent pre-tRNA transcript might have access to acquire the m22G26 modification, retrograde tRNA transport should theoretically allow iterative access to Trm1 (see [54]). Therefore, the mechanism by which cells maintain Trm1 activity in a functionaly limiting amount in minimal media is unclear. In any case, the data suggest that regulation of Trm1 can under certain conditions, differentially impact tRNA activity.
Others have developed means to identify RNA modifications from deep sequencing data sets [38]. The ablation of G26 misincorporations following deletion of trm1+ (Fig 3D), provided essential evidence that the correlation of G26 misincorporations with the suspected modification was indeed due to m22G26. Thus, this approach can be used by tRNA-HydroSeq and similar methods together with genetics to obtain quality and quantity information toward studying the biology of certain tRNA modifications. tRNA-HydroSeq detected misincorporations corresponding to m1G9, m22G26, m3C32, A34I, and m1A58 in tRNAs known to carry these modifications in yeast and/or other species. As was the case for G26, the G9 and A58 misincorporation levels varied in a tRNA-dependent manner (S3 Fig). However, unlike G26, we found no functional correlation of the other modifications with maf1+ expression, RNAP III activity, and suppression phenotype. Notably, A34I was distinguished from the other misincorporations in that there was uniformly efficient misincorporation in all tRNA substrates (S3 Fig). In addition, we found an intriguing tRNA-specific effect on A34I for the Rpc1 mutants and for growth media (S4 Fig). For 10 of the 11 tRNAs with A34 there was no significant difference in the Rpc1 mutants whereas a single tRNA, SerAGA showed reduced A34 misincorporations in both mutants (S4A Fig). We also observed significant difference in A34 misincorporation unique to tRNASerAGA in YES vs. EMM (S4B Fig). While the basis of this specificity and its significance is unknown, we note that while tRNASerAGA can decode both UCC and UCU codons and that the ratios of these differ greatly in high-expression vs. low-expression mRNAs [60], only the UCC (wobble) codon decoding is dependent on I34. These data provide examples of utilities of tRNA-HydroSeq beyond the ability to follow G26 modification.
Although mutations in RNAP III catalytic subunits cause HLD, their effects on the RNAP III transcriptome had not been reported [46]. We recreated two of these mutations in S. pombe at residues highly conserved from yeast to man. The mutations caused decreased transcription of the three tRNA genes examined and were associated with alterations of m22G26 and A34I modification efficiencies.
We wish to emphasize that these HLD mutations were introduced into S. pombe as a means to globally decrease RNAP III activity for the purposes of this study. Nonetheless it is tempting to speculate that these mutations might have similar effects on human RNAP III activity and possibly analogous changes in tRNAs. The results suggest that while RNAP III activity may be increased or decreased globally due to a number of mechanisms, the output with regard to tRNA activity may be asymmetric or nonumiform. In such cases, nonuniform alterations of tRNA activities on different subsets of codon-biased mRNAs may contribute to phenotype [1,9,10].
To the best of our knowledge maf1-antisuppression had been observed only in S. cerevisiae and for ochre sup-tRNAs of Tyr identity [22,30], which in S. cerevisiae carry m22G26. The present work extends this to S. pombe and sup-tRNASerUCA. The S. cerevisiae maf1-1 mutant was isolated from a mod5-mutant deficient for cytoplasmic i6A37 modification [29]. However, the hypothesis that i6A37 hypomodification was responsible for maf1-antisuppression had not been tested experimentally [19]. Our data show no i6A37 deficiency in S. pombe maf1Δ cells and that antisuppression occurs despite efficient i6A37 modification of sup-tRNASerUCA. Instead the data show that m22G26 hypomodification of sup-tRNASerUCA is responsible for maf1-antisuppression. In S. cerevisiae, the tRNATyrGUA, from which the ochre suppressor SUP11 was derived, contains G26 and is hypomodified in maf1Δ cells (Fig 7A). In S. pombe, sup-tRNASerUCA contains G26 and is hypomodified in maf1Δ cells. Over-expression of Trm1 in both S. cerevisiae and S. pombe substantially reverses their maf1-antisuppression phenotypes (Figs 4F and 7B). We note that antisuppression reversal by Trm1 over-expression was incomplete. Among other possibilities this suggests that other factors involved in TMS may be limited for suppression in the context of increased tRNA synthesis in maf1Δ cells.
Treatment of S. pombe maf1+ cells with rapamycin, which represses RNAP III via maf1+, increased suppression accompanied by m22G26 hypermodification. Similarly, human cells treated with rapamycin showed robust increase in m22G26 modification content. Serum-starvation led to increased m22G26 modification that decreased with serum replenishment (Fig 7C). Likewise, S. cerevisiae tRNATyr was m22G26 hypomodified in maf1Δ relative to MAF1 cells. In S. pombe, unlike the other modifications examined, only m22G26 varied with maf1+ expression concordant with TMS activity. This specificity is noteworthy since although a genetic screen uncovered GCD10 and TRM10, responsible for m1A58 and m1G9 tRNA modifications, as well as TRM1 [61], our data showed that m1A58 and m1G9 were not altered in maf1Δ relative to WT or +maf1+ whereas m22G26 levels were. In summary, the link between RNAP III activity and m22G26 modification efficiency appears to be specific and conserved.
S. pombe strains used are listed in S2 Table. Cells were grown in minimal media (EMM lacking uracil) or in rich media (YES) to an OD600 of 1.0. S. pombe cells were seeded to an OD600 of 0.4, and ten-fold dilutions were plated on the appropriate media as indicated. For liquid growth, overnight cultures were diluted to OD600 of 0.25 and incubated for two hours after which rapamycin (AG Scientific Inc., R1018) at 200 ng/ml, or DMSO alone (Sigma, D2650) was added one hour prior to RNA isolation.
The trm1+ open reading frame was amplified from S. pombe genomic DNA using a forward primer containing sequence for 3X FLAG peptide and XhoI site and a reverse primer with XmaI site. The PCR products were digested with XhoI and XmaI and ligated into XhoI-XmaI digested pREP4X.
The maf1+ gene and its 600 bp upstream region was PCR amplified from genomic DNA and cloned into the XhoI-PstI sites of pRep4X (removing the nmt1+ promoter) resulting in plasmid CB235.
Total RNA was isolated using hot phenol. In short, 50 ml cultures grown from an OD600 of 0.1 to 0.5 were harvested, washed with water and resuspended in 300 μl TES buffer (10 mM Tris Cl pH 7, 10 mM EDTA, 1% SDS). 300 μl water-equilibrated phenol was added and incubated at 65°C for 45 minutes, with vortex every 15 min. To the samples, 300 μl chloroform was added and centrifuged. Supernatant was extracted twice with acid-phenol-chloroform and once with chloroform before precipitation with ethanol.
Total RNA was resolved in 6% NuPAGE TBE-Urea gel (Life Technologies) and transferred to positively charged nylon membranes. Probing and washes were done as described previously [14].
Using affinity-purified anti-i6A from rabbit [31] (a kind gift of Anita Hopper, OSU) at 1:50 and processed for chemiluminescence was as described [28].
For tRNA isolation, 50 μg of total RNA was separated on a 6% TBE-Urea polyacrylamide gel (21 cm X 18 cm X 0.15 cm) followed gel purification of RNA shorter than 5S rRNA. 300 ng purified tRNA was hydrolyzed in 10 mM bicarbonate buffer pH 9.7 at 90°C for 5 min. The hydrolyzed RNA was dephosphorylated using calf intestinal alkaline phosphatase (NEB) followed by 5' phosphorylation by T4 polynucleotide kinase (NEB). Barcoded, pre-adenylated, 3’ blocked Illumina adapters were ligated to the 3’ end using T4 Rnl2(1–249)K227Q enzyme (NEB). After heat inactivation, all ligated samples were pooled in ethanol and precipitated. In parallel, two size marker RNA oligos of 19 and 35 nt were radiolabeled using T4 polynucleotide kinase and ligated to pre-adenylated adapter and pooled. Ligated tRNA samples and markers were resolved in a 15% TBE urea acrylamide gel followed by gel purification of tRNA samples between 19 nt and 35 nt guided by the marker lanes. 5’ adapter was ligated to the gel purified samples using T4 RNA ligase I (Thermo). RNA with ligated 5’ and 3’ adapters were gel purified, subjected to reverse transcription (Superscript III, Life Technologies), amplified by PCR (10–12 cycles) and the band was gel purified and sequenced using Illumina HiSeq 2500.
tRNA read depths for all samples generally varied with tRNA gene copy number; from 278 for the lowest read from the single copy ArgCCG gene in one replicate to greater than one-million reads for highly abundant tRNAs. Sequence reads were mapped to a reference file comprising sequences of all 61 unique mature S. pombe tRNAs. Sequences that did not map to this file were then mapped to a file containing all 150 unique S. pombe precursor-tRNA genes. Read count tables were made using the mapping data and analyzed using DEseq software.
The fraction of reads that mismatched the reference gene sequence at each position were tabulated. The average values of the fraction of misincorporation among replicates was calculated and plotted (e.g., Fig 3C).
S. pombe cells were grown in 10 ml of the noted media to A600 of 0.7–1, washed with water and resuspended in 400 ul of 20% trichloroacetic acid (TCA). Glass beads (0.5 mm) were added and vortexed for 1 min. The beads were separated from the material, washed with 5% TCA and the wash was pooled with the material recovered. This was centrifuged at 6000 RCF for 10 minutes, the supernatant discarded and the pellet washed twice with 1 ml acetone. The pellet was air dried and dissolved in 400 ul of 1X SDS sample buffer containing fresh beta-mercaptoethanol. Aliquots were resolved on a 4–12% Bis-Tris gel (Life technologies) followed by transfer to PVDF membrane. The blot was blocked for 1 hour with 5% non-fat milk in PBS. Polyclonal anti-Trm1 antiserum raised against the C-terminal peptide, GPKSKPGKRTIAEVDSKS, in rabbit (Thermo Fisher Scientific, Waltham, MA; animal #PA9064, day 56 bleed) was used at 1:500 in blocking buffer with 0.1% Tween 20. Anti-Tubulin (Sigma, #T5168) was used at 1:4000. After 1 hr. the blot was washed 4 times with PBS-Tween. Appropriate secondary Abs (LI-COR) of different fluorescent emittances were used at 1:20,000 in 5% milk solution in PBS with 0.2% Tween-20 and 0.01% SDS for 1 hour followed by 4 washes in PBS-T. The washed blot was scanned using LI-COR Odyssey Clx system and the images processed and bands quantified using ImageStudioLite software.
The glyceraldehyde phosphate dehydrogenase (GPD) promoter followed by a 3X-HA tag was amplified from the pYM-16 plasmid (PCR-toolbox, EUROSCARF) and cloned into the SacI-NotI sites of pRS426 to generate the pRS426GPD vector. The S. cerevisiae TRM1 gene starting from the ATG start codon to 855 bp downstream of the stop codon was amplified from genomic DNA and digested with NotI and XhoI present in the primers used. The fragment was inserted in the corresponding site of pRS426GPD. The empty vector (pRS426GPD) and the Trm1 clone were used to transform S. cerevisiae maf1Δ strain, MB159-4DΔ [57] which carries SUP11 sup-tRNATyrUUA, and plated on SC agar lacking uracil with 10 mg/l adenine.
|
10.1371/journal.ppat.1005826 | The Invertebrate Lysozyme Effector ILYS-3 Is Systemically Activated in Response to Danger Signals and Confers Antimicrobial Protection in C. elegans | Little is known about the relative contributions and importance of antibacterial effectors in the nematode C. elegans, despite extensive work on the innate immune responses in this organism. We report an investigation of the expression, function and regulation of the six ilys (invertebrate-type lysozyme) genes of C. elegans. These genes exhibited a surprising variety of tissue-specific expression patterns and responses to starvation or bacterial infection. The most strongly expressed, ilys-3, was investigated in detail. ILYS-3 protein was expressed constitutively in the pharynx and coelomocytes, and dynamically in the intestine. Analysis of mutants showed that ILYS-3 was required for pharyngeal grinding (disruption of bacterial cells) during normal growth and consequently it contributes to longevity, as well as being protective against bacterial pathogens. Both starvation and challenge with Gram-positive pathogens resulted in ERK-MAPK-dependent up-regulation of ilys-3 in the intestine. The intestinal induction by pathogens, but not starvation, was found to be dependent on MPK-1 activity in the pharynx rather than in the intestine, demonstrating unexpected communication between these two tissues. The coelomocyte expression appeared to contribute little to normal growth or immunity. Recombinant ILYS-3 protein was found to exhibit appropriate lytic activity against Gram-positive cell wall material.
| Innate immune defenses against bacterial pathogenesis depend on the activation of antibacterial factors. We examined the expression and relative importance of a gene family encoding six invertebrate-type lysozymes in the much-studied nematode C. elegans. The ilys genes exhibit distinct patterns of tissue-specific expression and response to pathogenic challenge and/or starvation. The most abundantly expressed, ilys-3, exhibits constitutive pharyngeal expression, which we show is essential for efficient disruption of bacteria under non-pathogenic growth conditions, and consequently it contributes to normal longevity. ilys-3 is also strongly up-regulated in intestinal cells after starvation or exposure to Gram-positive pathogens such as Microbacterium nematophilum and acts as a ‘slow-effector’ in limiting pathogenic damage from intestinal infections. We show that this induction by pathogens depends on the action of an ERK-MAPK cascade, which acts in pharyngeal rather than intestinal cells; this implies communication between pharynx and intestine. Tagged ILYS-3 protein was detected mainly in recycling endosomes of intestinal cells and in the intestinal lumen after starvation. ILYS-3 was also expressed in coelomocytes (scavenger cells) but we found that these cells make little or no contribution to defense. We examined the enzymatic properties of recombinant ILYS-3 protein, finding that it has lytic activity against M. nematophilum cell-walls.
| Most animal epithelia possess innate defense systems that sense pathogenic and toxic insults and transmit stranger/danger signals to activate appropriate counter measures. The efficacy with which animals can respond to pathogens determines whether organismal homeostasis can be maintained and microbial clearance achieved. In vertebrates, the transcriptional programs that control the epithelial production of antimicrobials and mucosal homeostasis both act via coordinated activation of common innate immune receptors: Toll-like receptors (TLR), Nod-like receptors (NLR) and the cytosolic helicases RIG-1 and MDA5. Some of these also function as sensors of endogenous or exogenous damage-associated molecular patterns [1].
The invertebrate C. elegans has proved a valuable model to deconstruct biological processes that deal with the way animals detect danger signals and respond to life-threatening events such as toxic chemicals, DNA damage, metabolic stress and pathogens. Damage inflicted on nematode tissues can trigger a variety of functionally conserved molecular events aimed at limiting and repairing damage or sustaining viability while adverse conditions persist (reviewed elsewhere [2–6]).
This nematode is a bacterial feeder that spends much of its life in decomposing vegetable matter and depends on microbes as its source of food. To survive an environment rich in potentially damaging microorganisms, C. elegans has evolved an epithelial defense system that is able to carry out complex functions. Thus surface-exposed tissues such as epidermis [7], pharynx [8], intestine [8,9], vulva [10], and hindgut [11] are not simple physical barriers, but have all been shown capable of eliciting appropriate immune defense when exposed to potential pathogens. A crucial regulatory challenge to the nematode is the requirement for an adequate discrimination between real threats and innocuous edible food sources.
Much of what we know about the way this nematode is able to cope with pathogen insults comes from studies with intestinal infections. The response of intracellular signaling pathways within the intestine to microbial pathogens has been studied extensively in C. elegans and has been reviewed elsewhere [4,12]. Intestinal epithelia recognize and respond to extracellular pathogens by activating specific intracellular programs that include families of MAP kinases such as p38 and JNK, among others, that result in expression of candidate immune effectors that fight pathogens.
Predicted antimicrobial proteins elicited in C. elegans after immune challenge include among others, C-type lectins, CUB-like domain proteins and lysozymes. Many of them probably help the organism maintain intestinal homeostasis under normal intestinal growth conditions [13]. However, and except for the saposins ssp-1 [14] and ssp-5 [15,16], for all the others their critical effector function in antimicrobial defense remains to be demonstrated.
We have studied the response of C. elegans to a Gram-positive bacterium that adheres to the rectal and anal cuticle, Microbacterium nematophilum [17]. This nematode-specific pathogen has provided a means to analyze the complex biochemistry of the surface coat and the underlying cuticle [3,5,18–20] as well as innate immune responses. Orally ingested M. nematophilum alone instigates inflammation and tail swelling [21]. Despite the fact that the most obvious response to infection is rectal colonization and the induction of inflammation in the rectal tissues, this bacterium also establishes itself in the gut of the worm. This makes it a good system to investigate complex effects that occur in the digestive tract associated with chronic gut colonization.
We, and others, have previously found a cluster of invertebrate lysozymes that include ilys-2 and ilys-3, (and the probable pseudogene ilys-1) is upregulated upon exposure to many different bacteria, namely M. nematophilum [22], Staphylococcus aureus ([8] and [23]), Bacillus thuringiensis DB27 [23], Enterococcus faecalis [24], Cronobacter sakazakii [25], Salmonella enterica serovar Typhimurium [26], Photorhabdus luminescens [27] as well as fungi Candida albicans [28] and Drechmeria coniospora [24].
Lysozymes are important bactericidal effectors present in phylogenetically diverse organisms from vertebrates to bacteriophages. In addition to promoting digestion, lysozymes confer direct broad-spectrum antimicrobial properties and perform essential functions in innate immunity. They are effective at targeting the cell wall of many Gram-positive bacteria by hydrolyzing the 1,4-β-glycosidic linkages between N-acetylmuramic acid and N-acetylglucosamine that make up the carbohydrate backbone of cell wall peptidoglycan. Although primarily associated with defense against Gram- positive bacteria, Gram-negative bacterial cell walls can also be attacked by these enzymes [29,30]. Based on their structural, catalytic and immunological differences, lysozymes have been classified into 6 sub-types. Major lysozymes found in animals belong to chicken(c), goose (g), invertebrate (i) or protist-type groups. All these four groups have been identified in invertebrates. C. elegans has numerous lysozyme genes belonging to both the protist-type (lys-1 to lys-10) and invertebrate type (ilys-1 to ilys-6) lysozymes reviewed by [31] and [32] and this study. Besides their ability to hydrolyse 1,4-β-glycosidic bonds in the glycan moiety of peptidoglycan, invertebrate lysozymes can also exhibit an isopeptidase (known as destabilase) activity, which specifically targets the isopeptide bonds established between D-Glutamate and L-Lysine present in some Gram-positive peptidoglycans [32–35]. An additional activity attacking isopeptide-type substrates can potentially enhance the lytic activity of invertebrate lysozymes. However, in most cases such synergy is still elusive and awaits further evidence.
While the protist-type lysozymes have been shown to play a role in host defense in interactions that involve C. elegans and B. thuringiensis [36], Serratia marcescens [37], Salmonella typhimurium [38], S. aureus [8] and M. nematophilum [22], little is known about the function of the i-type lysozymes in the worm. We therefore investigated expression, induction and function of ilys-1-6 focusing our attention on ILYS-3.
Here we show that C. elegans invertebrate lysozyme ILYS-3 has important roles in both healthy and diseased worms; it is required to enable normal pharyngeal grinder function and to defend against bacterial pathogens. We find that ILYS-3 is a slow effector that is induced by danger signals generated both by bacterial pathogens and starvation. Under non-pathogenic conditions, in the gut ilys-3 expression levels undergo a post-developmental regulatory oscillation. Levels increase after L1 hatching, decline after the L2 transition, and after L4 transition increase again becoming abundantly expressed in the intestine of adult worms.
Furthermore, we provide evidence that the M. nematophilum-mediated intestinal induction of ilys-3 is regulated through the pharyngeal action of the extracellular signal-regulated MAP kinase (ERK) pathway in a non-cell- autonomous manner. This effect implies a mechanism whereby pharyngeal cells sense and propagate danger signals and further instruct antimicrobial responses in distal tissues. This systemic immune regulation of ilys-3 reveals an unexpected communication between the pharyngeal and intestinal tissues in this multicellular organism.
Finally, using recombinant ILYS-3 fusion proteins we demonstrate their in vitro hydrolytic activity against the peptidoglycan of Gram-positive Micrococcus luteus and M. nematophilum. Such lytic activity will impact on the integrity of the cell walls and bacterial viability, which ultimately will affect proliferation in the nematode intestine. In addition, ILYS-3 probably contributes to the ability to digest and cope with the large amount of peptidoglycan fragments generated by a Gram positive diet (either pathogenic or non-pathogenic).
To provide insights into their site of action and response to infection, we investigated the expression pattern of the six invertebrate lysozymes present in the C. elegans genome. To this end we made transcriptional reporter fusions using DsRed2, CFP or GFP proteins (Fig 1 and S1 Fig). We found that all the invertebrate lysozymes have in common an intestinal expression pattern (Fig 1B and 1C and S1C Fig and S1E Fig, S1F–S1H Fig and S1K Fig and S1 Table). However, one subset appeared to have additional pharyngeal expression whereas the other exhibited some neuronal promoter activity. The four ilys genes ilys-1, -2, -3 and -6 exhibited distinct patterns of constitutive pharyngeal expression. ilys-1 expressed in the pm3 cells in the procorpus, the pm4 cells in the anterior bulb pharyngeal muscles (sieve) and the marginal cells mc1 and mc2 (S1A Fig and Fig 1); ilys-2 expressed in the pharyngeal muscle pm3 and in the nerve ring (S1B Fig); ilys-6 in the pharyngeal gland cells and in the coelomocytes (S1I and S1K Fig). A detailed characterization of ilys-3 expression is presented below. The second subset included ilys-4 and -5, and revealed neuronal-specific expression pattern. S1D Fig shows the expression of ilys-4 in a set of interneurons, the presumptive AVE/AVA and RIG/ALA cells that send processes that extend the entire length of the ventral and dorsal nerve cord respectively (S1E Fig and S1 Video). S1G Fig and S2 video depict the expression of ilys-5 in interneurons in the head, the presumptive pharyngeal neuron I1 and AIM/AIY.
This analysis revealed an unexpected variety in the ilys expression patterns, suggesting that these genes may have distinct biological roles. Anatomically, expression of ilys-1,-3 and -6 covers the whole pharynx suggesting that they may have complementary functions. Given the considerable chemical, organizational and architectural diversity within bacteria cell walls and peptidoglycans (PGNs), one might expect little or no redundancy among ilys genes.
We found distinct allelic forms of ilys-1 in various natural isolates of C. elegans. These included several exonic SNPs and a 10 bp insertion in the second intron that is likely not to impact on the coding region. Unlike all the other ilys genes, ilys-1 does not encode a signal sequence, its first exon consists only of two methionines, and this together with the accumulation of several polymorphisms suggests that this invertebrate lysozyme is a pseudogene, despite its functional promoter.
The functional relevance of the neuronal expression of invertebrate lysozymes is unknown. However, lysozymes have been thought to play a role in neurodegenerative diseases as a part of amyloid plaque pathology. In a fly model of Alzheimer's disease for example, neuronal co-expression of lysozyme and amyloid-β1–42 diminished soluble and insoluble amyloid species, and prolonged survival of Aβ-expressing flies [39].
In this study, we focused our attention on the ilys-3 gene because this was one of highly expressed genes induced in C. elegans upon exposure to M. nematophilum and its reporter showed strong induction under various conditions (as described below).
To determine its spatial distribution and responses to infection, we generated four expression reporters. Two of these consisted of a short (1.05 kb) and a long (4.5 kb) version of the ilys-3 5' sequence (presumed promoter) driving GFP expression. The third and fourth types were translational reporters with GFP or mCherry fused to ILYS-3 ORF at the N- or C-terminus, respectively. Both translational reporters used the 4.5 Kb ilys-3 promoter.
The short promoter drove GFP expression mainly in the pharynx grinder muscles pm7, the isthmus marginal cell mc2 and muscle cell pm5 (Fig 1A and 1B). Grinder action is believed to disrupt bacterial cells. In addition, all intestinal cells exhibited ilys-3 promoter activity, with some variability (Fig 1B and 1C). Intestinal expression of ilys-3 appeared to be temporally regulated. GFP was first detected in intestine of L1 hatchlings but declined at the L2 transition. By the L4 stage, the intestinal GFP signal became stronger and continued through adulthood. The intensity of GFP was very high in aging adult worms that were no longer self-fertile. Whereas the intestinal expression of ilys-3 changed during larval growth and adult life, the pharyngeal GFP reporter expression remained constant in larval and adult stages. Weak epidermal promoter activity was also detected.
The long promoter revealed additional GFP expression in the six scavenger coelomocyte cells, but at a significantly lower level in other tissues (Fig 1K). This reporter included only part of the 28 Kb sequence between ilys-3 and its closest upstream neighbor, C45G7.4, but most of the region is occupied by a large cadherin gene cdh-10, on the other strand, whose 3' end is only 1.2 Kb from the predicted ATG of ilys-3.
N and C terminal tags showed essentially the same spatial expression patterns. However, higher levels of fluorescence intensity were seen in transgenic animals harboring extrachromosomal arrays of the C-terminal mCherry translational fusion, so this reporter was used in most subsequent experiments.
Steady-state levels of ILYS-3::mCherry were barely detectable until late L4 stage. When the animals entered adulthood, the mCherry signal became stronger in the pharynx, the intestine and in the coelomocytes. The epidermis also expressed mCherry albeit at low levels (S2 Fig).
Representative images of this expression are presented in Fig 1D–1I. ILYS-3::mCherry was abundant in the intestinal lumen in old adults (Fig 1E) and in animals subjected to starvation (Fig 1H). Such luminal secretion of ILYS-3 by intestinal cells was most conspicuous in dauers induced on nutrient-depleted plates. By contrast, we noted surprisingly little of the ILYS-3::mCherry signal in the intestinal lumen of animals under normal growth. To attain additional evidence that tagged ILYS-3 accumulates in the intestinal lumen, we fed animals with fluorescent microspheres just prior to their imaging. As shown in Fig 1F and 1I ingested microspheres accumulated in the intestinal lumen marked with red fluorescence.
The translational mCherry fusion protein shows a punctate expression pattern in most tissues in the pharynx and intestine (Fig 1D and 1E). In the intestinal cells, ILYS-3::mCherry positive vesicles were enriched in the cytosol.
Together these data show that ILYS-3 has a dynamic expression pattern under standard growth conditions and is found primarily in the digestive tract of C. elegans in the pharyngeal cells and the intestine, indicating that it may play roles both in early stages of bacterial grinding and in further lysis required for nutrient digestion.
Given the unexpected paucity of the ILYS-3::mCherry signal in the intestinal lumen of young animals, we examined the subcellular trafficking of ILYS-3. For this, we used well-established markers for early endosomes (GFP::RAB-5), late endosomes (GFP::RAB-7), and Trans-Golgi Network (TGN) and apical recycling endosomes (GFP::RAB-11) (Fig 2). The intestine of C. elegans is a polarized epithelium with apical membranes and microvilli facing the lumen and a basolateral surface contacting the body cavity (pseudocoelom).These two domains are separated by apical junctions. Molecules that reach the lumen are believed to be delivered through complex sorting machineries in four distinct ways: 1. directly, 2. indirectly, via specifically dedicated vesicles, 3. via apical recycling endosomes or 4. through transcytosis by means of membrane-bounded carriers.
As shown in Fig 2 only a subpopulation of ILYS-3::mCherry colocalized and appeared as puncta around the rim of the GFP::RAB-7 labeled late endosomes (Fig 2A–2C). This population of mCherry-positive vesicles was distributed through the cytoplasm, in the middle focal plane. We found a Pearson's correlation coefficient of 0.39, indicating a low degree of colocalization with the total population of RAB-7-positive endosomes. Some of the RAB-11-positive recycling endosomes that appeared scattered in the cytoplasm were also mCherry positive. Pearson's coefficient at 0.47, suggested some degree of correlation. However no ILYS-3::mCherry signal could be detected at the apical membrane and its associated puncta (Fig 2D–2F). Likewise, no mCherry and GFP-double labeling was detected with the transmembrane marker for the apical intestinal membrane OPT-2 [40].
A separate but pertinent population of ILYS-3-positive vesicles was associated with the acidic lysosome-related organelles (LROs) in the intestine and were LysoTracker Green-labelled (Fig 2G–2I). The Pearson's coefficient was 0.62, implying a greater degree of correlation of ILYS-3 signal with these organelles. LROs are destination vesicles for degradation of extracellular macromolecules from both the apical and basolateral trafficking routes. Fig 2I illustrates also the lack of ILYS-3::mCherry in the apical membrane in contrast to the positive green-labelled-LysoTracker, internalised from apical uptake.
At its simplest, the surprising absence of strong apical and luminal mCherry signal in young animals could indicate that these animals simply degrade the tagged ILYS-3 more efficiently than older siblings. However, the presence of the ILYS-3:mCherry in endosomal and LRO compartments suggests that there could be fast turnover and/or re-uptake of the luminal ILYS-3 through an endocytic recycling route. Hence, the protein could still be present in the apical compartments, but below thresholds of detection.
Endocytic recycling of proteins from endosomes to the plasma membrane is important in cellular processes such as nutrient uptake and immune functions [41]. We next sought to determine whether the reason why we were unable to detect apical mCherry was because ILYS-3 was associated with endosomes that perform basolateral recycling. We used the rme-1 mutant that specifically accumulates abnormally high numbers of basolateral recycling endosomes but does not accumulate increased numbers of early, late, or apical recycling endosomes in their intestines [42]. In the rme-1 mutants a large number of ILYS-3::mCherry vesicles associated with a basolateral tubule-vesicular network were detected (Fig 2K) relative to WT background (Fig 2J). Together these results suggest that at least part of ILYS-3::mCherry associates with the interconnected endosomal tubules and that basolateral recycling of ILYS-3 does not require RME-1. Rather than being actively delivered to the apical plasma membrane domain, ILYS-3::mCherry appeared to be dispatched, via the basolateral route, to recycling endosomes. We also found that M. nematophilum infection induced high ILYS-3::mCherry signals in the tubular-vesicular recycling endosome network of the intestinal cells, strongly resembling animals lacking RME-1 (S3 Fig). The basolateral recycling endosomal location of ILYS-3 was surprising given that abundant ILYS-3::mCherry was observed in the lumen of old animals and dauers.
RAB-10 has been shown to function in basolateral recycling in the intestine upstream RME-1 [43]. rab-10 mutants exhibit enlarged early endosomes that are likely to play roles in receiving, sorting and also recycling cargo received from both the apical and the basolateral domains. We therefore tested whether a defect in such key endocytic route could interfere with the distribution of basolateral-destined ILYS-3::mCherry. In rab-10(ok1494) deletion mutants we observed an aberrant accumulation of the red signal on the apical membrane and its associated vesicles of the intestine (Fig 2L). This was consistent with an impairment on basolateral recycling that characterizes this genetic background. We interpret these results, as an indication that ILYS-3 can be routed to the apical side of the intestinal cells, albeit at low levels in WT backgrounds.
As described above, only in old adults and dauer larvae do we see abundant luminal accumulation of ILYS-3::mCherry. The simplest explanations for this are that the turnover of the protein is impeded in intestinal epithelial cells of these animals and/or it resulted from the leakage due to apical cell damage. Tight junctions act as fences, preventing leakage of basolateral components into the apical compartment. However, if their integrity is compromised, leakage may ensue. While these two scenarios seemed possible in old animals, they appeared unlikely in dauers. To address this issue we first examined where ILYS-3 accumulated in dauers. For this, we co-expressed ILYS-3::mCherry and the intestinal marker RAB-11 tagged to GFP, which labels the apical plasma membrane. As shown in Fig 3A–3C and Fig 3J, most of red signal conferred by ILYS-3::mCherry accumulated within the extracellular milieu, i.e. in the pharyngeal and intestinal lumina. In the intestine mCherry was detected between cells marked by GFP::RAB-11 and the intensity profiles of the two proteins have little overlap (Fig 3J). Steady levels of high luminal ILYS-3::mCherry could still be observed in 25-day old dauer larvae, suggesting that the protein was not actively degraded in these animals.
Strikingly, ILYS-3::mCherry expression pattern in the lumen of dauers rapidly returned to the original localization when animals resumed development after re-exposure to OP50. As illustrated in Fig 3D mCherry began to accumulate intracellularly even before animals resumed feeding in the first hour after they encountered food. The figure is representative of a recovering dauer larva transferred to a bacterial lawn mixed with fluorescent microspheres. No fluorescent beads were detected in the intestinal lumen, indicative of no feeding activity. The transgenic animal shown co-expressed ILYS-3::mCherry and the GFP::RAB-7, a marker for early endosomes near plasma membrane and late endosomes deeper in the cytoplasm. Red fluorescence signal was seen in the lumen and the cytosol at the most anterior and posterior intestinal cells, respectively. Three hours later, fluorescent beads denoting feeding activity could be found in the intestinal lumen, where mCherry was still highly abundant (Fig 3E and 3F). Upon overnight feeding the punctate expression pattern of ILYS-3::mCherry in the cytosol was re-established (Fig 3G–3I).
Taken together these data show that the vesicular and luminal ILYS-3::mCherry distribution is modifiable and regulated by nutritional conditions. This invertebrate lysozyme is secreted in the intestinal lumen but it can also be detected deeper in the cytosol in LROs and recycling-associated vesicles, both of which are known final destinations for internalized fluid-phase cargos. Our observations suggest an intracellular role for the C. elegans ILYS-3 in addition to luminal activity, which includes intracellular intestinal compartments where it could be acting to degrade residual peptidoglycan fragments.
Previous microarray analysis showed that during early infection by M. nematophilum, ilys-2 and ilys-3 were upregulated by 3 and 4- fold, respectively [22]. To further investigate the role of ilys-3 gene in host-pathogen defense, we visualized ilys-3 transcriptional activation in vivo using L1 transgenic worms expressing the GFP reporter transgene (short ilys-3 promoter). We used this promoter because the corresponding transgene showed stronger inducibility than the longer version.
We found that ilys-3 is highly responsive to immune challenges by different Gram-positive bacterial pathogens. Under steady state conditions, when animals develop on lawns of E. coli OP50 (Fig 4Ai), only low levels of GFP expression are detected in the gut. In contrast, 48 h after exposure to both virulent and attenuated strains of M. nematophilum, CBX102 (Fig 4Aii) and UV336 (Fig 4Aiii) or to M. luteus DMS20030 (Fig 4Aiv), strong GFP expression is induced in the intestine, irrespective of whether these pathogens induce an inflammatory swelling response in the worm rectum. The increased levels of GFP signal were quantified by measuring the fluorescence intensity of the intestinal cell int8 in individual transgenic worms that express the ilys-3 reporter (Fig 4B). Single-cell measurements revealed a 3-fold induction of ilys-3::GFP in response to M. nematophilum CBX102 and UV336.
To quantify the pathogen-induced activation of ilys-3 gene transcription and in non-transgenic animals we used quantitative real-time polymerase chain reaction (qRT-PCR) analysis of whole worm extracts, a method that allows a more reliable quantification of endogenous ilys-3 mRNA than reporter genes. Strong induction of ilys-3 (between 370 and 590x) was observed after animals were challenged for 24h (S4 Fig) and 48h by the three Gram-positive bacteria strains, CBX102, UV336 and M. luteus (Fig 4C).
A clear increase in transcriptional response (above 50x) was also seen for ilys-2 and less markedly for ilys-6 (above 10x, S4 Fig), whereas for ilys-4 and ilys-5 no significant changes were observed. Neither ilys-2,-3 nor-6, were changed in expression levels after exposure to the moderately lethal pathogen P. aeruginosa PAO1 (S4 Fig).
M. nematophilum CBX102 causes a distinctively cell swelling in the tail and infected animals become constipated 24 hours after feeding on this bacterial pathogen. In contrast, worms feeding on the attenuated strains M. nematophilum UV336 or M. luteus DMS20030 appear superficially healthy, but their rate of development is slower compared to E. coli OP50 controls, and they take one extra day to reach adulthood.
We hypothesized that the activation of ilys-3 following bacterial exposure could be caused by a stress response due to nutrient deprivation resulting from low quality food ingested by the worms. To test this we monitored the levels of expression of ilys-3 GFP reporter (GFP driven by the short promoter) in animals subjected to starvation. A 24 hour-starvation regime in young adults (Fig 4Avi) or in arrested L1 larvae (Fig 4Aviii and 4Ax) provoked a significant intestinal induction of the GFP signal (Fig 4D). Nutrient depletion led to a 14.5-fold and a 2.3-fold increase in the intestinal GFP signal detected in adults and larvae, respectively. The addition of live E. coli OP50 (Fig 4Av), [or HB101 (even better quality food, as reported by Shtonda & Avery [44]) restored the low levels of basal GFP expression of ilys-3 in the intestine.
A strong GFP expression of the ilys-3 reporter was also seen in animals fed on diets of heat-killed OP50, or heat-killed CBX102 (S5A and S5C Fig). Heat-killed bacteria are more likely to be harder-to-eat and to provide a bad quality diet. In fact, at 72 hours post-hatching all animals on dead bacterial cells had only reached the L2-L3 transitional larval stage. The response elicited to dead bacteria therefore represents a starvation rather than a pathogen response.
The starvation-induced GFP pattern was specific to the ilys-3 reporter lines as all the other ilys transgenes failed to show induction when worms were starved for 24 hours. In fact the ilys-5 reporter was repressed whereas transgenic animals expressing ilys-2 and ilys-4, CFP or GFP levels remained unchanged. The strong GFP signal present in the intestine of ilys-5 transgenic animals was completely abolished when animals were starved for 24 hours (S6 Fig).
Taken together, our results indicate that ilys-3 activation is responsive to both Gram-positive bacteria and starvation signals. Although a previous microarray analysis identified a rapid up-regulation of ILYS-3 in response to M. nematophilum [22], our real-time PCR analysis revealed a much larger delayed increase of mRNA levels 24 and 48 h after bacterial challenge (590 and 430x, respectively).
To investigate the function of the ilys-3 gene we analyzed the phenotype of the recessive mutation ok3222, which has an 855 bp deletion that removes most of the coding region. On standard bacterial food OP50 the mutants appeared superficially healthy, with no gross abnormalities and with fertility (self-progeny brood size) only slightly reduced relative to wild-type animals (S7 Fig).
If ilys-3 is expressed in the pharyngeal tissue because it is required for efficient lysis then the mutant might be expected to show signs of defective grinder activity and consequently accumulate live bacterial cells in the gut. To address this possibility we compared the kinetics of bacterial accumulation in the gut of mutant and WT animals of the same age. L4 animals were exposed to GFP-expressing E. coli and SYTO 13-labeled CBX102 and visually scored at different times for presence of a green signal in the intestinal lumen. Although intestinal E. coli can be normally detected in old WT animals [45,46], they are rarely seen in the younger animals used in this experiment indicating efficient bacterial lysis. In contrast, we observed that ilys-3 mutants accumulated substantially more un-lysed E. coli::GFP positive cells in their guts relative to WT (Fig 5Aiii–5Aiv and 5Ai and 5Aii). These differences were clear within four hours of feeding on GFP-expressing E. coli. Furthermore, an even higher intestinal bacterial load was observed when ilys-3 mutants were challenged with CBX102 labeled with SYTO 13 (Fig 5Bii and Fig 5Bi). SYTO 13 is a cell-permeant nucleic acid stain that permits visualization of unlysed live bacterial cells. The high green fluorescence of CBX102 cells seen in the lumen of ilys-3 mutants suggests that most bacteria that have passed through the pharyngeal grinder remain viable.
Bacterial survival in the alimentary tract was also measured directly by homogenizing washed whole animals and counting colony-forming units. In one-day old adults CFU numbers of E. coli in ilys-3 mutants were two-fold higher than WT controls [note that these counts include bacteria from both pharyngeal and gut lumen, and therefore underestimate the difference in intestinal burden] (Fig 5C). The intestinal proliferation of CBX102 was also assayed in animals of the same age and we found that ilys-3 mutants had significantly higher bacterial loads than WT (Fig 5F). We estimated that mutant animals had around 3.6-fold more cell counts than WT. Restoration of ILYS-3 in ok3222 mutants using the transgene eEx752 rescued this phenotype and transgenic animals were able to grind bacteria more efficiently (Fig 5Av and 5Avi, Fig 5Biii and Fig 5F). In addition, the accumulation of viable CBX102 cells in ilys-3 transgenic animals carrying the eEx752 array was comparable to that observed for WT (p = 0.4090).
We therefore concluded that ILYS-3 is required in the pharynx for efficient disruption of live bacteria and to reduce bacterial colonization in the intestine of C. elegans.
Standard E. coli OP50 is commonly used as food despite the fact that it can be mildly pathogenic and its accumulation in the intestine is responsible for much variability in the lifespan of C. elegans [47]. Given this, we compared the life expectancy of wild-type and ilys-3 null mutant strains grown on OP50 and M. nematophilum lawns.
Under our experimental conditions, the mean lifespan of wild-type C. elegans propagated from L1 on OP50 was approximately 17 days. However, the deletion mutation in ilys-3 decreased the mean lifespan by 24% to 13 days (p < 0.0001) (Fig 5D and 5E and S2 Table). The reduced life expectancy observed in ilys-3 mutants was restored to that of wild-type in transgenic animals carrying transgenes eEx752 or eEx754. Moreover, both transgenes extended average survival of mutant animals growing on OP50 by 20% (p < 0.0001), probably as a consequence of ILYS-3 overexpression.
Despite their identical average lifespans on pathogen lawns of CBX102 (10 days) N2 animals overexpressing ILYS-3 with eEx754 also exhibited increased adult longevity relative to control animals without this transgene. Compared to N2 the average lifespan of ilys-3 mutants on pathogen lawns decreased by 30% (from 10 days to 7 days, p < 0.0001), (Fig 5G and 5H and S3 Table). Interestingly, supernumerary copies of ilys-3 seemed to have a protective effect mainly during adulthood. Animals carrying transgenes eEx752 and eEx754 showed statistically significant survival curves relative to WT in both ilys-3 and WT backgrounds (S3 Table).
The translational fusion ILYS-3 protein appears to be fully functional and reverses the shortened lifespan of ilys-3(ok3222) when mutants were fed either OP50 or CBX102 diets.
To address the possibility that the reduced lifespan of ilys-3 mutants was a consequence of subtle developmental defects and/or to a reduced fitness of animals, we measured the developmental rate of WT and mutants on live bacterial cells of OP50 and on the pathogen CBX102 (S8 Fig). Synchronized animals were added to each of the diets as L1 larvae and their developmental age was assessed 48 hours later. No significant differences were found between wild-type and mutants fed on live OP50 or CBX102 (p = 0.9 and p = 0.7, respectively, 99% confidence level). On pathogen lawns both WT and mutants exhibited similar developmental progression rates at 48 hours and the majority of the animals were of the same developmental age, at the L2 stage.
We therefore concluded that the pathogen-response differences between WT and mutants that we observed were acquired after larval development rather then early in life. Thus, infection by M. nematophilum might therefore pose a higher risk in mature and elder animals when immunity wanes.
Altogether these results show that ILYS-3 has an important role in the pharyngeal grinder for proper lysis of bacterial cells. The reduced enzymatic activity by this tissue results in increased bacterial colonization in the gut, and reduced lifespan.
Although the major tissues that express ILYS-3 are associated with the digestive tract, we also detected expression in the coelomocytes. These are scavenger cells in C.elegans and have been shown to perform unspecific endocytosis of fluid-phase markers and macromolecules from the body cavity [48]. Although similar cells are believed to have an immune function in many invertebrates, in C. elegans their specific involvement in phagocytosis of whole bacteria seems unlikely and this arm of innate defense is apparently missing. Nevertheless, how did ILYS-3 appear in these cells? One possible scenario would be that ILYS-3::mCherry is secreted by intestinal and pharyngeal tissues and eventually taken up and degraded by the coelomocytes. Proteins with a secretion signal, such as ILYS-3 are translated and processed in the ER and might be secreted into the pseudocoelom before being endocytosed by coelomocytes. However, several lines of evidence indicate that coelomocytes directly produce ILYS-3. First, we observed coelomocyte expression in transgenic animals expressing the long promoter version of the GFP transcriptional reporter, thus suggesting that they are among the primary sites of ilys-3 transcription activity (Fig 1K). Second, we found that absence of functional coelomocytes did not result in an increased level of ILYS-3:mCherry in the body cavity; for this we used the coelomocyte-deficient strain, NP717 (kind gift of H. Fares). NP717 animals have genetically ablated coelomocytes due to expression of a variant of the Diphtheria toxin A fragment (E148D) driven by the coelomocyte-specific unc-122 promoter. These animals also express an ssGFP that is secreted by the body wall muscles and accumulates in the body cavity, demonstrating lack of coelomocytes and scavenging [48]. No accumulation of ILYS-3::mCherry in the pseudocoelom was detected in CB7137 animals (NP717 derivative strain) with ablated coelomocytes relative to their respective control counterparts. Finally, we found that in a rme-1 deletion mutant which has impaired coelomocyte up-take of endocytosis markers, the population of ILYS-3::mCherry positive vesicles was unaffected (S9C Fig). These mutants exhibit an accumulation of the soluble GFP secreted from body-wall muscle cells in their pseudocoelom [42]. However, examination of ILYS-3 expression in this genetic background revealed that compromising endocytic function did not affect the fluorescence intensity, the size or the number of the coelomocytic vesicles. These findings indicate that coelomocytes directly produce ILYS-3.
In the coelomocytes, ILYS-3 is present in presumptive endosomes and lysosomes. This was observed by injecting Alexa-488 BSA into the body cavity of animals expressing ILYS-3::mCherry (S9A Fig). An hour after injection, a population of endocytosed Alexa-488-BSA bearing vesicles co-localized with mCherry positive compartments. These were subsequently confirmed as late endosome/lysosomes vesicles by doubly labeling ILYS-3::mCherry transgenic animals with the endocytic compartment marker GFP::RAB-7, which detects early/late endosomes and lysosomes (S9D–S9F Fig). We also found that in the cup-5 mucolipin-1 mutants (ar465) the population of large vacuoles that correspond to late endosome-lysosome hybrids, were positive for ILYS-3::mCherry (S9B Fig). In these mutants, although lysosomes are able to fuse with late endosomes, the autolysosomes fail to degrade endocytosed material, thus resulting in accumulation of large vacuoles [48,49].
Given the ILYS-3 coelomocyte expression pattern, we reasoned that perhaps these cells might confer some protection against microbial pathogenesis, similar to their mammalian counterparts. To test this hypothesis we examined the effect of removing these cells on the lifespan of animals exposed to M. nematophilum and compared their response to that of wild-type worms. Depleting coelomocytes resulted in a wild-type response to the pathogen and the median lifespans of coelomocyte-ablated NP717 and N2 were indistinguishable (approximately 7 days) (S10 Fig). In contrast, in the coelomocyte-minus ilys-3 double mutants (CB7137, an NP717 derivative strain) an enhanced susceptibility to the bacterial pathogen was observed relative to NP717 and to N2. The overall median lifespan of both NP717 and CB7137 animals decreased from 7 to 5 days, (p < 0.0001), and ilys-3 double mutants were no different from ilys-3 single mutants. However, a greater proportion of double mutant larvae succumbed to the pathogen at early time points, while older animals that managed to escape the initial bacterial burden exhibited similar mortality risk rates to those of the ilys-3 single mutants (S10 Fig). When analyzed the two early survival curves were statistically different (p = 0.004). We concluded that removing coelomocytes per se did not shorten survivorship of wild-type worms and therefore these cells do not seem to provide a protective response to the pathogen M. nematophilum. However, when combined with a defective ilys-3 function, lower survival rates in the younger cohorts were observed. Combined absence of functional ilys-3 and coelomocytes contribute to disease pathogenesis, presumably due to a compound effect of gut luminal colonization and failure to remove/recycle unwanted macromolecules in younger animals. It is therefore likely that these cells play a more important role in hatchlings then in their older siblings and specially in situations where peptidoglycan overload occur.
The activation of the extracellular signal-regulated kinase (ERK) mitogen-activated protein (MAP) kinase cascade is required to mount a protective response against M. nematophilum and consequently loss-of-function mutants of MPK-1, the C. elegans ortholog of the mammalian ERK1/2, are more susceptible to this bacterial pathogen [50,51].
We therefore tested whether the ERK pathway could account for the up-regulation of ilys-3 after M. nematophilum infection. We found that induction of ilys-3 in the intestine of mpk-1(ku1) mutants exposed to M. nematophilum was severely compromised (Fig 6Ai and 6Aii). This was confirmed by measuring the relative fluorescence intensity of ilys-3p::GFP transcriptional reporter in the intestinal cell int8 of WT and mutant animals (Fig 6C), respectively. Inactivation of mpk-1 resulted in a significant reduction or absence of the ilys-3 transcriptional activity in the gut of animals grown on OP50 as well as on CBX102 (Fig 6Ai and 6Aii). In contrast, pharyngeal ilys-3 remained largely unchanged (as shown below).
These results were confirmed by assaying the effect of the MEK inhibitor U0126 on the expression of ilys-3 reporter in the gut of L4 WT animals grown on HB101 or on 10% CBX102 (Fig 6Bi and 6Bii). Exposure of the worms to the MEK inhibitor resulted in a statistically significant decrease of GFP in the intestinal cell int8, when compared to DMSO controls. This effect was seen when worms were grown on E. coli or challenged with CBX102 (Fig 6D). We also confirmed the effect of inhibition of ERK signaling pathway on the reduction of intestinal ilys-3 expression, by feeding L4 animals with mpk-1 dsRNA. Consistent with our mutant analysis, knockdown of this gene also diminished the GFP transcriptional reporter intensity in the intestine but not in the pharynx (S11 Fig). We concluded that intestinal induction of ilys-3 triggered by M. nematophilum infection is dependent on MPK-1 activity.
ERK activation has also been shown to mediate the detrimental effect of starvation in pharyngeal muscle of C. elegans, via the signaling transduction pathway muscarinic acetylcholine receptor-Gqα-PKC-MAPK [52]. We hypothesized that if activation of ilys-3 was also due to a response to starvation, then the loss-of-function gpb-2 mutants that are hypersensitive to muscarinic signaling and starvation should show increased ilys-3 activity in the gut. We found that this was the case. gpb-2 is the ortholog of vertebrate G β5 and encodes the β subunit of a heterotrimeric G protein that binds the Gqα EGL-30, functioning to negatively regulate signaling in the pharyngeal muscle. In these mutants ilys-3 was overexpressed (Fig 6E). GFP was abundant not only in the pharynx (Fig 6Eii), intestine and rectal epithelium (Fig 6Eiii) but also in other tissues such as epidermis (Fig 6Ei). In gpb-2 mutants the muscarinic acetylcholine signal cannot be down-regulated and starvation has detrimental effects in the pharynx [52,53].
We next asked whether hyperactivation of the pharyngeal MAPK pathway could affect the worm responses to CBX102. In assays with 100% lawns of CBX102, gpb-2(ad541) L1 larvae die in 3 days whereas WT had an average lifespan of 10 days. This early death is likely due to unrestrained pharyngeal damage [52], despite the fact that in this mutant background expression of an ilys-3 reporter was markedly induced following M. nematophilum infection (Fig 6F and 6G). We estimated a two-fold increase of the ilys-3p::GFP transcriptional reporter in the intestinal cells of transgenic mutant animals, relative to uninfected siblings (S12 Fig).
Overall, our results revealed a transcriptional program leading to ILYS-3 induction upon M. nematophilum infection, which is mediated at least in part by the ERK-MAPK and the negative regulator GPB-2 (Fig 5H). Loss-of-function in GPB-2 causes MAPK hyperactivation, damage to the pharynx with probably enhanced susceptibility to both starvation and pathogen despite increased ILYS-3 expression. Our results also suggest that ilys-3 is a useful in vivo sensor for monitoring responses to starvation as well as to pathogens.
The C. elegans mpk-1 gene is expressed in the nervous system and in many tissues including the pharyngeal-intestinal valve, intestine, body wall muscles and rectal epithelium ([54] and https://www.wormbase.org). For this reason we asked whether intestinal expression of mpk-1 alone could modulate the induction of the intestinal ilys-3 in the mutant background in response to the pathogen, thus working in a cell autonomous manner. Surprisingly, we found that this was not the case. Overexpression of MPK-1 in intestine under the intestine specific promoter mtl-2 (mtl-2p::MPK-1) did not rescue GFP reporter levels in the gut of mpk-1(ku1) mutant animals (Fig 7A). The levels of fluorescence in the intestinal cells int2 and int8 of single and double transgenes remained largely unaltered (Fig 7B and S13 Fig, respectively).
We also tested whether increased expression of the MAP kinase cascade in the rectal epithelium could lead to increased ilys-3p::GFP signal in the intestine in view of the known importance of ERK signaling in the rectal cells. For this, we used a strain that bears a rectal enhancer element of egl-5 coupled with the pes-10 minimal promoter driving expression of LIN-45(S312A,S453A) (Raf) (RE::LIN-45*), [55]. Although constitutive activation of the swelling response was seen in transgenic animals, as expected, no increase in levels of ilys-3p::GFP were observed in the gut (S14 Fig).
We next tested whether restoring MPK-1 specifically in the pharynx of mpk-1(ku1) mutants had an effect on intestinal induction of ilys-3p::GFP. We used a strain that expressed the wild-type mpk-1 from the pharyngeal specific promoter myo-2, a myosin heavy chain isoform gene expressed specifically in pharyngeal muscles (kind gift of L. Avery). We measured the fluorescence of intestinal ilys-3 in int2 and int8 cells in mpk-1 and wild-type single and double transgenic animals and compared to levels of expression in their single transgenic siblings (Fig 8 and S15 Fig). Pharyngeal expression of MPK-1 completely recapitulated the intestinal expression of ilys-3 pattern generated in the wild-type reporter strain (Fig 8A). Statistically significant high levels of intestinal GFP were detected in mpk-1 double transgenic animals compared to their single transgenic sibling mutants (Fig 8B). Furthermore, and as shown in Fig 8B and S15 Fig, no significant differences were found in the average levels of intestinal ilys-3 fluorescence between mutant and wild-type double transgenes. Thus, presence of MPK-1 in the pharynx resulted in an increased expression of ilys-3p::GFP and also rescued reporter gene induction in the intestine of the mpk-1(ku1) mutant. To corroborate the results described above, we used quantitative real-time PCR analysis to determine ilys-3 transcriptional activity in whole organism in mpk-1 mutants harboring the tissue specific transgenes that rescue MPK-1 activity in the pharynx and in the intestine (S16 Fig and S4 Table). Consistent with our confocal microscopy analysis we were able to observe high ilys-3 mRNA induction levels in the mpk-1 mutant only when the pharyngeal but not the intestinal MPK-1 transgene was present. Particularly high levels of ilys-3 basal transcription activity were detected in animals harboring the pharyngeal MPK-1 cDNA transgene reared on OP50 lawns. This is also in agreement with the microscopic analysis using the ilys-3 GFP reporter.
Overall, these results indicate that for its response to infection by M. nematophilum, intestinal ILYS-3 induction depends on MPK-1 activity in the pharynx. High levels of MPK-1 activity in the pharynx of mpk-1 mutant resulted in generally detrimental effects in most animals, making it impossible to complete pathogenicity assays, as the transgenic line was difficult to maintain. We therefore could not directly test whether activation of pharyngeal MPK-1 was protective against bacterial infections. This deleterious effect is consistent with previous reports [52].
Altogether these results showed that intestinal MPK-1 activity is not required for ilys-3 pathogenic induction in this tissue, whereas pharyngeal MPK-1 activity is.
However, under nutrient depletion, and in contrast to the bacterial-mediated response, inactivation of ERK-MAPK by mutation did not block induction of the ilys-3 reporter in the intestine (S17 Fig); thus indicating that immunity and nutrient depletion can be uncoupled and are presumably, under the control of two distinct regulatory networks.
In view of the evident importance of this invertebrate lysozyme, both in healthy and diseased worms we next wanted to analyze the enzymatic properties of ILYS-3, and prepared a purified recombinant protein. A construct was designed that allowed expression of the full length of ILYS-3 (139 amino acids) fused with E. coli maltose-binding protein (MBP, 367 amino acids) via a 47 amino acid linker. The purity and approximate molecular mass of (r)ILYS-3 were assessed by performing SDS–PAGE on a 10–12% gradient gel (Fig 9). The size of the fusion protein was approximately 58 kDa (including MBP-tag; 42.5 kDa), which was consistent with the predicted size of the fusion protein. Expression of rILYS-3 was accompanied by proteolytic activity resulting in an additional band of 43kDa in size that corresponded to MBP protein fused with the first 5 amino acids of ILYS-3 (Fig 9A).
Next rILYS-3 was tested in vitro for its ability to cleave peptidoglycan in cell walls of the standard bacterial strain, M. luteus. For this, we employed a zymogram assay, an electrophoretic method that is based on an SDS polyacrylamide gel impregnated with bacterial cells as substrate, which becomes hydrolyzed by the protein during the renaturation period. Methylene blue staining of the gel reveals sites of hydrolysis as white bands on an otherwise dark blue background [56]. Hen egg-white lysozyme (Hen Lys) served as a positive control in this assay. As shown in Fig 9B both Hen Lys and rILYs-3 display cell-wall degrading lytic activity, suggesting that ILYS-3 is capable of releasing peptidoglycan fragments in the same way as Hen Lys. A recombinant ILYS-3 fused to GST at the N-terminus, was also made and IPTG-induced bacterial cell extracts were found active on dead cells from both M. luteus (Fig 9B) and M. nematophilum (Fig 9C). In contrast, Hen Lys was only active on M. luteus.
The pH of the C. elegans lumen ranges from 5.96 ± 0.31 in the anterior pharynx to 3.59 ± 0.09 in the posterior intestine [57]. We investigated whether ILYS-3 is active under these conditions. For this we performed zymogram analysis with proteins renatured on gels exposed to buffers at different pH values. We found it maximally active between pH4.5 and 5.0, although a pronounced activity of the rILYS-3 was also seen at pH3.0. This coincides with the luminal pH of C. elegans intestine and the highly acidic digestive organelles lysosomes. The zymograms shown in Fig 9B and 9C depict rILYS-3 activity at pH5.0.
The hydrolytic activity by ILYS-3 is in agreement with its high sequence homology to the bivalve invertebrate lysozymes in Tapes japonica (PDB No 2DQA) and Meretrix lusoria (PDB No. 3AB6), which have muramidase activity [Percentage of pairwise sequence identity with ILYS-3 were of 51%, and 49% with 7.00 × 10−24 and 6.00 × 10−23 blast expectation values, respectively) [(https://www.predictprotein.org/)] (S18 Fig). The tertiary structural model of ILYS-3 shown in S18A Fig was constructed by performing homology-based modeling using the iTasser server. The model matches the crystal structure of the NAG-bound lysozyme from Meretrix lusoria (Chain A; PDB No. 3ab6) that has six α-helices and two β-sheets.
We concluded that the C. elegans ILYS-3 appears to be largely similar to the invertebrate lysozymes from bivalves [29,58]. Given that an intact cell wall is essential for bacterial viability and that the invertebrate lysozyme ILYS-3 has an in vitro lytic activity, we propose that ILYS-3 is likely to be an effector that affects this cell integrity thus reducing bacterial proliferation in the nematode and aiding in the digestion of shed PGN.
Attempts to demonstrate direct bactericidal action of recombinant ILYS-3 on live M. nematophilum cells were not successful, probably due to protein aggregation and perhaps a need for additional digestive factors that are present in the natural situation.
Central to the capacity of intestinal epithelial cells to maintain immune barrier functions is an ability to produce inducible antimicrobial effectors. Our results show that the intestinal cells of C. elegans are capable of producing and inducing a candidate bactericidal effector, ILYS-3 that helps prevent Gram-positive bacterial pathogen colonization, by cleaving and perhaps recycling the load of peptidoglycan fragments which might otherwise overwhelm the whole epithelia. Moreover, we show that pharyngeal ILYS-3 has an important role for lysing both Gram-positive and Gram-negative bacteria.
In multicellular organisms danger inflicted on cells results in alarm signals that activate innate immune responses. Thus, a host mechanism that propagates danger signals to neighboring cells and tissues is advantageous for pathogen clearance and damage repair. In the present study, we show that during M. nematophilum infection, the activation of the intracellular ERK-MAPK signaling cascade in the pharynx results in a signal that leads to ilys-3 transcription in a distal tissue, the intestine. This communication between pharyngeal muscle and intestinal epithelium enables host cells to increase the expression of effector ILYS-3, at a site where bacterial colonization occurs.
M. nematophilum infection represents a suitable model system to study chronic colonization of the intestinal epithelium in C. elegans. This is because in contrast to what is seen with bacterial pathogens that rapidly kill the nematode, M. nematophilum affects the worm fitness and elicits immune reactions as a consequence of its persistent infection in the intestine. Faced with continuous pathogen stimulation, the intestinal epithelium of infected worms thus requires a balance between mounting an exacerbated response or maintaining steady-state levels of antimicrobial expression.
Here we have elucidated the biological function of the C.elegans ilys-3, a member of the invertebrate type of lysozymes. We have found that one of the biological roles of ilys-3 is to aid bacterial grinding and lysis, which are required for nutrient digestion. Firstly, ilys-3 is expressed primarily in digestive tissues of C. elegans: the pharyngeal grinder and the intestine. Furthermore, upon ingestion of non- or mildly-pathogenic bacteria, animals with depleted levels of ILYS-3 showed increased intestinal bacterial cell counts, indicative of enhanced bacterial proliferation. This intestinal bacterial overgrowth is likely to be associated with the acquisition of the aging-related phenotype observed in ilys-3 mutants which exhibited reduced lifespan on both OP50 and CBX102. This is in line with findings in Drosophila [59] and C. elegans [46,60]. In these two models of aging, bacterial overgrowth acts on the organismal lifespan. Manipulations on the microbial dynamics with antibiotics were seen not only to prevent bacterial cell proliferation within their hosts but also to increase fly and worm lifespans.
Several lines of evidence indicate that ILYS-3 works as immune effector in C. elegans. First, we provide biochemical evidence that ILYS-3 has in vitro hydrolytic activity for the Gram-positive bacterial cell walls of M. luteus and M. nematophilum. Peptidoglycan consists of glycan strands of alternating β-1,4-linked N-acetylglucosamine (GlcNAc) and N-acetylmuramic acid (MurNAc) residues cross-linked to each other by short peptides made of L- and D-amino acids. C. elegans ILYS-3 has the conserved Aspartate and Glutamate residues responsible for muramidase activity and is probably capable of generating PGN fragments by hydrolyzing the β(1,4) linkages between N-acetylmuramic acid and N-acetylglucosamine residues in PGN. However, we have not established the exact sites of ILYS-3 cleavage in the complex peptidoglycan polymer. Likewise, we have not demonstrated that purified ILYS-3 can act alone as bactericidal protein. Animal lysozymes have been implicated in peptidoglycan hydrolysis and innate immunity, via the generation of soluble and immunogenic PGN fragments, which themselves act as ligands and can be detected by the host immune system [61,62]. In mice, the cutaneous innate defense response and subsequent NOD2 detection of the extracellular pathogen S. aureus, requires the intracellular delivery of peptidoglycan-derived muramyl dipeptides through the insertion of the pore-forming toxin α-hemolysin into the host membranes [63]. The second domain of catalytic activity predicted in invertebrate lysozymes is isopeptidase. However, our homology searches revealed that of the six invertebrate-type lysozymes present in the C. elegans genome, only ilys-4 has intact isopeptidase putative residues (S95 and H123). All the others appeared to have lost the Serine residue, which has been substituted by Alanine. Therefore they are expected to lack isopeptidase activity.
Second, the prolonged induction of ilys-3 suggests that sustained high levels of this antimicrobial are necessary to provide protection against a persistent pathogen such as M. nematophilum. This suggests that ILYS-3 is a slow and long-lasting effector, which takes longer to attain maximal levels and strengthen its antimicrobial function. Such sustained antimicrobial induction could be designed to entail an efficient protection against more persistent pathogens that resist the first line of effectors and remain resident within the host tissues. As mentioned above, ILYS-3 can also be induced by the Gram-positive S. aureus, which is capable of activating a strong transcriptional response in C. elegans intestinal epithelial cells. Work by Irazoqui showed that ilys-3 is highly responsive to this pathogen and its transcriptional activity started early in time by 4 hours, but continued to increase by their final assayed time point (12 hours) [8].
Intestinal function is reinforced by the luminal secretion of antimicrobials which disrupt essential features of bacterial biology and creates the first line of immune defense. We wished to address the localization of ILYS-3 and to identify the machinery responsible for its transport in the polarized intestinal epithelia and secretion to the gut lumen. Surprisingly, and despite the many candidates for secretion into the intestinal lumen, there is no published direct evidence for a secreted gut enzyme in C. elegans. We have been unable to ascertain the route of the abundant luminal ILYS-3::mCherry secretion detected in old adults and dauers, as our experiments using markers for apical secretion were inconclusive. Although apical levels of ILYS-3::mCherry were usually below thresholds of detection, we found that the fusion protein can be routed to the apical domains of the intestinal cells. This conclusion was based on experiments that reduced the function of basolateral recycling vesicles in the rab-10 genetic background. We also observed that ILYS-3::mCherry can be recycled basolaterally in a rme-1 independent manner which suggests that ILYS-3 may work partly within the recycling endosomes and acidic vesicles of the intestinal epithelium. Digestion of PGN fragments and bacterial fragments may require the coordinated action of lysozymes that work effectively both in apical and basolateral environments. In this scenario, ILYS-3 would help detoxifying and clearing the remained PGN fragments that managed to escape the initial action of lytic enzymes and end up intracellularly and/or in the body cavity. This is in keeping with the observations that ILYS-3 is expressed in the recycling endosomes and the acidic vesicles LROs of the intestine but also in the late endosomes and lysosomes in the coelomocytes.
The expression patterns we observed were obtained using protein expressed from multicopy arrays, and therefore potentially affected by overexpression artifacts. However, the transgenes used efficiently rescued all mutant phenotypes and did not exhibit obvious dominant effects. Moreover, the rapid and reversible changes in distribution in response to nutrient availability indicate that the reporter reliably reflect endogenous protein distribution.
Third, disruption of ilys-3 renders mutant animals more susceptible to the bacterial pathogen M. nematophilum. Fourth, overexpression of ILYS-3 leads to an enhanced protection against M. nematophilum in both ilys-3 mutants and wild-type backgrounds.
It is intriguing that ILYS-3 is detected in coelomocytes, the six scavenger cells present in the adult hermaphrodite that continuously and nonspecifically recycle and endocytose macromolecules from the pseudocoelom [48]. These cells are believed to perform detoxification, acting like hepatic cells in higher animals [64]. In addition, they have been shown to work as sensors for dietary restriction-mediated longevity in C. elegans [65]. Our results suggest that coelomocytes are ILYS-3-producing cells but we ruled out the possibility that they contribute in any major way to the defense response of the C. elegans wild-type to M. nematophilum. Nevertheless, once the combined functions of ilys-3 and the coelomocytes were compromised, a small effect was observed. The effect could be due to the excessive tissue exposure to PGNs resulting from lack of active processing.
Actively growing Gram-positive bacteria release large amounts of peptidoglycan fragments as a consequence of cell remodeling during cell division. Work done in Bacillus megaterium and B. subtilis established that 30–50% of PGN can be shed into the growth medium in each bacterial generation during exponential growth [66]. The intestinal environment of C. elegans can not be reproduced in a liquid-based medium, and consequently bacterial growth may differ in these two situations. However, bacteria that reside and proliferate in the nematode intestine, like M. nematophilum, can be expected to shed PGN fragments as part of normal growth, which may represent a significant burden on the intestinal cell function.
We found that in the intestine ilys-3 has a post-embryonic regulated expression pattern, suggesting that it might be important for maintaining nutrient levels or immune homeostasis during certain developmental transitions. In conventionally reared animals, high levels of ilys-3 transcriptional activity were detected in the intestinal cells of L1 at hatching, but these declined as animals reached L2. In early adults however, levels of intestinal ilys-3 increased and were sustained thereafter. The physiological relevance of such transient expression pattern where steady state levels of ilys-3 are kept high in hatchlings and adults may reflect the distinct needs to provide protection against bacterial colonization during key post-developmental stages. In the early phase of post-embryonic life, the intestine of naïve L1 may need to be primed with effectors that ensure safe first contact with food, in adulthood, the hermaphrodite intestine may become more vulnerable as its resources become directed towards nurturing the germline with yolk proteins.
With the onset of larval diapause, dauers might also need to be primed with effectors and signaling molecules that help them withstand harsh conditions and/or prepare them to get ready for feeding. We showed that upon nutrient depletion, luminal secretion of ILYS-3::mCherry was both very dynamic and reversible in these larvae. The fusion protein was secreted and seen to accumulate in the intestinal lumen, only to return rapidly to its cytosolic default state when animals encountered food and resumed development. A similar changing pattern of secretory polarity has also been reported for INS-35 and INS-7 [67], two insulin-like peptides that suppress larval diapause. These peptides, are usually secreted into the pseudocoelom in larval stages during normal development, but during diapause they are routed to the intestinal lumen where they are degraded. However, we do not have any evidence for the luminal degradation of ILYS-3:mCherry. In fact, in 25-day-old larvae the signal of fusion protein did not decrease and only changed as larvae adapted to resume development upon feeding. Unlike in starvation-induced dauers, food deprivation during other larval stages did not provoke accumulation of ILYS-3 in the lumen. These changing patterns of ILYS-3 distribution reveal yet another facet of that adaptive changes that occur in dauer development and demonstrate remarkable plasticity in the intestinal cells.
In C. elegans, most intestinal immune effectors that protect against many pathogens appear to act locally and respond cell autonomously to the activation afforded by key regulators such as the p38 MAP kinase PMK-1 [68] [13,28,69,70], DAF-16 [71] and ZIP-2 [72]. Interestingly, in the S. aureus model of infection intestinal ilys-3 induction requires the proline hydroxylase EGL-9 which under normal O2 levels, functions in a conserved hypoxia-sensing pathway to convert the hypoxia-inducible factor, HIF-1 [73].
In contrast, our results indicate that for the M. nematophilum-induced ilys-3 transcription an ERK/MAPK activity is required just as the starvation-mediated response in the pharyngeal muscle of C. elegans does [52]. Intense induction of ilys-3 was found in gpb-2 mutant animals in which the ERK/MAP kinase signaling could not be down-regulated. However, in contrast to starvation-effects on pharyngeal muscles, the starvation/pathogen induction of ilys-3 occurs in the intestine. The transcriptional programs that govern pathogen- and starvation are likely to be distinct despite the fact that they share at least one downstream molecular target, ilys-3.
We discovered a cell non-autonomous regulation of the ilys-3 transcriptional response following M. nematophilum infection. ERK expression in the pharynx alone was sufficient to restore intestinal ilys-3 induction in the loss-of-function mutant mpk-1; while the activation of ERK MAP kinase in the intestine was not able to activate ilys-3 transcription in this tissue. How can ERK signaling in the pharynx trigger ilys-3 transcription in the gut? The ability of multicellular organisms to mount an efficient innate response against pathogen infections is likely to be aided by systemic communication between individual cells and tissues. In the intestine of C. elegans cell non-autonomous effects have been suggested in many contexts, ranging from the regulation of organismal response to oxidative and heat stress and lifespan [74–76] to the modulation of immune response against the bacterial pathogen P. aeruginosa [77]. In most of these examples however, the signaling between tissues requires a neuronal component. However, it has been shown that upon DNA damage, germ cells are capable of eliciting a systemic pathogen and stress resistance response in the somatic tissues of C. elegans [78]. Such induction depends on the germline specific activity of the ERK/MAK kinase mpk-1, which evokes the transcriptional activation of innate immune genes, similarly to the local response evoked by p38 MAPK upon intestinal infection by pathogens [68].
The results shown here suggest a communication between two non-neuronal distal tissues. This is reminiscent of a mammalian epithelial cell-to-cell communication strategy that propagates NF-kB and ERK- dependent pro-inflammatory signals from infected cells to bystander cells following infections by Shigella flexneri and Listeria monocytogenes [79,80]. However, in our context the signaling pathway that activates ilys-3 in the pharynx is not required for its action in the other tissue.
In conclusion, our data suggest a model whereby upon M. nematophilum stimulation, the pharyngeal muscles activate the MAPK pathway leading to the transmission of an alarm signal to the intestinal epithelium. This results in the activation of an innate immune response, with transcription of the antimicrobial ilys-3. Whether the pharyngeal signal induces other intestinal responses, or acts on other distal tissues, remains to be seen. We also cannot exclude more complex models, in which the primary pathogen detection occurs in the intestine or elsewhere, and the danger signal is then amplified by the pharyngeal muscle. However, the pharynx is well placed to act as a site for primary danger detection.
Our work demonstrates the importance of a specific lysozyme gene both to normal nutrition and to defense. The unexpected complexity of its expression, both in different tissues and subcellular compartments, suggest that it may have functions within cellular vesicles, as well as in the digestive tract.
Our work also reveals the existence of a new tissue-tissue communication acting at the organismal level and reveals a strategy whereby the pharynx activates an antimicrobial intestinal defense by propagating bacterial/danger signals.
Maintenance and manipulation of C. elegans strains were performed as previously described [81]. Unless otherwise specified, animals were cultivated at 20°C and fed with Escherichia coli OP50 as food. Wild-type is the C. elegans variety Bristol strain (N2). Mutant alleles were provided by the Caenorhabditis Genetics Center at the University of Minnesota, which is supported by the National Institutes of Health–Office of Research Infrastructure Programs (P40 OD010440).
The following genes and alleles were used in this work:
LGI: VC1026 [rab-10(ok1494)], GS2643 [arIs37 [myo-3p::ssGFP + dpy-20 (+) I; cup-5(ar465) III; dpy-20(e1282) IV]. DA541 [gpb-2(ad541)]. LGIII: CB6148 [mpk-1(ku1)], DP38 [unc-119(ed3)]. LG IV: VC2493 [ilys-3(ok3222)], obtained from the C. elegans Gene Knockout Consortium, outcrossed three times and sequence verified prior to phenotypic characterization. LG V: DH1201 [rme-1(b1045)]. Unknown LG: CB6603 eIs102[egl-5p::GFP::LIN-45*], RT311 {unc-119(ed3); pwIs69[vha-6p::GFP::RAB-11 + Cbr-unc-119(+)]}, RT476 {unc-119(ed3); pwIs170[vha-6p::GFP::RAB-7 + Cbr-unc-119(+)]}, NP871 {unc-119(ed3) III; cdIs66[pcc1::GFP::RAB-7 + myo-2p::GFP + unc-119(+)]}. The strain DA2200 carrying the integrated adIs2200 [myo-2p::MPK-1::GFP] was a gift from Leon Avery, University of Texas Southwestern Medical Center, Dallas. NP717 strain carrying the unc-119(ed3); arls37(myo-3p::ssGFP); cdls32(pcc1::DT-A(E148D) + unc-119(+) myo-2p::GFP was a gift of H. Fares (University of Arizona, Tucson, Arizona). New strains constructed specifically for this study are listed in S5 Table.
E. coli OP50, E. coli HB101 (DE3) and E. coli::GFP (kind gift of Jonathan Ewbank), M. nematophilum CBX102 and UV336; M. luteus DMS20030; P. aeruginosa PAO1. Bacterial cells were grown at 37°C in LB and lawns prepared from exponential phase growth cultures.
Reporter constructs were generated using a PCR fusion protocol [82]. Except otherwise stated transgenic lines were obtained from injections of unc-119(ed3) mutants with the PCR products obtained from co-amplification of the promoters and the fluorescent proteins. The plasmid unc-119 (+) (pDP♯MM016) was used as a genetic marker for the transgenes [83].
For ilys-3p::GFP transcriptional fusion, the ilys-3 (C45G7.3) genomic sequence containing the 1.05 Kb 5′-upstream of the first predicted methionine was fused to the green fluorescent protein (GFP) and the unc-54 3′-untranslated sequence from the vector pPD95.75 (A. Fire). This amplicon was cloned into the pJET1.2 vector to give pMGN26. eEx650 and eEx651 are transgenic lines obtained from the injections in unc-119(ed3). CB7163 contains eIs120, a gamma-integrated array derived from eEx650. This strain was outcrossed three times to the wild-type N2 before use. For the ilys-1 reporter, the 1.7 Kb promoter of ilys-1 (C45G7.1) was fused to DsRed2 from pMGN7 [18]. The PCR product was cloned into pJET1.2. Subsequent injections into unc-119(ed3) generated transgenic lines eEx652 and eEx653. Transgene eEx655 was obtained from the amplicon encompassing the 2.4 Kb upstream methionine sequence of ilys-2 (C45G7.2) fused to CFP from pPD133.51. For the ilys-4p::GFP transcriptional reporter the 3.3 Kb upstream sequence of ilys-4 (C55F2.2) was fused to GFP from vector pPD95.75. eEx670 is the transgene obtained. For the ilys-5 transcriptional reporter the 5.8 Kb sequence upstream methionine of ilys-5 (F22A3.6) was fused to GFP from pPD95.75. The amplicon was cloned into pJet1.2. Transgenes obtained from unc-119 injections were: eEx671 and eEx672.
To generate the translational fusion of ilys-3p::ILYS-3::mCherry genomic DNA fragment encompassing the full length ILYS-3 (4.5 Kb promoter and 925 bp ORF) was fused at the C-terminal with the mCherry amplicon from pFP10. The resulting PCR was cloned into Topo XL vector (Invitrogen) to give pMGN47 and injected into CB7029 [ilys-3(ok3222); unc-119(ed3)] or into CB7007 [ilys-3(ok3222)] adult worms at a concentration of 20 ng/μl. For the injections into ilys-3(ok3222) the co-injection marker sur-5::GFP (pTG96) was used at 60 ng/μl. Transgenic lines eEx752, eEx753 and eEx754 contain extrachromosomal versions of pMGN47.
To construct N-terminal GFP::ILYS-3 transgene driven by its own promoter the GFP vector pPD117.01 was used (Andy Fire). For this the 4.5 Kb ilys-3 promoter was PCR amplified with oligonucleotides containing XbaI and NotI specific sites. The resulting amplicon was subcloned into pJet1.2 to generate pMGN50. The ilys-3 promoter was then restriction digested and cloned into the XbaI NotI sites of pPD117.01 to generate the plasmid 4.5 Kb ilys-3p::GFP, pMGN52. The entire coding sequence of ILYS-3 containing its own 3'UTR was obtained from N2 genomic DNA and amplified with oligonucleotides containing NheI and ApaI specific sites. The amplicon was subcloned into pJet1.2 to give pMGN51, and then excised and directionally cloned into the NheI and ApaI sites, downstream of GFP, into plasmid pMGN52. To establish transgenic lines, 20 ng of the resulting plasmid, pMGN53 was microinjected into the unc-119 mutant along with the unc-119 co-injection marker. Transgenic line eEx779 contains extrachromosomal version of pMGN53 to give CB7210.
Construct pMGN37 was generated by fusion of the 598 bp sequence of the intestinal specific promoter mtl-2 to mpk-1 cDNA from plasmid pHN3e [11], The resulting amplicon was cloned into pJet1.2. The mtl-2 promoter was also fused to mCherry to give pMGN39. These two plasmids were co-injected at 5 ng/μl into unc-119(ed3) and the transgenic line eEx727 was crossed into mpk-1(ku1) to give CB7004. Complete plasmid sequences are available upon request.
For heat-killing of OP50 and CBX102, fresh overnight cultures were concentrated 10-fold and incubated at 100°C for 60 min. Following heat treatment, cells were plated to confirm inviability after overnight incubation on LB plates. For the assays,100 μl of dead bacterial cell suspensions were plated on NGM plates and approximately 50 L1 larvae from the test strains WT, ilys-3, or ilys-3p::GFP reporter were placed on each plate. Animals were analyzed from triplicate plates and experiments were repeated twice each with freshly prepared bacterial cells.
The fluid-phase marker BSA::Alexa 488 (Thermo Fisher Scientific), was injected at 1 mg/ml into the pseudocoelomic space in the pharyngeal region of adult worms as described by Zhang et al [84]. Injected worms were transferred to OP50 seeded plates at 20°C, and observed at different time points. The intracellular trafficking of the dye was stopped by moving the plates to ice. Injected animals were subsequently mounted on agarose pads and observed on the confocal microscope.
qRT-PCR was performed in synchronized populations obtained by the alkaline bleach method using gravid hermaphrodites and letting the eggs hatch overnight in M9 on NGM peptone-depleted plates. Synchronized naïve L1 larvae were transferred to bacterial lawns made of 100% OP50, M. nematophilum CBX102, UV336 or M. luteus DMS20030 and harvested 24 or 48 hours later. In each case, total RNA from triplicate samples were extracted using Trizol (Thermo Fisher Scientific). RNA concentration and quality was assessed with a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific). 1μg of total RNA from infected and non-infected worms was used for reverse transcription (SuperScript VILO, Thermo Fisher Scientific), and subjected to RNase H (NEB) treatment subsequently. mRNA levels were assessed using a duplexing qPCR with the TaqMan Gene expression assay (Applied Biosystems). This assay relies on two fluorescently labeled TaqMan probes: the two reporter dyes FAM and VIC were used for the detection of the gene of interest and endogenous control, respectively. Using the two probes the method allows the simultaneous amplification and quantification of two target sequences in the same sample. Fluorescent reporters were detected on the real-time qPCR platform StepOnePlus (Applied Biosystems). Results were normalized to E. coli OP50 and to the endogenous control gene rla-1, which did not vary under the conditions being tested. Expression levels were assayed using the 2-ΔΔCt method [85]. Means and standard error of the mean were calculated from at least 3 independent biological replicates. Statistical analysis was done using Holm-Sidak's multiple comparisons test. Primer sequences are available upon request.
Unless otherwise stated infections with M. nematophilum and M. luteus were carried out by adding synchronized L1 larvae to 100% bacterial lawns followed by incubation at 25°C. Surviving worms were counted every day.
To analyze bacterial accumulation in the intestine, wild-type, ilys-3(ok3222) and ilys-3; eEx752 animals were synchronized and allowed to grow to L4 on OP50 seeded plates. L4 were then transferred to NGM plates seeded with E. coli::GFP or CBX102 and grown at 20°C for 24 hours. For the E. coli::GFP bacterial counts, pools of 10 worms each were surface sterilized with 100 μg/ml gentamicin in PBS + 0.1% Triton X-100 (PBST) and subsequently washed in 10 μl drops of ice-cold 25mM tetramisole/M9 (0.1% Triton X-100). Manual disruption of the worms was carried out in PBST in a sterile 1.5 ml microcentrifuge tube using a pestle and glass beads. Triplicates of worm lysates corresponding to 10 animals per group were serially diluted in PBS and incubated overnight on LB ampicillin plates to determine the number of colony forming units (CFU). For CBX102 bacterial counts, pools of 10 worms each were sterilized with 50 μg/ml nalidixic acid in PBST, serially diluted, and bacterial suspensions plated on LB agar supplemented with nalidixic acid. Cultures were incubated at 37°C and colonies counted after 48 hours.
Bacterial cells of CBX102 were labeled with SYTO 13 (ThermoFisher Scientific), at a final concentration of 15 μM in 10mM Tris-HCl, 5 mM NaCl, 1 mM EDTA, pH7.5. Briefly, fresh overnight cultures were concentrated 5-fold and bacteria were stained with SYTO 13 for 45 min at room temperature. Aliquots (100 μl) of stained bacterial suspensions were added to NGM plates, approximately 50 one-day old adults from the test strains WT, ilys-3, or ilys-3; eEx752 were placed on each plate and animals were allowed to fed for 2 hours.
Before being assayed, worms were fed for two generations on HB101 and synchronized twice by bleaching. The resulting eggs were left overnight to hatch at 20°C in M9 on NGM peptone-depleted pates. For lifespan analysis, Day 0 was the day animals were exposed to bleach for synchronization. Assays were conducted at 20°C or 25°C. Approximately 100 animals were tested in each experiment (25 animals/ NGM plate without FUdR). During the first few days, worms were transferred to fresh plates daily. Animals were scored every other day, as alive or dead by gentle prodding with an eyelash. For ilys-3; eEx754 and +; eEx754, only GFP positive animals were analyzed. Worms that crawled off the plate were censored. Statistical analysis was performed using the software Prism 6 (GraphPad) and survival data were compared using the log-rank test.
Synchronized eEx650 L1 transgenic animals were left on E. coli HB101 or 10% M. nematophilum CBX102:HB101 for 48 hours and allowed to develop to late L4. These were then transferred to HB101 or CBX102:HB101 (at 1:10 ratio) NGM seeded plates containing either 50 μM U0126 (Sigma) or DMSO (as a vehicle control). Plates were incubated for 48 hours at 25°C and fluorescence intensity in the intestinal cell int8 was measured using confocal microscopy.
For the microsphere feeding assay, adults and dauers carrying the transgene eEx752 ILYS-3::mCherry reporter construct were transferred to NGM plates containing overnight lawns of OP50 mixed with Fluoresbrites yellow-green microspheres, 0.5 μm (Polyscience, Inc.) at a 1:1000 ratio (v/v) of beads to bacteria. Animals were allowed to feed for at least 45 min or longer, anesthetized in a drop of 100 mM tetramisole and mounted on agarose pad for imaging.
After bleaching, synchronized L1 worms carrying the ilys-3p::GFP reporter were transferred to peptone-depleted NGM plates and let starve for a specific amount of time (typically between 24–72 hours from transferal). For the L4 nutrient depletion assays, well-fed late L4 animals were washed five times in M9 to exhaust any residual bacterial cells, and subsequently transferred to unseeded peptone-depleted NGM plates where they remained till they were collected for light microscopy analysis. The fluorescence intensity of intestinal GPF of nutrient depleted animals was then compared to that of chronologically identical animals reared under normal and well fed conditions. Mean fluorescence value was subtracted from the background fluorescence for each animal and statistical analysis was performed using a Mann Whitney test, 95% confidence interval relative to controls.
The lysosome marker LysoTracker Green (ThermoFisher Scientific) was used at 1 μM concentration and 200 μl of green dye were added to 9 cm OP50 seeded-NGM plate and left to stabilize for 24 h. Animals were allowed to feed on labeled E. coli for 2 days and imaged as young adults.
Quantification of the ilys-3p::GFP fluorescence in the intestinal cells int2 and int8 of reporter transgenes was carried out using a Leica TCS-SP5 Laser Scanning Confocal microscope using a x63 oil immersion lens and the Argon 488 laser. The focal plane with the highest GFP signal was used to measure fluorescence intensity within a ROI set to 0.4 μ thickness, and a 10 μ or 40 μ diameter, for L1 larvae or adults, respectively. To make comparisons across samples, data are presented in boxes plots which define interquartile range (25% of the data above or below the median), bars represent expression range, and the thick line is the median. Identical exposure settings were used for all genotypes. GFP fluorescence was limited to 495/512 nm to diminish background autofluorescence from the animals. For each experiment, and on the same day, 10–15 animals/ treatment were imaged. GFP/mCherry colocalization assays were performed on adult worms expressing the fluorescent markers as previously described. The Pearson's coefficients were calculated using Leica LAS AF software. This correlation coefficient has been used to calculate the percentage of pixels that colocalize from each channel. For imaging ILYS-3::mCherry worms were examined under the HeNe 543 nm laser line.
Transgenic animals expressing the ilys-3p::GFP reporter were used in the RNAi feeding assay performed as described [86] using bacteria from the Ahringer RNAi library (Source BioScience) grown overnight in Luria broth supplemented with 100 μM ampicillin and 25 μM tetracycline, and 100 μl of the culture was added to NGM plates supplemented with 1 mM isopropyl β-D-1-thiogalactopyranoside, 30 μM carbenecillin. Five L4 animals were transferred to the RNAi lawn and allowed to self-fertilize. The F1 progeny was added to OP50 or CBX102 lawns and phenotypes analyzed. RNAi experiments were replicated twice, each with freshly prepared medium. The empty vector L4440 was used as control RNAi.
A construct was engineered that allowed expression of the full length of ILYS-3 (139 amino acids) fused with E. coli maltose-binding protein (MBP, 367 amino acids) via a 47 amino acid-long linker. The maltose-binding-protein (MBP) fusion system is reported to enhance solubility and proper folding of recombinant proteins. The recombinant pMAL-c2X::ILYS-3 protein coding sequences were verified by DNA sequencing. The resulting MBP-rILYS-3 fusion protein (58 kDa and MBP alone 43 kDa) was overexpressed in E. coli BL21 (DE3) and induced with 1 mM IPTG for 3 hours at 37°C and the cells were harvested by centrifugation at 5000 g for 20 min and disrupted by a French pressure cell. The homogenates were cleared at 20,000 g for 25 min and affinity purified using an amylose resin column and washed with binding buffer following manufacturer's instructions (NEB).
A second (r)ILYS-3 was also prepared with a DNA fragment containing ILYS-3 coding region excluding the first 19 aa signal sequence, and containing BamHI and XhoI restriction sites. The fragment was cloned into the expression vector pGEX-6T1 (GE Healthcare) which contains a N-terminal GST affinity tag. The resulting plasmid pMGN73 (GST::ILYS-3 signal less) was verified by sequencing. E. coli BL21 cells were transformed with pMGN73 and grown in LB containing ampicillin and 1% glucose and incubated at 30°C, until cell density reached OD600 0.6. To induce the expression of GST::ILYS-3 fusion protein, 0.2 mM IPTG was added and incubation was continued by shaking for an additional 4 h at 23°C. Recombinant GST::ILYS-3 signal less expressed as inclusion bodies and these were solubilized in 8M urea. After solubilization, the fusion protein was refolded by stepwise dialysis for 48 h at 4°C against 50 mM Tris, pH 8.0, 10% glycerol, 50 mM NaCl, 10 mM glutathione reduced, 1 mM glutathione oxidized to drop urea to 0 M concentration. The refolded GST::ILYS-3 signal less fusion protein was subsequently dialyzed against 25 mM Tris, pH 8.0, 10% glycerol, 50 mM NaCl to remove glutathione and purified using a GST Spin Trap column pre-equilibrated with binding buffer (GE Healthcare) following manufacture's instructions.
The purified recombinant proteins were analyzed by performing a zymogram assay based on an SDS–PAGE and subsequently stained with Coomassie Blue. Two mg/ml of lyophilized M. luteus or M. nematophilum cells, were resuspended and incorporated into a 10% SDS-PAGE gels as in [56]. 15 μl samples were loaded into each well. After electrophoresis gels were washed three times in dH2O for 30 min and then renatured for 16 hours in 250 mM Tris pH (ranged from 3.0–7.0), 20 mM glycine and 100mM NaCl at 37°C, under constant agitation. Next day, gels were stained in 1% methylene blue for 1 hour, and destained in dH2O until clear bands emerged. Methylene blue staining of the gel reveals sites of hydrolysis as white bands on an otherwise dark blue background.
Animals were anesthetized with 10 mM tetramisole/M9 buffer mounted on 2% (w/v) agarose pads. Images were captured by differential interference contrast (DIC) and epifluorescence microscopy on a Zeiss Axioplan 2 fluorescence microscope using a Zeiss AxioCam camera, or by confocal microscopy using a Leica SP5. Images captured using an AxioCam camera and AxioVision software (Carl Zeiss).
Statistical analysis and graphing was done on Prism 6 (GraphPad Software, San Diego, CA).
|
10.1371/journal.pgen.1004733 | A Complex Genetic Switch Involving Overlapping Divergent Promoters and DNA Looping Regulates Expression of Conjugation Genes of a Gram-positive Plasmid | Plasmid conjugation plays a significant role in the dissemination of antibiotic resistance and pathogenicity determinants. Understanding how conjugation is regulated is important to gain insights into these features. Little is known about regulation of conjugation systems present on plasmids from Gram-positive bacteria. pLS20 is a native conjugative plasmid from the Gram-positive bacterium Bacillus subtilis. Recently the key players that repress and activate pLS20 conjugation have been identified. Here we studied in detail the molecular mechanism regulating the pLS20 conjugation genes using both in vivo and in vitro approaches. Our results show that conjugation is subject to the control of a complex genetic switch where at least three levels of regulation are integrated. The first of the three layers involves overlapping divergent promoters of different strengths regulating expression of the conjugation genes and the key transcriptional regulator RcoLS20. The second layer involves a triple function of RcoLS20 being a repressor of the main conjugation promoter and an activator and repressor of its own promoter at low and high concentrations, respectively. The third level of regulation concerns formation of a DNA loop mediated by simultaneous binding of tetrameric RcoLS20 to two operators, one of which overlaps with the divergent promoters. The combination of these three layers of regulation in the same switch allows the main conjugation promoter to be tightly repressed during conditions unfavorable to conjugation while maintaining the sensitivity to accurately switch on the conjugation genes when appropriate conditions occur. The implications of the regulatory switch and comparison with other genetic switches involving DNA looping are discussed.
| Plasmids are extrachromosomal, autonomously replicating units that are harbored by many bacteria. Many plasmids encode transfer function allowing them to be transferred into plasmid-free bacteria by a process named conjugation. Since many of them also carry antibiotic resistance genes, plasmid-mediated conjugation is a major mechanism in the dissemination of antibiotic resistance. In depth knowledge on the regulation of conjugation genes is a prerequisite to design measures interfering with the spread of antibiotic resistance. pLS20 is a conjugative plasmid of the soil bacterium Bacillus subtilis, which is also a gut commensal in animals and humans. Here we describe in detail the molecular mechanism by which the key transcriptional regulator tightly represses the conjugation genes during conditions unfavorable to conjugation without compromising the ability to switch on accurately the conjugation genes when appropriate. We found that conjugation is subject to the control of a unique genetic switch where at least three levels of regulation are integrated. The first level involves overlapping divergent promoters of different strengths. The second layer involves a triple function of the transcriptional regulator. And the third level of regulation concerns formation of a DNA loop mediated by the transcriptional regulator.
| Bacteria exchange genetic material at high rates by different processes, which are collectively named Horizontal Gene Transfer (HGT). HGT can be beneficial for bacteria because the newly acquired DNA may endow them with novel features enabling them to adapt to changing conditions in the environment, i.e. rapid evolution. On the other hand, HGT is notorious for its role in the dissemination of virulence/pathogenicity determinants and antibiotic resistance. The main mechanisms responsible for HGT are transformation mediated by natural competence, transduction and conjugation [1]–[6]. The latter mechanism, -conjugation-, concerns the transfer of a DNA element from a donor to a recipient cell. Conjugative elements containing all the information required for DNA transfer of a donor to a recipient cell are often found on plasmids, but they can also be embedded within a bacterial chromosome. These latter forms are generally named integrative and conjugative elements (ICE).
Some basic features of the conjugation process are conserved among plasmids [for review see], [ 7]–[10]. In most cases, a single-stranded DNA (ssDNA), which is generated by a rolling circle-like mode of DNA replication, is transferred into the recipient cell through a membrane-associated intercellular mating channel, named transferosome, which is a form of type IV secretion system. Conjugative plasmids can be exploited for the construction of tools to genetically modify bacteria of clinical or industrial relevance that are reluctant to genetic manipulation by other ways. Besides its intrinsic scientific interest, a detailed understanding about how conjugation genes are regulated is crucial to design strategies helping to interfere with the rapid spread of antibiotic resistance, and for the construction of genetic tools based upon conjugative plasmids.
Various conjugative plasmids have been studied in considerable detail [for review see], [ 7]–[10]. Although most of the well-studied conjugative plasmids replicate in Gram-negative bacteria, an increasing interest in conjugative plasmids of Gram-positive bacteria has resulted in the recent analysis of conjugative plasmids from for instance streptococci, enterococci, staphylococci and clostridia [11]–[17]. However, conjugation systems present on the Gram-positive soil bacterium Bacillus subtilis had not been reported until recently. This is most remarkable taking into account that (i) it is one of the best-studied Gram-positive bacteria; (ii) it has important industrial applications; and (iii) it is closely related to pathogenic and fastidious bacilli [for review see], [ 18,19]. Moreover, several B. subtilis strains are gut commensals in animals including humans [20]. B. subtilis plasmids may therefore play an important role in HGT in different environments. We chose the B. subtilis plasmid pLS20 for our studies. Originally, this 65 kb plasmid was identified in the Bacillus subtilis natto strain IFO3335 that is used in the fermentation of soybeans to produce “natto”, a dish that is popular in South Asia [21]. Previous studies on pLS20 have shown that it is conjugative in liquid media as well as on solid media [22], [23]. The presence of pLS20 has a broad impact on the physiology of the host, and the localization of some components of the conjugation machinery has been determined [24], [25]. The replication region of pLS20 has been characterized, and it has been demonstrated that it uses a dedicated segregation mechanism involving the actin-like Alp7A protein [26], [27]. pLS20 encodes a protein, RokLS20, that suppresses the development of natural competence of B. subtilis [28].
Recently, we have reported a global view of the regulatory circuitry of the pLS20 conjugation genes. A conjugation operon encompassing more than 40 genes is located next to a divergently oriented single gene, rcoLS20, which encodes the master regulator of conjugation responsible for keeping conjugation in the default “OFF” state. Activation of conjugation requires an anti-repressor, RapLS20, that belongs to the family of Rap proteins. Inactivation of the rapLS20 gene on pLS20 severely compromises conjugation, and conjugation was enhanced when rapLS20 was expressed from an ectopic locus. The activity of RapLS20, in turn, is regulated by a signaling peptide, Phr*LS20. The small phrLS20 gene, located immediately downstream of rapLS20, encodes a pre-protein. After being secreted, PhrLS20 can be processed by a second proteolytic cleavage, resulting in generation of the functional pentapeptide, Phr*LS20, corresponding to the five C-terminal residues of PhrLS20. When (re)imported, this peptide inactivates RapLS20. Therefore, activation of conjugation is ultimately regulated by the Phr*LS20 signaling peptide. The Phr*LS20 concentration will be relatively high or low when donor cells are predominantly surrounded by donor or recipient cells, respectively. Hence, conjugation will become activated particularly under conditions in which recipient cells are potentially present. In addition, Phr*LS20 has a crucial role in returning conjugation to the default “OFF” state [29].
Despite identification of the players involved in regulation of the conjugation genes, our knowledge on regulation of the genetic switch responsible for activating conjugation is still very limited. Using a combination of various in vitro and in vivo approaches, we show that the genetic switch controlling pLS20 conjugation involves at least three layers of regulation. Together, they tightly repress the main conjugation promoter under conditions that do not favor conjugation, while maintaining the ability to accurately switch on the conjugation genes when appropriate conditions occur. The three layers involve coinciding or overlapping divergent promoters of different strengths, autoregulated expression of RcoLS20, which turns out to be a tri-functional transcriptional regulator, and formation of RcoLS20-mediated DNA looping. The sophisticated regulatory mechanism that combines three layers of control into a single switch is novel for plasmids of Gram-positive bacteria. The implications of the uncovered regulation mechanisms for conjugation are discussed in the context of regulatory systems present on other HGT elements and with other regulatory systems involving DNA looping.
Conjugation is a complex and energy consuming process, involving the generation and transfer of ssDNA, synthesis and assembly of a sophisticated type IV secretion system, and establishment of specific contacts with the recipient cell. Hence, the process of conjugation and expression of the genes involved are strictly controlled. Analysis of the regulation of conjugation genes present on ICEs in bacteria and those on plasmids of Gram-negative bacteria indeed indicates that this is the case [for review see], [ 5,7]. In our previous studies, we have sequenced and annotated plasmid pLS20cat of the Gram-positive bacterium B. subtilis and identified a large conjugation operon. We have also identified rcoLS20 as the gene encoding the master regulator of conjugation, RapLS20 as the anti-repressor required to activate the conjugation genes, and we showed that the activity of RapLS20 is in turn regulated by the signaling peptide Phr*LS20. In this study, we analyzed the underlying molecular mechanism of how the pLS20 conjugation genes are regulated. The results obtained provide compelling evidence that the conjugation genes of pLS20 are controlled by a complex genetic switch, which is composed of at least three intertwined layers. A scheme of the three layers is shown in Figure 9. One of the levels results from the relative positioning of the main conjugation promoter, Pc, and the divergently oriented promoter Pr, driving expression of the rcoLS20 gene (Fig. 9A). The presence of divergently oriented promoters is a common form of gene organization in bacteria, and the (likely) role of this organization in transcriptional regulation has long been recognized [34]. Nevertheless, direct proof for and detailed analysis of the implications on transcriptional regulation are restricted to only a minor fraction of the divergently oriented transcriptional units detected. Here, we identified the conjugation promoter Pc and showed that it is a relatively strong promoter, which is repressed by the master regulator of conjugation RcoLS20. Importantly, the position of promoter Pc coincides, or at least partially overlaps, with the divergently oriented weak Pr promoter. It has been demonstrated that an RNA polymerase can bind only to one of two overlapping promoters [35], [36]. Thus, in the special configuration of overlapping promoters the RNA polymerase may itself act as a transcriptional regulator. Recently, Bendtsen et al. [37] described theoretical scenarios backed up by experimental data that overlapping promoters indeed can result in a transcriptional switch, provided that they have different activities in the absence of the regulatory protein, combined with a regulator that has a strong differential effect on the regulation of both promoters. This is exactly the case for the Pc/Pr promoter pair; in the absence of the regulator promoter Pc is several hundred folds stronger than Pr, and the presence of the regulator strongly represses the Pc promoter while activating the Pr promoter.
The second level of regulation contributing to the genetic switch concerns the multiple roles that RcoLS20 plays in the Pc/Pr regulation (Fig. 9B). We showed that, on the one hand, RcoLS20 activates transcription of its own weak promoter, Pr, thereby generating a self-sustaining positive feedback loop. On the other hand, RcoLS20 functions simultaneously as an efficient repressor of the Pc promoter. The dual effect that RcoLS20 has on Pc and Pr maintains conjugation effectively in the “OFF” state. We also showed that the level of rcoLS20 induction from an inducible promoter required for efficient repression of the Pc promoter was about ten-fold lower than that required for maximum auto-activation of the Pr promoter. These differential effects of RcoLS20 on repressing and activating the Pc/Pr promoters will also contribute towards maintaining conjugation stably in the “OFF” state under conditions when conjugation should not be activated. Interestingly, we found that at elevated concentrations RcoLS0 inhibits its own transcription. This negative autoregulation probably functions to keep RcoLS20 within a low concentration range in order to respond accurately to the anti-repressor RapLS20 to activate the conjugation genes. The triple effects RcoLS20 has on the regulation of the Pc/Pr promoters will also play an important role when RapLS20 induces the system to switch to the “ON” state. In addition to relieving repression of the strong conjugation Pc promoter, this will simultaneously annihilate autostimulation of the Pr promoter, preventing further synthesis of RcoLS20, which in turn will contribute in pushing and maintaining conjugation in the “ON” state.
A third level contributing to the genetic switch to activate the conjugation genes involves the DNA looping mediated by simultaneous binding of RcoLS20 to operators OI and OII (Fig. 9C). DNA looping mediated by a transcriptional regulator has been reported for several other regulatory systems in prokaryotes and their analyses have revealed that several features are conserved and necessary for DNA looping to occur [for review see, 38]. Our results showed that the properties of RcoLS20 and the DNA in the Pc/Pr region comply with the necessary features for RcoLS20-mediated loop formation. First, using different techniques, we show that RcoLS20, -predicted to contain a helix-turn-helix DNA binding motif in its N-terminal region [29]-, is a DNA binding protein and that it binds specifically to two operators, OI and OII. Second, operator OI, which is located more than 85 bp away from promoters Pc and Pr, is required for efficient regulation of both promoters. Third, RcoLS20 binds cooperatively to both operators. Fourth, dephasing the positions of the two operators by inserting 5 bp in the spacer region destroys proper regulation of the conjugation genes. And fifth, we showed that RcoLS20 forms tetramers in solution. This will create a unit containing multiple DNA binding motifs, facilitating cooperative binding to multiple sites within the two operators.
The DNA loop in the Pc/Pr region of pLS20 is characterized by a small spacer region that separates RcoLS20 operators OI and OII. The spacer length can be used to classify DNA loops into two categories: short or energetic loops, and long or entropic ones. Due to intrinsic stiffness and torsional rigidity of the DNA, loop formation is normally unfavorable for those with spacer lengths shorter than the DNA persistence length (approximately 150 bp), because the curvature energy required for forming such small loops becomes too great. For such short loops to occur specific features like intrinsic static bending or binding of an additional protein inducing bending are required. In the case of pLS20, in which the operators OI and OII are separated by only 75 bp, we show that the spacer region contains a static bent.
The first experimental demonstration that a DNA loop can play a crucial role in transcriptional regulation was reported for the E. coli ara operon in 1984 [39]. Since then, some other operons have been shown to be also regulated by transcriptional regulator-mediated DNA looping [for reviews see], [ 38], [40]–[43], though the actual number of transcriptional systems for which DNA looping has been conclusively demonstrated is remarkably low. In the case of plasmids, reports demonstrating DNA looping systems are limited to only few cases. One of these includes regulation of initiation of DNA replication at the beta origin of the E. coli R6K plasmid [44]; and in the case of Enterococcus faecalis plasmid pCF10 it has been proposed that regulation of its conjugation system involves DNA looping mediated by the pheromone-responsive transcriptional regulator PrgX [for review see, 45]. Bio-informatic analyses suggest that DNA looping mediated regulation of transcription is likely to be more common than the few cases for which this has been demonstrated so far. For instance, Cournac and Plumbridge [38] have screened the E. coli genome for the presence of putative “simple DNA looping systems” in which looping would involve a single regulator (i.e., this analysis included only transcriptional regulators for which the operator sequence is known, and did not take into account the putative loops that would involve heterologous proteins and/or global transcriptional regulators). Under these restrictive settings, this survey identified 48 genes/operons in which DNA looping mediated regulation is likely to play a role. Interestingly, fourteen of them involve divergently oriented promoters. In the context of our studies, it is worth mentioning the regulation of the conjugation genes located on the integrative and conjugative element ICEBs1 that is present in several B. subtilis strains. The gene encoding the transcriptional regulator ImmR, and the excision and conjugation genes are expressed from two divergently oriented promoters that are separated by ∼130 bp. At low concentrations, the ImmR protein can bind to six regions, three being proximal to each promoter. It has been suggested that repression of the immR promoter might involve cooperative interactions between ImmR molecules bound to binding sites proximal to both promoters, i.e. DNA looping [46]. Based on the distribution of the operator sites, DNA looping could also be involved in the transcriptional regulation of the Gram-negative plasmids Ti or IncP-plasmids, where divergent promoters have been shown to be involved in controlling both the replication and transfer functions [47], [48].
What are the benefits of DNA looping in general and for the regulation of the conjugation genes of pLS20 in particular? A major consequence of DNA looping is that it results in a high local concentration of the transcriptional regulator at the right place, which would increase its specificity and affinity [for recent review see, 49]. Often, -and RcoLS20 is not an exception-, transcriptional regulators are produced in limited amounts per cell. Low numbers of regulators enhance the possibility of transcriptional fluctuations between individual cells within a population. In addition, the intrinsic stochasticity of transcription, -also referred to as noise-, affects the temporal effectiveness of transcriptional regulation; again this is especially prominent when the number of regulatory proteins involved is low. Recent evidences indicate that DNA looping contributes importantly to controlling temporal transcriptional noise, as well as dampening transcriptional fluctuations between cells within a population [50], [51]. Thus, DNA looping contributes to the tight regulation of promoters especially when levels of transcriptional regulators are low by diminishing stochastic fluctuations in transcription.
For some differentiation processes, cell-to-cell or stochastic variability in levels of transcriptional regulators form the basis for activation of these processes, resulting in different behavior of genetically identical cells within a population [52]–[54]. Examples of these processes are the formation of persister cells, development of natural genetic competence, spore formation and swimming/chaining. It is believed that such a bet-hedging strategy is beneficial for the fitness of the species because there will always be some cells that are prepared to cope with a deteriorating environmental condition that may arise in the near future. However, for other processes, there may not be such an advantage and it would then be important to tightly repress the process at times when conditions for that process are not apt. Conjugation probably is such a process because there is no benefit in activating the conjugation genes when there is no recipient present to receive the plasmid. The fact that the efficiency of pLS20 transfer during growth conditions antithetic to conjugation is below the detection limit (at least six orders of magnitude lower than those observed during optimal conjugation conditions) strongly indicates that conjugation genes are tightly repressed under such conditions. However, the tight repression of conjugation should not compromise the ability of rapidly switching to high expression of the conjugation genes when appropriate conditions occur. In pLS20 this is achieved by the constellation of DNA looping combined with autoregulated expression of RcoLS20 and overlapping divergent promoters of different strength.
A well-studied genetic switch involving DNA looping is the one that governs the switch from the lysogenic to the lytic state of the Escherichia coli phage λ [for review see], [ 55,56]. In the lysogenic or prophage state, phage λ replicates passively with the host while the lytic genes are repressed. This prophage state is extremely stable and can be maintained for many generations. Upon induction of the SOS response, however, a switch is made to the lytic cycle resulting in excision of the phage genome, followed by its amplification and eventually lysis of the cell and release of phage progeny. The early lytic phage λ genes are located in two divergently oriented operons, which are controlled by the lytic promoters PR and PL. A third operon, which encodes amongst others the CI transcriptional regulator, is located in between the two early lytic operons such that the promoter of gene cI, PRM, flanks the divergently oriented PR promoter driving expression of one of the two early operons. In several aspects, functional analogies exist between CI and RcoLS20 although they share only 16% of identity at their primary protein sequence level. Both RcoLS20 and CI stimulate and repress their own promoter at low and high concentrations, respectively, resulting in a self-sustaining positive feedback loop while keeping the transcriptional regulator in a low concentration range. Above, arguments have been given that for pLS20 this situation, together with the effects of the DNA loop, is important for the tight repression of the Pc promoter during conditions in which conjugation is not favourable, while maintaining the sensitivity to be able to respond rapidly to switch on the conjugation genes when appropriate conditions occur. The transcriptional regulation of λ appears to serve a similar purpose. Thus, on the one hand the lytic genes are tightly repressed since spontaneous switching to the lytic cycle occurs less than once every 108 generations [57]. On the other hand, mutations that specifically eliminate the negative autoregulation of cI expression impair prophage induction [58], [59]. Another analogy between the pLS20 and λ systems is that both the regulators RcoLS20 and CI, can form higher order oligomers, permitting them to bind cooperatively to multiple sites distributed in two operators, effectively resulting in DNA looping which plays an important role in the genetic regulation of the conjugation and the lytic operon, respectively. Taking the analogy further, it is interesting to note that these regulatory systems both control a process of horizontal gene transfer.
However, there are also several differences between the two systems. For instance, whereas regulation of pLS20 conjugation genes involves a short loop of 75 bp, regulation of the λ lytic genes involves a long loop of 2.3 kb. A second difference is that CI protein forms dimers in solution. A pair of CI dimers tetramerizes when binding to the binding sites in one operator and another dimer pair does the same when binding to the other operator. Upon DNA looping, interaction between the two tetramers constitutes a functional octamer. In addition, when a loop is formed another pair of dimers may bind to additional binding sites present in both operators, and this additional bridge is responsible for repressing PRM promoter. At present, we do not have such detailed insights in transcriptional regulations at the molecular level for RcoLS20. However, instead of dimers, RcoLS20 forms tetramers in solution, which probably means that the molecular mechanism by which the pLS20 promoters Pr and Pc are regulated is distinct from the way CI regulates λ promoters PR and PRM. Another argument supporting this assumption is the different configuration of the divergent promoters and the binding sites for the regulator protein. In pLS20, the position of promoters Pc/Pr overlaps and the RcoLS20 binding sites in OII overlap and flank these core promoters. In λ the binding sites for CI regulator in one operator overlap the PR promoter and are located upstream of the PRM core promoter sequences. Finally, a major difference between the DNA looping involved systems of pLS20 and λ is how the switches are induced. In λ, the switch is induced by an SOS response which results in RecA-mediated CI autocleavage. In the case of pLS20, the switch is dictated ultimately by intercellular quorum sensing signaling involving the signaling peptide Phr*LS20 that regulates the activity of RapLS20, the anti-repressor of RcoLS20 [29]. This quorum sensing system will lead to activation of the conjugation genes when donor cells are surrounded by recipient cells. However, high levels of Phr*LS20 will build up when the majority of the cells that surround a donor cell already contain pLS20, and this will inactivate RapLS20 and hence block activation of the conjugation genes.
Besides those described here, it is possible that the pLS20 conjugation genes are regulated by additional mechanism(s). For example, the conspicuously long 5′ untranslated region upstream of gene 28 is predicted to form complex secondary structures, which might modulate expression of the downstream genes in a variety of scenarios. Currently, a study to elucidate a possible role of this long 5′ untranslated region is carried out in our laboratory.
In summary, in this work we have provided evidence that regulation of the conjugation genes present on pLS20 is based on a unique genetic switch that combines at least three levels of control. These include (i) overlapping divergent promoters of different strengths, (ii) auto-stimulation and repression of the weak Pr promoter by the transcriptional regulator at low and elevated concentrations, respectively, combined with simultaneous repression of the divergent strong conjugation promoter, and (iii) DNA looping mediated by binding of RcoLS20 regulator to two operators separated by a short loop. Most likely, the combination of these different layers causes tight repression of the main conjugation promoter Pc when conditions for conjugation are not optimal, while allowing the system to switch rapidly to high expression of the conjugation genes when appropriate conditions occur.
Bacterial strains were grown in LB liquid medium or on 1.5% LB agar plates [60]. When appropriate, the following antibiotics were added to media or plates: ampicillin (100 µg/ml), erythromycin (1 and 150 µg/ml in B. subtilis and E. coli, respectively), chloramphenicol (5 µg/ml), spectinomycin (100 µg/ml), and kanamycin (10 µg/ml). Table S1 lists the B. subtilis strains used. All of them are isogenic with B. subtilis strain 168. Plasmids and oligonucleotides used are listed in Table S2 and S3, respectively. All oligos were purchased from Isogen Life Science, The Netherlands.
E. coli cells were transformed using standardized methods as described in Singh et al [61]. For standard B. subtilis transformations, competent cells were prepared as described by Bron [62]. Transformants were selected on LB agar plates with appropriate antibiotics.
Standard molecular methods were used to manipulate DNA [60]. Sequence analysis was used to verify the correctness of all constructs. The same strategy was used to construct B. subtilis strains containing a copy of lacZ fused to the entire or part of the rcoLS20-gene 28 intergenic DNA region. First, the region of DNA to be cloned was amplified using appropriate primers (see Table S3), purified, and digested with the appropriate restriction enzymes. Next, the fragment was used to prepare a ligation mixture together with the integration vector pDG1663 digested with the same enzymes. The ligation mixture was transformed into E. coli XL1-blue cells. The plasmid content of several ampicillin resistant transformants was checked and clones containing the insert with appropriate size and orientation were subjected to DNA sequencing to verify the absence of mutations. The names of the pDG1663 derivatives and their characteristics are listed in Table S2. Plasmid DNA of each pDG1663 derivative was used to transform competent B. subtilis 168 cells. Transformants were initially selected for resistance to erythromycin. Next, double cross-over events were distinguished from single cross-over events by selecting transformants sensitive to spectinomycin. The resulting B. subtilis strains containing a single copy of lacZ preceded by different regions of the rcoLS20-gene 28 region at the thrC locus of the B. subtilis chromosome are listed in Table S1. Next, plasmid pLS20cat was introduced into the different lacZ fusion strains by conjugation. B. subtilis strain PKS9 contains a single copy of the rcoLS20 gene under the control of the IPTG-inducible Pspank promoter at its amyE locus and this cassette is linked to the spectinomycin gene. Chromosomal DNA of strain PKS9 was used to transform competent cells of the various lacZ fusion strains in order to construct derivatives of the lacZ fusion strains containing the Pspank-rcoLS20 cassette.
The following strategy was used to construct a translational fusion of rcoLS20 with his(6). The rcoLS20 gene was amplified from pLS20cat by PCR using primers oPKS14N and oPKS8. The purified PCR product was digested with NcoI and SalI and cloned into the vector pET28b+ digested with the same restriction enzymes to produce plasmid pRcoLS20-His. B. subtilis strain GR90 contains the rcoLS20-his(6) under the control of the Pspank promoter at the amyE locus. To construct this strain rcoLS20-his(6) was amplified from pRcoLS20-His by PCR using primers oGR3 and oGR4. The PCR product was digested with NheI and SphI and cloned into the vector pDR110 digested with the same enzymes to generate pPspankrcoLS20-His. This plasmid was used to transform competent B. subtilis cells selecting for spectinomycin resistance. Double cross-over events were selected by loss of amylase gene.
β-galactosidase activities were determined as described previously [63]. Overnight grown cultures were diluted 100 times into fresh prewarmed medium and samples were taken every 45 min.
Conjugation was carried out in liquid medium as described previously [29]. The effect of ectopic expression on conjugation of a gene controlled by the IPTG-inducible Pspank promoter was studied as follows. Overnight cultures were diluted in prewarmd LB supplemented with IPTG at the indicated concentrations to an OD600 of ∼0.05. Next, samples were taken at regular intervals to determine OD600 and were subjected to matings with proper recipient cells.
Preparation of total RNA samples, RNA sequencing and Bioinformatic analysis of RNAseq data was done as described previously [29].
E. coli BL21 (DE3) cells carrying plasmid prcoLS20-His were used to inoculate 1 litre of fresh LB medium supplemented with 30 mg/ml kanamycin and grown at 37°C with shaking. At an OD600 of 0.4, expression of rcoLS20-his(6) was induced by adding IPTG to a final concentration of 1 mM and growth was continued for 2 h. Cells were further processed as described previously [28]. Purified protein (>95% pure) was dialysed against buffer B (20 mM Tris-HCl pH 8.0, 1 mM EDTA, 250 mM NaCl, 10 mM MgCl2, 7 mM β-mercaptoethanol, 50% v/v glycerol) and stored in aliquots at −80°C. Bradford assay was used to determine the protein concentrations.
In essence, the gel retardation assays were carried out as described before [28]. Thus, different fragments of intergenic regions between gene 28 and rcoLS20 were amplified by PCR using pLS20cat as template. The resulting PCR fragments were purified and equal concentrations (300 nM) were incubated on ice in binding buffer [20 mM Tris HCl pH 8, 1 mM EDTA, 5 mM MgCl2, 0.5 mM DTT, 100 mM KCl, 10% (v/v) glycerol, 0.05 mg ml−1 BSA] without and with increasing amounts of purified RcoLS20His(6) in a total volume of 16 µl. After careful mixing, samples were incubated for 20 min at 30°C, placed back on ice for 10 min, then loaded onto 2% agarose gel in 0.5XTBE. Electrophoresis was carried out in 0.5X TBE at 50 V at 4°C.Finally, the gel was stained with ethidium bromide, destained in 0.5XTBE and photographed with UV illumination.
Determination of the transcription start sites by primer extension was performed essentially as described [64]. In brief, total RNA (30 µg) was mixed with 4 pmol of end-labeled oligonucleotide that served as primer; the mixture was heated at 70°C for 5 min and allowed to anneal for 5 min at 23°C. The annealed RNA was ethanol precipitated, resuspended and primer extension was performed with 30 U of AMV reverse transcriptase (Promega) at 42°C, as recommended by the supplier. The extended cDNA products were analysed by electrophoresis on a denaturing 6% urea-polyacrylamide gel, in parallel with a DNA sequence ladder performed by chemical sequencing [65] of a DNA fragment encompassing the mapped promoters (see below). The primer used to map promoter Pc was 5′-ttctagttctttttacac, while that used for promoter Pr was 5′-tctctattgcccacttat. Oligonucleotides were end-labeled with [γ-32P]-ATP and T4 polynucleotide kinase as recommended by the supplier (New England Biolabs). The 186 bp DNA fragment that served as sequence ladder was PCR amplified with primers 5′-acggtctagcgcttacaat and 5′-ttctagttctttttacac, the last one labeled at its 5′ end.
DNaseI footprinting assay was carried out as described [66]. The Pc/pr promoter encompassing region was amplified by PCR using primers p28_Δ16 and Prom28UpBam, and pLS20cat as template. One of the ends was radio-labeled by digesting the fragment with BamHI and subsequently filling in the end with exo− Klenow fragment in the presence of [α-32P]-ATP.
Presence of conserved motifs was searched by using motif-identification programs MEME [30] and BIOPROSPECTOR [31]. Prediction of the static bending properties of DNA sequences was carried out by calculating the global 3D structure according to the dinucleotide wedge model [67]. All graphics work was done by using Adobe Photoshop CS2 and adobe illustrator. Graphs were plotted using Excel program.
Sedimentation velocity assay. Samples in 20 mM Tris-HCl, 250 mM NaCl, 10 mM MgCl2, 1 mM EDTA and 100 mM glycerol, pH 7.4, were loaded (320 µL) into analytical ultracentrifugation cells. The experiments were carried out at 43–48 krpm in an XL-I analytical ultracentrifuge (Beckman-Coulter Inc.) equipped with UV-VIS absorbance and Raleigh interference detection systems. Sedimentation profiles were recorded at 280 nm. Sedimentation coefficient distributions were calculated by least-squares boundary modelling of sedimentation velocity data using the continuous distribution c(s) Lamm equation model as implemented by SEDFIT 14.1 [68]. Experimental s values were corrected to standard conditions (water, 20°C, and infinite dilution) using the program SEDNTERP [69] to get the corresponding standard s values (s20,w).
Sedimentation equilibrium assay. Using the same experimental conditions as in the SV experiments, short columns (90 µL) SE experiments were carried out at speeds ranging from 7,000 to 10,000 rpm and at 280 nm. After the last equilibrium scan, a high-speed centrifugation run (48,000 rpm) was done to estimate the corresponding baseline offsets. Weight-average buoyant molecular weights of protein were determined by fitting a single species model to the experimental data using the HeteroAnalysis program [70], and corrected for solvent composition and temperature with the program SEDNTERP [69].
|
10.1371/journal.ppat.1000292 | Differences in APOBEC3G Expression in CD4+ T Helper Lymphocyte Subtypes Modulate HIV-1 Infectivity | The cytidine deaminases APOBEC3G and APOBEC3F exert anti–HIV-1 activity that is countered by the HIV-1 vif protein. Based on potential transcription factor binding sites in their putative promoters, we hypothesized that expression of APOBEC3G and APOBEC3F would vary with T helper lymphocyte differentiation. Naive CD4+ T lymphocytes were differentiated to T helper type 1 (Th1) and 2 (Th2) effector cells by expression of transcription factors Tbet and GATA3, respectively, as well as by cytokine polarization. APOBEC3G and APOBEC3F RNA levels, and APOBEC3G protein levels, were higher in Th1 than in Th2 cells. T cell receptor stimulation further increased APOBEC3G and APOBEC3F expression in Tbet- and control-transduced, but not in GATA3-transduced, cells. Neutralizing anti–interferon-γ antibodies reduced both basal and T cell receptor-stimulated APOBEC3G and APOBEC3F expression in Tbet- and control-transduced cells. HIV-1 produced from Th1 cells had more virion APOBEC3G, and decreased infectivity, compared to virions produced from Th2 cells. These differences between Th1- and Th2-produced virions were greater for viruses lacking functional vif, but also seen with vif-positive viruses. Over-expression of APOBEC3G in Th2 cells decreased the infectivity of virions produced from Th2 cells, and reduction of APOBEC3G in Th1 cells increased infectivity of virions produced from Th1 cells, consistent with a causal role for APOBEC3G in the infectivity difference. These results indicate that APOBEC3G and APOBEC3F levels vary physiologically during CD4+ T lymphocyte differentiation, that interferon-γ contributes to this modulation, and that this physiological regulation can cause changes in infectivity of progeny virions, even in the presence of HIV-1 vif.
| Some host cell proteins can hinder, or restrict, the life cycle of HIV-1. APOBEC3G and APOBEC3F are cellular enzymes that decrease HIV-1's ability to replicate in a subsequent target cell if they are present in the virus particle. As a countermeasure, HIV-1 virion infectivity factor (vif) induces degradation of APOBEC3G and APOBEC3F, thereby preventing them from getting into the budding virus. Although vif-defective viruses cannot evade the antiviral effect of APOBEC3G, such viruses are very rarely present in HIV-1-infected humans. It is not yet known whether physiological variation in APOBEC3G and APOBEC3F expression in CD4+ T lymphocytes is substantial enough to reduce vif-positive HIV-1 infectivity. In this study, we found that T helper type 1 (Th1) cells, a subtype of CD4+ lymphocytes, expressed greater amounts of APOBEC3G and APOBEC3F than T helper type 2 (Th2) cells. This difference led to a difference in infectivity of HIV-1 produced from the two cell types, whether vif was expressed or not. These results demonstrate that physiological regulation of APOBEC3G does restrict vif-positive HIV-1, as well as vif-negative HIV-1. In addition, this study reveals biological factors regulating expression of these proteins that may be exploitable for new therapeutic or preventive strategies against HIV-1.
| APOBEC3G (hA3G) and APOBEC3F (hA3F), two of several related cytidine deaminases, evolved to limit retrotransposition [1]–[3]. Although the HIV-1 accessory protein vif depletes hA3G and hA3F from the producer cell, hA3G and hA3F are packaged into vif-deleted HIV-1 and significantly impair virion infectivity [4]–[6]. IFN-α, and certain cytokines and mitogens, have been implicated in increasing hA3G and hA3F expression in certain cell types [7]–[12]. However, little more is known regarding the transcriptional regulation of APOBEC3s in CD4+ T lymphocytes [13]. We noted several potential binding sites for GATA family transcription factors [14], in addition to previously observed interferon-responsive elements [7],[12],[15], in the putative promoter regions of hA3G and hA3F. Since GATA3 is integral to the differentiation of naïve CD4+ T helper cells into Type 2 (Th2) effectors, we hypothesized that Type 1 (Th1) and Th2 effector lymphocytes differed in their expression of hA3G and hA3F.
After naive CD4+ T lymphocytes interact with their cognate antigen, IL-12 and interferon-γ (IFN-γ) signaling drive their differentiation to a Th1 effector phenotype. In contrast, IL-4 signaling after antigen recognition drives differentiation of naïve cells to a Th2 phenotype [16]–[18]. These subtypes of T helper cells produce distinct cytokine profiles after subsequent activation. Th1 cells, when activated, produce IFN-γ to activate cell-mediated immunity. Th2 cells, however, secrete IL-4 and other cytokines which augment humoral immune responses. The differentiation to a Th1 or Th2 phenotype is dependent on the regulated expression of two master transcriptional regulators, respectively: T Box expressed in T cells (Tbet) and GATA3 [19]–[21]. Relative differences in the ability of Th1 versus Th2 subtypes to produce infectious wild-type HIV-1 progeny have been reported previously in several studies and were not explained by differences in expression of chemokine co-receptors for HIV entry [22]–[25].
Although high level over-expression of hA3G has been reported to decrease infectivity of vif-positive virions produced from cell lines in vitro [4], [26]–[28], it is not known whether physiological increases in hA3G or hA3F can overcome the effect of vif in primary T cells. Reports conflict about whether differences in levels of hA3G and hA3F in lymphocytes in vivo are inversely associated with the level of wild-type HIV-1 RNA in plasma of untreated patients [29]–[31]. One of two reports has correlated provirus hypermutation attributable to hA3G and hA3F with plasma viral load, consistent with effects in vivo against at least some vif-positive viruses [32],[33]. Since an effect of variation of levels of hA3G and hA3F in a physiologically relevant range on wild-type, vif-positive HIV-1 replication has not yet been directly demonstrated, the present study aimed to define if such cellular differences occur during Th1 versus Th2 differentiation and may cause changes in HIV-1 infectivity that affect pathogenesis.
Naïve CD4+ T cells from five individual HIV-1 negative donors were transduced with HIV-derived lentiviral vectors that expressed either GFP alone (control), or together with Tbet or GATA3. Expression of GATA3 and Tbet were found to have opposing effects on the expression of hA3G and hA3F mRNA by qRT-PCR (Figure 1A and 1B). Whereas expression of GATA3 reduced the level of hA3G and hA3F, Tbet significantly increased the levels of both enzymes. Based on these results, confirmation that this was a statistically and biologically significant effect was sought by studying Th1 versus Th2 differentiation using more physiological cytokine polarization.
Th1 and Th2 cells were differentiated in vitro by culturing naïve cells from nine individual donors in polarizing cytokines. Staining for Th1- and Th2-associated intracellular cytokines (IFN-γ and IL-4, respectively) and surface markers (CXCR3 and CRTh2, respectively) (Figure 2A), verified the phenotypes of the cytokine-differentiated cells. Cytoplasmic RNA was isolated and the levels of hA3G and hA3F mRNA were determined relative to GAPDH expression by qRT-PCR. Th2 cells expressed significantly less hA3G and hA3F mRNA than Th1 cells (Figure 2B and 2C). Western blot analysis of the two helper cell subtypes revealed that Th2 cells also expressed lower levels of hA3G protein than Th1 cells (Figure 2D and 2E). The statistically significant difference in median hA3G mRNA levels, and in protein levels, between Th1 and Th2 cells was approximately 3-fold.
Since previous studies have observed that mitogen treatment increases hA3G expression [7], we tested whether T Cell Receptor (TCR) stimulation would increase hA3G and hA3F expression in Tbet and GATA3 expressing T-cells. Levels of hA3G and hA3F RNAs increased after TCR stimulation of control vector- and Tbet-transduced cells, while this did not occur with TCR activation of GATA3-transduced cells (Figure 3A). A defining characteristic of Th1 cells is their ability to produce IFN-γ upon activation, which then exerts autocrine effects [34]. It is also known that GATA3 diminishes IFN-γ expression. Therefore, the hypothesis that IFN-γ contributes to the observed increase in hA3G and hA3F expression after TCR stimulation was tested by performing TCR stimulation of control- and Tbet-transduced cells in the absence or presence of a neutralizing anti- IFN-γ antibody. The presence of neutralizing anti- IFN-γ antibody blocked the TCR-stimulated increased transcription of hA3G and hA3F, and reduced basal levels, in both control- and Tbet-transduced cells (Figure 3B and 3C). This suggests that IFN-γ contributes to maintaining the steady state level of hA3G and hA3F in Th1 cells, as well as in increasing expression after TCR activation.
We next tested whether the differential expression of APOBEC3s between Th1 and Th2 cells led to a difference in infectivity of HIV-1 virions produced from these cells. We infected TCR-activated, cytokine-derived T helper cells with vif-deleted or vif-competent HIV-1(NL4-3) produced from 293T cells (which do not express hA3G or hA3F). Infected cells were washed 12 hours after infection and new media containing reverse transcriptase inhibitors (didanosine and zidovudine) was added to prevent spread past the first-round infected cells. Twelve hours after the new media was added, the culture supernatant fluids were collected, normalized by Gag p24 capsid antigen concentrations, and used to infect the TZM-bl indicator cell line. Infectivity was determined by luciferase activity. Figure 4A quantitates infectivity of wild-type and vif-deleted viruses produced from Th1 and Th2 cells from one of nine donors studied. Vif-negative viruses produced from Th2 cells from this individual were five-fold more infectious than those produced from Th1 cells, whereas vif-competent virions from Th2 cells were three-fold more infectious than those produced from Th1 cells (Figure 4A). The median infectivity of virions produced from Th1 cells of all nine donors studied was significantly less than that of viruses produced from all the different Th2 cells, whether vif was present or not (Figure 4B and 4C). The magnitude of this difference varied across different individual donors' paired Th1 and Th2 cells, whether vif was present or not (Figure 4B and 4C; each donor's Th1 and Th2 cells are linked by a line). A3G protein levels also varied, with Th1 cells having higher levels and a broader range of hA3G protein than Th2 cells (comparing X-axis in Figure 4D and 4E). Despite the small number of subjects and variability of the assays, there was a suggestion of an inverse correlation between A3G protein expression and infectivity of virions produced from Th1 (+vif r = −.16, −vif r = −.18; all p>0.05) or Th2 (+vif r = −.28, −vif r = −.01; all p>0.05). We amplified a pol gene fragment from the TZM-bl cells infected for 60 hours with Th1- or Th2-produced vif-negative virions to quantify if effects of cytidine deamination differed by cell type. Hypermutation was not seen in HIV pol DNA amplified from cells infected with virus produced from either cell type (data not shown), using either population sequencing subsequent to standard PCR or 3D PCR [35]. Although these data are consistent with direct effects of hA3G and hA3F on infectivity of vif-competent HIV as well as vif-defective HIV-1, it is possible that other variables may affect infectivity of virions produced from these cytokine-polarized cells.
To confirm a causal role for hA3G in the observed virion infectivity differences we modulated expression of A3G in Th1 and Th2 cells by increasing expression in Th2 cells and decreasing expression in Th1 cells. We increased expression of hA3G in cytokine-derived Th2 cells by transduction with a hA3G-expressing lentiviral vector or an “empty” control vector for comparison. Transduction of Th2 cells with the hA3G-expressing vector increased hA3G levels 4 fold over Th1 cells and 7 fold over Th2 cells (data not shown). After expansion, the unsorted population of hA3G vector-transduced Th2 cells (Th2-A3G), as well as Th1 and Th2 cells, were infected. The vif-deleted virions produced from Th1, Th2 and Th2-A3G cells were concentrated and the relative levels of virion packaged hA3G were determined by Western blotting. Figure 5A demonstrates that vif-deleted virions produced from Th1 cells contain more hA3G than virions produced from Th2 cells. Th2-A3G cells produced virions with more packaged hA3G than Th2 cells (Figure 5A). Transduction with the empty vector (Th2-Empty) caused no increase in cellular or virion hA3G levels, relative to untransduced Th2 cells (data not shown). Virions produced from the Th2-A3G cells were significantly less infectious than those produced from the Th2 cells transduced with the “empty” control vector (“Th2-Empty”) (Figure 5B). There was an inverse correlation between virion (and cellular) hA3G levels by western blot and virion infectivity.
Neutralizing anti-IFN-γ antibody was used to decrease expression of A3G in Th1 cells (as seen in Figure 3). Incubation with neutralizing anti-IFN-γ antibody, concurrent with activation, reduced the expression of A3G in Th1 cells nearly 2 fold (relative to Th1 cells incubated with an isotype control antibody) (Figure 5C, open bars). Virions produced from Th1 cells with reduced hA3G had increased infectivity (Figure 5C, closed bars). Taken together, these data indicate that variation in infectivity of virions produced from cells is related to differences in hA3G expression.
In this study, we have shown that the expression and anti-HIV function of hA3G and hA3F vary with naïve CD4+ T helper cell differentiation to Th1 and Th2 effector cells. Cytokine polarization of naïve cells into Th1 and Th2 effectors had similar effects to transduction of naïve cells with Tbet or GATA3. In both cases, decreased expression of hA3G and hA3F was seen in Th2 cells relative to Th1 cells. These complementary methods demonstrate that the differences observed in the Tbet and GATA3 transduced cells were due to transcription regulated by those factors and not an artifact of over-expression. Such an opposing effect of differentiation on expression of hA3G and hA3F is consistent with earlier findings of opposing effects on the expression of several other genes in these two T helper subtypes [21],[36]. This hA3G and hA3F expression difference between Th1 and Th2 cells affected wild type, as well as vif-deleted, HIV-1 infectivity.
Expression of Tbet in naïve helper cells has been shown to lead to production of IFN-γ[Σζαβo, 2000 #35]. In turn, that IFN-γ can act in an autocrine manner on Th1 cells [34]. Extracellular neutralization of IFN-γ secreted by Tbet-transduced and control-transduced cells blocked basal and TCR-stimulated hA3G and hA3F expression. This is consistent with an autocrine effect of IFN-γ regulating hA3G and hA3F expression. GATA3 is known to inhibit the production of IFN-γ [37] and no effect was observed with neutralizing anti-IFN-γ antibody or TCR stimulation of GATA3-transduced cells. This may be due to a GATA3-mediated block to production of IFN-γ, or a direct effect of GATA3 binding to the hA3G and hA3F promoters. These possibilities remain to be directly tested.
We verified that the difference in expression in cytokine-derived T helper cells led to a biological difference: infectivity of HIV-1 virions produced from Th1 and Th2 effectors varied inversely with their relative levels of cellular and virion hA3G and hA3F. Removal of vif resulted in reduced infectivity of virions produced from both cell types. The greater reduction of infectivity of virions produced from the Th1 cells is consistent with the relative greater APOBEC3 levels in those cells. Over-expression of hA3G in Th2 cells reversed the relative decrease in virion hA3G and the consequent relative increase in infectivity of virions produced from Th2 cells. The magnitude of the effect of the ectopically-expressed hA3G is likely underestimated here, as not every cell in this population is expressing the transduced hA3G. In addition, reduction of hA3G in Th1 cells also correlated with an increase in infectivity. We used neutralizing anti-IFN-γ antibody to decrease hA3G expression in Th1 cells because shRNA against hA3G or nucleofection (for introduction of siRNA against hA3G) proved toxic to in vitro-derived Th1 cells, which are more prone to cell death than other cultured T cells [38]–[40].These results are consistent with the variation in virion infectivity being caused, at least in part, by the differences in cellular and therefore virion hA3G, rather than other effects of the cytokine derivation.
In this study, we observed reduction of infectivity associated with increased amounts of readily detectable virion hA3G without identification of any G-to-A hypermutation. Although hA3G and hA3F are cytidine deaminases, there is extensive evidence that hA3G also reduces HIV infectivity through other mechanisms that may be the major contributor to A3G's inhibition of reverse transcription [26],[33],[41],[42]. Previous studies that have observed hA3G-related hypermutation in vitro differed from the short term virus replication allowed here, and instead used prolonged serial passage of HIV in transformed cell lines over-expressing hA3G [43],[44]. Therefore, it is likely that the difference in infectivity based on cell source of virus observed here is due to the other antiviral activities of hA3G that are not measured by hypermutation.
A major issue concerning the role of hA3G and hA3F in HIV-1 pathogenesis is the question of whether in vivo variation in these cellular restriction factors affects replication of wild type (eg, vif-competent) HIV-1. Although high level over-expression of hA3G does impair replication of wild type HIV-1 in cell lines [4], more recent studies have not conclusively determined if there is a correlation between the variation in cellular hA3G expression observed across HIV-infected individuals' peripheral blood mononuclear cells and the plasma viral load in these subjects [29],[30],[32]. The present results clearly indicate that physiological variations in hA3G levels in primary cells are inversely correlated with hA3G content and infectivity of wild type virions. This more direct measure of biological relevance observed here supports the conclusions of earlier reports showing that greater A3G activity was associated with lower viral load set-point [32], and suggests that continued investigation of the effect of APOBEC3 restriction factors on vif-competent HIV-1 pathogenesis in vivo is warranted.
The present results are also consistent with earlier reports that HIV-1 spreads better through cultures of Th2 cells than Th1 cells [24]. This effect was most apparent in the prior studies using CXCR4 (X4) tropic viruses [25], such as the viruses used here, and not explained by differences in expression of that co-receptor between Th1 and Th2 cells. The present results suggest, however, that virions produced from Th2 cells may be relatively more infectious than those produced from Th1 cells because of their relatively lower hA3G content. In an earlier study [25], CCR5-tropic HIV replicated equally well in Th1 and Th2 cells. Th1 cells express higher levels of CCR5 coreceptor than Th2 cells [21],[22]. Indeed, X4-tropic viruses were chosen for study here to minimize possible difficulty in interpretation of opposing effects of both increased CCR5 co-receptor expression and increased A3G expression in Th1 cell cultures, though further investigation into how co-receptor tropism affects infectivity is certainly warranted. Moreover, the wide inter-individual variation in hA3G and hA3F expression in our results (a 14 fold range in hA3G protein expression in Th1 cells and a four fold range in Th2 cells) suggests that there may be polymorphisms in the regulatory regions of the APOBEC3 promoters [45], or in factors that can modulate hA3G and hA3F expression or function. We hypothesize that this variation in hA3G and hA3F may contribute to the wide variation of progression time to AIDS among different patients. The Th1/Th2 cell balance may also vary across individuals based on several factors. Autoimmunity may lead to a Th1 cell skewing and parasitic infections may cause aTh2 cell predominance. Our findings suggest that a shift in this balance prior to, or during, HIV-1 infection may lead to compounded pathogenic effects. Decreased relative expression of hA3G and hA3F in Th2 cells may lead to a greater rate of decrease in that cellular pool, decreasing CD4+ help to B cells for antibody production. Also, an individual's variation in Th1/Th2 balance may lead to differences in HIV-1 genetic variation due to hA3G- and hA3F-mediated sub-lethal cytidine deamination of viral genomes over repeated cycles of infection [46].
The present study indicates that the regulation of expression of hA3G and hA3F, and their functional effect on HIV-1 infectivity, depends on the cytokine-regulated differentiation state of CD4+ T helper cells. Further molecular characterization of signals that modulate hA3G and hA3F expression will be needed. The current results provide compelling evidence that increasing hA3G in primary T cells impairs HIV-1 replication despite the presence of Vif. This validates inducing higher hA3G expression as a novel strategy for prevention of infection and/or treatment of the vif-positive viruses present in infected humans.
Blood was obtained from healthy volunteers under a protocol approved by the Vanderbilt Institutional Review Board. PBMCs were isolated using Ficoll Hypaque (Amersham Biosciences). CD4+ cells were isolated by negative selection through magnetic separation using autoMacs (Miltenyi Biotec, Auburn, CA) or Robosep (StemCell Technologies, Vancouver, BC. Canada). Naïve cells were subsequently purified by staining with CD45RO-FITC and CD25-PE (BD Pharmingen, San Jose, CA) followed by sorting on a FACSAria (Becton Dickinson, San Jose, CA). For activation and expansion, naïve cells were plated in wells coated with an anti-CD3 antibody (OKT3; American Type Culture Collection, Manassas, Virginia, United States) in RPMI with 10% FBS supplemented with 1 µg/ml soluble anti-CD28 antibodies (BD Biosciences Pharmingen) and 50 U/mL human rIL-2 (obtained from Dr. Maurice Gately, Hoffmann - La Roche Inc. through the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH) [47]. DMEM with 10% FBS was used to culture TZM-bl cells (obtained from Dr. John C. Kappes, Dr. Xiaoyun Wu and Tranzyme Inc. through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH) [48].
Naïve CD4+ T cells were differentiated by transduction with HIV derived lentiviral vectors expressing Tbet, GATA3, or a control vector at the time of activation [49],[50]. The vectors express GFP alone (control), or the transcription factor and GFP, from an IRES. After infection and activation, cells were expanded for 10 days. Following expansion, cells were sorted on a FACSAria for GFP expression.
To achieve Th1 cell differentiation using cytokine polarization, naïve CD4+ T cells were plated on anti-CD3 (OKT3) coated plates in RPMI supplemented with anti-CD28 antibodies, 0.5 µg/mL neutralizing anti-IL-4 antibody and 30 ng/mL recombinant IL-12. For Th2 cell differentiation by cytokines, naïve cells were cultured in media supplemented with 2.5 µg/mL neutralizing anti-IFN-γ antibody and 50 ng/mL recombinant IL-4. Cytokines and neutralizing antibodies were obtained from R&D Systems, Minneapolis, MN. The cells were expanded for 10 days and differentiation was confirmed by intracellular cytokine staining for IL-4-PE and IFN-γ-APC (BD Pharmingen, San Jose CA.) as previously described [49], as well as surface staining for CXCR3-PE and CRTh2-APC (BD Pharmingen, San Jose CA) [51]–[53]. To increase APOBEC3G expression in cytokine polarized Th2 cells, differentiating cultures were transduced with an APOBEC3G-expressing HIV derived lentiviral vector at the time of activation. The vector was constructed as other HIV derived lentiviral expression vectors previously described to express hA3G and HSA as a marker of transduction [50]. To reduce APOBEC3G expression in cytokine polarized Th1 cells, fully differentiated Th1 cells were activated for 48 hrs with CD3/CD28 coated beads (Invitrogen) in the presence of 5 µg and 10 µg anti-IFN-γ antibody (R&D Systems).
Cytoplasmic RNA was isolated from cell pellets (Qiagen RNeasy, Valencia, CA). RNA was quantified by spectrophotometry on a GeneQuant Pro (Amersham Biosciences, Piscataway, NJ). RNA concentrations were normalized and TaqMan quantitative real-time RT-PCR was performed (Applied Biosystems Prism 7000 Sequence Detection System, Foster City, CA). Reverse transcription used hA3G and hA3F specific primers with the sequences 5′- GCGGCCTTCAAGGAAACC-3′ and 5′-TTTTAAAGTGGAAGTAGAATATGTGTGGAT-3′, respectively. The primer-probe set used for APOBEC3G real-time PCR was: forward: 5′-CTGCTGAACCAGCGCAGG-3′ reverse: 5′-GCGGCCTTCAAGGAAACC-3′ and probe: 5′-CTTTCTATGCAACCAGGCTCCACATAAAC-3′. The set for APOBEC3F was: forward: 5′-GCACCGCACGCTAAAGGA-3′, reverse 5′- TTTTAAAGTGGAAGTAGAATATGTGTGGAT -3′ and probe: 5′TTCTCAGAAACCCGATGGAGGCAATG-3′. Values are expressed as copies of target per million copies of GAPDH or calculated as fold change using the delta-Ct method [54].
Transduced or cytokine-derived T helper subtype cells were lysed in 50 mM HEPES, pH 7.4, 125 mM NaCl, 0.2% NP-40, 0.1 mM PMSF and EDTA-free protease inhibitor cocktail (CalBiochem, San Diego, CA). Protein concentrations were normalized based on results of a Bradford Assay (Bradford Assay reagent, Bio-Rad, Hercules, CA). Lysates were separated on a SDS-PAGE gel and proteins were subsequently transferred to a Trans-Blot nitrocellulose membrane (Bio-Rad, Hercules, CA). The membrane was then incubated with a polyclonal anti-APOBEC3G antibody [55], washed and probed with a goat anti-rabbit secondary antibody conjugated with Alexa Fluor 680 (Invitrogen Molecular Probes, Carlsbad, CA). Fluorescent signal was then measured using the Licor Odyssey system (LI-COR Biosciences, Lincoln, Nebraska). Membranes were subsequently probed with a monoclonal β-actin antibody (Sigma, St. Louis, MO) followed by a sheep anti-mouse secondary antibody conjugated with IR-Dye800 (Rockland Immunochemicals, Philadelphia, PA). APOBEC3G expression is expressed as fluorescent intensity (Relative Light Units, RLU) of APOBEC3G bands divided by the fluorescent intensity (RLU) of the β-actin band [56]. For quantification of virion packaged APOBEC3G, virions were concentrated by centrifugation of culture supernatants through a 20% sucrose cushion at 125,000×g for 45 minutes and normalized for their p24 content with viral lysis buffer [50 mM Tris (pH 8.0), 40 mM KCl, 50 mM NaCl, 5 mM Na2EDTA, 10 mM DTT and 0.1% (v/v) Triton X-100]. Lysates were blotted as described above with anit-APOBEC3G and an anti-HIV-1 capsid p24 antibody derived from the 183-H12-5C hybridomas (obtained from Dr. Bruce Chesebro and Dr. Hardy Chen through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH [57]). Data are expressed as fluorescent intensity (RLU) of APOBEC3G bands divided by the fluorescent intensity (RLU) of the HIV-1 CA p24.
HIV-1 was produced by calcium phosphate transfection of 293T cells using NL4-3 (obtained from Dr. Malcom Martin through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH) [58] and vif-deleted NL4-3 (a gift from the Chris Aiken Laboratory, constructed by Hevey and Donehower) [59]. After determination of the concentration of viral particles by HIV-1 CA p24 ELISA, 300 ng of p24-equivalents of HIV-1 were spinoculated (300×g, 30 min) on 1×106 Th1, Th2, or TH2-A3G cells that had been activated by anti-CD3/CD28 coated beads (Invitrogen Dynal, Carlsbad, CA) for 60 hours [60]. Twelve hours after infection, cultures were washed twice with PBS. The cells were then resuspended in RPMI media containing 10 uM didanosine (Sigma, St. Louis, MO) and 25 uM zidovudine (Sigma, St. Louis, MO) to limit virus spread. After another 12 hrs in culture, supernatant fluids were collected for p24 antigen ELISA. Equal p24 concentrations of viral supernatant were then used to infect the TZM-bl indicator cells [48] and luciferase activity was determined in cell lysates 60 hours after infection (Bright-Glo Luciferase assay substrate, Promega, Madison, WI; TopCount scintillation counter, Packard/Perkin Elmer, Waltham, MA). Data are shown as RLU per nanogram p24 CA added.
|
10.1371/journal.pntd.0002232 | Seroprevalence of Entamoeba histolytica Infection among Chinese Men Who Have Sex with Men | Men who have sex with men (MSM) were found to be one of the high-risk populations for Entamoeba histolytica (E. histolytica) infection. Accompanied by the prevalence of human immunodeficiency virus (HIV) among MSM, invasive amebiasis caused by E. histolytica has been paid attention to as an opportunistic parasitic infection. However, the status of E. histolytica infection among MSM has been barely studied in mainland China.
Seroprevalance of E. histolytica was determined using an enzyme-linked immunosorbent assay based on a cross-sectional study conducted in Beijing and Tianjin, China. Factors potentially associated with E. histolytica infection were identified by logistic regression analysis.
A total of 602 MSM were included in the study and the laboratory data on serostatus of E. histolytica were available for 599 of them (99.5%). 246 (41.1%) and 51 (8.5%) of the study participants were E. histolytica seropositive and HIV seropositive, respectively. Univariate analyses suggested preferred anal sex behaviors were associated with E. histolytica seropositivity. In multivariate logistic regression analysis, “only has receptive anal sex” (OR: 2.03; 95% CI: 1.22 3.37), “majority receptive anal sex” (OR: 1.83; 95% CI: 1.13, 2.95), and “sadomasochistic behavior (SM)” (OR: 2.30; 95% CI: 1.04, 5.13) were found to be significantly associated with E. histolytica infection.
High seroprevalence of E. histolytica infection was observed among MSM from Beijing and Tianjin, China. Receptive anal sex behavior and SM were identified as potential predictors. Therefore, E. histolytica and HIV co-infection needs to be concerned among MSM due to their sharing the common risk behaviors.
| Entamoeba histolytica (E. histolytica) is a very common human gastrointestinal parasitic disease which affects 50 million people worldwide. Men who have sex with men (MSM) have already been found to be one of the high-risk populations with E. histolytica infection. Previous studies have reported an increased risk for E. histolytica infection and invasive amebiasis in HIV seropositive MSM. This pilot study aimed to investigate the serology of E. histolytica among MSM from mainland China. High prevalence of E. histolytica infection (41.1%, 246/599) was observed among the study population, receptive anal sex behavior and sadomasochistic behavior were found to be associated with the E. histolytica serostatus. Although HIV infection was not found to be associated with E. histolytica infection in this pilot study, studies from other countries had reported increased risks for E. histolytica infection and invasive amebiasis among HIV-positive MSM. Our findings suggest E. histolytica infection control needs to be concerned with respect to the increasing HIV prevalence among Chinese MSM population.
| Entamoeba histolytica (E. histolytica) has a worldwide distribution and is endemic in most developing countries. Invasive amebiasis (IA) caused by E. histolytica is a very common human gastrointestinal parasitic disease which affected 50 million people worldwide and caused greater than 100,000 deaths annually. High risk populations for developing IA include infants, travelers from endemic area, and patients who are taking immunosuppressant [1], [2]. In mainland China, E. histolytica infection was also very popular in general population. The average prevalence of E. histolytica infection was 0.95%, ranged from 0.01% to 8.12% [3].
In 1967, the association between amebiasis and homosexuality was suggested for the first time [4]. Men who have sex with men (MSM) population had already been found to be a high risk population with E. histolytica infection before 1990. Homosexuality and oral-anal sex have been most frequently reported as potential risk factors for E. Histolytica infection [5]–[13]. Accompanied by the transmission of human immunodeficiency virus (HIV) in MSM population, the prevalence of IA caused by E. histolytica are increasing and getting the attention as an important opportunistic parasitic infection. Recent studies from Australia, Japan, Korea and Taiwan reported increased risks for E. histolytica infection and IA among HIV-positive MSM [14]–[18]. Hung CC and colleagues recently reviewed the status of E. Histolytica infection in MSM [19]. By the end of 2011, China had about 780,000 people living with HIV/AIDS and 17.4% of them were MSM. The estimated new HIV infections in 2011 are 48,000 and 38.1% were MSM [20]. Case report of IA suggested the risk of E. histolytica prevalence among Chinese MSM, especially in the HIV positives [21]. However, to our knowledge, the prevalence of E. histolytica infection among MSM population has not been investigated in mainland China. The major aim of this study was to assess the seroprevalence of E. histolytica infection and the potential impact factors among MSM from China.
The study was approved by the Ethics Committees of the Institute of Pathogen Biology, Chinese Academy of Medical Sciences & Peking Union Medical College. All participants were the adults of men who had sex with men. Written informed consent was obtained before the interview and testing.
A cross-sectional study was conducted in Beijing and Tianjin, China. Six hundred study participants were recruited between March and July 2010, though local non-government organizations (Beijing Rainbow Volunteers Workstation and Tianjin Deep Blue Volunteers Workgroup). Participants' recruitment and inclusion criterion had been addressed in previous study [22].
Questionnaire administered by the trained interviewer were performed for each study participant, the data acquired from the questionnaire includes socio-demographic characteristics (e.g., age, income, ethnicity, education, employed, and marital status), sexual orientation and homosexual act, sexual behaviors in the past 6 months, history of STDs and HIV infection. Anilingus behaviors were defined as sexual stimulation involving oral contact with the anus and sadomasochistic behaviors (SM) were defined as behaviors which aimed to enhance sexual gratification from inflicting or submitting to physical and emotional abuse.
Blood samples were collected for E. histolytica and HIV serology. Serum samples were stored at −80°C until tested. Each study participant was assigned a unique identification number that was used to link the questionnaire and specimens. The HIV infection status was screened by an enzyme immunoassay (Wantai Biological Medicine Company, Beijing, China), and positive tests were confirmed by HIV-1/2 Western blot assay (HIVBlot 2.2 WB; Genelabs Diagnostics, Singapore). Qualitative screenings of serum immunoglobulin G (IgG) antibodies to E. histolytica were retrospectively performed using the commercial enzyme-linked immunosorbent assay (ELISA) kit (Shanghai Fengxiang Biological Technology Co. Ltd. Shanghai, China). Purified E. histolytica antigen was used to coat microtitration wells, incubated for 30 minutes at 37°C after adding 10 uL serum samples to wells. Washing and removing non-combinative antibody and other components, then combined HRP-conjugate reagent, then incubated and washed again. The substrate solution was added to each well, after 15 minutes at 37°C, stop solution was added to arrest color development and a ELISA reader was used to measure the absorbance at 450 nm. Each sample was tested in duplicate and the average optical density (OD) value was calculated. Test validity was evaluated as: the average of the positive controls should ≥1.00 and the average of negative controls should ≤0.10. The cut off value was set as the average of negative controls +0.15 according to the instruction of the kit.
Questionnaires were double entered and compared with EpiData software (EpiData 3.02 for Windows, The Epi Data Association Odense, Denmark). After cleaning, the data were then converted and analyzed using Statistical Analysis System (SAS 9.2 for Windows; SAS Institute Inc., NC, USA).
Study population was characterized by site with respect to age, ethnicity, education, marriage status, and current status of HIV. Differences between sites in these variables were assessed with Pearson chi-square test. The associations of E. histolytica infection with the characteristics of demographics, sexual behaviors, diagnosed STDs including HIV, and current status of HIV were estimated using Pearson chi-square test. Variables related with E. histolytica serology (p<0.1) in the univariate analyses were included in a multiple logistic regression model additionally adjusted for age, site, and HIV infection status. The Cochran-Armitage test was used to evaluate the trend of OD value in E. histolytica antibody test associated with anal sex behaviors categories.
A total of 607 participants were interviewed and signed the informed consent, 6 of them were excluded (3 did not complete sample collection and 2 Vietnamese living in Beijing). Finally, 602 were used for the analyses (302 from Beijing and 300 from Tianjin). Laboratory data on serostatus of E. histolytica and HIV were available for 599 (99.5%) and 598 (99.3%) of the study participants respectively. Age, ethnicity, education, marriage status, HIV-1 serostatus and E. histolytica serostatus of the study population were compared by site (Table 1). No significant difference was found for any characters (p>0.05). Therefore, participants from the two sites were pooled together for further association analyses. Age of the participants ranged from 16 to 72 years, with a mean age of 27.9±8.3 years. The participants showed the following characteristics: 95% (570/600) were Han nationality; 51.5% (309/600) had more than 12 year's education; 77.7% (467/601) were single. Laboratory data suggested two hundred and forty six (41.1%) and fifty one (8.5%) of the study participants were E. histolytica seropositive and HIV seropositive respectively. Twenty three participants were E. histolytica and HIV co-infected. Age, ethnicity, site, registered residence, education, marriage status, and alcohol use were not found to be related with E. histolytica serostatus (Table 2).
Homosexual men accounted for 71.0% (427/602), and bisexual men for 24.3% (146/602) of the study population. 73.6% (181/246) of participants with E. histolytica seropositive were homosexuals. Age of the first homosexual act ranged from 5 to 55 years old, with a mean age of 21.4±5.0 years. The median number of their homosexual partners was eleven and 43.1% of participants had stable homosexual partners before baseline survey. In the past one year, sixty one participants (10.2%) had group sex, thirty eight participants (6.3%) had ever received money for sex with male partners, and twenty three (3.8%) had ever provided money for sex with male partners. In the past 6 months, 133 (22.2%) participants reported had sexual behavior less than 1 time per month and only 1.3% (8/602) of participants insisted on using condoms in the process of insertive or receptive anal sex and oral sex. 22.5% of participants had ever been diagnosed sexually transmitted diseases. The association between sex behaviors and E. histolytica infection were also assessed. Self-reported preferred anal sex behaviors were classified to four types (only has insertive anal sex, majority insertive anal sex, majority receptive anal sex, and only has receptive anal sex). Univariate analyses suggested preferred anal sex behaviors were associated with E. histolytica infection. HIV infection was not found associated with E. histolytica infection (Table 3).
The variables associated with E. histolytica infection in the univariate analyses (P<0.1) were included in a multivariate logistic regression model. Age, site, and HIV infection status were fixed in the model as well. In the multivariate logistic regression model, only has receptive anal sex (OR: 2.03; 95% CI: 1.22, 3.37), majority receptive anal sex (OR: 1.83; 95% CI: 1.13, 2.95), and SM (OR: 2.30; 95% CI: 1.04, 5.13) were found to be significantly associated with E. histolytica infection after adjusted for the other variables (Table 4).
The OD in ELISA test was used for further analysis. Five hundred and eighty three participants reported their preferred anal sex behavior. As shown in figure 1, the OD values increased from the group of only had insertive anal sex (median OD, 0.036), majority insertive anal sex (median OD, 0.049), majority receptive anal sex (median OD, 0.117) to only had receptive anal sex (median OD, 0.131). The trend of increasing was significant (p<0.001), which was consistent with the result of multivariate analysis.
This pilot study investigated E. histolytica seroprevelance in MSM from China; potential factors associated with E. histolytica infection were evaluated as well. In a total of 602 study participants, E. histolytica seroprositivity was found to be 41.1%. Types of preferred anal sex behavior (only has receptive anal sex and majority receptive anal sex) and SM were identified as significant predictors for E. histolytica infection. In addition, significant different antibody levels were observed between subgroups with respect to the preferred anal sex behavior.
The first observation of a relation between enteric protozoan infections and sexual behavior was reported in 1968 [4]. Epidemiological studies conducted in the developed countries showed homosexuals or MSM had significant higher risk of E. histolytica infection. Using microscopy for diagnosis, the prevalence varied from 20% to 32% among MSM without gastrointestinal symptoms [5], [7], [8], [23]–[27]. However, microscopy is not sensitive or specific enough for the detection of E. histolytica in clinical specimens, especially for the differentiation E. histolytica from E dispar and E moshkovskii in the epidemiological studies with a large sample size. Therefore, serological tests were used to detect the infection though measuring anti-E. histolytica antibodies and seroprevalence ranging from 0.2% to 21% using ELISA test among HIV negative MSM were reported in several developed countries [18], [19]. Our results, for the first time, suggested a high prevalence (41.1%) of E. histolytica infection among MSM community from China. A recently published study conducted among general population in seven provinces in China showed that the seroprevalence of E. histolytica infection varied from 6% to 11% [28]. Although antibody test could not distinguish the past or current infection status and maybe overestimated the epidemic status, the fact that amebic liver abscess and latent infection had become one of the common opportunistic infection diseases among Chinese MSM AIDS patients reminds us to pay attention to the co-infection of E. histolytica and HIV. [21], [29].
In the present study, homosexual behaviors were mostly classified according to participants' tendency. Receptive anal sex behavior was found to be related to higher prevalence of E. histolytica infection. This finding and its underlying mechanisms should be further studied in the future. Homosexuals and history of anilingus had been demonstrated to be the risk factors of E. histolytica infection. In 1978, a study from the New York city reported that 20% of eighty nine sexually active homosexual men had amebiasis and the presence of infections associated with history of anilingus [5]. Another study from a venereal-disease-clinic population compared the prevalence of E. histolytica infections in homosexual men, bisexual men and heterosexual men. Homosexuality and oral-anal sex were found to be the most important risk factors for E. histolytica infection [7]. However, such an association was not observed in our study population. Interestingly, SM was found to be associated with E. histolytica infection in the present study. The data on the specific behaviors during the process of SM has not been well studied in China due to the potential issues of social culture and discrimination. Several published studies had revealed that people who had engaged in SM were more likely to have experienced oral-anal sex and other sexual risk practice [30]–[32]. In addition, fecal-oral contamination in these sexual behaviors maybe occurs and increases the opportunity of pathogen infection. Keystone JS' study showed cleaning of anus before anal sex was associated with a significant lower prevalence of infection [23]. But it is difficult to explore the factors linked SM behaviors to the infection susceptibility in our present study because only 30 participants (5.0%) reported SM behaviors. Developing targeted prevention and control strategies, such as developing sanitary habit before sexual behavior, may decrease the opportunity of pathogen infection. Therefore, it is necessary to further study potential risky behaviors associated with health problems among Chinese MSM.
Several limitations of this study should be kept in mind. First, potential bias due to the inaccurate response to the questionnaire, especially to the questions on sexual behaviors, could not be excluded completely. Second, our study participants might not represent the general MSM population due to the potential limitation of enrollment methods. Therefore, potential selection bias should be considered when interpret our results. Third, serology could not clearly identify the infection status as current infection or past infection, potential bias caused by such misclassification could not be excluded. Although statistically significant difference of antibodies levels was observed with respect to the preferred anal sex behaviors (p<0.001), however, the smaller sample size in each subgroups and a broad range of OD value should be considered. Further studies are needed to explore the underlying mechanisms for the observed relation between receptive anal sex behaviors and E. histolytica infection. Fourth, cross-sectional study design has its limitation on association analysis. Therefore, our results need confirmation by further large-scale case-control studies or prospective studies.
In conclusion, high prevalence of E. histolytica infection was observed among MSM from Beijing and Tianjin, China. Receptive anal sex behaviors and SM were identified as significant predictors for E. histolytica infection. Prevention and control of E. histolytica infection among Chinese MSM should be concerned because this special population confronted with high risk of HIV infection.
|
10.1371/journal.ppat.1001131 | Dengue Virus Ensures Its Fusion in Late Endosomes Using Compartment-Specific Lipids | Many enveloped viruses invade cells via endocytosis and use different environmental factors as triggers for virus-endosome fusion that delivers viral genome into cytosol. Intriguingly, dengue virus (DEN), the most prevalent mosquito-borne virus that infects up to 100 million people each year, fuses only in late endosomes, while activation of DEN protein fusogen glycoprotein E is triggered already at pH characteristic for early endosomes. Are there any cofactors that time DEN fusion to virion entry into late endosomes? Here we show that DEN utilizes bis(monoacylglycero)phosphate, a lipid specific to late endosomes, as a co-factor for its endosomal acidification-dependent fusion machinery. Effective virus fusion to plasma- and intracellular- membranes, as well as to protein-free liposomes, requires the target membrane to contain anionic lipids such as bis(monoacylglycero)phosphate and phosphatidylserine. Anionic lipids act downstream of low-pH-dependent fusion stages and promote the advance from the earliest hemifusion intermediates to the fusion pore opening. To reach anionic lipid-enriched late endosomes, DEN travels through acidified early endosomes, but we found that low pH-dependent loss of fusogenic properties of DEN is relatively slow in the presence of anionic lipid-free target membranes. We propose that anionic lipid-dependence of DEN fusion machinery protects it against premature irreversible restructuring and inactivation and ensures viral fusion in late endosomes, where the virus encounters anionic lipids for the first time during entry. Currently there are neither vaccines nor effective therapies for DEN, and the essential role of the newly identified DEN-bis(monoacylglycero)phosphate interactions in viral genome escape from the endosome suggests a novel target for drug design.
| Dengue virus infection is a growing public health problem with up to 100 million cases annually, and neither vaccines nor effective therapies are available. To search for the ways of preventing and treating dengue infections we need to better understand their molecular mechanisms. As with many other viruses, dengue virus enters cells by fusion between the viral membrane and the membrane of intracellular vesicles (endosomes). In this work we explored the fusion stage of dengue virus entry in different experimental systems ranging from virus fusion to artificial lipid membranes to fusion inside the cells. While earlier work on dengue virus entry has focused on viral protein that mediates fusion, we found that effective action of this protein requires specific lipid composition of the membrane the virus fuses to. In effect, this lipid dependence allows virus to control intracellular location of the fusion event and, thus, the place of its RNA release by exploiting cell-controlled differences between lipid compositions of different organelles the virus travels through. The essential role of the interactions between dengue virus and its lipid cofactors during viral entry suggests that these interactions may be targeted in drug design.
| With almost half of the world's population at risk for dengue infections, including life-threatening dengue hemorrhagic fever and dengue shock syndrome [1], the lack of vaccines and effective therapies lends urgency to the search for new targets for antiviral drugs. In this work we have focused on the membrane fusion stage of DEN infection. As many flaviviruses and alphaviruses, DEN enters mosquito and human cells by receptor-mediated endocytosis [2]–[5]. Fusion between the DEN envelope and the endosomal membrane is mediated by an envelope glycoprotein E [6]–[9] structurally similar to E protein of other flaviviruses and to E1 protein of alphaviruses such as Sindbis virus (SIN) [8], [10]. Acidification of endosomal content triggers a fusogenic restructuring of protein fusogens and fusion in flaviviruses and alvaviruses. DEN E undergoes a major conformational change that starts with the dissociation of the homodimeric form of E. Separated monomers rise from their initial positions parallel to the viral envelope to positions perpendicular to the envelope [6]–[9]. E monomers interact with the target membrane via their hydrophobic fusion loops and assemble into homotrimers that bridge the viral and target membranes. Subsequent refolding of the E trimer into its final hairpin structure, with the transmembrane domains and fusion loops at the same end of the rodlike molecules, bends the target and viral membranes towards each other and primes them for fusion [6], [8], [11]–[12].
Several observations have suggested that DEN fusion, in addition to low pH dependence, may involve as-yet-unidentified cofactors. Early research, confirmed in our preliminary experiments, found that low pH application to adjacent cells with DEN virions bound to the cell surface resulted in cell fusion for mosquito cells [13]–[15] but not for mammalian cells [13]. Furthermore, during viral entry into mammalian cells most DEN particles fuse only when they get to late endosomes/lysosomes [3]–[5], and, thus, they neither fuse nor become inactivated in early endosomes, which are expected to have pH low enough to trigger conformational changes in DEN E and induce DEN-mediated fusion between mosquito cells [15]. We suggested that in analogy to alphavirus fusion dependence on lipid cofactors (namely, cholesterol and sphyngomyelin) [16]–[19], the apparent differences in fusogenic properties of DEN towards different target membranes may reflect differences in their lipid composition. In contrast to the outer leaflets of plasma membranes of mammalian cells and to the inner leaflets of membranes of early endosomes, the outer leaflets of the plasma membranes of mosquito cells and the membranes of late endosomes of mammalian cells have high concentrations of the anionic lipids (AL) phosphatidylserine (PS) [20] and bis(monoacylglycero)phosphate (BMP) [21], respectively.
In this work, we explored the dependence of fusion of DEN, serotype 2, strain TH-36 (below referred to as “DEN”) on the lipid composition of the target membranes and found this fusion reaction to require the presence of AL. Liposomal, plasma, and intracellular membranes that did not support DEN fusion became fusion-competent upon addition of AL such as BMP and PS. The discovered dependence on AL was also observed for dengue virus serotype 2, strain New Guinea C and for dengue virus, serotype 4, strain H241. AL acted downstream of low pH-dependent stages of DEN fusion and promoted an advance of the fusion process beyond the earliest hemifusion intermediates. Endocytosed viral particles are acidified, and thus the restructuring of DEN E towards its fusogenic conformations is expected to be triggered before DEN comes into contact with the AL-containing membranes of late endosomes. For many viruses, including influenza virus [22] and SIN [17], premature activation of the viral fusion machinery by an acidic medium in the absence of a target membrane results in a quick inactivation of the machinery. In contrast, we found that DEN particles treated with a low-pH medium in the presence of AL-free target membranes retain their fusogenic properties for a relatively long time (>30 min). This unusually slow inactivation may explain the preservation of fusogenic properties of DEN for the duration of viral trafficking from early to late endosomes. We propose that the AL dependence and the delayed inactivation of the fusion machinery of DEN play an important role in defining the timing and location of the fusion stage of viral entry.
The fusogenic activity of diverse enveloped viruses is often characterized by measuring fusion between adjacent cells that is mediated by viral particles associated with the cell surface either through non-specific binding or through virus–receptor docking. For viruses that fuse in acidified endosomes, activation of viral fusogens and cell-to-cell fusion yielding syncytia is triggered by a short-term application of acidic pH. In agreement with earlier reports [13], DEN effectively fused mosquito cells C6/36 (Fig. 1A,B). Fig. 1B, curve 1 represents the pH dependence of DEN-mediated fusion of these cells and, as expected [15], already shows significant fusion at moderately acidic pH.
In striking distinction from insect cells, DEN produced in either insect or mammalian cells did not induce syncytium formation for any mammalian cells we tested (Vero, BHK21, CHO-K1, MA104, NIH 3T3, HAb2, BS-C-1, U967 (monocytes), and Raw (macrophages)) in the experiments in which cells pre-incubated with DEN were treated for 1 to 15 min with an acidic medium of pH values ranging from 4.9 to 6.5. Syncytium formation involves both the fusogen-dependent opening of fusion pores and their cell-machinery-dependent expansion, yielding a cell-size lumen [23]. We found that the absence of DEN-induced syncytia for mammalian cells reflects the inability of DEN to form fusion pores large enough to pass GFP and mRedFP, as evidenced by the lack of double-labeled cells (shown for Vero cells in Fig. 1C).
In parallel experiments, we contrasted the fusogenic activity of DEN with the well-characterized fusogenic activity of SIN, a virus that utilizes similar fusion machinery [8]. As expected, SIN effectively fused both mammalian cells and mosquito cells C6/36 (Fig. 1A; B, curve 2; and 1C; see also [24]).
We explored whether the inability of DEN to fuse mammalian cells can be explained by inefficient virus–cell-surface binding. To evaluate the binding (including non-specific receptor-independent binding), we incubated Vero cells (Fig. 1D) or CHO-K1 cells (not shown) with mildly biotinylated DEN or SIN particles, applied fluorescence-tagged streptavidin, and then carried out flow cytometry analysis (Fig. 1D). We found virus–cell binding for DEN to be somewhat higher than for SIN, the virus that effectively fuses the cells indicating that cell fusion for DEN is blocked downstream of cell-surface binding.
This conclusion was further substantiated by comparing DEN binding to C6/36 cells that were readily fusable by the virus with DEN binding to mammalian cells (MA104 cells) that did not fuse in the presence of DEN. In these experiments, we blocked internalization of DEN virions labeled with a fluorescent lipid DiD by low temperature (10°C) and quantified viral binding by measuring DiD fluorescence associated with the cells (Fig. S1). To take into account the very different sizes of C6/36 and MA104 cells, we normalized the amount of DiD fluorescence associated with the plasma membrane of the cells to the fluorescence of membrane probe NBD-tagged phosphatidylcholine (NBD-PC). This normalization is based on the assumption that NBD-PC similarly partitions into the outer leaflets of plasma membrane bilayers of C6/36 and MA104 cells and, thus, the cell-associated NBD fluorescence is proportional to the total area of plasma membranes. As shown in Fig. S1, under the conditions of our cell-to-cell fusion experiments on C6/36 cells and on MA104 cells, the ratios of DiD- and NBD- fluorescences, and thus the surface densities of DEN virions at plasma membrane, were very close. This finding argues against the hypothesis that DEN fuses C6/36 cells but does not fuse mammalian cells because of a large difference in the amounts of cell-bound virions.
To summarize, while DEN binds to and effectively infects both mosquito and mammalian cells, in our experiments it fused mosquito but not mammalian cells. The inability of DEN to fuse the plasma membranes of the virus-permissive mammalian cells indicated that the fusogenic activity of DEN depends on the target membrane. To identify the cofactors that have to be present to support DEN fusion, we concentrated on the unusual lipid composition of the late endosomal membranes that DEN fuses with to inject its RNA into cells. Using liposomes, we explored the dependence of DEN fusion on the composition of the target membrane.
To study the role of membrane lipid composition in DEN fusion, we measured low pH-induced lipid mixing between DEN particles labeled with a self-quenching concentration of a fluorescent lipid, DiD, and unlabeled liposomes of different composition. Only slow and inefficient lipid mixing between DEN and liposomes was observed for liposomes formed from lipids characteristic of the outer leaflet of the plasma membranes of mammalian cells (PM composition) (Fig. 2A,B). In contrast, we observed robust lipid mixing between DEN and liposomes containing phospholipids characteristic of late endosomal membranes (LEM composition). A much higher efficiency of DEN fusion to LEM liposomes than to PM liposomes was also observed for dengue virus serotype 4, strain H241 (Fig. S2) and for dengue virus, serotype 2, strain New Guinea C (not shown).
As expected [19], including Chol into the LEM lipid mixture (LEM-Chol composition) did not significantly change the initial rate or the extent of lipid mixing detected at t = 10 min (Fig. 2A,B). Inactivation of DEN particles by pre-incubation at pH 4.5 for 30 min (37°C) or by a histidine-modifying reagent diethylpyrocarbonate (DEPC) [25], resulted in a loss of DEN fusogenic properties (Fig. 2A). We also found that lipid mixing between DEN and LEM liposomes was inhibited by DEN monoclonal antibody 4G2 [26] (data not shown).
The inefficient fusion between DEN and liposomes containing lipids characteristic of plasma membranes contrasted with the robust fusogenic activity of SIN (Fig. S3). At pH 5.3, the initial rate and extent of lipid mixing with PM liposomes at t = 10 min for SIN were 10 fold and 3 fold higher than those for DEN. Strong lipid mixing between SIN and these liposomes is in agreement with earlier reports [17] and is also consistent with efficient fusion between alphaviruses and plasma membranes of mammalian cells (see above and [27]–[29]). Lipid mixing between SIN virions and PM liposomes developed slower than that between SIN and LEM-cholesterol liposomes but reached similar extents. Note that the rate of SIN-liposome lipid mixing detected in our experiments using the DiD dequenching assay is notably lower than that reported in [17], [30] and closer to the rates reported in [31]–[32]. This divergence most likely reflects the differences in assays used by different groups.
We hypothesized that LEM liposomes support DEN fusion because these liposomes contain an AL BMP, a specific lipid marker of multivesicular late endosomes that is present in different membrane domains of these organelles in concentrations between 20 and 70 mol% [21]. To test this hypothesis, we used liposomes of simpler compositions. While DEN did not fuse with liposomes that were formed from phosphatidylcholine (PC), liposomes that in addition to PC contained 30 mol % of either BMP or phosphatidylglycerol (PG) or PS supported lipid mixing (Fig. 2C), indicating that DEN fusion to liposomes depends on the negative charge of the target membrane rather than on a specific polar group of AL.
We then tested whether the inability of DEN to fuse mammalian cells can be explained by the lack of AL in the outer leaflet of the plasma membrane. We incubated CHO-K1 cells with the virus; then, after a wash, we placed the cells at 4°C and treated them with exogenous AL, PS. After a wash, we applied a low-pH medium at room temperature to trigger fusion. DEN effectively fused PS-treated cells but did not fuse untreated cells (Fig. 3A,B). While PS added to the outer leaflet of plasma membranes of mammalian cells undergoes inward transmembrane redistribution mediated by ATP-dependent aminophospholipid translocases [33], at the time of low pH application significant percentage of exogenous PS remained in the outer leaflet, as evidenced by its accessibility for extraction with bovine serum albumin, BSA (data not shown). Furthermore, DEN fusion of mammalian cells was similarly supported by addition of PG, an AL that is not an aminophospholipid (Fig. 3B). In control experiments, DEN did not mediate fusion between CHO-K1 cells if cell membranes were not supplemented with exogenous lipid or were supplemented with PC (Fig. 3B). In similar experiments carried out with SIN, we found cell fusion mediated by this virus to be almost unaffected by addition of either PS or PG (Fig. S4).
Since AL were applied only after removal of unbound DEN, the dramatic increase in fusion after PS treatment cannot be explained by a better DEN binding. We directly verified that cells treated and untreated with PS carried similar amounts of DEN particles in experiments with fluorescence-labeled viral particles. CHO-K1 cells were incubated with DiD-labeled virus at 10°C and washed. The temperature was lowered to 4°C and cells with associated virus were either treated or not treated with PS and then solubilized with 1% Triton X-100 at room temperature to dequench the DiD. The levels of fluorescence measured for samples from PS-treated and untreated cells were statistically indistinguishable.
As for many other enveloped viruses, fusion of DEN to the plasma membrane can be indirectly evaluated by measuring cell infection caused by low pH-induced fusion at the cell surface under conditions where acidification of the endosomes, essential for the biologically relevant entry pathway, is blocked [34]. We found the application of this fusion-infection assay (FIA) to DEN and mammalian cells (MA104, Vero and BHK21) to require a very high concentration of virions: 300 infectious units/cell (at least 10 times higher than the number of infectious viral particles per cell we had to use in FIA for mosquito C6/36 cells, data not shown). In this setting, FIA, while indirect, may be a very sensitive assay for fusion since fusion of a single virion out of hundreds may result in infection. As for DEN-mediated cell-to-cell fusion, the anionic lipids (PS or PG)-supplemented MA104 cells demonstrated much higher (15-fold) levels of fusion in FIA (Fig. 3C). Note that DEN-dependent fusion of PS-treated cells was still dependent on low pH. We observed a similar promotion of fusion-infection for Vero and BHK21 cells (data not shown).
The importance of AL for DEN-plasma membrane fusion was further confirmed in the experiments on mosquito cells. As mentioned above, C6/36 cells were reported to expose unusually high concentrations of AL at their surface [20]. In agreement with this study, we found a much higher cell-surface labeling with R-phycoerythrin -tagged annexin V for C6/36 than for MA104 cells (Fig. S5). Blocking PS and perhaps other AL [35] exposed at the surface of C6/36 cells with annexin V inhibited DEN-mediated fusion between these cells (Fig. 3D). While SIN-mediated fusion between C6/36 cells was also inhibited by annexin V perhaps because of a steric hindrance, inhibition of the DEN-fusion was much stronger. These findings support the hypothesis that the known ability of DEN to fuse C6/36 cells reflects the elevated concentrations of externalized AL in their membranes.
In brief, as with DEN-liposome fusion, the efficiency of DEN fusion to plasma membranes correlates with the accessibility of AL.
The effects of AL on fusion events during viral infection were studied in MA104 cells and BS-C-1 cells with pre-bound DEN labeled with DiD in a self-quenching concentration. Viral fusion events along the endocytic pathway diluted DiD and, thus, significantly increased cell fluorescence (Fig. 4A and S6). While first fusion events were observed less than 5 min after the rise in temperature to 37°C, most viral particles fused only at later times, with a median waiting time of ∼15 min (Fig. S6), consistent with an earlier study [3]. As expected, no intracellular fusion was observed when endosomal acidification was blocked by chloroquine or when DEN was DEPC-inactivated (Fig. 4A,B). In agreement with reports suggesting that DEN fusion reactions take place in late endosomes [3]–[5], we found that microtubule-depolymerizing nocodazole known to disrupt endosomal trafficking from early to late endosomes/lysosomes [36] inhibits intracellular fusion of DEN. When cells with pre-bound DEN were treated with AL (5-min, 4°C) before warming up to 37°C, we observed a dramatic increase in the average cell fluorescence (Fig. 4A,B). As for untreated cells, the increase in fluorescence of PS-treated cells was inhibited by chloroquine. The increase in DiD fluorescence reflects a higher efficiency of DEN–endosome fusion and suggests that in the AL-treated cells DEN fuses in early endosomes that normally would not have AL. This conclusion was further substantiated by the finding that intracellular fusion of DEN in AL-treated cells was much less sensitive to inhibition of endosomal trafficking by nocodazole than in untreated cells (Fig. 4B). Promotion of intracellular fusion for PS-supplemented cells was also observed for dengue virus of serotype 2, strain New Guinea C (not shown).
One may expect the delivery of DEN virions to late endosomal compartments that support viral fusion to be blocked by a dominant negative Rab7 (DN Rab7) that disrupts late endosomal/lysosomal biogenesis. Indeed, at least for some strains of DEN, DN Rab7 inhibits the infection of mammalian cells and DEN fusion [3]. As expected, we found that EGFP-tagged DN Rab7a S22N expression inhibits intracellular fusion of DiD-labeled DEN in MA104 cells, as evidenced by a much lower DiD fluorescence associated with DN Rab7-expressing cells detected by the EGFP fluorescence (Fig. S7A). Treating the cells with PS alleviated the DN Rab7 inhibition, as evidenced by the appearance of cells displaying both DiD- and EGFP- fluorescence (Fig. S7B). These findings further substantiate our hypothesis that DEN delays its fusion until entry into late endosomes because of the AL-enriched lipid composition of their membranes and does not fuse in early endosomes because these organelles normally do not have anionic lipids.
Since fusion is an early stage of DEN infection, promotion of DEN fusion in endosomal compartments may promote DEN infection. We extended our work from fusion assays to infection analysis and explored the effects of AL on the physiologically-relevant pathway of infection via endocytosis. A strong (∼16-fold) promotion of infection for cells supplemented with PS or PG (Fig. 4C) indicates that the AL dependence of DEN fusion results in a corresponding dependence of viral infection.
Conformational changes of DEN E and DEN fusion with insect cells are already triggered at the moderately acidic pH (∼pH 6.0) (Fig. 1B) that the endocytosed virus is expected to encounter in early endosomes. While premature activation of most viral fusogens inactivates them [37], inactivation of DEN has to be relatively slow to keep viral fusion machinery functional until the virus reaches late endosomes. Indeed, we found that more than half of DEN particles remained fusogenic towards LEM-liposomes after a 15-min pre-incubation at pH 5.5 either in the absence of the target membranes or in the presence of fusion-incompetent PC liposomes (Fig. 5A). Low pH exposure of DEN at the surface of MA104 cells resulted in similarly slow inhibition of subsequent intracellular fusion (Fig. 5B). Since most of the endocytotic cargo reaches late endosomes within 15 min [38], we concluded that the inactivation of DEN at acidic pH is slow enough to preserve the virus's fusogenic properties during its trafficking from early to late endosomes.
Which of the stages of the fusion pathway mediated by DEN E protein depends on the AL presence in the target membranes? Diverse fusion processes start with a local merger of contacting leaflets of two membranes, a stage referred to as hemifusion [39]. However, lipid flow through the earliest hemifusion intermediates may be restricted by the proteins surrounding the fusion site [39]. These ‘restricted hemifusion’ intermediates (RH) can be transformed into complete fusion with chlorpromazine (CPZ). Inverted-cone shaped CPZ preferentially partitions into the inner leaflets of cell membranes and breaks hemifusion structures composed solely of these leaflets [40]. To test whether DEN E forms RH intermediates, we used the experimental system of HAb2 cells with pre-bound human red blood cells (RBC) [41]–[42] utilized in earlier studies on RH [43]. HAb2 cells express HA0, an uncleaved form of influenza hemagglutinin that mediates very tight binding between HAb2 and RBC but is fusion-incompetent [43]. We allowed DEN to bind to HAb2 cells for 30 min at 10°C, then added RBCs labeled with a fluorescent lipid PKH26 and 15 min later washed out unbound virus and RBCs. Low pH application to HAb2–DEN–RBC complexes yielded no lipid mixing, unless the cells were pre-treated with PS (Fig. 6A). However, robust fusion was observed for cells not treated with PS when low pH pulse was followed by an immediate application of CPZ indicating that DEN-mediated HAb2-RBC fusion was blocked at the stage of RH intermediates.
The extents of lipid mixing gradually decreased when we extended the time interval between the end of low pH application and the CPZ pulse, indicating that RH formed by DEN E dissociated with time (Fig. 6B). This is similar to RH in fusion mediated by influenza and SIN viruses [43]. Likewise, 40% of cell complexes developed lipid mixing when PS was applied 5 min after the end of a 1-min pulse of pH 5.3, with much lower lipid mixing observed when PS was applied 30-min after the end of the low pH pulse. These findings indicated that pre-lipid-mixing fusion intermediates formed by low-pH-conformations of DEN E advanced to yield lipid mixing in the presence of AL independently of low-pH but dissociated with time in the absence of AL.
To summarize, DEN-AL interactions facilitate the transition from RH to more advanced fusion intermediates. The AL dependence of this transition distinguishes fusion machinery of DEN from that of SIN. As expected [43], SIN effectively mediated lipid mixing between RBC and HAb2 cells not treated with PS and CPZ application strongly promoted SIN-mediated fusion only for suboptimal pH showing that for pH≤5.8 most of the RH intermediates advanced to yield lipid mixing without CPZ application (Fig. S8).
This study developed from our analysis of the intriguing finding that while DEN effectively infects mammalian cells, it has a very low level of fusogenic activity towards the plasma membranes of these cells. We hypothesized that the late endosomal/lysosomal membranes that DEN fuses with during infection [3]–[5] contain cofactors for DEN fusion machinery missing in the plasma membrane. In our search for these cofactors, we have focused on the unusual lipid composition of the late endosomal membrane and found that the AL highly enriched in these membranes are indeed required for low-pH-dependent DEN fusion. Since in mammalian cells DEN comes into contact with AL-enriched membrane leaflets only in late endosomes, the AL dependence of DEN fusion likely determines the timing and the place of the release of the viral RNA into the cytosol.
All membrane fusion reactions are expected, and in many cases shown, to depend on the lipid composition of the membranes. Some dependencies are conserved among fusion processes driven by very diverse fusogens [39]. In contrast, the essential role of AL for DEN fusion is not conserved even within the same class (II) of viral protein fusogens. In spite of similarities between the structures of the fusogens utilized, another flavivirus, TBE, and alphaviruses such as SIN effectively fuse with AL-free target membranes [44]–[45]. Interestingly, although it is AL-independent, the fusion of alphaviruses including SIN does require the specific lipids cholesterol and sphingomyelin as cofactors for the protein machinery [46]–[48], and neither of these lipids is required for DEN fusion.
The specific mechanisms by which DEN E interactions with AL control E restructuring and fusion remain to be clarified. To start with, the local pH near a membrane containing AL is lower than the pH in the bulk of the solution because of the accumulation of protons near the negatively charged lipid headgroups. The changes in ion concentration at a charged surface are described by the Gouy-Chapman theory that yields the relation between the surface charge density, the electrostatic potential, the ionic strength of the solution and ion distributions near charged membranes [49]. An estimate based on the Gouy Chapman theory suggests that for a bilayer containing 30 mol% AL in 100 mM NaCl buffer, the apparent pH dependence of fusion may be shifted to a less acidic pH by up to 0.7 units. However, while DEN fusion with liposomes containing 30 mol% PS was already observed at pH 6.8, there was almost no lipid mixing for AL-free PC liposomes even at pH 4.5 (not shown). Thus, the AL-induced shift of local pH does not fully explain the role of AL as a prerequisite for DEN fusion. Note that pH dependences of fusion observed for dengue virus in different assays somewhat differ (see for instance, Fig. 1B and Fig. 2B). These differences likely reflect different numbers of activated fusion proteins required to reach detectable fusion stages for different target membranes.
In the fusion pathway mediated by several viruses, RH is followed by more-energy intensive stages that culminate in opening of an expanding fusion pore [39], [43]. While DEN forms the RH intermediates in the absence of AL, the later stages of DEN fusion require interactions between E protein and AL. In alphavirus fusion, the lipid cofactor cholesterol promotes the insertion of E1 fusion loops into the target membrane and formation of functional E1 homotrimers [47]. In analogy to this mechanism, we propose that the completion of DEN fusion requires E assembly into stable trimers dependent on interactions between the fusion loop of DEN E and AL in the target membrane (Fig. 7). Indeed, for a synthetic peptide representing the fusion loop region of DEN peptide, insertion into the lipid bilayer and peptide–peptide interactions at the bilayer surface are promoted in the presence of AL [50].
To facilitate the development of antivirals targeting the fusion stage of DEN entry it is important to have a reliable quantitative approach that may be used in a high throughput in vitro screening. In the most recent studies [11], [51], DEN fusion to liposomes has been detected by exposure of the viral core protein to liposome-encapsulated trypsin. A successful application of this approach (first developed for another virus in [52]) to DEN is an important advance. However this approach is difficult to quantify and still needs to be validated by excluding the possible role of leakages [52] that may accompany DEN-target membrane interactions at acidic pH [51]. For many viruses, including SIN and some flaviviruses, fusogenic activity can be conveniently monitored by measuring lipid mixing between viruses and liposomes [16]–[17], [19], [30]–[32], [44], [52]–[55]. However, development of a lipid mixing assay for DEN fusion has proved to be surprisingly challenging [56]. Our work explains the very low efficiency of lipid mixing between DEN virions and AL-free liposomes [11] and describes a simple fluorescence dequenching assay of fusogenic activity of DEN towards AL-containing liposomes.
In addition to a lipid mixing assay that characterizes fusion between DiD-labeled DEN and liposomes by measuring DiD dequenching, we also used DiD-labeled virions to develop the intracellular DEN fusion assay. Our assay is based on earlier work [3], [57] that elegantly explored the DEN entry pathway in living cells by following a single DiD-labeled DEN particle using real-time fluorescence microscopy. In our approach, instead of detailed characterization of the time course and localization of the DEN fusion events for ∼50–100 virions achieved in [3], [57], we have focused on developing a much simpler approach characterizing intracellular DEN fusion as averaged over thousands of virions and cells. We expect our quantitative assays of DEN fusion within cells and with liposomes to be of help in screening for potential anti-DEN drugs.
Viruses have developed different strategies to prevent the premature release of the conformational energy of their protein fusogens that would result in irreversible “discharge” or in undesirable fusion. For many viruses, including DEN, which utilize low pH-dependent fusogen proteins, these proteins are synthesized in an inactive form and then are converted to a mature fusion-competent form by proteolytic cleavage of the fusogen itself or an accessory protein [6]–[7], [10]. However, viral fusogens such as DEN E that can be activated at pH values close to neutral may need additional mechanisms for avoiding premature release of the conformational energy stored in the proteins. Our finding that DEN neither fuses nor rapidly inactivates in the absence of AL-containing target membranes suggests that AL dependence of DEN prevents functional inactivation of the virus until it reaches AL-enriched late endosomes (Fig. 7). RH connections provide an additional receptor-independent docking mechanism that holds viral and endosomal membranes in tight contact. Within multivesicular endosomes, the virus can fuse either to the limiting membrane or to the internal vesicles that contain the highest concentration of AL BMP [21]. In the latter case, viral RNA release requires an additional fusion event: BMP-dependent but likely DEN E-independent back-fusion between the internal vesicle and the limiting membrane. Back-fusion has been proposed as a mechanism for endosome-to-cytosol transport of RNA of vesicular stomatatis virus ([58], but see [59]).
Our work concentrates on fusion mediated by DEN serotype 2, strain TH-36. While experiments on virus/liposome lipid mixing for dengue virus of serotype 4, strain H241 (Fig. S2) and experiments on virus/liposome lipid mixing and intracellular fusion for dengue virus of serotype 2, strain New Guinea C suggest that these viruses have a similar dependence on AL, DEN virions of different serotypes and strains may use alternative pathways for their entry into mammalian cells [60]–[61]. Future research will clarify the applicability of the AL-dependent mechanism of timing viral fusion to the entry into late endosomes for different serotypes/strains of DEN.
To summarize, the AL dependence of DEN fusion identified in our work suggests a novel mechanism allowing viruses to exploit cell-controlled changes in membrane lipid composition. In this mechanism, internalized virus uses the specific lipid composition of late endosomes highly enriched in AL as a way of timing fusion to deliver viral RNA to its translation-replication sites effectively. We hope the assays developed in this study to directly characterize DEN fusion will help in developing antivirals including those targeting DEN-AL interactions to block or prematurely activate DEN E refolding. Furthermore, interactions between DEN E and AL may have implications for the pathogenesis of dengue hemorrhagic fever, which is characterized by activation of endothelial cells and extracellular exposure of PS.
Vero, MA104, BHK21, BS-C-1, CHO-K1, RAW 264.7, U967, and NIH3T3 cells (American Type Culture Collection (ATCC), Manassas, VA), and HAb2 cells (a kind gift of Dr. Judith White, University of Virginia), a line of NIH 3T3 cells stably expressing A/Japan/305/57 influenza hemagglutinin (HA) in the immature fusion-incompetent HA0 form [41] were grown in Advanced DMEM medium (ADMEM) supplemented with 10% fetal bovine serum, 25 mM HEPES, 2 mM glutamine, and antibiotics (complete medium) at 37°C and 5% CO2. Aedes albopictus C6/36 (American Type Culture Collection, Rockville, MD) were cultured in the complete medium at 28°C and 5% CO2. To facilitate detection of cell fusion events by coplating differently labeled cells, we developed Vero, CHO-K1, and BHK21 cells stably expressing either EGFP or mRedFP according to the standard procedure using pEGFP and pmRFP plasmids (kind gifts of Dr. Eugene Zaitsev, NIH, NICHD).
Human red blood cells (RBC) were freshly isolated from whole blood obtained from the National Institutes of Health (Bethesda, MD) blood bank and labeled with a fluorescent lipid PKH26 (Sigma, St. Louis, MO), as described in [62]–[63].
If not stated otherwise, we used dengue virus of serotype 2, strain TH-36. In some experiments we used dengue virus of serotype 2, strain New Guinea C and dengue virus of serotype 4, strain H241. All viruses were purchased from ATCC. We propagated DEN by inoculating monolayers of C6/36 cells or Vero cells in complete medium in the presence of 0.005% Pluronic F-127 (Invitrogen, Eugene, OR) at a multiplicity of infection (MOI) of 0.1. DEN particles released from the cells were harvested 5 days postinfection and cleared from cell debris by means of low-speed centrifugation. Working virus stocks with titers between 107 and 108 infectious units (IU)/ml were kept at 4°C for less than two weeks. SIN strain AR-339 (ATCC) was used for infection of Vero or C6/36 cells with a low multiplicity of infection to propagate the working virus stocks with titer between 108–109 IU/ml. Virus was collected 24-hours post-infection. DEN and SIN were titrated for Vero cells using a fluorescent focus assay.
Viruses were concentrated by overnight centrifugation (SW28 rotor, 21 K rpm, 4°C) on a 55% cushion of Optiprep Density Gradient medium (Sigma, St. Louis, MO) buffered with 20 mM Tricine-HCl and 140 mM NaCl, pH 7.8 (buffer TrN) supplemented with 0.005% Pluronic F-127.
To label viral particles with a self-quenching concentration of a fluorescent lipid, we mixed 10 µl of a 1 mM DiD solution from a Vybrant cell-labeling kit (Molecular Probes, Eugene, OR) with 1 µl of 10% Pluronic F-127 and bath-sonicated the dispersion for 5 min. Freshly prepared DiD dispersion was injected into 1 ml of virus stock (approximately 4×108 IU) under intensive vortexing. The mix was incubated for 30 min at room temperature and then for 2 hr at 4°C. Labeled virus was purified from unincorporated dye and from non-viral membranes and proteins by centrifugation (SW55 rotor, 1.5 h, 53 K rpm, 4°C) on a 40%–25%–20%–15% step gradient of Optipep density medium in TrN supplemented with 0.005% Pluronic F-127. We collected the band between 20% and 25% densities containing DiD-labeled virus. BSA (final concentration 1%) was added to stabilize the preparation. Labeled virus was used within 3 days. Before experiments, the viral suspension was passed through a PES Millipore 0.22 µm filter to remove viral aggregates. In fluorescence microscopy examination, the number of DiD-labeled spots practically coincided with the number of fluorescent spots observed when viral particles (DEN or SIN) were visualized by immunofluorescence with DEN antibody (MAB 8705) or SIN antibody (Sindbis Hyperimmune Ascetic Fluid). DiD-labeled virus retained infectivity, but with an approximately two-fold titer decrease.
All lipids used: Chol, PC (1,2-dioleoyl-sn-glycero-3-phosphocholine); PE (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine); SPM (N-oleoyl-D-erythro-sphingosylphosphorylcholine); PI (L-α-phosphatidylinositol, Soy); BMP (bis(monooleoylglycero)phosphate, S,R Isomer); PS (1,2-dioleoyl-sn-glycero-3-phospho-L-serine); and PG (1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol)) were purchased from Avanti Polar Lipids, Alabaster, AL. Liposomes were formed from the following lipid mixtures: PM composition: PC, phosphatidylethanolamine (PE), sphingomyelin (SPM), cholesterol (Chol) in a molar ratio of 4/1/0.5/4.5; LEM composition, PC/PE/phosphatidylinositol (PI)/BMP in a molar ratio of 5/2/1/2; LEM-Chol composition, PC/PE/PI/BMP/Chol in a molar ratio of 4/1/1/2/2) and from binary mixtures of PC with either BMP, PG, or PS in a 7/3 molar ratio. To form large unilamellar liposomes, lipid mixtures dissolved in benzene/methanol (95:5) were frozen in liquid nitrogen and then freeze-dried under vacuum overnight. Dry lipid powder was re-suspended in TrN buffer by vigorous vortexing. We subjected the lipid suspension to 10 freeze-thaw cycles by alternating immersion into liquid nitrogen and hot water. Finally, we extruded the lipid suspension 10 times through double-stacked track-etched 100 nm-pore polycarbonate filters (GE Osmonics) using a LIPEX extruder. Liposome sizes were checked using dynamic light scattering on N4 plus submicron particle size analyzer (Beckman Coulter, USA). Liposomes were kept on ice and used within a day.
In all experiments that involved virus-to-cell binding we incubated cells with viruses at 10°C. We found this temperature to be optimal for DEN binding because at lower temperatures (for instance, at 4°C) viral binding was very inefficient, and at higher temperatures we observed internalization of viral particles by cells.
To compare cell binding for DEN with that for SIN, we surface-biotinylated DEN and SIN particles by incubating viruses (106 IU) with 10 mM EZ-Link Sulfo-NHS-SS-Biotin (Pierce Biotechnology, Rockford, IL) in 0.5 mL phosphate buffer solution (PBS) supplemented with 20 mM Tricine, pH 7.8 (PBS-T buffer) for 30 min at room temperature. The reaction was quenched with 100 mM glycine in PBS. Biotinylated virus (105 IU of DEN or SIN) was allowed to bind to lifted CHO-K1 cells (104 cells) at 10°C for 30 min. Cells were washed three times with cold serum-free ADMEM medium supplemented with 1% BSA by pelleting at 4°C. After removing the unbound virus, we incubated the cells with streptavidin Alexa Fluor 488 conjugate (Molecular Probes, Eugene, OR) as recommended by the manufacturer and carried out flow cytometry analysis using a FACSCanto fluorescence-activated cell sorter (BD Biosciences).
In another experimental approach, we incubated the C6/36 and MA104 cells with DiD-labeled virus in 35 mm plates at 10°C for 1 hr in the ADMEM medium. Unbound viral particles were removed by washing, treated with 2.5 µM NBD-PC at 4°C for 5 min. The cells were washed with cold PBS. Then the cells were lysed and DiD fluorescence dequenched by replacing PBS with 200 µl of PBC supplemented with 1% Triton X100. Samples were cleared from debris by centrifugation. C6/36 cells are much smaller than MA104 cells. To compare surface densities of bound DEN particles at plasma membrane of C6/36 and MA104 cells, we normalized the DiD fluorescence that provides a measure of the amount of bound virus to the NBD fluorescence that provides a measure of the total area of plasma membranes accessible for NBD-PC insertion. This normalization is based on the assumption that NBD-PC similarly partitions into the outer leaflets of plasma membrane bilayers of different cells.
The effects of AL on virus infection were quantified in MA104, Vero and BHK21 cells by fluorescent focus assay. Briefly, serial dilutions of DEN in the ADMEM medium were incubated with confluent monolayers of the cells at 10°C for 60 min. After removing unbound viruses, the cells were treated with 2.5 µM of PS or PG in the ADMEM medium for 5 min at 4°C. The cells were incubated at 37°C for 1 hour in the ADMEM medium, then overlaid with the medium containing 2% FBS and supplemented with 0.75% carboxymethyl-cellulose. The cells were grown for 3 days at 37°C and fluorescent focus units were detected by immunostaining with the primary monoclonal antibody 4G2 and fluorescent secondary antibodies against mouse IgG. Data were normalized to control infection observed for the cells not treated with any exogenous lipids.
Cells grown to high confluency were incubated with DEN or SIN in complete medium for 30 min at 10°C. Unbound virus was removed by three washes with cold PBS-T, and fusion was triggered at room temperature by application of serum-free ADMEM medium adjusted to different pH values with MES and acetic acid. In unsuccessful attempts to achieve DEN-mediated fusion of mammalian cells, we increased the MOI to 1,000 and extended the duration of the low-pH application to 15 min.
To detect virus-mediated fusion between mammalian cells by redistribution of aqueous contents, we co–plated cells expressing either EGFP or mRedFP at a 1∶1 ratio. A day later, we incubated the cells with virions (DEN, MOI of 300 or SIN, MOI of 40) as described above, applied a 5-min low-pH pulse, and incubated cells for 30 more minutes in the complete medium at 37°C. The average number of co-labeled cells per microscopic field was normalized to the average number of contacts between differently labeled cells in the control experiment, in which cells were not treated with low pH. For each condition, we carried out 3 independent experiments and analyzed at least 10 microscopic fields in each experiment.
To score fusion between mosquito cells C6/36 with bound DEN (MOI of 100) or SIN (MOI of 40) virions we treated the cells with media of different pH for 15 min and then re-neutralized the cells. After a 2-hour incubation at 37°C in complete medium we quantified the efficiency of syncytium formation by measuring a decrease in the number of mononucleated cells. More than 10 fields of view were analyzed for each experimental condition.
In the experiments on virus-mediated HAb2-RBC fusion, cells with associated virions were incubated with PKH26-labeled RBCs to achieve 0–2 bound RBC per cell [62]. After three washes with PBS-T to remove unbound RBC, HAB2–virion–RBC complexes were treated with PBS titrated with citrate to an acidic pH for 5 min and then re-neutralized with PBS-T. Fusion was quantified as the ratio of dye-redistributed bound RBC to the total number of bound RBC.
While the inability of DEN to fuse mammalian cells is illustrated in the figures only for CHO-K1 cells and Vero cells and HAb2-RBC fusion, we carried out similar experiments and observed no DEN-mediated fusion for MA104, BHK21, BS-C-1, RAW 264.7, U967, and NIH3T3 cells.
Fusion of DEN virions to the plasma membrane of MA104 cells, Vero and BHK-21 cells was evaluated by a fusion-infection assay (FIA) [34]. This assay is based on measuring infection caused by low-pH-induced fusion between viral particles and plasma membrane under conditions when endocytotic entry of virus is blocked by inhibitors of endosomal acidification. We plated the cells on Lab-tek II 8-well chambered coverglass (Nalge Nunc) a day before experiment, then pre-treated them with 50 µM chloroquine (Sigma, St. Louis, MO) for 30 min at 37°C. Viral particles were allowed to bind to the cell surfaces at 10°C for 60 min (MOI = 300). After removal of unbound virus, the cells were treated (or not treated) with 2.5 µM of PS or PG in the ADMEM medium for 5 min at 4°C. Fusion was triggered at room temperature by replacing the medium with the ADMEM medium acidified to pH 5.3. 5 min later the cells were neutralized, incubated at 37°C for 4 hours in the ADMEM medium supplemented with chloroquine. By that time, virions that did not infect the cells by low pH-induced fusion to plasma membrane, have been internalized and, because of the blocked endosomal acidification and, thus, fusion, have already passed the endosomal compartments allowing productive RNA release and infection. After this 4 hour incubation in the presence of chloroquine, we overlaid the cells with the ADMEM medium supplemented with 2% FBS and 0.75% carboxymethyl cellulose to prevent virus spread. Fluorescent focus units were detected 3 days later by immunostaining with the antibody 4G2. Data were normalized to those obtained for the cells that were not treated with AL. We found FIA assay for DEN and mammalian cells to be very sensitive to the concentration of chloroquine (and other inhibitors of endosomal acidification) and the duration of the application of these inhibitors (longer applications results in cytotoxity) and to MOI used.
MA104 and BS-C-1 cells were grown on the coverglass bottom of 35-mm tissue culture dishes (MatTek, MA) to high confluency. DiD-labeled virus was added to cells (MOI of 100) and allowed to bind for 30 min at 10°C. Unbound virus was removed by three washes with cold PBS-T. The cells were warmed up to 37°C to allow virus internalization. After incubation of cells at 37°C for different times, the cells were fixed with 4% paraformaldehyde and analyzed by fluorescence microscopy. Fusion of DiD labeled virus within endosomes leads to dequenching of DiD and appearance of bright fluorescent spots throughout the whole cell but mostly in the perinuclear region. To quantify fusion efficiency, we imaged cells with an iXonEM+ 885 EMCCD Camera (Andor Technology, CT) on an AxiObserver inverted fluorescence microscope (Zeiss, Germany) using a Cy5-4040A filter set (Semrock, NY). To capture signal from all fused viruses, we collected image stacks throughout the cell with 250 nm spacing between slices. We analyzed the acquired images using an ImageJ macro developed in-house to measure the total fluorescence signal from bright spots per imaging field (20 fields per experimental condition were analyzed in each independent experiment; each experiment was repeated at least three times).
We assayed lipid mixing between DiD labeled viral particles and liposomes as DiD dequenching in four-clear-sided methacrylate cuvettes (Fisher Scientific, Pittsburgh, PA). The medium in the cuvettes was continuously stirred with a magnetic stirring device and thermostatted at 37°C. We mixed 10 µl (∼105 IU) of purified labeled virus with liposomes (final concentration of lipid 30 µM) in 2 ml of TrN buffer. We initiated the fusion reaction by adding a pre-titrated amount of MES/acetic acid buffer to reach the desired pH. We recorded fluorescence for at least 15 min at excitation and emission wavelengths of 620 and 665 nm, respectively, using an Aminco Bowman Series 2 luminescence spectrometer (Rochester, NY). At the end of each recording, we added Triton X-100 to a final concentration of 0.1% to fully dequench DiD (“100% lipid mixing”). We routinely verified that under our conditions the efficiency of lipid mixing (rates and extents) was not limited by the concentration of the liposomes used.
DEN was inactivated by a 15-min incubation at room temperature in PBS-T supplemented with 2 mM DEPC (Sigma, St. Louis, MO) added from freshly prepared stock solution in cold ethanol. DEPC was reported to inhibit vesicular stomatatis virus by modifying histidine residues on viral protein fusogen [25].
To add exogenous lipids (PS or PC or PG) to plasma membranes of CHO-K1, MA104, BS-C-1 cells and HAb2–RBC pairs, we incubated the cells with associated virions in a PBS supplemented with 16:0-06:0 NBD PS (1-palmitoyl-2-{6-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]hexanoyl}-sn-glycero-3-phosphoserine), 16:0-06:0 NBD PC (1-palmitoyl-2-{6-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]hexanoyl}-sn-glycero-3-phosphocholine) or 16:0-06:0 NBD PG (1-palmitoyl-2-{6-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]hexanoyl}-sn-glycero-3-[phospho-rac-(1-glycerol)]), all lipids purchased from Avanti Polar Lipids, Alabaster, AL). A 2.5 mM stock solution (1 µl) of PS, PC or PG in ethanol was injected into 1 ml of PBS-T under intensive vortexing. The cells were cooled down to 4°C and placed into this lipid-supplemented medium for 5 min, still at 4°C. We inferred from the levels of cell-associated NBD fluorescence observed with fluorescence microscopy that PS, PC and PG incorporated into cell membranes to similar concentrations (see also [64]). For virus-mediated cell–cell fusion, the lipid-supplemented buffer was replaced with warm (room temperature) serum-free ADMEM medium adjusted to different pH values by titration with MES and acetic acid. After the end of low pH application the cells were incubated at 37°C for 30 min and fusion was quantified by fluorescence microscopy. We verified that at the time of low pH application a significant part of exogenous lipids remain extractable by delipidated BSA. For the intracellular fusion assay, the cells with bound DiD-labeled DEN or SIN virions were treated or not treated with exogenous lipids as described above. The temperature was raised and, after incubation at 37°C for different times, we fixed the cells and analyzed them by fluorescence microscopy.
To reveal DEN-mediated RH between HAb2 cells and PKH26-labeled RBCs, a 5-min low pH pulse was followed by a 1-min application of a 0.5 mM solution of CPZ (Sigma, St. Louis, MO) in PBS-T. The percentage of HAb2–RBC pairs demonstrating lipid mixing (PKH redistribution from RBC to HAb2 cell) was assayed with fluorescence microscopy 20 min after the end of the low–pH pulse.
Annexin V is widely used to evaluate the expression of PS on cell surfaces [65]. We used this protein in two different experimental approaches. To inhibit DEN interactions with PS in the outer leaflet of plasma membranes of C6/36 cells, we first incubated the cells in annexin-binding buffer (BD Pharmingen, San Jose, CA) at room temperature for 30 min and then allowed DEN or SIN virions to bind to the cells. After removal of unbound virions, we incubated the cells with 50 µg/ml recombinant annexin V (BD Pharmingen, San Jose, CA) for 30 min at 10°C, washed the cells from unbound annexin, and then triggered fusion by applying an acidic pH medium. We verified that annexin V treatment had no effect on virus–cell binding using DiD-labeled virions.
To compare PS expression at the surfaces of different cells, we used R-phycoerythrin -tagged annexin V (BD Pharmingen, San Jose, CA), as recommended by the manufacturer.
In some experiments on intracellular fusion, to block endosomal acidification we treated the cells with the lysosomotropic agents chloroquine (Sigma, St. Louis, MO; 50 µM, 30 min, 37°C) or bafilomycin-A1 (Sigma, St. Louis, MO; 2 µM, 30 min, 37°C) prior to applying viral inoculum.
Cells were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions with a EGFP-Rab7a S22N plasmid [66], a kind gift from Julie Donaldson, NIH, on Lab-tek II 4-well chambered coverglass (Nalge Nunc). 18 hours later, DiD-labeled viral particles were allowed to bind to the cell surfaces at 10°C for 60 min. After removal of unbound virus, cells were treated (or not treated) with 2.5 µM of PS in the ADMEM medium for 5 min at 4°C and then incubated in the ADMEM medium at 37°C for 40 min. Cells were fixed and nuclei were stained with DAPI (Invitrogen). Analysis by fluorescence microscopy allowed us to identify the transfected cells by their EGFP fluorescence and to detect DEN fusion within endosomal pathway as intracellular structures displaying DiD fluorescence.
MA104 cells were preincubated with nocodazole (Sigma, St. Louis, MO; 60 µM in complete medium, 30 min, 37°C) prior to application of DEN and exogenous lipids. Nocodazole was present throughout the entire experiment.
|
10.1371/journal.pntd.0002435 | Hospital-Based Surveillance for Viral Hemorrhagic Fevers and Hepatitides in Ghana | Viral hemorrhagic fevers (VHF) are acute diseases associated with bleeding, organ failure, and shock. VHF may hardly be distinguished clinically from other diseases in the African hospital, including viral hepatitis. This study was conducted to determine if VHF and viral hepatitis contribute to hospital morbidity in the Central and Northern parts of Ghana.
From 2009 to 2011, blood samples of 258 patients with VHF symptoms were collected at 18 hospitals in Ashanti, Brong-Ahafo, Northern, Upper West, and Upper East regions. Patients were tested by PCR for Lassa, Rift Valley, Crimean-Congo, Ebola/Marburg, and yellow fever viruses; hepatitis A (HAV), B (HBV), C (HCV), and E (HEV) viruses; and by ELISA for serological hepatitis markers. None of the patients tested positive for VHF. However, 21 (8.1%) showed anti-HBc IgM plus HBV DNA and/or HBsAg; 37 (14%) showed HBsAg and HBV DNA without anti-HBc IgM; 26 (10%) showed anti-HAV IgM and/or HAV RNA; and 20 (7.8%) were HCV RNA-positive. None was positive for HEV RNA or anti-HEV IgM plus IgG. Viral genotypes were determined as HAV-IB, HBV-A and E, and HCV-1, 2, and 4.
VHFs do not cause significant hospital morbidity in the study area. However, the incidence of acute hepatitis A and B, and hepatitis B and C with active virus replication is high. These infections may mimic VHF and need to be considered if VHF is suspected. The data may help decision makers to allocate resources and focus surveillance systems on the diseases of relevance in Ghana.
| Ghana is endemic for yellow fever and lies between two Lassa fever endemic areas — Guinea, Liberia, Sierra Leone, and Mali in the West, and Nigeria in the East. Ebola hemorrhagic fever has been documented in the neighboring Cote d'Ivoire. Thus, it is plausible that the latter VHFs also occur in Ghana, and there have been rumors of cases, which present like VHF in the north of the country. Our study aimed at verifying or disproving this suspicion. At 18 hospital-based study sites in the Central and Northern part of the country, samples from 258 patients with VHF symptoms were collected and tested for various VHF by PCR. As viral hepatitis is an important differential diagnosis of yellow fever, we also tested for several serological and molecular hepatitis markers. Rather surprisingly, VHFs were not detected, indicating that, even if they are endemic in the North of Ghana, they do not significantly contribute to hospital morbidity. However, a large fraction of patients showed markers of acute hepatitis A, and active hepatitis B and C. Children were mainly affected by hepatitis A, while adults were affected by hepatitis B and C. Hepatitis A and B are vaccine-preventable, and chronic hepatitis B and C are treatable diseases. Further efforts are needed to reduce the burden of these diseases in Ghana.
| Viral hemorrhagic fevers (VHF) are acute viral diseases associated with bleeding, organ failure, and shock. The syndrome is caused by RNA viruses belonging to the families Filoviridae (Ebola and Marburg virus), Arenaviridae (Lassa virus), Bunyaviridae (Crimean-Congo hemorrhagic fever [CCHF] and Rift Valley fever [RVF] virus), and Flaviviridae (yellow fever [YF] virus). The case fatality rate depends on the causative virus and may be as high as 90% [1].
Several VHFs are endemic in West Africa, such as Lassa fever [2], Ebola hemorrhagic fever [3], CCHF [4], [5], RVF [6], and YF [7], [8]. Within the region, Ebola virus infection was documented so far only once in Cote d'Ivoire [3]. Lassa fever is endemic in the countries of Guinea, Sierra Leone, Liberia, Mali, and Nigeria [2], but has not been documented in Ghana. YF is endemic in Ghana [7], [8]. Since 1950, three major outbreaks — in 1969–70, 1977–80, and 1982–83 — affected the country and caused more than 400 deaths [9]. Whether other VHFs are endemic in Ghana, is not known. The list of differential diagnoses is long, because clinically, VHF is not easily distinguished from other febrile diseases in Africa. In particular, liver damage due to viral hepatitis may hardly be distinguished from YF. The presence of hepatitis B, C, and E virus in Ghana is documented by seroprevalence studies [10]–[19], while there is no published data on hepatitis A.
For several years, there have been anecdotic reports of cases presenting with VHF symptoms in the north of the country. The scarcity of reliable data on suspect cases is in part due to the lack of diagnostic tools and active surveillance systems. Therefore, we established PCR diagnostics for VHF at Noguchi Institute and conducted a hospital-based surveillance study to determine the etiology of illnesses presenting with VHF symptoms in the north of Ghana. In addition, viral hepatitis being an important differential diagnosis of hemorrhagic fevers was included in the study.
The study was approved by the Institutional Review Board of the Noguchi Memorial Institute of Medical Research (NMIMR-IRB 003/07-08). All subjects provided written informed consent.
The study was carried out from 2008 through 2011 at 18 hospitals in the Ashanti, Brong-Ahafo, Northern, Upper West, and Upper East regions in the Central and Northern sectors of Ghana (Fig. 1). In total, 285 patients were enrolled during the study period, of whom the first 27 patients were tested in a pilot activity on a reduced set of parameters. There was no pre-defined study group size. VHF patients often do not bleed, may present with a multitude of symptoms including jaundice, encephalopathy, and renal failure, and may be normo- or hypothermic in the terminal stage [1], [8], [20]–[23]. Therefore, a broad case definition was designed as a guideline for the local clinicians and health staff to screen and enroll the patients in the study sites. Criteria for including patients were severe illness with fever or history of fever and at least one of the following conditions: hemorrhage, jaundice, encephalopathy, renal involvement, absence of malaria, and lack of response to antibiotics and antimalarials. These criteria also include symptoms of severe liver disease due to viral hepatitis A, B, C, and E (jaundice, hepatic encephalopathy, bleeding due to impaired synthesis of coagulation factors, and renal failure due to hepatorenal syndrome). The principal investigators from Noguchi Institute regularly visited the study sites and trained local hospital staff on the case definition to minimize a selection bias due to different interpretations of the criteria.
A blood specimen was taken and basic clinical and demographic data were recorded on the case investigation form. Serum was separated by centrifugation. If a centrifuge was not available, blood was kept in the refrigerator until there was complete retraction of the clot and serum could be removed. Serum was stored in the refrigerator for a maximum of one week at the study site and transported in a cool box with ice packs to the laboratory at Noguchi Memorial Institute. In the laboratory, sera were stored at −20°C.
Viral nucleic acid was extracted from 140 µl serum using the QIAamp viral RNA kit (Qiagen). All PCR assays were performed in a volume of 25 µl with 2.5 µl or 5 µl nucleic acid extract as a template. Nested PCRs contained 1 µl PCR product of the first round RT-PCR as a template. Reagents, cycle numbers, primer sequences, target region, and amplicon length are shown in Table 1. Conventional RT-PCR assays for Lassa virus [24], RVF virus [25], flaviviruses including YF and dengue virus [26], CCHF virus [27], [28], and filoviruses including Ebola and Marburg virus [29] were performed at Noguchi Memorial Institute using a GeneAmp PCR System 2700 (Applied Biosystems). Testing for hepatitis A virus (HAV), hepatitis B virus (HBV), hepatitis C virus (HCV), and hepatitis E virus (HEV) was performed at Bernhard-Nocht-Institute. First, samples were tested using RealStar real-time PCR kits for HAV, HBV, HCV, and HEV (kindly provided by Thomas Laue, altona Diagnostics, Germany) on a LightCycler 480 (Roche) according to the manufacturer's instructions. In addition, conventional PCR assays were performed using a Primus25advanced thermocycler (PeqLab, Germany): HBV PCR [30] and two nested HEV RT-PCR assays, ORF2-457 PCR [31] and ORF2/3-137 PCR [32], were done on all samples, while HAV [33], [34] and HCV [35] RT-PCR assays were done only on samples positive for these viruses in the real-time RT-PCR.
Serological tests for HAV, HBV, and HEV were performed in 96-well ELISA format using commercially available kits according to the manufacturer's instructions: HAV IgM ELISA (DRG, Germany), HBsAg one Version Ultra (Dia.Pro, Italy), HBc IgM (Dia.Pro, Italy), and RecomWell HEV IgG and IgM (Mikrogen, Germany).
Serum levels of total protein, albumin, total bilirubin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), lactate dehydrogenase (LDH), amylase, creatinine, and urea nitrogen were determined with a Spotchem EZ SP-4430 analyzer (Arkray, Japan) at room temperature. Samples had been stored for ≥3 years at −20°C before analysis. Reference values were defined according to Kratz et al. [36].
PCR products generated in the conventional PCRs were sequenced on both stands using the PCR primers. The sequences were assembled using Lasergene software (DNASTAR) and automated base calling was proofread by visual inspection of the electropherograms. Phylogenetic analysis included the novel sequences of HAV (VP1/2A junction; n = 8), HBV (N terminal P protein region; n = 43), and HCV (NS5B region; n = 10) as well as sequences available from GenBank by December 2012. Three categories of GenBank entries were retrieved: (i) sequences with close relationship to the novel sequences as determined by BLAST search, (ii) sequences from Ghana, and (iii) representative sequences of genotypes and sub-genotypes [33], [37], [38]. The program jModelTest 0.1.1 [39] identified the general time-reversible model of sequence evolution with a gamma distribution of among-site nucleotide substitution rate variation (GTR+gamma) as the substitution model that best describes the data in the nucleotide sequence alignment for HAV (37 taxa, 467 sites), HBV (117 taxa, 684 sites), and HCV (85 taxa, 337 sites). Phylogenies were inferred by the Bayesian Markov Chain Monte Carlo method implemented in BEAST software [40]. Initially, BEAST was run with the uncorrelated relaxed lognormal clock to measure the ucld.stdev parameter which gives an indication of how clock-like the data is (HAV, ucld.stdev 0.16; HBV, ucld.stdev 0.63; HCV, ucld.stdev 0.31) [41]. The final analysis was performed using the parameters: GTR+gamma; constant population size; strict clock for HAV, relaxed exponential clock for HBV, and relaxed lognormal clock for HCV; 107 steps with sampling every 104th step; and 3 independent runs combined (effective sampling size >200 for all parameters).
Quantitative variables were analyzed using non-parametric descriptive statistics and statistical tests. Statistical comparison of unpaired groups was performed for continuous parameters with the Mann–Whitney U test and for frequencies with two-tailed Fisher's Exact test. A critical p value of 0.01 was considered appropriate, given that several tests were performed on the data set. If a study participant had missing data for a specific variable, he/she was excluded from the analysis of this variable, but not of the other variables. Relative frequencies (e.g. in %) for a variable were calculated with the denominator: group size minus number of participants with missing data for this variable.
A representative set of 6 HAV, 28 HBV, and 10 HCV sequences has been sent to GenBank and assigned the accession nos. KC632110–KC632153.
The project started in 2008 with a pilot study involving 27 patients. VHF was not detected, but 7 (26%) patients had acute hepatitis B or an exacerbation of chronic hepatitis B as evidenced by presence of HBV DNA, HBsAg, and anti-HBc IgM or seroconversion to HBsAg/HBV DNA-positive. Therefore, the design of the main study was adjusted to incorporate a wider range of tests for viral hepatitis.
The main study was conducted from 2009 to 2011 and included 258 patients from 18 study sites. The number of patients per site ranged from 1 to 72 (legend to Fig. 1). The male-to-female ratio was 68∶32 and median age was 23.5 years. Fever was reported in 74%, jaundice in 48%, and hemorrhage in 13% of the cases. None of the patients tested positive by PCR for Lassa virus, filoviruses including Ebola and Marburg virus, flaviviruses including dengue and YF virus, RVF virus, and CCHF virus (Fig. 2). However, 105 (41%) of the patients had evidence of HBV infection: 100 (39%) were HBsAg-positive and 59 (23%) were HBV DNA-positive, of whom 56 (22%) were positive for both markers (Table 2). Anti-HBc IgM was detected in 21 (8.1%) of HBV DNA and/or HBsAg-positive patients indicating acute or fulminant hepatitis B or severe exacerbation of chronic hepatitis B. Presumed chronic hepatitis B with active virus replication was diagnosed in another 37 (14%) patients, who were positive for HBsAg and HBV DNA, but negative for anti-HBc IgM. In total, 26 (10%) patients had evidence of acute hepatitis A: 25 (9.7%) were anti-HAV IgM-positive and 19 (7.4%) were HAV RNA-positive, of whom 18 (7.0%) were positive for both markers. HCV RNA was detected in 20 (7.8%) patients, indicating hepatitis C with active virus replication. Serological HEV markers were found in 10 (3.9%) patients showing anti-HEV IgM and 15 (5.8%) patients showing anti-HEV IgG. However, none showed both markers. In addition, none of the study patients was positive for HEV RNA upon testing with 3 different PCR assays.
Hepatitis A mainly affected children with a median age of 5, while hepatitis B and C mainly affected young adults with a median age around 30 years (Table 2). Compared to patients without diagnosis (median age 18 years), patients with hepatitis A were significantly younger (p = 0.001), and those with hepatitis B and C were significantly older (p = 0.00004 and p = 0.005, respectively). Patients with hepatitis B more frequently originated from the Northern region (p<0.0001) and less frequently from the Upper West & East regions (p = 0.0004) compared to patients without diagnosis (Table 2). The only symptom that showed a statistically significant association with a diagnostic category was jaundice; it was about twice as frequent among patients with acute hepatitis A or B than among patients without diagnosis (p = 0.0005 and p = 0.0003, respectively) (Table 2).
To provide additional evidence for an association of the virological diagnoses with the disease observed in the patients, clinical chemistry parameters were measured retrospectively for a subset of patients (n = 67) with a diagnosis of viral hepatitis (Table 3). Although the prolonged storage at −20°C and the freeze-thaw cycles predictably led to some loss of serum enzyme activities [42], [43], pathological changes were seen in most patients. AST and LDH were elevated in the majority of hepatitis cases. ALT, which is particularly sensitive to storage at −20°C [43], was elevated in a large fraction of patients with acute hepatitis A and B. Albumin was specifically decreased (relative to total protein) in most patients, while total bilirubin was elevated in particular in patients with acute hepatitis. The latter corresponds to the frequent finding of jaundice in this group. Between 60 and 100% of hepatitis patients showed 2 or more pathological values for parameters that indicate liver disease, namely albumin, total bilirubin, AST, ALT, and LDH (Table 3). In some patients, minor elevations of amylase, creatinine, and urea nitrogen were found. Overall, the clinical chemistry data show that the majority of patients with viral hepatitis has biochemical evidence of liver disease.
To determine the genotype of HAV, HBV, and HCV circulating in the study area, the PCR products were sequenced and subjected to phylogenetic analysis (Fig. 3). The Ghanaian HAV strains are closely related to each other and form within genotype IB a separate clade (posterior probability value of 1), which also includes strains from France. The HBV strains from Ghana belong to genotype A and E. The genotype E strains are scattered over the phylogenetic tree and are related to other genotype E strains from West Africa. However, due to the close relationship among all genotype E viruses, the phylogenetic relationships within this clade could not be resolved. The genotype A strains from Ghana show a well-supported sister relationship (posterior probability values of 1) with strains from Nigeria, Cameroon, and Haiti, which have previously been classified as sub-genotype A5. The HCV strains belong to genotypes 1, 2, and 4, are highly divergent, and difficult to classify into sub-genotypes. The Ghanaian genotype 4 strain is related to sub-genotypes 4a and 4c and shows a close relationship to strains from Egypt. Within genotype 2, two strains are related to sub-genotype 2c, one is related to sub-genotype 2d, and two are related to sub-genotype 2l. The remaining two genotype 2 strains could not be sub-classified. Similarly, the sub-genotype of the two strains within genotype 1 could not be determined. Genotype 1 and 2 strains are somewhat related to previously described HCV strains from Ghana [17].
This study was conducted to determine if VHF or other infectious diseases presenting with similar symptoms are of medical importance in Central and Northern areas of Ghana. The main conclusion from our data is that VHF hardly contributes to hospital morbidity in the study area indicating a low incidence of severe VHF. This does not prove that VHF is absent and in fact, YF is endemic in the country and cases may be seen at any time. In addition, mild VHF cases may not attend the hospital. Instead of VHF, a high incidence of viral hepatitis was found.
The case definition to screen for potential VHF patients was intentionally broad. On the one hand, clinical complications of VHF can mimic other diseases [1]. For example, renal failure is a frequent complication in Lassa patients [20], central nervous system disturbance such as seizures or coma is frequent in Lassa fever and RVF [21], [22], and hepatic injury with jaundice is typical for RVF and YF [8], [23]. On the other hand, the pathognomonic bleeding is rare in some VHF [1] and even fever cannot be considered a conditio sine qua non, as patients presenting in the terminal stage of Lassa fever — and probably other VHFs as well — are often normo- or hypothermic [20]. In the end, the broad case definition facilitated the sampling of patients with other serious diseases, including acute and chronic viral hepatitis.
Whether the diagnosed HAV, HBV, or HCV infections actually caused the clinical symptoms is difficult to prove. As the study was focused on diagnostic aspects, only minimal clinical information was collected and additional invasive or non-invasive diagnostic procedures were not performed. However, the retrospective analysis of clinical chemistry parameters provides some clues. For patients positive for anti-HAV IgM and/or HAV RNA or anti-HBc IgM plus HBV DNA, there is little doubt that the clinical symptoms are due to HAV or HBV infection, respectively. Anti-HBc IgM is a classical marker of acute or fulminant hepatitis B, but may also indicate an acute exacerbation of chronic hepatitis [44], [45]. Anti-HAV IgM and presence of HAV RNA are diagnostic markers of acute hepatitis A. Liver damage in our hepatitis A patients and anti-HBc IgM-positive hepatitis B patients is evidenced by jaundice as well as elevated AST, ALT, LDH, and bilirubin levels in serum. The hemorrhage reported in a few patients might be interpreted as a sign of liver failure, although this speculation cannot be supported by additional data.
A large fraction of study patients showed HBsAg and HBV DNA without being anti-HBc IgM positive, suggesting chronic hepatitis B with active virus replication. The clinical chemistry data show that about 60% of these patients had liver damage (≥2 pathological values for albumin, total bilirubin, AST, ALT, and LDH). Similarly, about 85% of patients with active HCV infection, as evidenced by virus detection, had biochemical evidence of liver damage (≥2 pathological values). In particular, AST and LDH levels were elevated and albumin levels were decreased in both groups. Thus, it is plausible that a large fraction of anti-HBc IgM-negative hepatitis B patients and hepatitis C patients attended the hospital due to the related liver disease. However, superinfection by another pathogen may not be excluded as cause of the disease.
Acute hepatitis A and B may lead to fulminant liver failure in a small fraction of patients [46]–[49] and chronic hepatitis B and C are the worldwide leading causes of liver cirrhosis and hepatocellular carcinoma [50]. Reports from other West African countries confirm that hepatitis B and C are among the major causes of hepatocellular carcinoma, liver cirrhosis, and fulminant hepatic failure in the region [51]–[55], which is supported by our data. Hepatitis B is already part of the Expanded Program on Immunization in Ghana. Each child should receive three doses of DPT-HepB-Hib (pentavalent) vaccine formulation at 6, 10, and 14 weeks after birth (http://www.afro.who.int/en/ghana/country-programmes/3215-expanded-program-of-immunisation-epi.html). An immunization coverage survey conducted in 2008 showed a national coverage for the third dose (Penta 3) of 72% (range 40 to 89% at district level) [56]. Thus, the hepatitis B incidence is expected to decrease in the long term. In spite of this progress, the implementation in clinical practice of established treatment options for hepatitis B and C [57], [58] should be a strategic goal to reduce morbidity and mortality from both infections [50]–[55].
The vast majority of hepatitis A patients were young children. This is consistent with the known epidemiology of this infection. HAV is transmitted by the fecal-oral route, e.g. via contaminated food or water. In the developing world, infections are most frequently acquired during early childhood, resulting in a high proportion of adults that are immune to HAV [59].
The prevalence of HBsAg carriers among blood donors in the Northern and Ashanti regions is 11% [10], [11], [18], [19]. The HBsAg prevalence increases in Ghana after birth up to 20–35% in individuals aged 11–15 years and then decreases with age, while the percentage of anti-HBc, a marker of past or present HBV infection, steadily increases up to 70 and 90% in individuals aged 20 and 40 years, respectively [10], [11], [16], [19]. This pattern suggests an acquisition of HBV infection with advancing age predominantly through horizontal transmission in childhood [16]. The epidemiology of HCV is similar, though not identical with HBV. The HCV seroprevalence in the Ashanti region ranges from 3 to 10% [11], [14], [18]. It is already high in children and does not show a clear age-dependent increase [11], [14], implying that many HCV infections occur during childhood. In summary, a large percentage of HBV and HCV-infected individuals in Ghana carry the virus since childhood and presumably develop overt symptoms of liver disease in adulthood. This epidemiology is consistent with the higher age of our hepatitis B and C patients; half of them were older than 30 years. However, no correlation between HBV and HCV markers was previously found in Ghana [11], which is in agreement with the low frequency of HBV/HCV co-infections in our study.
The incidence of hepatitis B in the study hospitals of the Northern region and the Upper West & East regions was higher and lower, respectively, relative to the control group of patients without diagnosis. We have no good explanation for this finding. The three regions are the poorest of the country and comparable with respect to many indicators [60] (see also legend to Fig. 1). A close inspection of the HBV phylogenetic tree reveals one genotype E cluster that nearly exclusively comprises sequences from the Northern region (Fig. 3, HBV-Ghana sequences no. 40, 22, 36, 38, 11, 25, 21, 23, 20, and 24 from top to bottom), while all other clusters showed no geographic pattern. Whether this Northern region-specific lineage is associated with a higher incidence remains to be studied. A sampling bias due to differences in the patient selection procedure in the various study sites may also be taken into account.
A few patients showed serological markers of HEV infection. However, all PCR assays were negative, HEV immunoassays are prone to false reactivity [61], [62], and none of the patients showed both IgM and IgG. Usually, both serological markers and HEV RNA are present during acute infection ([62], [63] and validation data for the test used in this study [64]). Therefore, we conclude that acute HEV infection does not significantly contribute to hospital morbidity in the study area.
Though a minor aspect of our study, the phylogenetic analysis revealed some peculiarities of HAV, HBV, and HCV strains circulating in Ghana. All HAV strains are closely related to each other and belong to genotype IB, though the patients originate from Ashanti, Brong-Ahafo, Northern, or Upper West region. This suggests that a predominant HAV strain circulates in large parts of Ghana. In agreement with previous reports from Ghana and other West African countries [12], [13], the HBV strains of our study belong to genotype A and E. The genotype A strains also cluster with other strains from West Africa and may belong to sub-genotype A5. The HCV sequences reported here and previously from Ghana and other West African countries show that the strains in the region are extremely diverse, belong to various genotypes (1, 2, and 4) and sub-genotypes, and often cannot be classified into sub-genotypes, as they are too divergent from the reference strains [14], [17], [65], [66]. Further studies are warranted to clarify the interesting molecular epidemiology of HCV in Ghana.
|
10.1371/journal.pcbi.1002179 | Computational Modeling of Allosteric Communication Reveals Organizing Principles of Mutation-Induced Signaling in ABL and EGFR Kinases | The emerging structural information about allosteric kinase complexes and the growing number of allosteric inhibitors call for a systematic strategy to delineate and classify mechanisms of allosteric regulation and long-range communication that control kinase activity. In this work, we have investigated mechanistic aspects of long-range communications in ABL and EGFR kinases based on the results of multiscale simulations of regulatory complexes and computational modeling of signal propagation in proteins. These approaches have been systematically employed to elucidate organizing molecular principles of allosteric signaling in the ABL and EGFR multi-domain regulatory complexes and analyze allosteric signatures of the gate-keeper cancer mutations. We have presented evidence that mechanisms of allosteric activation may have universally evolved in the ABL and EGFR regulatory complexes as a product of a functional cross-talk between the organizing αF-helix and conformationally adaptive αI-helix and αC-helix. These structural elements form a dynamic network of efficiently communicated clusters that may control the long-range interdomain coupling and allosteric activation. The results of this study have unveiled a unifying effect of the gate-keeper cancer mutations as catalysts of kinase activation, leading to the enhanced long-range communication among allosterically coupled segments and stabilization of the active kinase form. The results of this study can reconcile recent experimental studies of allosteric inhibition and long-range cooperativity between binding sites in protein kinases. The presented study offers a novel molecular insight into mechanistic aspects of allosteric kinase signaling and provides a quantitative picture of activation mechanisms in protein kinases at the atomic level.
| Despite recent progress in computational and experimental studies of dynamic regulation in protein kinases, a mechanistic understanding of long-range communication and mechanisms of mutation-induced signaling controlling kinase activity remains largely qualitative. In this study, we have performed a systematic modeling and analysis of allosteric activation in ABL and EGFR kinases at the increasing level of complexity - from catalytic domain to multi-domain regulatory complexes. The results of this study have revealed organizing structural and mechanistic principles of allosteric signaling in protein kinases. Although activation mechanisms in ABL and EGFR kinases have evolved through acquisition of structurally different regulatory complexes, we have found that long-range interdomain communication between common functional segments (αF-helix and αC-helix) may be important for allosteric activation. The results of study have revealed molecular signatures of activating cancer mutations and have shed the light on general mechanistic aspects of mutation-induced signaling in protein kinases. An advanced understanding and further characterization of molecular signatures of kinase mutations may aid in a better rationalization of mutational effects on clinical outcomes and facilitate molecular-based therapeutic strategies to combat kinase mutation-dependent tumorigenesis.
| The phenomenon of allosteric communication is fundamental to many biological processes and is recognized as an important factor governing molecular regulation of signal transduction networks [1], [2]. Theoretical and computational studies of allostery in biomolecular systems have witnessed a recent renaissance fueled by the growing interest in the development of quantitative models of allosteric communication in proteins and biological networks [3]–[7]. Sequence-based approaches have unveiled that protein allostery may be mediated by coupled motions of evolutionary conserved yet sparse networks of functional residues which constitute signal communication pathways in proteins [8], [9]. Recent network-based structural studies have also demonstrated that allosteric pathways may be formed through interactions of evolutionary conserved residues that are energetically coupled to mediate long-range communication [10]–[12]. Mechanistic understanding of collective protein motions and allosteric transitions at the molecular level has been significantly advanced by the employment of elastic network models (ENM) and normal mode analysis (NMA) [13]–[25]. These approaches have been further integrated with the information-based Markovian theory of signal propagation [26], [27] and have provided a generalized formalism of allosteric communication in proteins [28]–[31]. Structure-based ENM approaches combined with sequence-based bioinformatics analyses have identified that conserved low-frequency modes of collective motions are robust to sequence variations and capable of transmitting molecular signals over long distances [23]–[25], [31]. Allosteric communication mechanisms can range from a sequential model, where binding of a molecule at one site causes a sequential propagation of conformational changes across the protein, to a fully cooperative model, where structural changes are tightly coupled. More recently, an intermediate, “block-based” model was proposed, where sparse clusters of closely interacting residues can maintain a weak association to other blocks of residues and thus pass information between more distance regions of a protein [32].
Collectively, these studies have shown that allosteric networks orchestrating cooperative protein motions can be formed by evolutionary conserved and sparsely connected groups of residues, suggesting that rapid transmission of allosteric signals through a small network of distantly connected residue clusters may be a universal requirement encoded across protein families. Statistical analyses of motions in allosteric proteins with known inactive and active crystal structures have quantified the magnitude of allosteric effects, revealing a strong preference toward weakly constrained regions such as loops and protein surface regions [33]. Subsequent graph-based analysis and a global communication network model have shown that small-world allosteric networks have sparse connectivity and long-range protein communication is determined by specific residue clusters playing critical roles in the transmission of functional signals [34], [35]. The global communication network (GCN) model has integrated tertiary (residue-scale) and quaternary (subunit-scale) structural changes, providing a more general representation of allosteric communication mechanisms that allowed to simplify atomistic simulations and proven useful in guiding experiments probing allosteric function [35]. Data mining and machine learning methods using support-vector models have helped to infer rules that can distinguish structural hotspots of functionally important allosteric residues [36]. Computational biophysics studies of allosteric regulation have explored a functional linkage between simulations of protein dynamics and energetics of allosteric coupling [37]–[45]. Thermodynamics-based approaches that linked structural perturbations with free energy changes of allosteric coupling have provided quantitative insights into allosteric mechanisms of conformational switching [46]–[49]. A physics-based perturbation method, the Rotamerically Induced Perturbation (RIP), can generate concerted protein motions by applying local torsional perturbations to individual residues [50]. Nonequilibrium methods can monitor concerted protein motions and determine the distribution of signaling pathways, while avoiding long simulation times required in conventional molecular dynamics (MD) simulations [51]. Graph-based analysis of protein allosteric communication can reduce complexity and yield a convenient characterization of the protein architectures as one-dimensional maps comprised of nodes (residues) connected by edges (inter-residue “interactions”) [52]–[61]. These methods have shown that protein structural graphs form small world networks [52]–[55], characterized by high local residue connectivity and a small number of long-range connectivity. Network and graph-based approaches have been employed in predicting protein-protein interactions [54], [55] catalytic sites in enzymes [56], [57], protein structure, energetics and evolution [58]–[61].
The allosteric regulation of protein kinases serves as an efficient strategy for molecular communication and event coupling in signal transduction networks. The regulatory interactions have a major role in determining the conformational dynamics of the kinase domain and activation mechanisms [62]–[69]. Protein kinase regulation may be controlled by a dynamic coupling of two spatially distributed yet conserved and functionally important intermolecular networks between the N-lobe and the C-lobe forming a hydrophobic regulatory spine and a catalytic spine [66]–[69]. The wealth of structure-functional studies about protein kinases has demonstrated that protein kinase activity can be tightly regulated via dynamic interconversion between closely related active and highly specific inactive kinase states - a structural hallmark of the kinase domain which is critical for its normal function [70]–[79]. High-resolution nuclear magnetic resonance (NMR) spectroscopy can complement X-ray crystallography studies by probing protein dynamics on multiple time scales and detecting a site-specific ligand signature that allows differentiation between competitive and allosteric inhibitor binding [80]. NMR studies have detected protein kinase motions in the active and inactive forms on multiple time scales, suggesting that conformational mobility is vital for regulatory control of kinase activity [81]. The dependence of chronic myeloid leukemia (CML) on the translocated BCR-ABL kinase is associated with unique drug responses to small molecule inhibitors [82]. The mechanism of protein kinase regulation via dynamic equilibrium between structurally different functional states has been successfully exploited in the discovery of selective inhibitors targeting inactive conformations of the ABL kinase [83]–[94]. A large number of point mutations that impair the binding of Imatinib (Gleevec) to ABL have been described [83], [84], suggesting that some drug resistant mutations could exist before treatment, and may contribute to tumorigenesis. Structurally conserved gate-keeper mutation ABL-T315I is a dominant cancer-causing alteration, leading to the most severe Imatinib resistance by favoring the active form of the ABL kinase. These findings guided the design of the second-generation ABL inhibitors Dasatinib and Nilotinib [85]–[91]. While these inhibitors are effective against most ABL mutants, the ABL-T315I mutation is still resistant to all three therapies. Most recently, a third-generation of rationally designed analogs and hybrids of Imatinib and Dasatinib, including Ponatinib, DCC-2036 and HG-7-85-01 [92]–[94] were shown to recognize a broad spectrum of inactive kinase conformations and retained potency against ABL-T315I. Nevertheless, activating mutations that destabilize the inactive conformation of ABL (most notably ABL-T315I) still result in reduced binding affinity of these inhibitors. Although the vast majority of protein kinase inhibitors bind to the ATP binding site of the catalytic domain, a considerable effort has been recently invested to discover inhibitors associated with a specific kinase and disease [95]–[100].
Unlike ATP-competitive kinase inhibitors, allosteric inhibitors typically bind outside the catalytic domain and affect kinase activity by eliciting global conformational transformations, which may confer a greater specificity and allow for a subtle modulation of kinase regulation [101]. Allosteric regulation mechanisms in protein kinases may include stabilization of the inactive MEK kinases by targeting the adjacent to the ATP binding pocket in MEK-1, MEK-2 [102], [103] and JNK kinase [104]–[106]; allosteric binding to the myristoyl-binding pocket of ABL and regulation via formation of multidomain ABL-SH2-SH3 complexes [107]–[114]; activation mechanism via formation of regulatory complexes in cyclin-dependent kinase 2 (CDK2) [115], [116], EGFR [117]–[122], HER2/Erb2 [123], HER4/ErbB4 [124], [125]; and allosteric regulation of AKT [126]–[131] and PDK1 kinases [132], [133] via docking of a phosphorylated hydrophobic motif to a hydrophobic pocket on the N-terminal lobe in the catalytic domain. Activation processes in the ABL kinase is linked with the formation of multi-protein regulatory complexes with the SH2 and SH3 domains. Crystallographic studies have determined that in the downregulated inactive state of the ABL-SH2-SH3 complex the SH3-SH2 unit is docked onto the kinase catalytic domain [107], [108]. In contrast, small angle X-ray scattering (SAXS) analysis has detected a dramatic structural rearrangement in the active ABL complex accompanied by the release of the inhibitory interactions and disengagement of the SH2-SH3 domains [107], [108]. Hydrogen exchange mass spectrometry (HX MS) investigation of the ABL-T315I dynamics has provided the first evidence of long-range conformational disturbances caused by activating mutations and allosterically transmitted to the remote protein regions [111]. Recently discovered allosteric inhibitors GNF-2, GNF-5 of the ABL kinase can bind to the myristoyl-binding pocket and independently inhibit kinase activity [112], [113]. HX MS studies of the ABL-T315I dynamics in the presence of ATP competitive inhibitor Dasatinib and GNF-5 have revealed long-range cooperativity between the myristate-binding site and the ATP-binding site induced upon allosteric inhibitor binding that allowed for the effective synergistic inhibition of the ABL-T315I mutant by a drug combination [114].
Structure-functional studies of EGFR kinase domains have revealed that the formation of an asymmetric kinase dimer is critically associated with an activated kinase conformation and is essential for tyrosine kinase activation [117]–[122]. A recent crystal structure of the HER2 kinase domain [123] has provided additional support to allosteric activation via asymmetric dimerization, similar to activation mechanisms in the EGFR and HER4 kinases [124], [125]. In common to the crystal structures of EGFR, HER2 and HER4 kinases, activation mechanisms may exploit an asymmetric head-to-tail dimer, in which the C- lobe of one monomer acts as a “donor” monomer (activator) that interacts with the N-lobe of an adjacent “acceptor” monomer (receiver), stabilizing conformational changes that activate the receiver molecule. Moreover, an asymmetric structural arrangement of a functional EGFR dimer is highly similar to the complex formed by the receiver CDK2 kinase with its activator, cyclin A [115], [116]. Recent studies have shown the importance of the intracellular juxtamembrane EGFR region in promoting activation of an asymmetric dimer via forming a “juxtamembrane latch” between the N-terminal lobe of the receiver and the C-terminal lobe of the activator, allowing to “glue” two kinase monomers and potentiate activation of the receiver molecule [120], [121]. Hence, a unifying structural mechanism associated with asymmetric tyrosine kinase arrangements in regulatory complexes could underlie the activation mechanism of the EGF and ErbB protein families and explain a linkage between ligand-induced receptor dimerization and kinase activation [134]–[137].
Mechanisms of protein kinases regulation have been also studied in computational investigations of c-Src kinase [138]–[145], adenylate kinase [146], ABL kinase [147], CDK5 kinase [148], KIT kinase [149], PKA kinase [150], and AKT/PKB kinase [151], RET and MET kinases [152], [153] and EGFR kinase [154]–[160]. These studies have suggested that functional coupling between collective motions and local structural changes can rationalize the experimental data and provide molecular insights into allosteric mechanisms. We have previously reported that the impact of the gate-keeper mutant on conformational dynamics of ABL may spread far beyond the immediate site of mutation leading to functional changes in conformational mobility at the remote kinase regions [155]. These results corroborated with the HX MS experiments of ABL regulatory complexes [111], pointing to a potential allosteric effect of the activating mutations in the ABL kinase.
Despite recent progress in computational and experimental studies of protein kinase structure and function, the molecular mechanism and dynamics of mutation-induced allosteric kinase activation by regulatory complexes remain mostly qualitative. In this work, we have investigated mechanistic aspects of allosteric activation mechanisms in ABL and EGFR kinases by integrating the results of multiscale simulations with the principal component analysis and computational modeling of signal propagation in proteins. We show that mechanisms of allosteric activation in the ABL and EGFR kinases may be determined by a functional cross-talk between the organizing αF-helix and conformationally adaptive αI-helix and αC-helix. These structural elements form a dynamic network of efficiently communicated clusters that may control the long-range coupling and allosteric activation in the interdomain regulatory complexes. The results of study may reconcile current experimental data pointing to general mechanistic aspects of activating transitions in protein kinases.
In this work, molecular dynamics (MD) simulations, principal component analysis (PCA) and computational modeling of signal propagation in proteins [161], [162] were employed to elucidate molecular principles of allosteric communication in ABL and EGFR kinases and determine allosteric signatures of the gate-keeper cancer mutations at the increasing level of complexity - from catalytic domain (ABL-T315I, EGFR-T790M) to multi-domain regulatory complexes (ABL-T334I, EGFR-T766M). The following specific objectives were pursued in the present study: (a) perform a comparative analysis of the collective protein motions and allosteric communication profiles obtained from simulations of the catalytic domain and regulatory complexes; (b) determine key structural elements and functional residues in ABL and EGFR kinases involved in collective motions and long-range allosteric coupling; (d) analyze and compare long-range communications and allosteric signatures of mutation-induced kinase activation by the gate-keeper mutations in ABL and EGFR. We employed concepts of the absolute and relative Long Range Communication Capability (LRCC) associated with the protein residues in the context of a computational model of signal propagation in proteins [161] (see Materials and Methods for a detailed description).
According to this model, two remote protein residues (or residue clusters) are defined to have a high communication propensity (or communication capability) if the mean-square fluctuation of their inter-residue distance would vary within a relatively small range over long time MD simulations. The higher the fraction of residues that have high communication efficiency with a given residue at an empirically chosen threshold of efficient and fast long-range communications, the greater would be the absolute LRCC of this residue. Under this assumption, a perturbation in one residue in a pair of well-communicated residues should be consistently “communicated” to the “partnering” residue located at a significant distance. Conversely, two residues would not efficiently communicate when the thermal fluctuations of their inter-residue distance would be large and inconsistent with the level of displacements for respective residues, e.g. the inter-residue distance would change in a wide spectrum of values inconsistent with the amplitude of thermal fluctuations of individual residues. This would amount to a slow and inconsistent propagation of a perturbation signal from one residue to the other. A meaningful metric of long-range communication would display fluctuations of the inter-residue distance corresponding and reflecting the order of per residue fluctuations. According to our model, mutational changes can modulate the energy landscape of a protein and alter communication propensities among pairs of residues. The communication propensity histograms of the different protein states scan LRCC that represent the fraction of residues that may have high communication efficiency with a given residue and located at distances greater than a defined threshold from that residue. We hypothesize that the residue clusters characterized by the peaks in the communication efficiency profile may be important for long-range cooperative interactions.
In this section, we analyzed collective motions and long-range communications in the ABL kinases using the results of our recently reported simulations of the ABL catalytic domain and multidomain complexes [155] in the following functional states: inactive ABL state (PDB ID 1IEP) [70], the active ABL state (PDB ID 1M52) [71], [72]; the active form of the ABL-T315I mutant (PDB ID 2Z60) [75]; the inactive autoinhibited form of ABL-SH2-SH3 complex (PDB ID 2FO0) and the active form of the ABL complex (PDB ID 1OPL) [108]. For clarity and completeness of the discussion, we summarized our earlier results of 20 ns MD simulations of the ABL complexes [155] in the context of the current objectives (Figure S1). According to our results, a structurally stable bundle of α-helices in the C-terminal could be dynamically coupled via regulatory spines with the regions of larger thermal fluctuations corresponded to the P-loop, αC-helix and the activation loop. Moreover, mutation-induced modulation of protein flexibility in the inactive state may be compounded by the increased structural stability of the active form [155].
To better characterize the nature of collective motions between functional kinase regions in regulatory complexes, we analyzed here collective motions of the ABL complexes using PCA of the covariance matrix, calculated from 20 ns MD trajectories of the complete systems (Figure S2). The correlation matrix described the linear correlation between any pairs of Cα atoms as they move around their average position during simulations. A positive correlation between two atoms could reflect a concerted motion in the same direction, whereas a negative correlation may indicate an opposite direction motion. We noted important similarities and differences between the correlation profiles of inactive (Figure S2 A) and active forms (Figure S2 B). The covariance map of the active ABL complex displayed an increased and more broadly distributed level of positive correlation, both within the catalytic core, and also between the catalytic domain and the SH2 domain. Overall, the correlated motions along the first eigen mode in the active ABL complex represented a more uniform level of motion between sub-domains with lower amplitude fluctuations as compared to the autoinhibited form. The “breathing” inter-lobe motion of the catalytic core was coupled with the motions of the activation loop, αC-helix and the αG-helix of the C-terminal. The inter-lobe motions were also coupled with the collective inter-domain motions between the catalytic domain and the SH2 domain.
It is worth noting that the crystal structure of the active ABL complex (PDB ID 1OPL) is completely missing SH3 domain [108]. Consequently, to ensure a consistent comparison of long-range communications between inactive and active complexes, we focused our analysis on the catalytic domain residues and a comparison of their communication profiles derived from simulations of the isolated catalytic core (Figure 1) and ABL complexes (Figure 2). Structural mapping of long-range communications in the ABL catalytic domain was done using a reference distance of 30 Å (Figures 1 A–C) revealing a coupling between the N-terminal αC-helix (residues 280–292) and a C-terminal core cluster comprised by the αF-helix (residues 418–433), αI-helix (residues 486-496), and αE-helix (residues 337–356) (Figure 1). According to our analysis, allosteric communication between a flexible β-strand of the N-terminal lobe, the αC-helix and the P+1 loop could be controlled by the integrating αF-helix. The relative LRCC values between the inactive and active WT forms (Figures 1 D–F) reflected a partial loss in the communication capabilities in the active state and the increased mobility of the αF-helix and the αE-helix, indicative of a more flexible active kinase form. We observed that this effect may be partly offset by a moderate increase in communication propensities of the hinge region (residues 310–335) and the αC-helix (Figure 1 D, E). An interesting finding from this analysis was an appreciable impact of ABL-T315I on long-range communications, manifested in a broadened network of allosterically coupled residues (Figures 1 D–F). This effect could be seen by inspecting changes in the density of allosterically coupled clusters between the inactive states (Figure 1A), the active form (Figure 1B) and the active form of the ABL-T315I mutant (Figure 1C). The αC-helix, αF-helix, and αI-helix regained their integrating role in enabling long-range communications in the mutant form. A similar pattern of long-range communications between structurally rigid αF-helix and conformationally adaptive αI-helix, αC-helix and P+1 loop was detected from simulations of the ABL complexes (Figures 2 A, B). The impact of the ABL-T334I mutation on the inactive complex manifested in a decline of the LRCC values and the increased mobility of the αF-helix (residues 437–453), αE-helix (residues 356–376), αI-helix (residues 511–528), and αC-helix (residues 299–311) (Figures 2 A,C). This observation is consistent with MD simulations of ABL complexes [155], where ABL-T315I was shown to weaken the “rigid-clamp” arrangement and destabilize the inactive complex.
The efficient long-range communications between the organizing αF-helix and conformationally adaptive αI-helix and αC-helix is not merely a result of small thermal fluctuations in the respective segments. A dynamic network of long-range interactions between these regions may rather have a functional significance in coordinating collective inter-lobe and inter-domain motions, both of which are known to be important to allosteric communication [35]. A broader and denser network of long-range communicated clusters in the ABL-T315I included also the hinge region, the hydrophobic spine and the catalytically critical Asp-Phe-Gly (DFG) motif from the activation loop (Figure 1C). The critical role of the integrated cluster formed by the hydrophobic and catalytic spines, both anchored to the integrated αF-helix, is well recognized as an organizing element regulating protein kinase dynamics and activity [66]–[69]. The cooperative interactions between the αF-helix and the αC-helix may control a dynamic connection between the two lobes of the catalytic core and be important for a dynamic assembly and disassembly of the hydrophobic spine regulating the protein kinase activity. The combined analysis of the correlated motions and long-range communications in the ABL complexes is consistent with a mechanistic model of kinase activation involving cooperative assembly of the hydrophobic spine, the formation of the Src-like intermediate structure, and a cooperative breakage and formation of characteristic salt bridges [155]. It is worth stressing that coupling between rigid and flexible protein regions and correlation of various motions may generally lead to both increases and decreases in thermodynamic stability. A broader network of concerted motions and long-range communications in the mutant form is consistent with our previous finding that all free energy components may act concertedly to enhance the thermodynamic stability of the active ABL-T315I [155].
Analysis of long-range communications allowed to highlight a functional role of stabilizing interdomain contacts in the inactive and active ABL complexes (Figure 3). In the inactive ABL complex, the SH2 domain is docked closely onto the kinase domain by repositioning and rigidifying the αI-helix of the C-terminal and forming a dense network of hydrogen bonds and packing interactions (Figure 3 A,B). We observed high average occupancies for the major interdomain contacts that maintained their stability throughout a long simulation period. These specific contacts included hydrogen bonding between side-chain of Arg-153 of the SH2 domain and the backbone carbonyls of the kinase domain residues Gln-517 and Glu-518. Additional hydrogen bonding was formed between Arg-189 of the SH2 domain and Asp-523 of the αI-helix. This hydrogen bonding network was further strengthened by packing interactions between Tyr-158 of the SH2 domain, which aromatic ring was perfectly stacked against Tyr-361 from the αE-helix of the kinase domain (Figure 3 B). Importantly, the high occupancies of these interdomain contacts were significantly reduced for the ABL-T334I mutant (Figure 3 C). The interdomain interactions of the active ABL complex included Ile-164 of the SH2 domain interacting with Thr-291 and Tyr-331 of the kinase domain. Interestingly, the occupancies of these core interactions were sustained in the active “top-hat” ABL complex at a relatively high level and even further consolidated for the mutant complex (Figure 3 D).
In agreement with the experimental data, this analysis provided another evidence of a detrimental impact of the activating mutation on structural integrity of the inactive ABL complex that could promote conformational transformation to the active state. Conversely, a stabilizing role of this mutation may be seen in the enhanced structural rigidity of the interface in the active ABL form. Based on this analysis, we could suggest that the activating ABL-T334I mutation could perturb the interdomain interface via allosteric coupling between αF-helix and αI-helix and lead to a significant destabilization of the “rigid-clamp” form of the ABL-SH2-SH3 complex. Hence, the impact of the gate-keeper mutation may be allosterically transmitted to the interdomain regions located at a considerable distance from the mutational site, supporting allosteric nature of the mutation-induced ABL activation. Collectively, these factors could contribute to the mutation-induced allosteric effect that may perturb the thermodynamic equilibrium away from the inactive form towards alternative conformational states and thus serve as a catalyst of activation. These results corroborated with the crystallographic and functional studies of the ABL-T315I mutant [74], [75] confirming an activating nature of the gatekeeper mutation [76]. These findings may have certain relevance in the context of drug resistance effects and design of ABL inhibitors. Our results suggested that the ABL-T315I mutation could allosterically strengthen and coordinate distinctive structural elements of the kinase core, leading to the enhanced structural consolidation of the constitutively active kinase form. As a result, design of ABL inhibitors binding to the active form of the enzyme would inevitably have to overcome competition from cellular ATP. Novel ABL inhibitors of ABL-T315I that bind to the inactive conformation could experience weaker competition from ATP and may act by preventing kinase activation, rather than by inhibiting kinase activity directly.
The evidence of efficient long-range communications in active ABL complexes may be of importance given the rapidly growing interest in developing novel and specific kinase inhibitors inhibition targeting allosteric regions. Indeed, our study may have specific implications in light of recent experimental studies of allosteric kinase inhibition and cooperativity between the myristate- and ATP-binding sites of ABL [113], [114]. HX MS analysis of ABL-T315I in the presence of Dasatinib and allosteric inhibitor GNF-5 demonstrated that binding in the myristate-binding site can elicit allosteric alterations in the conformational dynamics of the C-terminal αI-helix that are propagated to the β-strand of the C-terminal lobe and the ATP-binding site [114]. The analysis of collective motions pointed to a possibility of concerted motions between the β-strand in the N-terminal lobe (residues 260–280 in the inactive complex) and the αI-helix from the C-terminal (residues 446–463 in the inactive complex) (Figure S2). Interestingly, our analysis also suggested that allosteric coupling between a flexible β-strand of the N-terminal lobe, the αC-helix and the P+1 loop may be mediated and controlled by the integrating αF-helix. Moreover, we found that the C-terminal αI-helix and the β-strand of the N-terminal lobe could be involved in the long-range communication of the down-regulated ABL complex and allosteric coupling of these functionally important binding sites could be modulated by the gate-keeper mutation. The importance of these results may be appreciated in the context of experimentally detected allosteric effect of the GNF-5 inhibitor that binds to the myristate-binding site and can allosterically affect the thermodynamic stability of the ATP-binding site residues from the β-strand [113], [114]. Hence, our results are in accordance with a mechanistic view of allosteric ABL activation emerging from the experimental data. We propose that allosteric inhibitor binding with ABL-T315I may lead to concerted changes of conformational mobility in these regions, thereby restoring structural arrangement of the ATP-binding site compatible with Dasatinib binding. A detailed analysis of allosteric ABL inhibition by small molecules is being currently pursued in conjunction with the experimental verification by our collaborators, which a subject of a separate investigation that extends beyond the scope of the current study and will be presented elsewhere.
In this section, we analyzed allosteric signatures of the EGFR kinase catalytic domain using the results of MD simulations in the following functional states: the inactive EGFR form (PDB ID 1XKK) [77]; the active EGFR form (PDB ID 2J6M) [78], the active form of the EGFR-T790M mutant (PDB ID 2JIT) [79]. The analysis of long-range communications the EGFR catalytic domain revealed similar coupling between structurally rigid αF-helix and conformationally adaptive αI-helix, αC-helix of the catalytic core (Figure 4). This effect was seen from inspecting changes in the distribution of communicated residue clusters in the inactive state (Figure 4A), the active form (Figure 4B) and the active form of the EGFR-T790M mutant (Figure 4C). Mutation-induced amplification of protein flexibility in the inactive state could be accompanied by the counter-effect of restoring structural stability of the active mutant form (Figures 4 D-F). We found that structural elements of the catalytic core involved in long-range communications may be common in ABL and EGFR, e.g. structural architecture of the kinase fold could determine the basic topology of cooperative interaction network. It is well recognized that the ‘on–off’ equilibrium between the inactive and active EGFR states can be altered by activating mutations, resulting in a net increase in kinase activity. Crystallographic studies have proposed that this equilibrium shift may be a result of structural alterations induced by activating mutations [77]–[79]. Our data support these conjectures by showing that the gate-keeper mutation may allosterically enhance protein mobility in the inactive state and then restore structural integrity of the activated form. This result may be of interest in rationalizing the existing mechanisms of the EGFR-T790M resistance that can substantially suppress the inhibitory effects of EGFR-based drugs Erlotinib and Gefitinib in the treatment of lung cancer [163], [164]. Importantly, this mutation can promote oncogenic activation, uncontrolled cell proliferation and tumorigenesis even in the absence of the selective pressure from the kinase inhibitors. In fact, a recent study showed that the EGFR-T790M harboring resistant clones may be found even in untreated lung cancers [165]. Two different molecular mechanisms were offered to explain how EGFR-T790M could confer drug resistance. Initially, it was proposed that the gate-keeper mutation may detrimentally alter the topology of the ATP-binding pocket that would prevent binding of reversible EGFR inhibitors due to steric hindrance [163], [164]. A number of recently discovered irreversible EGFR inhibitors BIBW2992 [166], PF00299804 [167], and HKI-272 [168], [169] could still inhibit the T790M mutants via covalent binding at the catalytic pocket of EGFR, which was at odds with the steric hindrance mechanism. Another study revealed that EGFR-T790M could increase the ATP affinity back to the EGFR-WT level, which may lead to a reduced potency of any ATP-competitive agent [79], [170]. According to this report, the increased ATP affinity may be a primary mechanism by which EGFR-T790M could confer drug resistance. Furthermore, it was suggested that irreversible binding may not be required for effective inhibition of the T790M mutant [170]. A novel reversible EGFR inhibitor XL-647 can bind EGFR-T790M mutant with an affinity sufficient to compete with ATP [171]. Our results are in line with these studies suggesting that the restored long-range communications and reacquired structural rigidity of the EGFR-T790M mutant may prompt the increased ATP affinity towards this mutant and related drug resistance effects.
The initial investigations indicated a potential “negative” impact of the activating mutation on conformational dynamics in the symmetric dimer form [154], [155]. We expanded previous studies and report here the results of 20 ns MD simulations based on the crystal structures of asymmetric and symmetric EGFR dimers (PDB ID 2GS6) in the normal and oncogenic states [117]. The most recent crystal structures of the EGFR kinase domain revealed binding of the extended juxtamembrane latch of the receiver kinase to the activator kinase [120], [121]. 20 ns MD simulations were also performed using the extended crystal structure of a symmetric EGFR dimer PDB ID 3GT8 [120]. The residues 669–682 of the JM-B segment were crystallographically resolved under the same conditions in the crystal structures of both asymmetric and symmetric dimers [117]. For clarity, the analysis focused on MD simulations of these asymmetric and symmetric dimers that included important for activation residues of the JM-B motif. We set out to determine the effect of EGFR-T766M mutant on conformational dynamics of regulatory dimer complexes and to understand the molecular basis of mutation-induced allosteric activation in the functional asymmetric dimer.
A comparative analysis of conformational mobility demonstrated an increased structural integrity and a greater stability of an asymmetric dimer that could be enhanced by the activating mutation (Figure 5). While the mutational effect in a symmetric dimer led to the increased flexibility as evident from root mean square deviation (RMSD) values, a reduction in thermal fluctuations of the mutant form was seen for an asymmetric EGFR dimer (Figures 5 A,C). Protein flexibility variations were also computed from the root mean square fluctuation (RMSF) of the backbone residues (Figures 5 B, D). The RMSF profiles showed a higher degree of structural variations upon activating mutation in a symmetric dimer, reflected the increased mobility of the activation loop with RMSF = 5 Å in the mutant as compared to RMSF = 1.5 Å–2 Å for EGFR-WT. Other regions of the enhanced mobility corresponded to the αI-helix at the C-terminal part of the two monomers. In contrast, a global reduction of the conformational mobility extended beyond the immediate site of mutation for an asymmetric EGFR dimer, suggesting that the gate-keeper mutation may allosterically strengthen structural integrity of the functional EGFR form.
The effect of the activating mutation could be further illustrated by monitoring differences in the conformational mobility for two important regulatory elements: the activation loop and the αC-helix (Figure S3). Smaller thermal fluctuations of the activation loop were seen not only in the monomer A, that occupies a critical part of the interdomain interface, (Figure S3 A), but also in the monomer B, which is located away from the mutational site and the inter-domain interface (Figure S3 B). The intrinsic mobility of the αC-helix is critically important in the activation mechanism as one of the central mediators of allosteric changes [117]–[122]. The αC-helix of the receiver monomer A is a key component of the inter-monomer interface in the functional asymmetric dimer. A reduction of the thermal motions in this segment was observed on a longer time scale for both monomers of the asymmetric dimer. Nevertheless, there still remained a certain level of residual mobility in the αC-helix from the activator monomer (Figure S3 C, D). In contrast, simulations of a symmetric dimer indicated that the αC-helix of the receiver and the αH-helix of the activator could be more mobile on a longer time scale. These results also agreed with recent computer simulations of ErbB family kinases [158]–[160]. Additionally, structural conservation of the critical salt bridge Glu738-Lys721 plays an important role in kinase regulation [117]. This characteristic salt bridge is fully intact in both inactive and active EGFR conformations, and could only break briefly during the transformation between functional states [155]. A stable behavior of this critical salt bridge was observed in both WT and mutant forms of the asymmetric dimer (Figure S4). The mutation-induced stabilization effect was especially pronounced in the monomer B of an asymmetric EGFR dimer (Figure S4 B). The dynamics of the inter-monomer interface in the asymmetric dimer may be also controlled by motions of the juxtamembrane region of the receiver that undergoes moderate thermal fluctuations. In the context of these observations, it may be worth pointing out that the increased flexibility of an asymmetric dimer form of the HER2 kinase could lead to a less stable active conformation as compared to EGFR and the low intrinsic catalytic activity [123]. Overall, these results are in line with the mechanism of allosteric EGFR activation, according to which direct contacts between the C-lobe of activator and the N-lobe of the receiver could destabilize autoinhibitory interactions involving the activation loop of the receiver and, as a result, no phosphorylation may be required for activation [134]–[137].
To highlight the principal motions of the active and inactive EGFR dimers, PCA was performed and identified the most relevant displacements of groups by emphasizing the amplitude and direction of the dominant protein motions, through projection on a subset of principal components (eigenvectors) of the covariance matrix calculated from the MD ensemble (Figure S5). We quantified correlated motions within the same monomer and between two monomers for both inactive and active EGFR dimers. In the active, asymmetric dimer a positive correlation emerged not only within each monomer, but also between the activator and receiver monomers (Figure S5). Conversely, in the symmetric dimer positive correlated motions between monomers could be suppressed. The long-range positive cross-correlations extended beyond the intra-monomer regions and could signal the presence of a more diffuse communication network which would favor concerted rigid body motions in the EGFR asymmetric dimer. We observed that asymmetric dimer motions projected onto the principal eigen vector (PC 1) may be determined by coordinated “breathing-like” motions between the activator and the receiver monomers as rigid bodies. The concerted motions of the monomers were accompanied by the low amplitudes “breathing” motions between N-terminal and C-terminal lobes within the monomers, mostly in the activator molecule. These motions are likely determined by the underlying topology of the catalytic domain fold and involve coordinated moves of the P-loop, αC-helix, and the activation loop. However, the reduction of thermal motions in the asymmetric dimer, most notably in the αC-helix and activation loop of the receiver, resulted in mostly suppressed intra-monomer motions of the receiver. This is consistent with the mechanistic role of the activator that via direct interactions with the N-lobe of the receiver can induce and “lock” the characteristic active conformation in the receiver molecule. Our observations also agreed with recent studies of correlated motions in EGFR [157] and provided additional useful insights concerning hierarchy of functional motions in the EGFR regulatory complexes.
The analysis of correlated motions was supplemented by modeling of long-range communications in the EGFR regulatory dimers at a range of reference distances from 20 Å to 70 Å. For clarity, we focused discussion on two representative cases by making the following assumptions. The reference communication distance of 30 Å was used to analyze primarily long-range intra—domain (intra-monomer) communications and interfacial inter-domain communications (Figure 6). We adopted a reference distance of 60 Å to highlight the effect and contribution of very long-range inter-monomer communications (Figure 7). Analysis of long-range communications revealed important attributes that could distinguish active and inactive EGFR dimers. A dense network of long-range inter-monomer communications (reference communication distance of 60 Å) could be seen in an asymmetric dimer (Figure 7 A, C), whereas a sparse network of long-range communications was observed in a symmetric dimer (Figure 7B). An asymmetric EGFR-WT dimer could be characterized by the enhanced intra-monomer (Figure 6A, B) and inter-monomer long-range communications (Figures 7A, B) as compared to the symmetric EGFR-WT. Structural mapping revealed the enhanced long-range inter-monomer communications for an asymmetric EGFR dimer (positive ΔLRCC values correspond to improved long-range communication) across a broad spectrum of reference communication distances (Figure S6). The introduction of the EGFR-T766M mutation could enhance long-range communications in the asymmetric dimer both at the reference distance of 30 Å (Figure 6 D) and 60 Å (Figure 7 D). The increased structural integrity of the asymmetric dimer induced by EGFR-T766M could contribute to further stabilization of the active complex. A network of efficiently communicated clusters included the “integrating” αF-helix, the “supporting” αH-helix and the “mediating” αC-helix (Figure 6C). The inter-monomer coupling is due to stabilizing interactions between the αC-helix of the N-lobe of the receiver and the αH-helix and αI-helices of the C-lobe of the activator. Overall, the combined analysis of correlated motions and long-range communications pointed to the long-range inter-monomer coupling in the asymmetric dimer as an important factor that makes this structural arrangement functionally relevant for activation.
The key structural elements that may control the long-range interdomain coupling and allosteric activation include the “integrating” αF-helix and the “mediating” αC-helix, potentially playing a role of “dispatchers” in regulation of allosteric EGFR activation. Structural analysis indicated that the proper positioning of the αC-helix for activation may be controlled through the αC-β4-loop (744-SVDN-747) immediately following the α-helix [117]–[125]. In both monomers of an asymmetric EGFR dimer, the α-helix could be stabilized via hydrogen-bonding between the main-chain atoms of Ser-744 and Asp-746 of the αC-β4-loop, and the side-chain atoms of Arg-807 and Tyr-803 of the αE-helix in the C-lobe, respectively. We noticed that the residues of the αC-helix and αC-β4-loop were involved in long-range inter-monomer communication. The central interactions stabilizing the inter-monomer interface of an asymmetric dimer are formed by the αC-helix of the receiver and the αI-helix, αH-helix of the activator. These structural elements could also contribute to the network of cooperatively communicating regions that may be strengthened by the activating mutation (Figures 6, 7). Structural environment of the αC-helix of the receiver may be a key “mediator” of long-range communications that control allosteric activation of the EGFR dimer. Activation by binding to a hydrophobic patch in the N-lobe of the receiver EGFR interface is a common theme discovered in various crystal structures of an asymmetric dimer [120]–[125]. In contrast, we found that the electrostatically stabilized symmetric dimer may be lacking the effective inter-monomer communication, as the mediating αC-helix was conspicuously absent among long-range interacting regions (Figure 7C). This may be a possible mechanistic reason explaining irrelevance of this structural form for activation.
Overall, our results suggested that the effective communication and functional cross-talk between the “integrating” αF-helix and the “mediating” αC-helix, may present an important organizing principle that controls the long-range inter-domain coupling and allosteric activation. The αF-helix along with the hydrophobic and catalytic spines defines the kinase architecture and, together with the αC-helix, may control global motions of the kinase fold. Indeed, all functional motifs in the C-terminal lobe including the activation loop, the catalytic loop, the P+1 loop are anchored to the αF-helix [66]–[69]. We found that the strategic position of the αF-helix may be utilized not only as an integrating scaffold for structural arrangement of other regulatory motifs but also for long-range communications and allosteric activation. Accordingly, using the network-based description of proteins [34], [35], the αF-helix and the αC-helix may be considered as communicating hubs of the regulatory complexes and, as such, may have an impact on allosteric coupling.
Recent structural studies showed a critical importance of the “juxtamembrane latch” interactions involving the JM-B segment for activation of the EGFR and HER4 kinase domains [117], [120], [121]. The juxtamembrane segment of human EGFR is formed by the N-terminal half known as the JM-A motif (residues 645 to 663) and the C-terminal half referred as the JM-B (residues 664 to 682) [117]–[125]. However, the nature of allosteric coupling between juxtamembrane segment and the kinase domain is not fully understood. Our simulations and analysis included only the JM-B segment because this region was crystallographically resolved in both the active and inactive EGFR dimers. In an asymmetric dimer arrangement, the activator monomer B makes contacts with the receiver monomer A through interactions involving the αH-helix and αI-helix of the activator as well as the juxtamembrane region and the αC-helix of the receiver (Figures 7A, 8A). We found that the second half of the juxtamembrane segment (JM-B) of the receiver molecule could also contribute to the network of allosterically communicated residues (Figure 7A), thus revealing a functionally relevant role of this region in promoting long-range cooperativity and activation of the asymmetric dimer. In addition to the JM-B residues of the receiver molecule, the residues of the αI-helix of the activator interfacing with the JM-B segment were involved in both intra-domain (“short” range cooperativity) and inter-domain communication (“long-range” cooperativity). The recent crystal structures of the EGF receptor and HER4 kinase domains with their juxtamembrane segments have indicated that the JM-B segment could extend from the N-terminal lobe of the receiver to interact with the C-terminal lobe of the activator, thus promoting allosteric activation of the receiver [120], [121]. Our results corroborated with these recent structural studies that pointed to the importance of the juxtamembrane region in activation of an asymmetric dimer. Furthermore, our analysis could offer an additional molecular insight by showing that the juxtamembrane region could facilitate asymmetric dimer formation by assisting the central mediator αC-helix in establishing efficient long-range allosteric communication between monomers. This result was also in agreement with the biochemical experiments that showed that kinase activity of EGFR could be compromised by deletion of the juxtamembrane region [135], [136].
The concerted motions and long-range communications between the activator and receiver molecules could ensure a dynamically enhanced stabilization of the asymmetric dimer required for activation [120], [137]. The JM-B segment of the receiver molecule may act as an “allosteric linker” that could coordinate rigid body motions between the monomers and thus control the dynamics of the dimerization interface critical for activation. These observations are in line with the notion that the juxtamembrane region of EGFR plays a key part in the allosteric activation mechanism by promoting dimerization and further stabilization of the asymmetric dimer [134]–[137]. Our observations are also consistent with the evidence that mutation-based disruption of the electrostatic hook in a symmetric dimer by D979K/E981R and E980R/D982K could reactivate autophosphorylation of EGFR [110], [124]–[127]. An important function of the JM-B segment in modulating mutation-induced EGFR activation may be partly due to its dynamic role acting as the “inter-monomer hinge” during allosteric changes.
According to a recently proposed generalized model of EGFR activation, JM-A segments of both the receiver and the activator may further potentiate asymmetric dimerization and be also required for activation [120], [137], It was proposed that the JM-A segments may form coiled-coil dimer that could further enhance stabilization of the asymmetric dimer and result in activation. Analysis of allosteric communications in the framework of such generalized model of EGFR activation presents a significant computational challenge as it would require currently lacking high-resolution crystallographic information of the complete juxtamembrane region to allow for more accurate biophysical modeling. We currently pursue combined homology modeling and computer simulations of the EGFR complexes involving a complete juxtamembrane region. This investigation extends beyond the scope and focuses of the present work and will be presented elsewhere.
The hydrophobic dimerization interface of an asymmetric dimer involves the bottom of the C-lobe of the activator molecule docked on the top of the N-lobe of the receiver molecule (Figure 8). We observed that the key residues contributing to the inter-monomer interface could be involved in efficient allosteric communication and the principal interactions between these residues may be allosterically stabilized by the activating mutation (Figure 8 A, B). For instance, we observed a high occupancy of the interfacial interactions between Tyr-740 from the αC-helix (monomer A) and Asp-918 from the αH-helix (monomer B). The effect of the activating mutation could be seen in further strengthening of this contact (Figure 8 B, C). Similarly, the effect of the activating mutation was reflected in further consolidation of the interfacial hydrogen bonding interactions Glu-841--Arg-949 and Glu-842--Arg-946 between the activation loop residues (monomer A) and the αI-helix residues of the monomer B (Figure 8 B, C).
Hence, the effect of the gate-keeper mutation may be allosterically transmitted to the inter-monomer interface residues from the αC-helix (monomer A), αH-helix, and αI-helix (monomer B), supporting allosteric nature of the mutation-induced activation mechanism. Mutation-induced structural stabilization of the interdomain interface coupled with the enhanced long-range cooperativity in a functional asymmetric dimer could rationalize the existing experimental data. Indeed, EGFR activation may be suppressed by mutations in the αH-helix and αI-helix of the monomer B (R938E, I942E, and K946E) that caused the loss of kinase activity [117]. In the head-to-head structure of a symmetric EGFR dimer, two kinase monomers are stabilized by a dense network of salt bridges and hydrogen bonds that connect the kinase monomer through the C-terminal fragments (Figure 9 A) [117], [120]. Based on this crystal structure of a symmetric EGFR dimer (PDB ID 2GS7 [117]), it was initially proposed that the “electrostatic hook” formed between the C-terminal tail (residues 979–990) and the hinge region in the kinase domain may stabilize structural topology of a symmetric dimer (Figures 9 A, B). The inter-monomer interface is formed by Asp-988, Asp-990 from the electrostatic hook, Lys-822 and Lys-828 of the N-terminal of the monomer A and Lys-799, Arg-938, and Lys-946 of the C-terminal lobe of the monomer B.
We found that the inter-monomer long-range communication propensities were significantly impaired in this structural form (Figures 10 A, B). Interestingly, the EGFR-T766M mutation resulted in a noticeable reduction of the inter-monomer interaction occupancies, which were mostly determined by the “electrostatic hook” residues (Figure 9B, C). Hence, the activating mutation may lead to the weakening of the electrostatic hook, which is a critical “stapling” element protecting the inactive symmetrical dimer. We also analyzed long-range communication profiles obtained from simulations of the most recent crystal structure of a symmetric EGFR dimer that included AF-2 helices in the C-terminal tail (Figure 10) [(PDB ID 3GT8 [120]). Similarly, it appeared that the inter-monomer long-range communication (at the reference threshold of 60 Å) was reduced and only the local hydrophobic environment of the αF-helix could contribute to a long-distance cross-talk between monomers (Figure 10 A,B). In this crystal structure the formation of the juxtamembrane latch may be compromised by the “electrostatic hook” and AF-2 helices in the C-terminal tail [120]. The “electrostatic hook,” which is located near the αC/β4 loop of the kinase domain, includes Asp-979, Glu-980, and Glu-981. These residues form salt-bridges with Lys-822, Lys-828, His749, and His-826 of the kinase domain (Figure 10 C, D). The results indicate that in a symmetric dimer arrangement, which is protected by the electrostatic hook residues, the αC-helices could be blocked from establishing long-range communication and, hence, their mediating role in promoting activation could be compromised. This may present a feasible mechanism preventing formation of alternative dimer arrangements [110]. The efficient long-range cooperativity and allosteric communications may be thus an important attribute of the functional regulatory complex.
We have studied molecular mechanisms of allosteric regulation in the ABL and EGFR protein kinases by integrating multiscale simulations with computational modeling of long-range communications in the regulatory complexes. The results have unveiled organizing principles of mutation-induced activation in the ABL and EGFR kinases that may be orchestrated by a cross-talk between the integrating αF-helix and the mediating αC-helix, responsible for coordination of the inter-domain coupling between key regulatory regions. These findings are in agreement with the central involvement of αF-helix and αC-helix in regulatory functions and allosteric activation. We have shown that collective motions and the efficient inter-monomer communications between the activator and receiver molecules could allow for a dynamically enhanced stabilization of the asymmetric dimer required for activation. Hence, the effective communication between the “integrating” αF-helix and the “mediating” αC-helix may coordinate coupling between the intra-domain and the inter-domain motions and thus be important for allosteric activation. The results of this study have unveiled that structurally conserved gate-keeper mutations may serve as catalysts of kinase activation by increasing long-range communication capabilities and promoting the enhanced stabilization of the active kinase form. We suggest that structural architecture of the regulatory kinase complexes and the intrinsic dynamic equilibrium between major conformational states can define the topology of allosteric networks, while specific communication pathways may be modulated by the mutation or binding partner. The results of our study reconcile recent experimental studies of allosteric kinase mechanisms and provide a useful molecular insight into hierarchy of functional motions and mechanistic aspects of allosteric kinase signaling at the atomic level.
The coordinates of the ABL and EGFR kinase catalytic domain and regulatory complexes in various conformational states were obtained from the Protein Data Bank (PDB) (www.pdb.org) [172]. In MD simulations of the ABL and EGFR kinase domains, we used the following crystal structures : PDB ID 1IEP (inactive ABL structure) [70], PDB ID 1M52 (active ABL structure) [71], [72], PDB ID 2G1T (Src-like inactive ABL structure) [109], PDB ID 2Z60 (the active form of the ABL-T315I mutant) [75], PDB ID 1XKK (Src/Cdk-like inactive EGFR structure) [77], PDB ID 2GS7 (Src/Cdk-like inactive EGFR structure) [109], and PDB ID 2J6M (active EGFR structure) [78], and PDB ID 2JIT (EGFR-T790M mutant) [79]. In MD simulations of the ABL complexes, we employed the crystal structure of the ABL-SH2-SH3 complex in the inactive form (PDB ID 2FO0) and the active form (PDB ID 1OPL) [107], [108]. In MD simulations of the EGFR regulatory complexes we utilized the crystal structures of an asymmetric, active dimer (PDB ID 2GS6 [117]) and inactive symmetric dimers (PDB ID 2GS6 [117], PDB ID 3GT8 [120]). The juxtamembrane segment of human EGFR is formed by the N-terminal half (JM-A motif, residues 645 to 663) and the C-terminal half referred (JM-B motif, residues 664 to 682). The residues 669-682 of the JM-B motif have been determined in the crystal structures of asymmetric and symmetric EGFR dimers [117] and were included in MD simulations and subsequent allosteric communication analysis. We have also analyzed crystal structures of the EGFR asymmetric dimer in the presence of its complete juxtamembrane segment (PDB ID 3GOP [121]), asymmetric dimer of the HER4 kinase (PDB ID 2R4B [119]). All crystallographic water molecules, bound inhibitors, and other heteroatoms were removed. The retrieved structures were examined for missing and disordered residues. The missing residues and unresolved structural segments were modeled using the program MODELLER which is an automated approach to comparative protein structure modeling by satisfaction of spatial restraints [173].
MD simulations of the EGFR regulatory complexes (each of 20 ns duration) were performed from the crystal structures of an asymmetric and symmetric dimers (PDB ID 2GS6 [117], PDB ID 3GT8 [120]). MD simulations were carried out using NAMD 2.6 [174] with the CHARMM27 force field [175], [176] and the explicit TIP3P water model as implemented in NAMD 2.6 [177]. The VMD program was used for the preparation and analysis of simulations [178]. The employed MD protocol was described in full details in our earlier study [155]. In brief, the EGFR dimers were solvated in a water box with the buffering distance of 10 Å. Assuming normal charge states of ionizable groups corresponding to pH 7, 39 sodium (Na+) and 23 chloride (Cl−) counter-ions at physiological concentration of 0.15 mol/L were added to achieve charge neutrality in MD simulations of the asymmetric EGFR dimer. 33 Na+ and 17 Cl− counter-ions were added in MD simulations of the symmetric EGFR dimer. All Na+ and Cl− ions were placed more than 8 Å away from any protein atoms and from each other. Equilibration was done by gradually increasing the system temperature in steps of 20K starting from 10K until 310K and at each step 10,000 steps of equilibration was run keeping a restraint of 10 Kcal mol−1 Å−2 on the protein alpha carbons (Cα). Thereafter the system was equilibrated for 150,000 steps at 310K (NVT) and then for further 150,000 steps at 310K using Langevin piston (NPT) to maintain the pressure. Finally the restrains were removed and the system was equilibrated for 500,000 steps to prepare the system for simulation.
An NPT simulation was run on the equilibrated structure for 20 ns keeping the temperature at 310K and pressure at 1 bar using Langevin piston coupling algorithm. The integration time step of the simulations was set to 2.0 fs. The SHAKE algorithm was used to constrain the lengths of all chemical bonds involving hydrogen atoms at their equilibrium values and the water geometry was restrained rigid by using the SETTLE algorithm. Nonbonded van der Waals interactions were treated by using a switching function at 10 Å and reaching zero at a distance of 12 Å. The particle-mesh Ewald algorithm (PME) as implied in NAMD was used to handle long range electrostatic forces.
The covariance matrix between residues i and j represented by the atoms was calculated for each of the 20 ns MD simulation trajectories by averaging motions of atoms deviating from the mean structure. A total of 500 snapshots from were taken from trajectories and only was used for analysis. Translational and rotational degrees of freedom are eliminated and the average atomic coordinates, i = 1,..., N, are calculated along the MD trajectory [161]. Essential dynamics (ED) analysis reduces the dimensionality of the covariance matrix by diagonalization. This method describes global protein motions that are represented by the matrix eigenvectors and eigen values. The essential directions of correlated motions during dynamics are calculated by diagonalizing the covariance matrix .
The correlation matrix represents the correlation between the motion of atom i and of atom j, obtained from the reduction and normalization of the covariance matrix.
The eigenvectors represent the directions in the multidimensional space that correspond to independent modes of atomic motion, while the eigen values represent their corresponding amplitudes [179], [180]. The magnitudes of eigenvectors are represented by their eigen values and the principal components of protein motions are analyzed by projecting MD trajectories onto directions corresponding to the largest eigen vectors. The correlation value is the normalized covariance matrix, ranging from −1 to 1. Since PCA makes it possible to identify amplitude and direction of relevant protein motions, the projections of the first principal component (PC1) have been analyzed. The calculations were performed using the CARMA package [181] and PCA_NEST web-based service [182].
Signal propagation and inter-domain communication events in proteins can be linked to the fluctuation dynamics of atoms, defining the communication propensity (CP) of a pair of residues as a function of the fluctuations of an inter-residue distance. To evaluate communications between protein residues, we computed a communication propensity between two residues and that can be defined as the mean-square fluctuation of the inter-residue distance :where is the distance between the Cα atoms of residue i and residue j.
We investigate the modulation of long-range communication propensities as a function of activation mutation using the theoretical approach previously developed for the analysis of signal propagation [161]. To determine whether a communication between two residues is efficient, we introduced the communication threshold , an important parameter in our analysis that is considering only values associated with close amino acids. To determine the value of communication threshold, for each residue i we considered only its communication propensities with the neighbor residues between i-4 and i+4:where m is the total number of terms taken into account in the sum; Nres is the total number of residues. Therefore, represents the average communication capability between nearby protein residues and, assuming that such residues communicate well, we propose to exploit this quantity as a measure to assess if two spatially separated amino acids in the protein can communicate effectively. Exploiting this threshold definition, we say that a couple of residues efficiently communicate if and only if their communication propensity is lower or equal to . The computed average communication capability between nearby protein residues was about 0.025. We set this value as the threshold for discriminating fast communications.
By considering a given reference distance δ, we define LRCC as the density for a given residue i obtained from the fraction of residues that efficiently communicate with it () at distances larger than δ. The higher is the fraction of such residues, the higher is the capability of residue Ri to establish efficient long-range communications. This quantity simplifies analysis of the communication propensity calculations and allows to readily monitor changes in the communication capabilities of the residues upon mutations. It is helpful to consider a difference between Absolute LRCC values associated with two different simulations and obtain Relative Long Range Communication Capabilities (designated in the Results/Discussion as ΔLRCC). We typically consider the LRCC values associated with a simulation of the WT protein kinase as a reference. The resulting relative measure may highlight mutation-induced differences in communication patterns of the two proteins. The positive ΔLRCC values between simulations of the mutant and WT systems correspond to the enhanced long-range communications in the mutant.
To evaluate signal communications between protein residues in the ABL and EGFR catalytic domains, we have computed the communication propensity between all protein residues and . Exploiting a spectrum of reference threshold distances (20 Å, 30 Å), we have then computed the absolute Long Range Communication Capability (LRCC) for each protein residue in the catalytic domain, that is defined as a fraction of residues that efficiently communicate with it () at distances larger than δ = 20 or 30 Å respectively. MD simulations of EGFR dimers employed the crystal structures with missing loop residues between 724 and 726, and from 968 and 980. Additionally, MD simulations of a symmetric dimer considered two (one per monomer) additional five-residue loops (E-I-Y-G-E) that were structurally present yet disconnected from the rest of the protein system in the crystal structures. For the sake of clarity, these loops were omitted in the allosteric communication analyses. For each simulation taken into account, we determined the communication propensity between all the protein residues (excluding the five-residue loops in the “symmetric” dimer) and then computed Absolute LRCC with the reference distance of 20, 30, 40, 50, 60 and 70 Å. It is important to note that we used 20, 30 and 40 Å as a reference distance to analyze long-range intra--domains communications (or simply, long-range communications) whereas we adopted 50, 60 and 70 Å as a reference distance to highlight the effect and contribution of long-range inter--domains communications (or simply, very long-range communications). The resulting histograms allowed to scan communication efficiencies, where each bin refers to a residue and gives the fraction of residues that have high communication efficiency with it at distances larger than the cutoff.
|
10.1371/journal.pbio.1001256 | Multi-Cellular Rosettes in the Mouse Visceral Endoderm Facilitate the Ordered Migration of Anterior Visceral Endoderm Cells | The visceral endoderm (VE) is a simple epithelium that forms the outer layer of the egg-cylinder stage mouse embryo. The anterior visceral endoderm (AVE), a specialised subset of VE cells, is responsible for specifying anterior pattern. AVE cells show a stereotypic migratory behaviour within the VE, which is responsible for correctly orientating the anterior-posterior axis. The epithelial integrity of the VE is maintained during the course of AVE migration, which takes place by intercalation of AVE and other VE cells. Though a continuous epithelial sheet, the VE is characterised by two regions of dramatically different behaviour, one showing robust cell movement and intercalation (in which the AVE migrates) and one that is static, with relatively little cell movement and mixing. Little is known about the cellular rearrangements that accommodate and influence the sustained directional movement of subsets of cells (such as the AVE) within epithelia like the VE. This study uses an interdisciplinary approach to further our understanding of cell movement in epithelia. Using both wild-type embryos as well as mutants in which AVE migration is abnormal or arrested, we show that AVE migration is specifically linked to changes in cell packing in the VE and an increase in multi-cellular rosette arrangements (five or more cells meeting at a point). To probe the role of rosettes during AVE migration, we develop a mathematical model of cell movement in the VE. To do this, we use a vertex-based model, implemented on an ellipsoidal surface to represent a realistic geometry for the mouse egg-cylinder. The potential for rosette formation is included, along with various junctional rearrangements. Simulations suggest that while rosettes are not essential for AVE migration, they are crucial for the orderliness of this migration observed in embryos. Our simulations are similar to results from transgenic embryos in which Planar Cell Polarity (PCP) signalling is disrupted. Such embryos have significantly reduced rosette numbers, altered epithelial packing, and show abnormalities in AVE migration. Our results show that the formation of multi-cellular rosettes in the mouse VE is dependent on normal PCP signalling. Taken together, our model and experimental observations suggest that rosettes in the VE epithelium do not form passively in response to AVE migration. Instead, they are a PCP-dependent arrangement of cells that acts to buffer the disequilibrium in cell packing generated in the VE by AVE migration, enabling AVE cells to migrate in an orderly manner.
| The mouse visceral endoderm (VE) is a simple epithelium in the egg cylinder stage mouse embryo. Many functions associated with epithelia require them to undergo extensive remodelling through changes in the shape and relative positions of constituent cells, a process about which we understand relatively little. The anterior visceral endoderm (AVE) is a specialized group of cells in the simple epithelium of the VE, and their stereotypic migratory behaviour is essential for establishing the orientation of the anterior-posterior axis in the early mouse embryo. We show that AVE migration is linked to changes in cell packing in the VE and an increase in “rosettes,” which are striking collections of five or more cells meeting at a central point. To probe the role of rosettes during AVE migration, we have developed a mathematical model of cell movement in the VE. Simulations suggest that rosettes are not essential for AVE migration, but are crucial for the orderliness of this migration. We also explored the role of Planar Cell Polarity (PCP) signalling, which is known to coordinate cell polarization and rearrangement in many different tissues. We find that mutants in which PCP signalling is disrupted have fewer rosettes, altered epithelial packing, and abnormal AVE migration. We suggest that rosettes in the mouse VE are a PCP-dependent arrangement of cells that act to buffer the disturbances in cell packing generated by AVE migration, thereby enabling AVE cells to migrate in an orderly manner.
| Epithelia have structural and functional roles throughout embryonic development and adult life. Their organised, cohesive nature makes them ideal for lining structures and acting as selective barriers. Epithelia show distinct apical-basolateral polarity, with the apical domain characterised by junctional complexes that form tight junctions serving as a barrier to the flow of substances between cells. In addition, adherens junctions extend in a continuous belt around cells and provide structural integrity to epithelia. Many functions associated with epithelia during development, growth, disease, and repair require them to be highly dynamic whilst at the same time maintaining robust structural integrity. Most morphogenetic processes during development therefore involve extensive remodelling of epithelial tissues: branching morphogenesis in the developing kidneys, lungs, and mammary glands; development of sensory organs and ganglia from epithelial placodes; and the formation of the neural tube, to give just a few examples (reviewed in [1]–[5]).
The mouse visceral endoderm (VE) is an example of a simple epithelium with a critical developmental role. It covers the epiblast and extraembryonic ectoderm (ExE) of the egg-cylinder stage mouse embryo. Though the foetus is derived predominantly from the epiblast, it is cells of the VE that are responsible for specifying anterior pattern in the epiblast. The anterior visceral endoderm (AVE), a specialised subset of cells in the VE, is responsible for the correct orientation of the anterior-posterior axis in the mouse embryo (reviewed in [6]–[9]). At around 5½ days post coitum (dpc), cells at the distal tip of the VE differentiate to form the distinct subpopulation of the AVE, characterised by the expression of genetic markers such as Hex, Lefty1, and Cer-1 [10]–[12]. The AVE migrates proximally in a unidirectional manner and then comes to an abrupt stop at the junction between the epiblast and ExE [13]. From this position, the AVE induces anterior pattern in the underlying epiblast by restricting expression of posterior markers to the opposite side of the epiblast cup [6],[14]. In mutants such as NodalΔ600/LacZ and Cripto−/−, the AVE is correctly induced at the distal tip of the egg-cylinder but fails to migrate, leading consequently to posterior markers in the epiblast being incorrectly localised. Such embryos show severe gastrulation defects and fail to develop further [15],[16].
The driving force for AVE migration remains unclear. Dkk1, a secreted inhibitor of Wnt signalling, is expressed just ahead of migrating AVE cells and has been shown to act as a guidance cue for the AVE [17]. A relatively higher level of cell proliferation in the posterior VE has been suggested to provide the initial displacement of AVE cells towards the anterior and possibly drive their directional migration [12], however more recent results suggest that the proliferation rate in the posterior VE is not higher than that in other regions of the VE and therefore unlikely to be involved in the movement of AVE cells [18]. Time-lapse microscopy of embryos carrying a Hex-GFP transgene that marks AVE cells shows that they actively migrate over a period of 4–5 h and are extremely dynamic, showing robust protrusive activity in the direction of motion [13]. Once AVE cells reach the border of the ExE, they abruptly cease proximal movement and instead start moving laterally along the boundary, as if in response to a barrier to migration. During this lateral movement, AVE cells show fewer or no protrusions [13],[19].
Recent reports have shown that the VE retains epithelial integrity during AVE migration [19],[20]. The tight and adherens junction markers ZO-1 and E-cadherin are present continuously along all cell borders of the entire VE at all stages of migration. In addition, AVE cells must migrate within the plane of the epithelium, rather than on top of other VE cells, because the VE remains a simple epithelium only one cell layer in thickness [13]. It would therefore seem necessary for AVE cells to negotiate their way through the VE without breaking epithelial integrity. This has been verified by time-lapse studies that show that AVE cell migration involves cell intercalation [19],[20].
Our time-lapse studies of the non-AVE cells of the VE show that the cells just ahead of (more proximal to) the migrating AVE show neighbour exchange during AVE migration [20]. Moreover, like the AVE, these cells too are unable to move beyond the boundary with the ExE. VE cells overlying the ExE (ExE-VE) show dramatically different behaviour in comparison to VE cells overlying the epiblast (Epi-VE). While the Epi-VE shows robust cell movement and intercalation, the ExE-VE in contrast is relatively static and shows very little cell mixing [20]. The barrier to AVE migration therefore appears to be a region of VE (the ExE-VE) that is non-permissive of the cell rearrangements required for AVE migration.
These two regions of the VE also show differences in localisation of the molecular motors F-actin and Myosin IIA, and the Planar Cell Polarity (PCP) signalling molecules Dishevelled-2 (Dvl-2) and Vangl2 [20]. PCP signalling coordinates cell polarisation and rearrangement across fields of cells in many different contexts, such as the compound-eye and wings of Drosophila, and the mammalian neural tube (reviewed in [21]–[23]). Morphogenetic cell movements in an epithelial context have been extensively studied in the Drosophila wing-disc and germband. Convergent extension movements in the germ band are brought about by junctional remodelling that results in T1-neighbour exchanges [24] and the formation and resolution of multi-cellular rosettes (five or more cells meeting at a point, Box 1) [25]. Germband extension is also characterised by an increase in the anisotropy of cells, initially regularly packed cells becoming increasingly disordered in their packing and shape [26].
Epithelial tissues, including the mouse visceral endoderm, resemble two-dimensional networks of polygons [27],[28]. Vertex models, in which each cell is represented by a polygon with a limited set of properties, are therefore often used to simulate the tissue-level effects of forces and important cellular processes, such as growth, proliferation, and junctional rearrangements. Rauzi et al., for example, used a vertex model to show that tissue elongation can be driven by an anisotropy of cortical tension in combination with simple junctional rearrangements [29]. Aegerter-Wilmsen et al. meanwhile found that they were able to reproduce polygonal distributions in the Drosophila imaginal wing disc by including mechanical feedback as a regulator for cellular growth [30]. Several other authors have used vertex models to gain key insights into other biological phenomena [31]–[34].
Using a combination of mathematical modelling and experimental observations, we probe how the broader cell intercalation movements observed in the Epi-VE might influence AVE migration. By examining embryos at various stages of AVE migration and mutant embryos in which migration fails to take place, we show that AVE migration is specifically linked with changes in cell packing in the VE and an increase in multi-cellular rosette arrangements. To explore the role of rosettes during AVE migration, we have developed a mathematical model that simulates cell movements in the VE. This model extends previous vertex models by implementing an ellipsoidal surface to represent a realistic geometry for the mouse embryo. We also include a new type of junctional rearrangement, by allowing close vertices to join together, thus mimicking rosette formation. Simulations in which rosettes are allowed to form closely mimic experimentally observed AVE migratory behaviour. However, simulations in which rosettes are not allowed to form show abnormally disordered AVE migration (Box 1), suggesting that, while rosettes may not be essential for AVE migration, they are essential for the orderliness to this migration observed in actual embryos. These simulations closely recapitulate results from mutant embryos in which PCP signalling is disrupted and which have significantly reduced rosette numbers. AVE cells are still able to migrate to the anterior in these mutants, but do so in an abnormally dispersed, disordered manner. Our model and experimental observations together lead us to suggest that in the mouse VE, multi-cellular rosettes do not drive cell migration but rather buffer the disruption in cell packing arising from AVE migration, thereby enabling the AVE to migrate in an orderly manner.
To characterise in greater detail the changes in cellular packing in the VE that accompany and possibly influence AVE migration, we visualised apical boundaries of VE cells by staining fixed embryos for the tight junction marker ZO-1. We captured 3-D confocal image volumes of entire embryos and then opacity-rendered the image stacks. This provided volume renderings of entire embryos, so that the shape of individual cells of the surface VE and the junctions formed between them could be examined in the context of the cylindrical embryo as a whole. These experiments were performed with Hex-GFP transgenic embryos [35], in which the AVE is marked by GFP fluorescence.
In embryos in which the AVE had not yet commenced migration, cells were mostly regular in outline throughout the VE. In contrast, in embryos in which the AVE had migrated anteriorly, Epi-VE cells showed a great variety of shapes and irregular packing, though ExE-VE cells remained relatively regular in shape and packing (Figure 1A). This suggested the observed irregularities in cell shape might be related to the cell rearrangements in the Epi-VE that accompany AVE migration.
To quantify the differences in cell shape in the Epi-VE and ExE-VE at different stages of AVE migration, we counted the neighbours for each of the cells of the VE as a measure of the number of sides or polygon number of the cell [30]. Using the opacity renderings of fixed embryos, we manually identified each VE cell, noted whether it was located in the Epi-VE or ExE-VE and the number of cells that shared an edge with it. A hexagonal arrangement of cells (mean polygon number close to six) is considered to be the preferred or equilibrium packing of cells in an epithelium, and deviations from this are indicative of increased disequilibrium (Box 1) [26],[28].
We grouped embryos into four different stages of AVE migration using Hex-GFP fluorescence to determine whether the AVE had been induced and to what degree it had migrated. “Pre-AVE” embryos were those in which the AVE had not yet been induced. “Distal” embryos had the AVE induced at the distal tip, but it had not yet started migrating. In “migrating” embryos, the AVE was in the process of migrating, and in “anterior” embryos the AVE had reached the boundary of the ExE-VE (the proximal limit to migration) and was starting to spread laterally.
We compared polygon numbers in the ExE-VE and Epi-VE within each stage and found that in “pre-AVE” and “distal” embryos, the difference between mean polygon numbers in these two regions was not significant (p>0.07, Student's t test). However, in “migrating” and “anterior” embryos, the mean polygon number in the Epi-VE was significantly lower than that in the respective ExE-VE (p<0.002, Student's t test) (Figure 1B).
We compared polygon numbers across the different stages, and found no significant difference in the mean in the ExE-VE of the four stages (p = 0.25, ANOVA). However, there was a significant difference in the mean polygon numbers among the Epi-VE of the four stages (p = 0.0007, ANOVA). In pair-wise comparisons, the mean polygon number of the Epi-VE of “migrating” and “anterior” embryos were both significantly lower than that of the Epi-VE of “distal” embryos (p = 0.02, Student's t test) (Figure 1B′).
We next determined the frequency of the different polygon numbers in the Epi-VE and ExE-VE at the four stages in development. As with the mean polygon number, in “pre-AVE” and “distal” embryos, the distribution of polygon numbers in the Epi-VE was not significantly different from that in the respective ExE-VE (p>0.4, Kolmogorov-Smirnov test) (Figure S1A and B). By contrast, in “migrating” and “anterior” embryos, the distribution of polygon numbers in the Epi-VE was significantly different to that in the respective ExE-VE (p<0.02, Kolmogorov-Smirnov test), with a relatively higher proportion of four-sided cells (Figure S1C and D).
We compared polygon number frequencies in the Epi-VE across the four stages and found that it was not significantly different between “pre-AVE” and “distal” embryos (p = 0.5, Kolmogorov-Smirnov test) (Figure S2A). However, as with the mean polygon number, the frequencies of polygon numbers in the Epi-VE of both “migrating” and “anterior” embryos was significantly different from that in the Epi-VE of “distal” embryos (p<0.05, Kolmogorov-Smirnov test) (Figure S2B, C, and H). The Epi-VE of “migrating” and “anterior” embryos showed an increase in the proportion of four-sided cells at the expense of five- and six-sided cells as compared to “distal” embryos (Figure S2H), which would explain the significant reduction in mean polygon number in the Epi-VE of these stages.
The change we see in cell packing in the VE is localised to the region to which AVE migration is restricted (the Epi-VE) and to the stages during which AVE cells migrate (“migrating” and “anterior”). To verify if the change in packing of Epi-VE cells is linked specifically to AVE migration (as opposed, for instance, to the developmental stage of embryos), we examined NodalΔ600/lacZ and Cripto−/− embryos, two mutants in which the AVE is correctly specified but fails to migrate. Embryos comparable to “anterior” stage wild-type embryos in size (p = 0.41, ANOVA) and shape (Figure S3) were dissected at 5.75 dpc and their polygon numbers determined.
We found that VE cell packing in both NodalΔ600/lacZ and Cripto−/− embryos was more similar to that in “distal” embryos than to that in “anterior” embryos. In contrast to “anterior” embryos (but similar to “distal” embryos), neither NodalΔ600/lacZ nor Cripto−/− embryos showed a significant difference in mean polygon number between their Epi-VE and respective ExE-VE (p>0.18, Student's t test) (Figure 1B). The frequencies of polygon numbers in the two regions were also similar (p>0.21, Kolmogorov-Smirnov test) (Figure S1E and F). Furthermore, when compared to the Epi-VE of “distal” embryos the Epi-VE of NodalΔ600/lacZ and Cripto−/− embryos did not show a significant difference in mean polygon number (p>0.34, Student's t test) (Figure 1B′) or frequencies of polygon numbers (p>0.38, Kolmogorov-Smirnov test) (Figure S2D, E, and H). When compared to the Epi-VE of “anterior” embryos, the Epi-VE of stage-matched NodalΔ600/lacZ and Cripto−/− embryos did show a significant difference in mean polygon number (p<0.02, Student's t test) (Figure 1B′) and frequencies of polygon numbers (p<0.03, Kolmogorov-Smirnov test) (Figure S2F, G, and H). Both mutants had a lower proportion of four-sided cells and higher proportion of six-sided cells as compared to “migrating” and “anterior” embryos (Figure S2H). These data all point to a specific link between AVE migration and changes in cell packing in the Epi-VE.
To confirm this is indeed the case, we determined the polygon number of VE cells in living embryos undergoing AVE migration. We visualised cell outlines in the VE of cultured embryos by differential interference contrast (DIC) time-lapse microscopy. Embryos were transgenic for Hex-GFP, enabling us to monitor AVE migration. We captured images from five focal planes at each time-point so cell outlines could be visualised unambiguously. We imaged embryos every 15 minutes to achieve sufficient time-resolution to follow individual cells from one time-point to the next. Due to the strong curvature of the surface of the embryo, only a relatively small portion of the surface VE could be viewed in focus. We analysed five embryos, in which we tracked a total of 31 Epi-VE and 28 ExE-VE cells during AVE migration, over an average period of 4 hours. We then compared the mean polygon number of these cells at the start of the experiment (when the AVE was at the distal tip of the embryo) with the mean polygon number of the same cells at the end of the experiment (when the AVE was in the process of migrating). The mean polygon number of the tracked Epi-VE cells was significantly lower during AVE migration compared to before the AVE had started migrating (p = 0.036, Student's t test on paired samples) (Figure 1C). The change in the mean polygon number of the “control” ExE-VE cells tracked during this same period was not significant (p = 0.238, Student's t test on paired samples). These results, together with the results from fixed wild-type and mutant embryos, strongly suggest migration of AVE cells is specifically accompanied by a reduction in mean polygon number in the Epi-VE and a shift away from the equilibrium cell packing arrangement.
Renderings of the VE surface revealed a variety of junctions between cells. In addition to junctions where three cells meet at a point (typical of idealised hexagonal arrays of cells), we also frequently observed four-cell junctions and five or more VE cells meeting at a central point to form rosette arrangements (Figure 2A).
Rosettes typically comprised between five and seven cells, occasionally with one or two cells contributing to two distinct rosettes (Figure 2B). The majority of cells involved in rosettes were non-Hex-GFP expressing, though 8% of rosettes also included Hex-GFP cells (n = 51 rosettes). Examination of confocal sections and segmentation of rosettes to separately render individual cells in the context of the surrounding VE confirmed that rosettes are comprised of a single layer of cells, with all cells of the rosettes in contact with the epiblast (Movie S1).
Multi-cellular rosettes are characteristic intermediaries of long-range coordinated cell rearrangements during germband extension in Drosophila [25]. Together with the fundamental mechanism of T1 neighbour exchange [24], they are understood to drive convergent extension movements in the germband. To determine what role rosettes might play in the context of the mouse VE where no such convergent extension movements have been reported, we quantified rosette numbers in fixed embryos. As with our analysis of polygon numbers, we categorised embryos into four groups: “pre-AVE,” in which the AVE had not yet been induced; “distal,” in which the VE was at the distal tip, prior to migration; “migrating,” where the AVE was in the process of migration; and “anterior,” in which the AVE had reached the endpoint to proximal migration and had started moving laterally. We manually scored multi-cellular rosettes in opacity renderings of ZO-1 stained embryos for each category. To correct for any differences in the number of cells in the VE present and able to contribute to rosette formation, we divided the number of rosettes by the total number of VE cells for that embryo. We refer to this value as the rosette “density.” The average rosette density was then calculated for each group.
Average rosette density was significantly different across the four groups (p = 0.025, ANOVA). We found a progressive increase in rosette density from “pre-AVE” to “distal” to “migrating” stages (Figure 3A). “Migrating” embryos had a significantly higher rosette density than “distal” and “pre-AVE” embryos (p<0.05, Student's t test). Rosettes' density decreased slightly from “migrating” to “anterior” stages, but not in a significant manner (p = 0.14, Student's t test).
The significant increase in rosette density during AVE migration suggests that rosette formation might be linked specifically to AVE migration. To confirm that this is indeed the case, we assessed rosette numbers in a double blind manner in NodalΔ600/lacZ and Cripto−/− embryos, two mutants in which the AVE is correctly specified but fails to migrate [15],[16]. Embryos were dissected at 5.75 dpc, comparable to “anterior” stage wild-type embryos in size (Figure 3A′, p = 0.90, ANOVA) and shape (Figure S3) and rosette numbers determined. Mutants of both lines showed a significant reduction in rosette density when compared to both “anterior” and “migrating” stage embryos (Figure 3A) (p<0.01, Student's t test). Both mutants had a significant reduction in the average number of rosettes (Figure 3A′), leading to the observed reduction in rosette density.
To determine if rosettes are restricted to any one region of the VE, we plotted their distribution with respect to the future anterior and the boundary between the epiblast and ExE. Rosettes showed a strong bias in distribution with respect to the boundary between the epiblast and ExE, being located almost exclusively in the Epi-VE, the region to which AVE cell migration is restricted. Within the Epi-VE, they did not show any bias in distribution with respect to the presumptive anterior (Figure 3B).
Rosette numbers increase during AVE migration (Figure 3A′), suggesting they are not static features of the VE. They are found predominantly in the Epi-VE, which is characterised by cell mixing [20]. To determine if rosettes in the mouse VE form by cell movement (as opposed, for example, to stereotypic patterns of cell division, or apoptosis of one cell drawing surrounding cells into a central point), we visualised cell outlines in the VE of cultured Hex-GFP embryos by DIC time-lapse microscopy. As before, we captured images from five focal planes at each time-point so cell outlines could be visualised unambiguously, and with a 15-minute time-lapse to achieve sufficient time-resolution to follow individual cells from one time-point to the next. Again, due to the strong curvature of the surface of the embryo, only a relatively small portion of the surface VE could be viewed in focus. In five embryos that remained in focus and in the field of view continuously for between 2 and 7 hours, we recorded a total of five rosettes forming—one rosette in each of three embryos and two rosettes in a fourth embryo. All these rosettes formed as a result of VE cells intercalating so that five or more cells met at a single central point to form a rosette. We did not observe apoptosis or cell division leading to rosette formation in any of these embryos. Cell tracking confirmed that in forming rosettes, cells that initially were not in contact with one another became neighbours (Figure 4 and Movies S2 and S3). Consistent with the distribution of rosettes in fixed embryos, we observed rosettes forming only in the Epi-VE.
We did not observe any rosettes resolving in our time-lapse recordings, suggesting that if they do resolve, it is on much longer time scales to their formation. We quantified rosettes in opacity renderings of ZO-1 stained 6.5 dpc embryos, approximately 20 hours after AVE migration, and found that while overall rosette density was significantly lower compared to “anterior” embryos, the average number of rosettes per embryo was significantly higher (Figure S4).
Our experimental observations show that AVE migration is accompanied by a decrease in mean polygon number in the Epi-VE and an increase in the number of rosettes. To explore possible roles for rosettes, we created a mathematical model that represents AVE migration within the mouse VE. A critical feature of the model is the ability to adjust the number of rosettes that form during migration simulations by changing a single parameter. We are thus able to observe how varying rosette numbers affects the emergent migration behaviour, whilst keeping all other parameters constant. Such computational experiments were intended to demonstrate whether rosettes are an important part of the migration process, or merely coincidental. We have recently described a 2-D version of such a model [36].
In our model, the apical surfaces of cells of the VE are represented by polygons lying on the surface of an ellipsoid. The polygonal representation is an abstraction of the cell shapes observed in vivo and captures key features such as edge- and neighbour-numbers. This framework is one of a class of cell-based models, including, for example, the cellular Potts model [37] and the cell-centre model [38]. Of these models, the vertex representation is the most appropriate in the context of AVE migration as it permits the explicit modelling of junctional rearrangements including rosette formation.
The numerous forces acting on each cell in vivo are encapsulated by tension and pressure forces acting on the vertices of the polygonal cells. The directions in which these forces act in two-dimensions are shown in Figure 5A. To extrapolate to a three-dimensional ellipsoid, the forces act tangentially to the surface at each vertex (Figure 5B). Each cell also has a volume that is able to change over time. The cell's height along the apical-basal axis can be inferred by dividing the volume by apical surface area.
The equation for the tension force acting on a vertex due to one of the cells to which it belongs is given by:where CL and CP are constants, lc and la are the lengths of the clockwise and anti-clockwise edges, respectively, and p is the length of the cell perimeter (Figure 5B,C).
The pressure force equation, meanwhile, is given by:where CA, CH, and CD are constants, a is the cell area, at is a target area, H is the height-to-area ratio, θ is the average internal angle of the cell (θ = π(s−2)/s for an s-sided polygon), is the internal angle at the current vertex (see Figure 5A), and n1. and n2 are integers.
We note that the exact form of the force equations does not affect the qualitative behaviour of our simulations. More information about the tension and pressure force equations can be found in Text S1.
By summing the contributions to the total force from each cell, an equation of motion for each vertex can be formulated. In this type of biological system, viscous forces dominate, and we therefore make the simplifying assumption that inertial forces can be neglected. The only additional parameter in the equations of motions is thus a viscosity parameter. The equation of motion for a vertex i is given by:where μi is the viscous coefficient, xi is the vertex position, and Fi is the sum of all forces acting on the vertex.
The equations are solved iteratively, with vertices free to move anywhere in 3-D space. In vivo the cells of the VE adhere to the epiblast and extra-embryonic ectoderm below, maintaining the shape of the embryo. To simulate this restoring force, vertices are therefore projected back to the ellipsoid during each iteration (Figure 5D). The time-step in our simulations is kept sufficiently small so that this projection is small relative to the movement of the vertices.
In vivo, cells in the Epi-VE are highly labile relative to those of the ExE-VE [20]. To simulate this fact, we adjust the relative viscosity of the vertices in each half of the ellipsoid. A higher viscosity μ in the ExE-VE ensures that movement is more restricted in the proximal half of the embryo. In this way we are able to simulate the barrier to migration that occurs at the junction between the Epi-VE and ExE-VE.
Alongside the standard vertex movements driven by the forces described above, two types of junctional rearrangement have been observed experimentally, and are therefore included in the model. The first is a T1 transition, which has been used in many previous vertex models (e.g. Weliky and Oster [31], Farhadifar [32]). Secondly, an edge whose length falls below a certain threshold is allowed to contract to a single point, with the vertices at the ends of the edge joining together. Rosettes of various sizes occur when several neighbouring edges contract in succession. This is a key process in the model, allowing the effect of rosettes on migration to be investigated. The number of rosettes can be controlled by adjusting the threshold length at which vertices join together. Increasing the threshold leads to more rearrangements, while decreasing it leads to fewer rearrangements.
During AVE migration in vivo, cells grow in volume and proliferate, and the size of the embryo increases. In order for our model to be realistic it is important to include these processes. Each cell is assigned an initial volume, which grows logistically over time. Cell division is implemented stochastically, based on the ratio of cell volume to some target (see Text S1 for details). To simulate the concurrent increase in embryo size, the ellipsoid itself is allowed to grow over time. This requires an adjustment of the equations for the projection of vertices back to the ellipsoid surface (see Text S1 for details). The radius of the ellipsoid grows linearly, and over the course of migration increases by approximately 10%, in agreement with experimental observations.
We designate a subset of cells at the distal tip of our ellipsoid to be the AVE and induce them to migrate by adjusting the forces acting on their vertices (Figure 5D). This is achieved in practice by increasing the pressure force at one or more of the proximal-most vertices of each migrating cell. Increasing this force causes those vertices to move, which in turn affects the properties of the cell and results in the whole cell moving proximally. In reality migrating cells show protrusions in the direction of cell movement that can be several cell diameters long [13],[19]. Our migration force can therefore be thought of as the reaction of the main body of the cell to the directional cues provided by the protrusions.
We initially simulated AVE migration with the vertex-joining threshold set at a level that allowed rosettes to form at a similar density to that observed experimentally. The AVE cells migrated in a manner similar to that seen in embryos, as an orderly, coherent group of cells. It was also found that cells ahead of the AVE were pushed against the ExE-VE forming a “crescent” shape very similar to that observed in embryos (Figure 6A,C and Movie S4).
To further test if our simulations were reasonable representations of experimental observations, we quantified polygon numbers both early and late in simulation (roughly equivalent to “distal” and “anterior” embryos, respectively).
As in cultured wild-type embryos, during simulations the Epi-VE underwent a significant reduction in mean polygon number (p<0.001, Student's t test) (Figure 5E). We also compared the frequency of different polygon numbers in the simulations. Similar to our observations in embryos, there was a significant shift in the frequencies of polygon numbers in the Epi-VE late in simulation as compared to early in simulations (p<0.001, Kolmogorov-Smirnov test) (Figure 5F), with a marked increase in the proportion of four-sided cells at the expense of six-sided cells.
Simulations were then run with a small vertex-joining threshold distance, thereby reducing the number of rosettes that form. All other parameters were kept constant. In this case AVE cells were able to migrate round the surface of the ellipsoid to the boundary with the ExE-VE, but in a dispersed manner not normally observed in embryos (Figure 6B and Movie S4). In these simulations the AVE breaks up into several clumps of cells with non-AVE cells between them, rather than maintaining its structure as a single coherent group.
The simulations suggest that the formation of rosette arrangements in the VE during AVE migration is required for the normal, orderly migration of AVE cells.
PCP signalling coordinates cell polarisation and rearrangement across fields of cells in a variety of contexts. PCP signalling is disrupted in the ROSA26Lyn-Celsr1 mouse line [20]. To determine if rosette formation is perturbed in these mutants, we quantified them in mutant embryos dissected at 5.75 dpc, a stage comparable to the wild-type “anterior” group. Mutants had a significantly reduced rosette density when compared to “anterior” embryos (p<0.05, Student's t test) (Figure 7A). ROSA26Lyn-Celsr1 embryos are similar in size to “anterior” embryos and the reduction in rosette density is the result of a significant reduction in the average number of rosettes per embryo (p<0.001, Student's t test) (Figure 7A′).
The AVE migrates in ROSA26Lyn-Celsr1 mutants, but in the majority of cases (six out of eight embryos) was abnormally dispersed, in a manner reminiscent of simulations in our model when rosettes were not allowed to form (Figures 7B–C and 6B). These mutants also show a variety of other AVE migration abnormalities such as unilateral whorls or migration into the ExE-VE [20].
We determined the polygon numbers of VE cells in ROSA26Lyn-Celsr1embryos. As with wild-type “anterior” embryos, the mean polygon number was significantly lower in the Epi-VE compared to the ExE-VE (p<0.001, Student's t test) (Figure 7D). Similarly, the frequency of polygon numbers in the ExE-VE and Epi-VE was found to be significantly different (p≤0.001, Kolmogorov-Smirnov test) (Figure S1G).
Interestingly, when compared to the Epi-VE of wild-type “anterior” embryos, the polygon number in the Epi-VE of ROSA26Lyn-Celsr1 embryos was significantly lower (p<0.05, Student's t test) (Figure 7D), suggesting that there was increased disequilibrium in Epi-VE cell packing in the absence of rosettes.
Prior to AVE migration, the distribution of cell polygon number is comparable in the Epi-VE and ExE-VE, with a peak between five and six sides. This distribution is different from the equilibrium distribution reported by Gibson et al. for a variety of metazoan epithelia that have a distinct peak at six-sided cells [28]. One possible explanation for this difference is that while the epithelia considered by Gibson et al. are all relatively flat (Drosophila wing imaginal disc, Xenopus tail epidermis, and Hydra external epidermis), the mouse VE is very highly curved with an average of fewer than 20 cells around a circumference of about 300 microns. This is likely to impose different constraints on the packing of cells in the VE when compared to other epithelia.
During AVE migration stages, mean polygon number drops and polygon distribution shifts towards three- and four-sided cells, but only in the Epi-VE (Figure 1B, B′, and Figure S2H). The ExE-VE in contrast does not show so marked a reduction in mean polygon number. This is consistent with time-lapse data which show that the Epi-VE and ExE-VE are distinct in their behaviour, the former undergoing a great deal of cell mixing with cells continuously changing shape, while the latter is relatively static [20]. A specific link between AVE migration and changes in epithelial topology is reinforced by NodalΔ600/lacZ and Cripto−/− embryos in which the AVE fails to migrate and in which the mean polygon number in the Epi-VE remains close to that in wild-type embryos in which the AVE has not yet started migrating (Figure 1B and B′).
A reduction in mean polygon number is also observed in the Epi-VE of cultured embryos, where the same set of VE cells is monitored during AVE migration. This indicates that the reduction in mean polygon number is due at least in part to dynamic changes in the packing of existing VE cells taking place on the time scale of 4 hours rather than, for example, new cells with fewer cell edges arising through division. Again, the change in polygon number is restricted to Epi-VE cells, consistent with this being the region that is behaviourally labile and to which AVE cell migration is restricted [20]. These findings suggest that during AVE migration the Epi-VE is in a state of increased disequilibrium with respect to cell packing.
We observe multi-cellular rosettes in the Epi-VE, a striking conformation of cells that deviates greatly from the hexagonal packing considered to be the equilibrium arrangement of cells in epithelia. In the Drosophila germband, rosettes have been shown to be transient intermediaries of the long-range coordinated cell movements of convergent-extension [25]. There are, however, no convergent-extension movements in the mouse VE and rosettes appear to play a different role in this context.
The significant increase in rosettes during AVE migration in wild-type embryos and the reduction in rosettes in mutants with a failure of AVE migration point to a specific role for rosettes in AVE migration (Figure 3A). This is further supported by the observation that rosettes are predominantly found in the Epi-VE, the region of the VE to which AVE migration is restricted. However, rosettes are not restricted to the anterior region of the Epi-VE but more or less evenly distributed throughout the Epi-VE (Figure 3B), with only a minority of rosettes (8%) including any Hex-GFP positive AVE cells. This suggests that rosettes are not involved specifically in driving AVE cell movement or determining the direction in which they migrate, but play a more general role in the Epi-VE during AVE migration.
Our mathematical model predicts that rosettes are essential for ordered migration, in which the AVE cells migrate as a coherent group. When simulations are run with fewer rosettes, AVE migration still takes place, but in an abnormally dispersed manner. It is only when rosettes are allowed to form that AVE migration is much more orderly and closely resembles that seen in actual embryos. This is confirmed by experiments using ROSA26Lyn-Celsr1 mutant embryos in which PCP signalling is disrupted [20] and significantly fewer rosettes are formed. Such embryos exhibit AVE migration but in an abnormally disordered fashion. Rosettes in the mouse VE are therefore not essential to drive AVE migration (in the sense they are understood to contribute to convergent extension in the Drosophila germ band), but appear to have the subtler role of modulating AVE migration so that it occurs in a stereotypic, orderly manner.
AVE cells have been shown to migrate in response to a directional cue from Dkk1 [17]. AVE cells migrate within an intact epithelial sheet by cell intercalation [19],[20]. It is not only AVE cells that show this intercalatory behaviour, but also other surrounding cells in the Epi-VE [20]. This suggests that intercalation among AVE and non-AVE cells in the Epi-VE needs to be coordinated, to allow AVE cells to “negotiate” their way through a field of Epi-VE cells to arrive at the prospective anterior. Our time-lapse experiments show that rosettes form as a result of cell intercalation and that the majority of cells participating in rosettes, though in the Epi-VE, are not AVE cells. PCP signalling is active in the Epi-VE and influences AVE migration [20]. When PCP signalling is disrupted, there are significantly fewer rosettes though the AVE still migrates (albeit abnormally), suggesting that rosette formation is not a passive response to AVE migration but is actively dependent on PCP signalling.
We interpret these results to suggest the following working model of AVE migration. Though AVE cells migrate in response to an extrinsic guidance cue, since they have to migrate through an intact epithelium, this movement has to be achieved through cell intercalation that has to be coordinated between the migrating AVE cells and surrounding non-AVE cells. We suggest that the role of PCP signalling in the Epi-VE is to coordinate this intercalation, at least in part via the formation of rosettes. We suggest rosettes facilitate orderly AVE migration by buffering the increased disequilibrium in cell packing in the Epi-VE accompanying the directional movement of AVE cells. Consistent with this view, after AVE migration the mean polygon number in embryos with disrupted PCP signalling is significantly lower than that in the Epi-VE of equivalent stage wild-type embryos, indicative of increased epithelial disequilibrium in the absence of rosettes.
How might rosette formation buffer the disequilibrium of cell packing in the Epi-VE? One possibility is that it allows non-AVE cells to group together and behave as a single unit, which in some way makes it easier for AVE cells to migrate through the VE epithelium. Though we observe several rosettes forming in time-lapse experiments, we do not observe any rosettes resolving. This suggests either that once formed they are relatively static features or that they resolve over different time-scales than those over which they form. Rosette density in 6.5 dpc embryos (approximately 20 hours after AVE migration) is significantly lower than that in “anterior” embryos, but this is due to the significant increase in size of embryos between these two stages rather than to a reduction in the number of rosettes. A total of 6.5 dpc embryos have a significantly higher average number of rosettes per embryo as compared to “anterior” embryos (Figure S4), consistent with the notion that rosettes formed during AVE migration might accumulate over time rather than resolve. A detailed study of the dynamics of rosettes will help address how precisely rosettes aid in the orderly migration of AVE cells, the mechanistic basis for their formation, and clarify whether they resolve. Recent developments in high resolution, low photo-damaging imaging technology such as light sheet microscopy [39],[40] now make it feasible to monitor cell movements on the surface of the cylindrical embryo over extended time-scales and will help resolve these issues.
In contrast to convergent-extension movements where all cells undergo a coordinated medio-lateral intercalation leading to tissue elongation, during AVE migration a subset of cells migrates directionally within a larger field of cells that undergoes cell rearrangement without extensive changes to the overall shape of the epithelium. Since the VE is arranged as a cylinder, it provides an appropriate model for the study of cell movements in other epithelia on elongated curved surfaces, such as lung buds, ureteric buds, or developing intestinal villi.
Our mathematical model of cell movements in the VE, in combination with experimental intervention, provides a powerful tool for the study of directed cell movements within epithelia. It is built on simple assumptions, incorporating forces acting upon cells, cell division, directional movement of a subset of cells, a behavioural “barrier” to migration, and the ability of cells to rearrange to form rosettes. Although the cells in our model have volume and height, they are not fully 3-D, in the sense that forces act only on apical surfaces, and there is no consideration of the fact that neighbouring cells might be at different heights. As further biological data are obtained, 3-D vertex models such as that of Honda et al. [41] may become desirable in exploring the cellular dynamics of epithelia such as the VE. However, representing the tissue as a 2-D sheet as we have currently done has proved informative in exploring the role of rosettes. From just the starting conditions of our model, behaviour emerges in simulations similar to that observed experimentally—for example, the formation of a “crescent” where cells ahead of the AVE are pushed against the ExE-VE, the reduction in mean polygon number during migration, and the abnormally broad and disordered migration of AVE cells when rosettes are not allowed to form. This emergent behaviour reinforces the potential of our model as a tool in probing cell migration in the VE and other epithelia.
Genetically modified mice were maintained on a mixed C57Bl/6 CBA/J background. The Hex-GFP line was bred into the various mutant backgrounds to enable the AVE to be followed. Embryos carrying the Hex-GFP transgene were obtained by crossing homozygous Hex-GFP studs with CD1 females (Charles River). All mice were maintained on a 12 hour light, 12 hour dark cycle. Noon on the day of finding a vaginal plug was designated 0.5 dpc. Embryos of the appropriate stage were dissected in M2 medium (Sigma) with fine forceps and tungsten needles.
Secondary only controls were done to verify the specificity of secondary antibodies. Embryos were fixed in 4% PFA in PBS at 4°C for 30 minutes; washed at room-temperature thrice for 5 minutes each in 0.1% Triton-X100 in PBS (PBT); incubated in 0.25% Triton-X100 in PBS for 15 minutes; washed thrice in PBT; blocked with 2.5% donkey serum, 2.5% goat serum, and 3% Bovine Serum Albumin (BSA) in PBT for 1 hour; incubated overnight at 4°C in primary antibodies diluted in 100 µl PBT; washed five times in PBT for 5 minutes each, with a final additional wash for 20 minutes; incubated at room temperature in the appropriate secondary diluted in 100 µl PBT for 2 hours or overnight; washed in PBT five times for 5 minutes and once for 15 minutes; and finally mounted with Vectashield mounting media containing DAPI (Vector Labs H-1200). Antibodies used were: Rabbit anti-ZO-1 (Zymed laboratories 61-7300) 1∶100 and Alexa Fluor 555 donkey anti-rabbit IgG (Invitrogen A-31572).
Fixed samples were imaged on Zeiss LSM 510META and Zeiss LSM 710 confocal microscopes using 20×/0.75NA or 40×/1.2NA lenses as appropriate. DAPI was excited at 405 nm, EGFP at 488 nm, and Alexa Fluor 555 at 543 nm. Z-stacks of entire embryos were acquired at a 0.8 µm interval using non-saturating scan parameters. Z-stacks of embryos were opacity rendered as 3-D volumes using Volocity Software (Improvision, UK). Figures were prepared with Adobe CS2 Photoshop and Illustrator (Adobe Inc).
Opacity rendered views of embryos were rotated through 360°, printed out, and the polygon number of each cell determined manually as the number of neighbours it had. Each cell was given a unique reference number to avoid being counted twice. Data were tabulated in Microsoft Excel and Apple Numbers 2009. Statistical analysis was performed using SPSS Statistics 17.0 and Apple Numbers 2009.
Culture media consisted of 50% home-made heat-inactivated mouse serum and 50% CMRL (Invitrogen) supplemented with L-glutamine, equilibrated at 37°C and 5% CO2 for at least 2 hours prior to imaging. Embryos were transferred into the pre-equilibrated media in Lab-TekII Coverglass bottomed eight-well rectangular chambers (Nalge Nunc International) and imaged for up to 8 hours on an inverted Zeiss 710 confocal microscope equipped with an environmental chamber to maintain conditions of 37°C and 5% CO2. Embryos were imaged with a water immersion 40×/1.2 NA objective every 15 minutes. At every time point, a Z-stack of five focal planes separated by 10.78 µm was captured. EGFP marking AVE cells was excited at 488 nm and DIC images were acquired with the confocal's transmitted light PMT.
Antibody stained confocal imaged embryos were recovered from slides; washed in syringe filtered PBT thrice for 5 minutes; washed in lysis buffer (50 mM Tris HCl pH 8–8.5, 1 mM EDTA, 0.5% Tween-20) for 5 minutes; transferred into PCR strips containing lysis buffer (16 µl for 5.5 dpc embryos) and Proteinase K (1 µl 20 mg/ml PK per 25 µl of embryo lysis buffer); lysed at 55°C for 1 hour; and the Proteinase K inactivated by incubating at 95°C for 10 minutes. PCR genotyping was performed using 3 µl of lysed embryo as template, the appropriate primers, and Illustra PuReTaq Ready-To-Go PCR Beads (GE Healthcare Catalogue No. 27-9557-01 (0.2 ml tubes/plate 96)). Cripto mutants were identified by their failure of AVE migration phenotype.
Primers for ROSA26Lyn-Celsr1: R1: 5′AAAGTCGCTCTGAGTTGTTAT3′; R2: 5′GCGAAGAGTTTGTCCTCAACC3′; R3: 5′GGAGCGGGAGAAATGGATATG3′. Bands expected: 250 bp mutant (R1+R2) and 500 bp (R1+R3).
Primers for NodallacZ: LacZ-5: 5′CCGCGCTGTACTGGAGGCTGAAG3′; LacZ-3: 5′ATACTGCACCGGGCGGGAAGGAT3′; A: 5′ATGTGGACGTGACCGGACAGAACT3′; B: 5′CTGGATGTAGGCATGGTTGGTAGGAT3′. Bands expected: 750 bp mutant and 700 bp.
Primers for NodalΔ600: Δ600-5: 5′GCTAGTGGCGCGATCGGAATGGA3′; Δ600-6: 5′AAGGGAAGTGAACTGGAAAGGTATGT3′. Bands expected: 350 bp mutant and 950 bp.
|
10.1371/journal.pcbi.1004289 | PF2 fit: Polar Fast Fourier Matched Alignment of Atomistic Structures with 3D Electron Microscopy Maps | There continue to be increasing occurrences of both atomistic structure models in the PDB (possibly reconstructed from X-ray diffraction or NMR data), and 3D reconstructed cryo-electron microscopy (3D EM) maps (albeit at coarser resolution) of the same or homologous molecule or molecular assembly, deposited in the EMDB. To obtain the best possible structural model of the molecule at the best achievable resolution, and without any missing gaps, one typically aligns (match and fits) the atomistic structure model with the 3D EM map. We discuss a new algorithm and generalized framework, named PF2 fit (Polar Fast Fourier Fitting) for the best possible structural alignment of atomistic structures with 3D EM. While PF2 fit enables only a rigid, six dimensional (6D) alignment method, it augments prior work on 6D X-ray structure and 3D EM alignment in multiple ways:
Scoring. PF2 fit includes a new scoring scheme that, in addition to rewarding overlaps between the volumes occupied by the atomistic structure and 3D EM map, rewards overlaps between the volumes complementary to them. We quantitatively demonstrate how this new complementary scoring scheme improves upon existing approaches. PF2 fit also includes two scoring functions, the non-uniform exterior penalty and the skeleton-secondary structure score, and implements the scattering potential score as an alternative to traditional Gaussian blurring.
Search. PF2 fit utilizes a fast polar Fourier search scheme, whose main advantage is the ability to search over uniformly and adaptively sampled subsets of the space of rigid-body motions. PF2 fit also implements a new reranking search and scoring methodology that considerably improves alignment metrics in results obtained from the initial search.
| In recent years, advances in cryo-electron microscopy (cryoEM) and three-dimensional (3D) image reconstruction have made it possible to determine the structures of molecular complexes, to sub-nanometer resolutions. These reconstructed 3D cryoEM maps provide a unique challenge, since the resolutions are often sufficient to resolve a subset of the secondary structural features (e.g., long α-helices), but not high enough to unambiguously identify others (e.g., short α-helices, β-sheets, α-helix pitch, or overall connectivity of proteins). Various types of hierarchical structure refinement models of the data, including the atomistic model match and fitting, which is the subject of this paper are crucial in helping derive a more complete understanding of the structure and function relationships of biological complexes. Our protocol offers several advantages over existing fitting techniques. We introduce three new scoring terms to evaluate the quality of fitting, and the optimization of these functions lead to better predictions than existing tools like ADP-EM and Colores. Furthermore, we have adopted a non-uniform FFT-based search which is not only faster than regular FFT, but it also enables one to selectively perform more refined searches in localized regions, which is specially useful when fitting a small molecular component into a larger symmetric or asymmetric macromolecular structure.
| Protein structural data is available in primarily two forms. Atomistic scale structures (or atomic structures for short), acquired through X-ray or nuclear magnetic resonance (NMR) imaging, contain information fine enough to localize the position of most, if not all, the atoms of the protein. However these imaging modalities do not allow a complete picture of the protein’s solvent-induced state. Three dimensional (3D) electron microscopy (EM) maps, reconstructed by single particle (SP) or electron tomography (ET), are at a lower resolution but are easier to obtain and probably closer to the functional native state. A relevant problem of computational structural biology is to reconcile these forms of protein structure data, producing a refined protein model that combines the finer resolution information in the former with the native-state information at lower resolution in the latter. Different frameworks or computational pipelines like comparative modeling, e.g. [1–5] and ab initio modeling, e.g. [6], have played an increasingly important role in this kind of structure determination referred to as the fitting problem. The fitting problem can be solved for either rigid-body (6D) or flexible motions (6D rigid body motion + flexible dimensions) of the atomic structure. In this work, we address aspects common to both problems, and demonstrate results here only on rigid-body fitting.
Approaches to the fitting problem begin by defining a score between an orientation of the atomic structure 𝓟 and the 3D EM map 𝓜. A majority of past work uses the cross-correlation score (CCS) between 𝓜 and a synthesized 3D EM map 𝓜𝓟 generated from 𝓟. The CCS is widely used because it is intuitive, easy to implement, and amenable to Fast Fourier transform-based correlations, discussed below. Variants of the CCS include the core-weighted or the Laplacian-filtered CCS [1–4, 6, 7] or normalized cross-correlation (NCC) [8, 9]. There have also been a number of other scoring functions. For instance, the external-total ratio (ETR) measures the total number of atoms of 𝓟 outside a given iso-contour of 𝓜 [10], the vector matching score measures the inner product between a set of vectors representing 𝓟 and 𝓜 [11], while in [12] isosurfaces are matched by comparing surface normals.
A recent review of scoring functions for cryo-EM fitting can be found in [13]. All scoring functions depend on representing 𝓟 (respectively 𝓜), in terms that render it mutually intelligible to 𝓜 (respectively 𝓟). The usual choice, and not necessarily the best choice, for the representation involves blurring 𝓟 by placing a Gaussian at each of its atomic centers. We introduce two representations, termed non-uniform inclusion potential and scattering potential, and show that the scattering potential results in better prediction accuracy. We discuss the details of the terms in the next section and perform a comparative analysis in the Results section.
Once a scoring function is chosen, an algorithm searches for its optima over the space of rigid-body transformations of the protein. Hereafter, we refer to this space as the motion group SE(3). Search algorithms can be usefully distinguished by whether they find local or global extrema of the scoring function. Local optimization is typically synonymous with a variant of steepest ascent [10, 14], although more powerful techniques such as Powell optimization [15] and quadratic programming [5] have also been used. In global optimization, the contest is between Monte Carlo- and Fast Fourier Transform (FFT)-based algorithms. Monte Carlo-based fitting algorithms [4, 16, 17] are able to step past local optima on their way to a close-to-optimal solution; they are easy to implement and widely documented in the literature. Exhaustive or Fourier-based approaches exploit the fact that it is beneficial if the computation of the objective function can be done relatively fast. Fourier-based, deterministic approaches [3, 7, 18–20] guarantee that the found solution is within a user defined error margin of the optimum. Thus they offer a compelling trade-off between accuracy and computation time especially when combined with parallelization techniques or other hardware specific speed-ups, e.g. [9].
We adopt a variant of FFT, the non-uniform SO(3) Fourier transform (NFSOFT) [21] which not only provides better asymptotic computational complexity, but also is specially suited for better sampling of SE(3) and adaptive local searches.
An important aspect of the search procedure is a suitable sampling of the motion group SE(3). Usually the product property SE(3) = ℝ3 × SO(3) is exploited for these samplings, where SO(3) denotes the group of three-dimensional rotations, (cf. [22]). Crucial to sampling on SE(3) is sampling of the rotational subgroup SO(3). There are several existing techniques that, given an angular sampling criterion, provide a set of samples that are uniform with respect to accepted metrics of uniformity [23–25].
The paper [7] discusses fast rotational matching, i.e., it omits the translational part of the matching procedure which we incorporate. Hence, the series expansion of the scoring functions used in their work is different, as it uses spherical harmonics but not Laguerre polynomials for the radial part of the function. In contrast to that [19] considers rigid-body motion with rotation and translation. They use a ℝ1 × 𝕊2 × SO(3) parameterization of the motion group that is different from ours. Their affinity functions are expanded on terms of spherical shells of different radii while we use a decomposition directly on ℝ3 using radial wavefunctions in addition to spherical harmonics only. In addition to that our fitting algorithm uses adaptive low-discrepancy samplings, cf. [24] that better reflect the underlying geometries of sphere and rotation group.
After a suitable sampling is obtained the essential mathematical tool needed is the fast calculation of the discrete Fourier transform on the rotation group SO(3) to evaluate the correlation integral that is the objective function. There are several methods to efficiently evaluate Fourier transforms specifically on SO(3) [21, 26, 27]. There are also works that tackle Fourier transforms on the entire motion group SE(3), [28, 29]. The use of fast and efficient algorithms to evaluate the Fourier transform on non-uniformly distributed points, cf. NFSOFT [21] is another improvement of our algorithm. See also Section “Rotational and Rigid-body Correlations; Non-Uniform SO(3) Fourier Transforms”.
A schematic overview our algorithm package PF2 fit which solves the 3D EM map rigid fitting problem, is shown in Fig 1. It introduces the following innovations, each of which lead to improvements over the current state of the art in terms of accuracy and speed.
New FFT-amenable complementary scoring scheme. The complementary scoring scheme rewards overlaps between the volumes occupied by 𝓟 and 𝓜 as well as overlaps between the volumes complementary to 𝓟 and 𝓜. In this context, we introduce two scoring functions: the non-uniform inclusion potential and the complementary space score, both of which are computed on non-uniform grids. We also implement the scattering potential as an alternative to classical Gaussian blurring. The new scoring functions compare favorably to Gaussian-blur-based scoring across a variety of resolutions, in the presence and absence of noise. In particular, our FFT-amenable scoring functions result in lower RMSD than existing ones across a range of resolutions for synthesized density map fitting, and result in lower ETRs for microscope acquired density map fitting, also across a range of resolutions. Uniform and focused sampling and search with non-uniform FFT. All prior techniques require an equispaced/uniform angular grid for rotational search, a property that results in a highly non-uniform search of the space of rotations SO(3) which is likely to miss important regions of motions while oversampling others. By contrast, uniform sampling the space of rotations SO(3), requires non-uniform angular grids (cf. [24]) which is only amenable to a non-uniform, SO(3)-FFT-based search algorithm.
Furthermore, since our non-uniform FFT framework does not require uniformity of the translational and rotational grids, it enables focused searches in both translational and rotational space, thus combining the advantages of local and global fitting schemes. Information driven rerank scheme. Finally, to improve the accuracy of our fitting predictions, we rerank results from the search stage with respect to a scoring function based on matching the skeleton of 𝓜 with the secondary structural elements of 𝓟. In the reranking stage, we also include the well-known mutual information score [30].
Our reranking stage improves the rank of fitting poses obtained in the initial search stage at resolutions < 10Å. We expect the reranking stage to become more effective as more EM maps between 3 and 8Å are isolated.
We should also mention that, due to the improved sampling of SO(3), the time taken by PF2 fit for an average fitting exercise is comparable to most rival fitting schemes, taking 2–3 mins on an quad-core computer per fitting procedure. In particular, non-uniform inclusion potential takes advantage of the non-uniform search scheme to provide even faster (1.3 mins) runtimes with reasonably accurate estimates of the fitting pose while guaranteeing an exhaustive sampling of the space of available motions. Also, leveraging the focused search capability, PF2 fit can be applied to a vast range of problem types, from subunit-subunit, to subunit-assembly, to multiple subunit fitting. We have extensively compared PF2Fit to ADP-EM (Ref. [3]) in the experiments.
Executable programs as well as the source code for the entire software package PF2 fit and each of its components libraries are available to all academic users for free through our website. We made the sampling of SO(3) and SE(3), and the non-uniform FFT search libraries separately available so that users can adapt and modify all or some of them independently.
A typical fitting procedure starts with two inputs: an atomic structure 𝓟 and a 3D EM map 𝓜, normally at different resolutions. Let A : ℝ3 ↦ ℂ and B : ℝ3 ↦ ℂ be such scalar-valued functions derived from 𝓟 and 𝓜 respectively. Once A(x) and B(x): are defined, the best fit of the two molecules is obtained by maximizing the unnormalized cross-correlation score
C C S ( A , B ) = ∫ ℝ 3 A ( R x + t ) B ( x ) d x , (1)
where (R, t) is a rigid-body motion, i.e., a three-dimensional rotation R followed by a three-dimensional translation t. Applying the rigid-body motion which produces the maximum score will lead to the best fit.
In this paper, we shall refer to A(x) and B(x): as affinity functions. Our fitting procedure is divided into two main stages (cf. Fig 1): the exhaustive FFT-based search, and the reranking. We discuss each of these stages, their affinity functions and their advantages below. Detailed comparison and empirical results are presented in the Results section.
PF2 fit provides four choices for defining the affinity functions of A(x) and B(x): the Gaussian Agc (respectively Bgc); the scattering potential Asc (respectively Bsc); the non-uniform inclusion potential Anu(respectively Bnu); and the complementary (pocket) space potential Acs(respectively Bcs). The first three are based on the space occupancy of 𝓟 and 𝓜, and fourth is based on the complementary space of 𝓟 and 𝓜, denoted Anu and Bnu respectively. A typical depiction of the space and complementary space can be found in Fig 2. For a description of how a complementary volume is computed, we refer to [31]. We now discuss each of the affinity functions in detail.
The second important ingredient of the inital search stage of rigid-body fitting, is the search algorithm PFcorr (Polar Fast Fourier Correlation), first introduced in [29], to search over the space of rigid-body motions SE(3) of 𝓟. PFcorr is a family of rigid-body correlation algorithms based on non-uniform SO(3) Fourier Transforms, and it has many favorable attributes relative to classical FFT-based search algorithms, the most salient of which we discuss here.
In the second stage, results obtained in the correlation-amenable search stage are reranked with respect to scores that exhibit the following features: (A) They cannot be expressed in the general form of Eq (1). (B) The information they capture about a particular fitting orientation is additional to, or, ideally, independent of, each of the affinity functions maximized in the search stage.
We carried out experiments to compare different scoring and reranking functions implemented in PF2 fit as well as to compare PF2 fit with other publicly available fitting software, namely the Colores tool in Situs [18] and ADP_EM [3]. In this section, we describe the benchmark dataset, the experiment protocol and the metric used in measuring and comparing accuracy of fitting. The results and their implications are presented in the next section (Results and Discussion).
In this section, we discuss the relative performance of the different Cross Correlation Scoring (CCS) terms introduced in PF2 fit, in terms of the results obtained from the experiments on both synthesized and electron microscope acquired data and compare the performance of PF2 fit with Colores and ADP_EM [3]. We also discuss the effect of reranking shemes, skeleton-secondary structure score (SSS) and the mutual information score (MIS). Finally, we discuss some unique advantages offered by PF2 fit.
Gaussian blurring have usually been the method of choice for fitting software. In this paper, we introduced the scattering potential which provides a better model of the electron density than Gaussian. In our experiments on both synthesized and acquired EM data, we found that SCCS is a valuable alternative to the GCCS, one whose performance is similar and stable across a range of resolutions.
The addition of the complementary space cross-correlation score (CCCS), or the pocket score, (Fig 6) to the scoring function resulted in tangible improvements to the quality of the obtained fit across the range of resolutions. This was observed with the GCCS (Eq (5)), as well as the SCCS, and both in the presence and absence of noise. The improvement in the quality of the results obtained is most dramatic at values of R beyond 15Å.
Fitting is essentially an optimization problem in a high dimensional configuration space. In PF2 fit, and any other existing methods, the configuration space is discretized to a small number of discrete samples where the scoring terms are evaluated and the maximum/minimum is reported. Now, let 𝓕 be the scoring term and 𝓒 be the configuration space, from which N discrete samples are taken. Then if the true maximum value is m*(𝓕) = maxx∈𝓒𝓕(X), and the sampled maximum value is m𝓢(𝓕) = maxx∈𝓢𝓕(X)- then it is guaranteed that m*(𝓕)−m𝓢(𝓕) ≤ ω𝓒(𝓕, dN) (see Theorem 6.4 in [55]), where dN is the dispersion of the samples in 𝓒, and ω is a measure of the continuity of 𝓕. So, for a given 𝓕 and 𝓒, the error is directly correlated with dispersion of the N discrete samples.
In PF2 fit, we use the low-discrepancy and low-dispersion sampling scheme for SO(3) space described in [24]. By contrast, existing fitting software [3, 7, 47], due to their use of uniform-FFT, requires uniformity in the parameters used to represent the orientations in SO(3), for example uniform sampling of θ, ϕ and ψ Euler angles or icosahedral vertices on the sphere as in [3]. But such uniform sampling of the parameters might lead to non-uniform set of samples in SO(3), leaving large gaps (high-dispersion) in some places, showing high discrepancy or not sufficiently reflecting the topological structure of the underlying domain, cf. [24]. Since, PF2 fit uses a non-uniform polar FFT (NFSOFT [21]), it is able to handle non-uniformity in the Euler angles (which leads to uniformity, in low-discrepancy sense, in SO(3)). As a result PF2 fit can achieve high accuracies even with very limited number of samples.
In Fig 7(A) we report the results of applying PF2 fit on the synthesized EM dataset, with different number of low-discrepancy samples in SO(3). Notice that the results are stable with only around 5k samples (corresponding to about 15 degree separation in SO(3)) which runs in around 100 seconds in a single threaded execution. The Non-uniform inclusion potential cross-correlation score (NCCS) highlights another advantage of non-uniform FFT. NCCS incurs much smaller overhead than GCCS or SCCS, and the advantage becomes more pronounced when the overall runtime is smaller (see Fig 7(B)). For example, at 5k samples, the NCCS is about 20% faster than GCCS. Further advantages of NCCS, in terms of its speed-accuracy tradeoff is discussed in a later section.
We discuss the result of applying two separate reranking schemes, the skeleton-secondary structure score (SSS) and the mutual information score (MIS) in terms of improving the ranks of the predictions from PF2 fit’s FFT-amenable initial scoring phase which have low RMSD.
We tested PF2 fit against the publicly available ADP_EM and Colores. Though these three programms have correlation-based fitting schemes, the exact formulation of the affinity functions, and just as importantly the sampling scheme are not the same. For instance the translational step of PF2 fit is not directly comparable to the translational step of Colores. Also, the angular sampling density as well as the expansion degree affect the outcome. So, it is almost impossible to come up with parameters for each software which would result in the same measure of dispersion in the samples. Hence, we fixed the parameters of PF2 fit such that its runtime is similar to those of ADP_EM and Colores in their default settings, and then compared the results.
Suppose an experimenter wants to refine the cryo-EM map of GroEL at 7.7Å (𝓜 = EMD 1180, 192 × 192 × 192 voxels) by fitting a single subunit of GroEL (𝓟 = 1AONa) into it. This is the subunit-assembly problem, in which the translational uncertainty is roughly twice the size of 𝓟, whereas the rotational uncertainty is the range of rotations from 0 to 2π. One way to effect the refinement would be to segment from 𝓜 a 3D EM subunit of GroEL 𝓜s, to which 𝓟 could be fitted using PF2 fit —SE(3) (Fig 9A).
If a good segmentation is unavailable, a software like Chimera [10] could be used to refine an approximate placement. Another option is to use rigid-body fitting with Colores. Chimera surveys a fixed number (=50) distinct poses in its gradient-descent-based optimisation scheme, and the fit obtained is only locally optimal. On the other hand, Colores uses a uniform Cartesian grid with a default translational step equal to the voxel spacing of the map. With a default angular fineness of about 30° on a cubical Euler Angle grid, this results in 643 = 262144 translational samples and 864 rotational samples. Of these, several positions are redundant, as they lie outside the region of the initial guess. In general, there is no way in Colores to confine the translational search to a given region.
A third option is to use PF2 fit. The focusing property of PF2 fit mitigates both the above disadvantages (Fig 9B-9C). Since the GroEL assembly is symmetric, the experimenter could place the subunit approximately within the 3D EM map (Fig 9B), and then instruct PF2 fit —SE(3) to do a comprehensive search in the local region around the 3D EM map. In such an experiment, translations are completely disabled, and the rotational search space is uniformly sampled. Using the uniform SO(3) sampling technique in Mitchell [24] yields 14868 samples at 10° angular step. The result in Fig 9C is obtained.
Using PF2 fit in such scenarios has the following advantages. First, the comprehensive search in the local region essentially guarantees that PF2 fit —SE(3), unlike iterative gradient-descent-based optimization techniques, is not sensitive to an initial guess. Second, unlike global search routines, PF2 fit does not generate spurious rigid-body fits in regions that are spatially distant from the optimal fit. Third, the time-taken for the experiment is proportional to the number of local samples rather than for the (much larger) entire search space. PF2 fit thus combines the merits of local and global search paradigms in its focused search.
Note that many rotational-FFT-based schemes, e.g. [3] share the focusing property; however, since these techniques use a cubic Euler Angle grid, they do not ensure that the space of rotations is uniformly sampled.
The NCCS is a non-uniform-grid-based version of the envelope score in Vasishtan and Topf [13]. Along with PF2 fit, the non-uniformity inherent to the inclusion potential enables a very high speed search of the space of rigid-body motions SE(3) available to 𝓟. We explain this by first noting that since the quantity of information in Anu(x), cf. Eq (2), is exactly equal to the number of atoms in 𝓟s, a relatively low degree L in Eq (10) suffices to represent it. In general, while the GCCS and SCCS each demand a degree at least equal to L = 20, with best results for L ≥ 25, the NCCS requires only a degree L = 5 (see Fig 10).
By itself, however, this property is of limited use. In the uniform-FFT frameworks used in either [47] or [3], the expansion degree is keyed directly to the coarseness of rotational sampling, because the underlying FFT grids are only as fine as the expansion degree allows them to be. Using a degree L = 5 in either of these approaches would mean conducting a rigid-body search over an angular grid with separation 360/(2 × 5) = 36°, an unacceptably coarse value for most rigid-body fitting exercises. By contrast, PF2 fit, functioning as it does through the non-uniform SO(3) Fourier transform, enables an arbitrarily fine scan of the space of rigid-body motions at any expansion degree.
These advantages mean that the NCCS can play a central role in rigid-body fitting. If a coarse estimate of a fitted position of 𝓟 with respect to 𝓜 is desired, then a low expansion degree version of the NCCS can be used, whereas a more accurate estimate can be found using the SCCS at L ≥ 20. The typical time taken for a subunit-subunit fitting exercise on a single-threaded Macbook Pro at 2.5 GhZ with 8GB RAM is about 1.3 minutes.
The results of this paper has contributed to existing methods and software on rigid-body fitting. In particular:
Cross Correlation scoring functions. We have introduced the non-uniform inclusion potential CCS in Eqs (2), (3). This score has been shown to be preferable to standard fitting metrics in terms of speed (cf. Fig 10). We have also introduced the concept of complementary space matching, and introduced the complementary space scoring function (CCCS). The addition of the CCCS results in significant improvements in the prediction accuracy across a range of resolutions, regardless of the target-target scoring function used. Finally, we have compared the scattering potential (SCCS) with the typically used Gaussian potential (GCCS), finding that it performs favourably in our application compared to the latter in both synthesized and microscope acquired density map fitting scenarios and hence provides a valuable alternative. Search scheme. We have introduced a search scheme that is resolution-robust, capable of local fitting, and able to quickly and comprehensively survey the space of rigid-body motions SE(3) (cf. Results section). The search scheme we have introduced is capable of uniformly sampling the space of rigid-body rotations SO(3), where uniformity is defined according to a chosen metric. For instance, in the sampling technique from [24] we use throughout this work, uniformity involves the competing notions of local separation and global coverage. Equispaced Euler angular grids, the mainstay of all current rotationally exhaustive techniques, generate samples in SO(3) that possess neither of these desirable features. See also [29] for a more detailed consideration of sampling. Reranking stage. We have introduced the skeleton-secondary structure score (SSS), whose performance we expect to improve as the resolution of experimental cryo-EM 3D EM maps improves. Optional multi-basis framework. Our match and alignment (fitting) algorithms can use one of two popular basis expansions to perform an exhaustive search. PF2 fit —SE(3) and—SO(3) is compatible with existing FFT-based fitting schemes, while being general enough to subsume the approaches that use these schemes, approaches such as those by [3, 19, 47]. Because the NFFT is currently not as fast as the FFT, there may be situations in which the use of the FFT-based technique, regardless of its drawbacks, might be indicated. Suitable modifications of PF2 fit —SE(3) and PF2 fit —SO(3) would be applicable in these situations as well.
The PF2 fit software package along with a tutorial is free for academic users and available through our website: http://www.ices.utexas.edu/CVC/software/.
|
10.1371/journal.pgen.1005875 | Which Genetics Variants in DNase-Seq Footprints Are More Likely to Alter Binding? | Large experimental efforts are characterizing the regulatory genome, yet we are still missing a systematic definition of functional and silent genetic variants in non-coding regions. Here, we integrated DNaseI footprinting data with sequence-based transcription factor (TF) motif models to predict the impact of a genetic variant on TF binding across 153 tissues and 1,372 TF motifs. Each annotation we derived is specific for a cell-type condition or assay and is locally motif-driven. We found 5.8 million genetic variants in footprints, 66% of which are predicted by our model to affect TF binding. Comprehensive examination using allele-specific hypersensitivity (ASH) reveals that only the latter group consistently shows evidence for ASH (3,217 SNPs at 20% FDR), suggesting that most (97%) genetic variants in footprinted regulatory regions are indeed silent. Combining this information with GWAS data reveals that our annotation helps in computationally fine-mapping 86 SNPs in GWAS hit regions with at least a 2-fold increase in the posterior odds of picking the causal SNP. The rich meta information provided by the tissue-specificity and the identity of the putative TF binding site being affected also helps in identifying the underlying mechanism supporting the association. As an example, the enrichment for LDL level-associated SNPs is 9.1-fold higher among SNPs predicted to affect HNF4 binding sites than in a background model already including tissue-specific annotation.
| A large fraction of genetic variants that have been associated with complex traits are found outside of protein coding genes and likely affect gene regulation. Many experimental efforts have been dedicated to mapping regulatory regions in the genome but there are not many systematic methods that integrate functional data and regulatory sequences to predict the potential effect of any genetic variant on any given tissue and motif. Here we present a tissue and factor specific annotation that provides a predicted functional effect for both common and rare genetic variants. These predictions, certain of which are validated experimentally, show that the majority of genetic variants in gene regulatory regions are actually silent. Annotating those that are not silent allows us to investigate the molecular basis for the genetic architecture of many common traits and also to study the evolutionary properties that different types of regulatory sequences have across tissues or transcription factors. Overall, our study supports the concept that polygenic variation in binding sites for distinct classes of transcription factors has been a major target of evolutionary forces contributing to disease risk and complex trait variation in humans.
| Despite large ongoing efforts to characterize regulatory regions in the human genome (e.g., ENCODE [1], Roadmap Epigenomics [2]), the lack of a regulatory genetic code to discriminate functional from silent non-coding variants in regulatory sequences poses severe limitations in interpreting the results of many human and population genetic analyses. For example, large numbers of genetic variants associated with disease and normal trait variation have been identified through genome-wide association studies (GWAS) [3]; yet a formidable challenge remains in determining the specific molecular mechanisms underlying association signals in non-coding regions. Similar challenges also arise when exploring the evolutionary functional significance of non-coding variants, for example through analysis of differences in genotype distribution across populations [4, 5]. This is also complicated by the fact that GWAS hits and signals of selection are usually found in large regions of association and do not directly pinpoint the true causative variants. In general, we do not know in which cell types/tissues these variants may have a functional impact.
Computationally and experimentally derived annotations for regulatory regions have been used to functionally characterize GWAS hits [1, 6–12]. However, a simple positional overlap between a genetic variant and regulatory regions is a necessary but not a sufficient condition to demonstrate an impact on TF binding. Many experimentally derived annotations are very useful to identify broad genomic regions across many cell-types, but lack the resolution necessary to pinpoint the regulatory sequences. High resolution functional assays like DNase-seq and ATAC-seq combined with computational methods that integrate sequence motif models [8, 9, 13, 14] can effectively dissect the regulatory elements; yet the motif models for transcription factor (TF) binding are generally not sufficiently well calibrated to predict the binding impact of a sequence change. Alternative ChIP based approaches (such as ChIP-seq and ChIP-exo), may provide increased TF and regulatory element specificity, but rely upon the availability of antibodies to target specific TFs or tagged TFs [15, 16]. The consequence is that we cannot provide a satisfactory answer to the following questions: Which genetic variants are more likely to impact binding of specific TFs? What is the fraction of genetic variants in regulatory regions that are not neutral? If we can adequately answer these questions, we may further ask: Did polygenic adaptation occur at binding sites for the same TF? Do variants in certain types of TF footprints and tissues contribute to variation in specific complex traits?
To help answer these questions, we have extended the CENTIPEDE approach to generate a catalog of regulatory sites and binding variants encompassing more than 600 experimental samples from the ENCODE and Roadmap Epigenomics projects with DNase-seq data, and recalibrated sequence motif models for more than 800 TFs. We then incorporated ASH information to provide additional empirical evidence, to validate the accuracy of the computational predictions and to estimate the fraction of genetic variants in regulatory regions that are not neutral. Importantly, our annotation is specific at the motif level (i.e., TF-specific) and at the sample level (i.e., tissue-specific). We then compare our results with the only alternative TF-centric annotation that has been recently published [17], but we also compare with non TF-centric SVM derived annotations [18]. Using our new catalog, we then examined genomic properties of the annotations, identifying characteristics that predict variants that disrupt binding, and demonstrated the action of natural selection on TF binding sites. Finally, we annotated and interpreted variants associated with complex traits, and we validated their allele-specific enhancer activity by reporter gene assays.
The CENTIPEDE approach allows to predict TF activity by integrating sequence motif models together with functional genomics data, and gains the most information from high-resolution data such as DNase-seq or ATAC-seq [19]. The spatial pattern in which reads are distributed, or footprint, is specific for each TF and can be very useful for discriminating between classes of TFs with distinct profiles [13]. In the original CENTIPEDE approach, the sequence models are pre-determined; e.g, k-mers or previously defined position weight matrix (PWM) models. However, many sequence models in existing databases were created with very few instances of known TF binding sites and do not represent the full spectrum of sequence variation that can be tolerated without affecting binding. Here, we have extended CENTIPEDE to readjust the sequence models for TF binding (Fig 1 and S1 Fig) using DNase-seq data and sequence orthologs (Methods). Compared to the original motif models the consensus sequence is largely maintained in the recalibrated motifs (S6 Fig). However, when we consider ChIP-seq peaks as validation we obtain superior precision recall characteristics (S7 Fig, Section 6.1 in S1 Text) and a much higher correlation with the prior probability of binding calculated by CENTIPEDE (S8 Fig, Section 6.2 in S1 Text).
Across all 653 DNase-seq samples, we identified a total of 6,993,953 non-overlapping footprints corresponding to 1,372 motifs active in at least one tissue and collectively spanning 4.15% of the genome. Each individual sample contained, on average, 280,000 non-overlapping footprints for 600 motifs and spanning 0.162% of the genome, indicating that footprints are highly tissue specific. Considering all SNPs from 1000 Genomes Project (1KG) at any allele frequency (even singletons), we found 5,810,227 (0.19% of the genome) unique genetic variants in active footprints (footprint-SNPs), 3,831,862 (66%) of which are predicted to alter the prior odds of binding ≥20-fold (effect-SNPs) based on the logistic sequence model hyperprior in the CENTIPEDE model (Fig 1C and 1D, Equation 2 in S1 Text). Effect-SNPs are further classified as switch-SNPs (264,965) if the allele flips the prior odds of binding. Importantly, in any of these categories we retain for each prediction the motif identity (TF-specific) and the underlying sample (cell-type specific) information.
These functional categories we computationally defined provide an answer to the question of which genetic variants in DNaseI sensitive regions are more likely to affect binding. To experimentally assess the accuracy of our answer, we used Quantitative Allele-Specific Analysis of Reads (QuASAR) [20] to perform joint genotyping and ASH analysis within DNase I hypersensitivity (DHS) regions (S2 Fig). While the initial quality filtering is the same as for the CENTIPEDE analysis, the parameters of the QuASAR model also allowed us to detect tissues with chromosomal abnormalities or samples from pooled individuals (Section 4.2 in S1 Text). These DNase-I samples were therefore excluded from ASH analysis (S9 and S10 Figs, S6 Table). Across the remaining 316 samples suitable for ASH analysis, we identified 204,757 heterozygous SNPs (hSNPs) in DHS sites (DHS-hSNPs) with coverage > 10x and with MAF > 0.05.
Overlapping our predictions with the DHS-hSNPs, 55,044 are footprint-hSNPs, 26,773 of these are effect-hSNPs, and 5,991 of these are switch-hSNPs. Overall, our computational predictions are highly concordant with the direction of ASH; 75% of the sequence models show positive correlation between the predicted and observed ASH (S11 Fig, S7 Table, Section 5.4 in S1 Text). Each of the nested SNP functional categories have marked differences in p-value distribution (Fig 2A) for the QuASAR test of ASH. Compared to what would be expected from the null uniform distribution, effect-hSNPs and switch-hSNPs have 8x and 14x times more SNPs with p < 0.001 respectively, showing that our functional annotations can predict ASH. Furthermore, these enrichments for lower p-values are much higher than those of DHS-hSNPs (4x) and footprint-hSNPs (6x), indicating that identifying SNPs in DHS regions and/or footprints alone is not enough to predict functional effects on binding. A similar observation can be made using the observed allelic ratios across CENTIPEDE annotations (S12 Fig). The result that SNPs that are just located in footprints or DHS regions tend to be silent is also true for other existing annotations (S13 Fig) or if we change the threshold for discriminating between footprints-SNPs and effect-SNP (S14 Fig). We also see that conservation score alone is not accurate enough to predict which SNPs have a functional impact on binding (S16 Fig).
To quantify the fraction of genetic variants that in each annotation will truly affect TF binding, we used ASH p-values as input evidence and followed the strategy of Benjamini et al. [21] to perform multiple testing correction in each category separately using Storey’s q-value procedure [22]. At an FDR threshold of 20%, we detected 3,217 unique hSNPs displaying significant ASH (Table 1), hereafter referred to as ASH-hSNPs. Taking into account LD (R2 < 0.8) these ASH-hSNPs constitute at least 3,158 independent loci. Several of the ASH-hSNPs were significant in more than one cell-type, giving a total of 4,940 observations of ASH-hSNPs across all samples. The 20% FDR threshold was chosen because this data was not originally sequenced to the depth that is generally required to call ASH at a single site with high confidence. In this reanalysis, we instead focus on the aggregate distribution of p-values to estimate the proportion of true null hypotheses (Storey’s procedure π ^ 0 estimate). We estimate that 56% of the effect-SNPs show evidence of ASH. While this conservative estimate can be considered a lower bound, it is still much higher than the estimates for DHS-SNPs (2.1%) and footprint-SNPs (3.1%), indicating that most SNPs in DHS regions and even in the putative binding sites do not affect binding.
In addition to the DNase-seq ASH validation, we compared our annotations to the results of QTL analyses targeting DNase-seq sensitivity sites (dsQTLs, [23]), and CTCF binding sites from ChIP-seq [24]. For dsQTLs, using the same PROC analysis (see Fig 2B) as in [18] demonstrates that effect-SNPs have a good performance compared to SNPs identified using a SVM approach or CATO [17]. Note that we have not repeated the PROC analysis for the methods studied by [18], but we used directly the results provided by them, as PROC analysis could be sensitive to a redefinition of the underlying true labels of the set used to evaluate performance (see discussion in Section 7 in S1 Text). If we constrain the gk-SVM model to those predictions that overlap with our CENTIPEDE footprints, the precision (at 10% recall) improves to 80%. This indicates that SVMs are better sequence models than PWMs, but are not as specific without footprint information. To further investigate the TF-specificity accuracy of our annotations we used CTCF QTLs. CTCF is a very special type of TF with insulation [25], DNA loop organization [26], and barrier functions [27]. Compared to training an SVM on the DNase-seq data-set (non TF-centric), models that are TF-centric such as CATO and our effect-SNPs (integrating the footprint and sequence preferences) demonstrate a superior accuracy in discriminating dsQTLs that are also CTCF QTLs from those that may affect other factors (see Fig 2C). Among all CTCF footprint-SNP instances, all those that are also effect-SNPs are enriched for low CTCF QTL p-values and we predicted the correct direction (the allele with higher binding) in 100% of the cases (Fig 2D, Section 3.3 in S1 Text).
Some of the alternative methods include information such as conservation, distance to the TSS and allele frequency, however we have not included them in our annotation as we wanted to use those measures for analyzing the potential impact on organismal function and study differences among distinct TF motifs.
Regions of the genome with demonstrated molecular function (e.g. genic regions) generally show reduced diversity [28] and a site frequency spectrum skewed towards rare variants. This is due to negative (purifying) selection, which prevents alleles from reaching high frequencies in the population if the molecular trait translates to a negative impact on organismal function. We investigated whether a similar skew in the site frequency spectrum exists at functional non-coding variants (effect-SNPs). We observed that effect-SNPs display an enrichment for rare variants (<0.5%) comparable to what it is observed in coding regions (Fig 3A), where rare variants are 1 to 2 times more likely to be non-synonymous changes than synonymous [29].
eQTL studies have found that variants associated with gene expression tend to occur close to the transcription start site (TSS) [30–33]. We detect a similar trend among our annotations, with 83% of footprint-SNPs occurring within 100kb of the TSS. However, we find a 1.12-fold depletion of effect-SNPs within 300 bases of a TSS (Fig 3B), which represents the core promoter region [34]. Effect-SNPs in this region are also enriched among rare variants (MAF < 0.001, 1.15-fold enrichment, Fisher’s test p-value = 6.027 × 10−13). This is likely because effect-SNPs in these regions have a major impact on regulatory processes that are shared across tissues. Accordingly, we also discovered a 1.18-fold enrichment for effect-SNPs in footprints active in 5 or fewer samples and a 1.38-fold depletion for effect-SNPs in footprints active in 50 or more samples (Fig 3C).
Since allele frequency can be correlated with distance to the TSS or sequence conservation, and shared footprints may also be more common at the promoter region, we tested several features (individually explored in Fig 3) in a joint model (Methods). All tested factors are significant predictors when considered together in a multiple regression logistic model, and the direction of the effect is the same as when they are considered separately (S8 Table). These results support the hypothesis that factors binding closer to the TSS and/or active in many tissues are housekeeping factors and those that recruit the transcriptional machinery and as a consequence are less likely to harbor common regulatory variants.
To examine the distribution of ASH-hSNPs across the different regulatory factors, we calculated the ASH enrichment ratio for each TF defined as the fraction of ASH-hSNPs over hSNPs relative to the average fraction across all TF (S17 Fig, Section 8.3 in S1 Text). At a nominal p-value cutoff of p < 0.01 (Binomial test), we detected 32 motifs enriched for ASH and 56 depleted for ASH (Fig 4A; S9 Table). In cases where multiple motifs correspond to the same factor, we observe similar enrichment for ASH-hSNPs (S10 Table), most notably for the factor AP-1, showing a >2.5-fold enrichment for ASH SNPs in all but one of the seven motif models. We see the same pattern for motifs significantly depleted of ASH-hSNPs, such as CTCF (1.5-fold median depletion) and E2F (1.8-fold median depletion). ASH enrichment ratios are also consistent across factors with similar functions. For example, three factors in addition to AP1 with roles in the immune response, CREB [35], c/EBP [36], and NF-κB [37] are over 2-fold enriched for ASH-hSNPs within their binding sites (S11 Table).
We then examined the genomic characteristics at TF binding sites to identify features that distinguish motifs enriched for ASH versus those that are not. We found that motifs enriched for ASH are significantly farther from the TSS, having an average median distance to the TSS of 23kb compared to 17kb for those depleted (Mann-Whitney p = 3.2 × 10−8; Fig 4B). Furthermore, motifs enriched for ASH are active in significantly fewer samples, on average active in 20% vs 40% for those depleted (Mann-Whitney p = 1.9 × 10−7; Fig 4C), indicating that TFs with a high degree of ASH across their binding sites tend to be active in fewer tissues. This further confirms that changes in footprints active in a large number of tissues (constitutionally active) are more likely to have pleiotropic effects and therefore impact negatively the fitness of the organism and suggests polygenic mechanisms of evolution on motifs categories (i.e. groups of binding sites for a given TF or for TFs regulating genes with similar functions).
An important question in evolutionary biology is the extent to which selection has acted on cis-regulatory elements in humans [38–41]. While methods are being developed to address this question [42, 43], such methods have only been applied to a narrow subset of TFs, and, in the case of [43], rely on RNA expression data to classify mutations as up- or downregulating transcription relative to the reference enhancer sequence. Given our categorization of footprint-SNPs relative to their effect on factor binding, we performed an initial survey of selection across TF binding sites using a test similar to the McDonald-Kreitman (MK) test [44] (S3 Fig, Section 8.4 in S1 Text). Applying our modified motif-wise MK test, we obtained a selection score for TF motifs with a sufficient number of binding sites (Fig 5A, S12 Table). At an FDR of 1%, we observe 84 factors whose binding sites are enriched for fixed functional differences (higher selection scores), suggestive of positive selection acting on those sites. Among the top scoring motifs are several factors that regulate neural and neuro-developmental processes, including POU1F1, PHOX2B, DBX2, UNCX, and YY1 which were not previously seen [42]. Among the factors with the lowest selection scores, we find ARNT, RBPJ, CREB1, POU2F2, and MYC which match with what has previously been observed [42]. While the interpretation of a positive selection score is generally that of positive selection, interpreting negative scores is more challenging. Generally, deleterious alleles are much less likely to reach fixation in populations than neutral alleles, however a negative selection score could also be explained by relaxation of selection or balancing selection. To identify the most likely evolutionary scenario for variation in binding motifs with negative selection scores, we calculated the derived allele frequency (DAF) for SNPs in binding sites. We observed an excess of rare alleles for SNPs in binding sites with a negative selection score (Fig 5B, S19 Fig, Section 8.5 in S1 Text), suggestive of weak purifying selection, rather than relaxation of selection (similar DAF spectrum across categories) or balancing selection (excess of intermediate frequency alleles).
We next asked whether the excess of functional polymorphism relative to functional divergence were influenced by background selection from nearby genes (S18 Fig), as functional regulatory variants may occur closer to the TSS, compared to silent variants. We find a mild but significant positive correlation between selection score and median TSS distance (Spearman ρ = 0.16, p = 5.6 × 10−9). Additionally, there is a negative correlation between tissue specificity and selection score (Spearman ρ = −0.20, p = 1.2 × 10−13). While some of the selection signal may come from nearby genes, there does appear to be a pattern of selective constraint on broadly active factors binding in promoter regions.
Given that our annotations comprise predicted functional effects across multiple cell-types/tissues and are anchored at footprints for known TF motifs, we asked if they could help interpret genomic hits reported in the GWAS catalog. We first considered a gross overlapping approach that considers each variant in a GWAS hit region equally likely to be causal (using an r2 cutoff of 0.8 from 1KG Project data, as in Ward et al. [10]). In GWAS hit regions, we compared the proportion of effect-SNPs over footprint-SNPs and found a moderate 1.11-fold enrichment for effect-SNPs (p < 2.2 × 10−16, 95% CI: 1.10—1.14). These moderate but statistically significant enrichments are typical of other annotations as well and are likely due to the fact that: i) we only consider the strongest GWAS hits (missing variants with moderate and small effects), ii) not all the factors and tissues may have the same enrichment, and iii) lack of resolution, as expanding the GWAS hit region makes the enrichment effects more moderate. Nevertheless, if we add our annotation to category 2 SNPs from the RegulomeDB [8] (SNPs with multiple regulatory annotations, but not yet shown to be functional), we detect a 1.6-fold enrichment for GWAS hits compared to category 2 SNPs alone (p = 6.11 × 10−5, 95% CI: 1.27—1.99). This result demonstrates that our annotation adds relevant information as it filters genetic variants not likely to be functional, but the overlap approach employed cannot take full advantage of the resolution and contextual information provided by our CENTIPEDE predictions.
To better test if the annotated effect-SNPs can help fine-mapping and give a mechanistic support for variants associated with complex traits, we integrated them into GWAS meta analyses for 18 traits (see S13 Table) using the recently developed hierarchical model fgwas [45]. Importantly, in this analysis we used as input the association p-values measured or imputed to all known common variants in the genome. Furthermore, for each trait we compare to a baseline model [45] that considers previously defined annotations [11, 46] and confounders (e.g., distance to TSS, coding region, and others). For each trait, we identified factors whose binding sites were enriched for associated SNPs (Fig 6A and 6B, S20 Fig and S14 Table) over the baseline model (the enrichments reported by fgwas are log-odds ratios from the model parameters).
Overall, we observed high enrichments for biologically relevant factors. For example, the enrichment for effect-SNPs in OCT-4 (POU5F1, a TF with a key role in embryonic development and stem cell pluripotency [47]) regulatory sequences when considering genetic variants associated with human height is 6.6-fold higher (95%CI: 3.7-8.2) than in the baseline model. This is consistent with previous observations of genetic variants associated with height being enriched in embryonic stem cell DHS sites [48]. We also observed an enrichment for the developmental regulators TBX15 (3.9x), FOXD3 (3.9x), and NKX2-5 (4.7x) for genetic variants associated with height. From a study of low-density lipoprotein (LDL) levels in the blood, enriched factors include the liver-specific factor HNF4A (9.1x), as well as several regulators of immune function, including CREB1 (3.7x), IRF1 (6.2x), and IRF2 (7.1x).
Our high resolution annotations allowed us to dissect the most likely functional variant (posterior probability of association, PPA > 0.2) in 88 previously identified GWAS regions (S15 Table, S23 Fig). For all 88 but 2 of these SNPs we have at least a 2-fold increase on the posterior odds of picking the potentially causal genetic variant according to fgwas (8.5x median fold increase) when compared to the comprehensive baseline annotation used by [45]. We then performed reporter gene assays for 21 SNPs to validate the predicted allelic effect on gene expression and the underlying molecular mechanism (Fig 7A and 7B, S16 Table, Methods). Among the regions tested we validated that 11 have enhancer/repressor activity and 10 have variants with allele-specific activity (p < 0.05, BH-FDR = 10%). This corresponds to 48% validation rate which is much greater than the 5% that would be expected by chance (Binomial test p = 2.01 × 10−8). Overall the predicted effect on binding and the change in gene expression are well correlated (Spearman ρ = 0.612, p-value = 0.0032), and the three SNPs with opposite effects may represent binding sites for repressors. Spearman correlation is robust to outliers, removing potential outlier rs540909 results in ρ = 0.657 (p-value = 0.002). We also achieve a similar correlation when we use our predictions to evaluate mutations in enhancers from a previously published reporter assay [49] that match our CENTIPEDE footprints (Spearman ρ = 0.76, p-value = 4.37 × 10−5, S22 Fig, Section 9.4 in S1 Text).
As an example, rs4519508, associated with a 2.1cm decrease in height [50], is in a binding site for the cell-cycle regulator family E2F (Fig 6D). Our annotation increased the PPA from a baseline of 10.5% to 44.4%, and it is the highest associated SNP in the association block (S21A Fig). This E2F footprint is active in >300 tissues (most of them fetal) and we detected ASH at this SNP in lung fibroblasts, validating that the reference allele at rs4519508 confers stronger binding than the alternate. Interestingly, in the reporter assay we observed 1.5-fold increased expression in the presence of the alternate allele, suggesting that at this location, E2F is acting as a repressor. Finally, this SNP is located within the promoter of PPP3R1, a regulatory subunit of calcineurin important for cardiac and skeletal muscle phenotypes; and a SNP in the same region has been shown to be associated with endurance [51] in humans. The p-value of association for this GWAS locus (p = 8.1 × 10−6) does not reach genome-wide significance in the height meta-analysis data we used [50]; however, in a recent more extensive meta-analysis for height [52] this locus achieves genome-wide significance p = 8.4 × 10−10, demonstrating that our annotation can be useful to rescue relevant loci.
Finally, a SNP associated with LDL levels, rs532436, is within a footprint for USF, an E-box motif (Fig 6C). Adding our annotation increased the PPA of the SNP from 39.7% to 94.7% (S21B Fig). We found that the alternate allele, associated with a 0.0785 mg/dL increase of LDL in the blood, is predicted to have a lower binding probability and results in 1.8-fold lower expression, compared to the reference allele. This SNP is identified by GTEx [53] as an eQTL for two proximal genes in whole blood: ABO (p = 5 × 10−5) and SLC2A6 (GLUT6, a class III glucose transport protein; p = 8 × 10−5). The SNP has an opposite effect on expression of the two genes, with the alternate allele showing lower expression for ABO and higher expression for SLC2A6.
These results show that our integrated analysis provides support for likely mechanisms linking regulatory sequence changes to complex organismal phenotypes. Furthermore, these mechanisms can be directly investigated through molecular studies, providing additional support that these sequence changes are truly functional.
We have developed an approach for assessing functional significance of non-coding genetic variants in DNase-seq footprints. Our strategy integrates sequence information with functional genomics data to predict the impact of single nucleotide changes on tissue-specific TF binding. This is achieved while integrating footprint information that preserves the identity of the underlying factor with high specificity. By borrowing data from ENCODE and Roadmap Epigenomics, we generated one of the most comprehensive catalogs available to date annotating regulatory regions and functional genetic variants across the genome.
Thus far, most common approaches for identifying regulatory variants from functional genomics data assume that each SNP in a regulatory region is equally likely to be functional. A key finding in this study is that genetic variants in active regulatory sequences, as defined by DNase I sensitivity and footprinting, are mostly silent; only 2.1% of SNPs in DHS regions and 3.1% of SNPs in CENTIPEDE footprints are estimated to have ASH. This is analogous to SNPs in coding regions, where most genetic changes are synonymous and do not result in an amino acid change [29]. The sequence model developed in this study provides a very useful filter for non-coding genetic variants that are not functional, resulting in a tissue-specific and motif-specific annotation of effect-SNPs (56.5% of which are estimated to have an impact on ASH). This is crucial information to take into account when we attempt to understand the molecular mechanism behind GWAS hits and evolutionary signals of selection. As additional functional genomics studies are performed, across larger sample sizes, tissue types and cellular conditions, it will be important to further determine the functional subset of regulatory variants within binding sites to achieve greater power in functionally annotating genetic variants associated with complex traits.
We find that genetic variants that are predicted to impact TF binding are depleted in the core promoter regions, exhibit higher sequence conservation in closely related species, tend to have low allele frequency and are enriched in tissue-specific footprints. These properties largely reflect the family-wise characteristics of motifs, which are further reflected in signals of selection. Future studies could incorporate tissue breath, conservation and distance to TSS as features to further filter effect-SNPs that may not show ASH. It should also be noted that our definition of functional regulatory variants is connected to the predicted effect on binding in the specific subset of cell-types/conditions that were available. Analyzing the allelic effects of non-coding variants in the context of other tissue types, conditions and functional genomic assays may potentially identify a functional role for some of the sites here defined as silent. In this study, we treated each TF separately, but future work should further explore the combinatorial grammar that different groups of motifs may define by cooperative binding to determine tissue specific binding sites. This will probably require more complex sequence models (e.g., SVMs [18, 54] or deep neural networks [55, 56]) than the PWMs used here. Here we show that the footprint information helps in predicting functional variants by further identifying the underlying TF compared to a sequence-fits-all model. More sophisticated footprint models [57] may also offer additional improvements to dissect the complexity of the regulatory grammar.
As not all genetic variants that have an impact on binding may lead to changes in gene expression and ultimately an organismal phenotype, combining these predictions with eQTL data across several tissues or environmental conditions would be important to further refine this annotation. As an example, Wen et al. [33], using an early release of this annotation in lymphoblastoid cell-lines demonstrates that effect-SNPs are 1.49 fold (with 95%CI[1.38, 1.63]) more likely than baseline SNPs (SNPs that are not located in a footprint) to be eQTLs (p = 4.93 × 10−22); in contrast, silent footprint-SNPs are 1.15 fold (with 95%CI[1.04, 1.27]) enriched in eQTLs, comparing to baseline SNPs (p = 0.0035).
A key feature of our annotation is that it spans a large collection of tissues and transcription factor motifs. This allowed us to trace some of the evolutionary history of TF binding and identify evolutionary constraints on specific molecular functions, which may reflect selective pressures during human history. For example, we observed that immune TFs are enriched for ASH sites, which supports the hypothesis that this may be a consequence of human adaptations to pathogen exposures [58]. On the other hand, we identified neural development TFs that may have undergone positive selection in humans. The large number of regulatory variants predicted in our study, together with previously reported eQTL signals [59–61], and the overall relevance that they have in explaining complex traits provide further support for polygenic models of complex traits in humans. By taking advantage of the factor-specific annotations in our study, we identified motifs that are enriched for regulatory variants associated with relevant GWAS traits and we provide examples of molecular mechanisms behind the association signals; e.g., immune TFs in the lipids study, and developmental TFs for height. Finally, we show how regulatory annotations improve the identification of potential causal SNPs in GWAS. Overall, the GWAS meta-analysis and selection signals in our study support the concept that polygenic variation in binding sites has been a major target of evolutionary forces and a key contributor to disease risk and complex phenotypes in human populations.
We used 1,949 PWM sequence models (motifs) from the TRANSFAC [62] and JASPAR [63] databases to scan the genome for a set of representative motif matches (Section 3.1 in S1 Text). For each motif, we used the matching sequences to calculate a new PWM model which we then used to scan the genome and identify all genome-wide motif matches using a two step approach:
Step 1: Initial CENTIPEDE scan and motif recalibration. For each motif, we extracted DNase-seq data at sequence matches across 653 samples (corresponding to 153 unique tissues) publicly available from the ENCODE and Roadmap Epigenomics projects (Sections 1 and 2.1 in S1 Text). The motifs and samples used are summarized in S1 and S2 Tables. For each motif and only for this initial step, we used a reduced subset of motif matches that include the top 5,000 best sequence matches, and up to 10,000 additional low-scoring sequences (Section 3.1 in S1 Text, note that for Step 2 we will use all motif matches in the genome). To avoid overfitting and to heuristically reduce the search space, these low scoring motif instances are human sequences that have orthologous very high scoring motif instances in the chimp or rhesus genome. We then applied the CENTIPEDE model to survey TF activity for each 1,272,697 tissue-TF pair. For each pair we then determined that the TF is active if the sequence matches that exhibit a CENTIPEDE footprint can be predicted from the PWM score (Z-score > 5, S4 and S5 Figs). Using this criterion, we determined that 1,891 TF motifs are active in at least one tissue. The full list of motifs active in each tissue can be found in S3 Table. We then recalibrated the PWM model for each active motif using the sequences of all motif matches that have a DNase-seq footprint (CENTIPEDE posterior > 0.99).
Step 2: Full genome CENTIPEDE scan and genetic variant analysis. Using the recalibrated sequence models we scanned the human genome again for all possible sequence matches. We used the CENTIPEDE algorithm to assess the probability that each motif instance is bound by a TF, both to the reference and to alternate alleles when the match contained a genetic variant catalogued in the 1KG Project [29]. In this second step, we included all high and low scoring PWM matches down to the threshold corresponding to a CENTIPEDE prior probability of binding of 10% (Equation 2 and Section 3.2 in S1 Text).
To evaluate whether the updated sequence models derived from DNase-seq data are better at predicting TF binding than the original seed motifs, we compared to ChIP-seq data available for a small set of TFs from the ENCODE project (as these data are generated in independent experimental assays that should be highly TF-specific). Using precision recall operating characteristic (P-ROC) curve analysis (see Section 6.1 in S1 Text), we determined that for a given precision (precision = 1—FDR, false discovery rate), the updated sequence models have higher recall (sensitivity) than the original PWM in detecting ChIP-seq peaks (S7 Fig). Additionally, we compared the correlation between the prior probability of binding (calculated by CENTIPEDE based on the PWMs) and the number of ChIP-seq reads overlapping motif matches (S8 Fig, Section 6.2 in S1 Text).
We classified a SNP in a CENTIPEDE footprint (footprint-SNP) as having a predicted effect on binding (effect-SNP) if the difference in the prior log odds ratio (from the logistic sequence model in CENTIPEDE, Equation 2 in S1 Text) between the two alleles was ≥3, indicating a ≥20-fold change in the prior odds of TF binding. We further classified an effect-SNP as switching the likelihood of binding (switch-SNP) if the prior log odds ratio flips; i.e, if it is ≥0 for one allele and ≤0 for the other. To generate a final set of annotated SNPs, we aggregated the data from each sample and motif into one table. For cases where a SNP is within multiple predicted binding sites, we selected the factor whose CENTIPEDE likelihood ratio was the greatest, i.e., the factor most likely to be binding at that location.
Starting from raw sequencing reads, we used a custom mapper [23] to align the reads to the hg19 reference genome. As allele-specific analysis is extremely sensitive to mapping errors and PCR duplicates, we employed several methods to reduce these sources of potential bias (Sections 2.2—2.4 in S1 Text). To detect allele-specific hypersensitivity, we applied QuASAR [20] to the processed read data to infer genotypes for all 1KG SNPs and determine the likelihood of allelic imbalance at heterozygous sites. Note that we only test a SNP with QuASAR if it is covered by ≥10 reads. To adjust for multiple testing, we used the q-value method [22] on the p-values produced by QuASAR.
We overlapped heterozygous SNPs (DHS-hSNPs) identified by QuASAR with CENTIPEDE footprints-SNPs and effect-SNPs catalogued for each sample. SNPs were then partitioned based on their predicted effect on binding into three non-overlapping categories: 1) hSNPs in predicted footprints whose binding effect is in the direction predicted, 2) all other hSNPs in footprints, 3) all other DHS-hSNPs. Because each annotation has a different prior expectation of being functional, we re-adjusted for multiple testing within each annotation separately using the q-value method [22] on p-values produced by the QuASAR model. We denote as ASH-hSNPs those hSNPs with a q-value < 20% in any of the partitions.
To determine which features of a SNP are predictors of functional effect, we performed multiple regression analysis using a logistic model considering the dependent binary variable El, indicating whether the footprint-SNP, l, is also an effect-SNP.
logit ( E l ) ∼ C l + F l + T l + N l + P l
We considered the following variables related to the probability of a footprint-SNP being an effect-SNP: the footprint likelihood ratio (without the sequence model) (Cl); the minor allele frequency (Fl); the absolute distance to the nearest transcription start site (Tl); the number of tissues for which the motif containing the footprint-SNP was predicted to be bound (Nl); the phyloP conservation scores calculated from primates (Pl).
This model does not evaluate the sequence, rather it combines the results shown separately in Fig 2 into a single model to characterize the predictions made by CENTIPDE. The model was fit using the GLM function in R. The result of this regression analysis can be seen in S8 Table.
To identify divergent TF binding sites, we used the UCSC liftOver tool on binding sites without a known polymorphism to obtain orthologous regions in the chimpanzee genome. Using the PWM model, we calculated PWM scores and CENTIPEDE prior probabilities of binding on the chimpanzee sequences. Sites with a sequence change in the motif instance (prior probability of binding differs from the humans sites) were classified as divergent, and were further categorized by the difference in binding affinity: “functional” for sites that change ≥20-fold between species (analogous to effect-SNPs), and “silent” for those that do not. For the binding sites containing a polymorphism, we used the definition of effect-SNPs to identify functional for silent sites and footprint -SNPs for silent sites. For each factor motif, we then calculated the number of binding sites belonging to each of the four categories (divergent functional, divergent silent, polymorphic functional, and polymorphic silent) and calculated a selection score similar to the McDonald-Kreitman test (Section 8.4 in S1 Text).
To integrate functional annotations and GWAS results, we used the fgwas command line tool [45]. fgwas computes association statistics genome wide using all common SNPs from European populations in the 1KG Project, splitting the genome into blocks larger than LD. Summary statistics were imputed with ImpG using Z-scores from meta-analysis data. Using an empirical Bayesian framework implemented in the fgwas software, GWAS data were then combined with functional annotations. We then compared the informativeness of these annotations from each of the 1891 motifs with CENTIPEDE predicted regulatory sites to a baseline model (see Section 9.2 in S1 Text) consisting of previously used genomic annotations identified as relevant [45]. For each locus that contains at least one SNP with a PPA > 0.2, we only consider the SNP with the highest p-value or PPA from fgwas. Rather than look at a credible set, we pick a single SNP most likely to be causal and see if that SNP has a higher PPA with the annotation than without it. While reduction in size of the credible set is very important for assessing fine-mapping methodologies, here our focus is on combining annotations to identify the single most likely causal SNP per GWAS locus.
GWAS-relevant effect-SNPs located in active footprints in LCLs (the cell line used for transfection) were ranked on the Spearman correlation coefficient in S7 Table. We initially selected the top 25 SNPs with a positive correlation, but the assays for 4 of them failed for several technical reasons (e.g., cloning step failed). To test allele-specific effects on expression for the remaining 21 SNPs, we first constructed inserts containing the reference or alternate allele for each SNP of interest (see Section 9.3 in S1 Text). Cloning of these inserts in the pGL4.23 vector was performed using the Infusion Cloning HD kit (Clontech) and DNA was extracted using the PureYield kit (Promega). Transfections were performed into GM18507 using the standard protocol for the Nucleofector electroporation (Lonza). Luciferase activity was measured for up to 20 replicate experiments using the Dual-Glo Luciferase Assay Kit (Promega). We contrasted the activity of each construct to the pGL4.23 vector, to assess enhancer/repressor activity of each region. To evaluate allele-specific effects, we contrasted the activity of the reference allele to the alternate allele for each region and we used a t-test to assess significance at a p < 0.05 threshold. We used the Benjamini-Hochberg [64] procedure to assess FDR across all 21 SNPs tested.
Unless otherwise noted, tests for enrichment on two-way categorical variables are based on Fisher’s exact test. Tests involving multiple categorical, discrete or continuous variables use a logistic regression model and Wald’s test on each enrichment parameter, and are identified as such.
The generated annotation files are available as supplementary tables and at http://genome.grid.wayne.edu/centisnps/. All other relevant data are available in the manuscript and its Supporting Information files.
|
10.1371/journal.pntd.0001882 | Genetic Diversity within Schistosoma haematobium: DNA Barcoding Reveals Two Distinct Groups | Schistosomiasis in one of the most prevalent parasitic diseases, affecting millions of people and animals in developing countries. Amongst the human-infective species S. haematobium is one of the most widespread causing urogenital schistosomiasis, a major human health problem across Africa, however in terms of research this human pathogen has been severely neglected.
To elucidate the genetic diversity of Schistosoma haematobium, a DNA ‘barcoding’ study was performed on parasite material collected from 41 localities representing 18 countries across Africa and the Indian Ocean Islands. Surprisingly low sequence variation was found within the mitochondrial cytochrome oxidase subunit I (cox1) and the NADH-dehydrogenase subunit 1 snad1). The 61 haplotypes found within 1978 individual samples split into two distinct groups; one (Group 1) that is predominately made up of parasites from the African mainland and the other (Group 2) that is made up of samples exclusively from the Indian Ocean Islands and the neighbouring African coastal regions. Within Group 1 there was a dominance of one particular haplotype (H1) representing 1574 (80%) of the samples analyzed. Population genetic diversity increased in samples collected from the East African coastal regions and the data suggest that there has been movement of parasites between these areas and the Indian Ocean Islands.
The high occurrence of the haplotype (H1) suggests that at some point in the recent evolutionary history of S. haematobium in Africa the population may have passed through a genetic ‘bottleneck’ followed by a population expansion. This study provides novel and extremely interesting insights into the population genetics of S. haematobium on a large geographic scale, which may have consequence for control and monitoring of urogenital schistosomiasis.
| Schistosomiasis is a disease caused by parasitic blood flukes of the genus Schistosoma. Species that infect humans are prevalent in developing countries, having a major impact on public health and well-being as well as an impediment to socioeconomic development. More people are infected with Schistosoma haematobium than with all the other schistosome species combined, however mainly due to the inability to maintain S. haematobium in the laboratory system empirical studies on this parasite are minimal. The genetic variation of this Schistosoma species on a wide geographical scale has never been investigated. In this study, we have used a DNA ?barcoding? approach to document the genetic variation and population structure of S. haematobium sampled from 18 countries across Africa and the Indian ocean Islands. The study revealed a distinct genetic separation of S. haematobium from the Indian Ocean Islands and the closely neighbouring coastal regions from S. haematobium found throughout the African mainland, the latter of which exhibited extremely low levels of mitochondrial diversity within and between populations of parasites sampled. The data from this study provides a novel insight into the population genetics of S. haematobium and will have an impact on future research strategies.
| Schistosomiasis remains one of the world's greatest neglected tropical diseases (NTD). Schistosoma haematobium is one of the most widespread species of Schistosoma and causes urogenital schistosomiasis in humans. More people are infected with S. haematobium than with all the other schistosome species combined. Of the >110 million cases of S. haematobium infection in sub-Saharan Africa, 70 million are associated with hematuria, 18 million with bladder wall pathology, and 10 million with hydronephrosis leading to severe kidney disease [1]–[2] and even bladder cancer [3]. Despite the enormous numbers of people infected with S. haematobium and the pathogenesis of the parasite's infection, empirical studies on S. haematobium are minimal, compared to those of S. mansoni and S. japonicum, due, at least in part, to the inherent logistical difficulties of maintaining S. haematobium within the laboratory system [4]. However, as the whole-genome sequence of S. haematobium has recently been published [5], further research into this most neglected of the NTDs, at least at the genomic level, may now be facilitated.
S. haematobium has a large geographical distribution being found throughout Africa, parts of the Middle East, Madagascar and the Indian Ocean Islands and is transmitted by various intermediate snail hosts within the genus Bulinus [6]. As yet, the diversity of S. haematobium has been the subject of very few molecular studies [7]–[8], although one earlier study using enzyme analyses by isoelectric focusing in polyacrylamide gels to study 22 laboratory bred isolates of S. haematobium showed some regional variation and suggested mixing of parasite strains due to human population movements [9]. It remains imperative, however, that more investigations are conducted to elucidate the extent of genetic variation across the range of this parasite if we are to realistically understand its potential evolution, transmission and perhaps ultimate control.
Praziquantel (PZQ) remains the drug of choice for treatment of schistosomiasis and for the control of morbidity. It has a good safety and therapeutic record and is easy to administer (single oral dose), generally improving the health and well-being of schistosome-infected people. National control programmes in several sub-Saharan countries aim to alleviate the burden of schistosomiasis in highly endemic areas through large-scale administration of PZQ [10]–[13] and are likely to place strong and novel selective pressures on the parasites, which may be predicted to impact their population structure and genetics [14]–[16], [8].
In recent years, developments in molecular tools, and in particular advances in DNA sequencing, have allowed greater exploration and recording of the genetic diversity of schistosome species and their hosts thereof (e.g. [17]–[18], [16]). Sequence variation in the mitochondrial cytochrome oxidase sub-unit I (cox1) gene is commonly compared between sampled specimens to identify evolutionary differences, as well as, similarities. Such studies have benefited from knowledge of the complete mitochondrial genome of S. haematobium, as well as other schistosomes [19]–[21], thereby enhancing population-focused studies [22], [17], [4]. DNA ‘barcoding’ of a large number of populations of S. mansoni to date has revealed extensive population diversity with geographical structuring existing between populations [23]–[24]. A focused study on primarily laboratory passaged S. haematobium worms from Zanzibar also revealed substantial genetic diversity with the worms splitting into two distinct phylogenetic groups [25]. Taking a similar DNA ‘barcoding’ approach, the aim of this study was to document the genetic variation of S. haematobium from several areas geographically spread across Africa using historical collections of laboratory isolates but also mainly large collections of individual schistosome miracidia and cercariae sampled directly from their natural hosts, thereby avoiding the ethical and biological biases inherent within analyses of laboratory-passaged adult worms [18]. It was expected that the data would reveal any geographical structuring of the parasite populations and the extent of the genetic diversity within and between populations across Africa. Also, due to the wide geographic spread and extensive sampling of miracidia, primarily from infected school-aged children it was predicted that the same or more genetic diversity would be found within and between populations as that found in the study on Zanzibar [25]. By so doing the origin, evolution and spread of S. haematobium on the African mainland and the Indian Ocean Islands could be further elucidated.
Ethical approval was obtained from Imperial College Research Ethics Committee (ICREC), Imperial College London in the UK, in combination with the ongoing Schistosomiasis Control Initiative (SCI) activities. In Senegal, ethical approval was obtained from the ethical committees of the Ministry of Health Dakar, Senegal. In Niger, ethical clearance was obtained from the Niger National Ethical Committee. In Cameroon, ethical approval was obtained from the Commité National d'Ethiqué, Cameroon. In Kenya, ethical approval was obtained from the Ethical Review Board of National Museums of Kenya/Kenya Medical Research Institute. In Tanzania, ethical approval was obtained from the Ethical Review Board of National Institute of Medical Research (NIMR). In Zambia, ethical clearance was obtained from the University of Zambia ethics committee.
Before conducting the study, the MoH-approved plan of action had been presented and adopted by regional and local administrative and health authorities. Meetings were held in each village to inform the village leader, heads of the families, local health authority, teachers, parents and children about the study, its purpose and to invite them to voluntarily participate. According to common practice and with approval from the Imperial College Research Ethics Committee (ICREC), due to low levels of literacy all village leaders, teachers, parents and study participants gave oral consent for the studies to take place. Informed consent for the urine examinations was obtained from each study participant and their parents or guardians. Oral consent for each participant was documented by inscription at school committees comprising of parents, teachers and community leaders. All the data were analyzed anonymously and all schistosomiasis positive participants were treated with PZQ (40 mg/kg). In schools or classes where the percentage of infections were more than 50%, mass treatment of all children was carried out at the end of the study.
The historical material stored in SCAN has been derived from UK laboratory passage of original specimens collected in collaboration with local authorities abiding by the ethical standards and collection requirements of the day and all specimens were maintained and analyzed anonymously.
Laboratory animal use was within a designated facility at the NHM regulated under the terms of the UK Animals (Scientific Procedures) Act, 1986, complying with all requirements therein, including an internal ethical review process at the NHM and regular independent Home Office inspection. Work was carried out under the Home Office project license numbers 70/4687 (-2003), 70/5935 (2003–2008), 70/6834 (2008-).
As part of a European Commission Specific Research Project (CONTRAST) ‘A multidisciplinary alliance to optimize schistosomiasis control and transmission surveillance in sub-Saharan Africa’, parasitological surveys were conducted at 13 localities in a total of seven countries across Africa between 2007–2010 (Table S1). Eggs were sampled directly from urine samples, either individually or pooled, of infected children. Eggs were concentrated from each infected urine sample by sedimentation or filtration, then rinsed in saline before transfer into a clean Petri dish containing mineral water and exposed to light to facilitate hatching of miracidia. Using a binocular microscope, individual miracidia were then captured in 2–5 µl of mineral water, pippetted onto Whatman FTA cards and allowed to dry for 1 hour [18].
Individual cercariae from naturally infected snails were also sampled from a few localities (Table S1). Snails collected from known transmission sites were placed individually or pooled into pots of fresh mineral water and exposed to light to stimulate cercarial shedding. On inspection, using a binocular microscope, visually identified schistosome cercariae were captured in 2–5 µl of mineral water, pipetted onto Whatman FTA cards and allowed to dry for 1 hour.
Laboratory passaged adult worms from nine additional localities held in the Natural History Museum, London (NHM) liquid nitrogen schistosome repository, Schistosome Collections at the Natural History Museum (SCAN), were also utilised for molecular analysis (Table S1). A feature of the biology of schistosomes relevant to the use of laboratory passaged samples for molecular analysis is that there will be a high level of selection or genetic bottlenecks imposed on the population during the passaging process and so these samples cannot realistically be treated as individual samples nor used to represent the true within-locality variation from these particular isolates [18]. Nevertheless, we chose to include these additional archived adult worm samples to increase the geographic range and scope of the current study. Individual worms sampled from the same laboratory passaged isolate/NHM number were not treated as individuals, but the different haplotypes found were used as representative data from those localities.
Adult worms were thawed on ice and total gDNA was extracted from individual males and females Table S1, using the DNeasy Blood and Tissue Kit (Qiagen Ltd, Crawley, UK) and eluted in 100 µl of buffer giving a concentration of 3.6–31.5 ng/µl of gDNA from each worm, 2 µl of which was used for PCR.
To compare variation found within the mtDNA with nuclear DNA, the complete ITS (1+2) rDNA (927 bp) was amplified from a single male and a single female worm from all localities that had representative adult worms available (Table S1). This marker was amplified and sequenced with the forward and reverse primers ITS1 + ITS2 [30] using the PCR and sequencing conditions as used for cox1. In the analysis, published S. haematobium ITS data were also included from Senegal (FJ588861), Mali (Z21716) and Zanzibar (GU257398). Sequences were aligned and any nucleotide differences recorded.
There were no differences between the ITS1+2 sequences from any of the samples. This nuclear marker therefore proved uninformative as a population genetic marker for S. haematobium.
In total 1978 cox1 sequences were analyzed: 46 from cercariae, 241 from laboratory passaged adult worms (35 published haplotype sequences from previous studies of which 27 were from 214 individual worms isolated from Zanzibar [25] and 1869 from miracidia collected directly from their human hosts. The cox1 region (956 bp) analyzed contained 70 variable sites (53 parsimony informative), and the diversity found within the S. haematobium populations sampled was low with the sequences resolving into just 61 unique haplotypes. The percentage occurrence of each haplotype varied but there was a main common haplotype (H1) that was found across most of the localities and appeared frequently on mainland Africa, representing 1574 (80%) out of the 1978 overall sequences analyzed.
The minimum spanning TCS network (Figure 1) clearly shows the dominance of H1 and splits the haplotypes into two groups, one that is made up of 28 haplotypes from Zanzibar, Coastal Kenya, Tanzania, Mafia Island, Madagascar and Mauritius and one that is made up of 46 haplotypes from all localities sampled excluding Madagascar and Mauritius. The two groups could not be linked due to too many missing steps: the network for Group 1 is more basic with H1 being a dominant central point from which, many of the other haplotypes branch off by 1 single step (1 bp change). This closely linked network, linked by single bp changes, is made of haplotypes predominantly found in mainland Africa (except Zambia) and a few haplotypes from Zanzibar. The majority of these single mutations do not form links with other haplotypes suggesting that they are random mutations that come and go but do not persist within the populations. Exceptions to this are the Egyptian haplotypes, which again branch off from the main haplotype H1 by one mutation and forms two haplotypes separated by a single bp change. The other more significant exceptions are longer branches forming more complicated networks with haplotypes from Zanzibar and its neighbouring regions Coastal Kenya and Mafia Island and also samples from Zambia. Both branches are again closely linked to H1 by single mutations with one branch also forming links with a Malawi haplotype and the other incorporates the haplotypes from Zambia. Group 2 forms a more complicated network between haplotypes and is more exclusive containing haplotypes from only the Indian Ocean Islands and the neighbouring East African regions of coastal Kenya and Tanzania.
As with the TCS analysis the same general splitting of the haplotypes was found with all phylogenetic methods separating the 61 haplotypes into two distinct well-supported groups (Figure 2) with H1 being central to the majority of the mainland African haplotypes. The details of all the samples that represent H1 can be seen in the sub tree in Figure 2. The tree topologies show the separation of samples from Coastal Kenya, Zanzibar, Mafia Island and Zambia from the main cluster in Group 1 and also the clear separation of Group 2 containing samples from the Indian Ocean Islands and its neighbouring African coastal regions.
The net divergence between the groups (0.02145±0.00102) shows a relatively short time between the genetic separation of these two S. haematobium groups compared to that of the much larger divergence from their sister taxa S. bovis (Group 1: 0.11644±0.01978, Group 2: 0.10932±0.02365)
Measures of overall haplotype and nucleotide diversity, together with within region and locality diversity, are presented in Tables 1 and 2. Sample numbers will affect the diversity especially when very low levels of samples are used, however the data do show a clear difference in diversity between those localities where large numbers of individual larvae were sampled. The number of unique haplotypes found in the western regions of Africa compared to the east is extremely small even though sampling was biased towards the west and the diversity seen in the east comes mainly from the coastal Kenyan samples. There is also a vast contrast, with an extremely high diversity found within the populations sampled from the Indian Ocean Islands and the neighbouring coastal regions compared to the rest of Africa, where one dominant haplotype appears to persist throughout the mainland. We also tested for strong selection in our data and found no deviation from neutral expectations (p values were all > than 0.05).
The nad1 haplotypes sequenced from several of the samples (Table S1) also supported the findings from the cox1 data (Figure 3). The haplotypes again split into two distinct groups; 1) dominated by a central haplotype found throughout mainland Africa and Zanzibar with a few closely linked haplotypes forming short branches and a longer branch to the Zambian haplotype and 2) containing haplotypes from Zanzibar, Madagascar and Mauritius.
DNA ‘barcoding’ approaches are now commonly used to provide insights into population structure and diversity within species including schistosomes. Studies on S. mansoni have shown high levels of mtDNA diversity within and between populations from endemic areas with haplotypes segregating by geography [23]–[24].
This study is the first time that DNA cox1 ‘barcoding’ has been used to elucidate the genetic diversity of S. haematobium populations across Africa and also from several of the Indian Ocean Islands (Zanzibar, Madagascar, Mauritius, Mafia). Ninety seven percent of the data generated came from larval stages sampled directly from their human hosts from across 20 localities, with the remainder of the data coming from historical collections based on laboratory-passaged worms. The study has revealed that the genetic diversity of S. haematobium across Africa is unexpectedly low. There were only 61 unique haplotypes found in the 1978 samples collected from 41 locations and 18 countries. The haplotypes split into two distinct groups; one that contains haplotypes predominately from mainland Africa with a few haplotypes from Zanzibar (Group 1) and the other that is made up of samples exclusively from the Indian Ocean islands and the neighbouring African coastal regions (Group 2). The net divergence between the two groups was considerable and was strongly supported by both the cox1 and the nad1 data. This is equivalent to the net divergence seen between some S. mansoni groups spread across Africa, separated by thousands of miles [24].
The lack of the diversity found within and between the S. haematobium samples can clearly be seen in the TCS network and phylogenetic analyses with a single haplotype (H1 from 1574 samples) being dominant across Africa with greater diversity found within the samples from the Indian Ocean Islands and the neighbouring African coastal regions. The nuclear ITS data showed no diversity from any sample proving that this can not be used as a population genetic marker for this parasite [24] although such a nuclear marker is vital for the detection of interactions with closely related species and for confirming species identity [31].
The longer branches stemming from the main H1 haplotype in Group 1 on the TCS analysis show that the populations from coastal Kenya, Zambia and Mafia are quite separated from the main haplotype group. The highest diversity was found in the S. haematobium populations from coastal Kenya and Zanzibar with complicated networks linking these haplotypes and several nodes not being represented by a haplotype. This suggests that haplotypes may have become extinct or that they have not been sampled indicating there may be more diversity still to be discovered in these areas. However, the basic network of single links around H1 suggests that further sampling in the other areas, reported on in this study, is unlikely to reveal further discrete groupings. With exceptions of haplotypes from the Indian Ocean Islands, Coastal Kenya and Zambia, the network clearly shows a lack of geographic structuring with the same cox1 haplotypes being found in Far West, Central West, East, South of Africa and also in Sudan, Egypt and Zanzibar.
The distribution of the haplotypes must reflect in part past movements of people. Group I and 2 parasites have been isolated from the same geographical regions and from the same host. For example, the haplotypes from Mwanza, Tanzania (TA1a + b), resolve into both network groups, one haplotype (TA1a) being identical to the H1 haplotype (Group 1) and one (TA1b) being identical to a haplotype found on Mafia Island and sitting at the end of a long branch within Group 2. This suggests that in this geographic region as well as the main dominant S. haematobium genotype there has also been transfer of Group 2 schistosomes from the Indian Ocean Islands and the neighbouring coastal regions probably associated with the movements of people [32]. Similarly, the positioning of the Mafia haplotypes (Mafia1 + 2) in the two groups shows the close affinity of this population with S. haematobium from Zanzibar. This is not unexpected due to infected children from Mafia having a travel history to the Zanzibar mainland and that urogenital schistosomiasis is suggested to be imported and not endemic on Mafia.
The higher diversity found in the coastal Kenyan populations (15 haplotypes), the close clustering with the Zanzibar haplotypes and the separation of these populations into the two groups also suggests a close association between parasites from coastal Kenya and Zanzibar. There is probably mixing and movements of the populations between coastal Kenya and Zanzibar and vice versa with the movements of people between these areas due to the trade routes between the Indian Ocean islands and the neighbouring East African coastal regions [32]. The positioning of the haplotypes from Madagascar and Mauritius in Group 2 further supports the uniqueness of the haplotypes from the Indian Ocean Islands and neighbouring East African coastal regions compared to those from mainland Africa. The recognition and distribution of Group 2 schistosomes suggests movement of S. haematobium populations from endemic adjacent regions such as Madagascar and the Arabian Peninsula. There is a history of prolific trade links between the Arabian Peninsula, India, the Indian Ocean Islands and the East coast of Africa aided by the Monsoon trade winds, which probably would have facilitated the movements of people and their parasites between these areas.
In consideration of general population genetic theories, the extremely low levels of genetic diversity between the S. haematobium populations separated by 1000's of miles across continental Africa compared to the high diversity found within the populations from the more isolated Indian Ocean Islands and their closely neighbouring African coastal regions is unexpected. It is particularly striking that the dominant haplotype H1 occurs across Africa with 1574 samples analyzed not showing a single nucleotide mutation in the mtDNA analyzed. The success of this haplotype might be attributed to a founder effect following a population ‘bottleneck’ with only a few individual parasites surviving and participating in a later population expansion. The lack of diversity found across Africa and the restriction of the Group 2 haplotypes to coastal regions of East Africa and the Indian Ocean Islands suggests that this may have happened relatively recently in terms of the evolutionary history of these parasites. Given the close phylogenetic relationship of the Indian/Asian and African Schistosoma species [33]–[34] it is possible that the lack of genetic diversity found within and between the S. haematobium populations across mainland Africa is attributed to a re-invasion by a small number of individuals of S. haematobium into Africa from a larger population in Asia across the Arabian peninsula, with a subsequent rapid spread and population expansion across Africa from East to West. The new small re-established population in Africa would be more sensitive to genetic drift and increased inbreeding resulting in low genetic variation. Due to the lack of fossil records it is extremely difficult to accurately define the evolutionary history and phylogeography of the Schistosoma genus [24], [34]–[35] however, data such as that reported here do provide new insights into how these parasites evolved and spread and c learly it will be of interest to barcode samples from the Arabian Peninsula.
One factor that could have influenced the divergence of our populations into the two groups relates to compatibility with intermediate snails hosts. Intermediate host use of S. haematobium is very specific and varies in different geographical locations [6]. However, based on current day snail distributions there is no obvious correlation between intermediate host use and the observed parasite diversity. S. haematobium in East Africa and the Indian Ocean Islands is mainly transmitted by Bulinus africanus and B. forskalii group species while elsewhere the same species groups can be involved but snails of the B. truncatus tropicus complex may also play an important role in transmission. However, the study by [36] did show that the intermediate snail host, B. globosus, separates into distinct West and East African clades on a molecular phylogenetic tree, possibly suggesting that the distribution of East African B. globosus could be a limiting factor in the spread of Group 2 type parasites. It is clear that more studies are needed to investigate the role of the different snail species and geographical populations of Bulinus in the transmission of Group 1 and Group 2 type parasites. As well as intermediate snail host compatibility it will be important to determine whether the different groups give rise to infections which result in different pathologies or which respond differently to treatments as genetic diversity has been noted to possibly have an effect on such characteristics [37].
Though the small regions of nuclear DNA analyzed in this study proved highly conserved, it is likely that there could be more diversity found in other regions of the nuclear genome, which may or may not correlate to that found in the mt DNA. A recent study [8], using a small number of microsatellite markers did find diversity in S. haematobium miracidial populations from Mali conflicting the data presented here, however only laboratory maintained material from Mali was analysed in the present study and so a direct comparison cannot be made. Due to the difficulties in directly sampling natural schistosome populations there exists a strong sampling bias within this study with the majority of the data obtained from large larval schistosome populations collected from 6 countries. The other countries are represented by far fewer or laboratory maintained samples and whilst providing useful genetic data it cannot be concluded that they are representative of the true genetic diversity in these countries. More samples need to be analyzed from more areas and with more genetic markers to further elucidate the genetic diversity of S. haematobium populations from all it's endemic areas. It would also be beneficial to analyze both nuclear and mt DNA simultaneously from the same individual sample in future population genetic studies and the recent publication of whole-genome sequence of an Egyptian isolate of S. haematobium [5] will facilitate the development of many more nuclear markers for population genetic analyses at the genomic level.
The genetic diversity of schistosome populations can be influenced by a variety of factors such as; host water-contact patterns, host immunity and susceptibility and moreover, mass chemotherapy has a great potential to promote selection. [15], [8]. The impact of the large-scale administration of PZQ, through national control programmes [11]–[12] on the genetic selection of both S. mansoni and S. haematobium is an area of high interest with respect to the development of drug resistance [38]–[40], [14]–[15], [41]. A high population diversity would be expected to provide a wide genetic base for selection to act upon possibly increasing the rate of resistance to treatment developing, ultimately resulting in a decline in diversity over time to a few, non susceptible genotypes [42]. The relatively low level of diversity within S. haematobium across most of mainland Africa, as defined by the current genetic markers, may indicate that these parasites may be less likely to change under drug pressure, however the high genetic diversity found on Zanzibar and the neighbouring African coastal region could offer a genetic base for the development of PZQ resistance and hence changes in parasite diversity in relation to chemotherapy needs to be monitored in these highly diverse areas.
This study has reported on some very unusual findings in relation to S. haematobium mtDNA population genetics. It is clear that further sampling in many areas will not dramatically increase the mtDNA diversity found but in areas such as Zambia, the East African coastal regions and the Indian Ocean islands where more diverse populations of S. haematobium have been found, further sampling would add to our understanding of the parasite population movements and diversity. It is also important that the population genetics of the S. haematobium is monitored further to link diversity with morbidity and to provide information on the response of parasite populations to drug treatment pressures. The mtDNA diversity described here, together with other molecular markers, will be of value to monitor the impact of control interventions on different S. haematobium genotypes and may assist in understanding the introduction or re-introduction of parasites associated with human population movements.
|
10.1371/journal.pntd.0005152 | Transfected Babesia bovis Expressing a Tick GST as a Live Vector Vaccine | The Rhipicephalus microplus tick is a notorious blood-feeding ectoparasite of livestock, especially cattle, responsible for massive losses in animal production. It is the main vector for transmission of pathogenic bacteria and parasites, including Babesia bovis, an intraerythrocytic apicomplexan protozoan parasite responsible for bovine Babesiosis. This study describes the development and testing of a live B. bovis vaccine expressing the protective tick antigen glutathione-S-transferase from Haemaphysalis longicornis (HlGST). The B. bovis S74-T3B parasites were electroporated with a plasmid containing the bidirectional Ef-1α (elongation factor 1 alpha) promoter of B. bovis controlling expression of two independent genes, the selectable marker GFP-BSD (green fluorescent protein–blasticidin deaminase), and HlGST fused to the MSA-1 (merozoite surface antigen 1) signal peptide from B. bovis. Electroporation followed by blasticidin selection resulted in the emergence of a mixed B. bovis transfected line (termed HlGST) in in vitro cultures, containing parasites with distinct patterns of insertion of both exogenous genes, either in or outside the Ef-1α locus. A B. bovis clonal line termed HlGST-Cln expressing intracellular GFP and HlGST in the surface of merozoites was then derived from the mixed parasite line HlGST using a fluorescent activated cell sorter. Two independent calf immunization trials were performed via intravenous inoculation of the HlGST-Cln and a previously described control consisting of an irrelevant transfected clonal line of B. bovis designated GFP-Cln. The control GFP-Cln line contains a copy of the GFP-BSD gene inserted into the Ef-1α locus of B. bovis in an identical fashion as the HIGST-Cln parasites. All animals inoculated with the HlGST-Cln and GFP-Cln transfected parasites developed mild babesiosis. Tick egg fertility and fully engorged female tick weight was reduced significantly in R. microplus feeding on HlGST-Cln-immunized calves. Collectively, these data show the efficacy of a transfected HlGST-Cln B. bovis parasite to induce detectable anti-glutathione-S-transferase antibodies and a reduction in tick size and fecundity of R. microplus feeding in experimentally inoculated animals.
| The cattle tick Rhipicephalus microplus is a hematophagous ectoparasite, responsible for the transmission of lethal parasites such as Babesia sp, limiting cattle production in tropical and subtropical regions of the world. There is an urgent emerging need for improved methods of control for these currently neglected tick and tick borne diseases. It is hypothesized that a dual attenuated-live vector vaccine containing a stably transfected tick antigen elicits protective immune responses against the parasite and the tick vector in vaccinated cattle. Live Babesia vaccines based on attenuated parasites are the only effective method available for preventing acute babesiosis. On the other hand, glutathione-S-transferase from Haemaphysalis longicornis (HlGST) is a known effective antigen against Rhipicephalus microplus, the most common vector for B. bovis. This study describes the development and testing of a transfected, B. bovis vaccine expressing HlGST against the tick R. microplus. A B. bovis clonal line designated HlGST-Cln expressing HlGST and GFP/BSD, and separately a control transfected B. bovis clonal line expressing only GFP/BSD was used to vaccinate calves in two independent experiments. All immunized calves developed mild babesiosis, and only calves immunized with the HlGST-Cln parasite line generated anti-HlGST antibodies. Tick egg fertility and fully engorged female tick weight were reduced significantly in R. microplus feeding on HlGST-Cln-vaccinated calves. Taken together, these data demonstrates the ability of transfected B. bovis to elicit antibodies against a heterologous tick antigen in cattle and to induce partial protection in the vaccinated animals against the cattle tick for the first time, representing a step toward the goal to produce a live vector anti-tick vaccine.
| The cattle tick Rhipicephalus microplus is a hematophagous ectoparasite limiting cattle production in tropical and subtropical regions of the world [1–4]. Methods to lessen the impact of R. microplus are based almost exclusively on the use of several chemical acaricides, including arsenics, organophosphorus, carbamates, chlorinated hydrocarbons, pyrethroids, macrocyclic lactones and benzoyl phenyl ureas [5]. However, this approach generates undesired consequences such as the selection of acaricide resistant tick populations and contamination of the environment and animal products [6,7]. In this scenario, alternative tick control methods, such as vaccination, are increasingly needed [8,9].
Tick vaccines for the control of cattle tick infestations such as TickGARD and Gavac [10,11] became commercially available in the early 1990's, and are both derived from the R. microplus midgut membrane-bound recombinant protein Bm86. However, none of the Bm86 derived vaccines are consistently efficient worldwide and the need for new tick vaccines remains [12,13]. Consequently, a growing number of R. microplus vaccine candidate tick proteins have been identified and evaluated, including Bm86 orthologues and homologs [14–16], tick salivary proteins [17], embryo enzymes [18,19], ribosomal protein [20], and detoxification molecules [21,22], among others.
The glutathione-S-transferases are a class of enzymes involved in detoxification of electrophilic substrates by their conjugation with glutathione [23]. GSTs from distinct species have been investigated as vaccine candidates against several parasites, such as Necator americanus [24], Schistosoma japonicum [24,25], Schistosoma mansoni [26], Trichinella spiralis [27], and Wuchereria bancrofti [28]. The use of GST in experimental vaccines resulted in variable degrees of protection against the targeted parasites, demonstrating their potential for generating protective immunity [29]. Furthermore, an experimental tick vaccine based on recombinant Haemaphysalis longicornis glutathione-S-transferase (HlGST) [30] elicited partially protective responses in bovines against R. microplus [21,22]. An additional striking and positive feature of HlGST vaccination was an increase in cattle weight gain in comparison to control animals [22].
The impact generated by R. microplus on cattle health is dual, in part due to a direct effect of attachment and blood ingestion [31], and due to the high morbidity and mortality caused by the numerous pathogens transmitted by this tick, including Babesia spp. and Anaplasma spp [7,32]. Bovine babesiosis is an acute and chronic disease caused by protozoan parasites of the genus Babesia, including B. bovis and B. bigemina [33]. If natural exposure to Babesia occurs at an early age, cattle normally develop subclinical disease and become immune to subsequent homologous parasite challenge as adults [34]. In contrast, exposure of Babesia-naive adult animals usually results in fatal acute disease [35]. Several vaccination procedures based on attenuated strains are available and commonly used as control methods to prevent acute Babesia infections in several countries [2,36]. Vaccination with live attenuated B. bovis strains usually results in mild acute and persistent infections in vaccinated calves, and the elicitation of strong immune responses conferring long-term protection against challenge with homologous and heterologous strains of the parasite [2]. Despite the risk of reversion of virulence, an important safety issue in live vaccines, B. bovis live attenuated vaccines have now been safely used as field vaccines, without reversion to virulence [37, 38].
Efficient transfection methods, which allow the incorporation and expression of foreign DNA into a parasite host genome, have been developed for B. bovis, and can also be applied to vaccine development. It was previously proposed that a transfected B. bovis expressing heterologous parasite proteins can be used as carriers to deliver selected antigens to the bovine immune system [39]. Clearly, transfection methods together with other related gene editing tools allow production of specifically designed strains for developing alternative and better defined attenuated B. bovis strains [1], and live vector vaccines effective against other parasites [2]. Ideally, such foreign antigen delivery platforms could be applied for developing dual Babesia and tick vaccines by producing a B. bovis strain able to synthetize a tick protein that induces anti-tick immune responses during cattle infection as well as the expected anti-babesia immune response [2,39]. However the ability of transfected B. bovis parasites to serve as vaccine delivery platforms remains so far an untested approach. This study describes the development and testing of a recombinant B. bovis strain able to express the tick protein HlGST and its ability to protect against a tick challenge. The results represent a step toward the goal of producing a live vectored anti-tick vaccine.
The transfection plasmid pMSASignal-HlGST-GFP-BSD is represented in Fig 1. The “B” expression site of plasmid pMSASignal-HlGST-GFP-BSD contains a chimeric gene MSA1-HlGST encoding a 21 amino acid fragment of the B. bovis MSA-1 protein corresponding to the signal peptide fused to a 672-bp fragment encoding the 222 amino-acids of the full size HlGST protein of H. longicornis (Fig 1). Plasmid pMSASignal-HlGST-GFP-BSD also includes the GFP-BSD selectable marker fusion gene cloned upstream of the Ef-1α IG region on the “A” promoter side, and flanking 5′ and 3′ Ef-1α ORF sequences to facilitate integration of the two exogenous genes and the bidirectional Ef-1α promoter into the Ef-1α locus of the B. bovis genome [45,49] (Fig 1).
Babesia bovis T3B parasites were electroporated with plasmids pMSASignal-HlGST-GFP-BSD, and control plasmids pEf-msa-1-Bm86ep-gfp-bsd [39] or pBlueScript (pBS). Blasticidin resistant parasites electroporated with plasmid pEf-msa-Bm86ep-gfp-bsd, designated Tf-Bm86ep-gfp-bsd, or plasmid pMSASignal-HlGST-GFP-BSD, termed HlGST, emerged in in vitro cultures starting 16 days after electroporation (Fig 2A). Expression of green fluorescent protein (GFP) was evident upon fluorescence microscopy in both emerging blasticidin-resistant parasite lines (Fig 2B), Transfected fluorescent parasites were also used to verify evasion of parasites from infected RBCs (S1 video). In addition, simultaneous production of the reporter (GFP) and the tick protein (HlGST) by the pMSASignal-HlGST-GFP-BSD transfected parasites, termed HlGST, was confirmed by RT-PCR and Western blot analysis (Fig 3A and 3B). The RT-PCR amplifications demonstrated transcription of both GFP-BSD and HlGST genes in the HlGST parasites maintained in culture (Fig 3A, line 1 and 2). Consistently, GFP-BSD but not of HlGST transcripts were detected in the transfected control parasite line Tf-Bm86ep-gfp-bsd (Fig 3A, line 3 of GFP-BSD and GFP boxes), and no GFP-BSD nor HlGST transcripts were detectable in non-transfected, non-blasticicin selected parasites (Fig 3A, line 4 of GFP-BSD and HlGST boxes). Also, RAP-1 (rhoptry-associated protein 1) transcripts were detected in all parasite lines tested, and no transcripts were detected when transfection plasmids were used as template in the RT-PCR reactions (Fig 3A). Additionally anti-HlGST rabbit antibodies specifically recognize a protein of approximately 30 kDa, a size which is consistent with the predicted size of the MSASignal-HlGST chimera, only in the HlGST transfected parasites in immunoblots (Fig 3B, lines 1 and 2). An approximately 42 kDa band was detected in all B. bovis culture lysates when the blots were incubated with a control monoclonal antibody against the merozoite surface antigen-1 (MSA-1) from B. bovis (Fig 3B).
Integration into the Ef-1α locus was tested by sequencing PCR amplicons derived from HlGST-transfected parasite gDNA. The PCR primers for these experiments were designed to amplify regions that include both exogenous DNA insert and a B. bovis genomic region lying adjacent to the Ef-1α locus (Fig 4, EF-GST and GFP-EF boxes). Identical PCR reactions performed on gDNA from non-transfected B. bovis or transfection plasmid pBm86ep-gfp-bsd did not result in the production of any amplification product. Sequence analysis of the PCR products demonstrated insertion of the foreign transfected genes in the targeted Ef-1α locus (GenBank accession number: KX021742).
Stable transfection experiments using a transfection plasmid containing the RFP and eGFP genes (pEf-eGFP-RFP-BSD, S1 File) using identical plasmid architecture as plasmid pMSASignal-HlGST-GFP-BSD (S1A Fig), indicated that the plasmid design used to obtain the transfected HlGST parasites can be stably incorporated into the Ef-1α locus of transfected parasites using distinct alternative patterns of insertion. Fluorescence analysis indicates that some of the distinct patterns of insertion preclude the expression of both transfected genes (GFP and RFP-BSD) by all transfected parasites (S1 Fig). These data confirmed that the stably transfected parasite line HlGST is composed by a mix of parasite subpopulations containing distinct pattern of exogenous gene integration, with some transfected parasites lacking or unable to express the MSA-1-HlGST gene. The presence of such a heterologous parasite line composition can interfere with further in vivo infection studies, which ideally requires of a homogeneous parasite population expressing both exogenous genes. Based on these observations, the HlGST-transfected culture was then submitted to a cloning procedure using a FACS method [41] in order to obtain a transfected clonal line containing and expressing both, the GFP-BSD and HlGST genes. Screening of in vitro cultures derived from FACS separated cells using a PCR based on the amplification of the rap-1 gene, identified eight rap-1 positive culture wells out of the total of 192 wells analyzed (S2 Fig). However, whereas RT-PCR analysis was performed on RNA extracted from the eight rap-1 positive wells, rap-1 transcripts were detected in seven of the eight wells (Fig 5A), while HlGST transcripts concurrent with rap-1 transcripts, were detectable in just a single FACS-separated parasite clonal line (Fig 5A), which was expanded and termed HlGST-Cln. Analysis of Clone 5 was not included on Fig 5B, since the cultured parasites were lost before characterization. Importantly, expression of HlGST in HlGST-Cln parasites was also confirmed by Western blot analysis using anti-HlGST antibodies (Fig 5B). Taken together, these results confirmed the occurrence of a mixed parasite population in the transfected parasite line HlGST which was submitted to FACS sorting, and the ensuing isolation of the clonal line HlGST-Cln able to express the GFP-BSD and the HlGST genes simultaneously.
Analysis of the pattern of insertion of the transfected HlGST and GFP-BSD genes in the HlGST-Cln line was performed by Southern blot and PCR. Intact and BglII digested gDNA extracted from the lines HlGST, HlGST-Cln, GFP-Cln [41] and non-transfected, were analyzed by Southern blots hybridized with GFP, HlGST, and Ef-1α specific dig-labeled probes. The Southern blot data, shown in Fig 5C, indicates that there is only one fragment recognized by all tested probes in the HlGST-Cln line, suggesting the presence of a homogenous parasite population containing a single copy of the exogenous HlGST and GFP-BSD genes inserted into the expected Ef-1α locus. Both GFP and HlGST probes hybridized with the transfection plasmid but did not hybridized with any BglII digested DNA from non-transfected parasites, confirming the specificity of the probes to the exogenous DNA. However, the GFP probe recognized at least three distinct types of DNA fragments derived from gDNA of the HlGST parasite line. These three distinct patterns of hybridization, named 1, 2, and 3, may be due to the presence of a homogeneous population with multiple insertions, or from a mixed population containing distinct types of insertion. Importantly, the expected 19.3 kb fragment equivalent to the insertion of the exogenous material to the elongation factor region (fragment 1) was also present. Yet, the HlGST labeled probe recognized only two DNA fragments, named 1 and 3, in the BglII digested gDNA derived from the HlGST parasites (mixed population), suggesting the presence of at least one subpopulation of transfected parasites containing only the GFP-BSD, but not the HlGST gene.
In addition, an Ef-1α-specific probe was also used in order to confirm integration of transfected genes into the expected Ef-1α- locus. This probe hybridized with several restriction fragments derived from the HlGST parasite line, designated as 1, 2 and 4 in Fig 5C. Fragment 4 is of the same size as the fragment hybridizing in the non-transfected parasites; fragment 2 has a similar size as the fragment hybridizing with the GFP probe in the clonal line, while fragment 1 is larger than the fragments 2 and 4. Because fragment 1 co-localizes with the single hybridizing fragment of the HlGST-Cln parasite line, it suggests that this DNA is derived from the subpopulation of parasites that integrated the full set of GFP-BSD and HlGST genes in the expected pattern of integration. Fragment 2 is likely derived from parasites integrating only a part of the exogenous transfected DNA, only the GFP-BSD side of the plasmid. Whilst the presence of parasites lacking the GFP-BSD genes is unlikely since all parasites recovered from cloning technique were green fluorescent and resistant to blasticidin, the presence of parasites containing only the reporter/resistance gene occurs, as represented in fragment 2. Regarding fragment 4, it is likely that it might have originated from a subpopulation of parasites with GFP-BSD insertions occurring at an alternative site, different than the Ef-1α locus, or derived from wild-type parasites still present in the transfected population. Finally, BglII digested gDNA derived from the GFP-Cln parasites were not recognized by the GST probe, confirming the specificity of the tested probes.
Interestingly, and consistent with previous observations [41], the results collectively, confirmed exclusive stable integration of the transfected genes into the Ef-1α ORF gene/locus of B. bovis. Also, the absence of co-localization of fragments in the same size of the control containing only plasmid DNA confirms the lack of free transfection plasmid or transfection-derived episomal DNA in the HlGST-Cln parasites.
Together, the data confirmed the isolation of a B. bovis transfected clonal line, termed HlGST-Cln able to express both transfected GFP-BSD and HlGST genes. Furthermore, the demonstration of co-migrating unique bands with probes EF, GST and GFP in the Southern blots is consistent with a single site of integration of the exogenous transfected genes in HlGST-Cln. We thus conclude that stable insertion of the transfected genes in the genome clonal line HlGST-Cln likely occurred as a single copy in the expected Ef-1α locus.
The ability of the clonal line HlGST-Cln to effectively express the HlGST in the external membrane of the transfected B. bovis merozoites was tested by immunofluorescences (IFA) (Fig 6). The IFA data using non-permeabilized HlGST-Cln free merozoites demonstrates that HlGST, as well as MSA-1, are effectively targeted to the merozoite surface. In contrast, the data strongly suggests that GFP, which lacks a signal peptide, is not localized in the surface layer of the non-permeabilized HlGST-Cln merozoites by specific antibodies (Fig 6).
Collectively, these data indicates that the HlGST-Cln line is an appropriate candidate for testing whether transfected parasites are able to cause acute and persistent infection in bovines while eliciting antibody responses against the HlGST protein.
Two independent immunizations were performed. The first experiment was aimed to demonstrate that infection of cattle with the HlGST-Cln parasite lines cause acute and persistent infection, remaining genetically stable, and elicit antibodies reactive with recombinant HlGST. In this experiment, two calves were experimentally infected with 5×107 infected erythrocytes of the parasite HlGST-Cln line (calves b1 and b2) and one control animal was experimentally infected with the same amount of B. bovis T3B-derived clonal line GFP-Cln parasites [41] (calf b3). All animals presented an increase in rectal temperatures above 40°C at some point during the acute stage of the disease, and reduction in hematocrit 7 days after immunization (Fig 7A). The presence of B. bovis in the blood of experimentally infected animals was confirmed by PCR (Fig 7B). While PCR revealed the presence of circulating merozoites in the blood in all animals (Fig 7B), no parasites were visualized in blood smears from jugular blood samples. Overall, these data suggests that all three experimentally infected calves developed similar clinical symptoms of mild babesiosis.
Both B. bovis strains used in immunization were culture-recovered from the blood of animals 8 days after immunization and analyzed (S3 Fig). Blasticidin-resistant fluorescent parasites were detected 10 days after the establishment of the in vitro cultures from all animals. RT-PCR, gDNA PCR, western and southern blot were performed with the recovered parasites, showing that the recovered HlGST-Cln-recovered parasites remain genetically stable, retain the ability to express the GFP-BSD and HlGST genes, and the clonal characteristic of cell lineages (S3 Fig), and thus they appear to be similar to the inoculated HlGST-Cln parasites.
Serological detection of anti-HlGST antibodies was performed using bovine sera from vaccinated and control groups. Western blot analysis show the specific recognition of recombinant HlGST by antibodies in the bovine sera from both calves (b1 and b2) experimentally inoculated with the HlGST-Cln line (Fig 8A) beginning at day 12 post-inoculation, at a 1:10 dilution which was verified until day 56 post-inoculation. Presence of antibodies reactive with HlGST confirmed expression of the transfected protein during the infection. In addition, the production of antibodies against RAP-1 was also determined routinely for each animal using a cELISA [47,48]. Anti-RAP-1 antibodies were also detected starting at 12 days post-inoculation (Fig 8B).
Once it was demonstrated that the transfected parasites were able to elicit mild acute and persistent infection, remain genetically stable, and generate anti-HlGST antibodies, we investigated whether calves infected with the parasite line HlGST-cln are able to interfere with tick development upon tick challenge in a separate experiment. To test this, we infected a group of three age-matched Hereford calves with the transfected strain HlGST-Cln (animals B1, B2 and B3) and three age-matched calves with the GFP-Cln transfected control strain (animals B4, B5 and B6). All infected animals received a 5x107 parasite inoculum. Similar as in the previously described immunization experiment, hematocrit and the rectal temperature were measured every day during the first 10 days after immunization. All animals had a gradual reduction in hematocrit after immunization, a clinical characteristic signal of acute babesiosis (S4A Fig). Rectal temperatures were measured in the same period and according to the threshold for fever determination, only one of the six animals showed increased rectal temperature above 40°C (S4B Fig). Interestingly, upon comparison to the 3 Holstein calves used in the first immunization experiment, the 6 Hereford calves of the vaccination trial presented a lighter response to B. bovis infection. Fibrinogen, an important acute phase protein [50], was also measured daily during the 10 days after immunization as an indicator for the presence of acute infection. All animals presented significant increase in fibrinogen levels during days 3 and 4 (p<0.01) post-infection, compared with pre-vaccination levels (S4C Fig), but fibrinogen levels were reduced thereafter.
The six animals used in the vaccination trial experiment were also subjected to additional biochemical and hematologic exams prior to vaccination and 5 and 10 days after vaccination. This intensive clinical following was done in order to check the possibility of other Babesia-unrelated clinical conditions in the animals subjected to vaccination with the recombinant parasite. Urea, creatinine, aspartate aminotransferase, alkaline phosphatase, albumin, total protein and total globulin were tested using immunized animals serum. None of those assays presented a significant clinical change after vaccination (S1 Table). For hematological parameters a reduction in total leukocytes was verified in day 5 after immunization (P<0.05), but all animals already recovered at day 10 after immunization and by then, their leukocyte levels were indistinguishable from the levels prior to immunization (P>0.05) (S2 Table). Despite total leukocyte reduction, it was not possible to evaluate any specific reduction among neutrophils, lymphocytes, monocytes, eosinophils or basophils (Repeated measures ANOVA P>0.05). All hematologic values are described in S2 Table. Importantly, even with the reduction at day 5, the leukocyte total counting stayed at levels considered similar to the normal reference value determined for bovines (S2 Table).
Serological analysis on the HlGST-Cln vaccinated animals showed presence of detectable anti-HlGST antibodies in immunoblot analysis, corroborating with the previous immunization experiment (S5 Fig). All experimental animals were subjected to a tick challenge for further collection of engorged females thirty days after immunization. Evaluation in the number and weight of fully engorged females demonstrated a significant reduction in individual tick weight among ticks derived from the three animals experimentally infected with the HlGST-Cln line (p<0.05) (Table 1), even though no difference in total weight or tick number was detected (p>0.05) (Table 1). In addition, egg fertility was reduced in ticks obtained from the calves vaccinated with transfected parasites expressing HlGST (P<0.05) (Table 1) compared with the GFP-control group.
The B. bovis protozoan presents a highly complex life cycle that includes a bovine host and Rhipicephalus microplus tick vector. The ability of the tick to perform a transovarian transmission to the new generation represents an effective mechanism for babesial dissemination and reinforces the critical role of the tick as a vector. As a result, efficient tick vector control is an essential strategy for eradication of this disease [51].
Live vector vaccine approaches are well described using simple organisms such as virus and bacteria as delivery platforms [52–56]. This vaccine methodology has the potential advantage of presenting foreign antigen to the immune system in the context of an infection, which can induce a better immune response, and also be able to amplify the stimuli due to organism multiplication, which is different and potentially more effective than subunit vaccine approaches [56]. However, few studies using eukaryotes as live vectors are currently available in the literature. These include the use of Toxoplasma gondii as a live vaccine vector against Eimeria tenella infection in chickens [57], transgenic Leishmania tarentolae against the pathogenic strains L. donovani and L. infantum [58], and a construction of Neospora caninum stably expressing a T gondii protein for further evaluation of its protective effects against T. gondii infection in mice [59].
A limiting step to achieve an efficient system for delivery of heterologous antigens via a recombinant live vector is the availability of a genetic modification tool that permits the modification of the desired vectors, such as transfection. For apicomplexan parasites as Plasmodium sp and Toxoplasma sp efficient transfection and gene editing methods have been developed [60–64]. Unfortunately, much less progress has been achieved for the genetic manipulation of B. bovis. Overcoming our limited ability to genetically manipulate this organism is vital to the better understanding of the biology of this parasite. A B. bovis transfection system was previously developed and can be useful for both vaccine development and study of the parasite biology. As shown in the S1 Video, fluorescent transfected parasites of the line HlGST allow direct visualization of specific parasite mechanisms of interest such as infected red blood cell lysis, and probably erythrocyte invasion, as previously showed [65], using fluorescent microscopy techniques. In addition, transfection techniques can be instrumental for developing novel vaccine approaches, including the development of a vaccine delivery system based on transfected Babesia parasites. Ideally, such vaccines should be designed to contain a homogeneous population of parasites able to express a heterologous antigen of interest during the natural course of infection. Also, the gene coding for the heterologous antigen of interest should remain stably integrated to the genome of the vector parasite even after several replication cycles of the vaccine vector in the infected host.
It is also important to determine whether transfection results in fitness cost to the parasite. In previous papers [41] it was showed that the transfection targeting the B. bovis ef -1α locus, such as performed in this study, do not alter the growth of parasites compared to the non-transfected control T3Bo parasites. In addition, in vivo infection studies comparing such transfected vs non-transfected parental parasites, [66] suggested the lack of apparent fitness costs to the parasite. These studies concluded that transfected parasites are genetically stable, and possess the characteristics required for a recombinant attenuated B. bovis vaccine.
Transfection of plasmid pMSASignal-HlGST-GFP-BSD into S74-T3B B. bovis parasites resulted in the stable integration of exogenous genes into the genome of the parasites. This plasmid was designed for the insertion into the Ef-1α locus and for the expression of a chimera version of the HlGST gene driven by the Ef-1α promoter “B”. The chimera gene included the signal peptide of the B. bovis MSA-1 fused to the full size HlGST orf. This fragment coding for the MSA-1 signal peptide was added to the 5’ region of the gene coding for HlGST in order to facilitate surface expression of the protein, as previously demonstrated [39] a configuration likely resulting in improved immunogenicity.
The transfected plasmid encoding for HlGST was able to successfully integrate into the B. bovis genome. However, and likely as a result of the complexity of the transfection construct containing regions that can facilitate homologous recombination, not all transfected parasites have the same integration profile and not all of them were able to express both transfected proteins simultaneously, which became clearly evident when a similar dual fluorescent construction was tested (S1 Fig). This observation is relevant to vaccine development since, as mentioned before, ideally a live vaccine should be based on a single homogeneous population in order to avoid the possible occurrence of selection mechanisms for non-vaccine relevant parasite subpopulation during infection [67]. In this scenario, all parasites of the vaccine strain should also be able to constitutively express the antigen of interest during the infection in order to maximize antigen exposure to the immune system. In preliminary experiments using parasites transfected with the dual promoter controlling expression of the RFP-BSD and eGFP genes (S1 Fig), we found a majority of parasites growing in in vitro cultures selected with blasticidin only expressing RFP, likely because the plasmid can insert in the genome in alternative patterns, and the RFP gene is linked to the blasticidin resistance gene. Interestingly, this data is consistent with the genetic and expression analyses of several clonal lines derived from the HlGST transfected parasite lines indicating that a great proportion of the transfected parasites did not present insertion of the HlGST ORF and consequently were unable to express the heterologous protein, and thus, irrelevant components for a vaccine based on transfected parasites expressing heterologous antigens. Together, these findings emphasize the need for further parasite selection following transfection and blasticidin selection using parasite cloning methods.
The availability of a clonal line expressing both, the GFP-BSD and HlGST proteins allowed in vivo infection in bovines. The initial experimental infections study performed in Holstein calves showed that the HlGST-Cln parasites are able to cause mild acute and persistent infections in the bovine host, both desirable attributes of a live vaccine. Analysis of recovered parasites demonstrated that these parasites remained genetically stable, and able to express the heterologous protein. Importantly, the HlGST protein generated by the transfected parasites during infection was able to elicit humoral immune responses that recognize the recombinant HlGST protein. Thus, the data obtained in the first in vivo experiments supported further testing of the experimental vaccine using a larger number of animals, and followed by tick challenge after immunization.
The second experimental immunization study included, in addition to the traditional hematocrit and temperature measurements, a more intensive and multifactorial panel of clinical studies, in order to verify if animals subjected to vaccination with parasites of the HlGST-Cln line developed additional clinical alterations. All infected animals presented the classical signs of babesiosis (temperature increase and hematocrit reduction) but none of the animals were prostrated and only one of them presented a temperature above of the threshold considered as fever. Consistent with the previous experiment involving Holstein calves, the Hereford calves also develop mild disease upon infection, but to a lesser degree. Difference in babesiosis susceptibility is well characterized between Bos taurus taurus and Bos taurus indicus cattle, the former being more susceptible to babesiosis [68]. It is also known that there are differences in the response to Babesia infection among cattle belonging to distinct Bos taurus breeds [69], which could be responsible for the differences observed in the response to infection among the two groups of animals tested in these studies. Taken the data of the second bovine trial together, none of the infected animals presented alterations in the biochemical parameters measured in the study (urea, creatinine, etc.) suggesting that vaccination with transfected parasites did not compromise the overall fitness of vaccinated calves.
Calves experimentally infected with the HlGST-cln parasites in both immunization experiments developed relatively low antibody titers against HlGST, and both serum presented recognition of recombinant HlGST only at a 1:10 dilution. However, the second vaccination experiment also demonstrated anti-tick activity for ticks feeding in vaccinated animals. These data indicates that the humoral response against HlGST expressed by transfected parasites was relatively weak and markedly lower in comparison with the response generated by animals immunized with recombinant protein in previous investigations [21,22]. This outcome is similar to the findings described by Zou et al [57] that used an engineered strain of T. gondii designed to express the yellow fluorescent protein (YFP) in the cytoplasm in order to test protection of vaccinated chickens against another engineered pathogen, a strain of E. tenella also expressing YFP [57]. They report that animals immunized with the transgenic apicomplexan also developed a partial protection, but anti-YFP antibody titers in chickens immunized with the transgenic parasites were markedly lower than those in animals immunized with recombinant YFP protein [57]. At least for the HlGST transfected B. bovis parasites, it is not possible to discard the possibility that the low humoral response was due to reduced levels of expression of HlGST during infection, which could be related to the transfection plasmid design used in this study.
In this work, A DNA fragment coding for the MSA-1 signal peptide was added to the 5’ region of the gene coding for HlGST in order to facilitate surface expression of the protein, as previously demonstrated [39]. However, despite the confirmed expression of the HlGST in the surface of the transfected parasites of the clonal line in IFA experiments, poor immunogenicity was observed in our study. Collectively, these data suggests that surface exposure of the exogenous antigen might be a necessary but not per se a sufficient requirement for increased antigenicity. Consistently, other previous work using a live vector vaccine approach with a trypanosomatid-based delivery system [72] also showed that externalization of the antigen of interest in the outer membrane of the parasite was not sufficient to induce a strong humoral response. Only when the fusion of the antigen to the N-terminus of a protein responsible for extracellular secretion was done it was possible to see an increased humoral response. It is also possible that regulation of the expression of the ef-1α B promoter is different among cultured and in vivo developed parasites, but testing this possibility was beyond the scope of our study. Alternative solutions to this potential limitation include the use of alternative stronger blood stage promoters, and/or the use of high-gene copy-number expression plasmids. However, the latter approach might be difficult to achieve since larger DNA inserts might be unstable and can potentially compromise the overall fitness of the live vector [70]. Alternatively, it is also possible to target expression of HlGST on the surface of infected erythrocyte, rather than in the merozoite surface. This mode of presentation could continuously potentially expose the antigen to the immune system and subsequently induce stronger immune responses. The B. bovis variant erythrocyte surface antigen (VESA) is known to be exported to the external membrane of erythrocyte [71]. However, further analysis of the mechanisms used by this protein for erythrocyte surface exposure is needed in order to test this alternative strategy in the transfected antigen of interest. Finally, another possible alternative is targeting secretion of desired antigen to the extracellular milieu [72], however further analysis of the mechanisms used for protein secretion in B. bovis are also necessary before this approach can be tested.
Remarkably, and despite the presence of relatively low amounts of anti HlGST antibodies, the animals immunized with HlGST-cln parasites in the second vaccination experiment presented a statistically significant reduction in egg fertility and in individual fully engorged female tick weight in comparison with GFP immunized control animals upon challenge with tick larvae. In contrast, it was previously found that vaccination of cattle with recombinant GST [21,22] resulted in a strong anti HlGST humoral immune response and effective protection likely due to a drastic reduction in the amount of eggs produced in ticks feeding on immunized animals. Although vaccination using these two procedures is based on a similar subunit antigen approach, they use different delivery strategies, which may result in dramatic differences in the outcomes upon tick challenge [73]. These differences include conformation of the antigen, the amounts and timing of antigen delivered, adjuvant effects, the possible involvement of different population of antigen-presenting cells, etc. Several vaccines work effectively through eliciting antibodies in serum or on mucosa in order to induce protection, and consequently the presence of antibodies correlates with effective infection blocking. However protective outcomes not only depend on the quantity of antibodies, but also of its functional characteristics [74] which can be influenced by the method of delivery and antigen presentation mechanisms.
Another hypothesis that should be tested in the future is the use of a Babesia based live vector vaccine as a dual vaccine. However, the focus of this study was limited to the development of an anti-tick vaccine, and consequently the ability of this vaccine to protect against further B. bovis challenge was not analyzed. In order to exploit this dual vaccine characteristic, further transfection assays should be done using B. bovis attenuated strains, such the ones used in live vaccines formulations.
In summary, we described a transfected B. bovis strain able to express HlGST, a previously demonstrated protective tick antigen that elicits immune responses in the bovine host. Also, we demonstrated that vaccination of calves with the recombinant vaccine caused mild acute disease and did not compromise their general fitness. However, vaccination with HlGST resulted in weak antibody responses against HlGST. Importantly, the vaccine was able to interfere with the life cycle of the tick vectors feeding in the vaccinated animals despite of low HlGST antibody titers. Regardless of the comparisons among recombinant and vectored antigen presentation, this work suggests that the hemoprotozoan B. bovis can be used as a live vector, but its ability to elicit strong humoral responses against the target antigen needs to be improved [75]. In addition, the design of transfection plasmids should be optimized for unambiguous insertion of the transfected genes into the genome.
In conclusion, these experiments provided important information as the basis to guide further transfection plasmid construction in order to obtain a more fitted and antigenic transfected parasite to be used in a dual live vector vaccine against B. bovis, ticks or even distinct parasites.
B. bovis strain S74-T3B [40] and T3B-derived clonal line Tf-149-6 C6 [41] hereby renamed as GFP-Cln, were maintained as a cryopreserved stabilate in liquid nitrogen. Parasites were grown in long term at a stationary phase culture using 10% of bovine red blood cells (RBC) in HL-1 medium supplemented with bovine serum as described by Levy [42] and maintained at 37°C and 5% CO2.
The transfection plasmid pEf-msa-1-Bm86ep-gfp-bsd described by Laughery et al [39], was used as a backbone to construct the pMSASignal-HlGST plasmid for stable transfection. The SacII restriction fragment of plasmid pEf-msa-1-Bm86ep-gfp-bsd containing the MSA1-BM86 chimera gene was removed by restriction enzyme digestion with SacII and replaced by a DNA fragment coding for the MSASignal-HlGST fusion gene.
For MSASignal-HlGST insert construction the sequence of the B. bovis MSAI signal peptide containing the restriction sites BamHI (Invitrogen), NotI (Invitrogen) and SacII (Invitrogen) was designed, synthesized (Integrated DNA Technologies) and amplified with primers described in Table 2 (MSA-SigBam F and MSA-SigNot/Eco R). The amplicon was cloned into pCR 2.1-TOPO (Thermo Fisher Scientific) cloning vector. The MSASignal fragment cloned in pCR TOPO 2.1 was digested with BamHI (Invitrogen), purified and then ligated in plasmid pBlueScript (pBS) vector previously digested with same enzymes, and named pBlue-MSA plasmid. The HlGST sequence was amplified using HlGST-BamHI-SacII F and HlGST-PstI R primers (Table 2) using the plasmid pET43a-HlGST [30] as template. This PCR product was cloned in pCR TOPO 2.1, and termed p2.1-HlGST plasmid. p2.1-HlGST was digested with SacII (Invitrogen) and PstI (Invitrogen) yielding a restriction fragment containing the HlGST gene, which was ligated into the plasmid pBlue-MSASig previously digested with the same restriction enzymes. The resulting plasmid, termed pBlue-MSASig-HlGST was then digested with NotI (Invitrogen) and PstI (Invitrogen) for ligation into the backbone transfection plasmid pEf-msa-1-Bm86ep-gfp-bsd, also digested with the same enzymes. All constructs prepared during these steps were sequenced in order to assure the absence of mutations. The final plasmid obtained was designated pMSASignal-HlGST-GFP-BSD, and is represented in Fig 1. Plasmid pMSASignal-HlGST-GFP-BSD was purified using Plasmid Plus Maxi Columns (Qiagen) for transfections.
Plasmid pMSASignal-HlGST-GFP-BSD together with control plasmids pBlueScript and pEf-msa-1-Bm86ep-gfp-bsd [39] were used for transfections. Twenty μg of each plasmid were suspended in 25 μL of cytomix buffer (120 mM KCl, 0.15 mM CaCl2, 10 mM K2HPO4/KH2PO4, pH 7.6, 25 mM Hepes, 2 mM EGTA, and 5 mM MgCl2, final pH 7.6). Parasites were obtained from a flask expansion. The infected red blood cells (iRBC) were centrifuged at 500 g for 5 min to sediment the cells that were washed once in cold filter sterilized cytomix buffer. The final washed cell pellet was re-suspended in volume/volume of cytomix solution to be further added to plasmid. Electroporation was performed in a Gene PulserII apparatus (Bio-Rad) using 0.2 cm cuvettes containing the plasmid/iRBC/cytomix solution, and settings used were 1.2 kV, 200 Ω and 25 fixed capacitance [43,44]. 20 μg of plasmid were suspended in 25 μL of cytomix buffer and electroporated with 75μl of bovine iRBC with a 56% parasitemia.
Following electroporation, iRBC were incubated in 24 well plates containing 1 mL of culture medium and 100 μL of RBC. Four hours after electroporation the medium was changed and selective agent, blasticidin (Invitrogen), added to a final concentration of 4μg/mL. Parasitemia was checked, twice a week, by counting of Diff-Quik (Dade Behring) stained blood smear slides in an optic microscope as described by Suarez and McElwain [44].
Genomic DNA of transfected and control parasites was obtained from cultured parasites as described [45] and used as template for PCR assays designed for analysis of the insertion pattern of the foreign transfected genes into the B. bovis ef-1α locus (integration PCR). A PCR designed to determine the pattern of transfected sequences into the B. bovis ef-1α locus was performed using two pairs of primers: the first Ef-Pr F8 + GST-BamHI-SacII F and the second, UPS-Ef-probe-R + Tracer-EcoRV-gfp-F (Table 2). Both forward primers anneal in a sequence originally present in the plasmid used for transfection, and both reverse primers anneal in a B. bovis genome region located in the vicinity of the B. bovis Ef-1α locus. Amplification of GFP/GST ORF was performed with primers Tracer-gfp-EcoIF and EcoRV-bsd-R and HlGST was amplified using GST-BamHI-SacII F and GST-PstI R primers (Table 2). Amplification of RAP-1 transcript was used as a control for presence of gDNA and performed with primers BoNF and BoNR (Table 2). PCR products were analyzed in 1% agarose gels and cloned in Topo 2.1 vector (Invitrogen) for posterior sequencing.
Genomic DNA was also used for southern blot analysis. Digoxigenin-labeled probes representing the HlGST ORF (GST Probe), the GFP ORF (GFP probe) and a 300 bp region upstream of the Ef-1α locus (EF Probe), were prepared by PCR amplification using a PCR Dig-Probe Synthesis kit (Boehringer–Roche). The GST probe was prepared by PCR with GST-BamHI-SacII F and GST-PstI R primers (Table 2) using the pHlGST-pET43 plasmid as template. The EF and GFP-BSD probes were prepared as described by Suarez and McElwain [45]. Total DNA from B. bovis merozoites was digested with BglII, electrophoresed during 16h at 20V, capillary transferred to ZetaProbe nylon membranes (Bio-Rad) and hybridized with dig-labeled GST, GFP and EF probes, as previously described by Suarez and McElwain [45]. BglII do not the casset inserted into babesia genome. The gDNA extracted from a previously described B. bovis T3B-derived clonal line TF-149-6 C6 [41], and redenominated GFP-Cln in this work, and plasmid DNA obtained from pMSASignal-HlGST-GFP-BSD and pGFP/BSD/Ef (the plasmid used in TF-149-6 transfection) were all used as controls in the Southern blots.
Expression of the reporter gene GFP was analyzed by fluorescence analysis using a Zeiss Axioskop fluorescent microscope (Carl Zeiss Micro Imaging) on in vitro cultured transfected parasite as previously described [45].
Fluorescent parasites were analyzed by RT-PCR to check for the presence of GST, GFP-BSD and RAP-1 transcripts. B. bovis merozoite total RNA was extracted from in vitro cultures by the standard TRIzol (Life Technologies) procedure as described previously [46], and treated with RNAse-free DNAse (Ambion). cDNA was generated using the Superscript First-Strand Synthesis System kit (Invitrogen) from 1 μg of total RNA. A fragment of the GFP/BSD ORF transcript was amplified from the cDNA either with the primers Tracer-gfp-EcoIF and EcoRV-bsd-R and the GST transcript was amplified using GST-BamHI-SacII F and GST-PstI R primers (Table 2). Amplification of RAP-1 transcript, used as a wild-type and parasite-derived constitutive control, was performed with primers BoN-F and BoN-R [47] (Table 2). Products of RT-PCR were cloned into vector pCR TOPO 2.1 (Invitrogen) and sequenced.
Protein expression was determined by Western blot analysis using whole culture lysates as previously described [46]. Equal amounts of protein (5 μg) were applied per lane in a 4–20% pre packed gel (Bio-Rad) and submitted to SDS-PAGE. Immunoblots were developed in a nitrocellulose membrane with anti-HlGST rabbit serum at a dilution of 1:1,000, anti-GFP antibody (Invitrogen) at a dilution of 1:1,000 and goat anti-rabbit-immunoglobulin peroxidase conjugate (Life Biosciences). The anti-MSA1 monoclonal antibody BABB35 [39] was used as a positive control for the immunoblots at a concentration of 2μg/ml. Purified recombinant protein produced from pET43a-HlGST [30] was used for anti HlGST antibody production. One rabbit was inoculated four times at 15 days intervals by subcutaneous route with 100 μg of recombinant protein. Protein concentration was determined according to the Bradford technique.
Immunofluorescence of extraerythrocytic merozoites was performed using the HlGST clonal line. Merozoites were isolated from HIGST-Cln parasite line with parasitemia over 30% by centrifugation two times at 400 RCF to remove the RBC with a final centrifugation at 2,000 RCF to pellet the merozoites and washed in 3% bovine serum albumin (BSA) PBS. Half of the isolated merozoites were then fixed for 10 minutes using 100% acetone and permeabilized by incubation with Triton X-100 0.1%. The remaining free non-permeabilized merozoites were incubated in 10% BSA with a combination of either 1) anti-GST (1/500) and anti-MSA-1 (mAb BABB35) (7μg/ml) and 2) anti-GST (1/500) for one hour. The cells were then washed in PBS two times with a 400 RCF centrifugation and incubated with 1:1000 10% BSA dilutions of either 1) goat-anti-rabbit Alexa Fluor 555 and goat-anti-mouse Alexa Fluor 488 and 2) goat-anti-rabbit Alexa Fluor 555 and anti-GFP conjugated with Alexa Fluor 488 for one hour. The cells were again washed two times, dried to a slide and mounted with Prolong Gold anti-fade with DAPI. The slides with permeabilized cells were incubated with either 1) anti-GST or 2) pre-immune rabbit, anti-Tryp, a non-relevant monoclonal antibody, for one hour, washed two times in PBS, and then incubated with 1) goat-anti-rabbit Alexa Fluor 555 and anti-GFP conjugated with Alexa Fluor 488, or 2) goat-anti-mouse Alexa Fluor 488 and goat-anti-mouse Alexa Fluor for one hour. All slides were then analyzed with epifluorescence microscopy to produce merged images.
Flow cytometry was used to obtain a clonal line as described previously [41]. Briefly, 50 μL of a growing culture with 9% PPE was washed once in culture medium and diluted in medium to obtain a cell density suitable for single cell sorting with a FACSVantage cell sorter (Becton-Dickinson) with Diva Software. Two 96 well plates were prepared with 200 μL of a 10% solution of RBC in culture medium. After sorting, individual infected cells were deposited into 96 well culture plates prepared with 200 μL of a 10% solution of RBC in culture medium, and cultured in a 3% oxygen atmosphere. Screening of individual culture wells for parasite DNA was performed using PCR with RAP (BoN-F and Bon-R) primers. Positive wells were transferred to a 48 well plate and RNA and protein collected for expression analysis.
Holstein calves were obtained at 3–6 months of age from a Washington State dairy. Animal procedures were approved by the University of Idaho Animal Care and Use Committee (#2013–66) in accordance with institutional guidelines based on the U.S. National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals.
Hereford calves obtained at seven-month old of age were acquired from a naturally tick-free area, housed in individual tick-proof pens on slatted floors and maintained at the Faculdade de Veterinária, Universidade Federal do Rio Grande do Sul, Brazil. Animal care was in accordance with institutional guidelines. Animal procedures were approved by the Universidade Federal do Rio Grande do Sul ethics comitee (#26247).
Two four to five months old spleen-intact Holstein calves (bovine 1 and 2 –b1 and b2) were experimentally infected with cultured B. bovis parasites of the clonal parasite line HlGST-Cln and one age-matched Holstein calf (bovine 3- b3) was experimentally infected with the parasite line control GFP-Cln. All experimentally infected animals were infected with 5×107 infected erythrocytes, in a total volume of 3 mL, via intravenous route. All animals were monitored for signs of acute babesiosis: parasitemia, fever and hematocrit. Blood samples were collected daily after the infection for DNA extraction to monitor infection by PCR. Seven days after inoculation 250 mL of blood were collected, defibrinated and cultured in a 48 well culture plate, in a 3% oxygen atmosphere at 37°C, for recovery of parasites from the blood of infected animals.
Two weeks after immunization serum samples were collected to detect the presence of anti-GST or anti-MSA antibodies, using western blot (as described in expression analysis). A cELISA for the detection of B. bovis anti RAP-1 antibodies was performed using a kit provided by VMRD (Pullman, WA) on serum samples, as previously described [47,48]. cELISA was performed as described in [48]. 5ng of RAP-1 antigen was used for plate coating. Antigen-coated plates were blocked with PBS plus 0.2% Tween 20 containing 20% nonfat dry milk for 1 h at room temperature, followed by 100 μl of test sera in duplicate wells for 30 min. After the serum from each well was removed, 100 μl (50 ng/well) of BABB75A4 MAb was added, and the plates were incubated at room temperature for 15 min. The percent inhibition of the mean of test sample wells was computed as follows: 100 − [(the OD of the test sample/the mean OD of the normal control serum panel) × 100].
For the second animal trial involving tick challenge, six seven-month old Hereford animals were experimentally infected with 5×107 infected erythrocytes, in a total volume of 3 mL, using the intravenous route. These calves were randomly divided into two groups of three test (HlGST-Cln parasites–B1, B2 and B3) and three control (GFP parasites–B4, B5 and B6) animals. The calves were monitored for signs of acute babesiosis including parasitemia, rectal fever, hematocrit and fibrinogen, daily, during 10 days after immunization, and also prior to the inoculation to check basal levels. All animals were also tested for serological levels of creatinine, urea, aspartate aminotransferase, alkaline phosphatase, albumin, total proteins and globulins, and a complete hemogram panel. Physiological data was statistically analyzed using repeated-measures analysis of variance with a post hoc Tukey-Kramer. Blood and serum samples were collected before immunization and 5 and 10 days after the inoculation. Levels of GST-specific antibodies in the serum samples were assessed by dot-blot. Nitrocellulose membrane circles were coated with 3 μg of recombinant HlGST antigen. The membranes were dried and incubated for 1 hour with a 2.5% skim milk in PBST blocking solution prior to probing with sera from B1-B6 animals at a 1:10 dilution for 16h. Anti-IgG alkaline phosphatase (Sigma) conjugate was used as secondary antibodies and the results were visualized using NBT (Fermentas) and BCIP (Fermentas). Antibody binding was evaluated by membrane scanning using software Image J and used to compare the difference among pre-immune (day 1) and post-immunization (day 30) cattle sera from vaccinated and control groups. Color intensity difference data was statistically analyzed using repeated-measures analysis of variance with a post hoc Tukey-Kramer.
Thirty days after immunization, all six calves were infested with approximately 20,000 10-day-old tick larva (from 1g of R. microplus Porto Alegre strain hatched eggs) placed on the dorsal region of each calf. From day 20 after infestation until the end of adult tick feeding period, all tick females that had dropped from the host were collected, counted and weighed daily. A total of 5 g of engorged adult female ticks from each animal, per day, were kept in petri dishes at 28°C and 85% relative humidity to evaluate oviposition, through the calculation of egg laying capacity, egg hatching and calculation of egg fertility. Egg laying capacity was obtained by calculating the ratio between total weight of females placed for egg laying and the total weight of resultant eggs. Egg fertility was calculated as the ratio between total egg weight and weight of hatched larvae from those eggs. All data collected after infestation was analyzed using standard t-test.
|
10.1371/journal.ppat.1000007 | The Interferon Response Inhibits HIV Particle Production by Induction of TRIM22 | Treatment of human cells with Type 1 interferons restricts HIV replication. Here we report that the tripartite motif protein TRIM22 is a key mediator. We used transcriptional profiling to identify cellular genes that were induced by interferon treatment and identified TRIM22 as one of the most strongly up-regulated genes. We confirmed, as in previous studies, that TRIM22 over-expression inhibited HIV replication. To assess the role of TRIM22 expressed under natural inducing conditions, we compared the effects of interferon in cells depleted for TRIM22 using RNAi and found that HIV particle release was significantly increased in the knockdown, implying that TRIM22 acts as a natural antiviral effector. Further studies showed that TRIM22 inhibited budding of virus-like particles containing Gag only, indicating that Gag was the target of TRIM22. TRIM22 did not block the release of MLV or EIAV Gag particles. Inhibition was associated with diffuse cytoplasmic staining of HIV Gag rather than accumulation at the plasma membrane, suggesting TRIM22 disrupts proper trafficking. Mutational analyses of TRIM22 showed that the catalytic amino acids Cys15 and Cys18 of the RING domain are required for TRIM22 antiviral activity. These data disclose a pathway by which Type 1 interferons obstruct HIV replication.
| Interferons are produced by cells in response to challenge by foreign pathogens such as viruses. The molecular mechanisms by which Type I interferons (e.g., IFNβ) inhibit the replication of HIV-1 are not fully clarified. We identified a gene called TRIM22 that belongs to the tripartite motif (TRIM) family that was strongly induced by IFNβ. Using RNA interference to reduce the expression of TRIM22, we showed that TRIM22 is a key mediator of the IFNβ response when expressed at natural levels. We demonstrate that TRIM22 blocks the intracellular trafficking of the viral structural protein Gag to the surface of the cell, and that the antiviral activity of TRIM22 is dependent on two cysteine residues (Cys15 and Cys18) that are critical for the E3 ligase activity of RING-containing proteins. This report describes a mechanism by which Type I interferons block HIV-1 replication.
| The interferon system is a well studied branch of the innate immune system active against viruses. Infection of vertebrate cells by many viruses provokes synthesis and secretion of interferons (IFNs), which mediate induction of a cellular antiviral state that obstructs further viral spread. Type I IFNs (α and β) are produced by many cell types, while Type II IFN (gamma) is produced by immune cells. IFN-induced signaling pathways begin with IFN binding to IFN receptors at the cell surface. This results in signal transduction via the Jak/Stat pathway, which leads to activation of interferon-responsive genes and synthesis of effector proteins, including PKR, RNAseL, Mx, and many others (reviewed in [1],[2],[3],[4]).
For HIV, previous studies have implicated IFNs in blocking both early and late stages of the HIV-1 lifecycle [5],[6],[7],[8],[9],[10]. In many cell types, the inhibition of later steps in the life cycle following integration seems to be most potent (reviewed in [2]). Although the effector mechanisms acting late have not been fully clarified, one report suggested that the IFN-inducible protein ISG15 interfered with the endosomal trafficking pathway used by HIV-1 to exit 293T cells by blocking the interaction of TSG101 with HIV-1 Gag [11].
Many tripartite motif (TRIM) proteins may function in innate immunity to restrict viral replication. TRIM5α from rhesus macaque (RhTRIM5α) blocks early replication steps of HIV-1 and other retroviruses (reviewed in [12]); in addition, one group has proposed that RhTRIM5α acts late during HIV replication as well [13]. The early target of RhTRIM5α is the capsid (CA) protein, which forms the protein shell of the viral core. Anti-HIV-1 activity has also been reported for other TRIM family members. TRIM1 has been shown to target the CA protein at an early stage pre-reverse transcription, TRIM22 has been suggested to affect HIV transcription and TRIM19 and TRIM32 have been suggested to affect trafficking of viral proteins (reviewed in [14]). Recently, TRIM28 was shown to repress transcription from a retroviral promoter by binding to proviral DNA [15].
TRIM proteins contain a highly conserved RBCC motif comprised of a RING domain, one or two B-box domains, and a predicted coiled-coil region (reviewed in [14]). The RING domain contains a specialized zinc finger [16] and has been shown to possess E3 ubiquitin ligase activity [17],[18],[19],[20],[21],[22]. Little is known about the B-box domain, which is unique to TRIM proteins. The coiled-coil domain is believed to promote protein oligomerization [23]. Several TRIM proteins also contain a C-terminal SPRY domain that is proposed to be involved in protein-protein interactions and RNA binding [24],[25].
Many TRIM family members are inducible by IFNs, providing candidate mediators of IFN inhibition. In an effort to identify TRIM proteins mediating anti-retroviral activity, we carried out transcriptional profiling of IFN-treated human osteosarcoma (HOS) cells and found that TRIM22 was the most up-regulated TRIM queried on the microarray used. HOS-CD4/CXCR4 cells were studied because they support robust HIV replication and show a strong antiviral response to IFNβ. Human TRIM22 (also known as Staf-50) was previously identified from a search for IFN-regulated genes in Daudi cells. Over-expression of TRIM22 was reported to repress transcription from the HIV-1 LTR in a plasmid reporter system in COS-1 cells [26]. In a later study by another group, over-expression of TRIM22 was reported to inhibit HIV-1 replication in human macrophages and 293T cells and postulated to act by repressing transcription from the HIV-1 LTR [27]. TRIM22 has also been implicated in normal haematopoietic differentiation [28],[29] and in diseases such as systemic lupus erythematosus [30] and Wilms tumor [31]. The expression of TRIM22 is altered in response to a variety of stimuli in addition to IFN including T-cell co-stimulation by CD2 and CD28 [32], viral antigens or infection [27],[33],[34],[35] and inflammatory cytokines [28].
Here we investigate the mechanism by which TRIM22 inhibits HIV infection. We mapped the primary block to late stages of assembly and release in the viral life cycle using over-expression studies. Short hairpin RNA against TRIM22 substantially reversed the inhibition of late steps induced by IFN treatment, implicating TRIM22 as a functional effector when expressed at biological levels. TRIM22 blocked the release of HIV-1 Gag-only particles, and the release of HIV particles was dependent on the RING domain of TRIM22. These findings clarify an IFN-inducible effector arm that disrupts the late stages of HIV infection.
Human osteosarcoma cells modified to express CD4 and CXCR4 (HOS-CD4/CXCR4) support robust HIV-1 replication, which is potently restricted by pre-treatment with IFNβ. We wished to identify restriction factors induced by IFNβ, and so carried out transcriptional profiling of HOS-CD4/CXCR4 cells before and after IFNβ treatment. As expected, a large number of genes were induced (52 total with a false discovery rate of 7%; Table S1), including the well-known antiviral genes OAS1 (∼235-fold), MX1/2 (∼24-fold), ISG20 (∼37-fold), ISGF3G (∼12-fold), IRF7 (∼13-fold) and TRIM19zeta (∼18-fold).
We focused our study on TRIM22, which was induced at least 86-fold after IFNβ-treatment. Other TRIM genes that were notably up-regulated included TRIM19 variants (2.8 to 18-fold), TRIM21 (2.6-fold), TRIM34 variant 1 (2.4-fold) and TRIM5delta (2.4-fold), consistent with previous studies [26],[36],[37],[38],[39]. Gene chip data on TRIM22 induction was confirmed using Northern blot analysis (Figure 1A). These results support previous studies of other cell types, although we observed a higher level of IFNβ-induced TRIM22 expression than previously reported [4],[27].
To analyze TRIM22 inhibition of HIV-1 replication, we generated HOS-CD4/CXCR4 cells that stably express hemagglutinin (HA)-tagged TRIM22 protein (Figure 1B). There was no significant difference between the doubling times of the empty vector control cells (24.4+/−2.4 hours) and the cells expressing TRIM22 (25.5+/−1.8 hours) (P = 0.7, Mann-Whitney Test), indicating TRIM22 expression was not detectably toxic. Cells stably expressing TRIM22 were then infected with increasing amounts of replication-competent HIV-1. Virus released into the supernatant after several rounds of replication was detected using p24CA antibodies in a Western blot (Figure 1C). Cells expressing TRIM22 released substantially less virus than the control cells. Similar results were obtained using another HIV-1 clone LAI (data not shown). These findings parallel a previous report [27].
We next analyzed the step in the HIV-life cycle affected by over-expression of TRIM22. To test for inhibition of early steps (entry through integration), we infected cells with HIV-1 pseudotyped with the VSV-G envelope and transducing GFP. We infected two isolates of the HOS-CD4/CXCR4 cell lines expressing TRIM22 and found no reduction in the fraction of cells positive for GFP (Figure 1D), indicating that TRIM22 did not strongly affect HIV/VSV-G early steps.
To test whether TRIM22 interferes with late events in the HIV lifecycle, we co-transfected plasmids encoding TRIM22 and a replication-competent HIV provirus. Forty-eight hours after transfection, Western blots of cell lysates and extracellular supernatants were probed with an antibody against the p24 capsid component of the Gag polyprotein (Figure 2). Co-expression of TRIM22 blocked the release of virions from HOS-CD4/CXCR4 cells, despite the production of substantial amounts of intracellular Gag (Figure 2). Gag also failed to accumulate in the extracellular supernatant in transfected U2OS, 143B and HeLa cells.
The nature of the blocks differed among the cell types studied. After correcting with an internal transfection control (peGFP), Gag proteins accumulated to a reduced extent in U2OS and 143B cells in the presence of TRIM22, while Gag accumulated inside cells to near wild-type levels in HOS-CD4/CXCR4 and HeLa cells but were not released (Figure 2 and data not shown). In all cell types, however, accumulation of extracellular Gag was abolished by co-expression with TRIM22.
The above data and results in [27] indicated that ectopic expression of TRIM22 can interfere with HIV replication, but from over-expression data alone it is uncertain whether expression at physiological levels would result in antiviral activity. For this reason, we tested whether depleting TRIM22 using short hairpin ribonucleic acid (shRNA) in the context of an interferon response allowed increased production of extracellular HIV particles. HOS-CD4/CXCR4 cells were transiently transfected for 24 hours with either pLKO.1/TRIM22shRNA#1 or the empty vector control (pLKO.1). In control cells, TRIM22 RNA accumulation increased in the presence of increasing concentrations of IFNβ. In the cells expressing the TRIM22 shRNA, TRIM22 RNA levels were significantly reduced (Figure 3A and B). As a control, expression of the related TRIM34 gene was tested and found to be unaffected by the TRIM22 shRNA (Figure 3A).
The knockdown cells and the control cells were then tested for their ability to release HIV-1 virus after transfection with a plasmid encoding replication-competent HIV-1 (pR9). Accumulation of HIV Gag in cells and supernatants was monitored by Western blot using p24CA antibodies (Figure 3C). In the absence of IFNβ, the control cells released virus into the supernatant. In the presence of increasing concentrations of IFNβ, cells displayed a progressive reduction in the amount of virus released, reaching undetectable levels at 1000 units/ml of IFNβ. There was little effect of IFNβ on intracellular Gag expression levels, indicating that IFNβ acts at the level of virus assembly and/or release in HOS-CD4/CXCR4.
Cells knocked down for TRIM22, when treated with IFN, released substantially more virus into the cell supernatant than control IFN-treated cells (Figure 3C, 500 and 1000 units/ml of IFNβ). Figure 3D compares the fold decrease in particle release after treatment with 1000 units/ml of IFNβ in cells transfected with the empty vector control, an irrelevant target (eGFP) control (pLKO.1/eGFPshRNA), a scrambled shRNA control ( pLKO.1/scramshRNA) and two different TRIM22 shRNAs targeting a different region of the TRIM22 3′ UTR. Gag release was inhibited by IFN more strongly in cells transfected with each of the control shRNAs than in cells transfected with TRIM22 shRNA. Over 14 control experiments, the fold-inhibition values for 1000 units/ml IFNβ ranged from 3.0-fold to 7.4-fold. Over 8 experiments with two different TRIM22 shRNAs and 1000 u IFNβ, inhibition ranged from 1.2-fold to 1.6-fold. The difference between the means of the fold-inhibition values of TRIM22 shRNA (1.4-fold) and the pooled controls (4.5-fold) was highly significant (P = 0.0002 Mann-Whitney) (Figure 3E). These data indicate that TRIM22 is an active antiviral effector in the context of the normal response to IFNβ.
The HIV Gag polyprotein is transported to the plasma membrane after translation where it assembles into particles and buds from the cell surface. Gag protein alone is able to bud from cells when expressed in the absence of the other HIV-1 proteins (reviewed in [40]). We asked whether TRIM22 acted on the Gag polyprotein by testing inhibition of budding by Gag-only virus-like particles.
We co-transfected a codon-optimized HIV Gag expression plasmid in the presence or absence of TRIM22 and measured the accumulation of Gag proteins in the cells and in culture supernatants by Western blot (Figure 4A). In the absence of TRIM22, Gag protein was detected in the cell lysate and efficiently released into the supernatant. In the presence of TRIM22, Gag was readily detected in the cell lysate, but not the supernatant. This effect parallels the effect of IFNβ on Gag-only particle release in HOS-CD4/CXCR4 cells (Figure 4B).
To assess the specificity of TRIM22, we assayed effects on murine leukemia virus (MLV) and equine infectious anemia virus (EIAV) Gag. Co-transfection of TRIM22 with either MLV Gag-Pol or EIAV Gag-Pol failed to restrict the release of Gag-containing particles into the supernatant, whereas TRIM22 restricted the release of HIV-1 Gag (Figure 4C). Together these data indicate that TRIM22 acts selectively on HIV-1 Gag.
We next investigated whether TRIM22-mediated inhibition was associated with altered Gag trafficking by monitoring the subcellular localization of HIV Gag fused to GFP (pGag-GFP) [41],[42],[43],[44]. We transfected pGag-GFP in the presence or absence of pTRIM22, then 3 hours later treated the cells with cycloheximide for 3 hours. Forty-eight hours after release of the cycloheximide block, the localization of the Gag-GFP protein was visualized by fluorescence microscopy by taking optical slices through the center of cells (Figure 5A). In the control cells, 65% of GFP-positive cells had punctate fluorescence at or near the plasma membrane and 35% of the cells showed diffuse cytoplasmic fluorescence. In contrast, in cells expressing TRIM22, 12% of the cells had punctate fluorescence at or near the plasma membrane and 88% of the cells yielded diffuse cytoplasmic localization without visible puncta (Figure 5B). The observed difference in proportions was highly significant (P<0.0001, Chi-square test).
To test whether IFNβ altered Gag localization, as with TRIM22 expression alone, we compared the effects of pre-treating cells with 1000 units/ml of IFNβ followed by transfection with GagGFP and microscopy. In the absence of IFNβ, 64% of GFP-positive cells had punctate fluorescence at or near the plasma membrane and 36% of the cells showed diffuse cytoplasmic fluorescence. In contrast, 22% of the IFNβ-treated cells had punctate fluorescence at or near the plasma membrane and 78% of the cells yielded diffuse cytoplasmic localization without visible puncta (Figures 5A and 5B). The difference in the number of cells exhibiting punctate localization at the cell surface before and after IFNβ-treatment was highly significant (P<0.0001, Chi-Square) and paralleled the result with the expression of TRIM22 alone.
To determine whether the diminished accumulation of Gag at the plasma membrane is due to accelerated degradation of Gag, we performed pulse-chase analysis on intracellular Gag levels in HOS-CD4/CXCR4 cells (Figure 5C). The half life of Gag was 4.4 hours in cells transfected with the empty vector control (pcDNA) and 4.5 hours when TRIM22 was co-transfected. In addition, Western analysis also yielded similar levels of intracellular Gag protein in the presence and absence of TRIM22 (data not shown). These data indicate that inhibition of particle release by TRIM22 is associated with diminished accumulation of Gag at the plasma membrane and is not likely the result of increased turnover of Gag in HOS-CD4/CXCR4 cells.
The myristoylation of Gag is an important requirement for targeting Gag to the plasma membrane and for virus assembly and release [45],[46],[47],[48],[49]. We asked whether the TRIM22-mediated disruption of Gag trafficking is a result of a defect in Gag myristoylation. We co-expressed Gag with or without TRIM22 in the presence of 3H-myristic acid and immunoprecipitated Gag from the cellular extract with anti-p24CA. Gag myristoylation was readily detected both in the absence and presence of TRIM22, indicating that TRIM22 does not block the myristoylation of Gag (Figure 5D). A fraction of the immunoprecipitated protein was subjected to Western blot analysis using anti-p24CA, confirming the presence of comparable amounts of Gag protein in both of the samples.
The modification of proteins by E3 ligases is known to be associated with altered endocytic trafficking (reviewed in [50],[51]). Since TRIM22 alters Gag trafficking and the RING domain of TRIMs have homology with E3 ligases, we tested whether the E3 ligase active site, present in the RING domain, is required for the antiviral effects of TRIM22. We mutated two conserved cysteine residues (C15A/C18A) in the RING domain that have been shown to inactivate the ubiquitin ligase activity in other TRIMs [17],[52],[53].
We found that cells co-transfected with pR9 and either the empty vector control or the pTRIM22 C15A/C18A mutant did not block the release of virus, though wildtype TRIM22 blocked quite strongly (Figure 6). Gag accumulated inside cells to near wild-type levels (Figure 6, second panel) and the expression levels of wildtype TRIM22 and the C15A/C18A mutant were similar (Figure 6, second panel). These data implicate the catalytic cysteine residues at position 15 and 18 of the RING domain as important for the observed antiviral effects of TRIM22.
We next asked whether we could detect an interaction between TRIM22 and Pr55Gag. We immunoprecipitated a Gag-GFP fusion protein expressed in the presence of FLAG-tagged TRIM22 with anti-GFP antibodies and then performed a Western blot with anti-Flag antibodies and detected co-precipitation of FLAG-TRIM22 (Figure 7). Reverse immunoprecipitation using anti-FLAG pulled down Gag-GFP only when FLAG-TRIM22 was co-expressed. Treatment of the samples with RNaseA prior to immunoprecipitation did not interfere with the association of TRIM22 with Gag, indicating that an RNA bridge did not mediate the interaction. TRIM22 did not co-immunoprecipitate with either MLV or EIAV Gag (data not shown). Thus TRIM22 binds specifically to HIV Gag.
Here we present data that TRIM22 is an interferon inducible effector that is responsible at least in part for the late IFNβ block to HIV-1 replication. We observed, as have others, that over-expression of TRIM22 in cells obstructed HIV replication. We show here that knockdown of TRIM22 in HOS-CD4/CXCR4 cells in the context of a natural IFN response abrogated the late block to HIV replication, implicating TRIM22 as an antiviral effector. TRIM22 expression was able to block release of HIV-1 Gag-only virus-like particles, but not MLV or EIAV Gag particles, indicating that TRIM22 acts specifically on the HIV Gag protein. In binding studies, TRIM22 associated with HIV-1 Gag but not MLV or EIAV Gag. TRIM22 expression altered Gag localization–cells containing Gag-GFP fusions showed punctate structures at the plasma membrane, potential sites of virus assembly and budding [41],[54],[55]. In contrast, Gag-GFP expressed in the presence of TRIM22 alone or IFNβ showed only diffuse staining. TRIM22 with substitutions in the E3 catalytic residues were unable to restrict HIV-1 release, indicating that inhibition was dependent on the active site. Taken together, these data implicate TRIM22 as an effector arm of the IFN-mediated antiviral state that selectively disrupts Gag sorting and release.
Complications in studying mechanisms of the IFN response arise from i) the pronounced cell type specificity of IFN action, ii) differential effects of different doses, and iii) differences among the effects of interferon types. Previous studies have reported that Type I IFNs can block HIV gene expression, Gag sorting, and Gag release [9],[11],[26],[56],[57],[58],[59],[60],[61],[62]. Here we show, using HOS-CD4/CXCR4 cells, that TRIM22 acts on HIV Gag, allowing efficient Gag expression but blocking assembly and/or release. Similar results were seen in HeLa cells. However, in U2OS and 143B cells, TRIM22 expression resulted in reduced accumulation of Gag in cells, indicating that TRIM22 inhibited either Gag synthesis or increased Gag turnover. A previous study, analyzing TRIM22 in COS cells, indicated that TRIM22 inhibited LTR-driven transcription. A similar mechanism may operate in U2OS cells and 143B cells. Recently, rhesus TRIM5α (but not its human ortholog) was shown to cause the rapid degradation of Gag in 293T cells [63]. We found no evidence that TRIM22 drastically altered the stability of Gag protein levels in HOS-CD4/CXCR4 or HeLa cells, however it is possible that altered Gag sorting results in faster degradation in U2OS and 143B cells than seen in HOS-CD4/CXCR4 and HeLa cells.
Against this background of cell type specific effects, comparisons of our results to previous reports are quite tentative. In one study, addition of 500 units/ml of IFNα to 293T cells was reported to inhibit HIV particle release about two-fold, and an siRNA against the interferon-induced ubiquitin-like molecule ISG15 blocked IFN inhibition [64],[65],[66]. In HOS-CD4/CXCR4 cells treated with IFNβ at 1000 units/ml, the inhibition of particle release was stronger (3.0-fold to 7.4-fold), but the shRNA against TRIM22 did not completely reverse IFN inhibition. One possibility is that TRIM22 and ISG15 both act in the same pathway. For example, TRIM22 might act as an E3 ligase for ISG15. Further experiments will be needed to assess the relationship of inhibition by TRIM22 to that of ISG15.
Another recent report suggested that IFNα treatment caused HIV-1 virions to become tethered to cell surfaces, resulting in eventual re-uptake by endocytosis [67]. The viral Vpu protein was reported to overcome this block in some cell types. It is possible that the mis-sorting of Gag observed here in HOS-CD4/CXCR4 cells results from virions trafficking to the membrane followed by re-uptake, though in HOS-CD4/CXCR4 cells there was no evidence for Gag accumulating at membranes at early times as might be expected from the re-uptake model. Also, in the HOS-CD4/CXCR4 cell model, expression of Gag in the presence or absence of Vpu did not noticeably affect particle accumulation in the supernatant. It will be useful to identify the cell type or experimental differences responsible for these divergent observations.
How human TRIM22 alters intracellular Gag trafficking in HOS-CD4/CXCR4 cells is unknown but a fascinating problem for further study. One candidate model would be that TRIM22 acts by interfering with Gag binding to membranes via the linked myristate residue. Membrane binding of Gag has been reported to be regulated by a multimerization-dependent myristoyl switch [68],[69],[70],[71]. We found that Gag was myristoylated normally in the presence of TRIM22 (Figure 5C) but TRIM22 might affect Gag multimerization and resulting exposure of the myristate residue. We also found that that the E3 ligase activity of TRIM22 is important for restricting viral particle release, which implies that TRIM22 probably disrupts Gag trafficking by transferring ubiquitin (or an ubiquitin-related moiety) to one or more targeted proteins. Although TRIM22 interacts with Gag, we have yet to detect a convincing effect of TRIM22 on the levels of Gag ubiquitylation or a size-altering modification of Gag (data not shown), raising the possibility that a cellular protein involved in Gag trafficking is the target. Thus studies of TRIM22 link E3 ligases, IFN, and HIV late replication steps, providing a new experimental route to investigating these issues.
293T, HOS-CD4/CXCR4, HeLa, 143B, and U2OS were maintained in standard growth media (Dulbecco's Modified Eagle's Medium (DMEM), supplemented with 10% heat-inactivated Fetal Bovine Serum (FBS), 100 U/ml Penicillin and 100 µg/ml Streptomycin) at 37°C with 5% CO2. HOS-CD4/CXCR4 cells stably expressing TRIM22 was generated by transfecting the cells with pTRIM22 followed by selection with 2 mg/ml of Genetecin. pTRIM22 was generated by cloning the TRIM22 coding region (GenBank Accession X82200) into pcDNA3.1HA using MfeI and XhoI restriction sites to generate pTRIM22. pFLAG/TRIM22 was generated by cloning the same TRIM22 coding region into p3xFLAG-CMV-10 (Sigma) using HindIII and XbaI restriction sites. pT22-C15A/C18A was generated using the QuikChange Site Directed Mutagenesis kit (Stratagene) according to manufacturer's directions. pTRIM22 was used as the template and the oligos used were 5′ gag aag gag gtg acc gcc ccc atc gcc ctg gag ctc c 3′ (forward) and 5′ gga gct cca ggg cga tgg ggg cgg tca cct cct tct c 3′ (reverse). pGag/GFP was a generous gift from M. Resh (Sloan-Kettering Institute). The plasmids pLKO.1 (Cat. #RHS4078), pLKO.1/eGFPshRNA (Cat. #RHS4459), pLKO.1/TRIM22shRNA#1 (Cat. #RHS3979-9574742; GenBank Accession NM_006074, sequence: 5′ cgg agc act cat cta caa gtt ctc gag aac ttg tag atg agt gct ccg 3′) and pLKO.1/TRIM22shRNA#2 (Cat. #RHS3979-9574744; GenBank Accession NM_006074, sequence: 5′ gtc acc aaa cat tcc gca taa ctc gag tta tgc gga atg ttt ggt gac 3′) were obtained from Open Biosystems. pLKO.1/scrambledshRNA (Addgene plasmid 1864; sequence 5′ cctaa ggtta agtcg ccctc gctct agcga gggcg actta acctt agg 3′) was obtained from [72] through Addgene. pEIAV (“pEV53B”) was obtained from [73]. pMLV (“pCgp”) was obtained from [74]. The plasmid encoding codon-optimized Gag (p96ZM651gag-opt) (from Drs. Yingying Li, Feng Gao and Beatrice H. Hahn [75]) and the HIV-1 p24 monoclonal antibody (183-H12-5C) (from Dr. Bruce Chesebro and Kathy Wehrly [76],[77],[78]) were obtained through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH. Poly/MonoUbiquitin (FK2, #PW8810) and PolyUbiquitin (FK1, #PW8805) antibodies were obtained from Biomol International. Unless otherwise stated, all plasmid transfections were performed using Lipofectamine 2000 (Invitrogen). Co-transfections were performed at a 1∶10 ratio (pR9 or pGag or pGagGFP: pcDNA or pTRIM22 or pT22-C15A/C18A respectively).
RNA was harvested from HOS-CD4/CXCR4 cells in log-phase growth 24 hours after human interferon beta (PBL Biomedical Laboratories Cat. #11420-1) treatment using a Qiagen RNeasy Kit. Labeling of RNA was done using standard techniques as described by Affymetrix (Santa Clara, CA). Two independent RNA samples were hybridized with Affymetrix HG-U133A chips in duplicate. Significance Analysis of Microarray software using a cut-off difference of 2-fold and a false discovery rate of 7% was used to analyze changes in gene activity after human IFNβ treatment via the permuted t test method [79]. Microarray data is available upon request.
HIV-1/VSV-G vector particles were generated by calcium phosphate-mediated transfection of 293T cells with three plasmids: p156RRLsinPPTCMVGFPWPRE (encoding the HIV-1 vector segment) [80], pCMVdeltaR9 (the packaging construct), and pMD.G (encoding the VSV-G envelope) [81]. Replication-competent virus (R9 and LAI clones) was generated by calcium phosphate-mediated transfection of 293T cells with pR9 or pLAI plasmids. Virus was quantitated for reverse transcriptase activity using standard protocols [82].
Prior to infection, virus aliquots were digested with DNaseI (0.2 units/µl) for 1 hour at 37°C. Cells were infected with varying amounts of virus for 3 hours at 37°C in a minimal volume of media, after which the media containing virus was replaced with fresh media and the cells were then incubated for the desired time.
Virus released into the supernatant was pelleted and lysed as described in [82]. Briefly, 1ml viral supernatants were clarified by centrifugation at 350 g for 10 mins at 4°C and passed through a 0.45 µm filter. Virus was precipitated overnight on wet ice at 4°C using 0.5× volumes of PEG solution (30% PEG 8000, 0.4 M NaCl). Precipitated virus was pelleted by centrifugation at 800 g for 45 minutes at 4°C. Gag particles were prepared as previously described [42],[44]. Briefly, 800 µl of supernatant containing Gag particles were pelleted by centrifugation at 20,000 g for 90 mins at 4°C over a 500 µl cushion of 20% sucrose.
PEG-precipitated virus and sucrose-purified Gag particle pellets were lysed with 75 µl and 30 µl of lysis buffer respectively and 2 µl–5 µl of the lysed pellets was used for Western blotting. Lysis buffer consisted of a fresh mix of Buffer A and Buffer B (2∶1 ratio respectively). Buffer A contained 12.5 ml of 1 M Tris (pH 7.8), 1.25 ml of 0.1 M EDTA (pH 8.0), 1.25 ml of 10% Triton X-100, 250 ml of Glycerol, 0.77 g of DTT and 3.72 g of KCl with the volume brought up to 500 ml with ddH2O. Buffer B contained 45 ml of 10% Triton X-100 and 1.63 g KCl with the volume brought up to 500 ml with ddH2O.
For analysis of cell lysates, cells were centrifuged at 350 g for 5 mins, washed twice with phosphate-buffered saline (PBS) and lysed with 50 µl of RIPA buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1× Complete Protease Inhibitor (Roche), 1% Triton X-100, 0.1% SDS). Two micrograms of the lysed cell pellets were used for Western blotting.
Cells were lysed with cold non-denaturing lysis buffer for 20 mins (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 10 mM EDTA, 0.5% Igepal CA-630 (Sigma), 1× Complete EDTA-free proteinase inhibitor (Roche)) and pre-cleared by centrifugation at 16,000×g for 15 mins. Two micrograms of primary antibody was diluted in 500 µl of cold PBS with 0.01% Triton X-100 and mixed with 15 µl of a 50% protein A-sepharose at 4°C for 1 hour with gentle rocking. Beads were washed 3 times with PBS and once with lysis buffer. 5 µl of 10% BSA was added to the beads and mixed with 600 µg–1 mg of pre-cleared cell lysate in a 500 µl–1 ml volume at 4°C for 2 hours with gentle rocking. Beads were washed 3 times with cold lysis buffer and once with PBS.
Samples for Western blotting were mixed with loading buffer and separated on a 10% SDS-PAGE gel. Protein was transferred to PVDF membrane by wet or semi-dry transfer. Western blotting was carried out by blocking the membrane for 1 hour in 5% skim milk followed by a ∼16 hour incubation with 1∶1000 dilution of primary antibody. Detection was carried out using HRP-labeled secondary antibody (1∶5000 for 30 mins) and the Amersham ECL Plus Western Blotting Detection System. Membranes were exposed to Super RX film (Fujifilm) or scanned using a Storm 860 Phosphorimager (Molecular Dynamics). Densitometric analysis in the linear range of the scanned image was performed using IQMac (v1.2) software. Western blots requiring quantitation were verified using an IRDye-labeled secondary antibody (1∶20,000 for 30 mins) and the Odyssey Infrared Imaging System (LI-COR Biosciences) according to manufacturer's directions.
Nearly confluent HOS-CD4/CXCR4 cells in 12-well plates (1 ml volume) were transfected with 1.6 µg of pLKO.1, pLKO.1/eGFPshRNA or pLKO.1/TRIM22shRNA. Cells were cultured for 16 hours, after which they were washed and treated with increasing concentrations of human IFNβ (PBL Biomedical Laboratories Cat. #11420-1) for 20 hours. The cells were then washed and transfected with 0.16 µg of pR9 (encoding full-length HIV-1). After 16 hours, virus released into the supernatant (800 µl) or contained within 2 µg of the cell lysate were purified and analyzed by Western blot as described above.
Decreased expression of TRIM22 mRNA was verified by Northern blot. Briefly, total RNA was isolated using the RNeasy RNA isolation kit (Qiagen). Twenty micrograms of RNA was separated on a 1.2% formaldehyde agarose gel and transferred to a nitrocellulose membrane. A region in the 3′UTR of TRIM22 mRNA from nucleotides 1523 to 1983 (GenBank Accession #BC035582) was PCR amplified from pCMV-SPORT6/TRIM22 (Open Biosystems Cat. # MHS1010-7508483) and random-labeled with αP32 for use as a probe. Differences in loading amounts were assessed using a β-actin mRNA probe end-labeled with γP32. Membranes were analyzed using a Storm 860 Phosphorimager.
HOS-CD4/CXCR4 cells cultured in 12-well plates on 18mm coverslips were co-transfected with pTRIM22 and pGag/GFP (10∶1 ratio respectively) for 3 hours, washed and then treated with 100 µg/ml of cycloheximide (Sigma) for 3 hours as described in [44]. Cells were incubated in fresh media to release the cycloheximide block and incubated for 48 hours. The coverslips containing the cells were then washed twice with PBS, fixed for 10 mins in PBS containing 5% formaldehyde and 2% sucrose and then washed twice more with PBS. Coverslips were mounted onto glass slides with ∼15 µl of Vectashield (Vector Laboratories) containing 0.5 mg/ml DAPI stain and then sealed with nail polish. Slides were examined using 40× oil immersion using a Zeiss fluorescence microscope fitted with an Apotome.
HOS cells ∼90% confluent in 6-well plates were co-transfected with 0.4 µg pGag and 4 µg of pcDNA3.1 or pTRIM22 for 24 hours. Cells were then labeled with 100 µCi/ml of 3H-myristic acid (Perkin Elmer) for 16 hours. Cells were washed 4 times with PBS and subjected to immunoprecipitation using p24CA antibodies as described above. Samples were resolved on a 10% SDS-PAGE gel, impregnated with EN3HANCE scintillant (Perkin Elmer), dried and exposed to a phosphorimager screen for 3 days at room temperature. The screen was analyzed using a Storm 860 Phosphorimager and IQMac (v1.2) software.
The pulse-chase stability assay for Gag was carried out as previously described with some modifications [13]. Nearly confluent 10 cm dishes containing HOS-CD4/CXCR4 cells were transfected with 30 µg of pcDNA3.1 or pTRIM22 and 3 µg of pGag-opt. After 36 hours, cells were washed with PBS and then incubated in methionine-, cystine- and cysteine-free DMEM (MP Biomedicals) for 1 hour. Cells were labeled for 45 mins with 50 µCi/ml of [35S] EasyTag Express Protein Labeling Mix (Perkin Elmer), washed and incubated in standard DMEM growth media with 10% heat-inactivated FBS for 0, 1, 2 or 4 hours. Cells were processed for immunoprecipitation with monoclonal p24 antibodies as described earlier. Precipitated proteins were separated on a 10% SDS-PAGE gel, impregnated with EN3HANCE scintillant (Perkin Elmer), dried and analyzed by phosphorimager. |
10.1371/journal.ppat.1004192 | Protective Efficacy of Passive Immunization with Monoclonal Antibodies in Animal Models of H5N1 Highly Pathogenic Avian Influenza Virus Infection | Highly pathogenic avian influenza (HPAI) viruses of the H5N1 subtype often cause severe pneumonia and multiple organ failure in humans, with reported case fatality rates of more than 60%. To develop a clinical antibody therapy, we generated a human-mouse chimeric monoclonal antibody (MAb) ch61 that showed strong neutralizing activity against H5N1 HPAI viruses isolated from humans and evaluated its protective potential in mouse and nonhuman primate models of H5N1 HPAI virus infections. Passive immunization with MAb ch61 one day before or after challenge with a lethal dose of the virus completely protected mice, and partial protection was achieved when mice were treated 3 days after the challenge. In a cynomolgus macaque model, reduced viral loads and partial protection against lethal infection were observed in macaques treated with MAb ch61 intravenously one and three days after challenge. Protective effects were also noted in macaques under immunosuppression. Though mutant viruses escaping from neutralization by MAb ch61 were recovered from macaques treated with this MAb alone, combined treatment with MAb ch61 and peramivir reduced the emergence of escape mutants. Our results indicate that antibody therapy might be beneficial in reducing viral loads and delaying disease progression during H5N1 HPAI virus infection in clinical cases and combined treatment with other antiviral compounds should improve the protective effects of antibody therapy against H5N1 HPAI virus infection.
| The H5N1 highly pathogenic avian influenza virus has been circulating in poultry in Asia, the Middle East, and Africa since its first appearance in southern China in 1996. This virus occasionally infects humans with a high case mortality rate and poses a significant pandemic threat. Since neutralizing antibodies generally play a major role in protective immunity against influenza viruses, antibody therapy is a potential option for preventing highly lethal infection with the H5N1 virus in humans. Here we evaluated the protective potential of a human-mouse chimeric monoclonal antibody with strong neutralizing activity against H5N1 viruses in mouse and nonhuman primate models of lethal H5N1 virus infection. The therapeutic use of the neutralizing antibody resulted in reduced viral loads and improved survival in animals infected with highly pathogenic H5N1 viruses. It was noted that the protective effects were more prominent in immunosuppressed macaques, which might provide a model of protection against severe clinical disease in immunocompromised patients. In addition, combination therapy together with an antiviral drug reduced the selection of escape mutants. Collectively, this study suggests that antibody therapy may have beneficial effects in clinical cases of H5N1 HPAI virus infection in humans.
| Influenza A viruses are divided into subtypes based on the antigenicity of two envelope glycoproteins, hemagglutinin (HA) and neuraminidase (NA). To date, H1-H16 HA and N1-N9 NA subtypes have been found in wild aquatic birds, the natural reservoir of influenza viruses [1]–[3]. Of these HA subtypes, only some avian influenza viruses of the H5 and H7 subtypes are known to become highly pathogenic avian influenza (HPAI) viruses under natural conditions. While HPAI viruses cause an acute systemic disease in poultry with a mortality rate that often approaches 100%, avian to human transmission of HPAI viruses is limited and HPAI viruses had never been reported to cause lethal infection in humans until the first emergence of an H5N1 HPAI virus in southern China in 1996.
The H5N1 HPAI virus has been circulating in poultry for more than a decade since its reemergence in southern China in 2003, and has caused unprecedented outbreaks in wild birds and poultry in Asia, the Middle East, and Africa [4]–[10]. The H5N1 HPAI virus occasionally infects humans with a high case mortality rate and poses a significant pandemic threat [11], [12], [13]. Since 2003, 641 laboratory-confirmed human cases of H5N1 HPAI virus infection have been reported from 15 countries, with 380 fatal cases (as of October 8, 2013) [12]. In fact, prior to the emergence of the swine-origin H1N1 pandemic virus in 2009, the impact on animal and public health of the Asian origin H5N1 HPAI virus led to the prediction that a virus of the H5 subtype might cause the next pandemic, since this HA subtype is distinct from those of viruses circulating in the human population (i.e., subtypes H1 and H3) [13].
In recent years, passive immunization with human or humanized monoclonal antibodies (MAbs) specific to viral proteins has been tested in animal models and clinical trials, providing evidence of the effectiveness of MAbs for prophylaxis or treatment of infectious diseases [14]. Indeed, a humanized MAb specific to Respiratory syncytial virus F protein is already approved by the US Food and Drug Administration and used in clinical cases. Importantly, particular attention has been paid to antibody therapy against highly lethal diseases such as rabies [15]–[17], severe acute respiratory syndrome [18], [19], Hendra [20], Nipah [21], and Ebola viruses [22]–[25].
It is known that HA, which is responsible for both receptor binding and fusion of the virus envelope with the host cell membrane, is the primary target of neutralizing antibodies against influenza viruses. Since antibodies generally play a major role in protective immunity against influenza virus infection [26], antibody therapy might be a potential option for preventing lethal infection of humans by the H5N1 HPAI virus. In this study, we genetically modified a mouse MAb (m61) neutralizing the infectivity of H5N1 HPAI viruses to create human-mouse chimeric MAb (ch61), aiming at clinical application, and evaluated its protective potential in mouse and nonhuman primate models of H5N1 HPAI virus infection.
HPAI virus strains A/Hong Kong/483/1997 (H5N1) (HK483), A/Viet Nam/1194/2004 (H5N1) (VN1194), and A/Vietnam/UT3040/2004 (H5N1) (VN3040) from the repository of our laboratory, were propagated in Madin-Darby canine kidney (MDCK) cells from the repository of our laboratory and stored at −80°C until use. HK483, VN1194, and VN3040 belong to clades 0, 1, and 1 in a phylogenetic tree, respectively [27]. MDCK cells were grown in Eagle's minimal essential medium supplemented with 10% calf serum. All experiments using infectious viruses were performed in the biosafety level 3 facilities of the Hokkaido University Research Center for Zoonosis Control and Research Center for Animal Life Science, Shiga University of Medical Science.
Mouse MAb 61-2-1 (m61), was generated according to standard procedures. Briefly, six-week-old female BALB/c mice (Japan SLC) were immunized intramuscularly two times with 100 µg of formalin-inactivated purified virions and boosted intraperitoneally [23]. Spleen cells harvested 3 days after boosting were fused to P3U1 myeloma cells according to standard procedures. Hybridomas were screened for secretion of HA-specific MAbs by enzyme-linked immunosorbent assay (ELISA), and cloned by limiting dilution. The resulting cell clones were inoculated into BALB/c mice intraperitoneally to produce ascites. Antibodies were purified from ascites using the Affi-Gel Protein A MAPS II Kit (Bio-Rad). Mouse MAbs ZGP133 and ZGP226 used as control antibodies were generated as described previously [23].
Human-mouse chimeric MAb ch61 was generated and purified from culture supernatants as described previously [23]. Briefly, total RNA was extracted from mouse hybridoma cells producing MAb m61, and the variable heavy- and light-chain regions were amplified by RT-PCR with primers designed for the antibodies. The PCR products were cloned into an expression vector. Stable cell lines expressing recombinant MAb ch61 were obtained by transfection of CHO DG44 cells (Invitrogen, Carlsbad, CA). Chimeric MAbs (ch133 and ch226) specific for the Ebola virus glycoprotein were generated as control MAbs using the same methodology [23]. These human-mouse chimeric MAbs were purified from culture supernatants using rProtein A Sepharose Fast Flow (GE Healthcare) and EndoTrap red (Profos AG). MAb purity (>98%) and endotoxin levels (<1.0 EU/ml) were confirmed by performing SDS-PAGE and with an Endospecy ES-50M kit (Seikagaku Corporation), respectively.
Serially diluted antibodies (100 µl) were mixed with 200 plaque forming units (PFU) of H5N1 viruses for 1 h at room temperature, and inoculated onto MDCK cells. After 1 h, the inoculum was removed and the cells were overlaid with 1% Bacto-Agar (BD) in Eagle's minimal essential medium (MEM). Two days later, the number of plaques was counted and the percentage of plaque reduction was calculated.
Escape mutants were selected by culturing VN1194 in MDCK cells in the presence of MAb m61. Serial dilutions of VN1194 were mixed with purified MAb m61 (final concentration of 10 µg/ml), incubated for 1 h, and the mixtures were inoculated into confluent MDCK cells in 6-well tissue culture plates. After 1 h adsorption, the cells were overlaid with MEM containing 1% agar and MAb m61 ascites (final dilution of 1∶1000), and then incubated for 2 days at 35°C. Eight escape mutants were purified from single isolated plaques, and propagated in MDCK cells with serum-free MEM containing trypsin. The nucleotide sequences of the HA genes of the parent strains and the escape mutants were determined and the deduced amino acid sequences were compared among these viruses (H3 numbering).
Six-week-old female BALB/c mice were passively immunized by intraperitoneal injection with 200 µg of purified MAbs m61 or ch61 24 hours before, or 24 hours or 72 hours after intranasal challenge with 50 µl of 12.5×50% mouse lethal dose of HK483 under anesthesia with isoflurane. Control groups were administered with control antibodies (mixture of MAbs ZGP133/ZGP226 or ch133/ch226) or phosphate-buffered saline (PBS). Animals were monitored daily for weight loss and clinical signs. Five days after the challenge, mice were euthanized to obtain lung tissue samples. Lung homogenates (10% w/v) prepared in MEM were centrifuged at 3,000× g for 10 min, and then the supernatants were examined for virus infectivity. Virus titers were measured by a plaque assay using MDCK cells.
The animal experiments were conducted in strict compliance with animal husbandry and welfare regulations. Food pellets of CMK-2 (CLEA Japan) provided once a day after recovery from anesthesia and drinking water were available ad libitum. Animals were singly housed in the cages equipping bars to climb up and puzzle feeders for environmental enrichment under controlled conditions of humidity (60±5%), temperature (24±1°C), and light (12 h light/12 h dark cycle, lights on at 8:00 A.M.). Five- to seven-year-old female cynomolgus macaques (Macaca fascicularis) from the Philippines (Ina Research) were used. The cynomolgus macaques used in the present study were healthy adults. The absence of influenza A virus NP-specific antibodies in their sera was confirmed before experiments using an antigen-specific ELISA, AniGen AIV Ab ELISA (Animal Genetics), for currently circulating influenza virus. Three weeks before virus inoculation, a telemetry probe (TA10CTA-D70, Data Sciences International) was implanted in the peritoneal cavity of each macaque under ketamine/xylazine anesthesia followed by isoflurane inhalation to monitor body temperature. The macaques used in this study were free from herpes B virus, hepatitis E virus, Mycobacterium tuberculosis, Shigella spp., Salmonella spp., and Entamoeba histolytica. Individual macaques were distinguished by treatments and numbers: C: macaques injected with control MAbs, T: macaques treated with MAb ch61, IC: immunosuppressed macaques injected with control MAbs, IT: immunosuppressed macaques treated with MAb ch61, ICP: immunosuppressed macaques injected with MAbs and peramivir, ITP: immunosuppressed macaques treated with MAb ch61 and peramivir.
Macaques (2.4–3.1 kg) were inoculated (day 0) with VN3040 (total 3×106 PFU/7 ml) in their nasal cavities (0.5 ml for each nostril) and on their tonsils (0.5 ml for each tonsil) with pipettes and into the trachea (5 ml) with catheters under ketamine/xylazine anesthesia. MAb ch61 or control MAbs (a mixture of MAbs ch133 and ch226) were administered intravenously twice (20 mg/head/dose; 6.5–8.3 mg/kg) on days 1 and 3 after infection. Animals were monitored daily (approximately every 12 hours) for clinical scoring (Table S1). Serum samples were obtained on days −1, 1, 3, 5, and 7. For virus titration, cotton sticks (TE8201, Eiken Chemical) were used to collect fluid samples from the nasal cavities and tracheas under ketamine/xylazine anesthesia, and the sticks were subsequently immersed in 1 ml of PBS containing 0.1% bovine serum albumin (BSA) and antibiotics. A bronchoscope (Machida Endoscope) and cytology brushes (Olympus) were used to obtain bronchial samples. The brushes were immersed in 1 ml of PBS with BSA. Viral titers were determined by the tissue culture infectious dose (TCID50) in MDCK cells [28]. For immunosuppressive treatments of macaques, cyclophosphamide (CP) (Nacalai Tesque) and cyclosporine A (CA) (Novartis Pharma) were used [29]. CP (40 mg/kg) was administered intravenously by bolus injection on days −7, −5, −3, −1, and 0. CA (50 mg/kg) was administered orally into stomach using a catheter from day −7 to day 6. We confirmed that the treatment with CP and CA decreased the number of white blood cells in the macaques (Fig. S1). In some experiments, peramivir hydrate (30 mg/kg/dose, provided by Shionogi & Co., Ltd.) was administered intravenously by bolus injection once a day from day 1 to day 5 after infection [30]. Since patients with a severe respiratory illness might have a difficulty to intake or inhale drugs, we chose peramivir hydrate as an antiviral agent with intravenous injection. The concentrations of cytokines in sera and tissue homogenates were measured using the Milliplex MAP nonhuman primate cytokine panel and Luminex 200 (Millipore). Although the experiment was originally designed to collect samples from all animals for virology and immunology studies terminating on day 7, some animals were euthanized when their clinical scores reached 15 (a humane endpoint) and subjected to autopsy to collect tissue samples. Macaques that were unfortunately found dead during the intervals of the monitoring time points were also immediately subjected to autopsy. These animals (i.e., euthanized or dead) were counted as nonsurvivors.
After autopsy on indicated days after virus infection, lung tissue samples were fixed with 10% formalin, and embedded in paraffin. Sections were stained with hematoxylin and eosin (H & E). Influenza virus nucleoprotein (NP) antigens were stained with antisera of rabbits immunized with an NP synthetic peptide (AFTGNTEGRTSDMR at positions 428–441 of the NP sequence: GenBank accession number, ADC34563) after treatment in a pressure cooker in 0.01 M citrate-phosphate buffer. After incubation with anti-rabbit immunoglobulin antibody conjugated with horseradish peroxidase (Nichirei Bioscience Inc.), NP was detected with diaminobenzidine (Nichirei Biosciences Inc.).
Animal studies were carried out in strict accordance with the Guidelines for Proper Conduct of Animal Experiments of the Science Council of Japan. The animal experiments were conducted in strict compliance with animal husbandry and welfare regulations. The mouse study was approved by the Hokkaido University Animal Care and Use Committee (Permit number: 08-0234). The nonhuman primate study was also carried out in strict accordance with the Guidelines for the Husbandry and Management of Laboratory Animals of the Research Center for Animal Life Science at Shiga University of Medical Science and Standards Relating to the Care and Management, etc. of Experimental Animals (Notification No. 6, March 27, 1980 of the Prime Minister's Office, Japan). The protocol was approved by the Shiga University of Medical Science Animal Experiment Committee (Permit number: 2011-6-9HHH). All procedures were performed under ketamine and xylazine anesthesia, and all efforts were made to minimize suffering. Regular veterinary care and monitoring, balanced nutrition, and environmental enrichment were provided by the Research Center for Animal Life Science at the Shiga University of Medical Science. Macaques were euthanized at endpoint (7 days after virus inoculation for immunological and virological analysis) using ketamine and xylazine anesthesia followed by intravenous injection of pentobarbital (200 mg/kg). Animals were monitored twice a day during the study to be clinically scored as shown in Table S1. Animals would be euthanized if their clinical scores reached 15 (a humane endpoint).
MAb m61 showed neutralizing activities against HK483 and VN1194 (Fig. 1A). The 50% inhibitory concentrations of MAb m61 against HK483 and VN1194 were 0.42 and 0.92 µg/ml, respectively. To determine the epitope for MAb m61, escape mutants of VN1194 were selected in the presence of this MAb and the deduced amino acid sequences of the parent virus and mutants were compared. Lysine to threonine, asparagine, and glutamic acid substitutions were found at position 193 in 12.5, 25.0, and 12.5% of the cloned mutants, respectively, and 50% of the mutants had substitution from lysine to glutamic acid at position 222 (data not shown). The amino acid residue at position 193 is located near the receptor-binding site on the antigenic sites of HA molecules [31]–[34]. Accordingly, MAb m61 showed hemagglutination-inhibition activity (data not shown). We then converted MAb m61 into the human-mouse chimeric MAb ch61, and its neutralizing activities against HK483, VN1194, and VN3040 were analyzed in vitro (Fig. 1B). MAb ch61 significantly reduced the infectivity of these H5N1 viruses in a dose-dependent manner, whereas the negative control MAbs did not. The 50% inhibitory concentrations of MAb ch61 against HK483, VN1194, and VN3040 were 0.43, 1.00, and 2.29 µg/ml, respectively. These values were similar to those of the original mouse MAb m61, indicating that genetic modification of this MAb did not significantly affect the neutralizing activity in vitro.
We next investigated the potential of MAbs m61 and ch61 to protect mice from infection by HK483, known to be highly virulent for mice [35], [36]. Mice treated with these antibodies 1 day before or 1 day after virus challenge with a lethal dose of HK483 survived without clinical symptoms, whereas all control mice died (or were euthanized) within 9 days after the challenge (Figs. 2A, B). Control mice uniformly showed severe weight loss (>25%) (data not shown). Treatment at 3 days after infection also partially protected the mice (Fig. 2C), although 2 surviving mice treated with m61 showed moderate weight loss (<15%) (data not shown). All control mice exhibited severe weight loss (>25%) and succumbed to HK483 infection. Consistent with the survival data, lung virus titers of mice treated with these anti-H5 HA MAbs 1 day before virus challenge were significantly lower than those of mice given the respective control antibodies (Fig. 2D). While of statistical significance, treatment after infection only modestly reduced the titers (Fig. 2D). These results indicated that MAbs m61 and ch61 were highly protective against H5N1 HPAI virus in mice.
To examine therapeutic efficacy of MAb ch61 in a nonhuman primate model of H5N1 HPAI virus infection, VN3040 was used, since this virus causes severe, often lethal, disease in cynomolgus macaques [37]. Macaques were infected with VN3040 on day 0 and treated with MAb ch61 or control MAbs twice on days 1 and 3 after infection. Body temperatures rose upon infection and decreased after the first injection of MAb ch61, but rose again on days 4–5 (Fig. S2). One of three macaques injected with control MAbs (C3) died on day 4, whereas all three macaques treated with MAb ch61 survived until day 7 after infection (Table 1, Exp. #1). The viral titers in nasal, tracheal, and bronchial samples of macaques treated with MAb ch61 were lower than those of macaques injected with control MAbs after the first injection of MAbs (i.e., on days 2 and 3) (Figs. 3A–C). In one of the MAb ch61-treated macaques (T1), the virus was only slightly detected in the nasal and bronchial samples on days 3–7 (Figs. 3A, C). Although the virus was recovered from the nasal samples of the other treated macaques (T2 and T3), the titers were lower than those of macaques injected with control MAbs (C1, C2, and C3) (Fig. 3A). Infectious viruses were recovered from lungs of most of the macaques even on day 7 (Table 2). Interestingly, the viral titers in nasal, tracheal, and bronchial samples drastically increased after day 4 in one macaque treated with MAb ch61 (T3) (Figs. 3A–C). Similar phenomenon was partially observed in the other treated macaques. Viral titers in tracheal and bronchial samples were often higher in T2 and T3 than in control macaques on days 4–7 (Figs. 3B, C). Accordingly, relatively high titers of the virus was detected in their lungs collected on day 7 (Table 2). Two of the treated macaques (T2 and T3) lost their appetite after virus infection and their clinical scores were increased, but they temporally recovered after injection of MAb ch61 (Fig. 3D). These results indicated that MAb ch61 reduced viral titers in the respiratory secretions of all the treated macaques, although inhibition of viral propagation was temporary in two of the treated macaques. We confirmed that the MAb concentrations on days 3–7 after challenge were maintained at above 20 µg/ml in all treated macaques up to day 7 (Fig. 3E). Thus, to examine the appearance of escape mutants, we sequenced viral RNAs extracted from the tracheal samples collected from MAb ch61-treated macaques on day 5. We found amino acid substitutions identical to those seen in the escape mutants selected in vitro (i.e., K193N or K193E) in 83% (5/6) and 25% (3/12) of the cloned viral genes obtained from T1 and T3, respectively, indicating that viral escape occurred during the treatment period.
To further examine the protective potential of MAb ch61, we used an immunocompromised macaque model with influenza virus infection [29]. Macaques were pretreated with CP and CA and then infected with VN3040 on day 0. Increased body temperature was observed after infection in most of the macaques (IC1, IC2, IC3, IT1, IT2, IT3, and IT5) (Fig. S3). Body temperatures that rose upon infection decreased after the treatment with MAb ch61 in IT1 and IT2, but rose again on days 6–7. All three macaques injected with control MAbs succumbed to infection by day 5 (IC1, IC2, and IC3), whereas two (IT3 and IT4) of the five macaques injected with MAb ch61 also died on days 6 and 4, respectively (Table 1, Exp. #2).
Infectious viruses were consistently detected in the nasal, tracheal, and bronchial samples of macaques injected with control MAbs until death (Figs. 4A–C). On the other hand, the viral titers in the nasal samples of IT2, IT4, and IT5, and those in the tracheal samples of all five macaques treated with MAb ch61 decreased on days 2 and 3 (i.e., after injection of the antibody) (Figs. 4A, B). It was also noted that the viral titers in the bronchial samples of IT1, IT3, and IT5 were markedly reduced on days 2 and 3 (Fig. 4C). However, in the bronchial samples of IT2 and IT4, the titers on days 2 and 3 were similar to those of control macaques (Fig. 4C). Clinical scores in IT2 and IT5 were improved (clinical score = 0) on day 7 and IT1 slightly regained its appetite after MAb treatment (Fig. 4D). The viral titers increased on days 4–7 in the trachea and bronchial samples of some of the treated macaques (e.g., IT1 and IT2), as was the case with treatment of immunocompetent macaques. Furthermore, infectious viruses were detected in all lobes of their lungs, while the virus replication in the lungs of the other treated macaques was limited on day 7 (Table 2). Viruses with the K193R substitution in HA were recovered from the tracheal samples of IT1 and IT2 (11/11 and 17/18 of the cloned HA genes, respectively), whereas the concentrations of MAb ch61 circulating in the serum on days 3–7 after challenge were maintained at above 20 µg/ml in all treated macaques (Fig. 4E). These results indicated that treatment with MAb ch61 resulted in reduced viral loads and partial protection from lethal HPAI virus infection in immunosuppressed macaques, though this MAb treatment might select escape mutants.
Since escape mutants were frequently selected in macaques treated with MAb ch61 alone, we examined combination therapy with MAb ch61 and the neuraminidase inhibitor peramivir to further reduce viral replication and the emergence of escape mutants. CP- and CA-pretreated macaques were infected with VN3040 and then MAbs were injected on days 1 and 3 in addition to continuous administration of peramivir on days 1–5. Two macaques treated with peramivir alone had to be humanely euthanized on days 5 and 4 (ICP1 and ICP3, respectively), whereas one macaque that received the combined treatment also died on day 3 (ITP3) (Table 1, Exp. #3). Increased body temperature was observed in two control and one ch61-treated macaques (ICP1, ICP3, and ITP2) (Fig. S4). The viral titers in the nasal and tracheal samples of macaques treated with both MAb ch61 and peramivir were almost undetectable after day 3 (Figs. 5A, B). Unlike MAb treatment alone (Figs. 3 and 4), no increase of the viral titer or body temperature was observed on days 4–7 in surviving macaques treated with MAb ch61 together with peramivir (Figs. 5A–C and Fig. S4) and the concentrations of MAb ch61 in the serum were maintained at above 20 µg/ml on days 3–7 after challenge in these macaques (Fig. 5E). Accordingly, infectious viruses were only slightly detected in the limited parts of lungs of the macaques (Table 2) and escape mutations (i.e., K193N or K193E) were not found in the cloned viral genes (0/11) obtained from the MAb ch61-treated macaques. Along with the reduced viral recovery from the samples, clinical scores in ITP1 and ITP2 were generally improved on day 7. These results indicated that combination therapy with MAb ch61 and peramivir inhibited viral propagation more efficiently than MAb or peramivir treatment alone, which might also result in reduced selection of escape mutants and improved survival after H5N1 HPAI virus infection in macaques.
To determine the cause of death of the macaques, we examined inflammation by measuring IL-6 production in sera and lung tissues. In the immunocompetent macaque model, an elevated IL-6 level was observed on day 3 in the serum of one macaque (C3) that died on day 4 but not in the other macaques (Fig. 3F). In addition, the lung IL-6 level of C3 on day 3 was markedly higher than those of the other macaques (Fig. 3G). In the immunocompromised macaque model, a marked increase of IL-6 was detected in the sera and/or lung tissues of all macaques injected with control MAbs (Figs. 4F, G). Similarly, increased IL-6 levels were detected in IT4. In a MAb ch61-treated macaque that died on day 6 (IT3), bacterial infection was detected in the cerebral ventricle (data not shown) and the rapid IL-6 response was not observed, suggesting that this macaque died of bacterial meningitis, not virus infection. In the combination therapy experiment, increased levels of IL-6 were detected in the sera of ICP1, ICP3, and ITP3, all of which were humanely euthanized or died after infection (Fig. 5F). IL-6 levels in lung tissues were relatively high in ICP3 and ITP3 (Fig. 5G). Consistent with some human cases previously described [38], [39], these results suggested that increases of IL-6 in the serum and lungs might be associated with systemic inflammatory responses leading to death. While increased production of TNF-α and IL-1β were also seen in the macaques with severe disease, the other cytokines tested were unlikely correlated with disease severity of the macaques (Figs. S5, S6). Since elevated levels of IL-6, TNF-α and IL-1β are likely involved in a variety of systemic inflammatory states that are associated with endothelial barrier dysfunction, these cytokines could be important mediators of increased endothelial permeability, which might result in systemic organ failure caused by H5N1 HPAI virus infection.
To evaluate the progression of disease after the antibody treatment, we examined the lung pathology of the macaques subjected to autopsy. Macroscopically, dark red areas representing inflammation and congestion were larger in the lungs of control immunocompetent macaques (C1–C3) than in the lungs of two immunocompetent macaques treated with MAb ch61 (T1 and T2) (Fig. 6). The dark red area was larger in the lung of T3 than in the lungs of T1 and T2. These findings were concordant with virus titers in the lungs collected at autopsy (Table 2). In immunosuppressed macaques treated with control antibodies (IC1–IC3), the macroscopic lesions with inflammation, hemorrhage, and congestion in the lungs were much smaller than those in the lungs of immunocompetent macaques (C1–C3) (Figs. 6, 7). In immunosuppressed macaques treated with MAb ch61 (IT1–IT5), the reddish lesions were smaller than those in control macaques (IC1–IC3) (Fig. 7). In particular, macroscopic inflammation in IT5 was observed only around the central bronchus. In immunosuppressed macaques also treated with peramivir, macroscopic reddish lesions were smaller than those in immunocompetent and immunosuppressed macaques treated without peramivir (Figs. 6–8). The lung of ICP3, which died on day 4 after virus infection, had dark red, edematous lesions.
We then examined histological changes of the lungs collected from the infected macaques (Figs. 9–11). Severe pneumonia reducing air space was seen in a control immunocompetent macaque (C2) and a macaque treated with MAb ch61 (T1) (Figs. 9A, B). In high magnification images, lymphoid and neutrophilic infiltration, thickened alveolar walls, and alveolar edema were observed (Figs. 9C, D). The other macaques (C2, T2, and T3) euthanized on day 7 showed similar histological changes (data not shown). In immunohistochemical staining for the influenza virus antigens, NP-positive cells were widely distributed and accumulated focally in the lung of the control macaque 7 days after virus infection (Fig. 9E). By contrast, NP-positive cells were seen but did not accumulate in the MAb ch61-treated macaque (Fig. 9F). Reduced numbers of NP-positive cells were also seen in the lungs of the other treated macaques (T2 and T3) (data not shown).
In macaques under the immunosuppressed condition, lymphoid infiltration was very mild compared with immunocompetent macaques. In the lung tissue obtained from a control macaque (IC3) at 4 days after virus infection, pulmonary edema was seen in the alveoli, resulting in loss of air space (Figs. 10A, C). In a macaque treated with MAb ch61 (IT2), the air space was decreased and alveolar septa were thickened in part, but the air content was still preserved (Figs. 10B, D). NP-positive cells were seen in the alveolar epithelium of the control macaque more frequently than in that of the MAb ch61-treated macaque (Figs. 10E, F). The cuboidal epithelial cells positive for the NP antigen were type II alveolar epithelial cells. Less severe histological changes and virus infection were also seen in the other treated macaques (IT1 and IT3) compared with the control macaques (data not shown). These differences in the histological changes and frequencies of NP-positive cells between control and MAb ch61-treated macaques were also seen in the macaques treated together with peramivir (Fig. 11).
Current strategies for the control of influenza include vaccination and antiviral drug treatment. Neuraminidase inhibitors have been used for H5N1 HPAI virus infection in humans as well as seasonal influenza caused by viruses of the H1 and H3 HA subtypes. However, the efficacy of the neuraminidase inhibitors on the human H5N1 infections is unclear due to the inevitable lack of adequate control studies. Moreover, drug-resistant H5N1 viruses were indeed detected in patients [40], [41] and, importantly, H5N1 viruses with reduced sensitivity to neuraminidase inhibitors were also isolated from chickens in the endemic area [42]. Thus, alternative strategies for prophylaxis and treatment need to be developed for pandemic preparedness against the H5N1 influenza virus.
Passive transfer of neutralizing antibodies may provide an alternative strategy for both prophylaxis and treatment of pandemic influenza. It was reported that an H5N1 HPAI virus-infected patient recovered after treatment with convalescent plasma, suggesting that passive immunotherapy may be a promising option for the treatment of H5N1 HPAI virus infection [43]. The efficacy of mouse MAbs specific for H5 HAs been evaluated in a mouse model with promising results for both treatments and prophylaxis [44], [45]. However, for clinical use, induction of anti-mouse MAb-specific antibody responses should reduce the neutralizing capacity of given MAbs and also limit the repeated use of mouse antibodies. Thus, passive immunotherapy with human or humanized MAbs has also been tested in mouse and ferret models [46]–[49]. Nevertheless, the protective potential of anti-H5 MAbs remained to be elucidated in a nonhuman primate model of H5N1 HPAI virus infection.
To help develop a clinical antibody therapy, we also generated a human-mouse chimeric monoclonal antibody (MAb ch61) that showed strong neutralizing activity against H5N1 HPAI viruses isolated from humans and evaluated its protective potential in animal models. In particular, we used a cynomolgus macaque model, which simulates the H5N1 HPAI virus infection of humans more faithfully and thus has been used as an animal model for vaccine and pathogenesis studies on influenza virus infection [50]. We found that treatment with MAb ch61 reduced viral loads and partially protected macaques from lethal infection with the H5N1 HPAI virus. It was noteworthy that the protective effect was more prominent in immunosuppressed macaques, which might provide a model of protection against severe clinical disease in immunocompromised patients. Thus, this proof of concept study provides the first evidence that antibody therapy may have beneficial effects in clinical cases of H5N1 HPAI virus infection in humans. Importantly, however, mutant viruses escaping from neutralization by MAb ch61 were recovered from some of the macaques treated with MAb ch61 alone and became predominant by 7 days after infection, whereas reduced virus replication upon treatment with MAb ch61 was observed in most of the treated macaques during the initial phase of infection. These results suggest that, as was shown in a mouse model of H5N1 HPAI virus infection [51], combination therapy using two different MAbs might be needed to prevent the generation of escape mutants and would likely be more beneficial.
Taken together, the results obtained in the present study demonstrated that the therapeutic use of anti-H5 neutralizing MAb ch61 resulted in reduced viral loads and improved protection in a nonhuman primate model of lethal H5N1 virus infection. In addition, it was also shown that combination therapy with the antiviral drug provided better protection and reduced the emergence of escape mutants. Combination therapy with other antibodies recognizing different epitopes may also attenuate symptoms and prevent the selection of escape mutants.
|
10.1371/journal.pntd.0002794 | Designing a Community Engagement Framework for a New Dengue Control Method: A Case Study from Central Vietnam | The Wolbachia strategy aims to manipulate mosquito populations to make them incapable of transmitting dengue viruses between people. To test its efficacy, this strategy requires field trials. Public consultation and engagement are recognized as critical to the future success of these programs, but questions remain regarding how to proceed. This paper reports on a case study where social research was used to design a community engagement framework for a new dengue control method, at a potential release site in central Vietnam.
The approach described here, draws on an anthropological methodology and uses both qualitative and quantitative methods to design an engagement framework tailored to the concerns, expectations, and socio-political setting of a potential trial release site for Wolbachia-infected Aedes aegypti mosquitoes. The process, research activities, key findings and how these were responded to are described. Safety of the method to humans and the environment was the most common and significant concern, followed by efficacy and impact on local lives. Residents expected to be fully informed and engaged about the science, the project, its safety, the release and who would be responsible should something go wrong. They desired a level of engagement that included regular updates and authorization from government and at least one member of every household at the release site.
Results demonstrate that social research can provide important and reliable insights into public concerns and expectations at a potential release site, as well as guidance on how these might be addressed. Findings support the argument that using research to develop more targeted, engagement frameworks can lead to more sensitive, thorough, culturally comprehensible and therefore ethical consultation processes. This approach has now been used successfully to seek public input and eventually support for releases Wolbachia-infected mosquitoes, in two different international settings - Australia and Vietnam.
| In recent years, a number of new strategies using novel technologies for the control of dengue fever control have emerged. These strategies are notably different from their predecessors and not without controversy. Many also require open release field trials to test their efficacy. Public consultation and engagement are recognized as critical to the future success of these programs, but questions remain regarding how to proceed. In this paper we describe an approach to public engagement that uses social research to design an engagement framework and communication materials tailored to the concerns, expectations, and socio-political setting of potential trial release sites. This approach was developed and implemented in Australia (2008–2010) where the first publicly supported field trials occurred January 2011. We report here on the implementation of this approach in Vietnam (2009–2010) where the second release will occur in 2014. This paper describes the process, research activities, outcomes and key findings from the Vietnamese field site. It highlights key public concerns and expectations about engagement and authorization and shows how these were used to develop a more targeted, culturally appropriate and comprehensible engagement framework and communication materials. The paper demonstrates the viability of this approach to community engagement for new dengue control strategies, in a ‘developing’ country context.
| The Wolbachia strategy aims to ‘manipulate mosquito populations to make them incapable of transmitting dengue viruses between people’ (www.eliminatedengue.com). Its potential emerged following the successful transference of the insect bacterium Wolbachia pipientis from the fruit fly Drosophila melanogaster into the Aedes aegypti mosquito [1], [2], [3]. Later studies showed that the bacterium spread effectively into wild populations, had a life-shortening effect on the mosquito, blocked the development of some dengue viruses and some strains had a life-shortening effect on the mosquito [4], [5]. These properties would, in all likelihood, greatly reduce the mosquito's capacity to transmit the virus. To trial its effectiveness in real world conditions, required a series of field release through which Wolbachia-infected mosquitoes would be released into wild populations the aim being to replace these.
The Wolbachia method is one of several strategies to emerge in recent year that use a range of new technologies to combat dengue fever. While some focus on genetic modification, others, like Wolbachia, use biological control [1], [5], [6]. However, these strategies are very different from their predecessors, notably source reduction and insecticide use, and are not without controversy. Moreover, many require open field releases to test their efficacy and potential uses. Significantly, these need to occur in the locations where dengue vectors are found, most commonly the homes, and places of work, education, worship and leisure of local residents at a release site.
Most commentators recognize that the political and ethical complexities of community field trials are considerable and that public and government approval in conjunction with high quality science are of central importance. It is also widely acknowledged, that given the spread and increasing prevalence of dengue fever throughout the tropics, field trials will need to be undertaken in a variety of locales, regions and countries, both so called developed and developing. While public engagement is also recognized as critical to the use and future success of these strategies, many questions remain regarding how to proceed in ways that are ethical, and comprehensible to those being asked to trial these strategies in their homes and backyards.
In 2008 an approach to engagement drawing on anthropological methodologies and insights was developed for the Wolbachia strategy. It was implemented in Cairns, Australia from 2008–2010 [6] and in January 2011 the first field release of Wolbachia-infected Ae. aegypti commenced. Drawing on anthropological methodologies and insights, this approach recognizes that different communities will have divergent expectations, knowledge, concerns, political structures and cultural sensibilities, that need to be understood and taken into account, if one is to engage sensitively, ethically and effectively [6], [7], [8], [9], [10], [11] [12]. The most reliable way to do this, is to talk with residents at a potential release site about the new dengue control methods and ask what their concerns are, how they want to be engaged and what would constitute authorization [6], [11]. From this research, an engagement framework is developed that is sensitive to local needs, expectations, knowledge and concerns.
So, rather than simply adopting an engagement strategy that was developed elsewhere and implementing it in another setting, this approach uses social research to design an engagement framework and communication materials that are tailored specifically to potential release sites. In brief, it begins by undertaking systematic social research to: (a) document the socio-political context and identify the various publics and stakeholders at the potential release site, (b) determine how they want or expect to be engaged and the forms this should take, (c) explore what would constitute authorization, (d) identify any questions or concerns they might have about the Wolbachia strategy, (e) identify lay knowledge of the disease, its transmission, vectors, perceived risk, etc. and (f) develop responses to these. The results of this research are then used to design a community engagement framework tailored specifically to the sociopolitical setting, and the requirements and expectations of a given population [6].
This paper describes the use of this approach from June 2009 to September 2010 at the second potential Wolbachia release site - Tri Nguyen Island, in central Vietnam. It outlines the process, research activities, outcomes and key findings from the Vietnamese field site. It also highlights key public concerns and expectations about engagement and authorization and shows how these were used to develop a more targeted, culturally appropriate and comprehensible engagement framework and communication materials. Most significantly, the paper demonstrates the viability of this approach to community engagement for new dengue control strategies, in a ‘developing’ country context. It is hoped that by reporting on the methodology, process and results, that readers will be able to see the steps taken and assess the capacity of this approach to reflect and address local requirements and expectations, as well as its potential applicability to other programs.
Dengue fever has a long history in Vietnam and continues to represent a major public health problem [13]. Disease transmission occurs throughout the year in the south of the country but is limited to the warmer months in the northern and highland areas. Two vectors are active in disease transmission, the Ae. albopictus and Ae. aegypti mosquitoes [12] [14], [15]. Historically, dengue control in Vietnam has focused on source reduction, container management, insecticides and community mobilization – the later relying on household visits by collaborators and the management of water storage containers [15]. Since 1989, community-based biological control initiatives using Mesocyclops spp. to control mosquito breeding in household water containers have also been introduced [15], [16], [17], [18]. These have also included successful community mobilization around the management of water storage containers and the presence of Mesocyclops spp.
Tri Nguyen Island (TNI) or Hon Mieu (‘Island Shrine’), as it was known historically, is located to the southeast of the city of Nha Trang (NT) in Khanh Hoa province, central Vietnam (Figure 1). It was selected as a potential release site for the Wolbachia strategy for a number of reasons. These include its physical isolation, its proximity to the Pasteur Institute in Nha Trang, famous for its work on infectious diseases, and residents' previous involvement in mosquito ecology and vector studies. Since the late 1940s and during the war with France, people from other provinces such as Quang Nam, Quang Ngai, Binh Dinh, Phu Yen moved to TNI. Today the island is stratified into 3 hamlets each with its own leader, which together represents one sector of the Vinh Nguyen ward of Nha Trang city, in Khanh Hoa province. In 2009 the population of TNI was 3253 residents, living in 710 households spread across three hamlets, each of which had its own political leaders [19].
The social research activities described here were undertaken over 16 months (June 2009—September 2010) and included six weeklong fieldtrips to Tri Nguyen Island. Research activities centered on two key groups: a) Residents of Tri Nguyen island and b) health providers, government officials and scientists with responsibilities at the local, regional and national levels (hereafter, Leaders). It is widely established that qualitative research methods are the most appropriate for assessing the views of a population, in part because of their emphasis on context and their documentation of knowledge and attitudes in a given geopolitical setting. In this study, key or recurring themes from the qualitative research were explored further using quantitative measures (a household survey and anonymous questionnaire) and results were triangulated (compared, challenged or confirmed) across different methods: interviews, observations, questionnaires and a series of community meetings and workshops - styled on a focus group. Importantly, the findings presented here should not be seen as isolated research activities, but as a body of interconnected data developed over time using iterative processes and then contextualized, triangulated and crosschecked. An overview of the research activities undertaken at each phase, the issues they explored, how participants were recruited and the outputs they produced, is provided in Table 1.
In the following section we describe the methods used at each step in the research process and how the key results were used to design an engagement framework and communication materials tailored to this potential release site. We do so on the assumption that successful engagement leading to a release using new dengue control methods is still somewhat rare and that it is the process as much as the results that will be of interest to others looking to engage communities around new disease control strategies.
The first step in the process was to immerse the two social science staff in the science of dengue and the Wolbachia strategy and to identify any information about the history and demographics of the potential release site. This included an extensive literature review on dengue fever, bio-control, GM food and organisms in Vietnam and internationally, and the development of a database (Table 1).
In June 2009 a PowerPoint presentation was developed (Table 2). It used the same slides and followed the same narrative structure as the presentation used at the Australian field site, to which Vietnam specific information was then added. Graphics with small amounts of text were used to communicate key messages around the following themes: increasing prevalence of dengue (local, national, international); disease transmission and vectors; current control measures in Vietnam; the Wolbachia strategy; the Australian pilot release; a potential release on TNI. A discussion was then facilitated to identify any questions or concerns and seek guidance on how to engage, whom to engage, what would constitute authorization (Table 2).
In July 2009 this presentation was used at the first of three leaders workshops, with thirty national, provincial, district and commune leaders, scientists and local health providers in attendance. Participants were chosen purposefully, because of their roles as leaders or officials and formally invited to attend. They included Ministry of Health leaders and scientists, members of the Khanh Hoa People's Committee and Khanh Hoa Health Department, and community and union leaders from TNI and NT. Project scientists and social scientists from Vietnam and Australia were present at the workshop.
At the first Leaders Workshop (Hanoi, July 2009) project employees were introduced to participants, and presentations delivered on the impact of dengue fever in Vietnam, the science behind the Wolbachia method, the potential release strategy in Vietnam and progress at the Australian release site (scientific and engagement). The presentation was approximately 20 minutes long, after which a discussion was facilitated while the second social scientist made observations on body language; interactions between participants and audio recorded the entire event - presentation and discussion. Participants were asked if they had any questions, thoughts or concerns and what their expectations around the strategy, engagement and authorization might be. Input was also sought to identify key stakeholders as well as feedback on the presentation and project communication materials.
An anonymous questionnaire was distributed at the end of the leaders workshop. It asked participants to identify any concerns or questions, evaluate how acceptable the Wolbachia strategy was, how they wished to be engaged and what would constitute authorization. This questionnaire provided baseline data for evaluating responses to the Wolbachia strategy through time, and was an important mechanism for tracking responses to the project among the leaders group and later, local residents. This process was also used at the Australian field site [6].
In early September 2009, a senior entomologist working for the Wolbachia project, who was well known to the local community, introduced project staff to Tri Nguyen (TNI) residents. Limited information on the history and demographics of TNI was publically available so a purposive sample of 10 in-depth interviews on the history, socio-political structure, social demographics and dengue history of TNI was undertaken with local residents and leaders. Purposive sampling involves the deliberate selection of individuals because of the crucial information they can provide – in this case local leaders with a detailed knowledge of the history and socio-political make up of the TNI community. These interviews, alongside informal discussions with local health and mosquito control staff and results from the Leaders workshop, were used to develop a detailed stakeholder contact list, which was added to over time. It categorized individuals and groups according to: level of influence (local, national, international); local expectations around engagement; marginality; and accessibility. This helped to determine who was engaged and when. In addition, results from the interviews were also used to improve the PowerPoint presentation and communication materials to be used at future community meetings and workshops.
In the next stage of the process, the results from these interviews were used to develop a Household Survey that examined the following: political structure (leaders, groups, organizations); social demographics of TNI (name, age, gender, occupation, education level, religion, family structure); knowledge of dengue, its vectors, control methods and perceptions of risk; and local health issues of concern to residents. The survey provided a brief introduction to the Wolbachia strategy and sought to identify early responses and advice on engagement and authorization for a release. The survey was piloted with 10 residents, reviewed and later administered to 100 households randomly selected from a list of 710 provided by local authorities - approximately 14% of all households.
The second Leaders Workshop was held in the mainland city of Nha Trang, and attended by 33 participants representing local (TNI) and district leaders, government representatives, scientists, local health providers and mosquito control staff. An update on the progress of the science, the Australian risk assessment and the release was provided and further advice sought on stakeholders, forms of engagement, authorization and the presentation and communication materials. As noted above, a discussion was facilitated and any questions or concerns were noted. The event was also audio-recorded for later transcription and analysis and the anonymous questionnaire distributed.
During the next phase of the project, 46 community meetings, attended by 661 local residents, were held in TNI during four, one-week trips in January (T1), March (T2), May (T3) and July (T4) 2010 (Table 1). The aim of these meetings was to gauge the range of views on the Wolbachia strategy, the science, potential release, engagement and authorization using the same focus group style format as the Leadership Workshops. Discussion was facilitated around the following themes: questions raised, concerns, acceptability, how and whom to engage and authorization. The meeting was audio-recorded and the anonymous questionnaire distributed at the end (Table 1).
During the second visit (March 2010) local residents who had contracted dengue attended the meeting and spoke of their experiences during the presentation. In addition, new results from the independent Australian Risk Assessment and new experiments showing Wolbachia was not transmitted to predators who ingested the infected mosquitoes were added to the presentation. During the third (May 2010) and fourth visit (July 2010), results of Vietnamese experiments indicating that ingesting infected mosquitoes did not affect or lead to transmission of Wolbachia among local predatory species was included. By this time we also had more information about government approval processes (following the final Leaders Workshop) and the likely time frame for this, so this to was incorporated into presentation. Other than these additions, the presentation was the same at each visit.
For the community meetings on TNI, a small number of participants were approached directly and sampled purposefully (i.e. health staff, hamlet and local union leaders and members) based on the stakeholder list we had begun developing. However, the majority of participants were sourced through flyers, posters and announcements over the community loudspeaker prior to each visit. As such the sample was broadly representative, with participants self-selecting to be involved. We aimed to reach at least one person from every TNI household (Table 1).
During the second (March 2010) and third (May 2010) visits, 20 in-depth interviews were also undertaken with residents from TNI and NT (aged 18–60 years) who could not attend the meetings. We approached marginalized or harder-to-reach groups identified during the Leaders Workshops and early interviews (n = 10) with local leaders. This included fishermen who were often away from the island, women with domestic and employment duties and minority religious or ethnic groups who it was thought might otherwise not have been engaged. These interviews began with the PowerPoint presentation and explored the same issues as the workshops and meetings. They were audio recorded for transcription purposes.
The third and final Leaders Group Meeting was held in Nha Trang and attended by 33 local, district and national leaders, local health and mosquito control staff and scientists. Presentations on the results of both the social and scientific research were provided, and further advice sought on regulatory pathways and approval processes in Vietnam. The anonymous questionnaire was also distributed.
Two social scientists and at least one senior entomologist attended every meeting or workshop. Prior to any research or engagement, an extensive and detailed list of questions and answers posed by the public at the Australian field site, was made available to Vietnamese project staff. It was posted to the project's website in June 2009 (see http://www.eliminatedengue.com/faqs for the current version) and later, on the Vietnamese language version and developed into flyers provided to participants. As the research progressed, it was clear that this extensive list covered almost every question posed by participants in the Vietnam research. When new questions or issues did arise, they were answered, if possible. If it was not possible to answer a question, it was recorded so that a response could be sought from appropriate staff and later provided back to the person asking the question and the community. This practice helped to ensure that information across the field sites, project staff and research activities - meetings, workshops, interviews etc. - was accurate and consistent.
Results from the in-depth interviews (n = 10) and Household survey identified three active civic groups on TNI: the Women's, Youth and Farmer (includes fishing) Unions. They were well respected in the community and would in all likelihood, be central to future research activities as well as an important conduit for disseminating information. They were given priority in the engagement framework that was being developed. The Household survey indicated that 29% of those surveyed identified as ‘Buddhist’, and 71% as ‘Non-religious’. In addition, 89% of adults surveyed (over 18 years) had a primary or secondary education, 6% had completed high school and 5% self identified as non-literate. Fishing (a predominantly male occupation) was the primary source of income for 70% of households surveyed, with small scale trading enterprises providing income for 13%. Women ran most of these. Average monthly incomes per household ranged from up to i) 2,000,00 DN (USD $95) 39%; ii) from 2,001,000 (USD $96) to 4,000,000 (USD $190) 37%; iii) from 4,001,000 (USD $191) to 6,000,000 (USD $285) 16% and iv) more than 6,000,000 DN (USD $285) at 6%. In sum, 76% of households surveyed earned up to 4,000,000 DN (USD $191) per month, based on exchange rates in September 2009.
Participants from the leaders interviews (N = 10) and workshops advised that a presentation delivered in a meeting and styled on a focus group - like the one they had attended - was in fact an appropriate way to communicate the Wolbachia story to TNI residents. They also recommended project staff work through the highly structured networks of governance identified above and undertake extensive consultations with local health staff, unions and residents at TNI. In the Household Survey, TNI residents concurred with this finding, suggesting project staff work through local hamlet and union leaders who in turn would inform residents. As one local resident expressed it: “When we want to know any information, the first persons we always come to are the local leaders such as: hamlet leaders, women union's leader…. I think that they are in charge of responding to any issues related to our local community” (interview, TNI resident, 42-year-old woman). Many residents explained that the role of the hamlet and union leaders was not only to represent them, but also to inform and connect local people with community activities – an important finding for developing an engagement framework.
Interviews with local hamlet leaders and health workers signposted that they expected to be engaged early and regularly, so meetings with these individuals and groups were given priority in the research phase and the final engagement strategy.
Residents also expected to be widely consulted about the strategy:
In addition, results from the Household survey, interviews (n = 10) and leaders workshops provided a number of insights critical to understanding the multiple ‘publics’ at the potential release site. They were used to prepare a comprehensive community profile, and develop a stakeholder contact list, the later categorized and prioritized groups according to local expectations (i.e. health workers and union officials to be engaged early), marginality (i.e. women and religious groups) and accessibility (i.e. fishermen). It began to emerge that an engagement strategy for TNI would need work through established political structures and engage at least one person from every household.
Results from the Household survey (n = 100) indicated that residents were well versed on prevention activities and current control methods, i.e. covering water containers, insecticide use, bed nets etc. [20]. Although 65% of those surveyed correctly identified key domestic breeding sites, there was also a strong and recurring association between ‘dirty places’, namely sewers, forested areas, and refuse and the mosquitoes thought to transmit dengue. Although 65% were able to identify the mosquito primarily responsible for dengue transmission in TNI, only 35% were able to explain the transmission cycle or describe symptoms – both of which were central to understanding the Wolbachia strategy (for more details see Huong and McNaughton 2012.
The Household Survey (n = 100) revealed that most residents (93%) identified dengue fever as a dangerous disease within their community. The main reasons cited were that it can be fatal (83.9%) and can spread very fast (40.9%). Residents looked first to local health workers (95%), followed by television (55%) and local officials (41%) as trusted sources of information on dengue and health. These and other results were used to develop a more targeted PowerPoint presentation on the Wolbachia strategy that focused on symptoms, the transmission cycle and the habitats of the vectors, three key gaps in local understandings. This presentation was used at 46 focus-group style meetings with 661 residents (Table 1).
The most prominent and recurring issue for respondents across the residents' and leaders' meetings and interviews was the safety of the method for people, animals and the environment. Relatedly, participants wanted to know if it was safe to be ‘bitten’ by a Wolbachia-infected Ae. aegypti mosquito, if was safe to drink water with these mosquitoes, their larvae or pupae in it, and if this would lead to Wolbachia being transmitted into other organisms, especially people. For example, a member of the youth union asked “Is it a problem if we are bitten by Wolbachia-infected mosquitoes? Can Wolbachia be transmitted into our body?” Some also expressed concerns that Wolbachia-infected mosquitoes might become susceptible to or able to transmit other diseases: “After releasing the Wolbachia-carrying mosquitoes, the dengue fever may be reduced, but how about other diseases; will it cause any other disease to come to our Island?”
Responses to questions relating to the potential transmission of Wolbachia to humans, other organisms or the environment included but were not limited to the following:
A discussion about the role of many project staff in blood feeding large numbers of these mosquitoes in the caged trials and laboratories (including photos) often ensued.
Alongside safety, considerable discussion centered on why TNI had been chosen as a potential release site, if it would be the first to trial this strategy and who would be responsible if anything should go wrong. For example, “I heard many people who participated in your discussions ask each other why this method was not applied somewhere else but on Tri Nguyen Island. Is it safe if it is applied here?” (Male, 25 years, member of the Youth union). Another resident expressed concerns about safety and responsibility as follows:
For many participants, assurances were sought that Australia rather than Vietnam would be the first place to release these mosquitoes. In addition, residents wanted clear pathways of responsibility outlined so they knew whom to speak to should something go wrong. Several residents asked directly, “Which agency will be in responsibility in case the release strategy will cause additional impacts?” (CM, T3). Local government and health officials also wished to know who would be responsible in the event of any problems and sought greater clarity from each other and project staff and leaders, regarding their specific responsibilities during a pilot release.
Clear lines of responsibility had been established and these were relayed to residents with responses like the following:
Another common concern centered on the efficacy of the strategy, especially in the long term. One resident attending the group asked “…does it [Wolbachia] have any side effects after being introduced into mosquitoes? It is a bacterium, so it must be harmful to some extent”. (CM, T2). Many participants were also concerned that the life shortening effect of Wolbachia would impact on the success of the strategy, “How can Wolbachia-infected mosquitoes help prevent the disease when they die early after being released?” (CM, T1). “I am concerned that it may be difficult for Wolbachia-infected mosquitoes to find another mosquito to copulate with, or that they may die before they can lay their eggs” (CM, T2). Many participants were interested in eliminating all mosquitoes or why current control methods were no longer as viable: “Why don't you try to kill all mosquitoes? Why don't you spray chemicals to kill them all?”(CM, T3). The 2009 Household Survey (n = 100) had indicated that while 86% found the Wolbachia strategy acceptable, the use of insecticides either inside (67%) or outside (74%) their homes was also viewed positively (see Table 3).
Responses to questions relating to efficacy, focused in part on the role of the trials in determining the effectiveness of this strategy, and that results from the Australian releases would be reported back to the community during future engagement. They also included, but were not limited to, the following (for more details http://www.eliminatedengue.com/faqs):
The nature and scale of the pilot release were also prominent, recurring issues from the community meetings and interviews (n = 20). Respondents commonly sought a high level of detail regarding the release, its timing and scale. Questions focused on further details regarding how many mosquitoes would be released, if this would be in all or only some houses, and how long it would take for wild mosquitoes to be infected. There was a lot of discussion about what residents should do to assist the effectiveness of the strategy and what impact this might have on people's lives. For example:
This question was answered as follows:
During the Residents meetings (n = 46) and interviews (n = 20) assurances were often sought that the release would not negatively affect or inhibit local lives and livelihoods and that householders would be made aware of any activities they needed to undertake before or during a release. There was strong support for being advised and informed well in advance of a release “so that we are well prepared for it?” (CM, T2).
In general, we responded to these questions as follows:
The anonymous questionnaire, handed out at the end of each meeting included the question, “Do you have any concerns about the Wolbachia method?” which was used to track residents' perceptions of the project through time. As indicated in Figure 2, the number of concerned participants declined significantly as the Residents' Meetings and interviews continued. During the final two visits to TNI in May and July 2010, no participants objected to a release (Figure 2).
Participants were asked at the Leaders workshops (n = 3), Residents' Meetings (n = 46) and interviews (n = 20) how they would like to be engaged about the Wolbachia strategy. There was a strong desire for public consultation across all groups, consistent support for in-community presentations and a strong preference for face-to-face interaction with the project team and senior health officials. There was much less support for the use of media, posters, brochures and leaflets.
One of the most common requests related to the scale of the engagement. At the local level, participants consistently indicated that well before a release the project team should engage with every community member and provide ongoing information on the safety and benefits of the project well before a release. For example, “More people, all people should be invited. A small group of participants like this is not representative enough to make a decision. It is perfect if 100% of people agree” (CM, T3). Others suggested that, at minimum, one person from each household should be engaged. For example, “One person from every household should be invited. The main income earner in every household should be invited so that they can remember what they have heard and tell others. If you invite those who are too old, they may not have a good memory to tell others about what they have heard” (CM, T3).
Participants were also asked what would be the best format to engage people on TNI about the strategy and in the lead up to a release if regulatory approval was given. There was an expectation of ongoing consultation about the strategy among residents, leaders and health staff, where updates on the science, safety, risk assessment, regulatory approval, pilot release strategy, results from the Australian release and a well-defined structure around roles and responsibilities would be provided. Some were also concerned that without this, people might forget what they had learned about the strategy and how to respond to a release. Community leaders and health professionals suggested that residents would come to them for information and guidance, especially if things did not go to plan. As such they sought to have clear pathways on any future roles and responsibilities they might have negotiated, outlined and communicated to residents well before a release.
As well as calling for regular updates, participants consistently identified the importance of a large meeting attended by at least one representative from each household as well as local and provincial leaders – essentially a forum where people could raise their ideas, discuss benefits and concerns and make a collective decision (Table 4). There was also a strong preference for voting at such a forum, as one resident expressed it “Voting can be used. Those who agree will raise their hand. If the majority raises hands that means it is supported” (CM, T1) (Table 5). As such a large public meeting held in the community or a vote was identified as a mechanism through which the project and the release would gain final and collective approval from the TNI community, alongside support of government officials (regulators, Ministry of Health and scientists) (Table 5).
The anonymous questionnaire also asked whether Resident's would support a pilot release if (a) the Ministry of Health undertook a risk assessment and approval process, and (b) scientific data from the Australian release site proved to be positive. During the first phase of social research and engagement in January 2010, 80.2% were in favor of the pilot release. By the final phase in July 2010, this had risen to 99.4% (Figure 3). Of course, participants can and do change their minds and they could react differently when a release happens, and this is a limitation of this study.
However, results from the Australian research did allow us to predict quite successfully how people would react and there was no last minute call to stop the release in Australia. Although a release has not yet occurred in Vietnam, the most recent engagement with TNI residents (2013) - where one person from every household was interviewed - 99% of householders were still in favor of the release, only a few months shy of its eventuality (data not shown).
Participants in this study brought different forms and degrees of knowledge about science, dengue fever, its vectors and control to their encounter with the Wolbachia strategy. While breeding sites and control measures were relatively well known, and many participants connected mosquitoes to dengue fever, few people were able to explain the dengue transmission cycle. Understanding the transmission cycle, and the role of Wolbachia in blocking the development of the dengue virus, was essential to understanding the strategy, perceived risks and ensuring residents were well informed. Addressing these gaps and assumptions was given high priority in presentations and communication materials used in research phase and in the development of a TNI engagement framework. It was essential to ensure as much as possible, that the project communicated the Wolbachia story as accurately, effectively and transparently as possible so that community members were able to engage, critique and ultimately decide whether they wish to support such an initiative. As the quotations provided above indicate, a good understanding of the strategy was evident among many respondents.
The research identified a range of concerns regarding the safety of Wolbachia for people, animals and the environment and in particular, the potential for transmission of the bacterium through biting behavior or accidental ingestion (the later unique to the Vietnamese field site). Identifying these concerns well in advance of a release (or a formal engagement strategy) provided an opportunity to develop clear, consistent responses to these issues that were comprehensible to local populations.
Although the scientific team was confident about the safety of the strategy, new experiments examining the potential for Wolbachia to be passed into the human bloodstream through the mosquito's saliva during feeding [5] as well as testing Wolbachia's capacity to be transferred from mosquitos to predator and non-predator species such as spiders, fish, copepods and geckos, common to local environments were undertaken [21]. Findings from these studies were incorporated into the community presentation and communication materials and fed back to residents during the research phase. Results of an independent Australian risk assessment, which suggested the risk to people and the environment were negligible, were also added. This, coupled with confirmation that the Wolbachia strategy would be trialed first in Australia, demonstrated to residents that their concerns about safety and the location of the first field trial had been taken seriously. This was important to the success of the approach and to future engagement.
Results from the social research suggested that procedures for consulting communities were well established on TNI and this involved consulting first with leaders at the national, provincial, and hamlet levels before moving out into the community. Indeed, it appeared that a similar process was expected for any research or future engagement seeking community support and authorization. Most people we spoke to wanted the community to come together as a group or, as representatives of individual households, determine the benefits and risks and decide through a vote or similar mechanism, whether or not to support a release. It emerged that although the role of leaders, government officials and scientists in decision making was important to many residents, so too was the role of local residents in deciding household by household, on whether or not to use this strategy. This we learned, was the process most residents thought should be used to seek their support and authorization for a release. This was the approach that was taken the formal engagement phase post 2010.
The growing acceptability of the Wolbachia strategy and a release over the research phase suggests that this approach was effective (Figure 3). Engagement from 2011–2012 drew on all of the findings and lessons highlighted above. Following an update on the latest results from the science and the first field trials in Australia [22] a representative from every household on TNI was asked to provide their consent, or not, for a release. Of these, more than 95% agreed to support the release. In 2013 the Vietnamese government gave regulatory approval for an open field release in TNI.
The approach described here produced a number of critical insights that helped determine the nature, scale, style and form of an engagement framework tailored specifically to the needs and wishes of officials and residents and the potential release site in Vietnam. It used systematic social research and consultation to (a) identify, inform and involve the public; (b) listen to their responses, questions and concerns; (c) examine the deeper cultural assumptions that underwrite these responses, including lay knowledge of dengue; (d) explore ways of responding to these issues i.e. scientifically, through education, the media, schools programs or new forms of participation; and (e) explore and enact suggestions regarding future engagement, participation, communication and authorization.
Through this process we found that residents at the potential release site in Vietnam expected to be fully informed and fully engaged about the science, the project, its safety, risk assessments, the nature of the release and who would be responsible should something go wrong. Along with key health and government officials and representatives they provided advice on how best to engage their community and wanted the opportunity to meet with and ask questions of scientists involved in these programs and to have their concerns taken seriously and answered respectfully. This approach thus afforded the development of a more culturally appropriate and comprehensible engagement framework and communication materials that empowered those being asked to assess, critique and support a field trial or release. It has now been implemented at three socially and politically diverse and complex field sites (seven in Australia, one in Vietnam) in two countries, demonstrating its capacity to reflect local requirements and its potential for use in other programs and other regions.
|
10.1371/journal.pgen.1001235 | Endocytic Sorting and Recycling Require Membrane Phosphatidylserine Asymmetry Maintained by TAT-1/CHAT-1 | Endocytic sorting is achieved through the formation of morphologically and functionally distinct sub-domains within early endosomes. Cargoes destined for recycling are sorted to and transported through newly-formed tubular membranes, but the processes that regulate membrane tubulation are poorly understood. Here, we identified a novel Caenorhabditis elegans Cdc50 family protein, CHAT-1, which acts as the chaperone of the TAT-1 P4-ATPase to regulate membrane phosphatidylserine (PS) asymmetry and endocytic transport. In chat-1 and tat-1 mutants, the endocytic sorting process is disrupted, leading to defects in both cargo recycling and degradation. TAT-1 and CHAT-1 colocalize to the tubular domain of the early endosome, the tubular endocytic recycling compartment (ERC), and the recycling endosome where PS is enriched on the cytosolic surface. Loss of tat-1 and chat-1 function disrupts membrane PS asymmetry and abrogates the tubular membrane structure. Our data suggest that CHAT-1 and TAT-1 maintain membrane phosphatidylserine asymmetry, thus promoting membrane tubulation and regulating endocytic sorting and recycling.
| The process by which cells take up nutrients and other large molecules from the extracellular environment is known as endocytosis. At the cell surface, external molecules become enclosed in membrane spheres called endosomes. Early endosomes serve as a sorting station, directing the contents (cargo molecules) to the correct compartment within the cell. This is thought to be achieved by the formation of membrane structures with distinct shape and function. For example, cargoes destined for recycling and degradation are processed through tubular membrane structures and big vesicular compartments, respectively. However, it is poorly understood how early endosome membranes are shaped into different structures. Here we show that two proteins, CHAT-1 and TAT-1, regulate membrane structure and are important for normal endocytic transport in the nematode worm C. elegans. TAT-1 and CHAT-1 are found in tubular membrane structures along the sorting and recycling pathway, where they enrich the outer membrane layer with a lipid called phosphatidylserine (PS) and probably change the membrane curvature. Loss of tat-1 and chat-1 function disrupts the asymmetric distribution of PS, abolishes tubular membrane structures, and abrogates endocytic sorting/recycling. Our data support a role of TAT-1/CHAT-1–regulated membrane PS asymmetry in promoting membrane tubulation for endocytic cargo sorting and recycling.
| In eukaryotic cells, internalized cargoes are transported to early endosomes where they are sorted to be recycled back to the plasma membrane, degraded in lysosomes or delivered to the trans-Golgi network. Early endosomes display a complex and pleiomorphic organization with many tubular processes emanating from central vesicular elements as revealed by three-dimensional reconstruction [1]–[4]. Internalized receptors dissociate from their ligands in early endosomes which have a slightly acidic internal pH; subsequent segregation of the receptor and ligand is thought to be achieved by a geometry-based mechanism [5]. Receptors and other membrane proteins concentrate in the tubular extensions which contain most of the endosomal membrane, whereas soluble contents are enriched in the vesicular components which account for the bulk of the endosomal volume [1], [6], [7]. The recycling vesicles which arise from the tubular extensions may undergo fast recycling, by fusing directly with plasma membranes, or slow recycling, by transporting cargoes through the endocytic recycling compartment (ERC), a collection of tubular membrane structures arranged around the microtubule-organizing center [7], [8]. Both cargo sorting and subsequent recycling require extensive membrane remodeling to form tubular extensions, which have a high ratio of surface area to luminal volume, thereby effectively concentrating cargoes on recycling membranes. However, it is not clear at present how these tubular processes are formed and maintained.
Both proteins and lipids are required for shaping membranes into various structures including tubular extensions. For example, BAR (Bin/amphiphysin/Rvs) domain proteins, which are central regulators of membrane remodeling, are capable of inducing membrane tubulation [9]. Members of the EHDs/RME-1 family of ATPases, which are important regulators of endocytic recycling in mammals (EHD1-4) and C. elegans (RME-1), associate with vesicular and tubular membranes in vivo and tubulate liposomes in vitro [10]–[14]. On the other hand, phospholipids regulate membrane shaping by either recruiting and activating effector proteins on target membranes or directly affecting membrane curvature. For instance, membrane-shaping proteins like BAR proteins and dynamin are targeted to specific membrane compartments by binding to different phosphoinositides through either a lipid-binding domain (PH or PX) or by electrostatic interaction or both, while phospholipid-binding and membrane deformation by EHDs/RME-1 family proteins appear to be mediated through their helical domains [15], [16]. In addition to acting through a protein-recruiting mechanism, phospholipids can directly affect membrane curvature. It has been observed that addition of phosphatidylserine (PS) to ATP-containing erythrocyte ghosts stimulates the formation of endocytic vesicles [17]. Notably, phosphatidylserine is asymmetrically arranged between the two membrane leaflets, being enriched in the inner leaflet of cell membranes [18]. As the most abundant anionic phospholipid of cell membranes, PS regulates surface charge and protein targeting in cultured cells, where it is also observed on the cytosolic surface of endosomes and lysosomes [19]. However, it remains to be determined whether PS, or PS asymmetry, is involved in shaping membranes into tubular processes during sorting and recycling.
Previous studies suggest that establishment and maintenance of PS asymmetry require the activity of type IV P-type ATPase family proteins (P4-ATPases), which selectively sequester PS and phosphatidylethanolamine in the cytosolic leaflet of the membrane [17]. The P4-ATPases are a large family of putative aminophospolipid translocases with 14 members in human, 5 in yeast and 6 in C. elegans [20], [21]. The two founding members, mammalian ATP8A1 (ATPase II) and yeast Drs2p, were found to transfer spin- or fluorescent-labeled PS analogues from the inner to the outer membrane leaflet when purified and reconstituted into proteoliposomes, indicating that they have intrinsic flippase activity [22], [23]. Deletion of DRS2 causes defects in phospholipid translocation and protein transport from the TGN to endosomes and vacuoles in yeast [24]–[27]. Interestingly, loss of function of yeast Cdc50p, a transmembrane protein that does not contain sequence features indicative of a direct involvement in phospholipid translocation, results in similar defects in lipid asymmetry and vesicular transport to those caused by DRS2 deletion [28], [29]. Cdc50p and the related protein Lem3p were later found to form complexes with P4-ATPases Drs2p and Dnf1p, respectively, and were shown to facilitate transport of the complexes out of the ER [28], [29]. Moreover, a recent study suggested that Cdc50 proteins are integral components of P4-ATPases and directly participate in the ATPase reaction cycle [30]. Nevertheless, the cellular function of Cdc50 family proteins in multicellular organisms remains unknown.
In C. elegans, the P4-ATPase TAT-1 which is most closely related to yeast Drs2p and mammalian ATP8A1, maintains cell surface PS asymmetry, thus preventing appearance of PS in the outer leaflet of the plasma membrane [31]. Moreover, an early step of endocytosis and trafficking in the lysosome biogenesis pathway was found to be defective in tat-1(lf) mutants which accumulate large intestinal vacuoles with characteristics of late endosomes or lysosomes [32]. However, it is not known how these different endocytic trafficking processes are affected in tat-1(lf) animals, or whether the defects are caused by disruption of PS asymmetry.
In the present study, we identified a novel C. elegans Cdc50 family protein, CHAT-1, which acts as the chaperone of the TAT-1 P4-ATPase. Endocytic sorting is severely affected in chat-1 and tat-1 mutants, causing abnormal cargo recycling and degradation. CHAT-1 and TAT-1 associate with PS-coated tubular membranes of early endosomes, endocytic recycling compartments (ERCs) and recycling endosomes. Loss of tat-1 and chat-1 function disrupts membrane PS asymmetry and abrogates the tubular membrane structure of sorting and recycling compartments. Our data suggest that TAT-1 and CHAT-1 maintain membrane PS asymmetry to regulate membrane tubulation for cargo sorting and recycling.
PS is usually confined to the inner leaflet of the plasma membrane and only appears on the cell surface during apoptosis [33]. From a forward genetic screen for mutants which disrupt specific labeling of apoptotic cells by a PS-binding protein, we isolated the mutant qx36, and several alleles of tat-1, which encodes a C. elegans P4-ATPase [34] (Materials and Methods). Since P4-ATPases are known to regulate membrane PS asymmetry [17], we investigated whether PS distribution was disrupted in our mutants. Surface-exposed PS can be detected using the secreted fluorescent biosensor GFP::Lact-C2 or Annexin V, both of which bind selectively to PS [19], [31]. In wild-type animals, surface-exposed PS was only observed around apoptotic cells (Figure 1A, 1D). In qx36 and tat-1 mutants, PS labeling was seen on the surface of virtually all cells, indicating that plasma membrane PS asymmetry was disrupted and PS was exposed on the surface of both living and dying cells (Figure 1B–1F; Figure S1A–S1E) [31]. In contrast, no surface labeling was observed in either wild type, qx36 mutants or tat-1(qx30) mutants using a secreted biosensor GFP::Lact-C2(AAA) which does not bind PS (Figure S1H–S1J) [19].
We cloned the gene affected in qx36 mutants and found that it encodes a C. elegans homolog of yeast Cdc50p (Figure S2A, S2L; Materials and Methods). Since Cdc50p is a chaperone and non-catalytic subunit of Drs2p, the P4-ATPase homologous to TAT-1 [28], [30], we named this gene chat-1 (chaperone of tat-1). CHAT-1 is most similar to Cdc50p in yeast and CDC50A in human (Figure S2L). All three are predicted to be integral membrane proteins containing two transmembrane domains [35]. The chat-1 gene in qx36 mutants contained a C to T transition which resulted in a premature stop codon after Leu 94 (Figure S2A). Similar phenotypes were also observed in ok1681, a deletion mutant of chat-1, or with RNAi inhibition of chat-1 activity (Figure S1F, S1G, and S1K). Although 3 different alternatively spliced chat-1 transcripts are predicted (www.wormbase.org), only chat-1a overexpression rescued both the PS asymmetry and membrane trafficking defects of qx36 mutants (Figure S2A-S2G and see below).
To examine CHAT-1 expression, we generated a CHAT-1::CFP translational fusion under control of the chat-1 promoter (Pchat-1chat-1::cfp) which fully rescued the qx36 mutant phenotypes (Figure S2A). CHAT-1 was expressed from early embryogenesis through all larval and adult stages in various cell types where TAT-1 is also expressed (Ptat-1tat-1::yfp) (Figure 1G–1K; data not shown). When specifically expressed in intestine cells, CHAT-1 and TAT-1 were found to colocalize to both apical and basolateral membranes as well as various intracellular structures (Figure 1L; Figure S3A). In chat-1(qx36) mutant intestine, TAT-1::GFP was lost from the plasma membrane and mainly colocalized with the ER-specific marker mCHERRY::TRAM (Figure S3B, S3C). Similarly, in tat-1(qx30) mutant intestine, CHAT-1::GFP was completely trapped in the ER (Figure S3D, S3E). These data are consistent with previous findings that Cdc50 family proteins and P4-ATPases are mutually required for exiting the ER and that TAT-1 and CHAT-1 are predicted to function as a complex [28], [36].
In chat-1 mutants, many large vacuoles accumulated in intestinal cells from early larval stages onwards, a phenotype that was described previously in a tat-1(lf) allele (Figure S1K, S1L; Figure S4A-S4C) [32]. Overexpression of CHAT-1 (and TAT-1) driven by either the endogenous promoter or the intestine-specific vha-6 promoter fully rescued the vacuolation phenotype, indicating that CHAT-1/TAT-1 act cell-autonomously to regulate membrane trafficking processes (Figure S2A, S2H–S2K; data not shown) [32], [37].
We next examined the distribution of various endolysosomal proteins in chat-1 and tat-1 mutants. GFP::RAB-5 labels early endosomes, which appeared as small punctate structures in wild-type animals but became significantly enriched and aggregated in tat-1(qx30) and chat-1(qx36) mutants (Figure 2A–2C, 2P). GFP::RAB-5 was also found on the surface of most abnormal vacuoles (Figure 2B, 2C; Figure S4D–S4F). GFP::RAB-7 labeled a small number of punctate structures and many big ring-like vesicles in wild-type intestine, which likely represent early/late endosomes and early lysosomes, respectively (Figure 2D) [38]. In chat-1(qx36) and tat-1(qx30) mutants, however, GFP::RAB-7-positive puncta significantly increased in number and formed aggregates, with a concomitant reduction of the big ring-like structures (Figure 2E, 2F, 2Q). Many abnormal vacuoles were marked by GFP::RAB-7, and some of them were positive for LMP-1::GFP, a lysosome-associated membrane protein (Figure 2E, 2F; Figure S4G-S4H, -S4R) [32]. The RAB-7 aggregation was also observed in tat-1 and chat-1 mutants stained with anti-RAB-7 antibodies (Figure S5A–S5C). In tat-1(qx30) mutants, RAB-5 and RAB-7 extensively overlapped, whereas they colocalized only to a limited number of small puncta in wild type, suggesting that early to late endosome transport is affected (Figure 2U, 2V). Moreover, GFP::RAB-10, which associates with basolateral early endosomes, the Golgi and apical recycling endosomes, displayed a punctate staining pattern in wild type, but formed a few cytoplasmic aggregates as well as appearing on abnormal vacuoles in chat-1 and tat-1 mutants (Figure 2G–2I, 2R; Figure S4J–S4L) [38]. In addition, RAB-11, which overlaps with RAB-10 on apical recycling endosomes and the Golgi, also formed aggregates and labeled abnormal vacuoles (Figure 2J–2L, 2S; Figure S4M–S4O) [38]. Finally, we examined the distribution of RME-1 which labels basolateral recycling endosomes [13], [38]. In tat-1(qx30) and chat-1(qx36) mutants, the distinct punctate staining pattern of RME-1 along basolateral membranes was significantly reduced and RME-1 either became diffuse or formed large aggregates in the cytoplasm, indicating that RME-1-positive recycling endosomes are disrupted (Figure 2M–2O, 2T). The abrogation of endogenous RME-1 pattern was confirmed by the staining of tat-1 and chat-1 mutants with anti-RME-1 antibodies (Figure S5D–S5F). RME-1 and RAB-10 localized to distinct endocytic compartments in wild type but partially overlapped on cytosolic aggregated vesicles in tat-1(qx30) mutants (Figure 2W, 2X); further suggesting that endocytic trafficking from early to recycling endosomes is defective. Consistent with defective trafficking from early to late endosomes, the number of mature lysosomes strongly labeled by Lysotracker Red significantly decreased in the mutant intestine, whereas ER and Golgi markers appeared normal (Figure S6A–S6J). In epidermis, however, tat-1(qx30) and chat-1(qx36) mutants accumulated enlarged acidic compartments positive for Lysotracker Red, a phenotype which was previously observed in animals carrying the tat-1 loss-of-function allele kr15 (Figure S6K–S6M) [32]. Collectively, our data suggest that endocytic trafficking through early endosomes is severely affected in tat-1 and chat-1 mutants, leading to disruption of recycling and late endosomes.
Because early and recycling endosomes are affected in tat-1 and chat-1 mutants, we next investigated whether endocytic recycling is defective by examining the trafficking of hTfR, the human transferrin receptor (hTfR::GFP), and hTAC, the α-chain of the human IL-2 receptor TAC (hTAC::GFP), both of which are recycled in a RAB-10- and RME-1-dependent manner in the C. elegans intestine [38]. hTfR::GFP accumulated significantly in the cytosol in tat-1 and chat-1 mutants whereas it mainly localized to basolateral membranes in wild type (Figure 3A–3C, 3J). Similarly, increased accumulation of cytosolic hTAC::GFP was also observed in these mutants, albeit to a lesser extent than hTfR (Figure 3D–3F, 3K). These results suggest that recycling of hTfR and hTAC is compromised. Moreover, the glucose transporter 1 (GLUT1), which enters mammalian cells through clathrin-independent endocytosis and constitutively recycles via the Arf6 pathway that require the function of Rab GTPases and RME-1/EHD1 family proteins, primarily localized to apical and basolateral cell membranes when expressed in the C. elegans intestine (Figure 3G) [39]. In tat-1(qx30) and chat-1(qx36) mutants, however, cytosolic accumulation of GLUT1::GFP dramatically increased and some of the signal appeared in abnormal vacuoles, indicating that trafficking of GLUT1 to the plasma membrane was disrupted (Figure 3H, 3I, 3L). Intracellularly accumulated hTfR and GLUT1 were found on vesicles positive for RAB-5, RAB-10 or RAB-7 in chat-1(RNAi) animals, especially on vesicles that clustered together, suggesting that they may be trapped in abnormal early endosomes (Figure 3M–3R, 3U–3Z). Consistent with this notion, hTfR and GLUT1 colocalized with RME-1 on basal and lateral membranes in wild type, but partially overlapped with cytoplasmic RME-1 aggregates in chat-1(RNAi) animals (Figure 3S, 3T, 3Z1, and 3Z2). hTfR or GLUT1 did not significantly overlap with either CHC-1, a marker for clathrin-coated pits, or Lysotracker Red in wild type or chat-1(RNAi) animals (data not shown). These data indicate that loss of tat-1 and chat-1 function results in defective recycling of cargoes, which are mainly trapped within abnormal early endosomes.
As abnormal RAB-7 distribution and fewer mature lysosomes were observed in tat-1 and chat-1 intestines, we examined whether the degradative pathway is also affected by using the VIT-2::GFP reporter to monitor yolk trafficking and accumulation [40], [41]. The initial uptake of yolk in both mature oocytes and fertilized embryos was normal in tat-1 and chat-1 mutants with no obvious accumulation of VIT-2::GFP in the body cavity (Figure 4A–4E). However, the redistribution of yolk to the gut primordium and their degradation appeared to be affected as significantly more VIT-2::GFP was observed in both early and late embryos as well as L1 larvae in tat-1 and chat-1 mutants than in wild type (Figure 4E–4V; Figure S5G–S5I). Moreover, we found that aged tat-1 and chat-1 mutants but not wild-type animals (60 h post L4/adult molt), accumulated a large number of big yolk granules in the intestine (Figure 4W–4Z; Figure S5J–S5L). As tat-1 and chat-1 adults aged for a shorter period of time (12, 24 or 48 h post L4/adult molt) contained a similar level of yolk in the intestine as in wild type, our data are consistent with compromised yolk degradation in these mutants (data not shown). In addition, tat-1 and chat-1 intestines accumulated many LGG-1-postive structures, which were disrupted in animals lacking atg-3, atg-5 or atg-7, suggesting that the degradation of autophagic cargo may also be affected (Figure 4Z1–Z4; data not shown) [42].
To understand how disruption of CHAT-1 and TAT-1 function results in endocytic defects, we examined their subcellular localization in the intestine by coexpressing CHAT-1::CFP and TAT-1::YFP under control of the intestine-specific promoter vha-6 (Pvha-6chat-1::cfp +Pvha-6tat-1::yfp). These reporters fully rescued the vacuolation phenotype in tat-1 and chat-1 mutants (Figure S2A). TAT-1 and CHAT-1 colocalized to both plasma membranes and intracellular tubular and vesicular structures (Figure 1L; Figure S3A). The intestinal tubular and vesicular localization pattern was also observed when the expression of TAT-1 or CHAT-1 was controlled by the endogenous promoter (Figure S7A, S7B; data not shown). To determine the identities of the cytosolic compartments labeled by CHAT-1 and TAT-1, we coexpressed mCHERRY fusions of different endocytic markers together with CHAT-1::GFP and TAT-1 (Pvha-6chat-1::gfp +Pvha-6tat-1). TAT-1 was included to ensure efficient ER export of CHAT-1::GFP; this combination is subsequently referred to as CHAT-1::GFP for simplicity. CHAT-1::GFP displayed a tubular and vesicular staining pattern, which did not overlap with either the Golgi marker MANS or the lysosomal marker Lysotracker Red, indicating that CHAT-1 is not on the Golgi or mature lysosomes (Figure 5E, 5F). No CHAT-1::GFP was found on the RAB-7-positive ring-like structures, suggesting that it is not enriched on late endosomes or early lysosomes (Figure 5D). However, CHAT-1 partially overlapped with RAB-5 on punctate structures, but not on tubule-like structures that were negative for mCHERRY::RAB-5 (Figure 5A). Thus, a proportion of CHAT-1/TAT-1 may localize to RAB-5-positive early endosomes. We next examined colocalization of CHAT-1 and RAB-10, which associates with endosomes and Golgi compartments [38]. Interestingly, when coexpressed with CHAT-1::GFP, RAB-10 displayed a different staining pattern and mainly localized to CHAT-1-positive tubular structures instead of labeling small cytoplasmic puncta (compare Figure 2G and Figure 5B). Similarly, RAB-11 labeled punctate structures when expressed alone, but colocalized with CHAT-1 on abundant tubules when the two were coexpressed (compare Figure 2J and Figure 5C). Moreover, in animals expressing both mCHERRY::RAB-10 (or RAB-11) and CHAT-1::GFP, the tubular structures became more evident and extensive than in animals carrying only CHAT-1::GFP or animals coexpressing CHAT-1 and RAB-5 or RAB-7 (compare B and C with other panels of Figure 5). These data suggest that RAB-10 and RAB-11 may act together with CHAT-1/TAT-1 to promote extension of the tubular structures.
As CHAT-1 tubules were labeled by RAB-10 and RAB-11, but not RAB-5 or RAB-7, we reasoned that they may be tubular early endosomes and/or endocytic recycling compartments (ERCs). We first examined the tubular structures in rab-10(lf) mutants, in which endocytic transport from early to recycling endosomes is disrupted, and found that CHAT-1 tubules were totally abolished; instead, CHAT-1::GFP overlapped completely with RAB-5 on enlarged early endosomes (Figure 6B, 6E). Inactivation of rab-10 also disrupted tubules labeled by CHAT-1 and RAB-11 (Figure S7E). Next, we examined the CHAT-1 tubules in rme-1(lf) mutants, in which trafficking from recycling endosomes to the plasma membrane is affected, and found that the tubules were not disrupted (Figure 6C). Instead, CHAT-1- and RAB-10- (or RAB-11) positive tubular structures became even more extended when rme-1 function was lost (Figure 6G; Figure S7D). Conversely, loss of rab-5 activity completely abrogated the tubular structures in either wild type or rme-1(b1045) mutants (Figure 6H; data not shown). RAB-5 and RAB-10 are required in early endosomes and for trafficking from early to recycling endosomes, while RME-1 acts downstream of them to promote membrane fission for releasing recycling carriers [14], [38]. Our data are consistent with the idea that CHAT-1/TAT-1 associates with the tubular membrane of early endosomes and ERCs. Finally, we examined the co-localization of CHAT-1 and RME-1, which is enriched on recycling endosomes [13]. We observed that CHAT-1::GFP overlapped with mRFP::RME-1 on basolateral tubulo-vesicular structures, indicating that CHAT-1 also localizes to RME-1-positive recycling endosomes (Figure 7A).
The formation of tubular extensions in early endosomes, ERCs and recycling endosomes is crucial for sorting and transporting recycling cargoes. We observed that the recycling cargo GLUT1 labeled CHAT-1-positive tubules near basolateral membranes and in the cytoplasm, supporting a role for these tubules in sorting and recycling (Figure 7E; data not shown). GLUT1::GFP was also found on RME-1-positive tubulo-vesicular recycling endosomes (Figure 7F). The tubular membrane structures containing GLUT1::GFP were completely disrupted in tat-1 and chat-1 mutants, in which GFP stained cytoplasmic punctate structures (Figure 7G–7I). Moreover, in tat-1 and chat-1 mutants, the remaining basolateral RME-1-positive puncta completely lost their tubulo-vesicular morphology and became globular (Figure 7B–7D). These data indicate that TAT-1 and CHAT-1 are required for forming and/or maintaining the tubular extensions of sorting and recycling compartments.
To investigate whether TAT-1 and CHAT-1 regulate endocytic sorting and recycling by maintaining membrane PS asymmetry, we examined PS asymmetry across endomembranes. We first determined PS distribution by expressing the biosensor GFP::Lact-C2 specifically in intestine cells (Pges-1GFP::Lact-C2) [43]. The cell membranes and surfaces of virtually all internal vesicles were labeled, indicating that PS was exposed on the cytosolic surface of both plasma membranes and various intracellular compartments (Figure S8A). We next coexpressed GFP::Lact-C2 with different endolysosomal markers and found that it labeled intracellular structures that were positive for RAB-5, RAB-7, RAB-10, RME-1 or Lysotracker Red (Figure 8A–8E). Thus, PS appeared on the cytosolic surfaces of recycling, early and late endosomes as well as lysosomes. mCHERRY::Lact-C2 coincided well with CHAT-1::GFP on tubular membranes, indicating that the tubules are coated by PS (Figure 8F).
In coelomocytes, which are scavenger cells that actively endocytose and degrade soluble material, GFP::Lact-C2 (Punc-122GFP::Lact-C2) stained plasma membranes and the surfaces of endosomes and lysosomes, as observed in intestine cells (Figure S8D, S8E). To determine whether PS is absent from the luminal leaflet of endomembranes in wild type and whether PS asymmetry is affected in tat-1 and chat-1 mutants, we focused on coelomocytes, which contain abundant endocytic vesicles that are larger than those in intestine cells. To detect luminal PS in endocytic vesicles, we examined expression of ssGFP::Lact-C2 driven by the myo-3 promoter (Pmyo-3ssGFP::Lact-C2), which after secretion from body wall muscle cells is taken up by coelomocytes through endocytosis and transferred to lysosomes via endocytic transport [44]. In wild-type coelomocytes, the endocytosed GFP::Lact-C2 mainly accumulated in lysosomes as indicated by an endocytic cargo ssCHERRY, whereas no clear GFP signal was detected in endosomes, suggesting that PS is likely absent from the luminal side of endomembranes (Figure 8G; Figure S8F, S8G). Therefore, like the plasma membrane, PS is preferentially distributed on the cytosolic side of endomembranes. Remarkably, we found that in both tat-1(qx30) and chat-1(qx36) mutants, the internalized GFP::Lact-C2 labeled endosome membranes with a ring-like staining pattern, indicating that PS appeared on the luminal side of endomembranes (Figure 8H, 8I). By contrast, expression of GFP::Lact-C2(AAA), which is deficient in PS binding, gave no or very faint and diffuse GFP signal in endosomes (Figure S8H–S8J). These data suggest that in tat-1 and chat-1 mutants, PS asymmetry across endomembranes is disrupted, causing PS to appear on both cytosolic and luminal leaflets of the membrane.
From a genetic screen for altered distribution of a PS-binding protein, we recovered mutant alleles of chat-1 and tat-1, which displayed identical PS asymmetry phenotypes on both plasma membranes and endomembranes. Mutants of chat-1 and tat-1 also showed identical endocytic defects. CHAT-1 and TAT-1 are expressed in the same tissues and colocalize to both plasma membranes and various intracellular compartments in the intestine. Furthermore, CHAT-1 and TAT-1 are co-dependent for exiting the ER, similar to P4-ATPase and Cdc50p proteins in yeast and mammalian cells [28], [36]. Therefore, like Drs2p and Cdc50p, TAT-1 and CHAT-1 may act as a complex to regulate membrane PS asymmetry and endocytic traffic.
We observed pleiotropic phenotypes associated with the presence of multiple endocytic vesicles in tat-1 and chat-1 mutants, many of which can be attributed to defective endocytic sorting. For example, in tat-1 and chat-1 mutants, enlarged early endosomes accumulated while recycling and late endosomes were disrupted. The abnormal vacuoles appear to be heterogeneous, since they were labeled by markers of early, late and recycling endosomes as well as early lysosomes [32]. Cargo recycling and degradation are also defective in these mutants. Therefore, loss of tat-1 and chat-1 function likely disrupts endocytic sorting through early endosomes, thereby affecting subsequent trafficking through both recycling and degradative pathways.
In a recent study, TAT-1 was found to be required at an early step of endocytosis [32]. Consistent with this, we observed a defect in endocytosis of fluid cargo from both basolateral and apical intestinal cell membranes in tat-1 and chat-1 mutants (data not shown). However, defective endocytosis was not seen in oocytes when yolk was taken up. Instead, we observed defects in yolk redistribution and digestion in tat-1 and chat-1 embryos (Figure 4).
How is endocytic sorting through early endosomes regulated by the P4-ATPase TAT-1 and its chaperone CHAT-1? The tubular elements of early endosomes serve as sorting platforms to enrich and transport transmembrane cargoes. Several lines of evidence indicate that TAT-1 and CHAT-1 are required for generating and/or maintaining the tubular membrane structure. Firstly, tat-1 and chat-1 mutants disrupt endocytic transport through early endosomes, leading to the accumulation of enlarged early endosomes positive for RAB-5, which labels the vesicular but not the tubular element. Secondly, TAT-1 and CHAT-1 associate with tubular membranes at sorting and recycling compartments, which contain the recycling cargo GLUT1. Thirdly, loss of tat-1 and chat-1 function abolishes the tubular membrane structure containing GLUT1 and disrupts the tubulo-vesicular morphology of RME-1-positive recycling endosomes.
Phospholipids can have significant effects on membrane curvature when their distributions between the two membrane leaflets are altered. It was observed that addition of exogenous phospholipids including PS to the outer leaflet of discoid platelets caused expansion of the outer surface in the form of numerous extensions [45]. In another case, incorporation of PS, phosphatidylethanolamine and phosphatidylcholine into the outer leaflet of discoid erythrocytes increased the outer membrane surface area and induced a crenated shape with a higher ratio of external to internal surface area, whereas transverse diffusion of exogenous phospholipids from the outer to the inner leaflet reversed the shape change [17], [45]. Our findings that CHAT-1-assoicated tubules are coated by PS and that loss of tat-1 and chat-1 function disrupts PS asymmetry of endomembranes and abrogates tubular extensions strongly suggest a role of PS and/or PS asymmetry in membrane tubulation. As an aminophospholipid transporter, TAT-1/CHAT-1 may catalyze the active translocation of PS from the luminal to the cytosolic leaflet of endomembranes, which results in a high ratio of PS in the cytosolic leaflet versus the inner leaflet and an increased outer monolayer area, leading to the deformation of membranes into tubular extensions [46] (Figure 9). Consistent with this, we observed extensive tubular membrane structures labeled by the PS biosensor mCHERRY::Lact-C2 in animals overexpressing CHAT-1 and TAT-1.
Our data suggest that RAB-10 and RAB-11 may also contribute to tubule formation or extension as both of them, when overexpressed, become enriched on CHAT-1::GFP-positive tubules and enhance tubule extension, whereas loss of rab-10 function completely disrupts tubular structures and traps CHAT-1 on enlarged RAB-5-positive early endosomes (Figure 9). However, we do not know whether RAB-10 plays a direct or indirect role in this process, or how its function is achieved. One interesting possibility is that RAB-10 and RAB-11 may contribute to tubule formation by regulating membrane-bending proteins, as BAR proteins are found to interact with a variety of effectors of small GTPases including Rabs [9].
RME-1 and the BAR protein AMPH-1 were recently shown to cooperatively regulate endocytic recycling in C. elegans intestine and to tubulate liposomes in vitro [14]. We found that PS-coated RME-1-positive recycling endosomes lost their tubulo-vesicular morphology in tat-1 and chat-1 mutants, suggesting that TAT-1/CHAT-1-regulated PS translocation may be involved in membrane tubulation mediated by RME-1/AMPH-1 (Figure 9). In addition, our observation that CHAT-1-RAB-10 (or RAB-11) tubules are further extended in rme-1(lf) mutants is consistent with the proposed function of EHD/RME-1 proteins in mediating membrane fission to generate recycling carriers [14], [16].
Our data support a role of TAT-1/CHAT-1 in membrane tubulation for cargo sorting and recycling. However, we do not know how the sorting of cargo for degradation may be affected by loss of tat-1 and chat-1 function. Interestingly, it was reported that tat-1(lf) mutants accumulate giant multi-vesicular bodies (MVBs) in hypodermal cells [32]. As translocation of PS and PE across the membrane bilayer is thought to provide the driving force for membrane bending [17], [47], [48], we suspect that loss of TAT-1 and CHAT-1 function may lead to defective budding from multi-vesicular compartments of early endosomes, thereby affecting the sorting of cargo into the degradative pathway. Further experiments need to be performed to test this hypothesis, especially the examination of MVBs in intestine cells.
Strains of C. elegans were cultured and maintained using standard protocols [49]. The N2 Bristol strain was used as the wild-type strain except for polymorphism mapping which used Hawaiian strain CB4856. Mutations used are described in C. elegans II [50] unless otherwise indicated. Linkage group I (LGI): rab-10(dx2) [38]. LGIII: dpy-18 (e364am), bli-5(e518), tat-1(qx30), tat-1(qx23), tat-1(qx24), tat-1(tm3110) (this study). LGIV: dpy-9(e12), unc-17(e245), dpy-13(e184sd), unc-8(n491sd), chat-1(qx36), chat-1(ok1681) (this study). LGV: unc-62(e644), dpy-11(e224), rme-1(b1045) [13]. LGX: lon-2(e678), unc-27(e155).
bIs34 (pcc1RME-8::GFP) [51], cdIs73 (pcc1RME-8::mRFP) [52] and cdIs97 (pcc1mCHERRY::CUP-5) [53] were kindly provided by Dr. Hanna Fares (University of Arizona, Tucson, AZ). pwIs216 (Pvha-6mRFP::RME-1), pwIs112 (Pvha-6hTAC::GFP) [38], and pwIs717 (Pvha-6hTfR(short)::GFP) [14] were kindly provided by Dr. Barth Grant (Rutgers University, Piscataway, NJ). yqIs25 (Plgg-1LGG-1::GFP) was kindly provided by Dr. Chonglin Yang (Institute of Genetics and Developmental Biology, Beijing).
Other strains carrying integrated or transgenic arrays used in this study are as follows:
pwIs50 (Plmp-1LMP-1::GFP) [52], bIs1 (VIT-2::GFP) [40], qxIs162 (Pges-1mCHERRY::TRAM), qxIs110 (Pges-1mCHERRY::RAB-5), qxIs111 (Pges-1mCHERRY::RAB-7), qxIs213 (Pges-1mCHERRY::RAB-10), qxIs195 (Pges-1GFP::RAB-10), qxIs154 (Punc-122GFP::Lact-C2), qxIs165 (Pges-1GFP::Lact-C2), qxIs150 (Pchat-1CHAT-1::CFP), qxIs188 (Ptat-1TAT-1::YFP +Pchat-1CHAT-1::CFP), qxIs143 (Ptat-1TAT-1::GFP), qxEx1753 (Pvha-6CHAT-1::GFP), qxEx2966 (Pvha-6TAT-1::YFP + Pvha-6CHAT-1::CFP), qxEx2622 (Pvha-6CHAT-1::GFP + Pvha-6TAT-1), qxEx1736 (PhspssGFP::Lact-C2), qxEx1999 (PhspssGFP::Lact-C2(AAA)), qxEx1398 (Pmyo-3ssGFP::Lact-C2), qxEx1533 (Pmyo-3ssGFP::Lact-C2(AAA)), qxEx2841 (Pvha-6GFP::RAB-5), qxEx2616 (Pvha-6GFP::RAB-7), qxEx1317 (Pges-1GFP::RAB-11), qxEx2247 (Pvha-6GLUT1::GFP), qxEx2726 (PhspssCHERRY), qxEx1207(Ptat-1MANS::mCHERRY), qxEx1914 (Pvha-6mCHERRY::Lact-C2), qxEx1867 (Pvha-6mCHERRY::RAB-11), qxEx3265 (Pvha-6GLUT1::CFP + Pvha-6CHAT-1::YFP + Pvha-6TAT-1).
TTR-52 is a secreted protein that specifically recognizes apoptotic cells through its binding to surface exposed phosphatidylserine (PS) [34]. In wild-type embryos carrying PhspTTR-52::mCHERRY, apoptotic cells are surrounded by mCHERRY, which is absent from the surface of living cells. In order to understand how the PS engulfment signal is regulated, we performed a forward genetic screen to look for mutants which disrupt or alter the staining of apoptotic cells by TTR-52::mCHERRY. From this screen, we isolated the qx36 mutant and several alleles of tat-1, which resulted in TTR-52::mCHERRY staining of virtually all cells, both dying and living [34].
qx36 was mapped to linkage group IV. Two rounds of three-point mapping were performed using unc-17 (−3.11) dpy-13 (0.00) and dpy-13 (0.00) unc-8 (+3.29), which mapped qx36 to a small genetic interval between 0.00–1.02. Transformation rescue experiments were performed and one fosmid clone in this region, WRM0637aA04, rescued the qx36 defect. Long PCR fragments covering different open reading frames within this fosmid were tested and only the fragment covering R08C7.2 possessed rescue activity. R08C7.2 encodes a Cdc50p-like protein of 348 amino acids, which we named chat-1 (chaperone of tat-1). Sequencing of the locus in the qx36 mutant identified a C to T transition, which resulted in a premature stop codon after Leu 94. Given that similar phenotypes were observed in ok1681, a chat-1 deletion mutant containing an 1152 bp deletion that removes the region from exon 2 to intron 5 of the chat-1 gene, or when chat-1 is inactivated by RNAi, qx36 is probably a null or strong loss-of-function mutation of chat-1.
qx30 was mapped to the right arm of linkage group III (LGIII). Further three-factor mapping and subsequent single nucleotide polymorphism (SNP) mapping were performed using dpy-18 (8.65) bli-5 (21.21) and various SNP markers. qx30 was mapped between genetic map positions 17.26 (Snp-Y49E10(2)) and 17.85 (Snp-Y111B2(9)). Since the tat-1 gene locates at 17.59 of LGIII, a transformation rescue experiment was performed and a DNA fragment containing the tat-1 gene fully rescued the qx30 defect. The sequence of the tat-1 gene was determined in qx30 and three other mutants, qx22, qx23, qx24, which are allelic to qx30. The qx22 and qx23 mutants carry the same mutation that results in substitution of the Ala at codon 627 with Val, whereas the qx24 mutant has a splicing mutation (G to A transition) at the junction of intron 12 and exon 13. The qx30 mutant has a G to A transition that results in a premature stop codon after Leu 1028.
To inactivate tat-1 and chat-1 by RNAi, dsRNA synthesized in vitro (550 ng/µl) was injected into the gonad of young adult hermaphrodites (see Table S1 for primer sequences). Embryos laid between 16 to 24 h post-injection were either used for analyzing PS asymmetry or cultured until the L4 larval or young adult stage for examining intestinal phenotypes. We found that tat-1 RNAi significantly diminished the expression of tat-1. For example, 65% of embryos (n = 40) transgenic for Ptat-1tat-1::gfp had bright GFP fluorescence before injection, but only 4% of them showed similar GFP intensity after injection. RNAi treatment of tat-1 also resulted in a 56% reduction of TAT-1::GFP expression in larvae (n = 40). Similarly, 100% of animals carrying an integrated Pchat-1chat-1::gfp reporter had strong GFP fluorescence before injection, but after chat-1 RNAi treatment, 99% had only weak GFP signal (n = 30). For rab-5 and rab-10 RNAi, the bacterial feeding protocol was used as described before [54]. Briefly, L3 larvae were treated with either rab-5 RNAi (I-4J01) or control RNAi (pPD129.36) and adult animals of the same generation were scored as most F1 progeny die due to inactivation of rab-5. For rab-10 RNAi, L3 or L4 larvae were treated with either rab-10 RNAi (I-3G23) or control RNAi (pPD129.36) and F1 progeny were examined at the adult stage.
The ex vivo staining of dissected gonads by Annexin V was performed as described before [55]. To determine membrane PS asymmetry in embryos, mixed stage animals carrying PhspssGFP::Lact-C2 or PhspssGFP::Lact-C2(AAA) were incubated at 33°C for 1 h followed by recovery at 20°C for 2.5 h before examination. To examine PS asymmetry across endomembranes in coelomocytes, mixed stage P0 animals carrying both Pmyo-3ssGFP::Lact-C2 and PhspssCHERRY were incubated at 33°C for 0.5 h to induce the expression and secretion of ssCHERRY. Culture was continued at 20°C for one more generation before F1 adults were examined. This long incubation allows the complete uptake and transport of ssCHERRY to lysosomes, which otherwise is seen in the body cavity and many tissues other than coelomocytes. To examine yolk uptake in oocytes, L4 hermaphrodites from the strains indicated were aged for 12 and 24 h and confocal images were taken with the same exposure time. The uptake of yolk was observed at the same two time points in wild type and tat-1 and chat-1 mutants. Images of animals that were aged to 24 h post L4/adult molt are shown in Figure 4. To examine yolk accumulation in the intestine, L4 larvae were aged for 12, 24, 48, and 60 h and images were taken with the same exposure time. No obvious difference in yolk accumulation was observed in wild-type, tat-1 or chat-1 intestines when animals were aged less than 60 h.
Intracellular accumulation of hTfR::GFP, hTAC::GFP and GLUT1::GFP was quantified by determining the number of labeled structures within a 250 µm2 area in the intestine. Lysotracker Red-positive granules and mRFP::RME-1-positive vesicles were counted within a unit area of 150 µm2. LGG-1 puncta and GFP::RAB-10- or GFP::RAB-11-labeled aggregated structure (clustering of >2 labeled puncta) were scored within a unit area of 500 µm2. Five different areas were chosen and quantified in each animal at the L4 larval or young adult stage and 8 animals were scored for each strain. The average total intensity per unit area of GFP::RAB-5 and GFP::RAB-7 in the intestine and VIT-2::GFP in fertilized early embryos was measured using Image J 1.42q software. For GFP::RAB-5 and GFP::RAB-7, 5 different areas (40 µm2 each) in the intestine of L4 larvae were chosen for each animal and 8 animals were quantified. For VIT-2::GFP in fertilized early embryos, 3 different regions (20 µm2 each) were chosen for each embryo and 32 embryos were scored. Axiovision Rel. 4.7 software (Carl Zeiss, Inc.) was used to quantify average total intensity per unit area of VIT-2::GFP in 4-fold stage embryos or in the intestine of aged adult (60 h post L4/adult molt). 3 different regions (12.6 µm2 each) were chosen for each animal and 32 embryos or 28 intestines were quantified. Student's two tailed unpaired t-test was performed and the P value was indicated in the figure legend.
CHAT-1(74-306) or full-length RAB-7 protein tagged with six Histidine residues (CHAT-1(74-306)-His6 or RAB-7-His6) was expressed in and purified from E. coli and used to raise rat polyclonal antibody against CHAT-1 or RAB-7. The antibodies were further purified by incubating 3 ml rat serum with nitrocellulose membrane strips containing 5 mg CHAT-1(74-306)-His6 or RAB-7-His6 protein. Bound antibodies were eluted with 100 mM glycine-HCl (pH 2.5). Purified anti-RAB-7 antibody recognized a single band of expected size (24 KD) in a western blot analysis using lysate prepared from mixed staged wild-type worms, while purified anti-CHAT-1 antibody failed to detect endogenous CHAT-1 expression in wild-type animals, but recognized CHAT-1::GFP (64 KD) or CHAT-1- His6 (39 KD) when the western blot was performed using lysate prepared from transgenic animals expressing CHAT-1::GFP or CHAT-1-His. In a whole-mount immunostaining experiment, anti-CHAT-1 antibody stained plasma membranes in wild-type but not chat-1(qx36) embryos or oocytes. RAB-7 antibody staining revealed a specific pattern reminiscent of GFP::RAB-7 in the intestine of wild-type but not rab-7 RNAi-treated animals except for the staining of apical membranes which appears to be non-specific. Monoclonal anti-GFP antibody was purchased from Roche (USA) and anti-RME-1 antibody was obtained from Developmental Studies Hybridoma Bank (University of Iowa, USA). For immunostaining, mixed stage embryos or dissected intestines were fixed with methanol/acetone followed by blocking in phosphate buffered saline (PBS) containing 1% BSA and 10% fetal calf serum for 2 h at 4°C. The samples were then incubated with primary antibodies in blocking buffer at 4°C overnight with 1:50 dilution for RAB-7, RME-1, CHAT-1 antibodies, 1:200 for GFP antibody and 1:500 for LGG-1 antibody. After washing three times with PBST (PBS + 0.2% Tween 20), the samples were incubated with secondary antibody conjugated to Alexa-488 or Alexa-546 (Molecular Probes) at a 1:50 dilution for 2 h at room temperature. The stained samples were washed three times as before and mounted in 15% VECTASHIELD mounting medium with DAPI (VECTOR) and visualized using a Zeiss LSM 510 Meta inverted confocal microscope.
DIC and fluorescent images were captured with a Zeiss Axioimager A1 equipped with epifluorescence and an AxioCam monochrome digital camera and were processed and viewed using Axiovision Rel. 4.7 software (Carl Zeiss, Inc.). A 100x Plan-Neofluar objective (NA1.30) was used with Immersol 518F oil (Carl Zeiss, Inc.). For confocal images, a Zeiss LSM 5 Pascal inverted confocal microscope with 488, 543, 514, 458 and 405 lasers was used and images were processed and viewed using LSM Image Browser software.
The sequences of the PCR primers mentioned in this section are presented in 1. Ptat-1tat-1::gfp was constructed by ligating the tat-1 minigene (tat-1A) containing 2.4 kb of the promoter sequence digested from Ptat-1tat-1::flag to pPD49.26-gfp2 through the Hind III and Nhe I sites [31]. To generate Ptat-1tat-1::yfp, the 2.4 kb promoter sequence of tat-1 was first amplified using primers PJCY16/17 and cloned into pPD49.26-yfp2 via the Hind III/Xma I sites. The resulting Ptat-1yfp was then ligated with full length tat-1 cDNA (tat-1A) amplified with primers PBHC154/PJCY4 through the Nhe I site. To construct Pvha-6tat-1, the tat-1 minigene (tat-1A) was amplified using primers PBHC123/124 and cloned into Pvha-6 through the Kpn I site. Pvha-6tat-1::yfp was generated by fusion PCR with three pairs of primers: (1) PPFG211/PWZ263 for amplifying the vha-6 promoter, (2) PWZ625/PWZ512 for amplifying sequences of the tat-1 minigene (tat-1A), yfp and the 3′UTR of the unc-54 gene and (3) PWZ624/PPFG199 for the final round of PCR amplification to create Pvha-6tat-1::yfp [56]. To generate Pchat-1chat-1a (Pchat-1chat-1), the 2 kb promoter sequence of chat-1 was amplified using primers PJCY21/9 and cloned into pPD49.26 through the Sph I/Xma I sites. The resulting Pchat-1 was then ligated with either the genomic fragment of chat-1a amplified by PJCY18/34 or the 1.1 kb genomic sequence of chat-1b amplified by PJCY18/35 through the Nhe I site to generate Pchat-1chat-1a and Pchat-1chat-1b. Phspchat-1c was constructed by ligating the 1.2 kb genomic fragment of chat-1c amplified using primers PJCY20/PYJ7 to pPD48.78 and pPD49.83 through the Nhe I and Nco I sites. To construct Pchat-1chat-1cfp, the 2 kb promoter sequence of chat-1 was first cloned into pPD49.26-cfp2 via the Sph I and Xma I sites, which was then ligated with the genomic fragment of chat-1 amplified with PJCY18/6 through the Nhe I site. The genomic fragment of chat-1 was similarly cloned into Pvha-6, Pvha-6cfp2 and Pvha-6gfp2 through the Nhe I site to generate Pvha-6chat-1, Pvha-6chat-1::cfp and Pvha-6chat-1::gfp, respectively. To construct various endocytic markers for specific expression in the intestine, genomic fragments (rab-5, rab-7, rab-10, rab-11.1) were PCR-amplified and cloned into Pvha-6gfp1, Pvha-6mcherry1, Pges-1gfp1 or Pges-1mcherry1 through the Kpn I, Kpn I/EcoR V (mcherry::rab-7) or Kpn I/Sma I (mcherry::rab-5) sites. The sequences of ER (TRAM: translocating chain associating membrane protein) and Golgi (MANS: mannosidase short) reporters were determined as described before [57], and amplified using primers PYJ44/45 and PYJ40/53, respectively. The resulting PCR fragments were then cloned into Pges-1mcherry1 and Pges-1mcherry2 to create Pges-1mcherry::tram and Pges-1mans::mcherry, respectively. To generate the GLUT1::GFP reporter, a 2.5 kb genomic fragment of R09B5.11 was amplified with primers PBHC204/205 and cloned into Pvha-6gfp2 through the Nhe I site. To generate GFP::Lact-C2 constructs for specific expression in coelomocytes and intestine, a Lact-C2 fragment was amplified from PVM-LACT-1 (Haematologic Technologies Inc.) using primers PBHC144/145 or PBHC147/146 and cloned into Punc-122gfp1 and Pges-1gfp1 through the EcoR V/Sac I and Kpn I sites, respectively. To construct a secreted GFP::Lact-C2, a Lact-C2 fragment was amplified with PBHC146/147 and cloned into Phspgfp1 or Pmyo-3gfp1 through the Kpn I site. The resulting Phspgfp::lact-c2 and Pmyo-3gfp::lact-c2 were then ligated with a 102 bp fragment containing a synthetic secretion signal amplified from pPD95.85 using primers PYZ203/206 through the Nhe I/Spe I sites to create Phspssgfp::lact-c2 and Pmyo-3ssgfp::lact-c2. The secretion signal was also cloned into Phspcherry to get Phspsscherry. To create the secreted Lact-C2(AAA) construct, point mutations (W26A, W33A, F34A) were introduced by site-directed mutagenesis (QuickChange; Stratagene, USA) into PVM-LACT-1 [19], which was further cloned into Phspgfp1 and Pmyo-3gfp1 using similar strategies as described above.
|
10.1371/journal.pntd.0002853 | Clinical and Parasitological Protection in a Leishmania infantum-Macaque Model Vaccinated with Adenovirus and the Recombinant A2 Antigen | Visceral leishmaniasis (VL) is a severe vector-born disease of humans and dogs caused by Leishmania donovani complex parasites. Approximately 0.2 to 0.4 million new human VL cases occur annually worldwide. In the new world, these alarming numbers are primarily due to the impracticality of current control methods based on vector reduction and dog euthanasia. Thus, a prophylactic vaccine appears to be essential for VL control. The current efforts to develop an efficacious vaccine include the use of animal models that are as close to human VL. We have previously reported a L. infantum-macaque infection model that is reliable to determine which vaccine candidates are most worthy for further development. Among the few amastigote antigens tested so far, one of specific interest is the recombinant A2 (rA2) protein that protects against experimental L. infantum infections in mice and dogs.
Primates were vaccinated using three rA2-based prime-boost immunization regimes: three doses of rA2 plus recombinant human interleukin-12 (rhIL-12) adsorbed in alum (rA2/rhIL-12/alum); two doses of non-replicative adenovirus recombinant vector encoding A2 (Ad5-A2) followed by two boosts with rA2/rhIL-12/alum (Ad5-A2+rA2/rhIL12/alum); and plasmid DNA encoding A2 gene (DNA-A2) boosted with two doses of Ad5-A2 (DNA-A2+Ad5-A2). Primates received a subsequent infectious challenge with L. infantum. Vaccines, apart from being safe, were immunogenic as animals responded with increased pre-challenge production of anti-A2-specific IgG antibodies, though with some variability in the response, depending on the vaccine formulation/protocol. The relative parasite load in the liver was significantly lower in immunized macaques as compared to controls. Protection correlated with hepatic granuloma resolution, and reduction of clinical symptoms, particularly when primates were vaccinated with the Ad5-A2+rA2/rhIL12/alum protocol.
The remarkable clinical protection induced by A2 in an animal model that is evolutionary close to humans qualifies this antigen as a suitable vaccine candidate against human VL.
| Human visceral leishmaniasis causes significant morbidity and mortality, constituting an important global health problem. Absence of safe and cost effective anti-leishmanial drugs, together with emergence of drug resistance and HIV co-infection have posed a serious challenge to the disease containment. Given the urgent need to prevent approximately 0.2 to 0.4 million new VL cases annually worldwide, all reasonable efforts to achieve a safe and effective Leishmania vaccine should be made. We have previously reported the protective properties of the rA2 protein against experimental L. infantum infections both in mice and canines. To further evaluate the efficacy of A2 in a more relevant animal model to human disease, we used the primate Macaca mulatta. Primates vaccinated with different rA2-based prime-boost regimes displayed varying degrees of protective immunity, as indicated by a marked reduction of symptoms and parasite burden in the liver. In particular the vaccination approach with non-replicative adenovirus vector expressing A2 (rAd5-A2) and boosted with the rA2 protein resulted in a more efficient control of parasites as well as resolution of hepatic immune granulomas at 24 weeks post-infection. The clinical efficacy provided by A2 in an animal model that is evolutionary close to humans qualifies this antigen as a promising candidate vaccine against human VL.
| Human VL is a severe systemic disease caused by protozoan parasites of the Leishmania donovani complex [1]. It remains one of the major infectious diseases primarily affecting some of the poorest regions of the world, with an estimated occurrence of approximately 0.2 to 0.4 million new cases of clinical VL annually worldwide, in addition to an estimated 20,000 to 40,000 VL deaths per year. At present, VL occurs in at least 83 countries or territories, but more than 90% of the global human cases were recorded in India, Bangladesh, Sudan, South Sudan, Ethiopia and Brazil. Although recognition of the geographic distribution of VL and its prevalence has increased during recent years, the disease is still grossly underreported [2]. Furthermore, most infections with the visceralizing Leishmania spp. remain asymptomatic or sub-clinical [3]–[5]. Frank disease (also known as kala-azar) is characterized by prolonged fever, diarrhea, hepatosplenomegaly, weight loss, and even death, if left untreated [6]. In addition to be partially influenced by the genetic background [7], [8], other risk factors such as young age, malnutrition, and immunosuppression [9]–[11] are important determinants of host susceptibility to VL. Chemotherapy is toxic and expensive, and a limited number of anti-Leishmania agents are available, to which drug resistance is documented [12], [13]. In addition, no proven successful vaccine for controlling human VL is in routine use [14].
The epidemiology of this disease is complex and can be altered by changes at any point in the transmission cycle that is formed by humans, the reservoir hosts and the phlebotomine sand fly vectors. In some parts of both the Old and New World, transmission occurs mainly in the peridomestic setting, where domestic dogs serve as primary reservoir host of L. infantum (syn. L. chagasi). Hence, measures employed to control zoonotic VL include mass elimination of seropositive dogs, but the impact of euthanasia programs on human and canine VL incidence is doubtful in theoretical and practical grounds [12], [15]. In other cases, the parasite is transmitted from human to human via infectious sand fly bites, as for L. donovani VL in India and Bangladesh and during epidemic spread in the East African region [2]. Thus, strategies employed to control anthroponotic VL have focused on active case detection and treatment and use of insecticide-impregnated materials [13]. However, a sustainable prevention of the disease using these control measures is costly and usually fails in developing countries [12], [13]. Nevertheless, most experts believe that prophylactic or possibly post-exposure vaccination will be essential for ultimate control of the disease [14], [16].
Several Phase III clinical trials testing crude vaccine approaches have given conflicting results [17]. Overall, the results vary from 0 to 75% efficacy against CL and little (< 6%) or no protection against VL [16]. Although host genetics can have dramatic effects on T-cell responses to existing vaccines [18], technical problems (including changes in the quality, stability and potency of the antigens) may provide explanation for some of the variation in efficacy observed in those human vaccine studies. To circumvent these obstacles, many recombinant vaccines using either subunit proteins in adjuvants, naked DNA and live vectors encoding genes for specific antigens have been tested for immunogenicity and protective efficacy in animal models of leishmaniasis [16].
In addition to crude parasite extracts, partially purified fractions containing secreted proteins of Leishmania or the Fucose Manose Ligant (FML) were shown efficacious and are currently used as commercial vaccines for canine VL [19], [20]. In addition, recombinant antigens such as A2, LACK, Cysteine Proteases A and B, or multicomponent vaccines including KMP-II, TRYP and GP63 or LeIF, LmSTI1 and TSA antigens have shown some level of protection in pre-clinical trials. A comprehensive list of the antigens along with immune responses and protection of respective trials are described in detail elsewhere [21]. Among the recombinant antigens selected as candidates for a prophylactic vaccine against VL, one of specific interest is the amastigote specific antigen A2 from L. donovani [22], [23]. The recombinant A2 (rA2) conferred protection in mice challenged with L. donovani, L. infantum or L. amazonensis when administered as recombinant protein [22], [24], DNA [25], viral vector [26], or transfected parasites (L. tarentolae) [27].
In the form of a currently licensed veterinary product (designated Leish-Tec), this rA2-saponin vaccine induced partial protection in the high dose L. infantum-beagle dog model [28]. Whether prophylactic immunization using A2-based vaccines can achieve similar levels of immunity against VL in genetically diverse human subjects has yet to be determined. Although the predictive value for any animal model in vaccine development ultimately depends on validating data from human trials, the primate M. mulatta, which diverged from humans approximately 25 million years ago, has been accepted as a system that more closely mirrors human immunity for vaccine-development studies against infectious diseases [29], [30]. In this communication, we provide evidence that rA2, as a single antigen, confers marked clinical protection in outbred macaques against L. infantum challenge, and may by itself constitute a promising vaccine candidate against human VL.
The experimental protocols involving monkeys and all the conditions of animal maintenance and handling were reviewed and approved by the Institutional Animal Care and Use Committee (CEUA-FIOCRUZ, resolution # P0048-00 and P.0215/04). All the invasive procedures were performed in accordance with the national guidelines for animal biosafety. Rhesus monkeys (Macaca mulatta) were obtained from a breeding colony from FIOCRUZ Primate Research Centre in Manguinhos (Rio de Janeiro, Brazil) and housed individually for experiments, in stainless-steel squeeze-back cages and fed daily with a commercially available primate diet supplemented with fresh fruits and vegetables. Water was provided ad libitum. The welfare of the primates was closely monitored by a veterinarian, under the supervision of nonhuman primate care specialists. All the procedures involving non-human primates were carried out according to the Brazilian guide for care and use of laboratory animals (Projeto de lei 3.964/97-www.planalto.gov.br), which is conformed to the recommendations of the Weatherall report for the use of non-human primates in research (http://www.acmedsci.ac.uk/images/project/nhpdownl.pdf). To minimize suffering before interventions, such as infectious challenge, sampling or clinical procedures, animals were anaesthetized with ketamine hydrochloride 10 mg.kg−1 (Cetamin, Synthec Vet, São Paulo, Brazil), and midazolam 0.10 mg.kg−1 (Dormonid, Farma-Roche, São Paulo, Brazil), both injected intramuscularly. Animals were submitted to euthanasia with a lethal overdose of thiopental sodium (Euthasol, Virbac Animal Health, Fort Worth, TX) administered intravenously.
The rA2 protein from L. donovani containing a tag of six histidine residues (A2-HIS) used for vaccination and for detecting A2-specific antibodies was purified from E. coli BL-21 containing pET16bA2 plasmid as reported elsewhere [31]. The pCIneo-A2 plasmid (DNA-A2) was constructed following the procedure described by Ramiro and co-workers [32]. The adenovirus recombinant vectors encoding either the L. donovani A2 or the Trypanosoma cruzi Amastigote Specific Surface Protein 2 (ASP2) genes were obtained as previously described elsewhere [26], [33].
The 17 males and 16 females outbred macaque, aged between five and seven years old, weighing around 6 kg, were acclimatized to the laboratory conditions for at least two weeks before the experimental procedures began. As indicated in Table 1 and Figure 1, different homologous and heterologous prime-boost vaccination regimens were used in this study. All vaccine and control formulations were prepared to give a final volume of 1 ml/dose. Briefly, primates were randomized by sex and assigned to seven groups. Group 1 contained three animals that received phosphate saline buffer (PBS). All other groups contained five animals each. The animals vaccinated with rA2 (rA2/rhIL-12/alum) or adenovirus and rA2 (Ad5-A2+rA2/rhIL-12/Alum) received, respectively, three and four subcutaneous doses with 21 days intervals. The animals vaccinated with DNA and adenovirus (DNA-A2+Ad5-A2) received four intramuscular injections in the left deltoid muscle region with 21 days interval. Forty days after the last boost, each macaque was inoculated intravenously with a single dose of 2×107 amastigotes/kg of body weight of a virulent L. infantum strain (MHOM/BR/2001/HP-EMO). Amastigotes were harvested from heavily infected hamster spleens, prepared as previously described [34], and typed by multilocus enzyme electrophoresis before use to challenge control and vaccinated primates.
Clinical follow-up was performed by accurate inspection of monkeys for the presence of typical signs of human VL (fever, diarrhea, body weight loss, hepatomegaly and splenomegaly). Additionally, blood collected into BD vacutainer tubes containing EDTA as an anticoagulant was used for assessment of hematological and blood chemistry parameters. The following blood components were measured with a computer-directed analyzer, using commercially available kits (CELM Cia Equipadora de Laboratórios Modernos, Barueri, SP, Brazil): cholesterol, urea nitrogen, total protein and albumin, alanine aminotransferase (ALT), aspartate aminotransferase (AST) activities. Total erythrocyte, leukocyte and haemoglobin counts were carried out with the cellular counter 530/550 (CELM Cia Equipadora de Laboratórios Modernos, Barueri, SP, Brazil). Commercial assays were conducted in accordance with the manufacturer's instructions. Animals were scored for clinical and laboratory signs on a semi-quantitative scale from 0 (absent) to 3 (severe), and the scores added up to give an overall clinical score for each animal. Monkeys with a total score of 0 to 3 were arbitrarily classified as being affected by sub-patent (low tissue parasitism) or asymptomatic patent infection (steady detection of parasite-positive specimens); those with a score of 4 to 18 were classified as suffering from symptomatic patent infection.
To ascertain the immunogenicity of rA2 antigen, the antibody response was evaluated by ELISA and serum samples from all experimental animals obtained at different time of the experiment. Animals were also assessed by Soluble Leishmania Antigen (SLA)-based ELISA to measure seropositivity for infection. The test procedure was that as described previously [35]. Briefly, ELISA plates (Corning, Tewksbury, MA) were coated with either 5 µg/ml of rA2 or 10 µg/ml of SLA, blocked with PBS 1% BSA, and then incubated with 100 µl of macaque serum diluted 1∶80. After washing three times, 100 µl/well of a peroxidase conjugate rabbit anti-rhesus monkey immunoglobulin G (Accurate Chemical & Sci Co, Westbury, NY, USA) diluted at 1∶20,000 was added, and incubated with the substrate OPD (Zymed, CA, USA). Absorbance at 490 nm was measured with a microplate reader (Model 680, Biorad Laboratories, Hercules, CA). A group of sera with previously known titers as control values, as well as naïve rhesus controls, were included in each test. SLA was prepared from stationary-phase promastigotes of L. infantum (MHOM/BR/2001/HP-EMO) as reported elsewhere [34].
The specificity of circulating anti-A2 antibodies in sera from vaccinated macaques was also assessed by Western blot analysis. Briefly, 2 µg samples of rA2 were loaded, run in a 10% SDS polyacrylamide electrophoresis gel (SDS PAGE) (Biorad Laboratories) and transferred to nitrocellulose sheet (Biorad Laboratories), as previously described by Towbin et al [36]. Nitrocellulose strips corresponding to different SDS PAGE lanes were incubated with serum samples diluted at 1∶200 and rA2-antibody specific binding revealed after incubation with a rabbit antibody anti-rhesus monkey IgG conjugated with horseradish peroxidase (Accurate Chemical & Sci Co). Anti-A2 monoclonal antibody was kindly provided by Dr. Greg Matlashewski (McGill University, Montreal, Canada.) and used as positive control.
For assessment of parasites, biopsy specimens were removed from liver at distinct stages of infection and processed for culture and histological examination or DNA isolation. Biopsy samples were cultured using NNN blood agar medium (Difco, Franklin Lakes, NJ) overlaid with complete Schneider’s Drosophila insect medium (Sigma-Aldrich Corporation, St. Louis, MO) prepared as reported elsewhere [34]. Relative parasite load quantification in terms of DNA amplification was carried out according to the procedure reported by Vitale and co-workers [37]. Briefly, DNA was extracted from the tissue samples using the Illustra tissue and cells genomic Prep Mini Spin Kit (GE Healthcare, Cleveland, OH), according to the manufacturer’s instructions. All samples were submitted to real time PCR with oligonucleotides synthesized by Life Technologies (Carlsbad, CA) for the macaque endogenous β-actin gene (5’- CTTCTACAACGAGCTGCGCG -3’ and 5’ TCATGAGGTAGTCGGTCAGG-3’) to normalize results. The Leishmania kDNA was amplified using oligonucleotides (5’-GGCGTTCTGCGAAAATCG-3’ and 5’- AAAATGGCATTTTCGGGC-3’) designed to amplify the conserved region of the minicircle. Standard curves were obtained from 500 ng to 1 pg (detection limit) of DNA for both targets. The threshold cycle was determined for each point. All real time PCR reactions were also submitted in parallel to gel electrophoresis and melting curves. Results were converted into ng of DNA based on the standard curve; kDNA amplification was then converted into number of parasites, assuming that 200 fentograms of DNA correspond to one parasite (10−6 ng = 1 fg; 2×10−4 ng – 1 parasite).
The experiment was terminated at week 24 post-challenge. Gross and light microscopic examinations of the liver and spleen were performed at necropsy. Paraffin sections from biopsy and necropsy tissues (fixed in 10% neutral buffered formalin) were stained with haematoxylin-eosin (Sigma-Aldrich Corporation).
Student’s t-test was used in comparative analysis of quantitative data and means were defined as significantly different when p-value < 0.05.
In order to test the ability of rA2 to protect non-human primates against a challenge with L. infantum we used different formulations and schedule of prime-boost protocols. The primates from negative control received four doses of PBS alone. In a second group animals received three doses of rA2 and rhIL-12 adsorbed in alum (rA2/rhIL-12/Alum). The respective adjuvant control from second group received three doses of rhIL-12 adsorbed in alum (rhIL-12/Alum). Another group of primates received two doses of adenovirus 5 encoding A2 (Ad5-A2) followed by two doses of rA2/rhIL-12/Alum (Ad5-A2+rA2/rhIL-12/Alum). As control, animals received two doses of adenovirus expressing an unrelated T. cruzi antigen (Ad5-ASP-2) followed by two doses rhIL-12/Alum (Ad5-ASP-2+rhIL-12/Alum). Finally, a group of primates received two doses of plasmid encoding A2 (DNA-A2) gene followed by two doses of Ad5-A2 (DNA-A2+Ad5-A2). As controls for the latter group, primates received two doses of wild type plasmids (DNA-wt) plus two doses of Ad5-ASP-2 (DNA-wt+Ad5-ASP-2). For additional details in route of vaccination, dose interval, and vaccine formulation please see Figure 1 and Table 1.
Upon immunization with different prime-boost regimens, apart from a rise in body temperature by 1–2°C recorded after the last boost, no other systemic adverse reaction in the monkeys was observed throughout the whole period of experiment. Post-vaccination local effect was observed only in two macaques that received a mixture of rhIL-12 and alum as adjuvants. A small transient nodule developed at the site of injection and self-resolved in approximately 10 days (Figure S1)
As shown in Figure 2, all animals vaccinated with either rA2/rhIL-12/alum or rAd5-A2+rA2/rhIL-12/alum protocols, but not with DNA-A2+Ad5-A2, showed higher A2-specific antibody response after the last boost and before infectious challenge. Interesting, the levels of circulating anti-A2 IgG antibodies in animals from rA2/rhIL-12/alum and rAd5-A2+rA2/rhIL-12/alum were decreased after challenge with L. infantum. As can be seen in Figure 3A, the specificity inherent of circulating A2-specific antibodies from immunized macaques was confirmed by immunoblot. The reactivity of a mAb anti-rA2 in lysates of cells infected with Ad5-A2 is also shown in Figure 3B. Following the infectious challenge, there was an initial increase of anti-SLA IgG antibodies in all the groups of monkeys at week 14 post-infection (day 193) (Figure 2).
The specific disease course was quite variable among macaques, ranging from mild to severe VL. This appears to result from the outbred genetics of macaques used in this study. Nevertheless, whereas 80% (12/15) of vaccinated monkeys had asymptomatic patent infections 6 weeks after the infectious challenge, at this time point 72% (13/18) of animals in the control groups were found symptomatic. Moreover, 61% (11/18) primates of the control groups were still considered symptomatic at week 24 post-infection, while only three symptomatic cases of the groups rA2/rhIL-12/alum and DNA-A2+Ad5-A2 clinically recovered from infections (Table 2). None of the macaques vaccinated with Ad5-A2+rA2/rhIL-12/alum were symptomatic at 24 weeks post infection.
Figure 4 shows the overall clinical score estimated for each monkey challenge-infected animal. According to their clinical condition, 6 macaques (with scores of 7–8) and twelve others (with scores of 4–6) in the control groups were classed as poly-symptomatic and oligo-symptomatic, respectively. Conversely, 9 (with scores of 1-3) out of 15 vaccinated monkeys were classed as asymptomatic cases. The most consistent clinical parameters observed in affected monkeys were an intermittent rise in body temperature by 1–3°C, diarrhea, decrease in body weight (12–30% change), anemia and increases in Alanine Aminotransferase (ALT) and Aspartate Aminotransferase (AST) (Table S1). These changes were evident by week 6 post-infection and became more pronounced in those with progressing disease (Table 2).
The impact of vaccination on establishment of infection was assessed through time by parasitological examination or real-time PCR of the liver at 6 and 24 weeks post-infection. As indicated in Table 2, monkeys in all groups had sustained course of infection, ranging from sub-patent (low parasitism and asymptomatic) to asymptomatic (patent parasitism) or symptomatic (patent parasitism and symptomatic). Nevertheless, steady detection by histopathology analysis occurred only in primates that remained with patent infection, i.e., amastigote-containing macrophages were found in post-mortem specimens removed from liver or lymphoid organs. In contrast, most of the cases clinically recovered from infection following vaccination displayed low or undetectable tissue parasitism. Accordingly, the relative DNA quantities of the parasite were significantly lower in immunized macaques than in PBS treated animals (Figure 5). Of note, a more marked reduction on parasite load was found in animals vaccinated with rAd5-A2+rA2/rhIL-12/alum, thus indicating that these animals more efficiently controlled parasite growth.
The main histopathological findings in the liver and spleen of challenged macaques are illustrated in Figures 6. Images shown in Figures 6A to 6G show liver images of immature (poorly differentiated) granuloma, immune (tuberculoid-type) granuloma, immune granuloma composed of epithelioid cells and Langhans-type multinucleated giant cells, immature granuloma containing parasitized macrophages, intrasinusoidal lymphocytosis, mononuclear infiltrate in a portal space, and reactions of Kupffer cells, respectively.
All animals at 6 weeks post-challenge developed poorly differentiated hepatic granulomas (Figure 7), typical of the initial stage of infection, thus confirming the establishment of L. infantum parasitism. These granulomas consisted of an aggregation of activated macrophages containing amastigotes, surrounded by lymphocytes and occasional plasma cells (Figures 6 and 7). Although not remarkable as the later stage of infection (Figure 8) differences were already seen when comparing vaccinated and control groups. In particular, immature granulomas were less frequent and contained less marked parasitised macrophages in macaques vaccinated with the Ad5-A2+rA2/rhIL-12/alum protocol.
At the chronic stage of infection (24 weeks post challenge), older hepatic granulomas composed of concentric layers of macrophages, epithelioid cells, Langhans-type multinucleated giant cells and lymphocytes (Figures 6 and 8) were documented only in groups of control macaques (i.e., PBS, rhIL-12/alum, Ad5-ASP-2+rhIL-12/alum and DNA-wt/Ad5-ASP-2), thus revealing that parasite persisted until the end of the experiment. At this time point, primates vaccinated with Ad5-A2+rA2/rhIL-12/alum exhibited almost complete granuloma resolution. Into a less extent, monkeys of the groups immunized with rA2/rhIL-12/alum or DNA-A2 + Ad5-A2 also displayed a regression of the hepatic lesions, as compared to those from control groups. The quantitative analysis of histological findings (Figure 9) are consistent with tissue liver parasitism (Figure 5) and clinical scores (Figure 4 and Table 2), all analyzed at 24 weeks post challenge, reflecting that protective immunity to L. infantum infection can be induced in heterogeneous macaque population by an A2-based vaccination.
In addition, we examined the lymphoid structure in vaccinated and non-vaccinated controls. Sections from the spleen revealed a high frequency of well organized lymphoid follicle (Figure 6H) in most of the vaccinated macaques, whereas non-vaccinated animals showed more often extensively disorganized lymphoid tissue with follicles decreased in number and size (Figure 6I), as well as sinusoidal congestion at the cortical zone (Figure 6J). Additional histological findings in controls (not vaccinated) included amastigote-containing macrophages in the subcapsular area and/or in the red pulp (data not shown).
On the base of compelling evidence that both CD4+ (including multifunctional Th1 cells and central memory CD4+ T-cells) and CD8+ T-cells are key players in the immune response to leishmaniasis, researchers have focused considerable efforts on the development of prophylactic vaccines that elicit T-cell responses [14], [16], [23] with the premise that such interventions will confer protective effects. Ample evidence supports the notion that heterologous prime-boost vaccination regimens can elicit greater immune responses than single immunization modalities. In this regard, combining DNA priming with a live vectored boost [32], [38], [39] or two different live vectors to prime and boost a response [40], [41] have been explored as a means of raising protective T-cell responses. Of note, sustained immunity elicited by these vaccines correspond to, in addition to the emergence of an specific Th1 response, CD8+ T-cells response [32], [39] that may also provide additional beneficial cytokines and/or their cytotoxic potential may allow release of amastigotes to facilitate killing by activated macrophages [14].
A variety of non-human primate models for both cutaneous leishmaniasis and VL have been used to assess the safety, immunogenicity, and protective efficacy of different vaccine protocols [30]. In most studies of this nature, it is difficult to accurately assess partial host immunity since clinical outcome, a highly variable parameter, is commonly used as a correlate of protection. Although Leishmania-specific T-cell responses can be induced safely in primates by vaccination, it depends on the particular protocol and may ranges from non-existing to full protection after the infectious challenge. However, it has become evident that the current parameters of cell-mediated immunity, i.e., delayed-type hypersensitivity skin tests, or in vitro recall T lymphocyte responses, do not always correlate with clinical recovery and resistance to infectious challenge [30]. Neither study in the L. amazonensis [42] or L. major [43] macaque models, nor those in the L. major-vervet monkey model [44] have resulted in a clear definition of what T-cell responses are required for vaccine-induced protection. Therefore, the only way to determine acquired resistance afforded by a candidate vaccine is to challenge the vaccinated animals with virulent leishmania parasites.
In the present study, we compared the potential efficacy of various A2 vaccination assays, using either recombinant protein, viral and DNA vectors. Our work showed that the vaccine preparations at the dose employed, apart from being safe and well tolerated, also stimulated specific antibody response to the rA2. The transient local adverse reaction recorded in two macaques that had received the recombinant antigen formulated in a mixture of rhIL-12 and alum is in agreement with the results obtained in our previous study [43], but differs from the findings reported by Kenney and co-workers [42]. The duration of these skin nodules was in general longer in their studies. These data are apparently accounted for the different antigen preparations (particulate antigens versus subunit proteins) and the amount of antigen used in the vaccine formulation.
Here, the vaccination protocols including rA2/rhIL-12/alum and Ad5-A2+rA2/rhIL-12/alum were highly immunogenic in that animals developed marked pre-challenge A2-specific antibody response. The lower number of A2 reactors in macaques vaccinated with DNA-A2+Ad5-A2 indicates that response to antigen in the monkey model is quite variable depending on the mode of immunization. For instance, it is well known that alum favor the induction of humoral responses, whereas Ad5 or DNA vaccination are known to induce a stronger T cell mediated immunity, and in particular CD8+ T cell responses. It is noteworthy that the anti-A2 antibody response was downregulated by infection. Likewise, differences in whether infection boosted (or not) the specific antibody responses to the recombinant leishmanial proteins Leish-110f, HI and HASPBI were obtained in a vaccine trail against experimental canine VL [45]. Although B lymphocytes can play an important role in shaping host defense against a number of intracellular pathogens through a variety of interactions with the cellular immune response [46], the precise value of high titers vaccine-induced parasite-specific antibodies in VL has yet to be fully defined [14].
Not surprisingly, macaques vaccinated with the L. donovani A2 antigen in different formulations and application regimens showed varying degrees of parasitological and clinical protection following infectious challenge. Overall, attempts to detect parasite-positive specimens through time by conventional diagnostic procedures (either by culture or direct microscopic examination) were less successful in vaccinated animals as compared to controls. Accordingly, the findings from the real time PCR-based quantification of L. infantum loads in liver samples revealed that most of the vaccinated animals had significantly lower parasitism following the time course of infection. This lower level of parasite burden correlated with reduction of L. infantum-induced granuloma formation in the liver and improvement of clinical conditions, particularly in macaques vaccinated with Ad5-A2+rA2/rhIL-12/alum. The efficiency of this specific regimen may be explained by the combined ability to induce antigen specific CD8+ T cells, and CD4+ Th1 cells, by Ad5-A2 and rA2 combined with rhIL-12/alum, respectively. All thought to be important immunological components in mediating protective immunity against Leishmania parasites.
The vaccine-induced clinical resistance was more evident at week 24 post-infection. At that time point, while only 15% (2/13) of the non-vaccinated macaques had recovered from symptomatic to asymptomatic patent infection, among the vaccinated groups 67% (10/15) animals had sub-patent infection with absence of clinical signs and lower serum levels of Leishmania-specific antibodies or reversion from a positive to negative serology for infection. It is well known that after clinical healing, immune responses likely maintain a state of persistent infection for the life of the host [14], thus suggesting that the protective immune response can control, but not fully eliminate, the sub-patent infection.
Our comparative analysis of the L. infantum-induced hepatic damage in groups of control and vaccinated macaques at week 24 post-infection indicates that all macaques in the control groups developed longstanding immune granulomas with structural properties remarkably similar to those seen in humans infected with this pathogen. Conversely, most of the vaccinated monkeys exhibited either almost complete resolution (Ad5-A2+rA2/rhIL-12/alum regimen) or marked regression (rA2/rhIL-12/alum and DNA-A2+Ad5-A2 regimens) of the poorly differentiated granulomas.
The immunologically active granulomas are thought to restrain the infection, kill the microbial target, and repair any accompanying tissue injuries. However, the overall antimicrobial efficacy of the granulomatous response to Leishmania appears to be variable, and ultimately depends on host determinants and pathogen virulence [47]. In L. donovani-infected mice, the development of effective (parasite-free) hepatic granulomas requires early IL-12-dependent IFN-γ production by T cells for the activation of monocytes/macrophages [48]. On the other hand, Foxp3−CD4+ T subset appears to be the dominant source of IL-10-mediated immune suppression in chronic forms of leishmanial disease in mice [49] and humans [50]. Despite these findings, the way in which IL-10 functions in uncontrolled growth of Leishmania-induced granulomas in infected non-human primates remains unclear [51].
Finally, the atrophy of lymphoid tissue and the disorganization of splenic microenvironments have been observed during canine VL [52], [53]. The mechanisms responsible for splenic protection against systemic infection are based on the clearly defined structural organization of the spleen into compartments [54]. In this study, whilst inflammation and structural changes of the splenic white pulp occurred in control animals, immunized monkeys exhibited well-organized lymphoid follicles, thus suggesting vaccine-induce protective immunity.
In conclusion, the results from this macaque vaccine trial testing different modalities and formulations by using the L. donovani A2 as amastigote specific antigen showed varying degree of protective immunity with respect to parasite load, hepatic granuloma resolution and clinical outcome. Combinations of priming with DNA or Ad5-A2 followed by a boosts with alum formulated subunit A2 protein plus rhIL-12 cytokine were safe, and showed promising protective effects. Giving the genetic variability of human T-cell responses across HLA haplotypes, monomeric vaccines can elicit variable protective immunity [18]. Therefore, a successful DNA and viral vectors as well as subunit protein-based vaccines will likely require a cocktail of proven immunogens. Accordingly, we are currently identifying novel amastigote specific immunogenic proteins that could be aggregated to A2 to further improve the level of vaccine-induced cell-mediated immunity and protection against VL [30].
|
10.1371/journal.ppat.1002210 | A Communal Bacterial Adhesin Anchors Biofilm and Bystander Cells to Surfaces | While the exopolysaccharide component of the biofilm matrix has been intensively studied, much less is known about matrix-associated proteins. To better understand the role of these proteins, we undertook a proteomic analysis of the V. cholerae biofilm matrix. Here we show that the two matrix-associated proteins, Bap1 and RbmA, perform distinct roles in the biofilm matrix. RbmA strengthens intercellular attachments. In contrast, Bap1 is concentrated on surfaces where it serves to anchor the biofilm and recruit cells not yet committed to the sessile lifestyle. This is the first example of a biofilm-derived, communally synthesized conditioning film that stabilizes the association of multilayer biofilms with a surface and facilitates recruitment of planktonic bystanders to the substratum. These studies define a novel paradigm for spatial and functional differentiation of proteins in the biofilm matrix and provide evidence for bacterial cooperation in maintenance and expansion of the multilayer biofilm.
| The bacterial multilayer biofilm consists of matrix-enclosed cells attached to each other to form large aggregates. The base of these aggregates may be attached to a living or non-living surface. The biofilm matrix most often contains at least one exopolysaccharide component and may also contain protein and DNA. While much is known about the exopolysaccharide component of the Gram-negative biofilm matrix, little is known about the function of biofilm matrix proteins. We hypothesized that the biofilm matrix might harbor proteins with diverse functions. Therefore, we undertook the first proteomic analysis of the biofilm matrix of a Gram-negative bacterium, V. cholerae. We subsequently focused on Bap1 and RbmA, two proteins that are abundant in the biofilm matrix. RbmA, which strengthens intercellular interactions, was found to be evenly distributed in the biofilm. In contrast, communally synthesized Bap1 was concentrated at the biofilm-surface interface and stabilized the association of the multilayer biofilm with the surface. Furthermore, the addition of purified Bap1 increased attachment of free-swimming cells to a surface. These studies provide evidence for spatial and functional differentiation of proteins in the biofilm matrix and suggest bacterial cooperation in stabilization of multilayer biofilm surface association and recruitment of new members.
| Bacterial biofilm formation is the process by which bacteria attach to abiotic surfaces, the surfaces of other unicellular organisms, the epithelia of multicellular organisms, and interfaces such as that between air and water. Surface adhesion enables bacteria to arrange themselves favorably in their environment and, therefore, is critical to environmental adaptation and survival.
Surface-attached bacteria may form either a single layered structure, known as a monolayer, or a multilayer biofilm [1]. Bacterial cells join monolayer and multilayer biofilms in response to distinct environmental signals, use distinct structures for adhesion in these two biofilms, and develop distinct transcriptional profiles within these two structures [2], [3]. However, the critical difference between these two types of biofilms is the extracellular matrix that surrounds cells in the multilayer biofilm. This matrix is comprised of biological polymers such as exopolysaccharide, protein, and DNA [4]. The matrix not only mediates bacterial aggregation and surface attachment but may also serve as a reservoir for extracellular, degradative enzymes and the nutrients released by their function. Therefore, the multilayer biofilm affords the bacterium advantages that monolayer biofilm does not.
Vibrio cholerae is a halophilic Gram-negative bacterium that causes the severe diarrheal disease cholera. V. cholerae makes two types of multilayer biofilms. One is dependent on environmental Ca2+ concentrations comparable to those found in seawater, while the other is dependent on the synthesis of an exopolysaccharide termed VPS [2], [5], [6], [7]. The genes required to synthesize VPS are primarily found in two large operons within the VPS island, one of which encodes the proteins VpsA through VpsK and the other of which encodes VpsL through VpsQ [8]. Transcription of these operons is controlled by a complex regulatory network, suggesting that the ability to limit biofilm matrix synthesis to a highly specific environmental niche confers a survival advantage [9], [10], [11], [12], [13], [14], [15], [16], [17].
While most studies suggest that the VPS-dependent V. cholerae biofilm is not important for colonization of the human intestine [18], [19], [20], this biofilm may be important for environmental persistence. Surface-attached V. cholerae predominate in the environment [21], [22]. Multiple avenues of evidence suggest that the chitinaceous surfaces of arthropods are an important substratum for V. cholerae biofilm formation [23], [24], [25], [26], [27]. Furthermore, V. cholerae is especially well adapted to life on chitin because of its many chitinolytic enzymes, the marked modulation of its transcriptome by the degradation products of chitin [28], and activation of its natural competence by chitin [28], [29], [30], [31].
Our laboratory and others have identified several environmental signals that activate VPS-dependent V. cholerae biofilm formation [2], [5], [12], [32], [33], [34], [35]. Among these are sugars transported by the phosphoenolpyruvate phosphotransferase system or PTS. Chitobiose and N-acetylglucosamine, which are degradation products of chitinaceous surfaces of arthropods, are transported exclusively by the PTS [36], [37]. Therefore, in the aquatic environment, association with arthropods is likely correlated with formation of a VPS-dependent biofilm.
To gain insight into the role of biofilm matrix-associated proteins in V. cholerae surface attachment, we set out to define the proteome of the V. cholerae biofilm matrix. Here, we present evidence that the biofilm matrix selectively retains secreted proteins. Furthermore, we show that RbmA and Bap1, two proteins of previously unknown function [3], [38], [39], [40], are present in the biofilm matrix. While RbmA functions similarly to previously identified biofilm matrix proteins in that it strengthens intercellular interactions [41], [42], [43], Bap1, which is jointly synthesized by biofilm-associated bacteria, is concentrated at the base of the biofilm where it reinforces the association of the biofilm with the surface and accelerates attachment of bystander bacteria not yet primed for biofilm matrix synthesis. These studies present evidence for specialization of proteins in the bacterial biofilm matrix and for bacterial cooperation in maintaining and expanding surface-associated biofilms.
Biofilm matrix proteins were isolated by a variety of methods. Briefly, biofilms were disrupted by vortexing in the presence or absence of 1 mm glass beads. Furthermore, biotinylation of extracellular proteins prior to biofilm disruption and subsequent neutravidin affinity purification were used to enrich for extracellular proteins. The protein mixtures prepared by these methods were analyzed by MS/MS. We then used in silico methods (Genome Atlas) to predict the subcellular localization of identified proteins. As shown in Figure S1, the proportion of recovered proteins that were predicted to be extracytoplasmic increased with biotinylation. Gentler methods of biofilm disruption also resulted in isolation of a larger proportion of predicted extracytoplasmic proteins. However, the most gentle disruption methods yielded fewer proteins overall and, therefore, a smaller number of secreted proteins.
The 74 predicted extracytoplasmic proteins identified by these methods are listed in Table S1. Based on either known function or bioinformatics, we predicted that 10 of these proteins were secreted and, therefore, were candidate biofilm matrix-associated proteins (Table 1). In addition, 17 of these proteins were located in the outer membrane (OM), and 26 of these proteins were located in the periplasm. The location of 18 proteins could not be predicted with certainty (Table 2). Citrate synthase (VC2092) and a putative acetyl CoA synthase homolog (VCA0139), which were predicted to have transmembrane domains, are most likely in the inner membrane. No additional inner membrane proteins were identified. NusA (VC0642), a transcription elongation factor we identified in the proteomic analysis, was predicted to be secreted. However, because of its function, we hypothesize that it is cytoplasmic.
Secreted proteins identified in our analysis included those forming bacterial appendages such as the mannose-sensitive hemagglutinin type IV pilus (MshA) and the flagellum as well as RbmA and RbmC, two proteins of unknown function that alter biofilm formation and are co-regulated with the VPS synthesis genes [3], [38], [39], [40]. Three proteins not previously associated with biofilms were also identified, namely a hemolysin (HlyA, VCA0219), a chitinase (VCA0027), and the hemagglutinin/protease (HAP; VCA0865).
We hypothesized that cell-associated proteins should be similarly represented in our analyses if they represented residual cellular material contaminating the biofilm matrix preparation. In fact, while 35% of the periplasmic and OM proteins identified were found in three or more biofilm matrix preparations, only 7% of all predicted cytoplasmic proteins were identified in 3 or more preparations. Furthermore, only two putative inner membrane proteins were identified. One possibility is that a common step in the purification process resulted in formation of spheroplasts, releasing outer membrane and periplasmic proteins into the supernatant during the purification process. Another possibility is that these OM and periplasmic proteins signify the presence of outer membrane vesicles in the biofilm matrix.
RbmC (957 aa), which was identified in our proteomic analysis, and its homolog Bap1 (691 aa) play uncharacterized and redundant roles in the observed colony morphology and biofilm phenotype of rugose V. cholerae variants [39]. The central portions of these proteins are 54% identical and 70% similar and include an EF hand domain, which is predicted to bind Ca2+, and a β-prism lectin-like domain surrounded by six FG-GAP domains (Figure 1A). RbmC is longer than Bap1 due to two N-terminal domains of unknown function that are also found in the E. coli mucinase StcE and a second C-terminal β-prism domain [44], [45].
We first confirmed that these proteins also serve redundant roles in biofilm formation in our V. cholerae strain MO10, which has a smooth rather than rugose colony morphology. As shown in Figure 1B, Δbap1 and ΔrbmC mutants formed a biofilm, while the double mutant did not. The biofilm defect of the Δbap1ΔrbmC mutant could be rescued by a plasmid encoding a wild-type allele of either Bap1 or RbmC (Figure S2). An rbmC allele with a truncation of the C-terminal β-prism domain not found in Bap1 (RbmC-C140) also rescued the biofilm defect of the Δbap1ΔrbmC mutant (Figure S2). These results suggest that, as previously noted for a rugose variant of V. cholerae, Bap1 and RbmC perform redundant functions in the V. cholerae biofilm. Furthermore, Bap1 represents the minimal protein required to rescue the Δbap1ΔrbmC mutant phenotype.
Our proteomic analysis identified ten candidate matrix-associated proteins (Table 1). ChiA-2, MshA, Bap1, RbmA, and the hemolysin HlyA were selected for further study. To determine whether these proteins were secreted by V. cholerae, the gene encoding each of these proteins was cloned between an inducible promoter and a C-terminal FLAG tag. As negative controls, we also cloned EIIAGlc (VC0964), a cytoplasmic component of the phosphoenolpyruvate phosphotransferase system, as well as Escherichia coli alkaline phosphatase (AP) and TcpG (VC0034), two periplasmic proteins. Each of these plasmids was introduced into V. cholerae. After culture in LB broth, the cells and supernatant were separated by centrifugation, and the presence of the tagged protein in each fraction was assessed by Western analysis (Figure 2A). The negative controls EllAGlc, TcpG, and AP were found in the cell pellet only. The secreted proteins chosen for further study were all found in the supernatant to varying degrees.
To determine if these secreted proteins were retained in the biofilm matrix, we formed biofilms with wild-type V. cholerae constitutively expressing affinity-tagged versions of each of these proteins. Biofilms were rinsed, and immunofluorescence was used to visualize the affinity-tagged proteins in the biofilm matrix. No fluorescence was observed for biofilms formed by strains carrying plasmids encoding the proteins EllAGlc, TcpG, or AP (data not shown). As expected, the pilus-forming protein MshA was visualized in the biofilm matrix. In addition, RbmA, Bap1, and HlyA were observed in the biofilm matrix. Although comparable amounts of ChiA-2 were secreted, much less was observed in the biofilm matrix (Figure 2B). This suggests that RbmA, Bap1, and HlyA are selectively retained in the biofilm, while ChiA-2 does not associate strongly with the biofilm matrix.
Bap1 and RbmA were previously found to alter biofilm formation [3], [38], [40]. Therefore, we hypothesized that their role in biofilm formation might be a structural one. To compare the native distributions of Bap1 and RbmA in the biofilm, we fused a FLAG tag to the C-terminal end of Bap1 and RbmA on the chromosome and visualized these tagged proteins in the biofilm by immunofluorescence. As shown in Figure 3A, RbmA was evenly distributed in the vertical dimension, while Bap1 was concentrated at the base of the biofilm. To objectively assess this difference, we measured the total fluorescence intensity in each transverse section. For each biofilm, this measurement was normalized to the transverse section with maximum fluorescence intensity and plotted as a function of distance from the substratum. As shown in Figure 3B, these measurements confirmed that Bap1 was concentrated at the biofilm-surface interface.
To determine whether the distinct vertical distributions of Bap1 and RbmA in the biofilm were the result of spatially heterogeneous transcription of bap1 and rbmA, we formed a biofilm with wild-type V. cholerae constitutively expressing Bap1-FLAG or RbmA-6XHis from a plasmid. As shown in Figure 4, the vertical distribution of Bap1 and RbmA in these biofilms was similar to that in biofilms expressing Bap1 or RbmA from their respective native promoters. However, with constitutive expression, more Bap1 was observed within the biofilm, most likely due to increased levels of protein. Taken together, our data suggest that the vertical distributions of Bap1 and RbmA in the biofilm are not the result of heterogeneous transcription of bap1 and rbmA within the biofilm. Rather, we hypothesize that Bap1 migrates to the biofilm-substratum interface after secretion from the cell.
To assess the transverse distribution of Bap1 and RbmA in the biofilm and the extent of co-localization of these two proteins, we combined equal numbers of a Δbap1 mutant expressing Bap1-FLAG from a plasmid and wild-type V. cholerae expressing RbmA-His from a plasmid. As shown in Figure 5, in transverse sections close to the substratum, Bap1 and RbmA were both distributed around the perimeter of cells, and some co-localization was observed. However, RbmA was more evenly distributed, while foci of increased intensity were observed for Bap1. Similar transverse distributions of each protein were observed in biofilms formed by a Δbap1 mutant expressing Bap1-FLAG from a plasmid alone and by a ΔrbmA mutant expressing RbmA-FLAG from a plasmid alone (data not shown). Based on these observations, we hypothesized that Bap1 might play a different role than RbmA in biofilm formation.
RbmA alters biofilm stability but not overall biofilm accumulation of rugose variants of V. cholerae [38]. In V. cholerae O139 strain MO10, we observed that deletion of rbmA had a small, statistically insignificant effect on biofilm formation. Rescue of a ΔrbmA mutant with a wild-type rbmA allele produced a biofilm that was similar to that of wild-type V. cholerae but significantly increased as compared with the biofilm of the unrescued mutant (Figure 6A). Vortexing completely dispersed the ΔrbmA mutant biofilm, while larger biofilm fragments were observed after similar treatment of the wild-type V. cholerae biofilm (Figure 6C). This ΔrbmA mutant phenotype could be complemented by expression of a wild-type rbmA allele in trans.
We hypothesized that, if secreted RbmA were essential for biofilm integrity, exogenous RbmA should rescue the biofilm defect of a ΔrbmA mutant. To test this, we first affinity purified RbmA (Figure 6B). We then allowed the ΔrbmA mutant to form a biofilm in the presence of increasing amounts of purified RbmA. Purified RbmA was able to rescue the biofilm defect of the ΔrbmA mutant (Figure 6C). We determined that rescue required an RbmA concentration of approximately 416 nM. Assuming all molecules of RbmA are functional, this corresponds to approximately 260,000 molecules per mutant cell.
In a standard assay, the biofilm formed by a Δbap1ΔrbmC mutant was indistinguishable from that formed by a ΔvpsL mutant (Figure 1B). However, we noticed that, unlike the ΔvpsL mutant, the Δbap1ΔrbmC mutant formed a pellicle on the liquid surface after 24 hours of static growth. Interestingly, mutation of bap1 and rbmC in a rugose variant of V. cholerae was not noted to preserve pellicle formation [39]. One possible explanation for this discrepancy is that, due to a difference in the surface chemistries of smooth and rugose variants, rugose variants do not interact as strongly with the air-water interface in the absence of Bap1 and RbmC. As shown in Figure 7, the pellicle formed by the Δbap1ΔrbmC mutant was loosely associated with the glass surface. Gentle shaking dislodged the Δbap1ΔrbmC mutant pellicle from the substratum sending it to the bottom of the tube, while the wild-type pellicle remained attached. Furthermore, vortexing of the Δbap1ΔrbmC mutant pellicle caused it to fragment into many small pieces. However, these pieces were larger than those observed when a ΔrbmA biofilm was vortexed. These defects were rescued by a wild-type bap1 allele provided in trans but not by rbmA (Figure 7), again indicating that Bap1 and RbmA have distinct roles in biofilm formation.
We hypothesized that if secreted Bap1 were responsible for adhesion of the biofilm to the surface, exogenously provided Bap1 should also rescue the Δbap1ΔrbmC mutant biofilm defect. To test this prediction, we used affinity chromatography to purify Bap1-FLAG as shown in Figure 8A. A Δbap1ΔrbmC mutant incubated in the presence of purified Bap1 formed a biofilm that was comparable to that of a Δbap1ΔrbmC mutant rescued by Bap1 expressed from a plasmid (Figure 8B). To determine the concentration of Bap1 required to restore biofilm formation to the Δbap1ΔrbmC mutant, we titrated purified Bap1-FLAG into a Δbap1ΔrbmC mutant culture and measured biofilm formation after 24 hours. As shown in Figure 8C, an 8.8 nM solution of Bap1-FLAG was sufficient to restore surface attachment. Assuming all Bap1 molecules are functional, this corresponds to approximately 5,500 Bap1 molecules per bacterial cell. Therefore, approximately 47 times less Bap1 was required than RbmA to form a biofilm with properties similar to that of wild-type V. cholerae.
To validate these quantifications in a native biofilm, we used Western analysis to estimate the relative quantities of Bap1-FLAG and RbmA-FLAG synthesized in biofilms formed with V. cholerae strains expressing either Bap1-FLAG or RbmA-FLAG from the native chromosomal location (Figure 8D). Two bands were always observed for biofilm-associated RbmA, suggesting that RbmA undergoes proteolysis in the biofilm. Including both RbmA bands in the calculation, we determined that there was approximately 16 times less Bap1 in biofilm preparations as compared with RbmA, recapitulating our results with purified protein. We hypothesize that less Bap1 is required in the biofilm because it principally associates with the base of the biofilm, whereas RbmA is distributed evenly throughout.
Because exogenously provided Bap1 restored biofilm surface adhesion to a Δbap1ΔrbmC mutant, we questioned whether Bap1 synthesis could be a joint venture in the biofilm community. To test this, we co-cultured a Δbap1ΔrbmC mutant with a ΔvpsL mutant. As shown in Figure 7, this produced a biofilm that was comparable to that of wild-type V. cholerae. We rationalized that (i) this biofilm might be comprised chiefly of ΔvpsL mutant cells because the Δbap1ΔrbmC mutant was providing it with the requisite biofilm exopolysaccharide, (ii) the Δbap1ΔrbmC mutant might predominate because the ΔvpsL mutant was providing it with the requisite Bap1 and/or RbmC, or (iii) approximately equal numbers of these two mutants might be found in the biofilm because each was providing the other with the requisite materials for biofilm formation. To determine whether Bap1, VPS, or both were shared resources within the biofilm, we performed a series of co-culture biofilm experiments using lacZ as a marker and determined the relative amounts of each species in the biofilm by plating on media containing 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-Gal). As shown in Figure 9A, although the lacZ+ strain always had a slight advantage in the biofilm, cells lacking Bap1 and RbmC were always found in the biofilm when co-cultured with cells that were able to produce these proteins. In contrast, cells that were unable to synthesize VPS were always excluded from the biofilm in spite of co-culture with cells that were able to synthesize VPS. Based on these findings, we hypothesize that Bap1 is a shared biofilm resource, but VPS is not.
To document communal Bap1 in the V. cholerae biofilm, we then co-cultured a Δbap1ΔrbmC mutant with a ΔvpsL mutant expressing GFP from a chromosomal location and Bap1-FLAG from a plasmid. The biofilms harvested from these co-culture experiments were visualized by microscopy after immunofluorescent staining of Bap1-FLAG and DAPI staining of bacterial DNA. As expected, approximately one GFP-labeled ΔvpsL mutant cell was observed in the biofilm for every GFP-negative Δbap1ΔrbmC mutant cell (Figure 9B). However, the perimeter of many Δbap1ΔrbmC mutant cells exhibited Bap1-FLAG-based immunofluorescence. To confirm that this observation was not the result of transfer of the plasmid from the ΔvpsL mutant to the Δbap1ΔrbmC mutant, we documented that all Δbap1ΔrbmC mutant cells in the biofilm remained sensitive to ampicillin (data not shown).
Our results confirm that Bap1-FLAG provided by a ΔvpsL mutant can be incorporated into the Δbap1ΔrbmC mutant biofilm. These findings indicate that Bap1 is a communal resource. In contrast, because ΔvpsL mutant cells were excluded from both Δbap1ΔrbmC mutant and wild-type V. cholerae biofilms, we conclude that VPS produced by neighboring cells is not available to the ΔvpsL mutant and, therefore, that unlike Bap1, the biofilm exopolysaccharide VPS is not a communal resource but instead tightly associated with the cell of origin.
We questioned whether Bap1 could also increase surface adhesion of bystander cells not yet committed to the sessile life style, as this would have implications for the role of Bap1 in biofilm expansion. We previously identified a medium in which V. cholerae does not synthesize enough of the biofilm matrix components to proceed past the monolayer stage of biofilm development [2]. We cultured wild-type V. cholerae, a Δbap1ΔrbmC mutant, or a ΔvpsL mutant in monolayer minimal medium with supplemented with purified Bap1. As shown in Figure 10A and quantified in Figure 10B, Bap1 increased surface adhesion of wild-type V. cholerae, a Δbap1ΔrbmC mutant, and a ΔvpsL mutant in monolayer minimal medium, while a control protein, bovine serum albumin (BSA), had no effect. This suggests that communal Bap1 secreted by nearby biofilm cells may also increase surface adhesion of bystanders that have not yet been reprogrammed for biofilm matrix synthesis.
The exopolysaccharide component of the bacterial biofilm matrix has been studied intensively [8], [46], [47], [48], [49], [50], [51]. More recently, components such as DNA and protein have been identified in the matrices of some bacterial biofilms. Here we provide the first proteomic analysis of a Gram-negative biofilm matrix. Our analysis revealed 10 secreted proteins, 43 periplasmic and outer membrane proteins, and 18 putative extracytoplasmic proteins whose location could not be predicted.
OM and periplasmic proteins were much more likely to be identified in multiple matrix preparations than inner membrane and cytoplasmic proteins, suggesting that these proteins may not be artifacts caused by cell lysis but rather the contents of biofilm-associated OM vesicles. Outer membrane vesicles are retained in the biofilms of Pseudomonas aeruginosa and Helicobacter pylori, and these vesicles appear to play a role in biofilm formation [52], [53], [54]. Furthermore, there is evidence that the compositions of membrane vesicles derived from the biofilm and from culture supernatants are distinct [52], [55]. V. cholerae has been reported to release outer membrane vesicles [56], [57], [58]. However, additional investigations are required to confirm the presence of these vesicles in the biofilm matrix and to determine their role in biofilm formation.
We studied four secreted proteins identified in our preliminary analysis in addition to MshA. The chitinase, ChiA-2, showed minimal retention in the biofilm matrix. However, three proteins of unknown function, Bap1, RbmA, and HlyA showed extensive association with the matrix. RbmA has no conserved domains of known function. Bap1, its homolog RbmC, and HlyA, which all contain at least one β-prism lectin domain, form a paralogous family in V. cholerae. We hypothesize that these secreted proteins are selectively retained in the biofilm, perhaps by binding to specific moieties in the polysaccharide scaffold.
Bap1 and RbmA were previously shown to play an undefined role in V. cholerae biofilm formation [3], [38], [39]. Here we show that RbmA and Bap1 have distinct distributions in the biofilm matrix. RbmA surrounds biofilm-associated cells throughout the biofilm and reinforces intercellular contacts from this location. In contrast, Bap1 concentrates around cells that form the biofilm-surface interface and stabilizes adhesion of the biofilm to surfaces. The distinct distribution of these proteins is not the result of heterogeneous expression within the biofilm. Rather, we hypothesize that it is the result of self-segregation after secretion from the cell. This is the first example of spatial and functional differentiation of secreted structural proteins in a Gram-negative biofilm matrix.
Biofilm matrix polysaccharide is considered to be a jointly synthesized, shared resource. We show here that this is not the case for the V. cholerae biofilm matrix. While the biofilm matrix protein Bap1 is a communal resource, VPS benefits only cells from which it is synthesized. Therefore, the V. cholerae biofilm exopolysaccharide is not freely secreted and available to the entire community.
Lastly, our results suggest that matrix-associated proteins may play an important role in expansion of existing bacterial biofilms on surfaces. Exogenous Bap1 increases surface adhesion of planktonic bystanders as well. Because nutritional signals and surface attachment are strong activators of the biofilm matrix synthesis genes, in aquatic environments, it is unlikely that planktonic Bap1 and RbmC would be synthesized by planktonic cells in quantities sufficient to increase surface attachment. Rather, we envision that Bap1 and RbmC secreted from an existing biofilm would condition surrounding surfaces, increasing the probability of bystander cell attachment.
These studies reveal a new paradigm for the bacterial biofilm matrix in which the biofilm exopolysaccharide forms a cell-associated scaffold to which communal biofilm matrix proteins adhere, possibly through carbohydrate-binding domains. These proteins may fulfill specialized structural roles or enable cooperative augmentation of the biofilm.
The bacterial strains and plasmids used in this study are listed in Table S2. Vectors used for protein expression included either an IPTG inducible promoter and a FLAG-tag (pFLAG-CTC, Sigma-Aldrich) or an arabinose inducible promoter and a 6X-His tag (pBAD-Topo, Invitrogen). Bacteria were cultivated either in Luria-Bertani broth (LB) or monolayer minimal media [2]. Where indicated, streptomycin (100 µg/ml), ampicillin (50 or 100 µg/ml), arabinose (0.04% wt/vol), and Isopropyl β-D-1-thiogalactopyranoside (1 mM) (IPTG) were added to the growth medium. A 0.1 M phosphate-buffered saline solution (PBS) (pH 7.0) was used in initial biofilm washes, and a 0.1 M Tris-buffered saline solution (TBS) (ph 7.0) was used to wash biofilms after biotin labeling.
10 mls of LB broth supplemented with streptomycin was added to a Petri dish and inoculated with V. cholerae. A biofilm including a pellicle formed over 48 hours of static incubation at 27°C. After incubation, the associated planktonic cells were removed. The remaining biofilm was washed by addition of PBS, agitation on a rotary shaker for 5 minutes with PBS, removal of PBS and non-attached cells, and addition of fresh PBS. This procedure was repeated twice. Matrix proteins were then prepared using each of the following four protocols (Figure S1). In preparation (i), the biofilm was disrupted in the presence of 1.0 mm glass beads (Biospec) and centrifuged to remove particulates. For preparations (ii), (iii), and (iv), a cell surface biotinylation kit (Pierce) was used to biotinylate extracytoplasmic proteins in the washed biofilm according to the manufacturer's instructions. After biotinylation, the biofilm was transferred to a 50 ml conical tube containing 2 mls of PBS. Disruption of the pellicle was carried out by ten sonication cycles of 10 sec (iii) or by vortexing in the presence (ii) or absence (iv) of 1.0 mm glass beads (Biospec) for one minute. The mixtures were then centrifuged at 20, 000× g for 30 min in the cold to remove particulates, the supernatants were applied to Neutravidin-agarose resin (Pierce), and the resin was washed several times with PBS. Biotinylated proteins were eluted from the resin by incubation with PBS to which 50 mM DTT had been added. This disrupts the disulfide bonds bridging biotin residues to extracellular proteins.
The four mixtures of proteins were precipitated with trichloroacetic acid, resuspended in SDS-PAGE loading buffer, run into a 4–20% gradient SDS-polyacrylamide gel, and then sent to the Taplin Mass Spectrometry Facility where the gel was cut into pieces and subjected to an in-gel trypsin digestion procedure. Peptides were extracted from the gel, dried in a speed-vac, and reconstituted in 5–10 µl of HPLC solvent A (2.5% acetonitrile, 0.1% formic acid). Each sample was loaded via a Famos auto sampler (LC Packings) onto a nano-scale reverse-phase HPLC capillary. Eluted peptides were subjected to electrospray ionization and then entered into an LTQ linear ion-trap mass spectrometer (ThermoFisher). Peptide sequences were determined by matching protein databases with the acquired fragmentation pattern by the software program, Sequest (ThermoFisher).
The ORFs of interest were amplified by PCR using primers including the start and stop codons of each gene of interest. For cloning into pBAD-Topo, PCR products were inserted into the expression vector according to the manufacturer's protocol (Invitrogen). For cloning into pFLAG-CTC, either NdeI and KpnI or NdeI and EcoRI restriction sites were included in the PCR primer pairs. The PCR products were then digested and ligated into the expression vector. The ligation products were transformed into E. coli TOP10 competent cells and selected on LB agar plates supplemented with ampicillin (100 µg/ml). The presence of the correct insert was confirmed by colony PCR and sequence analysis. Confirmed plasmids were electroporated into V. cholerae. V. cholerae strains harbouring a pBAD-Topo plasmid were grown in 0.02% arabinose
C-terminal fragments of bap1 and rbmA were amplified from the pFLAG-bap1 and pFLAG-rbmA plasmids, respectively, by the polymerase chain reaction with the following primers: Bap1 A: ATCGTCTAGAGTGTACGCGGGTTACTACGC and B: GACTGCATGCCAGACCGCTTCTGCGTTCTG and RbmA: A: AGTCTCTAGAGCCAGTGATTGAAGCAAATC and B: GACTGCATGCCAGACCGCTTCTGCGTTCTG. The resulting PCR products were digested with XbaI and SphI and ligated into the multiple cloning site of the suicide plasmid pGP704, and the sequence was confirmed. This plasmid was then integrated into the chromosome by single homologous recombination as previously described [37].
The ΔrbmC in-frame deletion mutant was constructed as previously described [12]. Briefly, the following primer pairs: Pair 1 A: TGGCGCCATATTCTATGACA and B: TTACGAGCGGCCGCATACACCCTTCGGCTTCATTC and Pair 2 A: TGCGGCCGCTCGTAATATTGGGCTCAACCCACTATG and B: GGCAGTTTAATGGCGATCAT were used to amplify two genome sequences spanning an in-frame deletion in the gene of interest. These DNA fragments were joined by the SOE technique [59], cloned into pCR2.1-TOPO and then subcloned into the suicide vector pWM91 by ligation after digestion with XhoI and SpeI. This suicide plasmid was used to generate an in-frame deletion in rbmC by double homologous recombination [12]. A similar procedure was used for generation of the ΔrbmA in-frame deletion mutant using the following primer pairs: Pair 1 A: CGTACTCGAGCACCCACAATTAGTGATCGCT and B: TAACGAGCGGCCGCACAACCATTTGTTTTTACAACTGG and Pair 2 A: TGCGGCCGCTCGTTATAAATTTACCTAGTCACTTAGTCGT and B: TCGACACTAGTCAAACTCTAGAACGGAACAAAA.
Biofilm quantification assays were performed as described previously with the following modifications [13]. Briefly, a single colony of V. cholerae was inoculated into 1 ml of LB broth and allowed to grow to mid exponential phase. The culture was then diluted in LB broth to yield an OD655 of 0.05 and divided into three disposable glass culture tubes (10 mm ×75 mm). These tubes were incubated statically at 27°C. After 24 hrs, planktonic cells were removed, and the OD655 of the cells was measured. Remaining biofilms were washed with PBS and then disrupted by vortexing in the presence of 1 mm beads. The OD655 of the resulting cell suspension was measured. For assays of biofilm integrity, biofilms were formed as described above and then either gently shaken or vortexed. All assays were performed in triplicate and statistical significance was determined by a student's t-test.
To evaluate protein secretion, V. cholerae was inoculated into 2 mls of LB broth supplemented with IPTG and ampicillin and grown for 6 hours at 37°C with shaking at 200 rpm. The OD655 of the final culture was measured, and then the cells were centrifuged at 4°C for 15 minutes at 4500 rpm. The supernatants and cell pellets were separated. Cell pellets were resuspended in the volume of PBS required to yield a final OD655 of 1. Five µl of this cell suspension were diluted in 20 µl 1x Laemmli buffer solution and boiled for 5 min. Supernatants were collected and filtered through a 0.25 µm filter. 10 µl of the supernatants were added to 2 µl 5x Laemmli buffer and boiled for 5 min. The protein mixtures in the cell pellets and supernatants were separated by electrophoresis on a 4–20 % precast SDS-PAGE gel (Pierce) and transferred onto a PVDF membrane (Millipore) with a semi-dry transfer apparatus using the Fast Semi-Dry Transfer Buffer (Pierce). The affinity tagged proteins were visualized as follows. Membranes were incubated overnight in a blocking solution consisting of PBS with 0.05% Tween 20 (PBS-T) and 5% skim milk-PBS. The membranes were then incubated with a 1∶10,000 dilution of Anti-FLAG M2-Peroxidase antibody in PBS-T for 1 hour on a rotary shaker. Membranes were washed once for 15 minutes and twice for 5 minutes in PBS-T and then developed using the ECL Plus Western Blotting Detection Reagent (GE Healthcare) according to the manufacturer's instructions.
To evaluate Bap1 and RbmA in the biofilm at native levels, a similar protocol was used with the following modifications: strains carrying either Bap1-FLAG or RbmA-FLAG on the chromosome were allowed to form biofilms for 24 hours in 2 ml of LB. After removal of planktonic cells and spent medium, biofilms were washed with PBS and resuspended in 500 µl of PBS. Biofilm cell extracts were prepared by sonication, and the protein concentrations of the extracts were determined by Bradford assay. 20 µg of each extract was diluted in 20 µL of MilliQ water and 5X Laemmli buffer and separated by SDS-PAGE. As a loading control, the RNA polymerase α-subunit was detected with an antibody raised against the α-subunit of E. coli RNA polymerase (Neoclone). Relative amounts of Bap1 and RbmA in the gel were approximated by densitometry analysis using ImageQuant 5.2 (Molecular Dynamics).
Wells of a 12 well microtiter dish were filled with 2 mls of LB broth supplemented with ampicillin and arabinose, where noted, and a tilted 18 mm ×18 mm glass cover slip was placed in each well. After 24 hours of static culture, the cover slips were placed in 6 well microtiter dishes and washed twice for 5 minutes with 2 mls of PBS on a rotary shaker. The cover slips were then incubated on a rotary shaker for one hour in a blocking solution consisting of PBS supplemented with 3% BSA. This solution was replaced with blocking solution containing Anti-6X His (1∶1, 000 dilution) (Abcam) and/or Anti-FLAG M2 (1∶1, 000 dilution) (Sigma-Aldrich), and the coverslips were then incubated for an additional hour. After this incubation, the cover slips were washed with PBS three times for 5 minutes each time. For labeling of FLAG-tagged proteins with DyLight549, biofilms exposed to the unlabeled Anti-FLAG M2 antibody underwent an additional 45 min incubation with DyLight 549 AffiniPure Rabbit Anti-Mouse IgG H+L (1∶500 dilution) (Jackson ImmunoResearch). For His-tagged proteins, the same procedure was used with an Alexa Fluor 488 Goat Anti-Rabbit Antibody (Invitrogen). The cover slips were then washed in PBS three times, for 5 minutes each time. Where indicated, the cover slips were also incubated with a 1 mg/ml DAPI solution for 5 min. Cover slips were mounted on concave glass slides filled with PBS and then sealed with nail polish. Confocal images were acquired at the Children's Hospital, Boston Imaging Core with a LSM700 microscope (Zeiss) equipped with a 63X objective and 405, 488, and 555 nm laser lines. A computer equipped with ZEN 2009 software was used to acquire and process images. As a control, a DAPI-stained biofilm was imaged before and after immunofluorescence manipulations. Very little change was observed after manipulation, demonstrating that the biofilm was not noticeably degraded by the immunofluorescence staining procedure (data not shown).
Wild-type V. cholerae carrying either a RbmA-FLAG or a Bap1-FLAG expression plasmid were grown overnight on an LB agar plate containing ampicillin. Several of the resulting colonies were inoculated into 100 mls of LB broth supplemented with ampicillin. When the culture reached mid-log phase, IPTG was added to a final concentration of 1 mM. After 4 hours of additional growth, the cells were pelleted at 5,000 rpm at 4°C (Sorvall, rotor SLA-600TC), and the recovered supernatant was distributed into two 50 ml conical tubes. 200 µl of Anti-FLAG M2 Affinity Gel prepared according to the manufacturer's instructions (Sigma-Aldrich) was added to each tube, and the tubes were agitated for 1 hour at room temperature to allow the protein to adhere to the resin. The resin was collected in 10 ml chromatography columns (Bio-Rad) and washed with 2×10 ml PBS. Proteins bound to the resin were eluted with 300 µl of 0.1 M glycine, pH 2.5 and instantly brought to pH 8 by addition of 10 mM Tris, pH 8. Protein concentration was determined by absorbance at 280 nm, and the eluate was analysed by SDS-PAGE using a 12% pre-cast gel (Pierce). After separation, the gel was stained with Imperial Stain (Pierce).
For quantification, equal numbers of lacZ+ and lacZ− V. cholerae strains were inoculated into LB-filled wells of a microtiter dish, and biofilms were allowed to form at 27°C over 24 hours. Biofilms were then disrupted with 1 mm glass beads, and serial dilutions of the resulting cell suspensions were plated for isolation on LB agar plates containing X-GAL. In the morning, numbers of blue and white colonies were recorded.
For microscopy, equal numbers of a ΔrbmCΔbap1 mutant and a ΔvpsL mutant carrying a chromosomally-encoded, constitutively expressed gfp allele and a plasmid-encoded bap1-FLAG allele were inoculated into LB-filled wells of a microtiter dish with a coverslip. Biofilms were allowed to form as described above. Biofilms formed on coverslips were subsequently removed and prepared for immunofluorescence as as described above. These biofilms were examined by confocal microscopy using the LSM700 microscope (Zeiss).
Cells were grown in a 24 well microtiter dish filled with minimal medium (MM) alone or supplemented with purified Bap1 or BSA. An Eclipse TE-2000-E phase contrast microscope (Nikon) equipped with a 20X objective and an Orca digital CCD camera (Hamamatsu) was used to obtain images. Surface area coverage was calculated using IP Lab software (Nikon). Two randomly selected fields were measured in each of three biological replicates.
Proteins listed in Tables 1 and 2 have the following Swiss Prot accession numbers. Table 1: MshA (Q60074), RbmA (Q9KTH4), RbmC (Q9KTH2), FlaB (P0C6C4), FlaD (P0C6C6), FlaC (P0C6C5), FlaA (P0C6C3), ChiA-2 (Q9KND8), HlyA (P09545), and HAP (P24153). Table 2: VC0174 (Q9KVH2), VC0430 (Q9KUT5), VC0483 (Q9KUN2), VC1101 (Q9KT04), VC1154 (Q9KSV2), VC1334 (Q9KSC4), VC1384 (Q9KS75), VC1523 (Q9KRW1), VC1834 (Q9KR13), VC1853 (Q9KQZ4), VC1887 (Q9KQW1), VC1894 (Q9KQV4), VC2168 (Q9KQ36), VC2517 (Q9KP59), VCA0026 (Y2826), VCA0058 (Q9KNA7), VCA0144 (Q9KN22), and VCA0900 (Q9KL48).
|
10.1371/journal.pntd.0000960 | Options for Active Case Detection of Visceral Leishmaniasis in Endemic Districts of India, Nepal and Bangladesh, Comparing Yield, Feasibility and Costs | The VL elimination strategy requires cost-effective tools for case detection and management. This intervention study tests the yield, feasibility and cost of 4 different active case detection (ACD) strategies (camp, index case, incentive and blanket approach) in VL endemic districts of India, Nepal and Bangladesh.
First, VL screening (fever more than 14 days, splenomegaly, rK39 test) was performed in camps. This was followed by house to house screening (blanket approach). An analysis of secondary VL cases in the neighborhood of index cases was simulated (index case approach). A second screening round was repeated 4–6 months later. In another sub-district in India and Nepal, health workers received incentives for detecting new VL cases over a 4 month period (incentive approach). This was followed by house screening for undetected cases. A total of 28 new VL cases were identified by blanket approach in the 1st screening round, and used as ACD gold standard. Of these, the camp approach identified 22 (sensitivity 78.6%), index case approach identified 12 (sensitivity – 42.9%), and incentive approach identified 23 new VL cases out of 29 cases detected by the house screening (sensitivity – 79.3%). The effort required to detect a new VL case varied (blanket approach – 1092 households, incentive approach – 978 households; index case approach – 788 households had to be screened). The cost per new case detected varied (camp approach $21 – $661; index case approach $149 – $200; incentive based approach $50 – $543; blanket screening $112 – $629). The 2nd screening round yielded 20 new VL cases. Sixty and nine new PKDL cases were detected in the first and second round respectively.
ACD in the VL elimination campaign has a high yield of new cases at programme costs which vary according to the screening method chosen. Countries need the right mix of approaches according to the epidemiological profile, affordability and organizational feasibility.
| For the elimination of any infectious disease (i.e., reduction of the burden of a serious public health problem to a minor problem which can be managed by the general health services) the right mix of public health tools has to be identified for the early detection and successful treatment of new cases as well as effective vector control (in the case of vector borne diseases) at affordable costs. The paper provides a powerful example of evidence building for cost-effective early case detection in the visceral leishmaniasis elimination initiative of Bangladesh, India and Nepal. It compares the camp approach (mobile teams testing in chronic fever camps for spleen enlargement and rapid diagnostic tests) with the index case approach (screening for new cases in the neighbourhood of reported visceral leishmaniasis patients) and the incentive based approach (where basic health workers receive an allowance for detecting a new case) using subsequent house-to-house screening for the identification of the real number of un-detected cases. By applying a mix of different study methods and an itinerate research process to identify the most effective, feasible and affordable case detection method, under different environmental conditions, recommendations could be developed which help governments in shaping their visceral leishmaniasis elimination strategy.
| India, Nepal and Bangladesh have committed to eliminate VL by 2015 and have adopted multiple public health strategies towards the elimination goal of reducing VL cases to less than 1 per 10,000 population [1]. Treatment with Miltefosine, the first oral effective drug, the possibility of an effective safe and affordable single dose liposomal amphotericin for treatment, the high specificity of rk39 rapid diagnostic test, and the absence of an animal host reservoir, makes VL a potential target disease for elimination [2]–[4].
VL control program strategies aim to reduce morbidity and mortality and involve a number of approaches. These include early diagnosis and prompt treatment of VL cases, surveillance for early detection of VL outbreaks, making available appropriate diagnostic facilities and drugs, promotion of health awareness, clinical and epidemiological research, and implementation of integrated vector management strategies including indoor residual spraying, use of bed-nets and improvement of environmental and housing conditions.
However, despite these efforts, VL transmission continues primarily in the absence of an effective surveillance system. Though VL patients ultimately report to health centers and hospitals, diagnosis is often missed in the early stages of infection and delayed due to lack of diagnostic facilities at peripheral levels of health systems with a consequent delay in treatment and sustaining the human reservoir [5], [6].
Shortage and supply delays of VL drugs, unavailability of diagnostics at the health center level often contributes to the community's poor public perception of the public health systems, a loss of faith and consequent low community participation/involvement in VL control program activities [7]. A passive surveillance for VL in the public sector is further hampered by an un-regulated private sector which fails to notify VL disease even where required by regulation.
This paper analyzes the feasibility, new case yield and costs of different active case detection strategies as applied in different country VL control program contexts at different VL endemicity levels.
The research program was approved by the Ethics Committees of all participating sites (Ethics Committee, Rajendra Memorial Research Institute of Medical Sciences; Ethics Committee, Institute of Medical Sciences, Banaras Hindu University; Institutional Review Board, Institute of Medicine, Tribhuvan University; Ethics Review Committee, ICDDR,B, and Ethics Review Committee, World Health Organization). Subjects participated in the study after a written informed consent.
The study was designed to test 4 different ACD strategies – camp, index case and blanket ACD strategies (see below) implemented serially in one health center area; and incentives based ACD strategy implemented in parallel in a geographically discontinuous health center area. The study was conducted from May to December 2009.
The study was conducted in the highly VL endemic districts of Saran and Muzaffarpur in Bihar in India, Sarlahi district of Nepal and Mymensingh district of Bangladesh. These regions were selected for their reported high annual VL incidence (per 10,000 population) varying from 20 – 25 in India, 5 – 8 in Nepal and 13 – 31 in Bangladesh. VL control strategies have been implemented variably in all the study districts. Indoor residual spraying has been sporadic and more focal in ‘hot spots’ both in the districts of India and Nepal but has not been done in Bangladesh. Miltefosine and amphotericin B was available in the study districts of India and Nepal but not yet in Bangladesh where at the time of the study, antimonials continued to be the mainstay of VL treatment. VL surveillance relied entirely on passive reporting.
The following 4 different ACD strategies were tested for early identification of VL cases in India, Nepal and Bangladesh:
A camp schedule was developed by the researchers in consultation with District and Health Center Officials. Community sensitization meetings were conducted through village health workers at the village level to solicit involvement of village elders and leaders in the organization of the camps. The Health Center Medical Officer was invited to conduct/assist in the camp. Anganwadi workers, Accredited Social Health Activists (ASHAs in India), health workers and other governmental and non-governmental village level functionaries were assigned roles for promotion and conduct of the camps. Public announcements using loudspeakers were used one day before and on the day of the camp for ensuring publicity for the camps. During the camp, all attendees were screened for fever more than 14 days and examined for splenomegaly. Patients with a positive rK39 test were referred to the nearest health center for treatment with subsequent follow up at home. Additionally, patients with a history of VL treatment in the past and suspected PKDL-like skin lesions were tested with rK39 RDT and referred for confirmation of diagnosis and treatment. The duration of camps was between half to one day; in Nepal, due to the long travel distances, 3 camps in different villages were held back to back while in India and Bangladesh the camp staff returned to their base over night.
House to house screening was used as a gold standard to assess how many new VL cases were missed by the camp approach. This was done including all households of the target villages the day following the camp. All households were numbered and members were screened by trained field staff for fever more than 14 days using a short screening questionnaire which also contained questions on VL disease which had occurred during the 12 months preceding the interview. Persons with chronic fever were examined for spleen enlargement by a physician/trained technician, and if positive were tested with rK39 and then sent for treatment to the closest health centre. Similarly, individuals with a past history of VL treatment with PKDL-like lesions were either referred to the health center or tested with rK39 test and if positive referred for confirmation of PKDL diagnosis and treatment.
To avoid a repeat survey in the same village following the blanket approach, the index case approach was simulated. A list of known VL patients in the target villages of the study treated in the past or currently on treatment was obtained from the health center records. These patients (defined in our study as ‘index’ case) were traced in the village and then visited in reality at their house where their presence was confirmed. All household identity numbers within 50 m radius or up to 100 households around the ‘index’ case were listed. This household list was then compared with the households screened in the blanket approach. Any individual from this household list, who had been detected by the house to house screening as a new VL case, was considered a new VL case detected by the index case approach.
This approach was implemented in India and Nepal (not in Bangladesh for political reasons) during a 4-month period in a well defined geographically distinct health center area. Health workers, Anganwadi workers, ASHA workers and community health volunteers were trained in the early identification and referral of suspected VL and PKDL cases in the community. Visual screening aids including photographs of PKDL cases were used for identifying patients for referral on an ongoing basis. An innovative incentive structure was developed in consultation with district health officials. Incentives (about USD 6) were paid for each new VL/PKDL case detected. All confirmed cases of VL/PKDL identified through this approach were ascertained by the researchers. After 4 months of incentive based ACD, a blanket house to house survey was conducted to ascertain the remaining VL/PKDL cases in the community not detected by the incentive based approach.
In order to assess the need for repeated camps in the same endemic villages, the camp approach with subsequent house to house screening was repeated 4 to 6 months after the first round in order to measure the yield of new cases after a certain period of time.
The WHO definition of suspected VL (fever >14 days with splenomegaly in a VL endemic area and a positive rK39) was taken as basis for initiating treatment as indicated in the national treatment guidelines of the three countries. Patients with PKDL like skin lesions with a past history of VL treatment and testing positive for rK39 were treated for PKDL.
The different ACD strategies were evaluated for their sensitivity of detecting new VL cases using the blanket approach as the reference. Assuming a VL incidence of about 12 per 10,000 population, a total of about 48 newly diagnosed VL cases are expected to be detected per study site which requires a screening population of 40,000 individuals per site. This sample will detect an average sensitivity of 80% with 8% absolute precision at a.05 significance level. On the basis of this assessment, it was decided to screen a population of approximately 40,000 people per site.
In order to calculate the costs of different forms of active case detection for the control program, only those direct costs were included in our assessment which would have to be paid from the budget of the governmental VL control program. The following cost elements were assessed: training, local travel and costs for preparation, costs for materials and allowance costs for personnel involved in the approach.
Data was entered in Epi Info at each site, preliminary cleaning was done before transferring the data base to the data management centre in Pune, India. A second check was performed and a joint data base was created. The data analysis was done using STATA.
The study was carried out in known VL endemic districts (annual incidence varying from 5 per 10,000 in Nepal to 30 per 10,000 in India and Bangladesh). The population in the study districts was poor and less educated compared to the national average. The age and sex composition of the study population across all sites was similar showing a relatively young population.
Sensitivity for the camp, index and incentive approach was calculated separately as percentage of new VL cases identified by each of the approaches divided by the new VL cases identified by the ‘gold standard’ blanket approach. Overall, in the 1st round, the camp approach detected 78.6% of all new VL cases in the study villages: 64.3% in Bangladesh, 83.3% in Saran, India but 100% in Muzaffarpur, India and Sarlahi, Nepal (table 2).
The index case approach detected less than half (42.9%) of new VL cases detected by the blanket approach (35.7% in Bangladesh, 66.7% in Saran, India) which covered a larger area than the 50 meter around index case houses. No new VL cases were detected in households situated within 50 m radius of any of the 19 index cases identified in Nepal (table 2) in contrast to Muzaffarpur in India where the index case approach was able to detect all the 3 new VL cases identified by the blanket approach in the community.
Overall, the incentive approach was able to detect 79.3% of new VL cases in the community (66.7% and 78.9% in the Indian sites and 100% in Nepal; table 2).
The 2nd round of screening was implemented 4–6 months after the 1st round. The yield of new cases detected through the camp approach was, with the exception of Saran, India, in all study sites lower in the 2nd round as compared to the 1st round (average number of patients detected with fever more than 14 days ranged from 1.6 and 3.5 patients per camp in India to 7.5 in Nepal and 4.6 patients in Bangladesh). The percent increase of VL cases by adding ACD to PCD was lower in the 2nd round (see below). Overall, out of a total of 196 fever cases detected, only 12 hitherto undiagnosed VL patients were identified which is 6.1% of the chronic fever cases.
Similarly, in the blanket approach of round 2, out of a total of 409 chronic fever cases, 20 hitherto undiagnosed VL cases were detected (4.9%; table 1).
A total of 20 new VL cases were identified in the 2nd screening round (India 8 in Saran and 2 in Muzaffarpur; Nepal – 3; and Bangladesh – 7) compared to 28 new VL cases in the 1st round. Regarding the sensitivity of different ACD approaches, there was a non-significant decrease from the first to the second round (table 2); the sensitivity of the camp approach was 60% and of the index case approach 20%.
The VL disease burden, estimated as the annual VL incidence by adding the “old” cases, which have been detected by PCD during the preceding 12 months and “new” cases detected by the blanket approach (which includes the case detected by camps) was similar across all sites: 23.4 per 10,000 in Bangladesh; 27.1 and 27.8 in the Indian sites and 28.8 per 10 000 in Nepal (which was due to a VL outbreak in one of the study villages with 96 cases in the preceding year). The percent increase of VL cases (i.e. ACD cases (x100)/PCD cases) was substantial with a 115% overall increase of case numbers (22x100/19) when adding ACD to PCD. In the second round the percent increase was less with 54.5% (12×100/22).
In the 1st round, a total of 42 new PKDL cases (all from Bangladesh) were detected by the camp approach and 23 additional new PKDL cases (Bangladesh – 18; Muzaffarpur, India – 5) in the subsequent blanket household screening. Interestingly, the index case approach was able to detect 15 of these new PKDL cases in the focal search around any of the VL index cases. The incentive based approach (not done in Bangladesh) did not yield any new PKDL case. A total of 9 new PKDL cases (Muzaffarpur, India – 2; Bangladesh – 7) were detected in the 2nd round using the blanket approach. The Nepal and the Saran, India site did not detect any new PKDL case during both rounds.
The point prevalence of PKDL (known + new PKDL cases based on the blanket approach) is estimated to be about 1.9 and 18.2 per 10,000 in India and Bangladesh respectively.
The camp approach in Bangladesh was able to detect 70% (42/60) of all new PKDL cases in the community compared to the blanket approach in the 1st round and 42.9% (3/7) in the 2nd round.
In the 1st round, the effort to detect a new VL case through the camp approach was highest in India (0.14 and 0.26 new VL cases detected per camp) compared to 0.60 and 0.83 cases per camp in Bangladesh and Nepal respectively.
In the index case approach the search in the neighborhood of an index case yielded about 0.05 to 0.06 new VL cases per focal search in India and Bangladesh respectively. Estimates for Nepal could not be determined as there were no new cases detected through the index case approach.
The number of households to be screened to detect one new VL case was higher for the blanket approach (1092 households per 1 new VL case; range 654 to 2433 households (non-incentive approach area) and 978 households, range 594 to 1888 households, in the incentive approach areas) than for the geographically restricted index approach areas where 648 households (range 174 to 1000 households) would have to be screened for detecting one new VL case (table 3).
The average cost (including costs for training, preparation and conduct) for a camp in the 1st round, ranged from USD 85.8 in Muzaffarpur, India to USD 195 in Nepal (table 4). The cost for detecting a new VL case by the camp approach ranged from USD 21.72 in Muzaffarpur, India to USD 661 in Saran, India. The estimated cost for detecting a new VL case by the index case approach (using cost simulations, see methods) ranged from about USD 149.10 in Bangladesh to about USD 200 in the Indian sites. The cost for detecting a new VL case by the incentive based approach ranged from USD 50 and 95 in India to USD 543.25 in Nepal. Similarly, the cost for detecting a new VL case by the blanket approach ranged from USD 112 in Muzaffarpur, India to USD 629 in Nepal.
Although we did not take a representative sample of our study populations, it became clear that the annual VL incidence in endemic districts in India and Bangladesh continues to be very high – more than 20 times the elimination target of 1 per 10,000 to be achieved by 2015 and in the range of previous studies [8], [9]. The burden of VL disease is grossly underestimated [10], [11] by the health systems in the Indian sub-continent due to over-reliability on passive surveillance. Active case detection strategies (household screening and index case based screening) were shown to provide a more realistic representation of the VL burden and are able to improve early diagnosis and potentially treatment of VL [12].This has been reconfirmed in our study where the increase of VL cases by adding ACD to PCD was more than double in the first screening round.
In the present study we could show that the blanket approach of household screening for VL yielded the highest number of new cases, though this ‘gold standard’ for active case detection was expensive and will be difficult to sustain by health systems. Moreover, the effort to conduct the blanket approach is much higher in terms of training manpower for the screening and supervision to ensure the quality of screening and subsequent actions such as spleen examination and Rapid Diagnostic Tests (RDTs).
The camp approach has been extensively used in public health care such as in pulse polio immunization, vitamin A prophylaxis and other public health interventions. This strategy is being tested for the first time in the context of disease surveillance for early diagnosis and treatment of VL. The camp approach, though largely standardized in our study, was implemented with some local variation in India, Nepal and Bangladesh and was successful in detecting three quarter of the unrecognized VL burden in the community. Only approximately 5% of patients with chronic fever had VL and this is a similar proportion of patients with chronic cough that have tuberculosis [13]. This means that it is cost-effective to have trained health staff in the team who can complete the VL diagnosis by doing spleen examination and the RDT. Although we did not compare different delivery mechanisms of chronic fever camps, it can be assumed that camp attendance can be improved by better pre-camp preparation to overcome an indifferent attitude of grass-root level village health workers and negative community attitudes towards public health interventions. The sensitivity of the fever camps for identifying new cases, although of acceptable levels in our study, may also be improved by using better communication skills to increase the demand of camp services. The costs of the camp approach were reasonable in our study. However, areas for savings can possibly be identified. Additional preparatory and planning efforts could also increment program costs.
The index case approach as conducted in our study limited the search for new cases to a 50 m radius around the houses of an index case. This is based on the study by Bern et al [14] who found the proximity to a previous VL case to be a high risk factor for a new infection and on unpublished data by Mondal et al. (personal communication) who found a sharp decrease of VL seroprevalence from Index case houses beyond a radius of 50 m. However, as long as we do not have reliable information on the flight range of vectors and the role of subclinical VL cases in the transmission dynamics, we will not be able to provide clear guidance for where and when to search for secondary cases around an index case. In our study the index case approach was able to identify less than half the VL burden in the community. This was mainly due to the fact that the other newly detected VL cases by house to house screening were living beyond the 50 m radius limit indicating the need for extending the search area in the index case approach of a public health program. A special case was Nepal where in one village, due to an earlier VL outbreak, many index cases were located and extensive case tracing had been done previously, so that no new cases could be detected. Though the pattern is not definitive, it appears that the index case approach performs reasonably well in moderate (Bangladesh) to high (India) VL endemic areas but further research is required for confirmation. The estimated costs of the index case approach were comparatively low though the approach needs to be modified to include geographical areas beyond the 50 m radius around index case houses. It will also be necessary to establish whether the chronic fever patient has to be sent to the health services for further assessment or can receive a full diagnosis, including spleen examination and RDT by a skilled health worker in the home village.
The incentive based approach was able to detect 80% of the VL burden in the community, the highest (100%) being in Nepal where the VL endemicity is relatively low [5]. Though not definitive, this suggests that the incentive based approach works well in low VL endemic areas. Incentives have been tried extensively in public health to improve utilization of services such as contraception, tuberculosis treatment and others. India has recently introduced performance based incentives into their health systems to achieve health care targets. The costs of incentive based approach for VL disease surveillance was lowest (except in Nepal) in our study yet effective in low disease burden settings. Most economic burden studies have focused on costs of VL patient management [15]–[17] and not on case detection. Further research is needed to assess cost effectiveness of the full VL elimination program strategies.
The study also provides insights into the periodicity with which active case detection should be applied. The second ACD rounds for camp and index case approaches showed a much lower percent increase of newly detected VL cases compared to the 1st round and likewise the number of VL cases detected per camp decrease from 0.36 (1st round) to 0.23 (2nd round). Likewise in the index case approach the number of new cases per index case was 1.66 times higher in the 1st round (0.05 new VL cases per index case) than in the 2nd round (0.03 new VL cases per index case). On the other hand, the number of newly detected cases in the second screening round by the camp approach was still substantial, so that affordability and staff availability will be important for the decision if ACD through camps should be done during the current VL elimination campaign once or twice per year in the same villages. Further research on different ACD delivery mechanisms and their costs can provide additional support for rational decision making.
The study provided further evidence on the huge burden of PKDL disease in the community especially in Bangladesh [18] and the ability of ACD to shed light on this largely unrecognized health problem. Further research is needed to better understand the low PKDL prevalence in India and Nepal and to determine cost effective treatment interventions.
In conclusion, the study provides evidence on the effectiveness and costs of different active case detection strategies. Preliminary evidence suggests use of the camp approach in high VL endemic settings, index case surveillance in high to moderate VL endemic settings and incentive based surveillance in low VL endemic settings. The study findings are useful for Country VL Elimination Programs to choose the most appropriate ACD method or tool mix for their communities, depending on the level of VL endemicity, feasibility of implementing these strategies and affordability within the context of their health systems.
|
10.1371/journal.pcbi.1000135 | Prediction by Graph Theoretic Measures of Structural Effects in Proteins Arising from Non-Synonymous Single Nucleotide Polymorphisms | Recent analyses of human genome sequences have given rise to impressive advances in identifying non-synonymous single nucleotide polymorphisms (nsSNPs). By contrast, the annotation of nsSNPs and their links to diseases are progressing at a much slower pace. Many of the current approaches to analysing disease-associated nsSNPs use primarily sequence and evolutionary information, while structural information is relatively less exploited. In order to explore the potential of such information, we developed a structure-based approach, Bongo (Bonds ON Graph), to predict structural effects of nsSNPs. Bongo considers protein structures as residue–residue interaction networks and applies graph theoretical measures to identify the residues that are critical for maintaining structural stability by assessing the consequences on the interaction network of single point mutations. Our results show that Bongo is able to identify mutations that cause both local and global structural effects, with a remarkably low false positive rate. Application of the Bongo method to the prediction of 506 disease-associated nsSNPs resulted in a performance (positive predictive value, PPV, 78.5%) similar to that of PolyPhen (PPV, 77.2%) and PANTHER (PPV, 72.2%). As the Bongo method is solely structure-based, our results indicate that the structural changes resulting from nsSNPs are closely associated to their pathological consequences.
| Non-synonymous single nucleotide polymorphisms (nsSNPs) are single base differences between individual genomes that lead to amino acid changes in protein sequences. They may influence an individual's susceptibility to disease or response to drugs through their impacts on a protein's structure and hence cause functional changes. In this paper, we present a new methodology to estimate the impact of nsSNPs on disease susceptibility. This is made possible by characterising the protein structure and the change of structural stability due to nsSNPs. We show that our computer program Bongo, which describes protein structures as interlinked amino acids, can identify conformational changes resulting from nsSNPs that are closely associated with pathological consequences. Bongo requires only structural information to analyze nsSNPs and thus is complementary to methods that use evolutionary information. Bongo helps us investigate the suggestion that most disease-causing mutations disturb structural features of proteins, thus affecting their stability. We anticipate that making Bongo available to the community will facilitate a better understanding of disease-associated nsSNPs and thus benefit personal medicine in the future.
| The introduction of large-scale genome sequencing technologies has dramatically increased the number of single nucleotide polymorphisms (SNPs) in public databases. For example, the NCBI (National Center for Biotechnology Information) dbSNP database [1], which is a major repository of human SNPs, contained data about ten thousand unique human SNPs as of Build 106 in 2002. By October 2007, there were about six and half million validated unique human SNPs, as of Build 128. Although the progress of collecting SNP data has been impressive, the pace at which disease-related SNPs are annotated is much slower. So far, only a few thousand SNPs have been associated with a human genetic disorder in the OMIM (Online Mendelian Inheritance in Man) database [2]. Further efforts are thus required to identify disease-associated SNPs in order to understand their effects on human health.
Genetic variations, such as SNPs, are likely to contribute to susceptibility to complex diseases such as cancer [3]. Single nucleotide variations in the coding regions that lead to amino acid substitutions, the so-called non-synonymous SNPs (nsSNPs), may be associated with a modulation of protein function. For example, extensive studies on point mutations in P-glycoprotein have shown that amino acid variations regulate its substrate specificity and lead to a variation of drug disposition among individuals [4]. As a consequence, attention has been focused on the study of the relation between nsSNPs and disease as well as predicting their phenotypic effects. Some early approaches exploited position-specific evolutionary information contained in multiple sequence alignments [5],[6]. Others have used predictive features of sequence and structure [7],[8], or machine learning algorithms [9]–[11] to classify SNPs. In addition, there are approaches that annotate nsSNPs at a genomic scale, such as LS-SNP [12]. Previous analyses have shown that methods that apply only sequence information may suffer significant reductions in accuracy when fewer than ten homologous sequences are available for the target protein [8]. Sunyaev et al. [13] have shown that disease-causing mutations often affect intrinsic structural features of proteins, while in an important study Wang and Moult [14] have demonstrated that most disease-associated mutations appear to affect protein stability rather than interfere directly with protein interactions. Following these results, others have focused on comparing the structures of wild-type and mutant-type proteins [14],[15] or have estimated the change of protein stability by using environment-specific amino acid substitution matrices that are derived from the three-dimensional structures of homologous proteins [16].
For analyzing structural effects of nsSNPs, we have developed an approach, Bongo (Bonds ON Graph, http://www-cryst.bioc.cam.ac.uk/˜tammy/Bongo), which uses graph theoretic measures to annotate nsSNPs. Graph theory has found many applications in the study of protein structures during the past two decades. For example, Ahmed and Gohlke used graphs to identify rigid clusters for modelling macromolecular conformational changes [17]; Canutescu and colleagues have predicted side-chain conformations by partitioning graphs in which vertices represent residues [18]; Vendruscolo and colleagues applied small-world networks to identify key residues that are important for protein folding [19]; Jacobs, Thorpe and their colleagues used graphs to describe bond-bending networks between atoms, so identifying the rigid and flexible regions in the proteins [20],[21]; Kannan [22]; and Brinda and Vishveshwara [23] used the graph spectral method to identify side chain clusters that are important for protein folding and oligomerisation ; Sol and colleagues used graphs to identify key residues for allosteric communication and modular connection by the edge betweenness algorithm [24],[25]. Bongo uses graphs to represent residue-residue interaction networks within proteins and to assign key residues that are important for maintaining the networks. The novelty lies in the application of a graph theory concept, vertex cover, by which key residues are identified for analyzing structural effects of single point mutations.
Here we begin by describing the use of interaction graphs to represent protein structures. We then introduce the ‘key residues’ that Bongo uses to evaluate structural impacts of point mutations, and explain their roles in terms of stabilising protein structures. We further describe the algorithm of Bongo, where a graph concept vertex cover was adapted to identify key residues, and we calibrate Bongo over eight single point mutations that result in a range of different structural changes in the p53 core domain. We evaluate the false positive rate of Bongo for 113 mutations where wild-type and mutant-type crystal structures have been demonstrated to have negligible differences in backbone conformation. Eventually, we evaluate the performance of Bongo by testing its ability to distinguish disease- and non-disease-associated nsSNPs in protein structures in the PDB (Protein Data Bank) [26]. Based on the benchmark results, we also analyse the percentage of disease-associated nsSNPs that are likely to cause structural effects in proteins.
A point mutation in a protein may often give rise only to a rearrangement of amino acid side chains near the mutation site, although sometimes a more substantial movement of polypeptide backbone locally or globally results. The former changes can be analysed by looking at the inter-residue interactions that a mutation creates or abolishes between its neighbouring residues. However the same approach may not be applicable to the latter, since simply paying attention to interactions immediately around a mutation site is not sufficient to predict structural effects on a larger scale.
In order to understand structural changes at a longer distance, we represent a protein as a residue-residue interaction graph, in which vertices represent residues and edges represent interactions between residues (Figure 1) (see more details in Methods). Of course, molecular dynamics calculations provide a powerful tool for identifying the impact of point mutations on the stability of the native states of proteins. However, these simulations are often time-consuming and require large computer power. Thus we have developed Bongo to provide an alternative approach by operating on interaction graphs, which are computationally more convenient. In our model, residue-residue interactions occur either through direct connection or through indirect links that involve intermediate residues. Such connectivity is based on ‘key residues’ that are important in maintaining the overall topology of the network, and thus the stability of the folded structure. These key residues eventually serve as reference points to evaluate whether a mutation can induce structural changes in a protein away from the mutation site.
Bongo measures the impact of a mutation according to its effects on key residues; it formulates the structural changes in a protein as changes of the key residues in a corresponding interaction graph. Here we adapt a variant of the vertex cover, defined in graph theory as a minimum set of vertices (residues) that are crucial to forming all the edges (interactions), to represent the key residues.
In Figure 2, we illustrate the notion of key residues and introduce the use of the difference between the vertex cover of wild and mutant type interaction graphs as a measure of the effects of a mutation. The example here is residue Y35 of protein 1BPI, a key residue forming several relatively strong interactions including pi-cation interactions with residues R20 and N44 and a hydrophobic interaction with residue A40 (Figure 2A and 2B). The mutation Y35G removes this amino acid from the set of key residues in the graph (Figure 2C) as its original interactions with other secondary structure elements no longer exist. Hence, residue 35 is no longer a key residue in the mutant interaction network. Therefore, this mutation is considered structurally damaging by Bongo; we discuss the exact criteria under which a mutation is deemed damaging below.
Bongo derives the interaction graph of a protein by considering each residue as a vertex and each residue-residue interaction, including hydrogen bonds, π–π, π–cation, and hydrophobic interactions, as an edge. The weight on each edge differs according to the total number of cross-secondary structure interactions as well as number of interactions with individual residues. The weighting scheme was calibrated against eight disease-associated mutations in the p53 core domain analysed by Fersht and co-workers [27],[28], as shown in Table 1. The optimised weighting of inter-secondary structure interactions is 0.8, 0.8, 0.8, 2.0 and 2.0 for H-bonds, π–π, π–cation, hydrophobic interaction, and hydrophobic core respectively. For internal interactions, H-bonds, π–π, π–cation interactions were given a weight of 0.6 and hydrophobic interaction a weight of 0.8. This distinction between inter and intra secondary structure interactions is used to reflect concerted movement of structural motifs within proteins. Thus, a single interaction loss among two densely interacting structures is less significant than one among two sparsely interacting ones.
Based on the above weighting scheme, Bongo defines the key residues as the minimum weighted vertex cover (see the definition of vertex cover in Methods), which represents the minimum necessary residues to establish the interaction network. However, finding the minimum vertex cover is known to be NP-complete and hence efficient algorithms only exist for approximate solutions [29]. Therefore, we use a selection scheme which adopts an approximation algorithm based on the greedy principle to identify the key residues. The approximation algorithm is known to give vertex covers that cost no more than H(|V|), where |V| denotes the size of a vertex set, times than the optimum solution where H(n) is the nth harmonic number. Compared to other graph theoretic constructs such as dominating sets [29], the vertex cover gives an intuitive notion of vertex importance. In fact, we have used more advanced techniques such as spectral decomposition [29] to identify structural information that is related to protein stability change, ΔΔG. However, the results were not better than those obtained by applying the vertex cover approach (data not shown). Indeed, we have observed in some cases (Figure 3) that the change of vertex cover after mutation correlates well with structural data. Therefore, we believe that the vertex cover can serve as a useful approach to estimating protein structural changes.
The key residues maintain the interaction networks in a protein, and each is assigned a priority value that measures its importance in determining the overall topology of the network (see Methods). When a point mutation is introduced into a protein, Bongo quantifies its structural effects according to the priorities of key residues affected. Thus we expect key residues, especially those with high priorities, to have important roles in stabilising folded protein structures. In order to check if the priority of key residues reflects their roles in forming structures, we calculated the correlation between the priority and the stability change (Each key residue was mutated to 19 other amino acids and the stability changes were calculated by I-mutant2.0 [30] (http://gpcr2.biocomp.unibo.it/˜emidio/I-Mutant2.0/I-Mutant2.0_Details.html), which has accuracy around 80% for predicting stability changes resulting from mutations when the three-dimensional protein structure is known. We consider only mutations that cause |ΔΔG|<3kcal/mol since they affect the stability without totally abolishing the overall structure of the protein. The median number of |ΔΔG|<3kcal/mol is used to calculate the correlation with the priority of key residues in order to avoid data skewness.), ΔΔG, of key residues identified from the p53 core domain (PDB: 1TSR). When we considered the top half of the key residues ranked by their priorities, ΔΔG relates to the priority of key residues with a Pearson correlation r = 0.61 and a significantly small p-value less than 0.001 (Figure 4A). This indicates that the correlation is statistically significant and also shows a good contrast to the low relation (r = −0.04) between assumptive priority (Since the non-key residues do not have priority values, they are assigned values according to those of the key residues that are nearest in the same secondary structures. If a non-key residue is flanked by two key residues, its assumed priority is the average of the priority values of its two neighbours.) and ΔΔG of non-key residues (Figure 4B).
We noticed that the correlation is weaker (r = 0.36) when the lower half of key residues, ranked by their priorities, is included. This is likely due to uncertainties in the definitions of key residues that are ranked with lower priorities: Since Bongo stops selecting key residues only when no edges are left in a graph, the key residues that have lower priorities may not have structural meaning but are simply chosen in order to complete the selection process (covering all the edges/interactions in the graph). In an attempt to exclude the uncertain key residues, we analysed how far the correlation is valid by gradually including key residues that have priorities in the lower half, in order of decreasing priorities. There is an acceptable correlation r = 0.52 when we consider up to three fourths of overall key residues, which suggests that the bottom one quarter key residues are not reliable indicators of structural effects. Thus Bongo does not consider the bottom quarter key residues so that their uncertainty does not affect the prediction results.
The distribution of key residues according to their location in secondary structures (Figure 4A) shows that the key residues in β-strands tend to have larger ΔΔGs and priority values compared to those in loops, whereas such differences are less clear for the case of non-key residues (Figure 4B). This suggests that, in general, protein stability should be more vulnerable to mutations in β-strands than those in loops, consistent with the observation that the β-strands in the p53 core domain are the major contributors to the core region of the protein. It also indicates that priority values and ΔΔG of key residues have consistent meanings in terms of protein structure.
Since the structures of the mutant proteins are not often available for nsSNPs, Bongo first uses Andante [31] to model the mutant-type protein structure by rearranging the side chain around the mutation site. The structural effects of a mutation are then analysed by comparing the wild-type and mutant-type key residues, denoted as Kwt and Kmt, respectively. If a key residue in Kwt is not found in Kmt, then it is considered to be affected by the mutation. Consequently the overall impact (I) of a mutation is calculated according to the key residues affected by the mutation, i.e.(1)where I is the total impact value, Kj is the priority of each key residue that is in Kwt but not in Kmt. N is the total number of key residues in Kwt, which normalise the size of proteins.
Thus each mutation is systematically quantified by its impact value I (an overview scheme of Bongo is shown in Figure 5). On deriving the impact value, Bongo considers mutations with I>1 to cause structural effects, which is the criterion calibrated over mutations in the p53 core domain.
In Figure 3, we give an example, the mutation Y35G in protein 1BPI, of how a mutation can have significant impact value. In addition to residue Y35, Bongo also predicts residues R42 to be affected by the mutation (Figure 3A). These two are at the ends of β strands and also in long loops linked to them. These regions undergo conformational changes when the mutation Y35G is introduced into the protein, where the biggest movement (4.2 Å) occurs between the wild-type and mutant-type Cα atom of residue G36 (The movement is measured when the wild-type (1BPI) and the mutant-type (8PTI) are superimposed by their Cα atoms.). Since the impact score calculated on the basis of these residues is greater than one, Bongo considers the mutation Y35G to cause structural effects in 1BPI, which corresponds to the experimental result.
In order to assess the errors due to the difference of a crystal structure of the mutant and a simulated one, we also compared the key residues of the two structures. It turns out that the differences of key residues between the modelled and the crystal structures are mostly located in the loop region, where structural changes occur when the mutation is introduced into the protein (Figure 3B). The overall distribution of the key residues that are specific for the modelled structure is similar to that of the key residues specific for the crystal structure. This suggests that the structural change at a longer distance can be captured in the interaction graphs by simply modelling a point mutation as rearrangement of side chains neighbouring to the mutation site.
In order to calibrate Bongo, we have used experimental data on the tumour suppressor p53 core domain, which is responsible for about 50% of mutations that lead to human cancers [32]. Owing to its importance, the wild-type and many mutant protein crystal structures have been determined. Several studies have been carried out for these point mutations within the domain, and thus make it a good calibration system for predicting structural effects of mutations. Furthermore, the structure of the p53 core domain is inherently unstable with a melting temperature of ∼42–44°C [33]. As a consequence, point mutations that cause either subtle structural changes or more dramatic effects are available for comparison.
For our study we identified eight nsSNPs (Figure 6) analysed experimentally by Fersht and co-workers [27],[28]. These mutations involve several different levels of structural change in the p53 core domain: (i) R273H has only a minor effect on the overall structure, with root mean square deviation (RMSD) ≤0.21Å in Cα positions between wild type and mutant type crystal structures; (ii) G245S, R249S, and R248A destabilise the p53 core domain by 1–2 kcal/mol and lead to local structural changes; (iii) C242S, H168R, V143A, and I195T destabilise the structure >2 kcal/mol and lead to global unfolding of the protein at body temperature. When the structure 1TSR in PDB was used as a calibration model, Bongo identified all mutations except R273H as causing structural effects in the p53 core domain (Table 1), which corresponds well with experimental data described in the literature.
For comparison we also used PolyPhen [5] to predict the effects of the same mutations. We consider PolyPhen as it uses multi-source data including three-dimensional structures, sequence alignments and SWISS-PROT annotations. Compared to other methods which either focus on protein structure or sequence information, it provides more comprehensive results. Of course, there are other methods that include even more information—for example, LS-SNP [12] also considers functional pathways, domain–domain interfaces, ligand–protein binding—but our purpose is to understand the usefulness of structural information by comparing it with a standard approach that mainly uses sequence and structural information. The results in Table 1 show that PolyPhen predicts all mutations except V143A to be probably damaging. PolyPhen's success in predicting R273H to be damaging is probably a consequence of the fact that R273 is functionally important for binding DNA and thus conserved in sequence for reasons that are not evident from consideration of the structure alone, whereas PolyPhen predicts V143A to be benign, probably as a result of comparatively weaker emphasis on structural information.
We further tested the application of Bongo to single point mutations that do not affect protein structure. Our benchmark set included 113 pairs of wild-type and mutant-type crystal structures in which each of them has RMSD in their backbone Cα atoms <0.4Å and the lower resolution of the two structures is ≤2.2Å (Dataset S1A). We chose these criteria in order to allow for experimental errors in the crystallographic solution of the structures of identical proteins, as suggested in the work of Hubbard and Blundell [35]. The benchmark result shows that Bongo predicts three of the single point mutations to cause structural effects, therefore yields a 2.7% false positive rate. Although this result may not be generalised to all the cases, it indeed encourages us to expect a low false positive prediction rate.
In the previous sections, we have shown that Bongo is able to predict structural effects of single point mutations with a low false positive rate. Here we further analyse the performance of Bongo in identifying disease-associated nsSNPs. Our test-set contains 506 disease-associated nsSNPs from the OMIM (Online Mendelian Inheritance in Man) database [2] and 220 non-disease-associated nsSNPs available in dbSNP database [1] which have no annotations in OMIM. All the nsSNPs in the test-set can be mapped to structures in the PDB (Dataset S1B and S1C) since Bongo uses structure as input.
For evaluation of Bongo, we calculated its sensitivity and specificity (with definitions explained in Table 2). By definition, if a method always classifies any mutation as ‘disease-associated’, it would achieve a sensitivity score of 100%. Similarly, a method could obtain a 100% specificity score by always predicting mutations as “non-disease-associated”. In order to avoid a biased analysis, we also calculated the PPV (positive predictive value) and NPV (negative predictive value; with definitions explained in Table 2); a better PPV or NPV implies a better performance in predicting positive or negative cases, respectively.
The overall test results (Table 2) show that Bongo has PPV and NPV of 78.5% and 34.5%, respectively, compared to that of PolyPhen of 77.2% and 37.6%, respectively. This indicates that Bongo and PolyPhen have similar accuracy in predicting disease-associated nsSNPs. Given the fact that PolyPhen also exploits sequence information that may take account of protein interactions with various substrates, macromolecules and other ligands, we believe this shows the potential of using interaction networks which consider structure alone. The similar predictive values suggest that, although the mechanisms by which nsSNPs induce diseases are complicated, structural change is an important factor in most cases. This is consistent with a previous study that shows most deleterious nsSNPs affect protein stability but not functionality [14], which indicates that structural impact is a more important factor in causing disease. In order to assess the performance of Bongo, we also compared the use of PANTHER [36], which is verified to have higher accuracy than PolyPhen by using Hidden Markov Model (HMM) for sequence scoring. The result shows that PANTHER has the PPV and NPV values (There are 48 disease-associated and 22 non-disease-associated nsSNPs for which PANTHER did not find an HMM model to do prediction; those nsSNPs are excluded from the calculation of PPV and NPV values.) comparable to those of PolyPhen and Bongo (Table 2), which further verifies the evaluation.
In addition to the predictive value, Bongo has a low sensitivity (28.1%) compared to that of PolyPhen (50.7%) and PANTHER (76.6%), and its specificity (82.4%) is high compared to that of PolyPhen (65.8%) and PANTHER (31.8%). This suggests that, although Bongo has a similar predictive value to that of PolyPhen and PANTHER, Bongo's high specificity and low sensitivity yields many less false positive predictions. We can thus be more confident about the cases that are predicted as disease-associated by Bongo than those predicted by PolyPhen. Regarding the low sensitivity of Bongo, we suppose this is due to the fact that Bongo is not able to predict mutations that only affect the function of proteins, e.g., the mutations in active or other interaction sites. We may improve Bongo's ability in predicting functional site mutations in the future work.
Among the 506 disease-associated nsSNPs in our test-set, Bongo predicted 142 of them to cause structural effects, which suggests that about 28% of nsSNPs that are involved in Mendelian diseases resulting from single protein mutations may cause extensive structural effects in proteins. However, the figure for nsSNPs involved in multigenic diseases like diabetes may not be so high as they exist individually in the population as a whole at high levels, but contribute only rarely to multigenic diseases when occurring with several other nsSNPs.
We have developed a method, Bongo, which uses graph theoretic measures to evaluate the structural impacts of single point mutations. Our approach has shown that identifying structurally important key residues in proteins is effective in predicting point mutations that cause extensive structural effects with a substantially lower false positive rate. Furthermore, our approach gives clues about the effects of nsSNPs on the structures of proteins, thus providing information complementary to methods based on sequence. By comparing our approach with PolyPhen and PANTHER in analyzing nsSNPs, we have also shown that structural information can provide results of quality comparable to those that use sequence and evolutionary information in predicting disease-associated nsSNPs.
In the residue-residue interaction graphs, Bongo considers structural information including hydrogen bonds, π–π, π–cation, and hydrophobic interactions, as well as secondary structure information. (1) Hydrogen bond: we use HBPLUS [37] to calculate hydrogen bonds, using its default settings for positioning hydrogen and minimum angles formed by the donor and acceptor at the hydrogen. (2) π–π interaction: aromatic side chains are considered to have π–π interaction if they have less than 6 Å between any atoms. We note that more accurate criteria could be applied at the expense of the calculation speed with similar results. (3) π–cation interactions are identified on the condition that there is a cation within 7 Å of any side chain atoms of an aromatic ring such that the angle between the cation and the normal vector of the aromatic ring is within 60°. The criterion is only an approximate one in order to speed up the overall calculation without sacrificing the accuracy of calibration. (4) Hydrophobic interactions are weighted according to Voronoi surfaces between non-polar residues calculated by an in-house program, Provat [38], while hydrophobic cores are identified when a non-polar residue shares non-zero Voronoi surfaces with only non-polar residues. (5) Secondary structure elements are assigned by DSSP [39].
The weighting of the interactions were optimised by using the Least-Squares Optimisation Tool in MATLAB (http://www.mathworks.com/products/matlab/), where the best solution was chosen on the basis of the best calibration result over the eight mutations listed in Table 1. Although calibration was carried out against only eight mutations, the performance of Bongo on the 506 disease-associated nsSNPs, which are distributed in proteins from many different families, is comparable to that of PolyPhen (Table 2).
Since the mutant-type structures are not usually available, we generate them computationally using Andante [31]. Andante predicts the structure by using evolutionary information to define rotamers in clusters of side chains that are structurally compatible, so rearranging the local structure around the mutation site. It should be noted that Bongo does not benefit from sequence information by using Andante, since the rearrangement of local side chains modelled by Andante simply introduces a local rearrangement to the residue-residue interaction network of a protein, which does not affect the overall structure of the interaction graph and is independent of the process of selecting vertex cover.
All the structural information is transferred into graphs by using Graphviz (http://www.graphviz.org/), which is an open source graph visualization project from AT&T Research.
Bongo derives the interaction graph of a protein by considering each residue as a vertex and each residue-residue interaction as an edge. More formally, an interaction graph G = (V,E) is a graph such that V is the set of residues and E is a set of edges. An edge (u,v) is defined between residue u and v if they exhibit one of the following interactions: backbone bonding, hydrogen bonds (H-bonds), π–π, π–cation, and hydrophobic interactions. Each edge is initially given a weight of 1. We then normalise interactions between two secondary structures by dividing the weight with the total number of cross-secondary structure interactions. Intra-secondary structure interactions are normalised in the same way. For interactions involving a group of residues, namely hydrophobic interactions, we normalise them by the Vonoroi surface area of each residues.
Since the key residues capture the vertices that are essential to maintain the interactions, we model them through the vertex cover set of the graph [29]. A vertex cover set S of a graph G = (V,E) is the set of vertices such that for every edge (u,v), either u or v is included in S. In the interaction graph terms, this amounts to picking a set of residues that covers every interaction in the graph. In Bongo, since the interactions are weighted, we consider the vertex cover problem G = (V,E,c) where c: V → R+ is the function that assigns weight to each vertex. A vertex cover set is said to be minimum if it contains the set of vertices that covers all interactions with smallest possible weight.
The algorithm used to select key residues captures the concept of pulling out one piece each time in a tower of wooden pieces, with the difference that in our case the pieces pulled out are key pieces but not redundant ones (Figure 7):
The algorithm reflects the importance of key residues in order of selection: key residues selected in an earlier time are more important, in terms of having higher priorities in maintaining the interaction network, than others that are identified later. Since there is a specific order of choosing vertices, the approximate vertex cover chosen by Bongo for a specific graph will be the same when Bongo repeats the selection process again. Taking advantage of the priorities assigned to each key residue, Bongo eventually quantifies the effect of a point mutation by considering the priority of key residues affected.
|
10.1371/journal.pntd.0003193 | Evaluation of a Dengue NS1 Antigen Detection Assay Sensitivity and Specificity for the Diagnosis of Acute Dengue Virus Infection | Currently, no dengue NS1 detection kit has regulatory approval for the diagnosis of acute dengue fever. Here we report the sensitivity and specificity of the InBios DEN Detect NS1 ELISA using a panel of well characterized human acute fever serum specimens.
The InBios DENV Detect NS1 ELISA was tested using a panel composed of 334 serum specimens collected from acute febrile patients seeking care in a Bangkok hospital in 2010 and 2011. Of these patients, 314 were found to have acute dengue by either RT-PCR and/or anti-dengue IgM/IgG ELISA. Alongside the InBios NS1 ELISA kit, we compared the performance characteristics of the BioRad Platelia NS1 antigen kit. The InBios NS1 ELISA Ag kit had a higher overall sensitivity (86% vs 72.8%) but equal specificity (100%) compared to the BioRad Platelia kit. The serological status of the patient significantly influenced the outcome. In primary infections, the InBios NS1 kit demonstrated a higher sensitivity (98.8%) than in secondary infections (83.5%). We found significant variation in the sensitivity of the InBios NS1 ELISA kit depending on the serotype of the dengue virus and also found decreasing sensitivity the longer after the onset of illness, showing 100% sensitivity early during illness, but dropping below 50% by Day 7.
The InBios NS1 ELISA kit demonstrated high accuracy when compared to the initial clinical diagnosis with greater than 85% agreement when patients were clinically diagnosed with dengue illness. Results presented here suggest the accurate detection of circulating dengue NS1 by the InBios DENV Detect NS1 ELISA can provide clinicians with a useful tool for diagnosis of early dengue infections.
| Infections by the mosquito-transmitted dengue virus continue to increase, threatening the health of people in over a hundred countries worldwide. The lack of effective mosquito control, licensed dengue vaccines or specific therapeutics to treat dengue infections presents challenges to reduce the burden of this disease. Rapid and accurate diagnosis of dengue infections is critical for the reduction of patients' morbidity and mortality. We present data that support the use of the InBios DENV Detect NS1 ELISA for the detection of dengue NS1 in patient serum. The InBios NS1 ELISA kit was tested against sera collected from acute fever patients seeking medical care in a Bangkok, Thailand hospital during 2010 and 2011. The data demonstrate the InBios DENV Detect NS1 ELISA accurately detects circulating dengue NS1 in the tested specimens, demonstrating high sensitivity and specificity. Nonetheless, the sensitivity of the NS1 ELISA kits was found to vary depending on the serological status of the patient (primary versus secondary dengue infection), time of specimen collection and dengue serotype. Further, the performance characteristics of the InBios DENV Detect NS1 ELISA were found to meet and exceed those of the commonly used Platelia Dengue NS1 Ag kit.
| Dengue fever (DF) and dengue hemorrhagic fever (DHF) are mosquito-borne illnesses caused by infection with four related, but antigenically-distinct, dengue viruses (DENV1, DENV2, DENV3, DENV4). The virus is thought to be responsible for close to 400 million infections per year worldwide, of which approximately 100 million are clinically apparent [1]. Although many dengue vaccines are currently under development, none have been licensed. Similarly, there are no specific licensed therapeutics against DF or DHF. The outcome of patients with DF or DHF depends significantly on early diagnosis, leading to appropriate clinical management. Currently, there is no US FDA-approved diagnostic test that can accurately detect dengue NS1 during the acute febrile stage of infection. Traditional serological diagnosis measuring levels of anti-dengue IgM and IgG is hampered by cross-reactivity especially in dengue endemic areas where more than one flavivirus co-circulate [2]. Moreover, serological approaches are based on the detection of antibodies, which can take several days to develop. RT-PCR is sensitive for diagnosis early in infection, but is relatively expensive and requires specialized equipment and skills that may not be available in resource-poor settings where dengue virus is endemic. An NS1 antigen capture ELISA, first developed in 2000 for DENV [3], was based on the premise it would act as a surrogate marker for viremia. Circulating NS1 in the serum of acute DF and DHF patients is an attractive target for diagnosis as it is a viral glycoprotein released from infected cells as soluble polymers as early as day 1 post-infection. It can remain circulating for several days after defervescence [4] and is detectable in the serum of patients with primary or secondary dengue infections [3]. A number of commercial kits have been developed and have been subjected to field evaluation [5]–[16] including a number of second-generation rapid assays for point-of-care use for early diagnosis of dengue infection [17]–[21]. The sensitivity and specificity of these assays vary significantly depending on the dengue serotype and lineage [21].
In this study, we estimated the diagnostic accuracy of the InBios DENV Detect NS1 ELISA (InBios, USA) and compared it to the widely used Platelia Dengue NS1 Ag kit (Bio-Rad, France). We used human serum specimens collected in Thailand during public health service testing during 2010 and 2011. These specimens were collected during acute febrile episodes and sent to our laboratories at the Armed Forces Research Institute of Medical Sciences (AFRIMS) for dengue laboratory confirmation. All specimens were tested using AFRIMS laboratory reference assays for dengue infection by IgM antibody capture ELISA and RT-PCR.
Clinical specimens used in this study were collected through routine public health service testing in Bangkok, Thailand during 2010 and 2011 from 334 patients suspected to have dengue fever within 7 days from onset of symptoms (Table 1). Clinical diagnoses by local Thai clinical staff were based on the 2011 SEARO World Health Organization [22] definition of acute dengue infection. Serum samples from patients were transferred to AFRIMS, Bangkok, Thailand and were maintained at −70°C until tested. All laboratory investigations were carried out at AFRIMS by experienced technicians following standard operating procedures. The technicians performing and interpreting the assays were blind to other test results and to any clinical information on the patients. Approval for the use of these specimens was obtained from the Institutional Review Boards of the Queen Sirikit National Institute of Child Health and of the Walter Reed Army Institute of Research. No private or confidential information was collected.
Paired acute and convalescent serum specimens were tested for anti-dengue IgM (MAC) and IgG ELISA, Japanese encephalitis virus [23], [24] and chikungunya [25] assays. Acute samples were also tested by nested RT-PCR for the presence of dengue virus and serotype identity [26], [27]. Although the ELISA assessment utilizes both acute and convalescent samples for infection status, only acute samples were used in this study for testing the InBios and Platelia Bio-Rad assays.
A patient was determined to have an acute DENV infection by identification of dengue virus genome by RT-PCR from an acute serum sample and/or detection of anti-DENV IgM antibodies by MAC and/or a ≥2 fold rise to ≥100 U in paired acute and convalescent samples [17], [28]. DENV IgM-positive cases were considered to be primary infections if the ratio of DENV IgM to IgG was ≥1.8 [23]. If the ratio was <1.8, it was considered a secondary infection. A sample with any positive criterion listed was used in the analyses as part of the composite positive control group except when discussing serotype where only RT-PCR positive results were used.
The InBios DENV Detect NS1 ELISA test is an assay for the detection of dengue virus NS1 antigen in human sera. Kits were provided by InBios (Seattle, WA). The test is based on the capture of NS1 antigen using a sandwich-type immunoassay and was performed by strictly following the instructions provided by the manufacturer. Briefly, 50 µL serum samples and controls were diluted 1∶2 with sample diluent buffer containing the secondary antibody and incubated at 37°C for 1 hr in microtiter plates pre-coated with anti-NS1 antibody. After washing, the wells were treated with a conjugate solution containing horse radish peroxidase (HRP) polyclonal antibody and incubated for 30 minutes at 37°C. Wells were washed and incubated with 3,3′,5,5′-tetramethylbenzidine (TMB) substrate solution in the dark at room temperature for 20 minutes. After addition of the stop solution, the optical density was read at 450 nm. The immune status ratio (ISR) was calculated from the ratio of the optical density of the test sample divided by the mean optical density of the cut-off control. ISR values ≥1 were considered positive for the presence of NS1 antigen.
In parallel, the same amount of each sample (50 µL) was assayed on the same day using the Platelia dengue NS1 Ag kit (Bio-Rad) according to the manufacturer's instructions. Briefly, 50 µL of sample and controls were diluted 1∶2 with sample diluent and combined with 100 µL of diluted HRP-labeled anti-NS1 monoclonal antibody. This solution was added to microtiter plates coated with anti-NS1 monoclonal antibodies and incubated at 37°C for 90 minutes. After washing, complexes between the capture antibody, NS1 and HRP-labeled antibodies were detected by a colorimetric reaction after incubation with TMB for 30 minutes. After the addition of a stop solution, the optical density of samples was read at 450/620 nm. A sample ratio was calculated by dividing the optical density of the test sample by the mean optical density of the cut-off controls. Sample ratios of <0.5, 0.5 to <1.0 and ≥1.0 were considered negative, equivocal and positive for the presence of NS1 antigen, respectively.
For data analyses, equivocal values were considered negative. Both the InBios and BioRad assays are not marketed as quantitative assays.
Test characteristics with their respective binomial 95% confidence intervals (CI) were calculated using standard formulas. Differences in assay performance were calculated by using McNemar's test [29]. Significance differences (p<0.05) in positivity rates relative to dengue virus serotypes were calculated using Pearson's chi-square or Fisher's exact test. SPSS for Windows version 19 and MedCalc version 12.4 software were used for analyses.
A total of 334 samples from individual subjects were evaluated (Table 1) from routine public health service samples collected in Bangkok, Thailand, between 2010 and 2011. The median age of the subjects was 9 years (range: 1 month to 24.7 years). DENV infection was confirmed in 314 (94.0%) subjects and serotype was determined by nested RT-PCR in 299 (95.2%). Fifteen (4.8%) subjects were negative by RT-PCR and confirmed as dengue positive by serological testing only. The serotype was unable to be determined in these 15 cases. Fifty one subjects (16.2%) had a primary infection, 260 (82.8%) had a secondary infection and serological status was not determined in 3 (1.0%) subjects as they were RT-PCR positive only. Clinical diagnoses by local Thai physicians were based on WHO SEARO 2011 guidelines [22]. There were 159 (50.6%) cases of DF and 141 (44.9%) cases of DHF/DSS with evidence of plasma leakage and/or deaths. Fourteen (4.5%) cases with laboratory confirmation of DENV infection were not given clinical diagnoses of dengue. No clinical diagnosis was given in 8 of the 14 subjects and there was one diagnosis each of acute bronchitis, acute gastritis, viral gastroenteritis, viral induced thrombocytopenia, viral illness and rickettsial illness. Sera samples were collected a median of 4 days from onset of illness (DOI). Twenty samples were negative for dengue infection.
Test performance characteristics of the InBios and Bio-Rad assays were compared against a composite reference standard including samples positive by RT-PCR positive and/or serological testing (Table 2). The overall sensitivity of the InBios assay was 86.0% (95% CI 81.7–89.4) and was significantly higher than the sensitivity of the Bio-Rad assay at 72.8% (95% CI 67.1–77.0) (McNemar's, p<0.0001). Both assays had specificities of 100% (95% CI 83.9–100.0) and there were no false positive results for either assay when compared to the composite reference standard. The InBios test was significantly more sensitive for patients ≤5 years of age at 95.1% compared to 83.8% for those >5 years (Chi-square, p = 0.02).
In Thailand, where the majority of presentations are secondary infections, we wanted to see if there was a difference in sensitivity between primary and secondary infections. The sensitivity of the InBios assay was 98.2% for primary infections (n = 51) and 83.5% for secondary infections (n = 260) which was significantly different (Chi-square, p = 0.002) (Table 3). This was not associated with a difference in the median day of presentation after illness onset which was 4 days for subjects presenting with either primary or secondary infection. The variation in sensitivity between primary and secondary infections was also seen with the Bio-Rad assay (96.3% vs 67.3% for primary and secondary infections, respectively; Chi-square, p<0.0001). There was no difference between the rate of NS1 detection for the InBios and Bio-Rad assays for primary infections, but there was a significant difference for secondary infections (McNemar's, p<0.0001). Sensitivity of NS1 detection with the InBios assay in IgG negative samples was 89.1% (251/282; 95% CI 85.0–92.2) and 55.2% (16/29; 95% CI 37.6–71.6) in IgG positive samples (Fisher's Exact Test, p<0.0001). A similar difference was also seen with the Bio-Rad assay with sensitivity of 77.2 (217/282; 95% CI 72.0–81.7) and 24.1% (7/29; 95% CI 12.2–42.1) for IgG negative and IgG positive samples, respectively (Fisher's Exact Test, p<0.0001). There was no difference in sensitivity for IgM positive or negative samples for either the InBios or Bio-Rad assays.
When all positive samples (n = 314) were stratified by their DOI, the sensitivity of the test decreased at later time points (Figure 1A) for both the InBios and Bio-Rad assays dropping to 50.0% and 37.5%, respectively, by Day 7. The difference seen between the assays is primarily due to differences in the sensitivity for secondary infections (Figure 1C). The specificity of both assays was 100.0% throughout for both primary and secondary infections.
When considering primary infections, the rate of detection of NS1 antigen remained close to 100% for both the InBios and Bio-Rad assays for all days tested (Figure 1B). However, for secondary infections the InBios assay positivity remained above 90% through Day 2 after symptom onset and dropped to 68% by Day 6 (Figure 1C). In contrast, the sensitivity of the Bio-Rad assay decreased starting on the day following symptom onset. The InBios test was significantly more sensitive than the Bio-Rad assay from Day 2 through Day 5.
Performance characteristics also varied by serotype (Table 4). When evaluating PCR positive results only, differences in sensitivity were seen between serotypes (Fisher's Exact Test, InBios, p = 0.04 and Bio-Rad, p<0.0001). Detection of DENV-4 was least sensitive with the InBios assay, whereas detection of DENV-2 was least sensitive with the Bio-Rad assay. The InBios assay was more sensitive than the Bio-Rad assay for each serotype, but this difference was only significant for DENV-2 (McNemar's, p<0.0001).
Out of 334 samples, 301 subjects were given a clinical diagnosis of dengue infection. Only one of these patients was given an incorrect diagnosis of DF and 14 of the 314 subjects positive for dengue infection were not identified clinically. A small decrease in sensitivity was seen in primary infections between DF and DHF/DSS disease when using the BioRad assay (Table 5), but was not significant (Chi-square, p = 0.246).
No single diagnostic assay can accurately detect dengue infection throughout its clinical course. A number of commercial tests for diagnosis of dengue infection are available, but only two have received regulatory approval [30], [31], none of which measures NS1. Diagnosis of dengue is done by detection of genomic material by RT-PCR early during infection and with serological assays (detection of IgM and IgG) at later time points [22], [32]. In the last decade, NS1 antigen testing has become common for early diagnosis of dengue infection [5]–[21], [33]–[35]. However, negative NS1 test results late in infection does not rule out an acute dengue infection, as it may be caused by low circulating NS1 antigens in the blood [17], [36]. In this study we evaluated the InBios DENV Detect NS1 ELISA, which we compared against the widely used Bio-Rad Platelia Dengue NS1 Ag kit, for the diagnosis of acute dengue infection using hospital-based passive surveillance samples collected in Bangkok in 2010 and 2011. We compared it against the Bio-Rad assay since it was the first commercial NS1 antigen detection assay, is one of the best characterized systems and is also the most widely used NS1 assay in Bangkok, Thailand. We evaluated the overall sensitivity of the assays against a panel of acute specimens with known serological status (i.e. primary vs secondary infection), number of days after onset of illness, DENV serotype, and clinical diagnosis or severity of infection.
The performance of NS1 assays, alone and in combination with detection of IgM or IgG, have been extensively tested [8], [13], [17], [21], [37], [38]. While results are not always consistent amongst different cohorts and assays, a number of general comments can be made. Sensitivity is highest in primary infections, when testing occurs shortly after onset of symptoms and when IgG is not detectable. Variations in sensitivity are dependent on the comparator used and dengue serotype. In previous studies of NS1 antigen detection, the sensitivity varied between 34% and 96% [5]–[12], [14]–[21], [33]–[35] while the specificity was always very high. In our study, we found the overall sensitivities for the InBios and Bio-Rad assays were in keeping with previous studies evaluating the performance of the Bio-Rad assay [7]–[12], [15], [16], [19]–[21], [33]–[35]. A number of samples collected on DOI 6 were found to be NS1 antigen positive. They were RT-PCR negative yet were determined to be positive for dengue infection by serological testing. This discrepant result between the NS1 and RT-PCR assays is likely due to NS1 antigen circulating in the serum for longer periods than viral RNA; thus extending the diagnostic window beyond that for RT-PCR testing.
The sensitivity in primary infections was significantly higher for both the InBios and Bio-Rad assays as has been seen in a number of previous studies [5], [11], [12], [16]–[21], [33], [38] and cannot be explained by when samples were collected in relation to symptom onset. More primary infections were seen in subjects ≤5 years (40.9% vs 10.3% in subjects >5 years) and likely explains the increased sensitivity in this age group. As expected, we saw the overall sensitivity of both assays drop off for samples collected later after symptom onset [7]–[9], [11], [12], [14]–[17], [19]–[21], [33], [34], [38]. For the InBios assay, overall sensitivity remained above 80% until DOI 6. When results were separated into primary and secondary infections, the sensitivity for both the InBios and Bio-Rad assays remained above 90% throughout for primary infections. Thus, decreased overall sensitivity for samples tested at later time points was mostly due to the influence of the decreased sensitivity in secondary infections. Differences in sensitivity may also be explained by antibodies that bind to NS1which may be different in each kit, although this information is not available to us.
Detection of NS1 antigen in secondary infections may be hampered by a rapid rise in antibody levels due to the anamnestic antibody response [39] resulting in the formation of immune complexes likely with IgG antibody, which prevents the binding of capture or detection antibodies to NS1 antigen. Increased sensitivity was seen in studies where a step to dissociate immune complexes was included [7], [40]. In keeping with previous studies [7]–[9], [14], [33], [35], we found the sensitivity of NS1 detection in acute samples with the InBios assay and Bio-Rad assays was lower in IgG positive compared to IgG negative samples with no difference between IgM positive or negative samples. This parallels what we see in primary infections where there are low levels of IgG and higher sensitivities than in secondary infections with higher levels of IgG and lower sensitivities.
Differences in sensitivity between DENV serotypes have been seen in some [10], [12], [17]–[19], [21], [33]–[35], but not other studies [8], [9], [14], [20]. In our study, we saw differences in sensitivity between serotypes, but the profile was different between the two assays. As seen previously, the Bio-Rad assay was least sensitive to DENV-2 [10], [12], [20], [33]. This may be partially explained by a larger proportion of secondary infections (88.2%) in the panel of DENV-2 samples tested. However, this is less likely as the larger proportion of secondary infections in DENV-2 did not affect the sensitivity of the InBios assay which remained high (89.5%). For the InBios assay, sensitivity was highest for DENV-1 as seen previously for other NS1 kits [8], [9], [12], [18], [19], [21], [33]. Similar to other studies, sensitivity was lowest for DENV-4 samples in the InBios assay, although there were only a small number of samples for this evaluation [18], [21], [35]. A number of other groups have found the sensitivity to DENV-3 to be decreased [8], [17], [19], but this was not seen in the current study.
Results have not been consistent when evaluating the sensitivity of NS1 antigen detection based on disease classification. Some groups have found no differences [7], [34] whereas Osorio et al. [21] found a decrease in sensitivity to NS1 detection in more severe cases, but is thought to be because more severe cases tended to be secondary infections and presented at later time points than non-severe infections. In our study, no difference in sensitivity was seen for either the InBios or Bio-Rad assays based on clinical disease classification.
There were a number of limitations to our study. Samples for testing were chosen from archived samples from Thai patients and may not represent circulating dengue viruses elsewhere. Due to sample availability, only a limited number of primary infections and DENV-4 infections were included. Dengue negative samples were tested for Japanese Encephalitis and Chikungunya infections, but were otherwise not fully characterized. More comprehensive testing of the InBios assay and other NS1 antigen detection assays needs to be done and should incorporate a larger number of primary infections and samples collected from both children and adults. To date, no cross-reactivity has been seen with other flaviviruses [8], [41], but this finding needs to be confirmed in a larger cohort of acute phase samples with better characterized flavivirus-positive samples. Future studies also need to consider differences in geographic area, circulating serotype (and genotype), patient ethnicity, viremia, immunological response and clinical severity.
This report show the InBios DENV Detect NS1 ELISA has comparable, if not better, performance characteristics to other NS1 antigen kits. Although its sensitivity varies depending on the serological status of the patient, date of specimen collection and serotype of the infecting virus, its use for accurate diagnosis of dengue infection should be considered by clinicians especially early in infection.
|
10.1371/journal.pgen.1005857 | Chromatoid Body Protein TDRD6 Supports Long 3’ UTR Triggered Nonsense Mediated mRNA Decay | Chromatoid bodies (CBs) are spermiogenesis-specific organelles of largely unknown function. CBs harbor various RNA species, RNA-associated proteins and proteins of the tudor domain family like TDRD6, which is required for a proper CB architecture. Proteome analysis of purified CBs revealed components of the nonsense-mediated mRNA decay (NMD) machinery including UPF1. TDRD6 is essential for UPF1 localization to CBs, for UPF1-UPF2 and UPF1-MVH interactions. Upon removal of TDRD6, the association of several mRNAs with UPF1 and UPF2 is disturbed, and the long 3’ UTR-stimulated but not the downstream exon-exon junction triggered pathway of NMD is impaired. Reduced association of the long 3’ UTR mRNAs with UPF1 and UPF2 correlates with increased stability and enhanced translational activity. Thus, we identified TDRD6 within CBs as required for mRNA degradation, specifically the extended 3’ UTR-triggered NMD pathway, and provide evidence for the requirement of NMD in spermiogenesis. This function depends on TDRD6-promoted assembly of mRNA and decay enzymes in CBs.
| Tudor-domain containing protein 6 (TDRD6) is a central component of the chromatoid body (CB) in male germ cells. Chromatoid bodies, which are present in spermatids, contain RNA and protein, are not enclosed by membranes, and typically reside close to the nucleus. Without TDRD6, a much distorted CB structure is observed, and this work asked for the functional contribution of TDRD6 to spermatids. We found that TDRD6 is required for localization of an RNA degradation machinery to the CB. This so-called nonsense mediated decay (NMD) machinery, known from somatic cells, destroys mRNAs that feature premature stop codons. Absence of TDRD6 significantly impairs one specific mechanism of NMD, which depends on long 3’ untranslated regions of the transcripts. Thus, the CB component TDRD6 acts in the assembly of the NMD machinery in the CB.
| During mammalian gametogenesis, substantial changes in chromosome structure and in gene expression profiles occur. Male germ cell transcription and transcriptomes have been studied quite extensively. Some transcriptome studies concern the whole testis, and thus represent combined somatic and germ cell data, which limits germ cell-specific conclusions. Other transcriptome reports describe meiosis- and/or postmeiosis-specific gene expression patterns (reviewed in [1]), some reports deal with the silencing of unsynapsed axial elements in early meiosis (reviewed in [2]), with the silencing associated with histone replacement by specific histone variants or by protamines (reviewed in [3]), with specific RNA species such as small RNA families (reviewed in [4]), with alternative splicing in germ cells (reviewed in [5]) or with transcription factors acting specifically in male germ cells (reviewed in [6]). Various RNA binding proteins were studied (reviewed in [7] and are mostly involved in processing of RNA species. Proper post-transcriptional processing of RNA molecules is essential for germ cell development and thus to producing gametes. For example, in late spermiogenesis, transcripts with short 3’ untranslated regions (UTRs) become preferred and are more stable [8] due to changes in 3’ UTR processing factors [9]. An example of such regulation is the Rnf4 transcript of which a long isoform is present in spermatocytes and a shorter isoform, due to truncation of the 3’ UTR at an upstream polyadenylation site, is present in spermatids [10]. The localization and regulation of mRNA translation is also specifically regulated. mRNAs coding for proteins important for the late stages of spermiogenesis, such as Prm1 and Prm2, are synthesized in the early round spermatid stage, but reside in translationally repressed mRNP complexes. The transcripts become competent for active translation only during the later stages [11]. The apparent reason for such regulation is the transcriptional inactivity of the elongating spermatids due to the extensive nuclear condensation,i.e. transcripts are produced early and stored for later translation. RNA-rich structures termed “nuage’” or “granules” appear in spermatogenic cells and seem to orchestrate this peculiar post-transcriptional program.
The cytoplasmic presence of “germ granules” is a unique feature of germ cells is. Germ granules are RNA-rich, non-membranous cytoplasmic structures that have been proposed to play important roles in RNA post-transcriptional regulation [12]. On the basis of structural features and protein composition, the different germ granules of mammalian spermatogenic cells are designated, for example, chromatoid bodies (CBs) in spermatids and intermitochondrial cement (ICM) in spermatocytes [13]. The CBs appear first as fibrous and granulated material in the interstices of mitochondrial clusters and in the perinuclear area of pachytene spermatocytes. After meiosis, the CBs condense into one single lobulated, perinuclear granule in round spermatids and disassemble later during spermatid elongation. It has been shown that CBs contain mRNA, miRNA and piRNA species [14,15]. Proteins that participate in RNA transport such as KIF17b, miRISC (miRNA induced silencing complex) proteins such as the Argonaute family proteins AGO2, AGO3 and Dicer, proteins implicated in piRNA biogenesis and function such as MIWI and MILI [14,16], RNA helicases/binding proteins such as MVH(DDX4), UPF1, GRTH(DDX25) and PABP and RNA decaying factors such as SMG6 [17] were recently found in CBs. According to their molecular composition, CBs were proposed to function in translational repression, RNA silencing and mRNA storage. Thus, CBs may provide a platform for various RNA processing enzymes and processes, which are, however, little described.
The CBs are enriched also in TUDOR domain (TDRD) containing proteins. It was proposed that the TUDOR domains of these proteins provide interaction interfaces and create a scaffold to organize the CB structure [18]. The association of TDRD5, TDRD9 and TDRD7 with piRNA biogenesis enzymes such as MILI and MIWI is essential for piRNA biogenesis and retrotransposon mRNA silencing, and established a role of CBs in piRNA processing [19–21]. A role of CBs in splicing was suggested based on the observation that TDRD1 participates in complexes with snRNAs in the context of CBs [22]. Previous work by us [23] showed that TDRD6 is major component of CBs and is required for its architecture. Ablation of TDRD6 disrupts the CB structure and leads to developmental arrest at the round-to-elongated spermatid stage. Altered presence of miRNAs was observed in Tdrd6-/- spermatocytes, but piRNA biogenesis and retrotransposon silencing were not affected.
However, whether TDRD6 is implicated in other mRNA metabolic processes that may occur within the CB was unknown. To gain insights into functions of TDRD6 and thus likely of the CB, we performed proteomics of purified CBs. Having identified UPF1 and UPF2 in the CB, which are key proteins in the nonsense mediated mRNA decay (NMD) pathway, we aimed at determining the contribution of TDRD6 to mRNA decay. Previous reports on processes associated with the 3’ end of mRNAs in spermatogenesis describe specific signals embedded in the 3’ UTR sequences or with individual proteins binding there (reviewed in [24,25]. However, the NMD pathway has not been described for mouse or human spermatocytes or spermatids. Processes and complexes that serve mRNA stability and function in germ cells are not sufficiently understood.
We show here that TDRD6 is essential for UPF1 localization to CBs and is critical for UPF1-UPF2 and UPF-MVH interactions. We report that a specific branch of NMD, the 3’ UTR length-triggered pathway, but not the downstream exon-exon junction dependent mode of NMD, is affected by absence of TDRD6 and thus CB distortion. We further show that association of some mRNAs with UPF1 is impaired in Tdrd6-/- spermatids, perturbing mRNA processing. We suggest that in spermatids TDRD6 is required for the specific long 3’ UTR dependent NMD pathway, which most likely acts within the CB.
TDRD6 was proposed to play an architectural role in the assembly of CBs such as a scaffold protein. Morphological studies showed that in Tdrd6 -/- spermatids, the CBs were found less compacted and of lower density [23]. We investigated the contribution of TDRD6 to CB composition by determining the protein constituents of CBs from Tdrd6+/- and Tdrd6 -/- spermatids. Based on a method described previously [15] we isolated CBs from adult Tdrd6+/- and Tdrd6 -/- testes. Testicular cell suspensions were chemically fixed to preserve the CB structures during the subsequent step of cell lysis in high stringency buffer. The lysates were centrifuged at low speed to acquire a CB-rich pellet. By immunostaining of MVH we could observe the presence of large, ring-like CB structures in Tdrd6+/- samples, and in Tdrd6 -/- samples smaller structures that may represent less compacted CBs as expected for in Tdrd6 -/- cells or precursor building blocks of CBs (S1Aii Fig). Next, Tdrd6+/- and Tdrd6 -/- CBs were attached to anti-MVH Dynabeads and immunoprecipitated (Fig 1A). Similar efficiency of immunoprecipitation from Tdrd6+/- and Tdrd6 -/- samples was obtained as seen by immunoblotting of the preparations for MVH (S1B Fig). The immunoprecipitated Tdrd6+/- and Tdrd6 -/- CB samples were resolved in SDS-PAGE gels and subjected to mass spectrometric analysis. Substantial differences in protein content were observed in TDRD6-deficient CBs compared to controls. We found 286 proteins in Tdrd6-/- CB preparations and 254 proteins in Tdrd6+/- CBs (S3 Table). We reckon that TDRD6 is key to supporting a normal protein composition of CBs. Only 96 proteins are present in both samples and thus do not require TDRD6 for their assembly in CBs. Those 96 proteins were excluded form the further analysis and we focused on the 158 proteins enriched exclusively in Tdrd6+/- CB samples (Fig 1B). The 190 proteins not present in unperturbed CBs but present in absence of TDRD6 may result from aberrant associations with MVH made possible by the removal of TDRD6.
We further analyzed the proteome data by focussing on proteins whose presence in the CB depends upon TDRD6. We performed Domain and GO term analysis through the DAVID platform [26] and QIAGEN’s Ingenuity Pathway Analysis (IPA, QIAGEN Redwood City). Analysis of enriched domains revealed that proteins with TUDOR domains, RNA binding domains, or helicase domains are enriched in the Tdrd6+/- CBs (Fig 1C) confirming previous studies [15]. The most represented GO terms and thus molecular and cellular functions are Cell Death and Survival, Cell Development, Protein Synthesis and RNA metabolism such as RNA trafficking, RNA Damage and Repair, RNA Post-Transcriptional Modification (Fig 1D). Thus, the CB localization of proteins bearing RNA binding domains and of proteins that participate in RNA post-transcriptional modification mechanisms was confirmed and is prominently affected by the loss of TDRD6.
Among the proteins identified in the “RNA Post-Transcriptional Modification” group are DEAD box RNA helicases (DDX21, DDX25), snoRNP related proteins (IMP3, DKC1, NOP56, NOP58), rRNA metabolism related proteins (EBNA1BP2, FBL, NCL), pre mRNA binding protein (HNRNPH3), exon junction complex proteins (EIF4A3, CASC3, RBM8A) and RNA decaying enzymes (UPF1, SMG6) (Fig 1E). UPF1 is a key factor of nonsense mediated mRNA decay (NMD). Initially the NMD pathway was considered as a quality control system that recognizes and degrades aberrant mRNAs with truncated open reading frames (ORF) due to the presence of a premature termination codon (PTC) [27]. However, recent studies demonstrated a general role of NMD in post-transcriptional regulation of non-aberrant mRNAs. Upstream ORF (uORF), introns in 3’ UTR and long 3’ UTRs have been identified as features that activate NMD [28]. UPF1 was found to be enriched at long 3’ UTR sequences [29,30] and increased association of UPF1 with the 3’ UTR triggers the decay of the mRNA [31]. Along with UPF1, UPF2 supports the decay of mRNAs with long 3’ UTR [32]. UPF1 was previously shown to be a component of the CB [17] and the fact that we identified UPF1 in intact, but not in disturbed CBs, motivated us to further analyze UPF1 and its partner UPF2 in spermatids.
Generally, very little is known about the presence and function of UPF complexes in germ cell development. To address whether the expression of Upf genes is developmentally regulated in the testis, we isolated and analyzed RNA from testis of successive days post partum, i.e. during the first wave of spermatogenesis and spermiogenesis (S2A Fig). Upf1 and Upf2 are weakly expressed in neonatal testes and their expression increases during the development of spermatocytes and spermatids. Upf1 and Upf2 expression increases moderately in meiotic cells, which were identified by Hormad1 expression at day 10 postpartum (pp) [33]. Later Upf1 and Upf2 expression peaks and coincides with high Tdrd6 expression at day 22 pp, which marks the appearance of early round spermatids [23]. Upf1 and Upf2 expression remain at high levels as round spermatids differentiate to elongated spermatids marked by Prm2 expression form 26 pp onwards. These data suggest that UPF complexes may play an hitherto undescribed role in the late meiotic and postmeiotic stages of spermatogenesis.
The parallel expression of Upf1, Upf2 and Tdrd6 led us to test whether the levels of Upfs are affected by TDRD6 deficiency. We isolated mRNA (S2B Fig) and protein extracts (S2C Fig) from Tdrd6+/- and Tdrd6-/- spermatids and compared mRNA and protein expression of UPFs. The absence of TDRD6 and thus the disruption of CBs did not affect levels of UPFs protein or mRNA.
We investigated the associations—direct or indirect—of TDRD6, MVH, UPF1 and UPF2 in Tdrd6+/- and Tdrd6-/- by co-immunoprecipitation in the presence or absence of RNAse A treatment. RNAse A treatment efficiency was assessed by RNA electrophoresis of the flow-through sample of the IPs (S2D Fig). We investigated the interaction between MVH, UPF1 and UPF2 by performing MVH immuno-precipitation (IP) (Fig 2Ai). Vinculin (VINC) a membrane-cytoskeletal protein was used as a loading and negative co-IP control (Fig 2Aii). The previously reported in vitro interaction between TDRD6 and MVH [23] was recapitulated by co-IP from spermatids and was independent of RNAseA inclusion (Fig 2Aiii). Reverse IP of TDRD6 (Fig 2Bi) showed also that TDRD6 associated with MVH (Fig 2Biii) irrespectively of RNAse A treatment. UPF1 co-IP with MVH (Fig 2Aiv) was observed specifically only in Tdrd6+/- spermatids but not in the Tdrd6-/- spermatids or in IgG control IPs and was dependent on RNA. Reverse IP of UPF1 (Fig 2Ci) showed also its association with MVH (Fig 2Cv), but only when TDRD6 and intact RNA were present. UPF2 co-immunoprecipitated with MVH (Fig 2Av) irrespectively of the genotype and RNAse A treatment and the reverse IP of UPF2 (Fig 2Di) demonstrated its association with MVH (Fig 2Dv). Thus, only UPF1, but not UPF2, requires a TDRD6-supported, intact CB for its association with the key CB component MVH.
Since we observed differential association of UPF1 and MVH in Tdrd6+/- and Tdrd6-/- spermatids, we investigated the interaction between TDRD6, UPF1 and UPF2 by performing TDRD6 IP (Fig 2Bi). UPF1 co-immunoprecipitated with TDRD6 (Fig 2Biv) specifically only in Tdrd6+/- spermatids but not in the Tdrd6-/- spermatids or in IgG control IPs. Reverse IP of UPF1 (Fig 2Ci) showed also its association with TDRD6 (Fig 2Civ). UPF2 co-immunoprecipitated with TDRD6 (Fig 2Bv) and the reverse IP of UPF2 (Fig 2Di) demonstrated its association with TDRD6 (Fig 2Div). The interactions of TDRD6 and UPF1 or UPF2 are resistant to RNAse A treatment. This data suggested the involvement of TDRD6 in complexes containing MVH, UPF1 and UPF2.
UPF1 binds directly to UPF2 via an UPF2-interacting domain [34], but upon IP of UPF1 from Tdrd6+/- and Tdrd6-/- round spermatids (Fig 2Ci), UPF2 was found to co-immunoprecipitate only in the Tdrd6+/- samples, UPF1 interaction with UFP2 is almost entirely abrogated upon loss of TDRD6 (Fig 2Ciii). Confirming these results, IP of UPF2 from Tdrd6+/- and Tdrd6-/- round spermatids (Fig 2Di) showed that UPF2 associated with UPF1 in the Tdrd6+/- sample but hardly in absence of TDRD6 (Fig 2Diii). UPF1-UPF2 association in Tdrd6+/- samples was not affected by the presence of RNAse A as expected. In conclusion, the absence of TDRD6, accompanied by distortion of CB structure, prevented UPF1-MVH and UPF1-UPF2 interactions.
Given the distinctly TDRD6-dependent associations of UPF1 and UPF2 shown above, the localization of UPFs in germ cells was determined by staining meiotic and postmeiotic cells with antibodies against UPF1 and UPF2. The localization of UPF proteins has been extensively investigated in mammalian cell lines where UPF1 is mainly cytoplasmic [35], but a fraction of UPF1 resides in the nucleus where it promotes DNA replication, S phase progression [36] and telomere stability [37]. More recently it was shown that UPF proteins localize to P-bodies in mammalian cells [38]. UPF2 is a cytoplasmic protein [35].
In meiosis I spermatocytes, positive for SYCP3, UPF1 localized to the perinuclear space of the cytoplasm and there was hardly any staining observed in the nucleus (Fig 3A). UPF2 was distributed in some clusters throughout the cytoplasm (Fig 3B). No apparent co-localization with TDRD6 was detected suggesting no participation in the precursor structures of CB in meiotic cells. The localization pattern of UPFs in meiotic cells remained unaffected by the loss of TDRD6 (Fig 3A and 3B).
In Tdrd6+/- mice, UPF1 was absent form the cytoplasm of round spermatids and was exclusively concentrated in CBs where it co-localized with MVH and TDRD6 (Fig 4A and 4Ci). However, in Tdrd6-/- round spermatids UPF1 failed to co-localize with MVH positive foci, i.e. with the distorted CBs found in Tdrd6-/- round spermatids, and remained diffuse in the perinuclear cytoplasm (Fig 4A and 4Cii). This suggested that TDRD6-positive, undistorted CBs are required for UPF1 re-localization from the cytoplasm of meiotic cells to the CBs of round spermatids. On the other hand, UPF2 (Fig 4B, 4Ciii and 4Civ) primarily co-localized with MVH in Tdrd6+/- and Tdrd6-/- CBs. UPF2 is a newly identified component of CBs. 100% (n = 67) and 97% (n = 76) of Tdrd6+/- CBs scored contain UPF1 and UPF2, respectively. In Tdrd6-/- round spermatids 0% (n = 41) of CBs contained mUPF1, while mUPF2 localized to 86% (n = 72) of Tdrd6-/- CBs (S3 Fig). This indicated a TDRD6 independent manner of localization of UPF2 to CBs, although CB presence of these proteins was slightly affected probably by the distorted architecture of the Tdrd6-/- CBs.
UPF1 is a key factor of nonsense mediated mRNA decay (NMD). Initially the NMD pathway was considered as a quality control system that recognizes and degrades aberrant mRNAs with truncated open reading frames (ORF) due to the presence of a premature termination codon (PTC) [27]. PTCs can arise from aberrant splicing events, 5’ UTR upstream open reading frames (uORFs) or by mutations. In principle, a termination codon residing more than ~55 nucleotides upstream of an exon-exon junction complex is considered a PTC and the transcript is a likely target for the so called downstream exon-exon junction stimulated (dEJ) NMD [27]. We hypothesized that mis-localization of UPF1 and failing interaction of UPF1 with UPF2 in the Tdrd6-/- strain would lead to accumulation of NMD sensitive transcripts in Tdrd6-/- round spermatids. To test whether the dEJ mode of NMD was affected by loss of TDRD6, we generated the mRNA profiles of germ cell populations enriched for round spermatids of Tdrd6+/- and Tdrd6-/- mice by deep sequencing. The MACS-purified population was more than 95% positive for expression of the marker hCD4 in both genotypes (S4A Fig), and the hCD4 is expressed at the same levels in Tdrd6+/- and Tdrd6-/- cells (S4B Fig). The preparations enriched for round spermatids contained approximately 70% round spermatids in both Tdrd6+/- and Tdrd6-/- samples (S4C Fig), and the fraction of primary and secondary spermatocytes was about the same. We used 4 biological replicates, i.e. round spermatid samples of four individual animals per genotype and acquired over 250 million RNA seq reads (S5 Fig). We aligned the RNA sequence reads with TopHat, assembled transcripts with Cufflinks and annotated them using Ensemble v67 [39–41]. Expression analysis was performed with Cuffdiff with a FDR of 0.1. We used 2 different approaches to classify transcripts with PTCs, which are putative dEJ NMD targets to be further analyzed. In the first approach, the mouse annotation of Ensembl v67 was used for the classification of the transcripts. Transcripts which had the biotype “Nonsense Mediated Decay” were extracted from the complete data set and used for the subsequent analysis. Here, if the coding sequence of a transcript finishes >50 bp from a downstream splice site, it is tagged as a putative NMD sensitive transcript. In the second approach, SpliceR version 1.12.0 was used for the annotation of transcripts with PTC [42]. We used the Cufflinks results files for SpliceR and filtered the isoforms with the setting "expressedIso" and "isoOK" within SpliceR. Furthermore SpliceR requires CDS information, which was retrieved with SpliceR internal function from UCSC. Annotation of transcripts was done with “annotatePTC” and transcripts were extracted, which were set to PTC equals TRUE. These transcripts were used for the comparison. There are 564 dEJ NMD sensitive transcripts identified only by Ensemble v67 database, 2004 transcripts identified only by the SpliceR pipeline and 1268 identified by both ways (S6A Fig).
1,520 (82%) out of 1,832 dEJ NMD sensitive transcripts by Ensemble v67 have a log2 fold change between -1 and 1 and 1,089 (59%) transcripts have a log2 fold change between -0.5 and 0.5 (Fig 5A). Similarly 2780 (85%) out of 3272 dEJ NMD sensitive transcripts by SpliceR analysis have a log2 fold change between -1 and 1 and 2031 (62%) transcripts have a log2 fold change between -0.5 and 0.5 (Fig 5A) indicating normal regulation of dEJ NMD transcripts between Tdrd6+/- and Tdrd6-/- round spermatids. Next, we looked at the expression values measured in FPKM for the dEJ NMD sensitive transcripts in both genotypes (Fig 5B). A Wilcoxon-Mann-Whitney test (p-value = 0.4789 for the dEJ NMD sensitive transcripts by Ensemble v67 and p-value = 0.8998 for the dEJ NMD sensitive transcripts by SpliceR analysis) showed that there is no difference in the FPKM values of the dEJ NMD sensitive transcripts between the TDRD6-proficient and -deficient samples.
We further examined a number of known NMD substrates to validate the high throughput analysis. Abnormal splicing events such as intron inclusion, exon skipping and splicing downstream of a normal termination codon can induce the dEJ mode of NMD. During an intron inclusion event, a PTC can be introduced either because it resides in the included intron or is generated due to frameshift of the physiological ORF. In an exon skipping event a frameshift of the ORF might produce a PTC. We tested PTC generation by intron inclusion and exon skipping events that characterized for specific transcripts in other murine tissues [43]. Performing RT-PCR using specific primers (arrows), which span intron inclusion events for Pkm2, Srsf2, Srsf3, Hnrpl, Brd2 and exon skipping events for Hnrnph3 and Mdm2, we investigated NMD sensitive transcript variants (marked by arrowheads). NMD sensitive or NMD resistant variants showed the same levels in Tdrd6+/- and Tdrd6-/- samples (Fig 5C).
Auf1 mRNA can also be used as a marker of NMD efficiency due to its unusual 3’ UTR architecture [44]. Splicing of exon 9 and exon 10 generates an exon junction more than 50 nt downstream of the normal termination codon, producing NMD sensitive transcript variants II and III (S6B Fig and [44]). We designed specific primers to map different splicing events and found that splicing events producing the NMD sensitive transcripts II and III occur largely the same in Tdrd6+/- and Tdrd6-/- samples (S6C Fig).
Finally, uORFs of a transcript would lead to premature translational termination and subsequent NMD. We compared the expression of transcripts with uORFs such as Atf5, Map3k14, Arfp1 and Dusp10, which were previously shown to be recognized by NMD in other cell types [45,46] by RT-qPCR. We found no difference in their expression levels in Tdrd6+/- and Tdrd6-/- round spermatids (Fig 5D). Together these data showed normal function of the downstream exon-exon junction dependent mode of NMD in TDRD6 deficient spermatids.
Although NMD was initially characterized in PTC dependent mRNA degradation as a quality control mechanism, there is evidence that NMD is implicated in the metabolism of normal mRNAs. A well studied feature of mRNAs which can elicit NMD is the long 3‘ UTR. UPF1 was found to be enriched at long 3’ UTR sequences [29,30] and increased association of UPF1 with the 3’ UTR triggers the decay of the mRNA [31] in an UPF2 and SMG6 dependent way [32]. Since we found TDRD6 to associate with UPF1 and UPF2, we assessed the effect of TDRD6 deficiency on the general mRNA transcriptome. We analyzed the expression of normal mRNAs in the transcriptome data described above, derived from germ cell populations enriched for round spermatids from Tdrd6+/- and Tdrd6-/- mice. We aligned the RNA seq reads with TopHat, assembled transcripts with Cufflinks and annotated them using Ensemble v67 [39–41]. Expression analysis was performed with Cuffdiff with a FDR of 0.1 and we found 2704 transcripts to be significantly (p-value <0.05) mis-regulated in absence TDRD6 and thus of intact CBs. More specifically, 1375 were up-regulated and 1329 down-regulated in Tdrd6-/- round spermatids (Fig 6A and S1 Table). Thus, TDRD6 is required for the presence of a proper mRNA repertoire in spermatids.
To further characterize the changes of the mRNA content of CB-disrupted spermatids, we grouped mis-regulated transcripts with p-values <0.05 into 3 groups according to the length of their 3’ untranslated region (UTR): short 3’ UTR <350 nt, medium 3’ UTR >350 nt and <1500 nt and long 3’ UTR >1500 nt. We analyzed the log2 fold distribution of these groups of transcripts. We observed that the majority of mis-regulated transcripts (514, 82% out of 628) with a long 3' UTR had a positive log2 fold change, i.e. they are present at higher levels in the Tdrd6-/- compared to the Tdrd6+/- round spermatids. The distribution of positive and negative log2 fold change of mis-regulated transcripts was not significantly altered for short and medium 3‘UTR length groups, but the log2 fold distribution of transcripts with long 3’ UTR, showing enrichement of upregulated transcripts in Tdrd6-/- samples was statistically different from the others (Wilcoxon-Mann-Whitney test p-value <2.2−16) (Fig 6B and S2 Table). We conclude that the mis-regulation of mRNAs in TDRD6 deficient, CB-disrupted spermatids correlates with an accumulation of transcripts carrying long 3’ UTRs.
Significantly mis-regulated transcripts with long 3’ UTRs >1500 nt correspond to 628 genes. 288 (46%) genes have a single transcript with long 3 UTR and 340 (54%) genes have multiple transcripts and transcripts with long 3’ UTRs among them (S6D Fig). These 340 genes code for 1176 putative transcript isoforms. These putative isoforms include the 340 long 3’ UTR transcripts tested previously, but in addition there are 167 isoforms that have 3’ UTRs shorter than 1500 nt and for the rest there is no reliable information on the 3’ UTR length. From the 167 short 3 ‘UTR isoforms of the genes with mis-regulated long 3’ UTR transcript isoforms, there are 15 transcripts, corresponding to 14 genes, showing a significant mis-regulation, while the large majority of 152 transcripts remained unchanged. Thus loss of TDRD6 affects specifically the long 3’ UTR isoforms of genes with multiple isoforms with different 3’ UTR lengths.
Further, we used the database of EMBL-EBI Expression Atlas and looked the expression analysis of different murine tissues. There are 10760 genes expressed in testis above a standard expression cutoff value of 0.5. We consider a transcript to be testis-specific when its expression is 10 times higher in testis than in any other tissue examined. There are 1851 genes that fall in this group. Of the 628 mis-regulated transcripts with long 3’ UTR there are 35 which can be considered testis specific (6%). 514 transcripts with long 3‘ UTRs are higher in the Tdrd6-/- round spermatids and 15 of them are testis-specific (3%). 113 transcripts with long 3’ UTRs are lower in the Tdrd6-/- round spermatids and 20 of them are testis-specific (17%).
UPF1 is an RNA helicase that can bind to all transcripts, although it preferentially associates with transcripts carrying long 3’ UTRs [30,47]. Our initial observation that mRNAs with long 3’ UTR tend to be up-regulated in CB distorted round spermatids led us to investigate particular mRNAs with 3’ UTR length >1000 nt with respect to UPF1 binding, mRNA levels and translational potential. We assessed the in vivo binding of UPF1 to selected mRNAs by performing anti-UPF1 RNA immunoprecipitation (RIP) from Tdrd6+/- round spermatids, followed by RT PCR. We examined 9 transcripts that carried long 3’ UTRs >1000 nucleotides and 2 transcripts with 3’ UTRs <350 nucleotides as a negative control. Positive signals in the anti-UPF1 RIP RT-qPCR were obtained for transcripts with long 3’ UTR as Spen (3’ UTR length = 1016 nt), Diap1 (3’ UTR length = 2244 nt), Mdc1 (3’ UTR length = 2040 nt), Ube2c (3’ UTR length = 1598 nt), Twsg1 (3’ UTR length = 3218 nt), Dixdc1 (3’ UTR length = 3438 nt), Daam1 (3’ UTR length = 2478 nt), Yap1 (3’ UTR length = 2497 nt) (Fig 7A) and Wnt3 (3’ UTR length = 1889 nt) (S7A Fig) and for transcripts with short 3’ UTR as Ecsit (3’ UTR length = 78 nt), Prss51 (3’ UTR length = 119 nt) (S7A Fig), suggesting that UPF1 binds in vivo to all these transcripts.
Next, we expanded the analysis of mRNA to UPF1 binding in the Tdrd6+/- versus Tdrd6-/- round spermatids. Significantly decreased binding to UPF1 in Tdrd6-/- round spermatids was observed for transcripts with long 3’ UTR such as Spen, Mdc1, Diap1, Ube2c, Twsg1, Dixdc1, Daam1 and Yap1 (Fig 7Bi–7Bviii). To assess the effect of impaired UPF1-mRNA binding on the mRNA levels, we performed RT-qPCR. The presence of mature mRNAs of Spen, Mdc1, Diap1, Ube2c, Twsg1, Dixdc1, Daam1 and Yap1, i.e. the transcripts with long 3’ UTR showing decreased association with UPF1 in Tdrd6-/- samples, was increased 2- to 3-fold in Tdrd6-/- round spermatids (Fig 7C). The pre-mRNA levels of these genes remained unchanged (Fig 7D), showing that the higher levels were not caused by increased transcription, but by increased stability. These results suggested that UPF1 binding to mRNAs carrying long 3’ UTR is perturbed upon TDRD6 deletion—and thus distortion of the CB and mis-localization of UPF1 –and correlates with increased mRNA stability possibly through decreased degradation.
We identified short 3’ UTR such as Ecsit and Prss51 (S7Bi and S7Bii Fig) and long 3‘UTR such Wnt3 (S7Bii Fig) to associate with UPF1 equally between Tdrd6+/- and Tdrd6-/- round spermatids. The expression levels of transcripts that show unaffected association with UPF1 among the different genotypes were mis-regulated both on the mature mRNA (S7C Fig) and nascent pre mRNA levels (S7D Fig) possibly through indirect mechanisms.
In RIP experiments using anti UPF2 antibody, we found that most of the mRNAs with long 3’ UTR that associate with UPF1 also associate with UPF2 in round spermatids (Fig 7E). The analysis of binding of UPF2 to long 3’ UTR mRNAs in the Tdrd6+/- versus Tdrd6-/- round spermatids showed significantly decreased binding of Spen, Mdc1, Diap1, Ube2c, Dixdc1 and Yap1 to UPF2 in absence of TDRD6 (Fig 7Fi–7Fvi). Thus, both UPF1 and UPF2 associations with long 3’ UTR mRNAs are affected by loss of TDRD6.
To assess the effect of impaired UPF1-mRNA binding on the translation potential of UPF1 bound mRNAs, we performed sucrose gradient fractionation to isolate translationally active fractions, which are those rich in polysomes (fractions #1–7), translationally inactive fractions that are rich in ribosomal subunits (fractions #8–10), and ribosome-free mRNPs (fractions # 11–12) (Fig 8A). Although the majority of UPF1 protein from yeast and human cell line cultures was shown to associate with polysomes [48], we found that in 26 dpp murine testis, enriched for round spermatids, UPF1 was underrepresented in polysome/ribosome fractions #1–10, confirmed by the presence of RPS6. The majority of UPF1 was detected in fractions #11–12 containing ribosome-free mRNPs, indicated by GAPDH. The same distribution was observed for MVH. The UPF1 and MVH association with ribosome-free mRNPs was not compromised by absence of TDRD6 (Fig 8B). To test the translational capacity of UPF1-associated mRNA species we extracted RNA from each fraction and performed RT-PCR. The transcripts with long 3’ UTR, which associated less with UPF1 in Tdrd6-/- samples such as Spen (Fig 8C and 8D), Diap1 (S8A and S8B Fig) Mdc1 (S8A and S8C Fig), showed relatively equal distribution in translationally active fractions #1–7 (52% for Spen, 43% for Diap1 and 53% for Mdc1) and translationally inactive ribosome fractions #8–12 (48% for Spen, 57% for Diap1 and 47% for Mdc1) in Tdrd6+/- samples. In contrast, in Tdrd6-/- samples, Spen (Fig 8C and 8D), Diap1 (S8A and S8B Fig) and Mdc1 (S8A and S8C Fig) showed increased abundance in the translationally active fractions #1–7 (75% for Spen, 68% for Diap1 and 59% for Mdc1) and decreased abundance in translationally inactive ribosome and ribosome-free fractions #8–12 (25% for Spen, 32% for Diap1 and 41% for Mdc1). Other transcripts with long 3’-UTRs, i.e Twsg1 (S8A and S8D Fig) and Yap1 (S8A and S8E Fig), are underrepresented in translationally active fractions #1–7 (37% for Twsg1 and 38% for Yap1) and overrepresented in translationally inactive fractions #8–12 (63% for Twsg1 and 62% for Yap1) in Tdrd6+/- samples. On the other hand, in Tdrd6-/- samples, Twsg1 (S8A and S8D Fig) and Yap1 (S8A and S8E Fig) mRNAs are more abundant in the translationally active fractions #1–7 (49% for Twsg1 and 48% for Yap1) and almost equals the abundance in translationally inactive fractions #8–12 (51% for Twsg1 and 52% for Yap1). Together this shows that mRNAs with reduced UPF1 binding in Tdrd6-/- spermatids such as Spen, Diap1, Mdc1, Twsg1 and Yap1 associated to a larger extent with polysomal fractions compared to the controls, suggesting these mRNAs were more actively translated in Tdrd6-/- spermatids.
On the other hand, mRNAs which showed no difference in UPF1 binding between Tdrd6+/- and Tdrd6-/- mice, such as Ecsit (S7Ei and S7Eii Fig), Prss51 (S7Ei and S7Eii Fig) and Wnt3 (S7Ei and S7Eii Fig) displayed a similar distribution pattern across all fractions in Tdrd6+/- and Tdrd6-/- samples (Ecsit: 45% in translationally active fractions #1–7 and 55% in translationally inactive fractions #8–12 in Tdrd6+/- and 52% in translationally active fractions #1–7 and 48% in translationally inactive fractions #8–12 in Tdrd6-/-; Prss51: 47% in translationally active fractions #1–7 and 53% in translationally inactive fractions #8–12 in Tdrd6+/- and 41% in translationally active fractions #1–7 and 59% in translationally inactive fractions #8–12 in Tdrd6-/- Wnt3: 43% in translationally active fractions #1–7 and 57% in translationally inactive fractions #8–12 in Tdrd6+/- and 35% in translationally active fractions #1–7 and 65% in translationally inactive fractions #8–12 in Tdrd6-/-). Overall, this data set shows that decreased binding of mRNA to UPF1 in Tdrd6-/- round spermatids correlates with increased mRNA stability and translational potential.
The aim of the present study was to define the role of a germ cell-specific protein, TDRD6, which in spermatids resides in the CB and is a main structural component of this cell organelle whose functions remained hitherto largely unknown. CBs are considered large RNP complexes in the cytoplasm close to the nuclei of round spermatids. CBs were proposed to be sites of accumulation of mRNPs exported from the nuclei [49]. It was postulated that these mRNPs are translationally repressed through piRNAs or miRNAs or by translational regulators such as Nanos, Pum and Gemin3 [49,50]. These mRNPs would be stored or targeted to other cytoplasmic sites [51]. The dominance of Tudor domain proteins in the CB and their interactions with PIWI and other proteins was suggested to provide the molecular scaffold for CB [18]. TDRD6 and TDRD7 were shown to be indispensable for CB architecture [20,23], and in our study we used TDRD6 deficient mice. The CB has been implicated in piRNA biogenesis and retrotransposon silencing [19]. The loss of TDRD6 results in male infertility and disruption of CB architecture of which a remnant “ghost body” is left. Genome methylation remains normal as did retrotransposon silencing, which depends on MIWI and MILI, suggesting that the proper architecture of the CB is required for other functions.
To decipher some of these functions, we used proteome analysis to determine differences between Tdrd6+/- and Tdrd6-/- CB compositions and determined distinct perturbations of the CB proteome in TDRD6-deficient samples. In CB preparations we identified 158 proteins that depend on TDRD6 for their enrichment in CBs. To compare these with the transcriptomics data, the 158 IPI protein IDs were converted to 139 Ensemble transcript IDs (88%). Next we looked at the expression values of these transcripts in our RNA deep sequencing analysis of the Tdrd6+/- and Tdrd6-/- round spermatid transcriptomes. Out of the total of 139 transcripts, 96 remained unchanged between the genotypes (70%). 27 transcripts expressed lower in the Tdrd6-/- round spermatids (19%) and 16 transcripts expressed higher in the Tdrd6-/- round spermatids (11%). The failure to identify the 24 transcripts, that expressed lower in Tdrd6-/- round spermatids, in the Tdrd6-/- CB may be due to the very low expression levels. However, the the vast majority of the proteins (82%) are normally or higher expressed in Tdrd6-/- round spermatids, so the failure to identify them in Tdrd6-/- CB is likely a consequence of CB distortion in this mutant.
Many of the proteins absent in TDRD6-deficient CBs bear RNA binding domains pointing to a critical role of CBs in RNA metabolism, i.e. post transcriptional regulation.
Among the RNA-related proteins found within the CB proteome were components of the RNA degrading machinery, which we further investigated. Our data are in agreement with a recent study [17] that provided first insights into the molecular composition of CBs. A major subset of proteins that localizes to CBs are those implicated in RNA degradation processes e.g. mRNA decapping enzyme DCP1a and RNA endonuclease SMG6. Indeed CBs share common features with P-bodies and the NMD core factor UPF1 was found in both structures [14,38]. Here we demonstrate that key NMD factors UPF1 and UPF2 are highly expressed in post-meiotic male germ cells and accumulate in CBs implying a key role of NMD for the completion of spermatogenesis. We provide evidence that the localization of UPF1 to CBs depends on a TDRD6-supported CB structure, while UPF2 is targeted to CBs via different mechanism(s). We analyzed protein-protein interactions in the presence and absence of TDRD6. In the wild-type situation, MVH and UPF1 associated with UPF2 and TDRD6. In absence of TDRD6, MVH and UPF2 interacted with each other localizing to the CB “ghost body”, but UPF1 failed to associate with them. Thus, TDRD6 supports the formation of UPF1-containing mRNPs in the CBs.
It is very unlikely that TDRD6 itself binds directly to RNA, since there are no RNA binding domains identified in this protein. TUDOR domains bind to methylated arginines or lysines [18]. One may speculate that such methylated residues in UPF1 would enable UPF1-TDRD6 interaction, which is subject of future investigations. In any case, TDRD6 likely provides a protein scaffold, where RNA binding proteins are brought into proximity so that correctly assembled mRNPs can be formed and stabilized.
Accumulation of UPF proteins in CBs indicated that CBs support NMD, for example CBs may serve as storage sites for NMD proteins or even as sites of active NMD. The loss of TDRD6 and subsequent perturbation of UPF1 interactions did not affect the levels of PTC containing transcripts, thus did not affect PTC induced, downstream exon-exon junction dependent NMD. In Tdrd6-/- round spermatids dEJ-triggered NMD is functional despite the compromised interaction of UPF1 and UPF2, suggesting an alternative pathway of UPF1 activation on a PTC containing transcript. However, we observed increased levels of transcripts with long 3’ UTR in Tdrd6-/- sample, suggesting that TDRD6 supports the long 3’ UTR triggered pathway of NMD. To our knowledge, this is the first mutant that discriminates between different modes of stimulating NMD. We also demonstrate that specific mRNAs with long 3‘ UTR associate with UPF1 and UPF2 in vivo in round spermatids, but this association is much reduced in Tdrd6-/- cells. The reduced association with UPF1 correlated with increased levels of these mRNAs and their increased translational potential in the Tdrd6-/- background. The presence of a few mRNAs with either long or short 3‘ UTR that bind to UPF1 in a TDRD6-independent manner but are nevertheless altered in levels in absence of TDRD6 suggests that TDRD6 regulates the levels of some mRNAs independently of UPF1 through a distinct pathway. It has been shown that the average 3‘ UTR length of transcripts required for spermiogenesis is shorter compared to transcripts required for pre-meiotic, meiotic or testicular cell development [8]. Transcripts with shorter 3’ UTR may be more stably stored for longer periods and thus may be particularly competent for efficient translation during the last stages of spermiogenesis.
NMD is important for many developmental processes as systemic depletion of the murine Upf1 gene results in complete loss of NMD and leads to post implantation embryonic death [52]. NMD is essential for hematopoietic stem cells and for B and T lymphocyte maturation, since conditional ablation of murine UPF2 in the hematopoietic system is detrimental to proliferation of progenitor cells and leads to up regulation of aberrant TCR and Ig locus recombination products [53]. On the other hand, NMD activity is down-regulated in neural stem cell upon neurogenic signaling to allow differentiation [54]. Thus, tissue- and cell-type specific roles of NMD exist, but are known in only a few instances. We provide the first evidence of NMD functioning in the regulation of transcripts during spermiogenesis. Successful completion of the spermiogenic program depends strongly on post-transcriptional regulation as the transcriptional production of RNA ceases from the mid to later stages because of the extensive nuclear compaction.
The use of mice was approved by the State of Saxony animal welfare officials, Az DD24-5131/339/6 and was performed according to the national and EU guidelines.
Construction of TDRD6‐deficient mice was described previously [23]. In all experiments, except otherwise noted, testes from postnatal day 26 (P26) Tdrd6+/- and Tdrd6-/- mice were dissected to be enriched in round spermatid cells. Tdrd6+/- mice used as control for the experiments do not exhibit any phenotype and provide the targeting vector with the hCD4 gene in frame with the Tdrd6 5′ UTR and ATG (start) codon, that allows isolation of TDRD6 expressing cells through an anti hCD4-MACS approach [23]. For cell preparations enriched in round spermatids, the Tunica albuginea was removed and seminiferous tubules resuspended in 10 ml PBS and passed subsequently through 100μm and 40μm stainers. Cells were washed once with PBS and hCD4-positive cells were magnetically labeled with CD4MicroBeads (Miltenyi Biotec) and MACS isolated (Miltenyi Biotec) according to manufacturer instructions.
Testes for immunostaining were fixed in freshly prepared 4% PFA for 1h on ice, briefly washed with PBS, and incubated O/N in 30% sucrose. Testes were embedded in OCT blocks, frozen on dry ice, and cryo-sectioned at 7 μm thickness.
CBs were isolated according to Meikar et al. (2010) with some modifications. hCD4 positive cells from Tdrd6+/- and Tdrd6-/- adult mice were fixed in 1% PFA (Sigma) solution for 10 min at RT. The reaction was stopped by adding glycine (Roth) pH 7 to a final concentration of 0.25 M. The fixed cells were lysed by sonication in 0.5 mL of RIPA buffer (50 mM Tris-HCl at pH 7.4 (Roth), 150 mM NaCl (Roth), 1% NP-40 (Sigma), 0.5% sodium deoxycholate (Sigma), 0.1% SDS (Roth), 1 mM EDTA (Sigma), 1 mM DTT (Roth), 5mM NaF (Sigma), 1mM Na2VO3 (Sigma), 1mM PMSF (Sigma), 1x protease inhibitor cocktail complete mini (Roche)) supplemented with 100U RNAse inhibitor (Invitrogen). The lysate was centrifuged at 300g for 10 min and the CB enriched pellet resuspended in 0.5 mL of RIPA buffer. The CBs were immunoprecipitated using Dynabead Protein G (Invitrogen) coupled to rabbit polyclonal anti-MVH (Abcam) O/N at 4°C. Dynabeads were washed 4 times with RIPA buffer and the crosslinks of the isolated CBs were reversed by incubation at 70°C for 45 min in 1x Laemmli buffer.
CB samples were separated in mini-protean TGX pre-cast gradient gels (BioRad) and stained with SimplyBlue SafeStain (Life Technologies). Gel pieces were excised from the sample lanes, followed by in-gel digestion with trypsin (Promega) and extraction of the peptides. The peptides were analyzed using LC-MS/MS with an Ultimate 3000 (Dionex Corp, Sunnyvale CA) nanoLC system connected to a LTQ Orbitrap mass-spectrometer (ThermoScientific Corp., San Jose CA) equipped with an automated nanoelectrospray ion source TriVersa (Advion BioSciences, Ithaca NJ). All MS/MS samples were analyzed using Mascot (Matrix Science, London, UK; version 2.2.04). Mascot was set up to search the ipi.MOUSE_V3.76_20110304 database assuming the digestion enzyme trypsin. Mascot was searched with a fragment ion mass tolerance of 0.50 Da and a parent ion tolerance of 5.0 PPM. Oxidation of methionine and propionamide of cysteine were specified in Mascot as variable modifications. Scaffold (version Scaffold_3.6.4, Proteome Software Inc., Portland, OR) was used to validate MS/MS based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95% probability as specified by the Peptide Prophet algorithm [55]. Protein identifications were accepted if they could be established at greater than 99.0% probability and contained at least 2 identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm [56].
Scaffold normalizes the MS/MS data between samples. Normalization is done on the MS sample level, which is the total sample run through the mass spectrometer. The normalization method that Scaffold uses is to sum the “Unweighted Spectrum Counts” for each MS sample. For the purposes of protein identification, Scaffold uses a ProteinProphet model, assigning the peptide exclusively to the protein with the most evidence. The result is that the peptide has a weight of 1 in one protein and a weight of zero in all other proteins. However, if there are two proteins, and each protein has the same peptide, then each spectrum for this peptide has ions contributed from both proteins. The “Unweighted Spectrum Count” option on Scaffold's Samples page will count this spectrum twice, once in the first protein and once in the second protein. This count is “unweighted” in the sense that the spectrum counts the same in each of the shared proteins. Scaffold counts unweighted spectra for determining protein abundance. These sums are then scaled so that they are all the same. The scaling factor for each sample is then applied to each protein group and adjusts its “Unweighted Spectrum Count” to a normalized “Quantitative Value”.
International Protein Index (IPI) accession numbers of proteins identified more than 2 fold enriched in Tdrd6+/- CB (S3 Table) were uploaded to DAVID platform [26], functional annotation for protein domains from PFAM database was performed with threshold count 3 and threshold EASE 0.1. The same list was uploaded to QIAGEN’s Ingenuity Pathway Analysis (IPA, QIAGEN Redwood City) and functional analysis was performed with custom parameters.
Immunofluorescence labeling of frozen sections of mouse testis was performed using rabbit polyclonal anti UPF1, rabbit polyclonal anti UPF2 [35], rabbit polyclonal anti MVH (Abcam), mouse monoclonal anti SYCP3 [57], guinea pig polyclonal anti C-term TDRD6 (this study). Sections were fixed using 4% PFA for 20 minutes, blocked and permeabilized with 2% BSA, 0.1% Triton-X100 in PBS and incubated overnight with primary antibodies. Slides were washed with PBST and probed for 2 h with secondary antibodies Alexa-566-labeled goat anti guinea pig, Alexa-488-labeled goat anti rabbit, or Alexa-647-goat anti mouse (Molecular probes, Invitrogen). For double immunostaining rabbit polyclonal antibodies were labeled using the Zenon Rabbit IgG Labeling Kits (Molecular Probes, Invitrogen). Slides washed again with PBST and nuclei were visualized with DAPI. Images acquired with a Zeiss LSM 510 confocal microscope and quantification of signal intensity was done with ImageJ.
Tdrd6+/- and Tdrd6-/- round spermatid cell suspension fixed in 1% PFA solution for 10 min at RT to capture RNA-protein and protein-protein interactions. Cells were lysed in 0.5 mL RIPA buffer supplemented with 100U RNAse inhibitor (Invitrogen) for 20 min on ice. The lysate was centrifuged 1000 rpm for 10’ at 4°C and the protein concentration of the supernatant was quantified. 150 μg of protein extract diluted in IP buffer (50 mM Tris-HCl at pH 7.4 (Roth), 150 mM NaCl (Roth), 0.25% Triton-X100 (Sigma), 1 mM EDTA (Sigma), 5mM NaF (Sigma), 1mM Na2VO3 (Sigma), 1mM PMSF (Sigma), 1x protease inhibitor cocktail complete mini (Roche)) to a final volume of 250 μL and antibodies coupled to Dynabeads used for immunoprecipitation: goat polyclonal anti UPF1 (Bethyl), rabbit polyclonal anti MVH (Abcam), rabbit polyclonal anti TDRD6 (Antibody Verify), rabbit polyclonal anti UPF2 [35], goat IgG (Invitrogen) and rabbit IgG (Invitrogen). The beads were then washed 4 times with IP buffer. The retained proteins were resolved by SDS-PAGE and immunoblotted with with the aforementioned antibodies and mouse monoclonal anti Vinculin (Sigma).
3 μg of total RNA per sample were used for library preparation. Ribosomal RNA was depleted by using the GeneRead rRNA Depletion Kit (Qiagen). RNA fragmentation, cDNA synthesis and further RNA-Seq library preparation was done with the NEBNext Ultra Directional RNA Library Prep Kit (New England Biolabs). After enrichment and XP bead (Agencourt AMPure Kit; Beckman Coulter, Inc.) purification, quality control was done using Fragment Analyzer (Advanced Analytical). The bar-coded libraries were equimolarly pooled and subjected to 75 bp single-end sequencing on Illumina HiSeq 2000, resulting in an average of 33 million reads per sample. Sequencing raw data were deposited in GEO database under the GSE63948 accession number. The “Tuxedo Suite” of Bowtie, TopHat, Cufflinks and Cuffdiff [39,40,58,59] was used for the alignment and expression analysis. We aligned the samples separately to the mm9 genome using the splice junction mapper Tophat (version 2.0.9), which used Bowtie 2 (version 2.1.0) for mapping. The Ensembl version 67 [41] was used as a support for the annotation during the alignment.
Total RNA was extracted using the TRIZOL Reagent (Invitrogen), according to the manufacturer’s instructions. The concentration and purity of the RNA samples were determined using spectrophotometer scan in the ultraviolet (UV) region. Total RNA (1 μg) was reverse transcribed (RT) with SuperScript II Reverse Transcriptase (Invitrogen) using random primer mix (NEB) according to manufacturer’s instruction. RT PCR amplification was carried out as follows with specific primers (S4 Table): 30” at 95°C, 20” at 60°C, and 30” at 72°C, for 30 cycles using DreamTaq Green DNA Polymerase (Fermentas). RT PCR products were visualized on 1% agarose gels by ethidium bromide staining. RT-qPCR amplification was carried out as follows with specific primers (S4 Table): 5” at 95°C and 30” at 60°C for 40 cycles using GoTaq qPCR Master Mix (Promega). Data analyses was performed with the ddCT method and the unpaired, one tail t-test was implemented.
We performed anti UPF1 RNA immunoprecipitation according to [30] with modifications. Briefly, testicular cell suspension was prepared in 20 ml ice-cold PBS and subjected three times to 150 mJ/cm2 UV-C light (Stratagene Stratalinker 1800). After irradiation, hCD4 positive cells were selected as described above and lysed in 0.75 ml RIPA buffer supplemented with 100U RNAse inhibitor (Invitrogen) for 20 min on ice. The cell lysate was centrifuged at 13,000g for 10 min (4°C). The supernatant was split in 3 samples: 0.25 ml for input, 0.25 ml for anti UPF1 RIP and 0.25 ml for control IgG RIP. RIP samples were diluted with IP buffer to a final volume of 2 ml and pre-cleared with 15 μl Dynalbeads Protein G (Life Technologies). Then, 5 μL of anti UPF1 antibody (Bethyl) or normal goat IgG (Santa Cruz) were added and rotated at 4°C for 4 h. Afterwards, 15 μl Dynalbeads Protein G (Life Technologies) were added and incubated at 4°C for 1 h. After IP, the beads were washed four times with IP buffer and incubated with 1 mg/ml Proteinase K (Roth). Then, RNA extraction, RT and RT-qPCR were performed as described above. RIP RT-qPCR data analysis was performed with fold enrichment method. Briefly, each RIP RNA fractions’ CT value was normalized to the Input RNA fraction Ct value for the same RT-qPCR assay (ΔCt) to account for RNA sample specific differences as ΔCt [normalized RIP] = (Ct [RIP]—(Ct [Input]—Log2 (Input Dilution Factor))). Then, the normalized [RIP] fraction Ct value was adjusted to the normalized background [control Ab RIP] fraction Ct value (ΔΔCt) as ΔΔCt[RIP/control RIP] = ΔCt [normalized RIP]—ΔCt [normalized control RIP]. Fold enrichment above the sample specific control was calculated as linear conversion of ΔΔCt: Fold enrichment = 2 (-ΔΔCt[RIP/control RIP]). RIP assays were conducted in 3 biological replicates and unpaired one-tailed t-test was implemented.
For UPF2 RIP, Tdrd6+/- and Tdrd6-/- MACS enriched round spermatid cell suspension fixed in 1% PFA solution for 10 min. After fixation, the cells were lysed in 0.45 ml RIPA buffer supplemented with 100U RNAse inhibitor (Invitrogen) for 20 min on ice, followed by 2x15 s sonication. Cell lysate was centrifuged at 1,000g for 10 min (4°C). The supernatant was split in 3 samples: 0.15 ml for input, 0.15 ml for anti UPF2 RIP and 0.15 ml for control IgG RIP. RIP samples were diluted with IP buffer to a final volume of 0.5 ml and 15 μl of serum containing rabbit polyclonal anti UPF2 or normal rabbit IgG (Santa Cruz) were added and rotated at 4°C for 16 h. Afterwards 30 μl Dynalbeads Protein G (Life Technologies) were added and incubated at 4°C for 2 h. After IP, the beads were washed six times with IP buffer and incubated with 1 mg/ml Proteinase K (Roth). RNA extraction, RT, RT qPCR and analysis were performed as described for UPF1 RIP.
Testicular extracts from Tdrd6+/- and Tdrd6-/- mice (P26) were subjected to sucrose gradient fractionation as described previously [60]. Briefly, testicular lysates (100 mM NaCl, 10 mM MgCl2, 20 mM HEPES, pH 7.6, 0.5% Triton X-100, 200U RNAseOUT) were centrifuged at 13,000 × g at 4°C for 2 min, and the supernatant was applied to the top of a 15–40% linear sucrose gradient. The gradient was centrifuged at 115,000 × g for 200 min (Beckman Coulter). Absorbance tracing at A254 was obtained with 759A Absorbance Detector (Applied Biosystems) and twelve fractions (1 mL) were collected manually. RNAs were extracted from 0.5 ml of each fraction using the TRIZOL Reagent (Invitrogen). Reverse transcription and RT PCR reactions performed as described above. Proteins were separated by SDS/PAGE, and Western blots were probed with rabbit polyclonal anti RPS6 (Antibody Verify) mouse monoclonal anti GAPDH (Santa Cruz), goat polyclonal anti UPF1 (Bethyl) and rabbit polyclonal anti MVH (Abcam).
Cell preparations from total testis or MACS purified hCD4+ were stained with FITC-anti-Human CD4 for 20 min at 4°C and subsequently with 1 μg/ml Hoechst 33342 for 30 min at 32°C. Cells were washed with PBS and resuspended in FACS buffer (PBS, 1% BSA and 1mM EDTA). Before the analysis, 1 μg/ml PI was added to exclude dead cells. Stained cells were analyzed on a BD LSRII (BD Biosciences) using FACSDiva software (BD Biosciences). Data were analyzed using FlowJo software (TreeStar).
|
10.1371/journal.pbio.1002274 | Scaling the Drosophila Wing: TOR-Dependent Target Gene Access by the Hippo Pathway Transducer Yorkie | Organ growth is controlled by patterning signals that operate locally (e.g., Wingless/Ints [Wnts], Bone Morphogenetic Proteins [BMPs], and Hedgehogs [Hhs]) and scaled by nutrient-dependent signals that act systemically (e.g., Insulin-like peptides [ILPs] transduced by the Target of Rapamycin [TOR] pathway). How cells integrate these distinct inputs to generate organs of the appropriate size and shape is largely unknown. The transcriptional coactivator Yorkie (Yki, a YES-Associated Protein, or YAP) acts downstream of patterning morphogens and other tissue-intrinsic signals to promote organ growth. Yki activity is regulated primarily by the Warts/Hippo (Wts/Hpo) tumour suppressor pathway, which impedes nuclear access of Yki by a cytoplasmic tethering mechanism. Here, we show that the TOR pathway regulates Yki by a separate and novel mechanism in the Drosophila wing. Instead of controlling Yki nuclear access, TOR signaling governs Yki action after it reaches the nucleus by allowing it to gain access to its target genes. When TOR activity is inhibited, Yki accumulates in the nucleus but is sequestered from its normal growth-promoting target genes—a phenomenon we term “nuclear seclusion.” Hence, we posit that in addition to its well-known role in stimulating cellular metabolism in response to nutrients, TOR also promotes wing growth by liberating Yki from nuclear seclusion, a parallel pathway that we propose contributes to the scaling of wing size with nutrient availability.
| What mechanisms control the sizes of animal organs? It is known that organ growth is the product of two systems: an intrinsic system that coordinates cell proliferation with the specification of cell fate (patterning), and an extrinsic system that synchronizes growth with nutrient levels. Developing organs integrate these two inputs to ensure that properly proportioned structures develop which are of the right scale to match overall body size. However, the mechanisms used to integrate these distinct growth control systems have remained largely mysterious. In this study, we have addressed how intrinsic and extrinsic systems combine to drive growth of the Drosophila wing. Focusing on the Target of Rapamycin (TOR) pathway—a major, nutrient-dependent regulator of organ growth—and Yorkie—the transcriptional activator downstream of the Hippo pathway and a key, organ-intrinsic growth regulator—we have identified a circuit in which TOR activity limits Yorkie’s capacity to promote wing growth, in part through a novel mode of transcription factor regulation that we term “nuclear seclusion.” We find that inhibiting TOR leads to the retention of Yorkie in the nucleus but diminishes its transcriptional activity by diverting it away from target genes. We posit that subjugating Yorkie in this way contributes to how fluctuations in TOR activity scale wing size according to nutrient levels.
| A universal property of animal development is the capacity to scale body size and pattern in response to environmental conditions as well as during evolution [1]. For example, when starved during the larval growth phase, Drosophila adults emerge at around one quarter of their normal size but are correctly proportioned. Likewise, Drosophila species differ over ~5-fold in body size but are highly similar in shape. The overt likeness of primate skeletons (differing 18-fold in length), as well as those of frogs (30-fold) and fish (1,600-fold) provide more dramatic demonstrations of scaling across vast taxonomic groups, but despite its generality, the genetic control of organ scaling is poorly understood. Animals possess distinct systems for controlling growth locally and systemically: organ-intrinsic signaling mechanisms couple growth to patterning and morphogenesis, defining organ shape and dimension [2–4]; conversely, humoral signals, produced on feeding, act globally to control body size [5]. Yet, the existence of scaling implies the two systems cannot be independent. Cells in developing organs must integrate local and global information and proliferate accordingly, generating organs—and entire animals—that are functioning, proportional wholes [6–9].
The Drosophila wing is a classical paradigm of organ growth [2,3]. Here, as in other animals, nutrients influence growth via Target of Rapamycin (TOR) signaling [5,10]. During larval life, this pathway is activated in wing cells by haemolymph signals produced in response to feeding, including Insulin-like peptides (ILPs) that act via the Insulin Receptor (InR)/PI3-Kinase/Akt pathway, as well as sugars and amino acids. These inputs converge to regulate TOR—an intracellular kinase with diverse roles in metabolism [5,10]. Starvation reduces TOR activity and scales wing size (and entire body size) downwards (Fig 1A and 1B), an effect mimicked by genetically inhibiting TOR (Fig 1C). Yet, wing growth is simultaneously governed by intrinsic signaling systems (e.g., Wnt, BMP, and Hh morphogens) that control wing size, shape, and pattern [2,3,6,8]. Many of these organ-intrinsic systems exert their effects at least in part via regulation of the Warts (Wts)/Hippo (Hpo) pathway [11–13]—a network of proteins that inhibit a growth-promoting transcriptional coactivator, Yorkie (Yki; orthologous to vertebrate YES-Associated Protein [YAP]) [14]. Hpo and Wts are kinases that act in sequence, Hpo phosphorylating Wts and Wts phosphorylating Yki to sequester Yki cytoplasmically. Inhibition of either kinase promotes growth by allowing Yki to evade cytosolic sequestration and gain access to the nucleus [15–17]. Nuclear Yki binds transcription factors including Scalloped (Sd, a TEAD [transcriptional enhancer activator domain] protein) [18,19] to up-regulate expression of genes that promote cell growth and proliferation. Morphogens [20,21], the protocadherins Fat and Dachsous [22–25], the Crumbs/Lgl epithelial polarity proteins [26–29], and mechanical strain [30] all modulate either or both Wts and Hpo, exerting effects on wing growth via Yki. But for the wing to scale, Yki activity, or the growth that Yki stimulates, must be contingent on TOR activity. Despite previous attempts to assess the links between Wts/Hpo and InR/TOR signaling [15,31–35], the logic by which cells in growing organs integrate these inputs to achieve organ scaling remains unclear.
Here, we report a novel molecular mechanism by which Yki can integrate the control of Drosophila wing growth by the Wts/Hpo and TOR pathways. As part of this mechanism, we present evidence for what is, to our knowledge, a previously unknown mode of transcriptional regulation, which we term “nuclear seclusion”, whereby a transcription factor (Yki) is retained in the nucleus but sequestered from target loci and hence rendered unable to induce their expression. In contrast to the Wts/Hpo pathway, which controls Yki activity by governing its access to the nucleus, our findings indicate that the TOR pathway activates Yki after it gains entry to the nucleus, by allowing it to escape nuclear seclusion and gain access to its growth-promoting target genes. We posit that this novel mechanism allows Yki to integrate local patterning inputs mediated by the Wts/Hpo pathway with systemic, humoral inputs mediated by TOR, comprising part of the system that scales wing growth in response to nutrition.
To determine the relationship between TOR and Yki in the developing Drosophila wing, we first asked whether the capacity of Yki to promote growth depends on TOR activity. To do so, we generated clones of cells mutant for Wts/Hpo pathway components that normally hold Yki activity in check and tested if the resulting increases in tissue mass and cell number depend on TOR activity. All such experiments that we performed indicate an obligate role of TOR activity for Yki-dependent wing growth. Specifically, clones mutant for the FERM (4.1 Ezrin Radixin Moesin) domain protein Expanded (Ex), which is required for normal Hpo activity, grow far larger than control clones (Fig 1F; S1 Data [37]). However, when such clones were also mutant for TOR (ex—Tor—clones), they were tiny (Fig 1G; S1 Data) and not significantly larger than Tor—clones (Fig 1D and 1E; S1 Data). The impaired growth of both Tor—and ex—Tor—clones is not due to apoptosis, since it is not detectably rescued by expressing the apoptosis inhibitor p35 in the mutant cells (Fig 1H–1K; S1 Data). Equivalent experiments, activating Yki by removing Wts and inhibiting TOR by removing the TOR-activating GTPase Rheb, gave similar results (S1A–S1H Fig). Likewise, overexpressing Yki throughout the wing primordium leads to excessive growth; however, co-overexpression of ΔP60, an inhibitor of InR signaling upstream of TOR [38], blocks this growth, yielding wings similar in size to those in which only ΔP60, alone, was overexpressed (S1I–S1O Fig and discussion in S1 Fig legend; S1 Data). Equivalent results were obtained in the developing head primordium: mutation of wts or ex leads to dramatic overgrowth of head tissue, which is suppressed when the head is also mutant for Tor or the InR pathway kinase Akt (S1P–S1X Fig). Thus, reducing the level of InR and TOR signaling, which is normally set by nutrient status, restricts Yki’s capacity to promote growth—suggesting a relationship that could contribute to scaling wing size with nutrient levels.
Why is Yki-driven growth limited by the level of TOR activity? TOR might be required upstream to facilitate Yki activity, or in parallel to trigger other, independent growth-related processes. Alternatively, TOR might be required downstream, with Yki promoting growth at least in part by elevating InR/TOR pathway activity. To test this latter possibility, we assayed whether conditions that abnormally increase Yki activity (loss of either Ex or Wts) have a corresponding effect on the levels of phospho-Akt (pAkt-S505), which are normally elevated by enhanced InR signaling (e.g., by removal of the InR pathway inhibitor PTEN; Fig 2A). We find that pAkt-S505 levels are normal in ex—or wts—clones (Fig 2B and 2C), as well as in protein extracts from the overgrown wing discs of homozygous ex—or wts—larvae (Fig 2D). Likewise, the levels of phosphorylated S6-Kinase (pS6K-T398), a readout of TOR activity, were not affected in ex—or wts—discs (Fig 2E). We therefore infer that Wts/Hpo regulated Yki activity does not stimulate either InR or TOR pathway activity in the wing, consistent with results from Drosophila cell culture [15]. This finding argues against a downstream role of InR/TOR signaling in mediating Yki-driven growth.
Further evidence that TOR activation is not dependent on Yki comes from the observation that ectopically activating the InR/TOR pathway via upstream components can suffice to increase the growth of yki—wing clones (rescued with p35; Fig 2F, 2G and 2J; S1 Data). Specifically, expressing Rheb (to activate TOR) resulted in a 22% increase in clone size (Fig 2I and 2J; S1 Data) and expressing Dp110 (to activate InR) resulting in a 26% increase (Fig 2H and 2J; S1 Data). In both cases, clonal expansion resulted solely from increases in cell size (Fig 2L; S1 Data), without effects on cell number (Fig 2K; S1 Data). Hence, TOR can be activated and promote detectable cell growth, albeit not cell proliferation, in the absence of Yki. Notably, control clones expressing Dp110 or Rheb with p35 that are wild type for Yki strongly overgrew (S2A–S2D Fig; S1 Data), and this was in part due to both their increased cell numbers (S2E Fig; S1 Data), as well as strongly increased cell size (S2F Fig; S1 Data). We infer that although InR/TOR pathway activity appears largely independent of Yki activity, ectopic InR/TOR activation nevertheless requires Yki function as a prerequisite in order to drive cell proliferation.
If Yki depends on, but does not activate, the InR/TOR pathway to promote growth, an alternative explanation is that the InR/TOR pathway acts in parallel to or upstream of Yki. Yki activity depends in large part on phosphorylation by Wts, which causes Yki to be sequestered in the cytosol, reducing its access to the nucleus [15–17]. Yki normally appears predominantly cytoplasmic in the developing wing (Fig 3A), consistent with its essential, but tonic, activity in sustaining wing growth. However, upon Wts inactivation, Yki accumulates in the nucleus and inappropriately up-regulates transcription of its growth-promoting target genes, causing pronounced tissue hyperplasia [15–17] (Fig 3B). Hence, inhibiting TOR activity might countermand the overgrowth phenotype caused by loss of Wts activity by compromising nuclear access of unphosphorylated Yki. In preparing to test this possibility, we were surprised to discover that reducing TOR activity has the opposite effect on Yki localization: instead of restricting nuclear access, it causes a dramatic increase in nuclear Yki, even as it blocks Yki-dependent growth.
In our initial experiments, we strongly inhibited endogenous TOR activity in the developing wing, using dpp.Gal4 to drive a stripe of expression of UAS transgenes encoding three proteins that have been widely employed to study the consequences of reducing TOR activity: a dominant negative form of TOR (TORTED) and the TOR corepressors TSC1 and TSC2 [32,40–46]. Co-overexpression of all three proteins causes strong nuclear Yki accumulation in the stripe (Fig 3C and 3D). Expressing either TSC1+TSC2, or TORTED, alone, in the stripe or in other patterns within the wing, also caused enhanced Yki nuclear localization (S3A–S3C Fig), albeit less strongly, suggesting that the level of nuclear accumulation depends on the strength of TOR inhibition, and furthermore is not region-specific. Note that we prefer the use of transgenes to inhibit TOR signaling in large sectors of tissue, rather than examining tiny Tor—clones, where interpretation is compromised by protein perdurance. We also observed enhanced Yki nuclear accumulation when InR signaling was inhibited, upstream of TOR (S3D–S3F Fig). Conversely, removal of two downstream metabolic targets of TOR, dS6K, and 4EBP/Thor [10] did not affect Yki nuclear accumulation (S3G–S3J Fig), indicating that the control of Yki localization by TOR involves a mechanism that is not mediated by these metabolic regulators. Thus, reduced InR/TOR activity results in enhanced nuclear accumulation of Yki—a result that is counterintuitive, because nuclear Yki is usually associated with increased, rather than reduced growth [15–17].
Typically, nuclear Yki up-regulates growth-promoting target genes, one of the best characterized of these being diap1, the gene encoding the Drosophila Inhibitor of Apoptosis protein [18,19]. However, the dramatic increase in nuclear Yki accumulation caused by TOR inhibition (Fig 3C and 3D) is associated with a strong reduction in DIAP protein levels (Fig 4A and 4B), consistent with reduced diap1 gene transcription. Diap1 transcription is normally driven by a Yki-Sd complex that binds ~4 kb downstream of the transcription start site of the predominant diap1-RA transcript [19]. A minimal reporter construct, 2B2C-lacZ, containing ~300 bp of this enhancer region, is up-regulated by Yki overactivation [19]. Conversely, RNAi knockdown of Yki results in reduced 2B2C-lacZ expression (S4A and S4A’ Fig). Like Yki RNAi knockdown, TOR inhibition caused a marked decrease in 2B2C-lacZ expression (Fig 4C and 4D), indicating a failure of the Yki/Sd-dependent 2B2C enhancer to maintain normal target gene expression.
TOR inhibition similarly compromises the transcription of other Yki target genes. four-jointed (fj) is the second target gene that we have confirmed is reduced by Yki knockdown (S4B Fig). Like Yki knockdown, TOR inhibition strongly reduced fj-lacZ expression (Fig 4E and 4F). A third reporter, exe1-lacZ, was unresponsive to TOR inhibition at 25°C, but raising the temperature to 29°C for several hours prior to dissection (to increase GAL4-dependent TORTED, TSC1, and TSC2 overexpression) sufficed to repress this reporter (Fig 4G and 4H). Further, in animals homozygous for the exe1-lacZ reporter allele, exe1-lacZ expression was clearly reduced by TOR inhibition at 25°C (S4C and S4D Fig). We note that such exe1-lacZ/exe1-lacZ discs have reduced ex gene function and hence abnormally elevated levels of nuclear Yki; nevertheless, a sharp increase in nuclear Yki can still be detected in the TOR-inhibited stripe, coincident with reduced transcription of the exe1-lacZ gene (S4E–S4G Fig). Finally, the effect of TOR inhibition appears specific to Yki target genes, since other proteins expressed in the developing wing, including Vestigial (S4I Fig), Armadillo (S4J Fig), Distal-less (S4J’ Fig), a Distal-less-lacZ reporter (S4K Fig), and indeed Yki itself (Fig 3C and 3D), were not reduced by strong TOR inhibition.
The unexpected (and counterintuitive) effects of TOR inhibition on Yki nuclear accumulation and target gene expression are reminiscent of another scenario in which enhanced nuclear access of Yki is associated with the loss of target gene expression. In this case, Yki can be forced to accumulate in the nucleus by overexpression of Sd, the site-specific DNA binding factor that normally mediates target gene activation by Yki in the wing blade (S6A Fig) [18,19]. Under this condition, the abnormally high levels of Sd sequester the limited levels of Yki available, exerting a classic dominant negative “squelching” effect [47] that titrates Yki away from Sd bound to target gene enhancers. As in the case of TOR inhibition, the level of nuclear Yki increases whilst target gene expression decreases (S6B and S6C Fig). This resemblance suggests that TOR inhibition may exert its effects on Yki target gene expression via a similar mechanism: by directing Yki to the nucleus whilst reducing its access to the relevant target gene enhancers.
To test this possibility directly, we focused on diap1, the locus where the general mechanism of transcriptional control by Yki is best characterized, with a defined enhancer element, 2B2C, that is controlled exclusively by Yki in complex with Sd (henceforth Yki-Sd) [14,18,19,48]. We prepared chromatin from control and TOR-inhibited wing discs (the latter coexpressing TORTED+TSC1+TSC2 under the control of a Flp-out Act5C>CD2>Gal4 driver) and immune-precipitated using a Yki antibody [15]. To produce enough tissue despite TOR inhibition, discs were allowed to grow to a large size before a strong heat shock was used to excise the stop cassette within the Act5C>CD2>Gal4 driver and initiate Gal4 expression in nearly all cells (Fig 5A and 5B; loss of magenta CD2 marks the Act5C>Gal4 expressing tissue), with dissection of 200 discs per genotype 8–10 hr later (see Methods).
In control discs, as expected, Yki was enriched ~2–3-fold within the 2B2C diap1 element, using primers centered around the CATTCA motif to which Yki-Sd binds (relative to primers defining unrelated DNA sequences, as well as an IgG control treatment; Fig 5C). Knockdown of Yki (using the same Act5C>CD2>Gal4 protocol to drive transient expression of a UAS.ykiRNAi transgene) abolished this peak of enrichment, confirming its dependence on Yki levels and also validating the specificity of the antibody (S8A and S8B Fig). Strikingly, in experimental, TOR-inhibited discs, Yki enrichment at 2B2C was also diminished, by almost half (p < 0.05, n = 4 fully independent experimental replicates; Fig 5C; S1 Data). This reduction occurred despite the clear increase in nuclear Yki resulting from TOR inhibition (Fig 5A and 5B).
These data provide evidence that inhibiting endogenous TOR activity reduces the amount of Yki bound to a canonical Yki-Sd target gene enhancer, even as it causes an increase in Yki nuclear accumulation. Hence, we propose that TOR inhibition renders Yki subject to a phenomenon that we term “nuclear seclusion”, whereby Yki is mobilized and/or sequestered in the nucleus but hindered from accessing its target enhancer sequences. We posit that this mechanism accounts at least in part for the reduction in DIAP1 expression observed when TOR is inhibited (Fig 4B) and suggest that it likewise explains the loss of expression of other TOR-dependent Yki-Sd target loci for which defined enhancers have not yet been characterized (e.g., fj, ex; Fig 4F and 4H).
It is important to note that in most of our experiments, we have strongly reduced TOR signaling by coexpressing TORTED+TSC1&2, to approximate, or at least approach, TOR loss-of function. These experiments reveal a basic requirement in wing cells for at least some level of TOR signaling in order to override nuclear seclusion of Yki and permit it access to drive transcription of target genes. However, we additionally propose that nuclear seclusion also operates within the normal physiological range of TOR activity as set by nutrient intake. First, weaker impairment of TOR signaling, by expression of TORTED alone, produces a wing size comparable to a moderately staved animal (Fig 1C). And yet it also leads to the two hallmarks of nuclear seclusion: increased Yki accumulation (S5D Fig), with concomitant reduction in expression of target genes (S5E Fig). Second, moderate reductions in TOR activity with RNAi transgenes directed against TOR and Rheb also reduce wing blade size within the range of normal trait scaling [5] (S5A–S5C Fig). Again, these manipulations cause detectable—albeit correspondingly much less pronounced—nuclear accumulation of Yki (S5F and S5H Fig) and reduced target gene expression (S5G and S5I Fig). Hence, while our assay of strong TOR inactivation (TORTED+TSC1&2) gives the clearest observable nuclear accumulation, and is the basis for most of our mechanistic insights regarding Yki nuclear seclusion, we posit that this phenomenon operates beyond a simple baseline requirement for TOR signaling and is modulated by fluctuations of TOR signaling within the normal range. Furthermore, we note that because Yki is predominantly cytoplasmic in wing cells (Fig 3A), the small fraction of active, nuclear, Yki might be effectively secluded by quite modest reductions in TOR signaling (without leading to the profound nuclear accumulation witnessed when TOR is strongly impeded).
What mechanism mediates Yki nuclear seclusion in response to TOR inhibition? Given the similar effect of Sd overexpression on Yki localization and target gene transcription, we considered the possibility that the mechanism mediating Yki nuclear seclusion could be a pronounced increase in Sd protein level. However, Sd is already at high level throughout the developing wing, and TOR inhibition does not cause a further increase, as monitored by the expression of a fully functional GFP protein trap allele of the endogenous sd gene (sdGFP; [49]) (S4J” Fig J; S8G and S8H Fig). Hence, we sought to assess two other possibilities: (i) that Yki, alone, is subject to nuclear seclusion, e.g., by being titrated away from enhancer-bound Sd, or (ii) that Sd, in addition to Yki, and perhaps in complex with it, is targeted for seclusion.
Using sdGFP wing discs and a GFP antibody, we assessed Sd-GFP enrichment at the 2B2C diap1 enhancer by ChIP. Like Yki, Sd-GFP is strongly enriched at 2B2C in wild-type discs (Fig 5D). However, inhibiting TOR activity caused a reduction in Sd-GFP enrichment at 2B2C to a similar degree as it does Yki enrichment (Fig 5C and 5D). Hence, it appears that TOR inhibition results in nuclear seclusion of Sd as well as Yki, rather than just Yki. One explanation for this observation is that Sd might need to be bound by Yki to bind 2B2C —indeed, such a cooperative interaction appears to be required for enhancer binding by Sd in combination with Vg, another Sd transcriptional coactivator [50]. However, we observed that RNAi knockdown of Yki, which as expected results in reduced Yki binding at 2B2C, has no effect on Sd enrichment (S8B Fig). That Sd binding to 2B2C apparently does not depend on Yki argues against cooperative target site binding as a possibility. We therefore posit two possibilities: that Sd in complex with Yki is subject to nuclear seclusion, or, alternatively, that TOR inhibition secludes Sd in addition to, but independently of, Yki. If this second scenario were the case however, we may expect Sd in complex with transcriptional activators other than Yki to be influenced by seclusion. Conversely, expression of vg, an autoregulated target of Vg-Sd but not Yki-Sd complexes in wing cells, is not affected when TOR activity is inhibited (S4H and S4I Fig). Hence, we favor the view that nuclear seclusion acts on Sd in complex with Yki, although this additional facet of the mechanism remains to be confirmed.
Yki is a modular protein with three regions that are conserved with vertebrate YAP and essential for normal function (Fig 5E): (i) a Wts phosphorylation site which, when phosphorylated, mediates cytoplasmic retention of Yki by 14-3-3 anchor proteins (ii) two WW domains that bind both Wts and Ex; and (iii) an N-terminal region, the Sd/TEAD binding domain (TBD), that mediates binding to Sd. Yki normally shuttles between the cytosol and nucleus in response to the state of phosphorylation at the Wts site [15–17], with phosphorylated and unphosphorylated forms accumulating, respectively, in the cytosol and nucleus. This shuttling is revealed by the build-up of nuclear Yki in wing cells following treatment with the nuclear export inhibitor Leptomycin B (S6D and S6E Fig; [51]). However, TOR inhibition is unlikely to cause abnormal Yki nuclear accumulation by reducing Wts-dependent phosphorylation of Yki, as this would mimic the consequence of Wts/Hpo inactivation [15–17], namely target gene induction rather than the observed repression. We therefore considered an alternative possibility that a different mechanism of Yki regulation, perhaps mediated by one or both of the other conserved Yki motifs, might be at play.
To shed further light on the mechanism of nuclear seclusion, we generated Tubα1. ykiGFP transgenes that express GFP-tagged forms of Yki at near-physiological levels (S8C Fig; S1 Data) and analyzed the consequences of mutating each of the three conserved functional domains on the subcellular accumulation of Yki. TOR inhibition caused nuclear accumulation of wild-type YkiGFP as expected (Fig 5F). However, introducing a mutation, P88L, in the N-terminal TBD that diminishes binding to Sd [19] prevented nuclear accumulation in TOR-inhibited cells (Fig 5G). This block to nuclear accumulation appears to be specific to the loss of TOR pathway activity, as YkiGFP-P88L protein accumulates in the nuclei of wts—wing cells that have normal TOR pathway activity (Fig 5K), and the same is true for YkiGFP-P88L protein that is additionally mutated for the Wts phosphorylation sites in otherwise wild type wing cells (Fig 5 and 5J). Likewise, mutating Yki’s two WW domains prevents nuclear accumulation following TOR inhibition (Fig 5H), even though the WW domains are known not to interfere with nuclear accumulation following Wts/Hpo pathway inactivation [52]. Thus, it appears that the Wts-phosphorylation site governs whether Yki has access to the nucleus: only Yki in which this domain is not phosphorylated by Wts can enter. In contrast, interactions mediated by the TBD and WW domains appear to determine whether any unphosphorylated Yki that gains access to the nucleus will be secluded there when TOR activity is inhibited. Accordingly, the abnormal nuclear accumulation of Yki resulting from TOR inhibition should represent a build-up of unphosphorylated Yki, even when Wts is present, as any unphosphorylated Yki that enters the nucleus will get trapped there. Consistent with this prediction, we observed a general reduction in phosphorylated Yki-S168 levels following TOR inhibition (S8D Fig). (Note that although we interpret this as a consequence of nuclear seclusion, we cannot rule out an additional, independent effect of TOR inhibition in dephosphorylating Yki. However, we note that if such an activity occurs, it has seemingly negligible function consequences for Yki activity; dephosphorylated Yki should increase target gene expression, but our data show that blocking TOR strongly reduces Yki target gene expression).
Based on these data, we infer that low levels of InR/TOR signaling activity result in the induction or function of a presently unidentified Nuclear Secluding Factor (NSF) that acts in wing cells to sequester Yki in the nucleus and to divert Yki-Sd from its target genes. The putative NSF might interact with Yki directly, via Yki’s N-terminus and/or WW domains and cause Yki nuclear seclusion by both increasing the binding of Yki to Sd and preventing the resulting NSF-Yki-Sd complex from accessing Yki-Sd target genes. Accordingly, NSF action would result in enhanced Yki nuclear accumulation and reduced Yki-Sd target gene expression. This mode of Yki regulation acts independently of the Hpo-Wts mediated cytosolic tethering of Yki to limit target gene expression. Indeed, even in Hpo pathway mutant tissue (e.g., loss of Ex, or Wts), where Yki is constitutively dephosphorylated, TOR inhibition still leads to increased Yki nuclear accumulation (S4E–S4G Fig) and decreased Yki target gene transcription (S4C–S4D Fig; S8E and S8F Fig).
The requirement for InR/TOR signaling activity to allow Yki to escape nuclear seclusion is revealed by genetic manipulations that compromise InR/TOR pathway activity and cause Yki to accumulate nonproductively in the nucleus. Importantly however, we do not think that TOR signaling promotes Yki-driven wing growth solely by this mechanism. Instead, we posit that this mechanism operates in parallel with other outputs of InR/TOR signaling that regulate a range of basic metabolic processes that are necessary, independently, for cell growth as well as rate-limiting for cell proliferation [5,10,53]. A clear demonstration of this parallel requirement for InR/TOR signaling is the observation that Yki-driven overgrowth of the wing (caused by removal of wts) could be suppressed by removal of S6-Kinase (dS6K), a TOR target that catalyses cap-dependent mRNA translation (S7A–S7D Fig; S1 Data), but which is not itself involved in nuclear seclusion (S3G, S3H and S3J Fig).
Hence, while TOR/InR activity is needed to override Yki nuclear seclusion, it is not sufficient to promote proliferative growth in the absence of concomitant activation of canonical metabolic targets like dS6K. Coupled to our reciprocal finding above, that TOR activity is unable to promote cell proliferation without Yki (Fig 2F–2L), we conclude that neither output of TOR is sufficient. Rather, both outputs are required, first to unleash Yki from nuclear seclusion, and second to provide the metabolic wherewithal for Yki target genes to promote growth. By concomitantly compromising both outputs, nutritional deprivation coordinates Yki target gene expression with decreased metabolic capacity to reduce the growth rate of the wing.
Our results describe how inhibiting TOR signaling reduces Yki function as part of the mechanism that might scale wing growth downwards in response to nutritional deprivation. To further elucidate how TOR and Yki might function together to control wing size, we asked whether superphysiological TOR activity might alter Yki localization and/or activity to scale wing size upwards. However, we were unable to detect any such changes in Yki using either of two well-established approaches to overactivate the InR/TOR signaling in vivo. Specifically, overactivating TOR by removing the negative regulator TSC1 in wing cells did not appear to alter the nuclear-cytoplasmic distribution of Yki as assayed either by immunofluorescence in clones of Tsc1—mutant cells (Fig 6A) or by quantitating the nuclear–cytoplasmic ratio of Yki in fractionated whole wing discs from entirely Tsc1—mutant larvae (Fig 6B). Similarly, we tested whether clonal removal of TSC1, or alternatively, clonal overexpression of the positive regulator Rheb, affects the expression of any of several Yki target genes, and again, observed no differences (Fig 6C and 6D; S9A and S9B Fig). Hence, superphysiological gains in TOR activity do not appear to influence either Yki localization or activity, suggesting that peak endogenous TOR signaling is sufficient to fully override Yki nuclear seclusion as well as any other InR/TOR-dependent constraints on Yki coactivator function.
Nevertheless, manipulations that generate superphysiological levels of TOR activity do reveal a further facet of the TOR-Yki relationship in the wing. We noticed that despite strongly elevating TOR signaling in wing discs (Fig 2E), loss of Tsc1 activity failed to alter overall wing disc size (S9C and S9D Fig). Instead, it appears to cause excess growth that is counterbalanced by apoptosis (S9C and S9D Fig). The same was true of discs overexpressing rheb in all cells under the control of the imaginal disc-specific headcase.GAL4 driver (hdc.G4; Fig 6J, 6K and 6N; S1 Data), which produces an equivalent level of TOR activation to loss of Tsc1 (compare phospho-S6K levels in S7I Fig to those of Tsc1—discs in Fig 2E). This increase in apoptosis can be attributed to a repressive effect of superactivation of TOR on the expression of DIAP1 protein and the bantam microRNA (Fig 6E–6G; S9F and S9F’ Fig), both targets of Yki regulation that have antiapoptotic activity [55]. However, the repression of DIAP1 and bantam does not appear to be due to enhanced Wts kinase activity, as Tsc1—discs showed a mild reduction rather than an increase in phospho-Yki S168 levels (Fig 6H). Moreover, the effects on DIAP1 and bantam are seemingly Yki-independent, since the repression of DIAP1 appears to be posttranscriptional (S9A and S9B Fig), and ChIP experiments indicate that Yki binding to its target enhancer in the bantam gene is not reduced in Tsc1—wing discs (Fig 6I) [56]. In principle, DIAP repression could be a secondary consequence of bantam repression, since DIAP1 protein is negatively regulated by the Head involution defective protein (Hid) [57], a bantam target [58]. However, we failed to observe any effect of overexpressing or removing bantam on DIAP1 protein levels (S9G and S9H Fig), from which it appears that TOR superactivation does not act via bantam to repress DIAP. Thus, TOR overactivation appears to cause cell death independently of the Wts/Hpo/Yki pathway, possibly mediated by reduced expression of both DIAP1 and bantam.
To gauge if abnormally high levels of TOR activity indeed repress DIAP and bantam to offset tissue overgrowth, these gene products were resupplied to discs overexpressing rheb under hdc.GAL4 control. Overexpression of diap1, alone, under hdc.GAL4 control produced no significant effect on wing disc size (Fig 6N; S1 Data). However, co-overexpression of diap1 together with rheb fully suppressed cell death that would otherwise result from the overexpression of rheb alone (Fig 6L), and disc size increased by 77% (Fig 6N; S1 Data). To resupply bantam, GAL4-dependent transgene expression was not used, since bantam overexpression itself induces growth [58,59], rendering synergistic outcomes difficult to interpret. Instead, a transgene was constructed that expresses GFP at low levels under the direct control of the Tubα1 promoter, with the 3’UTR containing a minimal 100 bp genomic bantam fragment that includes the 21 bp bantam microRNA [58]. The resulting Tubα1.GFP-bantam transgene does not itself stimulate overgrowth (Fig 6N; S1 Data) but produces enough bantam to rescue a homozygous null bantam animal to adulthood, producing a normally sized and patterned animal (S9O and S9P Fig). This suggests that a single copy of Tubα1.GFP-bantam produces a level of bantam activity comparable to homozygosity for the endogenous bantam gene. Strikingly, when introduced as a single copy into the hdc.G4>UAS.rheb background, disc size increased 70% (Fig 6N; S9N Fig; S1 Data), similar to the effect of co-overexpressing DIAP with rheb, although the addition of the Tubα1.GFP-bantam did not fully suppress cell death (compare Caspase staining in S9L and S9M Fig). Finally, combining co-overexpression of rheb and diap1 with Tubα1.GFP-bantam led to still more growth, with the wing disc more than doubling in size (110% larger; Fig 6H and 6N; S1 Data).
In sum, the effects of reintroducing DIAP1 and bantam reveal a novel effect of superphysiological TOR activity: the repression two key Yki target genes, albeit by Yki-independent mechanisms, to induce cell death and counterbalance the excessive tissue growth that would otherwise result from TOR hyperactivity. We suggest that this regulatory circuit normally safeguards against organ overgrowth and overriding it exposes a hidden capacity of superphysiological TOR signaling to stimulate tissue hyperplasia.
Organ size depends on organ-intrinsic mechanisms that couple cell proliferation to the specification of pattern, as well as on external signals such as nutrients that scale overall body size [6–9]. Our study reveals one way in which these two distinct developmental phenomena can be integrated in the developing Drosophila wing, via a circuit linking nutrient-controlled TOR (and upstream InR) activity to Yki target gene access. A key finding is our discovery that, in addition to the well-known cytoplasmic anchoring mechanism employed by the Wts/Hpo tumour suppressor pathway to constrain Yki nuclear access, the InR/TOR pathway operates by a wholly different mechanism to allow nuclear Yki to activate its target genes. Inhibiting each pathway results in the accumulation of Yki in the nucleus, but with opposite effect. Nuclear Yki resulting from reduced Wts/Hpo activity drives Yki target gene expression and promotes growth. But, strikingly and counterintuitively, nuclear Yki resulting from reduced InR/TOR activity causes a loss of target gene access and impedes growth.
Based on our results, and as summarised in Fig 7A, we propose that reduced TOR signaling causes Yki to bind a secluding factor that sequesters Yki in the nucleus but diverts it from acting in complex with Sd to bind target loci, thereby reducing gene transcription. We posit that, by combining TOR-dependent relief from Yki nuclear seclusion with canonical Wts/Hpo phosphoregulation of Yki nuclear access, wing cells are able to integrate nutrient levels (via TOR) with patterning/morphogenetic inputs (via Wts/Hpo) to achieve a level of Yki-Sd activity that matches metabolic capacity and scales wing size appropriately. Elucidating this mechanism further will depend on identifying the proposed NSF and determining how its action is mediated by Yki’s N-terminus and WW domains. NSF may bind Yki that shuttles into the nucleus, titrating it away from target genes; one possibility, raised by a reviewer, is that NSF acts by modifying Yki in some way, stabilizing the protein and preventing its degradation (thereby contributing to its observed accumulation in the nucleus) with a modification that blocks it from accessing target loci.
A model for how TOR and Wts/Hpo inputs are integrated during normal Drosophila wing growth is depicted in Fig 7B. When the native TOR pathway is signaling at its peak, the level of Wts/Hpo activity dictates the amount of unphosphorylated Yki that is free to enter the nucleus. TOR activity liberates the Yki-Sd complex from a default state of nuclear seclusion, permitting nuclear Yki-Sd to access and transcriptionally activate loci that promote proliferative growth. In parallel, TOR also stimulates cellular metabolic processes such as protein translation (mediated by targets such as dS6K), which are preconditions for tissue growth. Both outputs of TOR are required for wing cells to divide and gain mass under Yki-Sd control. Given the central role of TOR signaling in nutrient sensing, [5,10], we further propose that this circuit may contribute to the capacity of the developing wing to scale downwards under conditions of nutrient deprivation. Specifically, under nutrient-limiting conditions, we suggest that the consequent reduction in TOR activity renders the Yki-Sd complex subject to nuclear seclusion, thereby decreasing transcription of Yki-Sd target genes that promote growth. Reduced TOR activity would also diminish metabolism, matching metabolic capacity with the reduced activity of Yki-Sd target genes. Such a circuit would provide a sensitive means for developing tissues to integrate systemic and local inputs into growth. Our finding that excess activity of TOR appears to cause excess growth that is compensated by cell death via reduced DIAP1/bantam expression (Fig 7C) reveals that the circuit linking TOR and Yki is additionally stabilized by negative feedback, constraining organ growth in the event of erroneous, superphysiological TOR activity. Hence, a previously unknown facet of nutrient sensing by TOR is the incorporation of a self-limiting tumor suppression mechanism.
While our study delineates a circuit that integrates TOR and Yki activity during Drosophila wing growth, we note that it is only in this tissue that we find unequivocal evidence for nuclear seclusion of Yki, the means by which we posit this integration is achieved. Blocking TOR signaling in two other imaginal disc types, the eye and leg, does not appear to affect either Yki localization or target gene expression (S8I–S8K Fig). The same is also true even for other cell populations within the wing imaginal disc, notably those giving rise to the wing hinge and body wall. Hence, nuclear seclusion of Yki does not appear to be a general consequence of inhibiting TOR activity, and may occur primarily (or possibly only) in presumptive wing cells.
Intriguingly, the wing primordium is unique amongst imaginal disc tissues in expressing high levels of Sd under the control of Vg, a transcriptional coactivator that forms a complex with Sd to “select” the wing state. Although Yki-Sd complexes regulate growth control downstream of the Wts/Hpo pathway in the other Drosophila imaginal discs, Sd is expressed in these other tissues at tonic, low levels. Further, in the wing, Vg-Sd and Yki-Sd coexist, regulating distinct cohorts of target genes, and our present evidence suggests that inhibiting TOR does not alter the capacity of the Vg-Sd complex to gain access to, or activate, its distinct set of target genes (e.g., vg and sd itself). Hence, nuclear seclusion of Yki may represent an organ-specific mechanism to integrate nutrient-regulated InR/TOR and morphogen-dependent Wts/Hpo activities under conditions in which Sd plays a major, independent role in specifying cell state. It remains possible that TOR activity regulates nuclear seclusion of Yki to control growth in other tissues, and possibly in other animals, albeit in complex with other transcriptional effectors and primarily through the control of target gene access rather than nuclear access. Alternatively, TOR may limit growth in these other systems exclusively through its regulation of cellular metabolism, without additionally modulating the level of Yki activity. Studying how global chromatin occupancy by Yki changes in response to TOR pathway activity, and varies by tissue type, will be needed to evaluate these scenarios.
To our knowledge, nuclear seclusion represents a novel paradigm of transcription factor regulation. Although experimental overexpression of nuclear Yki cofactors (e.g., Sd [18,52]) can exert similar effects on Yki localization and target gene expression, thus far it has not been demonstrated that such a mechanism functions as a biologically relevant mode of transcription factor regulation operating in vivo. Presently, the identity of the factor mediating Yki nuclear seclusion is unknown. However, in both yeast [60] and mammals [61,62], mTOR functions as a classical kinase that phosphorylates diverse substrates, leading to their retention in the cytoplasm. We hypothesize that our proposed NSF is one such substrate. A potential avenue of future exploration will be systematically testing how TOR inhibition influences the increasing number of identified nuclear modulators of Yki—some of which, like Sd, affect Yki localization and function when overexpressed (e.g., Hipk [63,64], MASK [65,66] and Tgi [48,67]). As we have demonstrated for the regulation of Yki by TOR, nuclear seclusion offers a complementary mechanism for controlling transcription factors that are already subject to phosphorylation-dependent partitioning between the cytosol and nucleus. Nuclear seclusion may therefore be used more widely to achieve signal integration, both during development as well as in mature tissues.
The following mutant alleles, Flp Recombination Targets (FRTs), UAS transgenes, Gal4 drivers and reporter transgenes were used (source indicated in parentheses; see Flybase http://flybase.org/ for details).
All measurements were made blind, without knowledge of the genotype. When performing multiple t tests involving the same treatments, Bonferroni-corrected p-values were used to assess the level of significance. Clone and wing sizes were measured in Adobe Photoshop using the “lasso” tool, and cell numbers counted using the “count” tool. For whole wing cell number estimates, setae were counted within a 170 x 170 pixel box within the posterior intervein region. Total wing cell number = (wing size in pixels / (170 x 170)) x cell number in 170 x 170 pixel box.
Discs were dissected, fixed in 4% PFA and stained with the following antibodies: rabbit anti-Phosho-AKT (S505) (1:200, Cell Signalling Technology), rabbit anti-Distal-Less (1:3,000), rabbit anti-Yki (1:500; D. Pan), mouse anti-DIAP1 (1:500), rabbit anti-ß-galactosidase (1:6,000; Cappel), mouse anti-Ptc (1:150; DHSB), mouse anti-Armadillo (1:10; DSHB), guinea pig anti-Distal-less (1:3,000; R. Mann), mouse anti-CD2 (1:500; BD Biosciences). Alexa Fluor secondaries (Invitrogen) were used, and Hoechst was used to label nuclei. A Leica SP5 confocal microscope was used to capture all immunofluorescence images. For SEM, adult heads created by the EGUF/Hid technique were dissected into 70% ethanol, dried with an AUTOSAMDRI critical point drier, gold coated and imaged using the Hitachi S-3400N Variable Pressure SEM installed at the Microscopy Imaging Lab, University of Toronto.
For standard blotting of total wing protein, ~80 wing discs/lane were dissected from wandering third instar larvae into laemmli sample buffer supplemented with DTT, Complete Mini EDTA-Free protease inhibitor (“CPI” Roche) and Phosphatase Inhibitor Cocktail III (Calbiochem). Samples were boiled for 10 min and protein was quantified using the Pierce 660nm assay. Equal amounts of protein were loaded per lane. For fractionation of wing disc cells, ~80 wing discs were dissected per genotype, and washed twice with ice cold PBS+CPI. Discs were then washed briefly in 10μl hypotonic buffer (10 mM HEPES + 1.5 mM MgCl2 + 10 mM KCL + CPI), centrifuged (1,500 rpm, 5 min, 4°C), and incubated in 30 μl hypotonic buffer on ice for 10 min. Discs in buffer were then dounced ~30 times in a glass homogeniser. Nuclei were separated by centrifuging (15 min, 3,750 rpm, 4°C) and removing the cytoplasmic fraction. Nuclei were washed three times in hypotonic buffer (spinning down and resuspending each time), and finally resuspended in 30 μl hypotonic buffer. 40μl laemmli sample buffer + DTT + CPI was added to both nuclear and cytoplasmic fractions, which were then boiled for 10 minutes and quantified using the Pierce 660 nm assay. Equal amounts of nuclear and cytoplasmic protein were loaded per lane. Antibodies used were rabbit anti-Phospho-Drosophila Akt S505 (1:500; Cell Signalling), anti-Akt (1:1,000; Cell Signalling), anti-Phospho-Drosophila p70 S6 Kinase T398 (1:500; Cell Signalling), rabbit anti-Yki (1:1,000, K. Irvine), rabbit anti-phospho Yki S168 (1:1,000; D. Pan), mouse anti-DIAP1 (1:1,000; B. Hay), mouse anti-ß-actin (1:10,000; Abcam), rabbit anti-Histone 3 (1:10,000) and mouse anti-ß-tubulin (1:500; DSHB). HRP-conjugated secondaries (Invitrogen) were used, and blots were developed with an ECL Plus detection kit (Amersham).
For ChIP of Yki and Sd-GFP at the 2B2C locus, the following procedure was undertaken four times to generate four fully independent experimental replicates. Female larvae of the experimental genotype (sd-GFP/y w hs-flp; UAS-TorTED/+; act>CD2>GAL4/Tsc1&2), negative control female larvae (sd-GFP/y w hs-flp; act>CD2>GAL4/+) or YkiRNAi control (sd-GFP/y w hs-flp; UAS-YkiRNAi/+; act>CD2>GAL4/+) were allowed to reach the late larval stage before a 1 hr/37°C heatshock. This caused widespread >CD2> cassette excision leading to an extremely high frequency of clones expressing TORTED and TSC1 + 2, or YkiRNAi. Eight to ten hours later, 200 experimental and control larvae were into PBS + Complete Mini EDTA-Free protease inhibitor (denoted “CPI” Roche). Following the protocol of Estella et al. [69], batches of 25–30 larvae/tube were fixed (25 min; 1.8% formaldehyde in 50 mM HEPES + 1 mM EDTA + 0.5 mM EGTA + 100 mM NaCl + CPI), quenched (6 min; 0.125 M glycine in PBS + 0.01% Triton) and passed through a buffer series to prepare nuclei for lysis (2 washes 10 min Buffer A: 10 mM HEPES + 10 mM EDTA + 0.5 mM + EGTA + 0.25% Triton + CPI, 2 washes 10 min Buffer B: 10 mM HEPES + 200 mM NaCl + 1 mM EDTA + 0.5 mM EGTA + 0.01% Triton + CPI). Individual wing discs were then dissected from larvae and placed in sonication buffer (10 mM HEPES + 1 mM EDTA + 0.5 mM EGTA + CPI), and sonicated on ice. For each genotype, 10% of the fresh chromatin was removed for the input control, and the remainder was split into two samples, each diluted 1:1 with 2xRIPA buffer (280 nM NaCl + 20 mM HEPES + 2 mM EDTA + 2% glycerol + 2% Triton + 0.2% Sodium Deoxycholate). Samples were incubated overnight with either anti-Yki (1:315; D. Pan), anti-GFP (1:300; molecular probes) or anti-Rabbit IgG control (Cell Signalling Technology) antibodies. Antibody-bound chromatin was pulled down for 3–4 hr with Protein A Agarose (Roche), washed several times in 1xRIPA and once in TE. Chromatin was then eluted twice with elution buffer (50 ul 1% SDS, 0.1 M NaHCO3), once at room temperature and once at 55°C. Eluted chromatin was placed at 65°C to reverse crosslinks. Following a 3 hr treatment with Proteinase K, ChIP DNA was extracted with phenol/chloroform, ethanol precipitated and resuspended in TE. Quantitative real time PCR was performed using HotStart-IT SYBR Green qPCR Master Mix (2x) and an Applied Biosystems 7300 RT-PCR machine. For both genotypes from each of the three experimental replicates, DNA was amplified from total input control (serially diluted to avoid saturating the reaction), IgG-, GFP- and Yki-precipitated treatments, using primers for the 2B2C region of diap1 (2B2CF3 5’-AGAAAACTCGAAAGGCAGCTC/2B2CR3 5’-CCAAAACCAAACCAACGAAC) and a control locus (pyruvate dehydrogenase; PDH_F 5’-CGGAAGTGAAGCTGACCAAG/PDH_R 5’-GTAGGTCCATCCGTGGA CAC). The Δct method was used to calculate the degree of enrichment in each treatment relative to the input control DNA.
For ChIP at the bantam locus, chromatin was prepared from ~150 wild type and Tsc1Q87X/Tsc1PA23 mutant wing discs, and subjected to pulldown with ant-Yki or IgG control antibodies as explained above. An ~7 kb region around the microRNA was screened for Yki enrichment, and a strong peak was determined within a region 1,861–1,621 bp 5’ to the 21 bp hairpin, corresponding to positions 640419–640646 on chromosome 3L of the April 2006 assembly of the Drosophila melanogaster genome (BDGP R5/dm3). This region was targeted in qPCR using the primers: bantam3F 5’-CGGGAACAGTCATAAAAGTTG C/bantam3R 5’-CTTTGCCTGTTCTGCCAT CC. Note that this region lies inside the C12 enhancer region bound by Yki identified by Oh and Irvine [56].
To quantify Yki expression from the tub-Yki-GFP transgene, the cassette was removed in clones, and discs were stained for total Yki. The staining intensities of five wild type regions and five regions of clonal tissue were quantified in ImageJ and averaged. The ratio between clonal and wild type tissue was calculated as 1.69.
|
10.1371/journal.pntd.0004391 | Multi-Locus Sequence Typing of Bartonella bacilliformis DNA Performed Directly from Blood of Patients with Oroya’s Fever During a Peruvian Outbreak | Bartonella bacilliformis is the etiological agent of Carrion’s disease, a neglected tropical poverty-linked illness. This infection is endemic of Andean regions and it is estimated that approximately 1.7 million of South Americans are at risk. This bacterium is a fastidious slow growing microorganism, which is difficult and cumbersome to isolate from clinical sources, thereby hindering the availability of phylogenetic relationship of clinical samples. The aim of this study was to perform Multi Locus Sequence Typing of B. bacilliformis directly in blood from patients diagnosed with Oroya fever during an outbreak in Northern Peru.
DNA extracted among blood samples from patients diagnosed with Oroya’s fever were analyzed with MLST, with the amplification of 7 genetic loci (ftsZ, flaA, ribC, rnpB, rpoB, bvrR and groEL) and a phylogenetic analysis of the different Sequence Types (ST) was performed. A total of 4 different ST were identified. The most frequently found was ST1 present in 66% of samples. Additionally, two samples presented a new allelic profile, belonging to new STs (ST 9 and ST 10), which were closely related to ST1.
The present data demonstrate that B. bacilliformis MLST studies may be possible directly from blood samples, being a promising approach for epidemiological studies. During the outbreak the STs of B. bacilliformis were found to be heterogeneous, albeit closely related, probably reflecting the evolution from a common ancestor colonizing the area. Additional studies including new samples and areas are needed, in order to obtain better knowledge of phylogenetic scenario B. bacilliformis.
| The bacteria Bartonella bacilliformis is the etiological agent of Carrion’s disease, which is a neglected poverty-related disease, related to Mountain Andean valleys of Peru, Colombia and Ecuador. This disease, in absence of treatment presents a high mortality during the acute phase, called Oroya’s Fever. The second phase is characterized by the development of dermal eruptions, known as “Verruga peruana” (Peruvian wart). This bacterium is a fastidious slow growing microorganism, being difficult and cumbersome to isolate from clinical sources. Then, the available data about phylogenetic relationship in clinical samples are really scarce, but suggesting high variability. The aim of the study was to perform direct blood analysis of B. bacilliformis Multi Locus Sequence Typing (MLST), a genotyping tool, in patients with Oroya fever during an outbreak. The present study demonstrates that the direct blood PCR, followed by nucleotide sequencing and MLST is a technique useful in the phylogenic characterization of this fastidious microorganism endemic from Andean regions. In this study, we demonstrate that the outbreak of Oroya’s fever was caused by closely related Sequence Typing (ST) microorganisms and, additionally, new STs have been described.
| Carrion’s disease is neglected tropical neglected poverty-linked illness caused by Bartonella bacilliformis. This infection is endemic in low-income areas of Peru, specifically related to Andean regions from Peru, Ecuador and Colombia, covering roughly 145,000Km2 only in Peru, and it is estimated that approximately 1.7 million of South Americans are at risk [1–3]. This illness has two phases: the first, named Oroya’s Fever, mainly affects young children (>60% of cases) and is characterized by fever, acute bacteremia at about 60 days and severe hemolytic anemia [2,4]. Complications are common in this phase, and secondary infections are also frequent due to transient immunosuppression [5]. In the absence of adequate treatment, high levels of mortality (44% to 88%) have been reported [2,4]. The second phase is called “Verruga Peruana” (Peruvian Wart), in which the bacterium induces the proliferation of endothelial cells, resulting in a series of cutaneous lesions [6]. A variety of verrugal lesions are presented in the chronic phase: miliary, nodular and mular [1]. Asymptomatic carriers have also been described in the population from endemic areas (0.5–45%) [7].
B. bacilliformis is a fastidious slow growing microorganism, which is difficult and cumbersome to culture and isolate from clinical sources [2]. Thus, the data available about the phylogenetic relationship of clinical samples of B. bacilliformis are scarce and non-uniform. Indeed, to the best of our knowledge no studies on clonal relations based on Pulsed Field Gel Electrophoresis (PFGE) have been performed, and molecular approaches have been based on PCR methodologies, including Repetitive Extragenic Palindromic PCR (REP-PCR), Enterobacterial Repetitive Intergenic Consensus (ERIC-PCR), Amplified Fragment Length Polymorphism (AFLP), Infrequent Restriction Endonuclease Site PCR (IRS-PCR), analysis of the 16S-23S ribosomal DNA intergenic spacer regions or analysis of the sequence of specific genetic loci such as gltA, ialB and flaA [8–10]. This latter methodology resembles a Multi-locus sequence typing (MLST) technology. MLST approaches are based on housekeeping gene sequencing, being robust, standardized methodology useful to develop epidemiological and evolutionary studies [11]. In fact, MLST schedules have been developed to analyze the phylogenetic relationships of Bartonella henselae [12], and adapted to other Bartonella species, including Bartonella quintana [13] and Bartonella bovis [14]. Furthermore, the use of MLST has been useful in the identification of Bartonella ancashi, new specie of Bartonella genus, closely related to B. bacilliformis [15]. Regarding B. bacilliformis, a MLST schedule has recently been developed based on the sequence of 7 housekeeping genes (bvrR, ribC, ftsZ, groEL, flaA, rnP and rpoB) [16], with 8 different ST being detected in 43 isolates. However, it should be considered that due to the relative isolation of the Andean valleys, the population structure of B. bacilliformis might differ between different endemic areas.
The aim of the study was to perform direct blood MLST of B. bacilliformis from patients diagnosed with Oroya Fever during an outbreak in Northern Peru.
Seven blood samples from Cachachi (Department of Cajamarca in Northern Peru) were collected during March and April 2009 from patients clinically diagnosed with Oroya Fever. Additionally, another two blood samples were collected from Oroya’s Fever patients living in the Condebamba (Cajamarca Department, 50 Km from Cachachi) and Ancash Department in November and October 2011, respectively. Finally, two collection strains isolated in 1941 (CIP 57.19; NCTC12135) and 1949 (CIP 57.18; NCTC12134) from the Pasteur Institute Collection and previously described as belonging to Sequence Type 3 [16] were used as controls (Fig 1). The clinical data and disease presentation of some patients were obtained.
All adult participants provided written informed consent. The study were submitted, revised and approved by the Ethics and Research Committees of the Universidad Peruana de Ciencias Aplicadas in Peru and Hospital Clinic of Barcelona in Spain.
The presence of B. bacilliformis in all the blood samples was confirmed by PCR amplification of 438 bp of the 16S rRNA gene of B. bacillifomis (5’CCTTCA GTTMGGCTGGATC-3’ and 5’-GCCYCCTTGCGGTTAGCACA-3’) as previously described [17]. In all cases the identity of the amplified fragments was confirmed after being visualized in 1.5% agarose gel stained with Sybr Safe and gel recovered using Wizard SV gel and PCR clean up system, (Promega, Madison, WI, USA) following manufacturer's instructions and were sequenced by Macrogen (Seoul, Korea).
The DNA was extracted from 200 μl of blood sample and directly from the control bacterial strains using a commercial extraction kit (High Pure Kit Preparation template, Roche Applied Science, Mannheim, Germany). Blood and bacterial DNA obtained after extraction were eluted in 100 μl of nuclease free water and then processed or stored at -20°C until use.
Internal fragments of the 7 genetic loci (ftsZ, flaA, ribC, rnpB, rpoB, bvrR and groEL) included in the B. bacilliformis MLST schedule were amplified as previously described [16]. Reaction mixtures were exposed to denaturation at 96°C for 5 min followed by 50 cycles of 96°C for 40 sec, 55°C for 40 sec and 72°C for 50 sec, with a final extension step of 72°C for 10 minutes. Amplified fragments were visualized in 1.5% agarose gel stained with Sybr Safe and subsequently gel recovered using Wizard SV gel and PCR clean up system, (Promega, Madison, WI, USA) following manufacturer’s instructions and sequenced by Macrogen (Seoul, Korea).
Phylogenetic relationship analyses were conducted using MEGA version 5 [18]. The phylogenetic tree was constructed by UPGMA (Unweighted Pair Group Method with Arithmetic Mean Analysis). The phylogenetic tree was inferred from 500 bootstrap replicates. The sequences of all the alleles described previously were obtained from Genbank (accession numbers JF326267 to JF326294) and were ordered according to the corresponding Sequence Type (ST) in order to develop the phylogenetic tree.
The mean age of patients studied was 25.9 years (SD = 13.77, IC95% = 19.5–32.3), 44.4% being female. Among the 5 patients from whom clinical data were recovered, all (100%) presented fever (>38°C) and malaise, 4 (80%) reported chills, myalgia and pallor, 3 (60%) headache, 2 (40%) reported jaundice and arthralgia and only one patient (20%) presented vomiting (Table 1). In 3 cases the treatment was recorded, in all cases being ciprofloxacin alone (2 cases) or with ceftriaxone (1 case) during 14 days.
Among the 9 blood samples analyzed, a total of 4 different B. bacilliformis STs were identified. The most frequently found was ST1, present in 6 out of 9 (66%) samples, all from the Cajamarca Department (5 out of 7 belonging to the Cachachi outbreak, and that of Condebamba), while the sample from the Ancash Department belonged to ST4 (Fig 1). Additionally, two samples from the Cachachi outbreak presents a new allelic profile, belonging to new STs, which were classified as ST9 (1,2,1,1,1,1,1) and ST10 (1,1,1,1,1,3,1) respectively. The 2 collection strains were classified as ST3 (Table 2).
On determination of phylogenetic relationships between the ST9 and ST10 and the previously described ST, they were found to be closely related to ST1, differing in only 1 of the 7 alleles (Fig 2).
Studies on the clonality and phylogeny of B. bacilliformis are scarce. This may be due to the slow growth of this bacterium and a series of specific requirements which directly affect the culture. The present study demonstrates MLST studies of B. bacilliformis may be performed directly from blood samples thereby avoiding the difficult step of culturing this microorganism. However, a series of limitations that may limit the usefulness of this technique should be taken into account. Among these, definitive results may not be obtained in the hypothetical case of polyclonal infections when the infecting isolates belong to different STs. Along this line, although to the best of our knowledge coinfection by two different B. bacilliformis clones has not been described to date, coinfections by different Bartonella variants has been reported in cotton rats [19]. In the present report, new MLST were not related to artefactual overlapping of sequences belonging to different B. bacilliformis causing a concomitant infection, because no double peaks were observed in any DNA sequence. Another limitation is that direct blood PCR approaches in the study of asymptomatic B. bacilliformis carriers do not have enough power due to the low bacterial burden [20].
The present study demonstrates the heterogeneity of the B. bacilliformis population. Highly clonality has also been found in other species of Bartonella, such as B. quintana, a re-emerging pathogen causing trench fever [13]. Thus, 3 different ST (ST1, ST9 and ST10) were recovered amongst the samples analyzed from the Cachachi outbreak. However, it is of note that these 3 STs were closely related to each another, and thus may reflect the evolution from a common ancestor colonizing the area.
To date only one study has determined the MLST of B. bacilliformis isolates [16]. In this study ST1 was found to be widely distributed in central and northern areas of Peru, accounting for 46% of the samples analyzed, including samples from the 1960's. In addition, ST1 has been detected in the neighboring San Martin Department, and thus, its presence in the Cajamarca Department is not surprising. ST2, ST3, ST4 and ST8 have been described in the center of the country, similar the present sample belonging to ST4. Meanwhile, ST5 has been observed in southern isolates and ST6 and ST7 in the north of the country.
Some techniques such as PFGE or REP-PCR are useful for the description of clones and specific outbreak characterization, as for example virulence, being of special interest to study recent genetic events. On the other hand, techniques such as MLST classification describe ancient genetic differentiations that may underlie more in depth differences [11–13]. For example, specific STs could possess increased virulence or may have a greater facility to develop either acute or chronic infection, or to remain undetected in asymptomatic carriers. Unfortunately, the scarce data on STs of B. bacilliformis make it difficult to delineate these aspects.
The clinical data of only a few patients were recorded; however the symptoms reported are in accordance with those more extensively described, such as the presence of fever, pallor, malaise and headache in acute cases, [17]. Some of these symptoms are a consequence of hematological complications, such as pallor, while others like vomiting or jaundice are related to gastrointestinal problems [21]. Although this disease mainly affects children under 14 years of age (more than 60% of cases) [22], in our study the youngest patient of the Cachachi outbreak was 15 years old. This may be related to the outbreak nature of the samples. Fortunately, all patients who receipt treatment respond well to the treatment. The currently recommended treatment for the acute phase of Oroya’s Fever includes the use of ciprofloxacin as first line therapy in adults and children >14 years, while chloramphenicol, cotrimoxazole, amoxicillin plus clavulanic acid and ceftriaxone are used as a second line or children and pregnant women [23]. Fortunately, up to now, the levels of antibiotic resistance reported among B. bacilliformis have shown that this microorganism is highly susceptible to the antibiotics tested [24].
In summary, this is the first report of MLST of B. bacilliformis performed in direct blood samples, with two new ST variants being described. Present data highlight the need to extend the studies to new samples and geographical areas, in order to provide a better picture of the situation, which will allow specific STs of B. bacilliformis to be associated with clinical symptoms, and the severity or phase of the disease.
|
10.1371/journal.pntd.0002396 | Adhesins of Leptospira interrogans Mediate the Interaction to Fibrinogen and Inhibit Fibrin Clot Formation In Vitro | We report in this work that Leptospira strains, virulent L. interrogans serovar Copenhageni, attenuated L. interrogans serovar Copenhageni and saprophytic L. biflexa serovar Patoc are capable of binding fibrinogen (Fg). The interaction of leptospires with Fg inhibits thrombin- induced fibrin clot formation that may affect the haemostatic equilibrium. Additionally, we show that plasminogen (PLG)/plasmin (PLA) generation on the surface of Leptospira causes degradation of human Fg. The data suggest that PLA-coated leptospires were capable to employ their proteolytic activity to decrease one substrate of the coagulation cascade. We also present six leptospiral adhesins and PLG- interacting proteins, rLIC12238, Lsa33, Lsa30, OmpL1, rLIC11360 and rLIC11975, as novel Fg-binding proteins. The recombinant proteins interact with Fg in a dose-dependent and saturable fashion when increasing protein concentration was set to react to a fix human Fg concentration. The calculated dissociation equilibrium constants (KD) of these reactions ranged from 733.3±276.8 to 128±89.9 nM for rLIC12238 and Lsa33, respectively. The interaction of recombinant proteins with human Fg resulted in inhibition of fibrin clot by thrombin-catalyzed reaction, suggesting that these versatile proteins could mediate Fg interaction in Leptospira. Our data reveal for the first time the inhibition of fibrin clot by Leptospira spp. and presents adhesins that could mediate these interactions. Decreasing fibrin clot would cause an imbalance of the coagulation cascade that may facilitate bleeding and help bacteria dissemination
| Leptospirosis is probably the most widespread zoonosis in the world. Caused by spirochaetes of the genus Leptospira, it has greater incidence in tropical and subtropical regions. The disease has become prevalent in cities with sanitation problems and a large population of urban rodent reservoirs, which contaminate the environment through their urine. Understanding the mechanisms involved in pathogenesis of leptospirosis should contribute to new strategies that would help fight the disease. We show in this work that Leptospira strains, virulent, attenuated or saprophytic are capable of binding fibrinogen (Fg). The interaction of leptospires with Fg inhibits the formation of fibrin clot that may result of an imbalance in the haemostatic equilibrium. In addition, we show that plasminogen (PLG)/plasmin (PLA) generation on the surface of leptospires can lead to Fg degradation, showing evidence of possible route of fibrinolysis in leptospirosis. We also present six leptospiral proteins, as novel Fg-binding proteins, capable of inhibiting fibrin clot formation by thrombin-catalyzed reaction, suggesting that in Leptospira these multifunctional proteins could mediate Fg interaction. Our data suggest possible mechanisms that leptospires could employ to affect the coagulation cascade and fibrinolytic system that might lead to bacteria spreading.
| The spirochete Leptospira interrogans is a highly invasive pathogen and the causal agent of leptospirosis, one of the most widespread zoonosis of human and veterinary concern [1], [2]. The transmission occurs through contact with environmental water contaminated by leptospires shed in the urine of animal carriers [1], [3]. Humans are accidental and terminal hosts in the transmission process of leptospirosis. The leptospires enter the body via abrasions on skin or actively through mucosa, spreading to any tissue, and colonizing target organs [4], [3]. Leptospira can cause damage of the endothelium of small blood vessels, leading to hemorrhage and localized ischemia in multiple organs. As a consequence, renal tubular necrosis, hepatocellular damage and development of leptospirosis-associated pulmonary hemorrhage syndrome (LPHS) may occur in the host [1], [5], [6]. The mechanisms responsible for bleeding in leptospirosis are poorly understood. Hemolysins could play an important role in this toxic response and several genes coding for predicted hemolysins were identified in the genome sequencing L. interrogans [7]. Yet, when evaluated, these proteins failed to show hemolytic activity in human erythrocytes [8], [9].
Fg and PLG are key proteins in the coagulation cascade and fibrinolysis, respectively, and critical determinants of bacterial virulence and host defense [10]. Fg is the major clotting protein present in blood plasma with an important role in blood coagulation and thrombosis. PLG under the action of its activators generates PLA, a serine-protease capable of degrading ECM components, fibrin, facilitating the pathogen penetration and invasion [11].
We have previously shown that PLA-associated Leptospira renders the bacteria with proteolytic activity capable of degrading ECM components [12] that in turn, may help bacterial penetration and dissemination. Furthermore, PLA-coated leptospires have also shown to degrade IgG and C3b that could facilitate the bacterial immune evasion [13]. The adhesion of physiological osmotically induced Leptospira with Fg was described but their effect on fibrin formation was not ascertained [14]. The leptospiral proteins LigB and OmpL37 were shown to interact with Fg and LigB was reported to reduce fibrin clot formation [15], [16], [17].
We thus decided to evaluate if Fg-associated Leptospira was capable to inhibit the fibrin clot formation and the ability of seven recombinant proteins to act as leptospiral Fg-receptors. We report that Leptospira strains and the recombinant ECM- and PLG-interacting proteins, rLIC12238 [18], Lsa33, Lsa25 [19], Lsa30 [20], OmpL1 [21], rLIC11360 and rLIC11975 [22], are capable to adhere to Fg. We also show that this interaction inhibits fibrin clot formation by thrombin-catalyzed reaction. Moreover, PLA-coated Leptospira was capable to degrade Fg. Altogether, the results suggest possible pathways that Leptospira may interfere with the coagulation/bleeding process.
All animal studies were approved by the Ethical Committee for Animal Research of Instituto Butantan, São Paulo, SP, Brazil under protocol no 798/11. The Committee in Animal Research in Instituto Butantan adopts the guidelines of the Brazilian College of Animal Experimentation.
The non-pathogenic L. biflexa serovar Patoc strain Patoc 1, the pathogenic attenuated L. interrogans serovar Copenhageni strain M-20 and the virulent strains of L. interrogans serovar Kennewicki strain Pomona Fromm (LPF) and serovar Copenhageni strain Fiocruz L1-130 were cultured at 28°C under aerobic conditions in liquid EMJH medium (Difco) with 10% rabbit serum, enriched with L-asparagine (wt/vol: 0.015%), sodium pyruvate (wt/vol: 0.001%), calcium chloride (wt/vol: 0.001%), magnesium chloride (wt/vol: 0.001%), peptone (wt/vol:0.03%) and meat extract (wt/vol: 0.02%) [23]. Leptospira cultures are maintained in Faculdade de Medicina Veterinária e Zootecnia, USP, São Paulo, SP, Brazil. Unless otherwise stated, experiments were performed with leptospires resuspended in low salt – lsPBS (50 mM), instead of PBS that contains 137 mM NaCl, because it is an osmolarity condition closer to cultivation, which is approximately 35 mM.
For the immunofluorescence assay (IFA), live L. interrogans sorovar Copenhageni strain M-20 suspensions (∼108) were harvested at 5,000× g for 15 min, washed three times with lsPBS, resuspended in 200 µL lsPBS, and fixed with 200 µL with 4% paraformaldehyde for 40 min at 30°C. After the fixation, propidium iodide (Sigma) diluted 1∶50 in lsPBS was added to stain the nuclei, and the suspensions were incubated for 45 min at 30°C. After this time, the leptospires were gently washed three times with lsPBS and blocked with lsPBS containing 5% BSA for 90 min at 30°C, followed by incubation with 16 µg Fg in 200 µL lsPBS for 60 min at 37°C. The leptospires were washed three times with lsPBS plus 1% BSA and incubated with goat-produced antiserum against human Fg at a 1∶50 dilution for 45 min at 37°C. Leptospires were washed three times again and incubated with rabbit anti-goat IgG antibodies conjugated to fluorescein isothiocyanate (FITC; Sigma) at a dilution of 1∶400 for 45 min at 37°C. After this incubation, the leptospires were washed twice with lsPBS containing 1% BSA and once with distilled water. Finally they were resuspended in lsPBS-antifading solution (ProLong Gold; Molecular Probes). The immunofluorescence-labeled leptospires were examined by use of a confocal LSM 510 META immunofluorescence microscope (Zeiss, Germany). As a control for unspecific binding, Fg was absent from the reaction mixtures.
For accessing the leptospiral binding to soluble human Fg, ELISA plates were coated with 108 leptospires in lsPBS per well (L. interrogans sorovar Copenhageni strain Fiocruz L1-130, L. interrogans serovar Copenhageni strain M-20, and L. biflexa serovar Patoc strain Patoc 1), allowed to set for 3 h at 37°C and then blocked with 200 µL of phosphate-buffered saline containing 0.05% Tween (PBS-T) with 10% non-fat dry milk. After 2 h incubation, plates were washed with PBS-T and then 100 µL of a Fg solution (Fg- Sigma) (10 µg/mL in lsPBS) were added per well and incubation proceeded for 2 h. After extensive washing, 100 µL of solution containing goat anti-Fg (1∶25,000 in lsPBS) were added per well and incubation was carried out for 1 h at 37°C. After three washings, 100 µL of solution containing peroxidase (HRP)-conjugated rabbit anti-goat IgG (1∶50,000 in lsPBS) were added per well and the reaction continued for 1 h at 37°C. The wells were washed three times, and o-phenylenediamine (OPD) (1 mg/mL) in citrate phosphate buffer (pH 5.0) plus 1 µL/mL H2O2 was added (100 µL per well). The reaction was allowed to proceed for 10 min and interrupted by the addition of 50 µL of 8 M H2SO4. Readings were taken at 492 nm in a microplate reader (Multiskan EX; Thermo Fisher Scientific, Helsinki, Finland). The binding of Fg to each leptospiral strain was performed in triplicate and a negative control in which Fg was omitted was included in the experiment. In the no-cell control, BSA replaced leptospires. For statistical analyses, the binding of Fg to different strains and control BSA was performed using one-way ANOVA, followed by Tukey post-test for pairwise comparisons.
Live leptospires (L. interrogans sorovar Copenhageni strain Fiocruz L1-130, L. interrogans serovar Copenhageni strain M-20, and L. biflexa serovar Patoc strain Patoc 1), 109, 107, 105 cells/mL were harvested at 5,000× g for 20 min at room temperature, washed once with lsPBS, resuspended in 0.5 mL of lsPBS plus 1 mg/mL of Fg (Sigma), and incubated for 2 h at 37°C. ELISA plates were placed with 90 µL/well of leptospires plus Fg and 10 µL/well of thrombin (10 U/mL - Sigma). The fibrin clot formation was measured every 1 min for 10 min and then every 5 min for 35 min by an ELISA plate reader (Multiskan EX Thermo Fisher Scientific) at OD595nm. The positive control of the reaction employed Fg (1 mg/mL) plus thrombin (10 U/mL) while in the negative control thrombin was omitted. For statistical analyses, inhibition of fibrin clot formation was performed by one-way ANOVA, followed by Tukey post-test for pairwise comparisons. Three independent experiments were performed.
Virulent L. interrogans serovar Kennewicki strain Pomona Fromm (108 leptospires/sample) were treated in 200 µL lsPBS with the addition of: (a) 10 µg PLG (b) 3U urokinase (uPA) (c) 10 µg PLG and 3U uPA (generating PLA-coated leptospires) or (d) only lsPBS with no additions (untreated). The bacteria were incubated for 1 h at 37°C with the PLG, and for one more hour after the addition of uPA. The cells were washed three times with lsPBS to remove the free PLG, uPA and PLA. Then, the bacteria were resuspended in 100 µL lsPBS containing goat anti-human Fg human purified Fg (Sigma, USA) as substrate, and incubated for 16 h at 37°C. Three additional controls were employed: (a) one sample received 1 µg of the protease inhibitor aprotinin (Sigma, USA) to the Fg-leptospires incubation, (b) one sample contained only Fg without leptospires, and (c) one sample received the PLG and uPA treatment without the addition of leptospires to rule out the free PLA Fg degradation. The leptospires were removed by centrifugation, 20 µL of the supernatants were separated by 8% SDS-PAGE and then transferred to nitrocellulose membranes in semi-dry equipment. The membranes were blocked by incubating overnight at 4°C with 5% non-fat dry milk and 1% BSA. The Fg detection was performed by incubations with goat anti-human Fg antibodies and rabbit anti-goat secondary antibodies conjugated with HRP, followed by ECL (GE Healthcare) development.
In another assay, after the incubation for Fg degradation, the pelleted leptospires and the remaining cell-free supernatants were also evaluated for the presence of Fg by ELISA. The cells were resuspended in 100 µL lsPBS and coated onto ELISA plates overnight at 4°C, as well as the cell-free supernatants previously diluted 20 times. The plates were washed three times with PBS-T and blocked with solution containing 5% BSA and 5% non-fat dry milk for 2 h at 37°C, following incubations with anti-Fg polyclonal antibodies (1∶25,000) and secondary antibodies conjugated to peroxidase (1∶50,000). The reactions were developed by addition of 100 µL/well of solution containing 1 mg/mL OPD and 1 µL/mL H2O2, and stopped by addition of 50 µL/well H2SO4. The samples in which the leptospires received no treatment (only PBS) were considered as having 100% of Fg for comparative purposes.
Predicted coding sequences (CDSs) were analyzed for their cellular localization by PSORT program, http://psort.nibb.ac.jp [24], [25]. The web servers SMART, http://smart.embl-heidelberg.de/ [26], [27], PFAM, http://www.sanger.ac.uk/Software/Pfam/ [28], and LipoP, http://www.cbs.dtu.dk/services/LipoP/ [29] were used to search for predicted functional and structural domains within the amino acid sequences of the CDSs.
Amplification of the CDSs was performed by PCR from L. interrogans serovar Copenhageni strain M-20 genomic DNA using complementary primer pairs (Table 1). The gene sequences were amplified without the signal peptide tag and cloned into pAE expression vector [30]. The final constructs were verified by DNA sequencing on an ABI Prism 3730_L sequencer (Seq- Wright, Houston, TX) with appropriate vector-specific T7 (F: TAATACGACTCACTATAGGG) and pAE (R:CAGCAGCCAACTCAGTTCCT) primers. Detailed of cloning, expression and purification of the recombinant proteins Lsa33, Lsa25, rLIC12238, Lsa30, OmpL1, rLIC11360 and rLIC11975 has been previously described [18], [19], [21], [20], [22]. The proteins rLIC11360 and Lsa30 were kept at pH 12 required for their solubility.
Five female BALB/c mice (4–6 weeks old) were immunized subcutaneously with 10 µg of each recombinant protein adsorbed in 10% (vol/vol) of Alhydrogel (2% Al(OH)3, Brenntag Biosector, Denmark), used as adjuvant. Two subsequent booster injections were given at 2-week intervals with the same recombinant proteins preparation. Negative - control mice were injected with PBS plus Alhydrogel. Two weeks after each immunization, the mice were bled from the retro - orbital plexus and the pooled sera were analyzed by ELISA for determination of antibody titers.
The binding of the recombinant proteins to Fg was evaluated by a modified ELISA, as follows: 96-well plates (Costar High Binding, Corning) were coated overnight in PBS at 4°C with 100 µL of 10 µg/mL of the human Fg (Sigma); LigB7-12 protein of L. interrogans, known as having Fg-binding domain [31], [17], generous donation from Dr. Odir Dellagostin, UFPEL, RS, Brazil, was employed as a positive control. PspA, an outer membrane protein of Streptococcus pneumoniae [32], kind gift from Dr. Luciana Leite, Instituto Butantan, SP, Brazil, was employed as an unrelated negative control protein. Gelatin (Difco) was also employed as negative protein control. Plates were washed three times with PBS-T and blocked for 2 h at 37°C with PBS with 10% (wt/vol) non-fat dry milk. The blocking solution was discarded and 100 µL of 10 µg/mL recombinant proteins in PBS was incubated for 2 h at 37°C. Wells were washed three times with PBS-T and incubated for 1 h at 37°C with polyclonal mouse antiserum produced against each recombinant protein. The serum dilution used was determined by titration curve with the corresponding recombinant protein and the value of 1.0 at OD492nm was employed. These values are: 1∶1,000 for rLIC12238; 1∶750 for Lsa33; 1∶500 for rLIC11975 and rLIC11360; 1∶400 for Lsa30, 1∶800 for OmpL1, 1∶500 for Lsa25 and 1∶200 for PspA. After incubation, plates were washed again and incubated with HRP-conjugated anti-mouse immunoglobulin G (IgG), diluted 1∶5,000 in PBS. After three washings, 100 µL/well of 1 mg/mL OPD plus 1 µL/mL H2O2 in citrate phosphate buffer (pH 5.0) were added. The reactions were carried out for 15 min and stopped by the addition of 50 µL/well of H2SO4 (8 M). Readings were taken at OD492 nm. In another assay, the assessment of bound proteins was performed by incubation for 1 h at 37°C with monoclonal anti-polyhistidine-HRP (Sigma) at appropriate dilutions: 1∶5,000 for rLIC12238; 1∶10,000 for rLIC11975 and rLIC11360; 1∶1,000 for Lsa30; 1∶500 for Lsa33, 1∶400 for OmpL1, 1∶200 for Lsa25 and 1∶200 for LigB7-12. The reaction was developed with 1 mg/mL OPD plus 1 µL/mL H2O2, as described above. The rate of interaction of recombinant proteins to Fg was determined by measuring the reaction as a function of time. The OD492nm value after 2 h interaction was considered the maximal binding (100%) and was used for statistical analyses, using Student's two-tailed t test. For determination of dose-response curves of the binding of recombinant proteins to human Fg, protein concentrations varying from 0 to 4,000 nM in PBS were used. Binding was detected with polyclonal antibodies against each protein at the dilution described above, except for LigB7-12 where monoclonal anti-polyhistidine-HRP was used (1∶200). For statistical analyses, the binding of recombinant proteins to human Fg was compared to its binding to gelatin by Student's two-tailed t test.
96-well plates were coated with 1 µg of Fg in 100 µL of PBS and allowed to set overnight at 4°C. The wells were washed three times with PBS-T and then blocked with 200 µL of 10% (wt/vol) nonfat dry milk for 2 h at 37°C. Prior to the next step, the proteins were incubated for 1 h at 37°C with the respective antibodies diluted 1∶50 in 100 µL of PBS. After the incubation, each recombinant protein (1 µg) was added per well in 100 µL of PBS, and allowed to attach to Fg for 90 min at 37°C. Polyclonal serum obtained in mice against another leptospiral protein, DnaK (1∶50), was employed as a control that is not specific for the studied proteins. After washing six times with PBS-T, bound recombinant proteins were detected by adding monoclonal HRP-conjugated mouse anti-polyhistidine-HRP (Sigma) at dilutions described above. Incubation proceeded for 1 h at 37°C. The detection was performed with OPD, as previously described. BSA or gelatin was used as negative control (data not shown). For statistical analyses, the attachment of blocked recombinant proteins to Fg was compared to the binding with the untreated proteins by the two-tailed t test (* P<0.05).
ELISA plates were coated overnight at 4°C with 100 µL of 10 µg/mL Fg. Plates were washed three times with PBS-T and blocked with 200 µL of 10% (wt/vol) nonfat dry milk for 2 h at 37°C. The recombinant proteins were denatured by incubation at 96°C for 10 min; 1 µg of each was added per well in 100 µL of PBS. The recombinant proteins were allowed to attach to Fg at 37°C for 90 min. After washing six times with PBS-T, bound recombinant proteins were detected by incubation with mouse serum raised against the respective protein (dilutions described above) at 37°C for 1 h. After three washings with PBS-T, 100 µL of a 1∶5,000 dilution of HRP-conjugated rabbit anti-mouse IgG (Sigma) in PBS was added per well for 1 h at 37°C. The detection was performed with OPD, as previously described. BSA or gelatin was used as negative control (data not shown). For statistical analyses, the attachment of denatured recombinant proteins to Fg was compared to untreated recombinants binding by the two-tailed t test (* P<0.05).
The ELISA data were used to calculate the dissociation equilibrium constant (KD) according to the method previously described [33] based on the equation: KD = (Amax.[protein])/A) – [protein], where A is the absorbance at a given protein concentration, Amax is the maximum absorbance for the ELISA plate reader (when the equilibrium is reached), [protein] is the protein concentration and KD is the dissociation equilibrium constant for a given absorbance at a given protein concentration (ELISA data point).
The assay of thrombin-catalyzed fibrin clot inhibition was performed in the presence of recombinant proteins. We have employed the concentration of recombinant proteins in which there was a binding saturation to Fg. Proteins that do not bind Fg were employed at the same concentration range. Protein concentration employed: Lsa33 – 3,000 nM, OmpL1 – 3,000 nM, rLIC12238 – 3,500 nM, rLIC11975 – 2,000 nM, LigB7-12 – 1,500 nM, Lsa25 – 3,000 nM, PspA – 3,000 nM. Recombinant proteins were resuspended in 0.5 mL of PBS plus 1 mg/mL of Fg and incubated for 2 h at 37°C. ELISA plates were coated with 90 µL/well of recombinant proteins plus Fg and 10 µL/well of thrombin (10 U/mL). The fibrin clot formation was measured as previously described. Reduction of the fibrin clot formation was calculated by comparing the value of the last reading point, at 45 min, with the positive control (100%). Analysis was performed using one-way ANOVA, followed by Tukey post-test for pairwise comparisons.
ELISA plates were coated with Fg (1 µg/well) in PBS and allowed to set overnight at 4°C; the wells were then washed and blocked with 10% non-fat dry milk in PBS-T for 2 h at 37°C. The blocking solution was discarded, and the wells were incubated for 90 min at 37°C with increasing concentrations of recombinant proteins (0 to 4.5 µM). After three washings, 100 µL/well of 4×107 live L. interrogans serovar Copenhageni strain M-20 were added for 90 min at 37°C. The unbound leptospires were washed and the quantification of bound leptospires was performed indirectly by anti-LipL32 antibodies produced in mice (1∶4,000), based on the fact that LipL32 is a major expressed membrane leptospiral protein [34]; the procedure was followed by the addition of HRP-conjugated anti-mouse IgG antibodies, essentially as described in Atzingen et al., (2008) [35]. The detection was performed by OPD, as above described.
The ability of L. interrogans sorovar Copenhageni strain M-20 cells to bind human Fg was performed by immunofluorescence assay (IFA). Leptospires were visualized by propidium iodide staining (Fig. 1, panel A) followed by protein detection with goat anti- human Fg, in the presence of anti-goat IgG antibodies conjugated to FITC. Green fluorescence could be observed for Fg (Fig. 1 -Fg1B, Fg2B). The localization of the protein-green light within the leptospires was achieved by superimposing both fields and the results obtained are shown in Fig. 1 - Fg1C and Fg2C. The FgØ shows the control of the reaction in which Fg was absent.
We have evaluated the capability of soluble human Fg (10 µg/mL) to interact with immobilized strains of leptospires by ELISA. We performed the experiments using virulent L. interrogans serovar Copenhageni strain Fiocruz L1-130, pathogenic attenuated L. interrogans serovar Copenhageni strain M-20 and one saprophytic L. biflexa serovar Patoc strain Patoc 1. The binding of Fg to each leptospiral strain was performed in triplicate and the data represent the mean ± the standard deviation from one representative experiment is depicted in Fig. 2A. The measurements were performed in the presence (Fg+) and absence (Fg−) of human Fg while in the no-cell control, BSA replaced leptospires. For statistical analyses, the attachment of Fg to leptospiral strains was performed by one-way ANOVA followed by Tukey post-test for pairwise comparison, asterisks above the bars refer to comparison to BSA (no cell control) (**P<0.01). The comparison among leptospiral strains is also shown (#P<0.05 and ## P<0.01). The results show that all strains tested were able of binding human Fg, but the virulent strain was more efficient. We have assessed the effect of Leptospira-bound to Fg on the inhibition of thrombin-catalyzed fibrin clot formation. The reaction was analyzed with the same strains and readings were taken at OD595nm every 1 min for the first 10 min and then every 5 min for 35 min. The complete reaction, Fg plus thrombin, was used as a positive control while in the negative one, thrombin was missing. The determination was performed in two independent experiments and a representative assay is shown in Fig. 2B. The data show that all the strains studied promoted an inhibition of fibrin formation statistically significant compared to the reaction positive control. However, the small difference observed between the strains pathogenic and saprophyte was not statistically relevant.
The effect of PLA-coated virulent L. interrogans serovar Kennewicki strain Pomona Fromm on human Fg was evaluated by Western blotting using anti-human Fg antibodies (Fig. 3A). The results show that PLA generation on the surface of Leptospira cause degradation of human Fg (Fig. 3A, lane 5), which is not observed when at least one of the reaction components is missing (Fig. 3A, lanes 1, 2, 3and 4) and completely prevented in the presence of a serine protease inhibitor, aprotinin (Fig. 3A, lane 6). The data suggest that PLA-coated leptospires were capable to employ their proteolytic activity to interfere with one of the coagulation cascade substrate.
The Fg degradation by PLA-coated leptospires was also evaluated by ELISA. When leptospires were treated with PLG+uPA, the detection of Fg remaining in cell-free solution is decreased when compared to untreated, only PLG or only uPA-treated or when aprotinin was added to the Fg incubation (Fig. 3B). An additional control lacking leptospires was added in order to rule out the contribution of free PLA to the Fg degradation. As expected, the data showed that the remaining Fg level in this control was comparable with the non-proteolytic controls. Cell-bound Fg was also evaluated. PLA-coated leptospires retain Fg binding ability, though it seems to be diminished (Fig. 3B), probably due to Fg degradation.
The rationale for protein selection was mostly based on cellular localization, since surface proteins are potential receptors for Fg. We have selected seven proteins, all of them previously shown to be leptospiral adhesins and were already published, rLIC12238 [18], [36], Lsa33, Lsa25 [19], Lsa30 [20], OmpL1 [21], rLIC11360 and rLIC11975 [22]. Table 1 summarizes features of the selected proteins, gene locus, given name, gene conservation within the sequenced genomes, the sequences of primers used for cloning techniques and molecular mass.
The amplified coding sequences, excluding the signal peptide tags, were cloned and expressed as full-length proteins in E. coli. The recombinant proteins were expressed with 6×His tag at the N-terminus and purified by nickel affinity chromatography, as previously described [35].
Proteins of Leptospira have been reported to bind Fg [15], [16], [17]. We thus decided to investigate whether the selected surface-exposed proteins were capable of binding human Fg in vitro. Seven recombinant leptospiral proteins, expressed and purified in our laboratory, LigB7-12, used as positive control [31], [16], [17], and the negative protein controls gelatin and pneumococcal recombinant protein PspA, were individually placed onto 96-wells and incubated with previously immobilized human Fg. The binding was quantified by ELISA and the results obtained from three independent experiments are shown in Fig. 4. The protein bound to Fg was probed with the respective homolog polyclonal antiserum raised in mice (Fig. 4A) or with the monoclonal antibody anti-histidine tag (Fig. 4B). The percentage of binding for each protein with Fg, measured as a function of time is shown in Fig. 4C. The protein rLIC11360 promptly reacted with Fg with 40% of binding achieved after 5 min reaction, contrasting with rLIC11975 that showed very low binding activity at this time (Fig. 4C). The binding of recombinant proteins to Fg was also assessed after blocking the proteins with the corresponding antibody and after submitting them to denaturing conditions at 96°C for 10 min (Fig. 4D). The binding was totally inhibited by antibody-blocked protein in the case Lsa30, Lsa33, rLIC11975, rLIC12238 and rLIC11360, while a very low percentage of the binding remained for OmpL1 (8.4%), suggesting the participation of non- immunogenic epitopes on the interaction (Fig. 4D). Anti-DnaK serum, control employed as unrelated antibody, promoted a partial decrease on the binding of rLIC11360 and Lsa30, no effect with the protein Lsa33, while a slightly increase was detected with the proteins rLIC11975, rLIC12238 and OmpL1. At any rate, these differences were not statistically significant and are probably due to the presence of non-specific antibodies in the polyclonal serum. The adhesion of heat-denatured proteins to Fg was almost totally abolished in the case of rLIC12238 and OmpL1, 24% remained with rLIC11360, while 63–77% of the binding continued with Lsa30, Lsa33 and rLIC11975 (Fig. 4D). The results suggest that for some proteins the binding depends on their conformational structures while with others it probably relies on their linear primary conformation.
The interactions between the recombinant proteins and Fg was assessed on a quantitative basis, as indicated in Fig. 5A. Dose-dependent and saturable binding was observed when increasing concentrations (0 to 4,000 nM) of recombinant proteins rLIC12238, Lsa33, OmpL1, rLIC11975 and rLIC11360, or (0 to 2,000) of Lsa30, were allowed to individually adhere to a fixed human Fg amount (10 µg/mL) for 2 h. Saturation was reached with all except Lsa30 protein, due to the impossibility to achieve this protein at higher concentrations. In the case of LigB7-12, dose-response curve was measured using monoclonal anti-His tag antibodies, and the results depicted in the insert of Fig. 5A shows that saturation was not reached in the protein concentration range employed. Based on the ELISA data, the calculated dissociation equilibrium constants (KD) for the recombinant proteins with Fg are depicted in Fig. 5B; the highest and the lowest KD values were for rLIC12238 (733.3±276.8 nM) and Lsa33 (128±89.9 nM), respectively.
Since Leptospira bound to Fg inhibited thrombin-catalyzed fibrin formation, we decided to investigate whether the recombinant proteins bound to Fg were capable to mediate this interaction. The concentration of recombinant proteins used was the one in which there was a binding saturation to Fg (Lsa33 – 3,000 nM, OmpL1 – 3,000 nM, rLIC12238 – 3,500 nM and rLIC11975 – 2,000 nM). LigB7-12, previously shown to inhibit thrombin-catalyzed fibrin formation [16], [17], was employed as positive control at 1,500 nM. Recombinant proteins Lsa25 and pneumococcal PspA that do not interact with Fg, were employed as negative controls, at 3000 nM. Each recombinant protein was pre-incubated with 1 mg/mL of Fg at 37°C for 2 h. The reaction mixtures were used to coat ELISA plates and 10 U/mL of thrombin was added per well. The fibrin clot formation was measured every 1 min for 10 min and then every 5 min for 35 min. The percentage of inhibition was calculated as a function of time, taking the complete reaction at the last time-point, in the absence of proteins, as a 100% fibrin formation (positive control). In the negative control, thrombin was omitted from the reaction. The measurements were performed in triplicate and representative curves of two independent experiments are shown in Fig. 6. The proteins rLIC11360 and Lsa30 were not employed on these assays because they were kept at pH 12 for solubility and thrombin is not active above pH 10. The data show that the four proteins tested, Lsa33, rLIC12238, rLIC11975 and OmpL1, elicited an inhibition of 40–50% on the fibrin clot formation, similar to the one elicited by LigB7-12 (Fig. 6). This effect on clot formation by the recombinant proteins was similar with all of them, although a lag time was observed within the rLIC12238 inhibition curve. The inhibition promoted by all proteins was statistically relevant when compared to the positive reaction control. The results are consistent with the KD of the proteins towards Fg that are in the same order of magnitude. In contrast, Lsa25 and PspA, proteins that do not react with Fg, were not capable to inhibit thrombin-catalyzed fibrin production, resulting in curves similar to the positive control of the reaction (Fig. 6).
The inhibitory effect exerted by recombinant proteins on leptospiral adherence to Fg was quantified by ELISA. Fg-coated microtiter wells were incubated with increasing concentration (0–4.5 µM) of rLIC12238 (Fig. 7A), rLIC11975 (Fig. 7B), Lsa30 (Fig. 7C), rLIC11360 (Fig. 7D), Lsa33 (Fig. 7E) and OmpL1 (Fig. 7F) for 90 min prior to the addition of 4×107 L. interrogans serovar Copenhageni strain M-20. Wells were probed with anti-LipL32 serum, given the fact that LipL32 is a major membrane leptospiral protein [34]. The results are depicted in Figure 7 A to F, and show that the proteins caused a modest, but significant reduction in the number of leptospires adhering to Fg (*P<0.05) with 0.5, 1.0, 1.0, 0.1, 0.25 and 1.0 µM of rLIC12238, rLIC11975, rLIC11360, Lsa30, Lsa33 and OmpL1, respectively. We have performed three independent experiments with comparable results.
Fg is the major clotting protein present in blood plasma with an important role in coagulation and thrombosis. Several studies have suggested roles of Fg and PLG in host bacterial interactions [11]. The fibrinolytic system that generates PLA from PLG and its activators, decomposes the fibrin clot by degrading fibrin [37]. This system is negatively regulated by PLG activator inhibitors. A balance between coagulation system and anticoagulation is necessary to avoid pathophysiological conditions, such as, bleeding and thrombosis [38].
Several Fg-binding proteins of important pathogens, such as, Staphylococcus aureus [39], group B Streptococcus [40], Lactobacillus salivarius [41] and the spirochetes Treponema pallidum [42], T. denticola [43], [44], [45] have been reported and characterized. These Fg-binding proteins are bacterial surface or secreted adhesins that although acting through different mechanisms will ultimately lead to enhance the bacterial survival in the host [46].
The adhesion of virulent L. interrogans strain Fiocruz L1-130 to Fg has been shown to be induced by physiological osmolarity, but the effect of this binding on fibrin clot formation was not evaluated [14]. The leptospiral proteins, LigA and LigB [16], [17] and OmpL37 [15] have been described as Fg-binding proteins. In addition, the interaction of LigB with Fg was shown to inhibit thrombin-catalyzed fibrin formation [16], [17].
In this work, we show that L. interrogans serovar Copenhageni are capable of binding human Fg on their surface, as visualized by indirect IFA. Moreover, we have employed ELISA to assess the interaction of human Fg with virulent L. interrogans serovar Copenhageni Fiocruz L1-130, attenuated L. interrogans serovar Copenhageni M-20 and non-pathogenic saprophytic L. biflexa serovar Patoc Patoc 1, and the effect of this binding on the thrombin-catalyzed fibrin formation. Moreover, we studied seven leptospiral adhesin- and PLG-binding proteins, rLIC12238 [18], Lsa33, Lsa25 [19], Lsa30 [20], OmpL1 [21], rLIC11360 and rLIC11975 [22] for their capacity to bind Fg and to inhibit fibrin clot formation. We show that all strains tested bind Fg, including the saprophytic one, and that this interaction inhibits fibrin clot formation. Though the virulent strain appears to be more efficient, the values obtained were not statistically significant. Attachment of recombinant proteins to Fg was specific, dose-dependent and saturable with all proteins but Lsa30. The interaction of the proteins with Fg, similar to the Leptospira, promoted an inhibition on thrombin-induced fibrin clot formation. Thus, Leptospira, similar to other pathogens, express multiple Fg-binding proteins [38], [46].
As expected, recombinant proteins partially inhibited attachment of intact L. interrogans to immobilized Fg. The inhibitory effect exerted by the recombinant proteins was moderate, ranging from 0.1 to 1.0 µM of protein concentration to reach significance, and could be explained by the existence of additional L. interrogans binding proteins contributing to the leptospiral adherence to Fg. A correlation between inhibition and affinity was detected with the recombinant protein Lsa33, while no correspondence was seen with the others.
The inhibitory effect on fibrin formation observed with leptospires and with the six recombinant proteins was partial, similar to the LigB, fragment 7–12, employed in this work as positive control, and previously reported by Choy et al., 2011 [16], using LigB fragment 9–11. Our data differ with the total clotting inhibition promoted by the Fg-binding proteins, SdrG of S. epidermidis [47], ClfA of S. aureus [48] and LigBCen2R [17]. The data suggest that in Leptospira other mechanisms might be involved in fibrin formation and/or the main function of these Fg- binding proteins is not associated with the clotting.
We have described the interaction of Leptospira with fibrinolytic system and shown that it occurs with virulent, attenuated and saprophyte strains of Leptospira. Moreover, we demonstrated that this association renders the bacteria with proteolytic activity capable of degrading ECM components [12]. Several membrane proteins were identified as PLG-binding receptors capable of generating PLA in the presence of activator, suggesting that the interaction with the fibrinolytic system might be important during leptospirosis [49], [18], [50], [36], [19], [21], [20]. Indeed, the increased plasma levels of Fg degradation products detected in leptospirosis has provided evidence for fibrinolysis activity [51], [52]. We show now that Leptospira surface-associated PLA activity is capable to degrade Fg in vitro, suggesting one possible pathway to generate Fg metabolites during the disease.
Fg is considerably upregulated during inflammation or under exposure to stress such systemic infections [46]. The activation of coagulation cascade with increased levels of plasma Fg during leptospirosis has been detected [53], [51], [54]. It has been suggested that these findings are possibly associated to severe tissue damage, vascular endothelial injury or a compensating production by the liver in response to the augmented Fg utilization [55], [53], [51]. Our data show that Leptospira either through their Fg-binding proteins or coated with PLA activity would increase the consumption of Fg molecules by sequestering or degrading them. Under these circumstances, a reduction on fibrin clot formation is expected. In addition, leptospirosis patients with clinical bleeding were reported to have lower platelet counts when compared to other patients [54], a condition that would help decrease thrombosis, facilitate bleeding and help bacterial dissemination.
In conclusion, we have provided molecular evidence of the mechanisms that Leptospira could employ to interact with components of the coagulation cascade and the fibrinolytic system. In addition, we have shown that six adhesins could mediate the binding of Leptospira to Fg and impair thrombin- induced fibrin clot formation. We believe that our results should contribute to the understanding of the complex coagulopathy observed during leptospirosis.
|
10.1371/journal.pgen.1005838 | Continuous Influx of Genetic Material from Host to Virus Populations | Many genes of large double-stranded DNA viruses have a cellular origin, suggesting that host-to-virus horizontal transfer (HT) of DNA is recurrent. Yet, the frequency of these transfers has never been assessed in viral populations. Here we used ultra-deep DNA sequencing of 21 baculovirus populations extracted from two moth species to show that a large diversity of moth DNA sequences (n = 86) can integrate into viral genomes during the course of a viral infection. The majority of the 86 different moth DNA sequences are transposable elements (TEs, n = 69) belonging to 10 superfamilies of DNA transposons and three superfamilies of retrotransposons. The remaining 17 sequences are moth sequences of unknown nature. In addition to bona fide DNA transposition, we uncover microhomology-mediated recombination as a mechanism explaining integration of moth sequences into viral genomes. Many sequences integrated multiple times at multiple positions along the viral genome. We detected a total of 27,504 insertions of moth sequences in the 21 viral populations and we calculate that on average, 4.8% of viruses harbor at least one moth sequence in these populations. Despite this substantial proportion, no insertion of moth DNA was maintained in any viral population after 10 successive infection cycles. Hence, there is a constant turnover of host DNA inserted into viral genomes each time the virus infects a moth. Finally, we found that at least 21 of the moth TEs integrated into viral genomes underwent repeated horizontal transfers between various insect species, including some lepidopterans susceptible to baculoviruses. Our results identify host DNA influx as a potent source of genetic diversity in viral populations. They also support a role for baculoviruses as vectors of DNA HT between insects, and call for an evaluation of possible gene or TE spread when using viruses as biopesticides or gene delivery vectors.
| While gene exchange is known to occur between viruses and their hosts, this phenomenon has never been studied at the level of the viral population. Here we report that each time a virus from the Baculoviridae family infects a moth, a large number (dozens to hundreds) and high diversity of moth DNA sequences (86 different sequences) can integrate into replicating viral genomes. These findings show that viral populations carry a measurable load of host DNA sequences, further supporting the role of viruses as vectors of horizontal transfer of DNA between insect species. The potential uncontrolled gene spread associated with the use of viruses produced in insect cells as gene delivery vectors and/or biopesticides should therefore be evaluated.
| The genomes of large eukaryotic double-stranded DNA viruses contain high proportions of cellular genes resulting from host-to-virus horizontal transfers (HT) [1–4]. For example, at least 10% of giant virus genes and up to 30% of herpesvirus genes likely originated from eukaryote or prokaryote genomes [1, 5, 6]. Some of these genes have been shown to act as key factors in the etiology of viral diseases [7, 8]. Because cellular gene content can be quite different between closely related viruses and/or quite similar between distantly related viruses [1, 2, 9], viral co-option of host genes appears to be rather frequent during virus evolution. The cellular genes that have so far been identified in viral genomes result from relatively ancient host-to-virus HT events. From a population genetics perspective, these viral-borne host genes must have been inherited at low to intermediate frequencies over multiple rounds of viral replication until they finally reached fixation in the viral species, likely because they provided a fitness gain to the virus. In agreement with this hypothesis, many of these genes are thought to play a role in thwarting host anti-viral defenses, thus facilitating viral replication [10]. A corollary of this scenario is that many viral-borne host genes resulting from host-to-virus HT should be found at varying frequencies in viral populations. However, host-to-virus HT has never been investigated at the micro-evolutionary scale of the viral population. Therefore, the frequency of host-to-virus HT as well as the evolutionary and molecular processes underlying the capture and domestication of eukaryotic genes by viruses remain poorly understood.
Another outstanding question arising from host-to-virus HT is whether viral-borne host genes acquired from a given host individual can be transferred to the genome of another infected individual through virus-to-host HT. In other words, can viruses act as vectors of HT between their hosts? Hundreds of HT cases have been characterized in eukaryotes [11, 12]. Many of these transfers have generated evolutionary novelties and allowed receiving organisms to adapt to new environments [13–15]. Horizontal transfer of DNA is therefore increasingly appreciated as an important evolutionary force shaping eukaryote genomes. However, the mechanisms and the potential vectors involved in HT of DNA between eukaryotes remain poorly known, especially in multicellular eukaryotes. Viruses have been proposed as candidate vectors facilitating HT between eukaryotes because they are transmitted horizontally (and in some cases vertically) and they replicate inside host cells [16, 17]. In metazoans, the vast majority of HTs characterized so far are transfers of transposable elements (TEs), which constitute pieces of DNA that are capable of moving from one genomic locus to another, often duplicating themselves in the process [18]. Several studies have uncovered host TEs packaged in viral capsids or even integrated into viral genomes, suggesting that TEs can jump from host to virus during the course of a viral infection [19–24]. We discovered two such TEs from the cabbage looper moth (Trichoplusia ni) integrated at low frequency in genome populations of the baculovirus Autographa californica multiple nucleopolyhedrovirus (AcMNPV) following infection of T. ni caterpillars [23]. Importantly, these two TEs show signs of HT between several sympatric moth species that can be infected by baculoviruses in the wild. The search for T. ni sequences integrated into populations of AcMNPV was however restricted to the dozen of T. ni genes and transposable elements that were known at the time. Therefore, the number and diversity of moth TEs and non-TEs that become integrated into AcMNPV genomes during the course of an infection remains poorly characterized.
Here we report a comprehensive search for host sequences integrated in 21 genome populations of the baculovirus AcMNPV (Baculoviridae) following infection of caterpillars from two moth species. The Baculoviridae comprise large, circular dsDNA viruses infecting mainly Lepidoptera but also Hymenoptera and Diptera [25]. Most baculoviruses are transmitted as occlusion bodies (OBs), i.e. the virions are protected in a protein matrix allowing the virus to remain infectious in the environment for extended periods of time [26]. AcMNPV is a multiple nucleopolyhedrovirus, meaning that each OB typically contains dozens of virions, each enclosing multiple genomes individually packaged within nucleocapsids. This morphology allows the virus to initiate infection as a highly polymorphic population [27], and can foster the maintenance of deleterious genotypes [28]. Rather untypical for a baculovirus, AcMNPV is a generalist virus, able to infect moth species belonging to nine lepidopteran families [29]. The two moth species we used are Trichoplusia ni (Plusiinae) and the beat armyworm Spodoptera exigua (Noctuinae), which belong to the Noctuidae family and are known to be highly susceptible to AcMNPV. These agricultural pests are found in many regions of the world, and can occur in sympatry [30]. We performed in vivo experimental infections of both T. ni and S. exigua caterpillars to generate AcMNPV populations for deep sequencing. Our population genomics approach yields the first estimate of the frequency and spectrum of host sequences that can become integrated in the genome of a large dsDNA virus.
The first AcMNPV genomic dataset we analyzed was generated by sequencing the 134-kb AcMNPV genome at 187,536X average depth after in vivo amplification of the virus in T. ni larvae (G0 in S1 Fig). The viral population that produced this dataset was then independently passaged in ten lines of T. ni larvae and ten lines of S. exigua larvae, each line consisting of ten successive infection cycles (G10 in S1 Fig). OBs recovered from the last infection cycle of each line were sequenced at between 9,211X and 33,783X average depth for the ten T. ni lines (total depth = 145,386X) and between 3,497X and 35,434X average depth for the ten S. exigua lines (total depth = 163,610X). Viral sequencing reads were used as queries to perform Blastn searches against sequences from both moth species. Host sequences included RNAseq data corresponding to 70,322 T. ni contigs and 96,675 S. exigua contigs [13,14], as well as 469 and 486 contigs from T. ni and S. exigua, respectively, that were assembled in this study using sequencing reads that did not map onto the AcMNPV genome (see methods). All viral reads aligning to moth contigs were then used as queries for Blastn searches against the AcMNPV consensus genome [9] to identify chimeric reads (i.e. sequences containing both AcMNPV and moth DNA), as evidence of junctions between host and viral DNA. After applying various filters to eliminate false positives, we extracted a total of 27,504 chimeric reads from all 21 AcMNPV genomic datasets. Chimeric reads were identified in the initial AcMNPV population from T. ni (n = 9,464), as well as in all ten T. ni lines (n = 460 to 1,904; total = 12,219) and all ten S. exigua lines (n = 41 to 1,684; total = 5,821) (S1 Table; S1 Dataset).
The 27,504 host-virus DNA junctions involved 38 T. ni and 48 S. exigua contigs (S1 Table; S2 Dataset). Similarity searches and structure analyses revealed that 69 of these contigs are TEs (29 in T. ni and 40 in S. exigua) belonging to both major groups of eukaryote TEs (Table 1; S1 Table): retrotransposons (three superfamilies) and DNA transposons (10 superfamilies). The remaining 17 contigs did not show any sequence or structural similarity to any known TE or protein. However, their closest Blastn hits in the GenBank whole-genome sequence database were found in Lepidoptera, suggesting these 17 contigs indeed originate from the host genomes.
The large proportion of TEs among the host contigs found to be joined to viral DNA may indicate that transposition is the main mechanism of insertion of host DNA into viral genomes. Alternatively, junctions between host TEs and viral DNA could result from technical artifact leading to chimeric reads composed of viral and contaminating host sequences. Though we verified that the amount of contaminating host DNA was very low (if at all present) in all our samples (see ref [23] and S1 Text), we cannot totally exclude the presence of such contamination. If our samples were contaminated, and given that TEs make up the single largest fraction of eukaryote genomes [31], technical chimeras involving mainly host TEs might not be unexpected. However, the 27,504 chimeric reads correspond to 7,049 different junctions, as defined by their location in the viral genome and in host contigs. Indeed, 1,412 of these 7,049 unique junctions are covered by more than one read (two to 1,256 reads; S2 Fig). This strongly suggests the junctions we observe do not result from any kind of technical artifact. As duplicates generated by PCR during library construction were removed (S1 Text), it seems very unlikely that a technical error would generate multiple chimeras involving exactly the same virus and host DNA sequences at the same positions. Only amplification of junctions through in vivo viral replication provides a plausible explanation for these observations, ruling out the possibility that these junctions result from technical chimeras. To further assess the biological origin of host-virus DNA junctions, we sought to characterize the molecular mechanisms involved in the integration of host sequences in AcMNPV genomes.
For each host contig integrated into several distinct viral sites, we located the target insertion sites both in the viral genome and in the host sequence, and examined sequence patterns in their vicinity. We found that most inserted sequences were DNA transposons, for which the junctions with the virus genome clustered immediately before the 5’ terminal inverted repeat (TIR) or immediately after the 3’ TIR, as expected in the case of transposition events (Fig 1; S3 and S7 Figs; S3 Dataset). In addition, the integration sites in the viral genome were generally characterized by short (1–5 bp) highly conserved sequence motifs (Fig 1; S4 Fig) corresponding to known TE preferred insertion sites (e.g., TTAA for the Piggybac family, TAA for harbinger, CGNCG for transib). These patterns corroborate earlier findings [19–21, 23] and indicate that many DNA transposons are indeed able to integrate into viral genomes during the course of an infection through bona fide transposition. Overall, we identified 19,899 host-virus junctions resulting from transposition. Counting only once all host-virus junctions covered by more than one read yields a minimum of 6,579 junctions resulting from independent transposition events, out of 7,049. The mechanism underlying the vast majority of the host-to-virus HT detected in this study is therefore transposition.
Among the remaining insertions, 434 unique host-virus junctions were deemed highly unlikely to result from transposition. Contrary to the host-virus junctions resulting from transposition, which were all located at the extremities of the 5’ and 3’ TIRs, these junctions were scattered within the host sequences. A short sequence motif of 1 to 20 bp, identical between the insertion site and the host sequence, characterized 298 of these junctions (Fig 2). The length of these microhomology motifs is significantly longer than expected by chance (Khi2 = 4,523; p < 10−15, 20 d.f.) and argues against technical artifact as the main cause of these junctions (as technical error favoring microhomology are highly unlikely). We also note that 15 of the 434 junctions are covered by more than one read (two to 27 reads), indicating that some of these junctions were amplified through viral replication. The 158 remaining non-transposition junctions lacking microhomology could either have resulted from the ligation of blunt-ended sequences (n = 87), or were characterized by the presence of 1 to 2 nucleotides that apparently did not originate from either the host or viral genomes (n = 71, corresponding to negative microhomology lengths in Fig 2). The distribution of microhomology lengths suggests that in addition to transposition, host DNA can be integrated into viral genomes via a variety of recombination events, some of which (but not all) rely on microhomology motifs between virus and host DNA sequences. Whether such recombination events are mediated by viral factors (e.g. LEF-3, AN, PCNA [32, 33]), or by host-encoded DNA repair mechanisms [34] known to enhance baculovirus amplification [35], would be worth addressing at the functional level in the future. Most of the non-transposition junctions lie within DNA transposon sequences, which may appear intriguing. We speculate that on principle, any region of the host genome could be joined to viral DNA through microhomology-mediated recombination (including genes that may turn to be beneficial to the virus), provided that it contains a double stranded break. Yet, since DNA transposons have the capacity to excise themselves from the host genome, they may be among the most numerous extra chromosomal DNA fragments containing double stranded breaks ready to recombine with broken viral DNA.
We then mapped the independent insertions of host sequences along the AcMNPV genomes, that is, counting only once all insertions possibly resulting from amplification through viral replication. The map (Fig 3A; S5 Fig) shows that integrations occur virtually everywhere in the viral genome and that all 151 viral genes are disrupted by host insertions at least once. Remarkably, the local densities of insertions of S. exigua TEs strongly correlate to those of T. ni TEs (Fig 3B and S2 and S3 Tables; 54% of variance explained; p < 10−15). This correlation is not explained by variation in sequencing depth or density of the preferred transposition motifs (identified above) along the viral genome (S2 and S3 Tables). Two other causes may explain this correlation: (1) varying degrees of tolerance to insertions along the viral genome and (2) varying rates of transposition along the viral genome. Cause (1) implies that insertions are more likely to be replicated through viral replication in some regions than others because their impact on viral fitness would be lower. In this case, the correlation between local densities of insertions along the viral genome should be maintained or even higher when considering all insertions that possibly result from viral replication. However, the correlation almost disappears (0.1% of variance explained, S3 Table) when all insertions (including potentially replicated ones) were considered. Hence, although the fitness impact of insertions may well vary along the viral genome, these variations cannot explain the strong correlation we initially observed. This leaves cause (2) as the only explanation for the correlation of insertion frequencies from TEs of the two species. In other words, TEs from the two moths, in spite of being different (Table 1 and S1 Table), tend to transpose preferentially into the same regions of the AcMNPV genome irrespective of the density of target sites. This suggests that the pattern of integration may be shaped by structural properties of the viral genome. We propose that, as observed in eukaryotic genomes [36–38], the distribution of transposition-mediated integrations along the AcMNPV genome may be influenced by accessibility of the viral genome to host TEs, which itself likely depends on the structure of AcMNPV chromatin, known to be dynamically remodeled during viral replication [39].
Taking into account the number of chimeric reads per library, the total number of reads in each library, the size of the AcMNPV genome and the minimum alignment size that can be returned by Blastn, we calculated that on average 4.8% viral genomes (range 1.1% to 14.3%) carry at least one host sequence among AcMNPV populations (Table 1). Though the number of host-to-virus HT events that generated these frequencies is likely to be high, we cannot infer it precisely because we cannot tell how many of the host-junctions covered by more than one read were amplified through viral replication. Indeed, the same TE may insert several times at any suitable viral site. Furthermore, it is possible that a number of host-virus junctions, which cannot be evaluated here, were generated through subsequent transposition of viral-borne TEs (not coming from the host genome) into multiple copies of the virus genome.
The relatively high frequencies of AcMNPV genomes carrying host DNA fragments at any given time raise the question of whether such host sequences are inherited over viral infection cycles. We thus tested the residual presence of T. ni sequences (inserted at G0 in S1 Fig) in viral populations, that had subsequently been passaged 10 times in S. exigua (G10 datasets in S1 Fig). Using the libraries from S. exigua G10 viruses as Blastn queries against T. ni contigs revealed 24 new insertions that were not previously found in Blastn searches against S. exigua contigs (see S1 Text for details). These insertions likely involve S. exigua sequences homologous to T. ni but absent from the S. exigua contigs. None of these insertions were identical (in terms of position in the virus genome and host contig) to any found in the G0 virus population. The persistence of a given host DNA fragment in virus populations thus appears to be low, likely because of the deleterious effects large insertions have on the viral genome carrier. Although many new host sequences become integrated into AcMNPV genomes at each viral infection cycle, they are thus purified out of the viral population after only few infection cycles. Hence there is a high turnover of host sequences inserted into the viral genome each time the virus replicates in a host. Under natural settings, continuous host-to-virus flow of genetic material generates a significant proportion of recombinant viruses (Table 1). At the viral population scale, this represents a gene reservoir that could fuel host-virus coevolutionary arms race through co-option of a host sequence favoring the virus in a given environment. Our findings thus shed light on the first evolutionary steps underlying viral co-option of cellular genes.
Under the hypothesis that viruses can act as vectors of HT of TEs, once inserted in a viral genome, viral-borne TEs should then be able to jump from the viral genome to the genome of a new host organism. To evaluate the possibility that AcMNPV can shuttle TEs between insects, we first checked whether some TEs found integrated in our AcMNPV genome datasets have retained the structural features necessary for transposition. Among the 41 contigs inserted in at least 10 different viral sites, 11 correspond to TEs for which we recovered both TIRs and that encode a putative full-length intact transposase gene (Fig 1; S3 Fig). Provided that these TEs can be transcribed in a new host, they should thus be able to jump from the viral genome into the genome of this new host. However, it is important to note that the transfer would only be effective if the host survived viral infection in the first place. This is more likely to happen if the host shows resistance to the virus or if the virus harbors deleterious mutations.
We then reasoned that if AcMNPV is able to act as vector of HT of TEs between its insect hosts, the TEs we found integrated in the AcMNPV genome populations might have been horizontally transferred relatively recently between insects of various susceptibility to AcMNPV. To test this hypothesis, we assessed whether some of the TEs uncovered in this study have been horizontally transferred between T. ni and/or S. exigua and other insect lineages. We used the 69 TE sequences as queries to perform Blastn searches against the 144 non-Noctuidae insect genomes available in GenBank as of March 2015. For 21 of these TEs (14 S. exigua and seven T. ni TEs), we found highly similar copies (>85% nucleotide identity) in the genome of one or more other insects (Fig 4). The 21 TEs show a combination of features that are typically indicative of HT [40–42]. They have a patchy distribution in the insect phylogeny and, importantly, the between-species nucleotide identity calculated for each of these TEs is much higher (91% identity on average) than synonymous nucleotide identities calculated for 11 conserved genes between the same species (37% identity on average, Fig 4; S4 and S5 Tables). We conclude that at least 21 of the 69 TEs found integrated in AcMNPV have undergone one or multiple horizontal transfers between T. ni or S. exigua and one or several other insect lineages. Nucleotide identity between some T. ni or S. exigua TEs and those uncovered in other insects is very high (up to 99%), suggesting some of HT events took place very recently. Among the other insects involved in HT, we found four lepidopteran species known to be susceptible to AcMNPV infection (Fig 4) [43]. Our study therefore provides further compelling support for the role of baculoviruses as potential vectors of TEs between lepidopterans [19, 20, 23]. Indeed, given that on average 4.8% of baculovirus genomes harbor a host sequence and that a caterpillar typically becomes infected by ingesting thousands of baculovirus genomes, each non-lethal infection represents an opportunity for between-host baculovirus-mediated transfer of DNA.
In this study, we have shown that each time the baculovirus AcMNPV infects a caterpillar host, a large number of host TEs can transpose into its genome. Many TEs and other host sequences can also integrate into AcMNPV genomes through microhomology-mediated recombination events. Our work also demonstrates that the density of transposition events is not homogenous along the AcMNPV genome and that, while the influx of host sequences integrated into AcMNPV is continuous, each newly integrated host sequence is rapidly purged out of AcMNPV populations. Together, these observations are reminiscent of the well-known gene exchanges that take place between bacteria and bacteriophages [47], and indicate that such exchanges may also occur on a regular basis between eukaryotes and eukaryotic viruses. Our results also raise a number of questions worth addressing in future experiments. In particular, it would be interesting to monitor the evolution of the frequency of viral replicates carrying any given host sequence across successive infection cycles, as the host DNA sequence retention time affects the likelihood of such sequence being horizontally transferred between hosts. Furthermore, since the rate of host-to-virus HT is measurable at the population level, it would be worth investigating whether this phenomenon influences the within-host replication dynamics of AcMNPV. Another exciting question is whether this phenomenon is limited to moth-AcMNPV interactions or whether it also takes place in other host-virus systems. Finally, it is noteworthy that AcMNPV and other baculoviruses are used as biopesticides and developed as vectors for several biomedical applications such as gene or vaccine delivery [48, 49]. Our results therefore call for an evaluation of the risk of gene or TE spread through uncontrolled virus-mediated HT potentially generated by these approaches, which rely on mass production of the viruses in insect cells or in vivo.
The GenBank accession numbers of the 21 AcMNPV genomic datasets analyzed in this study are: SRS533250, SRS534469, SRS534534, SRS534575, SRS534677, SRS534587, SRS534590, SRS534631, SRS534673, SRS536572, SRS536571 and SRS534470, SRS534499, SRS534514, SRS534536, SRS534537, SRS534543, SRS534542, SRS536937, SRS534588 and SRS534589 [23]. They consist of 101-bp paired sequences (reads), except for dataset SRS533250, which consists in 151-bp paired reads.
These datasets were produced through experimental evolution, which consisted in generating ten per os infection cycles on ten lines of T. ni and S. exigua larvae using 2500 occlusion bodies from an AcMNPV stock derived from a viral sample originally isolated from a single Alfalfa looper (Autographa californica) individual collected in the field. The full experiment is described in Gilbert, Chateigner [23] and in S1 Fig. The AcMNPV DNA samples used to produce the 21 sequencing datasets were all extracted using the QIAamp DNA Mini Kit (Qiagen) after purification of AcMNPV occlusion bodies by a percoll-sucrose gradient, Na2CO3 dissolution and enzymatic removal of host DNA [23].
Data analyses were performed in R [50], unless another tool is mentioned. All Blastn searches were carried out under default settings.
To identify host DNA sequences integrated in genomes of the AcMNPV baculovirus, we used viral reads as queries to perform Blastn searches on T. ni and S. exigua transcripts generated by Pascual, Jakubowska [51] and Chen, Zhong [52]. In addition, to recover as many insertions of host DNA as possible, we assembled non-viral DNA elements present in the viral genomic libraries. These elements may represent inserted host DNA sequences absent from (or incomplete in) the available transcriptomes of both moth species.
We applied the following procedure on genomic libraries obtained from each moth species. We aligned all reads on the AcMNPV genome using the end-to-end mapping strategy of Bowtie 2 [53]. We used Samtools view on resulting alignment files to extract read pairs for which at least one read did not align. These unmapped reads were trimmed off low quality score bases with Trimmomatic [54], and assembled with SOAP deNovo 2 [55] using a kmer length of 71 bases, which showed good assembly statistics compared to other lengths.
We checked assembly quality by performing Blastn homology searches of assembled contigs against themselves, and found that many contigs differed only at one or both of their ends but were otherwise identical. Blastn searches of the contigs against the AcMNPV genome revealed that the contig ends that differed were similar to parts of the viral genome. We assumed that mostly similar contigs resulted from a genetic element inserted at different sites of the AcMNPV genome, and that these viral sites had been partly included into contig ends during the assembly process. We thus trimmed contigs from these viral regions, and reassembled them using the assembly feature included in Geneious 4.5 [56], allowing a maximum mismatch of 10% in overlapping regions. This yielded 469 contigs for genomic libraries generated from T. ni lines and 486 contigs for S. exigua lines. These contig sequences were added to known transcriptome sequence of the corresponding moth species [51, 52], which we hereafter simply refer to as “transcripts”, in order to constitute the host databases for the Blastn searches designed to identify junction between moth and viral DNA.
Blastn searches with default parameters [57] were carried out using the 21 AcMNPV genomic datasets as queries to identify similarities of at least 28 nucleotides (as defined by the default settings) between reads obtained from each viral line and the sequence database corresponding to its host. Each read showing similarities was then blasted against the AcMNPV reference genome, together with the other read of its pair (mate) so as to detect junctions occurring between paired reads.
For a given read listed in a blast output, we retained the alignment with the best score, randomly choosing between alignments of identical scores. This selection was done separately for alignments with transcripts and for alignments with contigs, in order to help selecting between homologous contigs and transcripts (see below). For a read to be considered as a junction between host and virus DNA, we imposed minimum lengths of alignment with the virus genome only, and with the host genome only, of 16 bp each (S6 Fig). Furthermore, at least 95 nucleotides of the read had to align with virus and host sequences (130 bp for 151-bp reads). The overlap between these alignments was set to involve at most 20 bp and at least -2 bp. These filters excluded reads from virus regions having similarities with host contigs, in which case the region aligning to a host contig would be included in that aligning to the virus genome.
To detect junctions that occurred between two paired reads (mates), we selected read pairs meeting the following conditions: (i) one mate must have a similarity with the virus genome of at least 95 bp (130 bp for 151-bp reads) and present no similarity with any host contig, and (ii) the other mate must have a similarity to a host contig of at least 95 bp (130 bp for 151-bp reads), and present no similarity with the virus genome.
We found that the sensitivity of our approach to detect junctions between host and virus DNA was highly dependent on the quality of the assembly of the host sequences that were used as reference. Thus, it will be important that future studies dedicate substantial effort to generate a high quality and comprehensive set of host sequences in order to find all possible host-virus chimeras.
We discarded all alignments (junctions) involving contigs or transcripts not meeting the following criteria. To ensure that a contig we assembled represented moth DNA, it had to be partly similar to a transcript of the corresponding host species, as determined by Blastn searches of contigs against host transcripts, or to align with at least one read of a pair that also aligned with a known host transcript, as determined by the Blastn search of reads against the contig and transcript databases.
Because the contigs we assembled are, as expected, partly similar to some host transcripts, a chimeric read may have similarities to a contig and to a transcript (after selecting the best alignment in each category, as explained above). In other words, contigs and transcripts can be candidates for the same insertions. To minimize redundancy between contigs and transcripts, any transcript sharing at least one chimeric read with a contig was discarded. We therefore retained transcripts that did not share any junction with any assembled contig. Finally, we discarded host contigs or transcripts having similarities with less than three chimeric reads (i.e., potentially inserted in the AcMNPV genome less than three times) or having a cumulative alignment length of less than 75 bp with chimeric reads.
For junction counts shown in S1 Table, we removed duplicates that may have resulted from PCR amplification of the same junction during library preparation (S1 Text). Junctions sequenced in both directions and appearing in two overlapping paired reads were counted only once.
Defining Pj as the average number of junctions per virus genome involved in the construction of a genomic library, the probability for a read from that library to cover a junction between a host sequence and the viral genome can be approximated as Pj × Lr/Lg, where Lr is the read length and Lg is the length of the virus genome. For a read to be chimeric under our criteria, a junction has to be at least 28 bp away from the read ends (S6 Fig). However, the overlap between alignments at the junction (S6 Fig), the mean length of which we denote as Ov, allows the junction to be slightly closer from the read end and to yield a 28-bp region of sequence similarity detectable by blast, so that the probability for a read to be chimeric is Pj ×Lr−56+OvLg.
Since this probability can be approximated as the ratio of the number of chimeric reads Nc over the number of viral reads N of a sequence library, we obtain Pj≅NcN × LgLr−56+Ov.
N was estimated by running samtools view [58] on alignment files obtained by mapping reads from each virus line on the AcMNPV consensus genome, using the local sensitive settings of Bowtie 2 [53]. Nc includes technical duplicates (PCR duplicates and overlapping paired reads, see above) because they contribute to N as much as they do to Nc.
We derived the proportion of virus genomes carrying at least one host DNA fragment by assuming that the number of inserted fragments per virus genome follows a Poisson distribution of mean Pj/2, as one insertion of host DNA into the circular AcMNPV genome should yield 2 junctions.
For simplification, we hereafter refer to contigs assembled in this study and to previously assembled host transcripts as “contigs”.
We identified each junction producing a chimeric read by the offset it involves between the virus genome coordinates and the host contig coordinates (S1 Text). Reads having the same offset, involving the same viral DNA strand, host contig, and coming from the same genomic library in the case of S. exigua lines (which do not share a ancestor on this host) were considered likely to come from viral amplification of the same original insertion of host DNA. In the following analyses, we selected only one chimeric read per original junction, favoring the read with best alignment score on the host contig.
These reads were mapped onto their corresponding contig by inserting gaps of appropriate length before their sequence, based on alignment coordinates reported by blast, to produce multifasta alignment files. Visualization of these files in Geneious [56] and BioEdit [59] (example shown in S7 Fig) showed that junctions clustered at one or two positions in a contig likely representing the end(s) of a TE. For many contigs, this was further supported by the presence of terminal inverted repeats (TIRs), which are typical of class II DNA transposons, the presence of long terminal repeats (LTRs), which are typical of LTR retrotransposons and by similarities to known TE protein motifs returned by blastx on the GenBank non-redundant protein database.
For each cluster of at least 10 junctions, which likely represent insertions of the same TE end, we analyzed sequence conservation at insertion sites in the virus genome. This was done by computing insertion coordinates based on the previously obtained offset (S1 Text), and by building sequence conservation logos [60–62] of 30-bp sequences around insertion sites. Sequence logos are provided in S4 Fig.
Some junctions did not form clusters (according to our criteria defined in S1 Text) and were scattered along host contigs (example shown in S7 Fig), suggesting that different fragments of these contigs were inserted. This concerned 434 junctions, most of which presented similarities between host and virus sequences at insertion points (Fig 2, yielding to the overlap shown in S6 Fig). To check whether these similarities were overall longer than expected by chance, we extracted from each chimeric read resulting from this type of junction the last 20 bp that aligned to the host contig (next to the junction point), and computed the lengths of similarities this 20-bp sequence had with 20 random 20-bp regions of the AcMNPV consensus genome. This allowed comparing the distribution of expected and observed identity lengths with a Khi-square test.
We explained the number of junctions in 1500-bp windows of the AcMNPV genome, combining all virus lines from S. exigua, with a generalized linear model including three covariates: the average sequencing depth in that window, the number of common TE targets found in S. exigua lines, and the number of junctions found in virus lines from T. ni, without considering interactions between terms.
Sequencing depth was estimated by running samtools mpileup [58] on mapping files obtained previously. We modified mpileup to allow greater depth than 8000. We counted the following frequent targets of S. exigua TEs, based on the logos we established previously (S2 Fig): “TTAA” (for piggybac TEs), “TTA”, “TAA” (for Harbinger TEs), and “TA” (for Mariner TEs).
A Poisson distribution was assumed for the number of junctions per genome window. We selected the best model on the basis of corrected Akaike Information Criterion (AICc) returned by the dredge function of the R package MuMIn [63] (S2 Table), and we submitted it to an analysis of deviance (S3 Table). We fitted this model twice: considering all junctions (including viral replicates), and only independent junctions based on their identifiers. In the latter case, the most likely model only included the number of junctions per window in T. ni as a covariate (S2 Table).
Sequencing depth of host contigs (Fig 1; S3 Fig) was computed by using alignments coordinates from results of Blastn search of reads against host databases (see above), using the same criteria to select a single alignment for reads having similarities with several contigs/transcripts. Depth was averaged over 20-bp sliding windows overlapping by 10 bp.
We assessed whether T. ni and S. exigua TEs found integrated in the AcMNPV genome underwent HT between insects. We used the T. ni and S. exigua TEs we identified as queries to perform Blastn searches against the 144 non-Noctuidae insect whole genome sequences available in GenBank as of March 17th 2015. We identified candidate HT events when a T. ni or S. exigua TE aligned to a sequence from another insect genome with at least 85% nucleotide similarity over at least 100 bp. To assess the level of neutral genetic distance expected under vertical inheritance between T. ni/S. exigua and all insect species in which we found Blastn hits meeting the above criteria, we calculated synonymous distances for 11 conserved genes between T. ni/S. exigua and those insect species using the non-corrected Nei-Gojobori method in MEGA 6 [64], following Gilbert, Chateigner (23). Overall we calculated 143 pairwise synonymous gene distances between S. exigua and 13 other insect species and 55 pairwise synonymous gene distances between T. ni and five other insect species.
|
10.1371/journal.pcbi.1002059 | How Structure Determines Correlations in Neuronal Networks | Networks are becoming a ubiquitous metaphor for the understanding of complex biological systems, spanning the range between molecular signalling pathways, neural networks in the brain, and interacting species in a food web. In many models, we face an intricate interplay between the topology of the network and the dynamics of the system, which is generally very hard to disentangle. A dynamical feature that has been subject of intense research in various fields are correlations between the noisy activity of nodes in a network. We consider a class of systems, where discrete signals are sent along the links of the network. Such systems are of particular relevance in neuroscience, because they provide models for networks of neurons that use action potentials for communication. We study correlations in dynamic networks with arbitrary topology, assuming linear pulse coupling. With our novel approach, we are able to understand in detail how specific structural motifs affect pairwise correlations. Based on a power series decomposition of the covariance matrix, we describe the conditions under which very indirect interactions will have a pronounced effect on correlations and population dynamics. In random networks, we find that indirect interactions may lead to a broad distribution of activation levels with low average but highly variable correlations. This phenomenon is even more pronounced in networks with distance dependent connectivity. In contrast, networks with highly connected hubs or patchy connections often exhibit strong average correlations. Our results are particularly relevant in view of new experimental techniques that enable the parallel recording of spiking activity from a large number of neurons, an appropriate interpretation of which is hampered by the currently limited understanding of structure-dynamics relations in complex networks.
| Many biological systems have been described as networks whose complex properties influence the behaviour of the system. Correlations of activity in such networks are of interest in a variety of fields, from gene-regulatory networks to neuroscience. Due to novel experimental techniques allowing the recording of the activity of many pairs of neurons and their importance with respect to the functional interpretation of spike data, spike train correlations in neural networks have recently attracted a considerable amount of attention. Although origin and function of these correlations is not known in detail, they are believed to have a fundamental influence on information processing and learning. We present a detailed explanation of how recurrent connectivity induces correlations in local neural networks and how structural features affect their size and distribution. We examine under which conditions network characteristics like distance dependent connectivity, hubs or patches markedly influence correlations and population signals.
| Analysis of networks of interacting elements has become a tool for system analysis in many areas of biology, including the study of interacting species [1], cell dynamics [2] and the brain [3]. A fundamental question is how the dynamics, and eventually the function, of the system as a whole depends on the characteristics of the underlying network. A specific aspect of dynamics that has been linked to structure are fluctuations in the activity and their correlations in noisy systems. This work deals with neuronal networks, but other examples include gene-regulatory networks [4], where noise propagating through the network leads to correlations [5], and different network structures have important influence on dynamics by providing feedback loops [6], [7].
The connection between correlations and structure is of special interest in neuroscience. First, correlations between neural spike trains are believed to play an important role in information processing [8], [9] and learning [10]. Second, the structure of neural networks, encoded by synaptic connections between neurons, is exceedingly complex. Experimental findings show that synaptic architecture is intricate and structured on a fine scale [11], [12]. Nonrandom features are induced by neuron morphology, for example distance dependent connectivity [13], [14], or specific connectivity rules depending on neuron types [15], [16]. A number of novel techniques promise to supply further details on local connectivity [17],[18]. Measured spike activity of neurons in such networks shows, despite high irregularity, significant correlations. Recent technical advances like multiple tetrode recordings [19], multielectrode arrays [20]–[22] or calcium imaging techniques [23], [24] allow the measurement of correlations between the activity of an increasingly large number of neuron pairs in vivo. This makes it possible to study the dynamics of large networks in detail.
Since recurrent connections represent a substantial part of connectivity, it has been proposed that correlations originate to a large degree in the convergence and divergence of direct connectivity and common input [8] and must therefore strongly depend on connectivity patterns [25]. Experimental studies found evidence for this thesis in a predominantly feed-forward circuit [26]. In another study, only relatively small correlations were detected [27] and weak common input effects or a mechanism of active decorrelation were postulated.
In recent theoretical work recurrent effects have been found to be an important factor in correlation dynamics and can account for decorrelation [20], [22]. Several theoretical studies have analysed the effects of correlations on neuron response [28], [29] and the transmission of correlations [30]–[34], also through several layers [35]. However, the description of the interaction of recurrent connectivity, correlations and neuron dynamics in a self-consistent theory has not been presented yet. Even in the case of networks of strongly simplified neuron models like integrate and fire or binary neurons, nonlinear effects prohibit the evaluation of effects of complex connectivity patterns.
In [36], [37] correlations in populations of neurons were studied in a linear model that accounted for recurrent feedback. With a similar model, the framework of interacting point processes developed by Hawkes [38], [39], we analyse effects of different connectivity patterns on pairwise correlations in strongly recurrent networks. Spike trains are modeled as stochastic processes with presynaptic spikes affecting postsynaptic firing rates in a linear manner. We describe a local network in a state of irregular activity, without modulations in external input. This allows the self-consistent analytical treatment of recurrent feedback and a transparent description of structural effects on pairwise correlations. One application is the disentanglement of the explicit contributions of recurrent input on correlations in spike trains in order to take into account not only effects of direct connections, but also indirect connectivity, see Figure 1.
We find that variations in synaptic topology can substantially influence correlations. We present several scenarios for characteristic network architectures, which show that different connectivity patterns affect correlations predominantly through their influence on statistics of indirect connections. An influential model for local neural populations is the random network model [40], [41], possibly with distance-dependent connectivity. In this case, the average correlations, and thereby the level of population fluctuations or noise, only depend on the average connectivity and not on the precise connectivity profile. The latter, however, influences higher order properties of the correlation distribution. This insensitivity to fine-tuning is due to the homogeneity of the connectivity of individual neurons in this type of networks. The effect has also been observed in a very recent study, where large-scale simulations were performed [42]. In networks with more complex structural elements, like hubs or patches, however, we find that also average correlations depend on details of the connectivity pattern.
Part of this work has been published in abstract form [43].
In order to study correlations in networks of spiking neurons with arbitrary connectivity we use the theory derived in [38], which we refer to as Hawkes model, for the calculation of stationary rates and correlations in networks of linearly interacting point processes. We only summarise the definitions and equations needed in the specific context here. A mathematically more rigorous description can be found in [38] and detailed applications in [44], [45].
We will use capital letters for matrices and lower case letters for matrix entries, for example . Vectors will not be marked explicitly, but their nature should be clear from the context. Fourier transformed quantities, discrete or continuous, will be denoted by , for example . Used symbols are summarised in Table 1.
Our networks consists of neurons with excitatory and inhibitory neurons. Spike trains of neurons are modeled as realisations of Poisson processes with time-dependent rates . We have(1)
where denotes the mathematical expectation, in this case across spike train realisations. Neurons thus fire randomly with a fluctuating rate which depends on presynaptic input. For the population of neurons we use the spike train vector and the rate vector . Spikes of neuron influence the rate of a connected neuron by inducing a transient rate change with a time course described by the interaction kernel , which can in principle be different for all connections. For the sake of simplicity we use the same interaction kernels for all neurons of a subpopulation. The rate change due to a spike of an excitatory presynaptic neuron is described by and of an inhibitory neuron by . The total excitatory synaptic weight can then be defined as and the inhibitory weight accordingly as . Connections between neurons are chosen randomly under varying restrictions, as explained in the following sections. For unconnected neurons . The evolution of the rate vector is governed by the matrix equation(2)
The effect of presynaptic spikes at time on postsynaptic rates is given by the interaction kernels in the matrix and depends on the elapsed time . Due to the linearity of the convolution, effects of individual spikes are superimposed linearly. The constant spike probability can be interpreted as constant external drive. We require all interactions to respect causality, that is for . The Hawkes model was originally defined for positive interaction kernels. Inhibitory kernels can lead to negative values of at certain times, so strictly one should use the rectified variable as a basis for spike generation. We assume further on that becomes negative only rarely and ignore the non-linearity introduced by this rectification. The effects of this approximation are illustrated in Figure 2. In the equilibrium state, where the expectation value for the rates does not depend on time, we then have
where we denoted the expectation of the fluctuating rates by for notational simplicity. An explicit expression for the equilibrium average rates is(4)
where refers to the identity matrix.
We describe correlations between spike trains by the covariance density matrix . For point processes it is formally defined as the inverse Fourier transform of the spike cross-spectrum, but can in analogy to the case for discrete time be written as(5)
and corresponds to the probability of finding a spike after a time lag , given that a spike happened at time , multiplied by the rate. The term represents chance correlations such that for uncorrelated spike trains for . Due to the point process nature of spike trains, autocovariance densities have a discontinuous contribution . This discontinuity is separated explicitly from the continuous part using the diagonal rate matrix with the constant elements (here denotes the Kronecker delta). For independent spike trains so that one recovers the autocorrelation density function of Poisson processes, . A self-consistent equation that determines the covariance density matrix is(6)
for . A key result in [38] is that, if the Fourier transform of the kernel matrix(7)
is known, (6) can be solved and the Fourier transform of the cross covariance density is given by(8)
The definition of the Fourier transform implies that and accordingly , where we introduced the shortcuts and for the integrated covariance density matrix and kernel matrix, respectively. They are, from (8), related by(9)
The rate Equation (4) becomes with these definitions(10)
Equation (8) describes the time-dependent correlation functions of an ensemble of linearly interacting units. In this work we concentrate on purely structure-related phenomena under stationary conditions. Therefore we focus on the integrated covariance densities, which are described by Equation (9). Differences in the shape of the interaction kernels which do not alter the integral do not affect our results. One example is the effect of delays, which only shift interaction kernels in time. Furthermore we restrict ourselves to systems where all eigenvalues of satisfy . This condition guarantees the existence of the matrix inverse in (9) and (10). Moreover, if the real part for any , no stable equilibrium exists and network activity can explode. For further details see Section 1 of Supporting Text S1.
The matrix elements and have an intuitive interpretation. The integrated impulse response corresponds to the average number of additional spikes in neuron caused by an extra spike in neuron .
The integrated cross-correlations , in the following simply denoted as correlations, equal, for asymptotically large counting windows , the covariances of spike counts and between spike trains and ,(11)
see for example [20], [46]. On the population level one finds for the population count variance normalised by the bin size, , that(12)
Strictly this is only true in the limit of infinitely large bin size. However, the approximation is good for counting windows that are large with respect to the temporal width of the interaction kernel. In this sense, the sum of the correlations is a measure for the fluctuations of population activity. Another measure for correlations that is widely used is the correlation coefficient, . In this context it is not convenient, as the normalisation over the count variance destroys the simple relation to the population fluctuations. Even worse, as count variances are, just as covariances, influenced by network structure, for example global synchrony is not captured by this measure.
We simulated networks of linearly interacting point processes in order to illustrate the theory, Figure 2. In this network connections between all nodes are realised with constant probability . Parameters were chosen such that net recurrent input is inhibitory. The full connectivity matrix was used for the rate and correlation predictions in Equations (9) and (10) and the population count variance, Equation (12). Further simulation details are given below. This figure demonstrates that the approximation that fluctuating rates stay largely above zero gives good results even in effectively inhibitory networks with strong synapses. There are nonetheless slight deviations between prediction and simulation. On the one hand, fluctuations of the variable around a positive mean can reach below zero. This factor is especially relevant if rate fluctuations are high, for example because of strong synapses and low mean input. On the other hand, strongly inhibitory input can result into a negative mean value of for some neurons. This can happen only for wide rate distributions and strong inhibition, since the ensemble average of is always positive. In Figure 2C it is shown that only few neurons have predicted rates below zero, and that deviations between predicted and simulated rate distributions are significant primarily for low rates. The correlations in panel D are hardly affected. We found that for a wide range of parameters Hawkes' theory returns correct results for most of the rates and correlations even in effectively inhibitory networks.
Simulations of linearly interacting point processes were conducted using the NEST simulator [47]. Spikes of each neuron were generated with a rate corresponding to the current value of the intrinsic variable . Negative values of were permitted, but resulted in no spike output. Neurons received external drive corresponding to a constant rate of . Incoming spikes resulted in an increase/decrease of of amplitude for excitatory/inhibitory spikes, which decayed with a time constant of . This corresponds to exponential interaction kernels with total weights and . Synaptic delay was . Simulation time step was for the correlation and rate measurement and for spikes shown in the raster plot. In Figure 2 total simulation time was . Data from an initial period of was dropped. Correlograms were recorded for the remaining time with a maximum time lag of (data not shown). The value for the correlations was obtained from the total number of coincident spikes in this interval. The total number of spikes was used for the measurement of the rates, while population fluctuations were determined from bins in the first .
In this section we address how recurrent connectivity affects rates and correlations. Mathematically, the kernel matrix is the adjacency matrix of a weighted directed graph. Single neurons correspond to nodes and connections are weighted by the integrated interaction kernels.
With the shorthand
Equation (9) becomes(13)
where the rates are given by (10), . For simplicity we normalise the external input, . The matrix describes the effect of network topology on rates and correlations. Under the assumptions stated in the methods section, can be written as a geometric series,
The terms of this series describe how the rates result from external and recurrent input. The matrix relates to the part of the rates resulting directly from external input. For , each of the single terms corresponds to indirect input of other nodes via paths of length . The element consists of the sum over all possible weighted paths from node to node in steps via the nodes (note that ). Since , the elements of describe the influence of neuron on neuron via all possible paths. Similarly(14)
with . The first term accounts for the integral of the autocorrelation functions of independent stationary Poisson processes, given by their rates. Higher-order terms in this series describe recurrent contributions to correlations and autocorrelation. The matrix elements of are(15)
In these expressions, a term like describes the direct effect of neuron on , taking into account the interaction strength and the rate of the presynaptic neuron. For example, in the term with and the elements describe indirect input of to via all . For , counts the common input of neurons and from all . Altogether, the series expansion of the correlation equation describes how the full correlation between neurons and results from the contributions of all neurons , weighted by their rate, via all possible paths of length to node and length to node , for all and .
These paths with two branches are the subgroup of network motifs that contribute to correlations. Further examples are given in Figure 3. The distribution of correlation coefficients depends on the distributions of these motifs. Note that larger motifs are built from smaller ones, hence distributions of different motifs are not independent.
As mentioned before, the sum (14) converges only if the magnitude of all eigenvalues of is smaller than one. This ensures that the feedback by recurrent connections does not cause runaway network activation. Both too strong recurrent excitation and too strong recurrent inhibition can lead to a divergence of the series. In such cases, our approach does not allow correlations to be traced back to specific network motifs.
Under this condition, the size of higher-order terms, that is the collective influence of paths of length and , decreases with their total length or order . This can be stated more precisely if one uses as a measure for the contribution the operator norm . After diagonalising we have(16)
where denotes the eigenvalue with the largest absolute value. If it is close to one, contributions decay slowly with order and many higher-order terms contribute to correlations. In this dynamic context the network can then be called strongly recurrent.
The average correlation across all pairs can be computed by counting the weighted paths between two given nodes. The average contribution of paths of length is(17)
Let us separate the contributions from rates to the autocorrelations and define the average correlation by(18)
The population fluctuations are determined by ,(19)
As a first approximation let us assume that every neuron in a given subpopulation projects to a fixed number of neurons in each subpopulation , denoted by . Furthermore, each neuron receives the same number of input connections from neurons of the two subpopulations, denoted by and . Synaptic partners are chosen randomly. These networks are called regular in graph theory, since the number of outgoing and incoming connections of each neuron, called the out- and in-degree, is identical for all neurons. This restriction can be relaxed to approximate certain types of networks, as we discuss in the respective sections. We set the external input . Then the total input to each neuron is . The shortcut corresponds to the average input each neuron receives from a potential presynaptic neuron.
Since input is the same for all neurons, all rates are equal. Their value can be obtained as follows by the expansion of (10),
In a similar manner, analytical expressions for the average correlations can be obtained. Explicit calculations can be found in Section 2 of Supporting Text S1. In particular, the average correlation and hence the population fluctuations only depend on the parameters and .
Closed expressions can be derived in the special case where there is a uniform connection probability between all nodes, i.e.(20)
With and one finds for the individual contributions(21)
and the average correlation(22)
Here, can be interpreted as the average direct interaction between two nodes and as the average common input shared by two nodes. Average correlations are determined by mean input and mean common input.
Equation (22) can be used as an approximation if the degree distribution is narrow. In particular this is the case in large random networks with independent connections, independent input and output and uniform connection probabilities. These conditions ensure that deviations from the fixed out- and in-degrees balance out on average in a large matrix. Numerical examples can be found in the following section.
Instead of purely random networks we now consider networks of nodes arranged in a ring with distance dependent connectivity. The type of each neuron is determined randomly with probabilities and , such that on average excitatory and inhibitory neurons are distributed over the ring. Outbound connections of each neuron to a potential postsynaptic neuron are then determined from a probability profile or , depending on the mutual geodesic distance on the ring. The average interaction between two randomly picked neurons at a distance is
A sketch for this construction scheme is depicted in Figure 6A. For the connection probabilities we use a boxcar profile, and , where denotes the Heaviside step function. Neurons with a distance smaller than are connected with a probability , where and depend on the type of the presynaptic neuron.
The stability of such a network depends on the radius of the bulk spectrum. Additionally and in contrast to the random network, besides the eigenvalue corresponding to the mean input of a neuron, a number of additional real eigenvalues exist outside the bulk spectrum. A typical spectrum is plotted in Figure 6B. These eigenvalues are particularly pronounced for locally strongly connected rings with large and belong to large scale oscillatory eigenmodes. The sign of these eigenvalues depends on the shape of the interaction profile. For short-range excitation and long-range inhibition (6C), that is a hat-like profile, these eigenvalues are positive and tend to destabilise the system. For the opposite, or inverted-hat case (6D), these eigenmodes do not affect stability, therefore stability is determined by the radius of the bulk spectrum. This can be seen as an analogue to the case of net inhibitory input in random networks.
As in a random network, the degree distribution of nodes in a ring network is narrow, hence Equation (22) is a good approximation for the average correlation if the total connection probability is independent on the neuron type,
In this case the average correlation does not depend on the specific connectivity profile. However, the full distribution of correlations depends on the connection profile, Figure 6E and F. For localised excitation the eigenvalues of oscillatory modes get close to 1, rendering the network almost unstable, and many longer paths contribute to correlations. Since for ring networks neighbouring nodes can share a lot of indirect input, while more distant ones do not, this leads to more extreme values for pairwise correlations.
We found that in networks with narrow degree distributions average correlations are determined by global parameters like the population sizes and overall connectivity , see Equation (22). In networks with broad degree distribution however, the regular-graph approximation is no longer valid. Thus, in such networks the fine structure of the connectivity will, in general, play a role in determining the average correlation. To elucidate this phenomenon, we use a network model characterised by a geometric degree distribution. The fine structure can then be manipulated without altering the overall connectivity. Specifically, the connection statistics of a given node will depend on the out-degree. The network model is defined as follows (compare Figure 8A). Out-degrees of excitatory and inhibitory neurons are chosen from a geometric distribution with a probability
where the parameter corresponds to the mean out-degree. The resulting distribution has a mean connection probability of and a long tail. Excitatory neurons are then divided into classes according to their out-degree. We will call neurons with out-degree hubs and the rest non-hubs to distinguish the classes in this specific example. Postsynaptic neurons for non-hubs and inhibitory neurons are chosen randomly from all other neurons. For each hub we fix the fraction of connections that go to other hubs. The number of connections to excitatory neurons is chosen from a binomial distribution with parameter . A number of the postsynaptic neurons are randomly chosen from other hubs, outputs go to non-hub excitatory neurons and connections to randomly chosen inhibitory neurons. By varying between 0 and 1, excitatory hubs can be chosen to form a more or less densely connected subnetwork. From the cumulative geometric distribution function, , the expected fraction of hubs is , which is about 0.35 for . If hubs are preferentially connected to non-hubs, otherwise hubs are more likely connected to each other.
By construction the parameters do not depend on . Hence terms with , including common input, are also independent of . The statistics for longer paths are however different. If excitatory hubs preferentially connect to hubs, the number of long paths within the excitatory population increases. The effects on correlations are illustrated in Figure 8. Densely connected hubs increase average correlations. While the contributions of smaller motifs do not change significantly, from the larger motifs all but the pure chain motif contributions are affected.
Different effects can be observed in networks of neurons with patchy connections and non-homogeneous spatial distribution of neuron types. A simple network with patchy connections can be constructed from neurons arranged in a ring. We consider two variants: one where all inhibitory neurons are situated in the same area of the ring, compare Figure 9A, and one where they are randomly distributed over the ring. For each neuron, postsynaptic partners are chosen from a “patch”, a population of neighbouring neurons which is located at a random position, with a probability . If neuron populations are not uniformly distributed, this leads to large variations in single neuron , even if average values are kept fixed. We compare networks where excitatory and inhibitory neurons are spatially separate, Figure 9A, versus randomly mixed populations. In Figure 9B average correlations are compared to correlations in networks with random connectivity. If excitatory and inhibitory neurons are distributed randomly, no significant increase is seen, but if populations are separate, correlations are increased strongly when patches are smaller. In Figure 9C is depicted which network motifs are responsible for the increase of correlations. It can be observed that the difference in correlation is mainly due to differences in contributions of symmetric common input motifs with , and to some extent of nearly symmetric ones (). The reason is that if neurons of the same type receive common input, firing rates of their respective postsynaptic targets will be correlated. If their types differ, their targets receive correlated input of different signs, inducing negatively correlated rate fluctuations. Patchy output connections lead to an increased fraction of postsynaptic neurons of equal type if populations are spatially separated. In this case average correlations are increased. This effect is a direct consequence of the spatial organisation of neurons and connections. The same effect could however be achieved by assuming that single neurons preferentially connect to a specific neuron type.
A comparison of motif contributions to correlations, Figures 8C and 9C, shows that different architectures increase correlations via different motifs. Asymmetric motifs play a role in the correlation increase for hubs, but almost none for patchy networks.
We studied the relation between connectivity and spike train correlations in neural networks. Different rules for synaptic connectivity were compared with respect to their effects on the average and the distribution of correlations. Although we address specific neurobiological questions, one can speculate that our results may also be relevant in other areas where correlated activity fluctuations are of interest, such as in the study of gene-regulatory or metabolic networks.
The framework of linearly interacting point processes in [38] provides a transparent description of equilibrium rates and correlations. It has been used previously to infer information about direct connectivity from correlations in small networks [44], as one amongst many other methods, see for example [49], [50] and references therein. Another application was the study of spike-time dependent plasticity [45], [51] and, in an extended framework, the description of spike train autocorrelations in mouse retinal ganglion cells [52]. An approach using linearised rate dynamics was applied to describe states of spontaneous activity and correlations in [53]. Correlations in populations of neurons have been studied in a rate model in [36] and in a point process framework in [37]. Hawkes' point process theory allows the treatment of correlations on the level of spike trains as well as the understanding of the relation of complex connectivity patterns to the statistics of pairwise correlations.
Although Hawkes' equations are an exact description of interacting point processes only for strictly excitatory interactions, numerical simulations show that predictions are accurate also for networks of excitatory and inhibitory neurons. Hence correlations can be calculated analytically even in effectively inhibitory networks in a wide range of parameters, as has already been proposed in [39]. One should note, however, that for networks with strong inhibition in combination with strong synaptic weights and low external input, low rates are not reproduced well.
The activity of cortical neurons is often characterised by low correlations [27], and can exhibit near-Poissonian spike train statistics [54] with a coefficient of variation near one. In theoretical work, similar activity has been found in balanced networks [41] in a certain input regime [40]. The level and time dependence of external input influences the general state of activity as well as pairwise correlations. In this study we are only concerned with an equilibrium resting state of a local network with asynchronous activity where external input is constant or unknown. We use Poisson processes as a phenomenological description for such a state and do not consider the biophysical mechanisms behind spiking activity, nor the reasons for asynchronous spiking on a network level. However, we found in simulations of networks of integrate and fire neurons of comparable connectivity parameters in an asynchronous-irregular state that correlations can be attributed to a large degree to linear effects of recurrent connectivity, although single neuron dynamics are nonlinear and spike train statistics are not ideally Poissonian (data not shown). Thus, although a linear treatment may seem like a strong simplification, this suggests that Hawkes' theory can be used as a generic linear approximation for the spike dynamics of complex networks of neurons. A similar point has been made in [53].
We quantified correlations by integrated cross-correlation functions in a stationary state. The shape of the resulting correlation functions, which has been treated for example in [30], [37], [55], was not analysed. The advantage is that our results are independent of single neuron properties like the shape of the linear response kernel. Specific connectivity properties that can be described by a graph, as for example reviewed in [3], can be directly evaluated with respect to their effects on correlations.
In Hawkes' framework, taking into account contributions to pairwise correlations from direct interactions, indirect interactions, common input and interactions via longer paths is equivalent to a self-consistent description of correlations. This interpretation helps to derive analytical results for simple networks. Furthermore it allows an understanding of the way in which recurrent connectivity influences correlations via multiple feed-back and feed-forward channels. In particular, we showed why common input and direct input contributions are generally not sufficient to describe correlations quantitatively, even in a linear model. We showed that average correlations in networks with narrow degree distributions are largely independent of specific connectivity patterns. This agrees with results from a recent study [42], where conductance based neurons in two-dimensional networks with Gaussian connectivity were simulated. There, the degree of single neurons was kept fixed and population averaged correlations were shown to be invariant to different connectivity patterns. For net-inhibitory networks, indirect contributions to correlations effectively reduce average correlations. A similar effect has been described in [20] and in [36] for a rate model. In networks with strong recurrence, characterised by eigenvalues of the connectivity matrix close to one, correlation distributions are strongly influenced by higher-order contributions. In these networks broad distributions of correlations arise. In contrast, in very sparsely connected networks correlations depend mainly on direct connectivity.
Can we estimate the importance of recurrence from experimentally accessible parameters? In [56] the probability of a single extra input spike to generate an additional output spike, corresponding to , has been measured in rat barrel cortex in vivo as 0.019. Additionally, the number of connections made by each neuron was estimated to be about 1500. We now consider a local network with a fraction of inhibitory neurons of 20%. We assume an inhibitory synaptic weight to balance the excitation, such that . The estimated mean degree is consistent with many different topologies. Let us consider the case of a uniform random network of 15000 neurons with connection probability 0.1. For comparison we also look at a densely connected subnetwork of just 2500 neurons with a connection probability of 0.6. The first model results in a spectral radius for the connectivity matrix , hence falling in the linearly unstable regime. In contrast, the second network displays a spectral radius slightly below one, which indicates linear stability. What can we conclude from this discussion? In the first place, this crude estimate of the spectral radius suggests that a value in the order of one is not an unrealistic assumption for real neural networks. This would call for a consistent treatment of long-range, higher-order interactions. This view is also supported by simulations of integrate and fire networks [31], which can yield similarly values for the spectral radius close to one. Our second example, although biologically less realistic, shows the range in which the spectral radius can vary, even if certain network parameters are kept fixed. This highlights the importance of the connectivity structure of local neural networks, as different network architectures can strongly affect the stability of a certain activity state.
We addressed ring networks with distance-dependent connection probability. Here, average correlations do not depend on the connectivity profile. However, for densely coupled neighbourhoods very broad correlation distributions can arise. A Mexican hat-like interaction has especially strong effects, since in that case higher-order contributions amplify correlations. This is not surprising since it is known that Mexican hat-like profiles can support large-scale activity patterns [57]. This implies that local inhibition increases network stability and leads to less extreme values for correlations. Distributions of correlations and distance dependence of correlations have been measured experimentally [20], [21], but they have not yet been related directly to anatomical connectivity parameters. In [19], the distance dependence of pairwise correlations as well as higher-order correlations has been measured experimentally. A generalisation of Hawkes' correlation equations in conjunction with the framework of cumulant-correlations discussed in [58] presents a promising route to study structure dependence also of higher-order correlations.
A generalisation to two-dimensional networks with distance dependent connectivity could be used to further investigate the relation between neural field models which describe large-scale dynamics [59]–[61] and random networks. However, the analysis using the full connectivity matrix allows to incorporate effects of random connectivity beyond the mean field limit. One example is that stability of networks is not only determined by mean recurrent input, but also by input variance.
Pairwise correlations affect activity in pooled spike trains [62]. We found that distance dependence of connectivity creates strongly coupled neighbourhoods and that population signals therefore depend on the connectivity statistics of the network. Such population signals could for example be related to local field potentials.
If the degree distribution is wide, networks can be constructed where connection probability depends on the out-degree of postsynaptic neurons. We considered networks where excitatory hubs, defined by a large out-degree, form a more or less densely connected subnetwork. Similar networks have been studied in [63]. In graph-theoretic terms, the connectivity between these hubs influences the assortativity of the network. A commonly used measure is the assortativity coefficient, which is the correlation coefficient between degrees of connected nodes. We calculated a generalised version for weighted networks, the weighted assortativity coefficient [64]. It can assume values between -1 and 1. Our networks have values between −0.22 and −0.05. Negative assortativity values are a consequence of the geometric degree distribution, but networks with more densely connected hubs have a higher coefficient. In our model, more assortative networks exhibit larger correlations than more disassortative ones. This illustrates how differences in higher-order statistics of connectivity can influence correlations, even if low order statistics do not differ.
In networks with patchy connections, an increase of correlations can be observed when populations of neurons are spatially non-homogeneous. Some information about how network architecture influences correlations can be obtained from examining contributions of individual motifs. In patchy networks mainly the contributions of symmetric motifs are higher, when excitatory and inhibitory neurons are separated, and therefore responsible for the correlation increase. In networks with hubs also asymmetric motifs play a role.
We found that fine-scale structure has important implications for the dynamics of neural networks. Under certain conditions, like narrow degree distributions, local connectivity has surprisingly little influence on global population averages. This suggests the use of mean-field models. On the other hand, broad degree distributions or the existence of connected hubs influence activity also on the population level. Such factors represent, in fact, major determinants of the activity state of a network and, therefore, should be explicitly considered in models of large scale network dynamics.
As considerable efforts are dedicated to the construction of detailed connection maps of brains on multiple scales, we believe that the analysis of the influence of detailed connectivity data, possibly with more refined models, has much to contribute to a better understanding of neural dynamics.
|
10.1371/journal.pntd.0005064 | Investigating the Contribution of Peri-domestic Transmission to Risk of Zoonotic Malaria Infection in Humans | In recent years, the primate malaria Plasmodium knowlesi has emerged in human populations throughout South East Asia, with the largest hotspot being in Sabah, Malaysian Borneo. Control efforts are hindered by limited knowledge of where and when people get exposed to mosquito vectors. It is assumed that exposure occurs primarily when people are working in forest areas, but the role of other potential exposure routes (including domestic or peri-domestic transmission) has not been thoroughly investigated.
We integrated entomological surveillance within a comprehensive case-control study occurring within a large hotspot of transmission in Sabah, Malaysia. Mosquitoes were collected at 28 pairs households composed of one where an occupant had a confirmed P. knowlesi infection within the preceding 3 weeks (“case”) and an associated “control” where no infection was reported. Human landing catches were conducted to measure the number and diversity of mosquitoes host seeking inside houses and in the surrounding peri-domestic (outdoors but around the household) areas. The predominant malaria vector species was Anopheles balabacensis, most of which were caught outdoors in the early evening (6pm - 9pm). It was significantly more abundant in the peri-domestic area than inside houses (5.5-fold), and also higher at case than control households (0.28±0.194 vs 0.17±0.127, p<0.001). Ten out of 641 An. balabacensis tested were positive for simian malaria parasites, but none for P. knowlesi.
This study shows there is a possibility that humans can be exposed to P. knowlesi infection around their homes. The vector is highly exophagic and few were caught indoors indicating interventions using bednets inside households may have relatively little impact.
| The primate knowlesi malaria has now emerged in human populations throughout South East Asia. Our limited knowledge of where and when people get exposed to the vector (Anopheles balabacensis) has resulted in poor control measures, although it is assumed that exposure occurs primarily when people are working in forest areas. We investigated the role of peri-domestic (outdoors but around the household) and domestic transmission. Mosquitoes were collected at 28 pairs households composed of one where an occupant had a confirmed knowlesi malaria infection and an associated “control” where no infection was reported. Most of the vectors were caught outdoors from 6pm - 9pm. The vectors were also significantly more abundant in the peri-domestic area than inside houses (5.5-fold), and also higher at case than control households. Ten Anopheles (out of 641) were found positive for primate malaria parasites. This study shows that humans can be exposed to knowlesi infection around their homes. Given the vectors are mainly outdoor biters, interventions using insecticide treated bednets inside households may have relatively little impact. A paradigm shift in control methods is required to reduce infection of this primate malaria.
| The success story of reducing malaria worldwide [1] has been marred by a few notable exceptions where bulk of disease is caused by zoonotic “neglected” malaria species with atypical transmission that makes them less easy to control. Zoonotic malaria, such as Plasmodium knowlesi from the long tailed macaque (Macaca fascicularis) in SE Asia [2], and P. brasilanum from NewWorld monkeys in South America [3], are a growing public health problem.
In South East Asia, the long tailed macaque harbours at least five simian malarias, namely, P. coatneyi, P. inui, P. fieldi, P. cynomolgi and P. knowlesi [4,5]. Plasmodium knowlesi is presently the main zoonotic malaria with the greatest public health importance, especially in Sabah, Malaysian Borneo which has recorded the highest growing number of P. knowlesi cases in humans, and most of these cases are clustered within one district, Kudat in the North eastern region [6,7]. Plasmodium knowlesi is morphologically similar to P. malariae and had been misdiagnosed as such for a long time [2]. A first case of naturally acquired human infection with P. cynomolgi has also been reported from peninsular Malaysia [8]. Thus it is a possibility that other primate parasites may also be soon contributing to human cases as previously predicted [9].
In Sabah, it has been confirmed that An. balabacensis is the primary vector [10] and the long tail macaques, which are the natural reservoir hosts for simian malaria parasites are present. Furthermore, Kudat district has many secondary forest areas surrounded by hilly areas, oil palm estates and rubber plantations which in general serve as habitats not only for long-tail macaques but also for Anopheles species. Close interaction between monkeys, mosquitoes and human increases the chances of being infected with P. knowlesi.
The vectors of P. knowlesi malaria in Malaysia comprise of five Anopheles species of the Leucospyrus group namely: An. hackeri, An. latens, An. cracens, An. introlatus and An. balabacensis [10–15]. In Vietnam Anopheles dirus of the Dirus group was recorded as the vector [16,17]. These vectors are found mainly in the forests and are outdoor biters, and likely to have low susceptibility to frontline control strategies which typically involve use of insecticides in homes.
In Malaysia, the National Malaria Eradication Program was launched in 1967, followed by state-wide malaria control programs during the 1970s and 1980s. Consequently, great reductions in malaria prevalence were recorded, from 240,000 in 1961 to around 50,000/year during the 1980s [11,18]. The success of the eradication programs was also reflected in Sabah, East Malaysia where malaria notifications decreased sharply, from peak notifications of 33,153 and 15,877 during 1994–1995 for P. falciparum and P. vivax respectively to 605 and 628 respectively in 2011. Similarly notifications of P. malariae/P. knowlesi also fell from a peak of 614 in 1994 to <100/year in the late 1990s/early 2000s [6].
Although Malaysia has shown considerable success in the control of human malaria and is on target towards elimination of malaria by 2020 [19], notifications of suspected P. knowlesi cases have increased from 59 notifications in 2004 to 996 in 2013, an overall increase of over 16-fold [6]. According to the Malaysian Ministry of Health, P. knowlesi is the predominant species occurring in the country comprising 62% of the cases in 2013 [20]. It was suggested that the increase in number of P. knowlesi notification in Kudat maybe be due to high awareness of knowlesi infection among physicians and availability of better diagnostic tools to identify this malaria parasite [20]. In other words what is reported now represents the true infection rate as compared in the late 1990s and early 2000s when P. knowlesi cases were misdiagnosed as P. malariae. However, recent findings by Fornace et al. had demonstrated a clear link between land use change and P. knowlesi incidence, which strongly supports the idea that this is not just a problem of poor diagnosis/changing awareness, but a real epidemiological change [21].
The frontline vector control methods practiced in Malaysia under the malaria elimination programme are same as those used in several other endemic settings within the region i.e. the application of insecticides in houses either through use of Long Lasting Insecticide Treated Nets (LLINs) or Indoor Residual Spraying (IRS). However there is little evidence to support that this is effective against P. knowlesi vectors.
Thus there is a need for more detailed entomological investigation to assess the relative importance of exposure to mosquito vectors at or away from home, and to design control measures accordingly. Our working hypothesis is that given An balabacensis is the primary vector, we would expect that infection risk is higher when they are present. Towards this end, a case-control entomological study was conducted to determine if P. knowlesi infection risk is linked to exposure to vectors in domestic and peri-domestic settings. Specific comparisons were included to test for differences in vector abundance, species composition, biting time and infection rate at case and control households. Further aims were to assess and characterize the biting behavior of P. knowlesi vectors near homes (time and place of biting e.g outdoors vs indoors), and to evaluate the potential for spread of other primate malarias in domestic settings. These findings will be useful for the control programme in designing vector control measures.
The study was conducted in Kudat district which is located in the northeasternn tip of Sabah (6°53'14.35" N 116°49'25.10" E) and covers an area of approximately 1,300km2 with a population of 84,000 people of predominantly Rungus ethnicity (2010 National Census). The climate is tropical and the area is mainly coastal, with a maximum elevation of 250 metres above sea level. Forest cover is highly fragmented and substantial deforestation has occurred through conversion of forest to agricultural land [21, 22]. The majority of the population lives in small villages (mean population 160±15, S1 Table) and the main livelihood activities are small scale farming and plantation work.
Plasmodium knowlesi is the main cause of human malaria in Kudat and, due to this relatively high incidence, this area is the focus for a number of interdisciplinary studies on the biomedical, environmental and social risk factors for P. knowlesi (http://malaria.lshtm.ac.uk/MONKEYBAR). This includes a case control study, in which clinical malaria cases were recruited from the district hospital and visited at their homes within two weeks of initial diagnosis [22]. As there is mandatory reporting and referral of all malaria cases to the district hospital, the majority of symptomatic cases are captured by hospital systems.
Approximately 180 P. knowlesi cases were identified through this active case surveillance between 2013–14; of which we randomly selected a subgroup of 28 for further entomological follow up (representing cases reporting February, July, 2014 from 23 different villages: Fig 1, S1 Table). The cases came from the age group 19–74 years old with a mean of 44. There was a preponderance of males amongst the cases, and many (78.6%) worked in agricultural sector, taking at least about 10–30 minutes to walk to their work place (Table 1)
Additionally, a matched “control” household was recruited in the vicinity of the case household for study which shared similar environmental characteristics in terms of surrounding vegetation and terrain, but where no occupants had reported with any malaria infection within the study period as indicated by records from the hospital and interviewing the residents. From the group of potential “control” households identified in the vicinity, one was randomly selected using a random table. The final choice was also dependent on the owner’s agreement. Within two weeks of the case detection, the selection of control house was accomplished and entomological work initiated.
Data on the types of crops or vegetation surrounding households was collected, as well as the distance between each pair of case and control household (S2 Table). The mean distance between the case and the control houses was 255±48 m (18–1000 m). As the villages were generally small in area, occasionally a control house could be located in a neighbouring village. Of the 28 case-control household pairs, five pairs occurred within the same village.
For all pairs of households, mosquito sampling was conducted by four workers each at case and control simultaneously on the same nights. At these pairs, sampling was conducted for one to three nights depending on the owner’s permission and logistic constraints.
Indoor collections were conducted at one station in the living room of houses (H), whereas outdoor collections were conducted at three selected stations (S1, S2 & S3) within the garden area surrounding the house. The distance of the stations from the house was 24±1.7 m for the case and 19±0.8 m for the control household. These outdoor stations were selected based on information provided by the family members about where they were most likely to spend time outdoors in the evening. Mosquitoes were baited using human landing catch (HLC) method, but only Anopheles spp were collected for further analysis, whilst other species were killed and discarded at the site. Here a volunteer collected mosquitoes by exposing his lower legs (from knee downwards) and collecting all mosquitoes that land upon them in a plastic specimen tube (2 cm diameter X 6 cm) which had a small piece of moist tissue at the bottom. Each station was manned by one person who would collect Anopheles for 12 hours straight (1800 to 0600 hr), and there was rotation of workers. The HLC workers at each case and control house were regularly monitored by a supervisor. The Anopheles were kept individually in a tube with a label which had information on the place, date, hour, location (indoors vs outdoors) and station of collection. The mosquito samples collected were recorded by hour in order to estimate the biting profile over course of night. The next morning the samples were taken to the laboratory to be processed.
Anopheles specimens were identified the next morning in the laboratory based on morphology characters using published identification keys. The key of Sallum et al. [23] was used for Leucosphyrus group, whereas keys developed by Rattanarithikul et al. [24] were used for other groups. The identified specimens were kept individually in a sterile 1.5mL microfuge tube and stored in -20°C until used for molecular analysis.
Each Anopheles specimen was cut into two parts: head-thorax and abdomen, and placed separately inside an autoclaved mortar and the tissue homogenized using pestle. The total DNA was extracted from each part following the method of Phillips and Simon [25] and stored in -30°C until PCR analysis. Detection of malaria parasites in the Anopheles specimens was performed using the nested PCR Plasmodium genus-specific method described by Singh et al. [26]. When a sample was found positive for malaria parasites, a second nested PCR was performed to determine the Plasmodium spp. using species specific primers in singleplex PCR [4, 8, 26,27]. Primers of nine species of Plasmodium namely P. coatneyi, P. inui, P. fieldi, P. cynomolgi, P. knowlesi, P. falciparum, P. vivax, P. malarie and P. ovale were used in this study. All these species have been recorded in Malaysia although P. ovale is an imported species, while P. knowlesi is the prevalent simian malaria infecting man.
Both PCR 1 and PCR 2 were performed with 25μl final volume. The reaction components were prepared by mixing 5.0μl of 5X PCR buffer (Promega), 0.5μl of dNTPs (10mM) mixture (Promega), 3.0μl of 25mM MgCl2 (Promega), 1.0μl each of 10μM forward and reverse primers, 0.3μl of Taq DNA polymerase (5U/μl), 2.0μl of DNA template and sterile dH2O up to 25μl final volume. After the first PCR was completed, 2.0μl of the first PCR product was used as a template in the second PCR. The PCR conditions used were: an initial denaturation at 95°C for 5 min, followed by 35 cycles of 94°C for 1 min, annealing for 1 min and 72°C for 1 min, and a final extension at 72°C for 5 min. The annealing temperature was set based on the optimum temperature of the primers (S3 Table).
Statistical analysis was conducted using R programming language for statistical analysis (version 3.2.2). Generalised linear mixed models (GLMM) were constructed to test for variation in the abundance of Anopheles between case and control houses, and indoor and peri-domestic settings. In the analysis, household type (case or control) and location (indoor or outdoor) were considered as fixed effects, while month, night and sampling station (site) as random effects. To identify the best model, both negative binomial and Poisson distributions, interaction between type and location, as well as zero inflation were fitted. Tukey's Post Hoc test was used to compare mean between fix effects (household type and location) as well interaction between these two effects.
We also analysed the proportion of mosquitoes that were caught feeding outdoors (Po), and the proportion of human exposure to An. balabacensis (Pe).
The proportion of mosquitoes that were caught outdoors (Po) was calculated as
Po=O18−06 h(O18−06 h+I18−06 h)
(1)
where O and I are respectively the number of mosquitoes caught biting outdoors and indoors during 6 pm– 6 am.
From interviewing the residents and observation, more than 50% of the villagers would be indoors by 8 pm, and out the next morning by 5 am as they go to the plantations to work. The proportion of human exposure to An. balabacensis (Pe) is thus calculated as the proportion of bites that happen outdoors during the time when people are likely to be outdoors, out of the sum of bites expected throughout the night as humans move between indoor and outdoor areas of their home:
Pe=O18−20, 05 h(O18−20, 05 h+I20−05 h)
(2)
where O18-20, 05 h represents the mosquitoes caught biting from 6–8 pm, and 5-6am, and I20-5h represents the number caught indoors between 8 pm– 5 am. This would give a comparison between the proportion of bites people exposed to when outdoors between case and control households.
GLMM with a binomial distribution and a logit link function was used to obtain the binary estimates of Po and Pe. In these models, household type (case or control) was fitted as a fixed effect, while sampling station of the case as a random effect.
This project was approved by the NMRR Ministry of Health Malaysia (NMRR-12-786-13048). All volunteers who carried out mosquito collections signed informed consent forms and were provided with antimalarial prophylaxis during participation. House owners also gave permission to use their houses for collection of mosquitoes.
Among those who worked in the agricultural sector (Table 1), more than double the number cases were employed on rubber or oil palm plantations than controls, while more controls worked in the vegetable farms than cases.
A total of 793 Anopheles belonging to 12 species were caught during the period of study, with An. balabacensis being the dominant species (81% of total), followed by An. maculatus, An. barbumbrosus, and An. donaldi (Table 2). Overall, more An. balabacensis were caught at case (total 392 or 1.81 bites per man per night) than control houses (total 249 or 1.15 bites per man per night). Ten and 9 different Anopheles species were collected at case and control houses respectively, compared to only 6 and 2 indoor collections from case and control. Higher numbers were recorded at Kg. Tinukadan Laut (CC24), Kpg. Paradason B (CC26), Kpg. Nangka (CC5) (S1 Fig).
GLMM analysis indicated that the negative binomial distribution gave a better fit than Poisson distribution. The log-likelihood values were for negative binomial and Poisson distribution -513 vs -529 respectively, while the Akaike information criterion (AIC) values were 1043 vs 1068. Adopting the model with a negative binomial distribution, the abundance of An. balabacensis was found to vary significantly between case and control (case 0.28±0.194 vs control 0.17±0.127, z = 4.62, p<0.001), and between the surrounding peri-domestic area and inside the house (0.56± 0.394 versus 0.09±0.063, z = 9.09, p<0.001) (Fig 2). The interaction between house type and location was not significant (z = 0.8, P>0.05)
More than 50% of An. balabacensis mosquitoes were caught biting in the early evening (6pm - 9pm) with the peak hour between 7pm - 8pm (Fig 3). After 8pm, the number rapidly decreased with approximately 84% of the total nightly catch being accumulated by midnight. Nevertheless, one or two individuals of An. balabacensis could still be caught until dawn.
In general, the proportion of bites taken by An. balabacensis outdoors (Po) was very high (>95%), and did not vary between case and control households (Table 3). Similarly the proportion of human exposure to bites (Pe) did not vary between case and control households.
A total of 793 Anopheles individuals were tested by molecular method and only ten An. balabacensis (out of 641 or 1.56%) were found to be positive for malaria parasites. Seven of them were caught at case houses (6 outdoors and one indoors), and 3 at control houses (all outdoors, Table 4). All these mosquitoes were found to be positive for simian malaria parasites (P. coatneyi, P. inui, P.cynomologi & P. fieldi); but none were with either the dominant zoonotic parasite reported in the area (P. knowlesi) or any human-specific Plasmodium. Ninety percent of infected An. balabacensis was caught biting outdoors between 6pm - 10pm (9 out of 10), and with one infected individual caught between 1am - 2am.
The proportion of infected An. balabacensis caught from case houses (7/392 or 1.79%) was slightly higher than at control houses (3/249 or 1.20%). However, the sample sizes were too low for any robust statistical analysis of differences.
We conducted a randomized case-control field study to test the hypothesis that there is an association between P. knowlesi infection risk and higher exposure to mosquito vectors in peri-domestic (outdoors surrounding houses) and within domestic (inside house) settings. Although 12 Anopheles species were caught by HLC, only 4 were detected in reasonably high abundance: An. balabacensis, An. maculatus, An. barbumbrosus and An. donaldi. Of these, only An. balabacensis and An. donaldi have been previously implicated as malaria vectors of human malaria in Sabah [18,28]. In contrast, An. maculatus is the main malaria vector of human malaria in peninsular Malaysia [29]. Anopheles balabacensis appears to be a widespread species found in almost all sites, although significantly high numbers were caught in Kpg Tinukadan Laut (CC24) especially in those areas near to forest fringes. About 95% of An. balabacensis was caught outdoors, similar to what was previously recorded (a ratio of outdoor:indoor catch of 24:1) in Kuala Penyu, another district in Sabah [28]. A recent study conducted in Banggi Island situated north of Sabah and in Kg Paradason in the interior of Kudat district [10] also revealed that An. balabacensis was the predominant species collected, followed by An. donaldi in both sites. However, the next most abundant species was An. vagus, in Banggi, but An. barbirostris group in Kg Paradason. Anopheles maculatus was not caught in Banggi.
GLMM analysis indicated a significant difference between the number of vectors caught at case and control houses, and between outdoor and indoor catches. The primary vector of P. knowlesi in the area, An. balabacensis, was present at higher abundance at households where cases were reported, which would suggest a higher risk at the case houses. Furthermore, as 90% of the infected mosquitoes were caught outdoors, it is likely that peri-domestic infection is an important risk factor. Although the indoor number of infective mosquitoes caught was small, getting infected indoors cannot be discounted.
In Sarawak, it was postulated that humans were likely to acquire infection of P. knowlesi from being bitten by infected An. latens while hunting in the forest or as they return from the farm around dusk since in their study no infective mosquitoes were obtained from the village [13]. However, in Sabah clustering of cases among family members have been reported and they postulated that people could be infected around their homes [30]. A recent study also in Sabah showed the presence of asymptomatic cases of P. knowlesi occurring among the community in Kudat [31]. Thus, the result of this study seems to support the hypothesis that it is also possible for people to be infected in and around their homes. Although An. balabacensis is highly exophagic with only one infected individual found indoors we should not dismiss the fact that the possibility of indoor infection does exist. Thus we need to possibly expand our paradigm about transmission of P. knowlesi to include the possibility of peri-domestic infection, and conduct further studies to evaluate simultaneously the infection risk in and around households, as well as in forest areas, so the relative contribution of all these routes could be formally quantified.
Many areas in Kudat district have undergone deforestation and clearance of vegetation for crop plantations, but it appears that An. balabacensis has remained the dominant species, with the exception of Kinabatangan area of Sabah where An. balabacensis was replaced by An. donaldi as main malaria vector as a result of deforestation and malaria control activity [18]. This suggests that the abundance of An. balabacensis in Kudat district was not greatly affected by the environmental changes. The impact of forest disturbance such as logging has been shown to increase the abundance of this disease vector and may partly explain the rapid rise in P. knowlesi cases in Sabah [32].
The feeding time of An. balabacensis appears to have changed since late 1960s when most of the area in Kudat district was still covered with forest. A study conducted then [29] showed that An. balabacensis was actively biting human at late night (10pm onwards), compared to early night with peak hour between 7pm to 8pm recorded now. In fact, this species starts biting human outdoors as soon as it starts to get dark. This change in feeding time could be due to An. balabacensis adapting to more people staying closer to forest fringe as more forested areas are cleared for crop plantation and housing. This could also be due to the introduction of insecticide treated bednets. Further study will be needed to confirm this.
Although we did not obtain an An balabacensis individual infected with P. knowlesi, as only ten individuals were found Plasmodium positive albeit for other simian malarias, this could be a sampling error. As such, we are unable to make a conclusive prediction about infection risk at case households. Given the generally low rates of P. knowlesi infection in the vector (eg 13/1482 or 0.88%, data also collected in Kudat) [10], thousands of samples are needed to obtain strong evidence to show that P. knowlesi was not present, and/or to compare infection rates between case and control households. Furthermore, as the densities of vectors in these settings are generally low, it would not have been feasible to achieve this sample size within the one year time span that the case control study was running. What data collected here do show however is that the primary vector is present at higher abundance in peridomestic settings where cases are reporting, on which basis the possibility of peridomestic transmission cannot be dismissed. This also indicates that people are routinely exposed to a variety of different primate malarias around their home; but that to date, only a couple (knowlesi and cynomologi) seem capable to causing any clinical infection. More research needs to be carried out to determine why these two primate malarias succeeded where the others fail so we can be proactive in the fight against future new simian malarias infecting man.
The difference between bites per man per night between case and control houses is 0.66 (1.81–1.15) which works out to be 241more bites per person in a year for the case house. Since the infective proportion of vector is 0.88 [10], this is equivalent to a higher entomological inoculation rate (EIR) of 2.12. In addition, more than double the number cases worked in rubber or oil palm plantations than controls. Perhaps these two factors may help explain why there was P. knowlesi infection in the case houses. However more research is needed to validate this.
As most P. knowlesi cases have been recorded from villages close to where macaques abound, and given that the primary vectors species, An balabacensis bite outdoors, a new paradigm in managing this malaria is needed. More attention should be focused on the ecology and biology of An. balabacensis in order to develop more effective control methods if the control or elimination of P. knowlesi malaria in Kudat district is to be successful. The current malaria control programme using ITNs might not have the desired impact as this species is mainly an exophagic species, and infection is more likely to occur outdoors in peri-domestic settings, in plantations and forest.
|
10.1371/journal.pcbi.0030212 | Self-organizing Mechanism for Development of Space-filling Neuronal Dendrites | Neurons develop distinctive dendritic morphologies to receive and process information. Previous experiments showed that competitive dendro-dendritic interactions play critical roles in shaping dendrites of the space-filling type, which uniformly cover their receptive field. We incorporated this finding in constructing a new mathematical model, in which reaction dynamics of two chemicals (activator and suppressor) are coupled to neuronal dendrite growth. Our numerical analysis determined the conditions for dendritic branching and suggested that the self-organizing property of the proposed system can underlie dendritogenesis. Furthermore, we found a clear correlation between dendrite shape and the distribution of the activator, thus providing a morphological criterion to predict the in vivo distribution of the hypothetical molecular complexes responsible for dendrite elongation and branching.
| Neurons elaborate two types of neuronal extensions. One is axon, which sends outputs to other neurons. Another is dendrite, which is specialized for receiving and processing synaptic or sensory inputs. Like elaborated branches of trees, the shape of dendrites is quite variable from one type to another, and different dendritic geometry contributes to differential informational processing and computation. For instance, neurons of the space-filling type (e.g., retinal ganglion cells) fill in an open space to pick up spatial information from every corner of their receptive field. Therefore, dendrite development is one of the representative examples of the emergence of function through morphogenesis. Previous experiments including ours showed that competitive dendro-dendritic interactions play critical roles in shaping dendrites of the space-filling type. In the present study, we incorporated this finding in constructing a new mathematical model, in which reaction dynamics of chemicals are coupled to neuronal dendrite growth. Our numerical analysis suggested that self-organizing property of the proposed system underlies formation of space-filling dendrites. Furthermore, we provided a morphological criterion to predict the in vivo distribution of the hypothetical molecular complexes responsible for dendrite elongation and branching. We have now found a substantial number of molecules involved in dendrite development, thus it is timely to discuss the prediction from this work.
| One of the primary interests in developmental biology is the emergence of function through morphogenesis. Morphological diversity of dendrites and its impact on neuronal computation perfectly represents the importance of this problem: shapes of dendrites are highly variable from one neuronal type to another, and it has been suggested that this diversity supports differential processing of information in each type of neuron [1–3]. Therefore, patterning of neuronal class-specific dendrites is a process to produce shapes that realizes the physiological functions of neurons. Recent advances in genetic manipulation at the single-cell level enabled us to identify genes whose loss of function affects neuronal morphology (reviewed in [4–6]); however, we are far from formulating an overall picture of the underlying mechanism of pattern formation.
Among various classes of dendrites is the “space-filling” type, which uniformly covers its receptive field. The concept of space-filling was introduced by Fiala and Harris [7], and we use this term with a slightly different meaning here. Neurons elaborating space-filling dendrites are found in various parts of nervous system, including retinal ganglion cells [8], trigeminal ganglion cells [9], Purkinje cells (Figure 8B) [10], and Drosophila class IV dendritic arborization (da) neurons (Figure 1) [11–14]. The space-filling type looks very complex morphologically, but can be regarded as being simple in their isotropic features and in their two-dimensionality. Most importantly, it shows distinctive spatial regulation of pattern formation: for instance, dendritic branches of Drosophila class IV da neurons avoid dendrites of the same cell and those of neighboring class IV cells, which allows complete, but minimal overlapping, innervation of the body wall (designated as isoneuronal avoidance and tiling) (Figure 1A and 1B) [11,13–15]. Our previous experiment together with studies by others demonstrated that competitive dendro-dendritic interaction underlies tiling, as shown by the fact that the da neurons reaccomplish tiling in response to ablation of adjacent neurons of the same class or to severing of their branches (Figure 1C) [11,14]. It should be noted that the qualitatively same inhibitory dendro-dendritic interaction is working between the adjacent neurons of the same type as well as between dendrites of the same neurons.
There are two types of the proposed mechanisms that support this repulsive behavior of dendrites: one is contact-dependent retraction of dendrites and the other is repulsion of dendrites via diffusive suppressors. The contact-dependent mechanism seems insufficient to a clear field splitting, because as far as dendrites do not make contacts (by passing under other dendrites, for example) they can invade neighboring territories. Moreover, time-lapse analysis showed that dendrites make a turn before they are about to cross nearby branches [16]. So we prefer diffusive signaling to a contact-dependent one. Similar mechanisms have been suggested to work in other model systems as well [9,17,18]. With all available information taken together, we considered the space-filling dendrite to be an ideally suited one for us to start modeling, due to the simplicity of its patterning and the experimentally verified mechanism of the pattern formation.
A number of mathematical models for neurite formation were previously proposed; and most of them assumed that dendrite development is a consequence of stochastic sprouting and subsequent growth arrest [19–22]. Different forms of branching functions were postulated and modified so that calculated dendrograms would fit dendritic arbors of real neurons in a quantitative manner. Those studies were descriptive and did not provide a comprehensive mechanism of pattern formation. In this study, we developed a new class of mathematical model for neurite formation to approach a principle of development of the space-filling dendrites. In our neurite growth model that is based on the aforementioned inhibitory dendro-dendritic interaction, various aspects of pattern formation, e.g., extension, orientation of growth, and branching of dendrites, are represented in a single framework. Computer simulation showed that our model develops dendritic extension and branching autonomously; furthermore, numerical analysis determined the conditions for dendritic growth.
As mentioned above, two-dimensionality is a characteristic of space-filling dendrites; thus we built our model in the 2D space, dividing the 2D space into two distinct compartments (Figure 2A), i.e., the compartment occupied by neurons (designated as the cell compartment or the cell region hereafter) and the extracellular compartment. This model is referred to as the “cell compartment model.” We assumed that growth of the cell region, which shapes the dendritic trees, is regulated by a hypothetical intracellular chemical, i.e., the activator (Figure 2A). We set a restriction in terms of the movement of the activator so that it diffuses only the inside of cells. The activator promotes the growth of the cell compartment when its concentration is higher than threshold (Tr in Figure 2D). To account for the inhibitory dendro-dendritic interaction, we hypothesized another chemical, the suppressor. The suppressor is produced when the concentration of activator is high, i.e., it is produced at the actively growing region of dendrites. The suppressor acts to decrease the concentration of activator, but the concentration of activator can increase by its autocatalytic production where the activator is locally concentrated. The reaction between activator and suppressor is the so-called “activator-inhibitor type” [23,24] (“1” in Figure 2A). The activator-induced production of the suppressor can be realized by either local translation of suppressor-encoding mRNA in dendrites [25] or secretion of suppressor proteins from intracellular organelles. The suppressor is secreted from the cell, diffuses throughout the extracellular space, and then binding of the suppressor to its receptor drives intracellular signaling to repress the production of the activator (“2” in Figure 2A). So, the suppressor mediates long-range inhibitory interactions between dendrites (Figure 2B). These settings endow the system with feedback-loop regulation at two different levels: one is between the two chemicals and the other is between the dynamics of these chemicals and the expansion of the cell compartment. The latter consists of the following reciprocal interactions: the activator controls growth of the cell region and growth of the cell region determines where the activator can diffuse further.
Our model can be written as the following equations:
where u and v are concentrations of the activator and the suppressor, respectively. Note that these equations are already non-dimensionalized, so d is the ratio of the diffusion coefficient between the two substances (see the section “Original equations”). As we hypothesize that diffusion of the suppressor is faster than that of the activator, d is larger than 1 [26]. c(x, t) is a symbolic variable to indicate the “core” of the cell (Figure 2C and 2D). The biological correlates of c could be microtubules that support structural integrity of the cell. The right-hand side of Equation 1c indicates that the dynamics of the cell state is bi-stable and that the two steady states are 1 and 0, indicating “core” and “not core,” respectively, and a(u) is the switching point at which the growth behavior of c is flipped. c quickly reaches 1 when u is higher than threshold (Tr). The symbol Ω is the 2D real space, and xc denotes a point in Ω, where c is larger than 0.5. Ωc, which is the region of the cell in Ω, and is defined by using xc as follows:
(γ is a rate constant to rescale time and space) [26]. R is the distance between core and the plasma membrane, and the cell compartment is represented by collective circular domains around the core with radius R (Figure 2C). We found that R = 0.004 realized the finest resolution of patterns, so we used this value of R throughout this study (see the section “R value” for details). Describing the cell growth as a rapid transition between bistable states is reminiscent of a way to solve moving boundary problems in phase-field models [27]. A difference between these models and ours is whether diffusion of the phase field is incorporated or not; a diffusion term does not appear in Equation 1c, because diffusion of the cell state is biologically unrealistic in this case.
f(u,v) and g(u,v) represent chemical reaction terms, where the partial derivatives satisfy the following conditions:
> 0 (autocatalytic production of the activator),
< 0 (inhibition of synthesis of the activator by the suppressor),
> 0 (production of the suppressor by the activator) and
< 0 (concentration-dependent degradation of the suppressor). We used the following formulas for f and g:
We assumed that the receptor is uniformly distributed over the dendritic surface and that the strength of the signaling follows the local concentration of the suppressor. We adopted the 0-fixed boundary condition for the activator at the cell boundary:
We used the periodic boundary for the other variables v and c at the boundary of the 2D square space to model the real 2D space Ω in numerical simulation.
We numerically calculated the model given by Equations 1a–1c with reaction terms of Equations 2a and 2b (see Materials and Methods). Computer simulation showed that the cell compartment model could autonomously generate quite distinct dendritic patterns depending on the set of parameters employed (Figure 3). In each case where the model produced dendritic patterns, they were generated through repeated cycles of elongation and branching of dendrites (two examples are shown in Videos S1 and S3). With one set of parameters, smooth branches were formed, where neighboring branches aligned themselves nearly parallel to each other (Figure 3A). In such a cell, the distribution of the activator is continuous and mostly uniform, except for every branch terminal, where the density of the activator is relatively high (arrows in Figure 3B; Video S2). With a different set of parameters, the dendritic branches showed a more rugged morphology (Figure 3D). Stubby and non-aligned branches were formed, and the activator was distributed in a punctate manner in that cell (Figure 3E; Video S4). We call each punctum, where the activator was highly concentrated, a “spot.” Dendrites elongated by generating new spots (arrows in Figure 3E) and bifurcated when spots fissioned (arrowheads in Figure 3E). The suppressor was concentrated where the density of the activator was high, and it was distributed more broadly than the activator (Figure 3C and 3F). This distribution underlies long-range inhibitory interactions between neighboring dendrites. The interactions appeared to control whether or not dendrites would branch and in which direction dendrites would elongate. As a result, the branching frequency considerably varied among branchlets (compare yellow and blue arbors in Figure 3A and 3D), whereas the branch density was kept almost constant throughout the dendritic trees.
In a separately prepared manuscript, we addressed more biological issues such as tiling (Figure 1A and 1B) and regeneration in response to branch severing (Figure 1C). Branches of multiple neurons in our computer simulation, when they appeared in the same 2D space, avoided each other and accomplished tiling and isoneuronal avoidance. The neurons in our computer simulation were even able to reaccomplish tiling after local destruction of dendritic arbors exactly as Drosophila class IV da neurons do. Furthermore, modifications of our model enabled reproduction of a wide range of space-filling dendritic trees and even a non–space-filling type. Taken together, our model succeeded in qualitatively recapturing development of space-filling dendrites.
In the all cells examined, u and v exhibited a linear relationship at the growing tip of dendrite (Figure 4A for smooth branches and Figure 4B for rugged ones). Starting from u = 0 at the distal margin of dendrite, u should increase with time and it is observed as spatial change in u from distal to more-proximal parts of dendritic terminals. In contrast, the spatial change in v cannot be explained by reaction dynamics: for instance, in a case of Figure 4A, u and v should increase and decrease, respectively, according to vector field. Nevertheless, the supply of the suppressor from proximal dendrites via its diffusion seems to counteract actions of reaction functions, resulting in the increase of v in the proximal direction. Thus, most likely diffusion plays an essential role in determining the dynamics of the suppressor at dendritic tips.
As described below, we conducted numerical analysis to examine the generality of our cell compartment model and to determine the conditions for growth of dendrites that could be common to various types of neurons.
We calculated the cell compartment model by using different parameter sets of reactions between the activator and the suppressor, and searched for those by which dendritic patterns were successfully generated (Figure 5A). We defined a dendritic pattern by the following two conditions: first, cellular extensions bifurcated. Second, the density of dendrites was less than a criteria value. Typical examples of patterns violating either of these conditions are shown in Figure 5B–5D. This analysis clearly shows that dendritic patterns could be generated in a large parameter region (closed circles in Figure 5A), and so formation of dendritic patterns in our model does not appear to depend on particular parameter sets.
As explained before, our model produced two different types of patterns: the well-aligned smooth pattern, in which the activator is continuously distributed (Figure 3A) and the poorly aligned rugged pattern, in which punctate distribution of the activator is seen (Figure 3D). Those patterns shown in Figure 3 are two extreme examples; and intermediate patterns could be generated, depending on parameters employed. Interestingly, our numerical analysis revealed a correlation between Turing instability [23] and the distinctive shape of dendritic patterns. Turing instability, a widely applied theory of pattern formation, indicates an ability of chemical (in this case, activator–suppressor) interactions to develop spatially periodic patterns. The condition of chemical reaction dynamics for Turing instability was addressed by considering the two-variable (u and v) dynamics in the uncompartmentalized 2D space (designated as no compartment model, that is, a conventional RD model), and then by numerically calculating a parameter region for potential Turing instability in the no-compartment model (see Equations A3a–A3d in the section “Conditions for Turing diffusion-induced instability” and region I in Figure 5A) [26]. We used typical values for other parameters such as pb because changing the pb value did not significantly alter the shape or the size of region I (unpublished data). The results of this analysis clearly showed that relatively rugged patterns were obtained by using the condition that satisfied Turing instability (region I in Figure 5A); on the other hand, better-aligned patterns were obtained by using the condition that did not satisfy it (region II in Figure 5A). Therefore, it is suggested that the difference in two typical dendritic patterns obtained in our computer simulation stems from whether chemical dynamics in themselves are able to develop spatially periodic patterns or not.
Furthermore, we noticed that the shape of dendrites reflected the intracellular distribution of the activator. From bottom-left to top-right of the (pe − pa) space (Figure 5A), the dendrite morphology became smoother; and distribution of the activator changed from punctate in nature to more continuous (Figure 5E1-5E4). Continuity in the activator distribution seems to strongly depend on the shape of local branches (Figure 5E2). Even within the same cell, the local distribution of the activator was punctate in branch-rich regions (e.g., right-enclosed branches in Figure 5E2), whereas it was more continuous in branchless regions (e.g., left-enclosed branch in Figure 5E2). Co-existence of two distinctive types of distributions, punctuate and continuous, in a single cell suggests that these two types of distributions are locally stable structures.
The above-mentioned analysis also indicated that the condition for developing dendritic patterns did not entirely cover region I. In addition, it is of particular interest that spatially non-homogeneous dendritic patterns were generated in region II, in which homogeneous distribution at the steady state should be stable in the two-variable (u and v) dynamics (see Discussion for details). Most likely this discrepancy of conditions for pattern formation in the cell compartment model and the no compartment one originates from the structure of cell and the feedback between the chemical reaction and cell growth in the model.
We further examined the relationship between our model and the Turing system. In general, the Turing system develops dot, stripe, or reverse-dot patterns in the 2D space, depending on parameters (e.g., the distance from the equilibrium point to the upper limitation of activator [Amax]) [28]. So we explored whether or not the conditions for dendritic pattern formation were related to the property of the Turing system to generate either a dot, stripe, or reverse-dot pattern.
By changing the upper limitation of activator (Amax) in the no-compartment model, we drew a phase diagram, in which each dot, stripe, and reverse-dot pattern was mapped to a different parameter region (Figure 6A). Subsequently we searched for parameter sets that developed dendritic patterns in the cell compartment model (circles in Figure 6A); and the results of this analysis indicated that dendritic patterns were obtained mostly in the dot domain (D in Figure 6A). Therefore the punctate distribution of the activator in rugged dendrites (Figure 3E) can be interpreted as the typical dot pattern of the conventional RD system being generated inside of the cell compartment. Dendritic patterns were not obtained in most of the stripe or reverse-dot domains (S or R in Figure 6A). Computer simulation with parameter settings in the stripe or reverse-dot domains generated patterns, which did not resemble the shape of dendritic arbors of real neurons (Figure 6B–6E). If conditions for Turing instability were not satisfied, dendritic pattern was produced in a parameter region adjacent to the dot domain. These results are consistent with an intuitive understanding of the process of dendritic pattern formation; that is, dendrites grow in pursuit of a track of locally activated molecular complexes for branching. In this sense, a punctate or terminally dense distribution of activator is favored, whereas the stripe or reverse-dot one is not.
It is worth evaluating whether the results of this study are specific to a particular dynamics or if they represent more general properties of the RD system. For that purpose, we tested several different forms of reaction terms and one of them was as given below:
Parameter settings for potential Turing instability in the linear dynamics described by Equations 3a and 3b were determined and plotted (region I in Figure 7A) as in Figure 5A.
Parameter dependency of dendritic pattern formation was examined, and we found that dendritic patterns were generated in both outside and inside of region I (Figure 7B and 7D, respectively). Therefore, classical Turing conditions were not necessary or sufficient for dendritic pattern formation in this linear dynamics, either. Furthermore, whether the function was linear or non-linear, the activator distribution well-correlated with the shape of branches (Figure 7C and 7E; compare them to Figure 5E); and dendritic patterns were generated preferentially in the dot domain, but not in the stripe or reverse-dot domain (unpublished data). Collectively, all of the results suggest that a wide range of parameter settings and different dynamics of chemical reactants allow development of dendritic patterns in the cell compartment model.
Finally we found that our cell compartment model provides a prediction for future experiments. As described before, the numerical simulation of the model unraveled a strong correlation between shapes of dendrite and distributions of the activator (Figure 5E and Figure 7E). We noticed that dendritic trees of some real neurons were reminiscent of those of the smooth type in our computer simulation (Figure 8A and 8B) and that terminal branches of some other real neurons were less aligned (Figure 8C and 8D). Accordingly, if the developmental machinery proposed by this study is actually functioning in vivo, the intracellular distribution of the hypothetical activator could be predicted on the basis of the morphological features of dendrites. More specifically, the distribution of the activator may be terminally dense in neurons of the smooth type and punctate in the rugged type (for instance, Figure 8A and 8B and Figure 8C and 8D, respectively).
To support the validity of our prediction, we set a quantitative measure called a “dispersion of orientation of branches” (DOB) to characterize dendrite morphology. DOB is the coefficient of variation of directions of branch segments in each local region of dendritic trees (Figure 9 and Materials and Methods); hence the smaller is the DOB, the better-aligned are the local branches. Quantification of the DOB for the smooth and rugged types of the obtained patterns in our computer simulation confirmed that it was significantly smaller in the former type (double asterisks in Figure 8E). We next quantified the DOB for real neurons and found that values for the smooth type (Figure 8A and 8B) were significantly smaller than those for the less-aligned type (Figure 8C and 8D; asterisks in Figure 8E). These results suggest that geometry of real neurons may also be characterized by DOB and that we can use DOB as a morphological measure for predicting the intracellular distribution of the activator in vivo.
In this study, we developed the first mathematical model that sheds light on autonomous pattern formation of neuronal dendrites. The cell compartment model, which is based on the experimentally verified dendro-dendritic interaction, autonomously develops dendritic elongation and branching. It should be noted that dendritic patterns are defined not only by the numerical parameters such as the terminal number, but also by other properties such as mutual avoidance. Our model places emphasis on the latter aspects of the space-filling dendrites, which are difficult to characterize by quantitative measures, and indeed qualitatively recaptures developmental regulation of the space-filling dendritic patterns. Collectively, we believe that this study offers a new concept in developmental biology, a self-organizing mechanism in neuronal dendrite pattern formation.
Many of the previous models assumed that elongation and branching of dendrites are controlled by probability functions, in which each parameter separately codes individual growth rules such as degree- or segment length- dependent rate of elongation and/or branching [19,20]. In contrast, dendritic patterns are autonomously generated without embedding different parameters to control each branching frequency, branch angle, and self-avoidance of dendrites in our model. Considering that we are presently far from understanding the entirety of the molecular mechanisms of chemical reactions occurring in vivo, the high performance of the proposed system obtained with diverse forms of reaction function takes on significance, because it may support a future application of the model to the dendritogenesis of a whole variety of real neurons.
Our numerical analysis showed that generation of dot patterns of the activator in rugged dendrites could be attributed to a property of chemical dynamics, which is supported by Turing instability. On the other hand, classical Turing diffusion-induced instability alone cannot give us a comprehensive explanation of the pattern formation in our model, because dendritic patterns were successfully developed even when the spatially homogeneous pattern at the steady state of chemical reaction dynamics was stable. We think that the compartmentalized structure in our model may increase instability of the dynamics of the cell growth. Actually, it was shown both in experiments and in computer simulation that a straight interface could become unstable to make complex spatial patterns in certain bistable dynamics [27,29]. Hence, analyzing the model based on the idea of front instability may be one way to understand the behavior of our model. From a viewpoint of experimental biology, these results suggest that simultaneous, high-resolution imaging analyses on molecular interactions and plasma membrane dynamics would be informative.
We introduced new criteria to categorize patterns of dendrites in real neurons and to predict the intracellular distribution of potential molecular complexes for dendrite growth. Two distinctive dendritic patterns were found in both computer-simulated and real neurons, and it is suggested that the distribution of the activator is characteristic of the shape of branches. Further advances in our understanding of the molecular mechanisms involved in dendrite development are required to address whether the prediction from our cell compartment model is valid or not. Yet, there are a couple of interesting observations that may indicate periodicity in dendrites of real neurons. For instance, Golgi apparatus is distributed in a punctate manner in da neurons and pyramidal neurons [30–32]; and its localizations at branch points are important for branch formation. In addition, staining for microtubule-associated protein 2 in the absence of detergents reveals that regions of high signal intensity are found in a spatially periodic manner along dendrites and that dendritic branch points are preferentially associated with these regions [33]. It would be interesting to review these observations in the perspective of our model.
Our cell compartment model is a simplified version of dendrite growth in vivo, and new elements can be installed depending on needs or researchers' interests. For instance, although generated patterns in the present model are highly homogeneous, less homogeneous patterns could be obtained if stochastic aspects or noise are strengthened (for example, by fluctuating Tr along dendritic branches). It is also interesting to extend our model to include activity-dependent processes, such as synaptotropic dendrite growth [34,35] and refinement of pre-existing branches during late stages of development [36,37]. Furthermore, we are now trying to reproduce development of non–space-filling type dendrites, which are anisotropic in terms of the direction of elongation and inhomogeneous in terms of coverage of a field, by incorporating a guidance mechanism and/or an RD system of intracellular activator and suppressor. Although we should bear in mind that overlaying these additional features could modify the properties of the system, we hope that combination of biochemical experiments with enlarged editions of this mathematical model may clarify the comprehensive logic underlying neuronal dendrite development.
Colony formation by Bacillus subtilis is a well-known example of dendritic patterning in biology. Bacillus subtilis generates distinctive colony patterns depending on the substrate softness and nutrient concentration [38], and formation of most of the colony patterns was well-reproduced by RD models [39,40] and a cell automaton model [41]. Similarity between neuronal dendrites and bacteria colonies is found not only in terms of their morphology, but also with respect to repulsive behaviors; i.e., when two colonies are in close proximity, they avoid each other just as do space-filling neurons [42]. In addition, interesting parallels can be also found between dendrite development and other branching morphogenesis such as coral [43], vertebrate lung [44], and trachea of Drosophila [45]. These systems accomplish physiological functions that can be regarded as similar to space-filling dendrites. For instance, trachea must elaborate its branches to deliver oxygen to the whole body. Mathematical models for these pattern formations have been proposed [43,44,46,47], and it is suggested that branching morphogenesis in general can be understood as the following: a part of the structure that happens to sprout due to some fluctuation locally speeds up its growth and eventually develops a visible branch. We observed a similar behavior of dendrites in our computer simulation. Furthermore, recent works revealed the molecular basis of lateral inhibition between the neighboring lung epithelium and between growing tips of trachea that may correspond to long-range inhibitory dendro-dendritic interactions in the development of space-filling dendrites [44–46]. Therefore, our model on neurite formation would be potentially informative in understanding the above-mentioned branching morphogenesis.
Despite the afore-mentioned similarities, there is one big difference between bacteria colony models and ours. The former relies on non-linearity in diffusion and reaction function for pattern formation [39]. On the other hands, dendritc growth in our model does not require such non-linearity (Figure 7). It might be that unambiguous boundary of the cell in our model plays an equivalent role to non-linear diffusion terms in bacteria colony models. Taking advantage of the fewer constraints in chemical dynamics in our model, we addressed the relationship between Turing instability and biological branching morphogenesis. Other branching morphogenesis might obey the conditions that were clarified in this study. Again, generality of the proposed mechanism would be significant for testing this possibility in other systems of interest.
To calculate the model, we used the finite difference method, a simple explicit scheme. The simulation starts from a small cell body. The initial value of the activator is 0.5± small random deviations in each position inside of the cell body and 0 in other places, whereas the value of the suppressor is 0.1± small random deviations in the cell body and 0 otherwise. Changes in initial conditions of the activator or the suppressor affected the results only slightly. Small noise was added to the diffusion coefficient of the activator in every calculation step to cancel the anisotropy of the grid in numerical simulation.
Image processing and measurement were done with ImageJ. First, we superimposed a square on individual dendritic trees (those in Figure 3A and 3D and Figure 8A–8D; see also Figure 9). The size of each square was normalized to that of the entire dendritic tree (the size of the tree was defined as that of a polygon connecting dendritic tips). As for the obtained patterns in computer simulation (those in Figure 3A and 3D), we skeletonized them and sampled four pairs of squares that were located at the same coordinates (“1” and “2” in Figure 9). Each branch segment was approximated by a line segment connecting two edges of the branch segment (“3” in Figure 9). We measured the angle of the line segment with respect to the horizontal direction, repeated measurement for all segments in each small square, and calculated the coefficient of variation, which we called the DOB. Average values of DOB for each dendritic tree are shown with means ± SD in Figure 8E.
Imaging and single cell labeling of Drosophila sensory neurons were done as described [11,13,48]. Strains used were NP7028 UAS-mCD8::GFP [11], ppk-GAL4 UAS-mCD8::GFP [49], elav-GAL4 UAS-mCD8::GFP hsFLP, tub-Gal80 FRT40A, and FRT40A [13].
Original equations of the activator-suppressor model were as follows:
where u and v are the concentration of the activator and that of the suppressor, respectively. du and dv are diffusion coefficients. Original chemical reaction terms were:
We non-dimensionalized Equations 4a–4c and Equations 5a–5b to obtain Equations 1a–1c and Equations 2a–2b.
The value of R determines the thickness of the branches as expected. Smaller R resulted in thinner branches, thus finer patterns. However, there seems to be a minimum value of R to support dendrite growth. The minimum value may be necessary to produce a new spot of the activator, which is separated from the pre-existing spot, in the vicinity of the cell boundary. We confirmed that the minimum value of R was independent of the spatial grid size in numerical simulation, and thus the above results are not an artifact of numerical simulation. So we used R = 0.004, which gave the finest dendritic patterns (R = 0.0041 yielded nearly equal results to those obtained with R = 0.004).
Conditions for Turing diffusion-induced instability [23] are the following:
where the partial derivatives of f and g are evaluated at the steady state (u0,v0) which satisfies f(u0,v0) = 0 and g(u0,v0) = 0 [26]. Equations 6a and 6b describe conditions for a stable equilibrium point in the absence of diffusion. Equations 6c and 6d describe conditions for an unstable periodic solution in the presence of diffusion.
|
10.1371/journal.pntd.0006620 | Dynamic changes in human-gut microbiome in relation to a placebo-controlled anthelminthic trial in Indonesia | Microbiome studies suggest the presence of an interaction between the human gut microbiome and soil-transmitted helminth. Upon deworming, a complex interaction between the anthelminthic drug, helminths and microbiome composition might occur. To dissect this, we analyse the changes that take place in the gut bacteria profiles in samples from a double blind placebo controlled trial conducted in an area endemic for soil transmitted helminths in Indonesia.
Either placebo or albendazole were given every three months for a period of one and a half years. Helminth infection was assessed before and at 3 months after the last treatment round. In 150 subjects, the bacteria were profiled using the 454 pyrosequencing. Statistical analysis was performed cross-sectionally at pre-treatment to assess the effect of infection, and at post-treatment to determine the effect of infection and treatment on microbiome composition using the Dirichlet-multinomial regression model.
At a phylum level, at pre-treatment, no difference was seen in microbiome composition in terms of relative abundance between helminth-infected and uninfected subjects and at post-treatment, no differences were found in microbiome composition between albendazole and placebo group. However, in subjects who remained infected, there was a significant difference in the microbiome composition of those who had received albendazole and placebo. This difference was largely attributed to alteration of Bacteroidetes. Albendazole was more effective against Ascaris lumbricoides and hookworms but not against Trichuris trichiura, thus in those who remained infected after receiving albendazole, the helminth composition was dominated by T. trichiura.
We found that overall, albendazole does not affect the microbiome composition. However, there is an interaction between treatment and helminths as in subjects who received albendazole and remained infected there was a significant alteration in Bacteroidetes. This helminth-albendazole interaction needs to be studied further to fully grasp the complexity of the effect of deworming on the microbiome.
ISRCTN Registy, ISRCTN83830814.
| Studying the relationship between soil-transmitted helminthiasis and gut microbiota is becoming more important as both have been implicated in modulating immune system in various inflammatory diseases. However, findings of previous studies of the effect of helminth on the microbiome are inconsistent. In this study, an optimal design, a placebo-controlled anthelminthic trial was conducted to dissect the effect of helminths and anthelminthic treatment on gut microbial profile. In addition, a novel statistical model was used to analyse the association by taking into account the correlation structures between bacterial categories by applying multivariate analysis whereby the multiple testing correction is not needed.
| Shortly after birth, the human body is colonized by a community of bacteria [1, 2] with relatively simple composition which increase in number and complexity with age [3]. The densest colonization with commensal microbes of the human body is found in the intestine [4] which has a beneficial impact on gastro-intestinal function and host health by providing support for host metabolism, protection against pathogenic microbes, integrity of intestinal mucosa, and modulation of the immune system [2, 3, 5]. Furthermore, it has been shown that intestinal microbiota is associated with dietary habits [6, 7], physiological factors such as age, gender and BMI [8, 9] as well as diseases, such as inflammatory bowel disease and obesity [1, 5, 10].
Apart from intestinal microbiota, certain pathogens such as soil-transmitted helminths (STH) may coexist in the human intestine. It is estimated that STH, largely represented by Ascaris lumbricoides, hookworm such as Necator americanus and Ancylostoma duodenale, and whipworm Trichuris trichiura, infect 2 billion people in the majority of developing countries and mostly children [1, 11]. These infections have been reported to cause impairments in physical, intellectual, and cognitive development [12]. At the same time, these parasitic worms have a long co-evolutionary interaction with their host. The result of this co-evolutionary trajectory, seems to be that helminths lead to immune regulatory responses that allow their long term survival within their host [13, 14]. Since intestinal microbiota and helminths share the same niche in their host, it is hypothesized that the presence or absence of intestinal helminths may affect their interaction with each other within the host. In an interesting study, evidence was provided for the beneficial effects of the microbiome on successful completion of whipworm life cycle [15]. Currently, there is also much interest to determine whether helminth infections affect the gut microbiome and whether the effects of worms on human health is mediated via alteration in the microbiome composition. It is becoming increasingly clear that the gut microbiota has important link to the immune system and several disease outcomes. With the mass drug administration programs underway to eliminate intestinal helminths in many endemic regions, it is essential to fully understand the consequences of deworming on community health by characterizing the effect on the gut microbial composition.
Recently, several studies investigated the relationship between the intestinal microbiome and intestinal helminth infections. In swines, a statistically significant association between Trichuris infection and the gut microbiome composition was shown [16, 17], evident from the altered abundance of the genus Paraprevotella and phylum Deferribacteres in the infected pigs. The chronic infection of Trichuris muis in C57BL/6 wild-type mice increased the relative abundance of Lactobacilli [18], while giving T. trichiura ova to macaques with chronic diarrhea increased the phylum Tenericutes and resulted in clinical improvement [19]. Therefore, in animal models, Trichuris infection seems to be associated with alternation in the gut microbiome. However, in humans, findings are not consistent. In an observational study in Ecuador, comparing the gut microbiome of infected and uninfected school children, no significant differences at various taxonomical levels were found [20]. On the contrary, two other observational studies in rural villages of Malaysia [21] and Zimbabwe [22] found a significant increase in diversity and abundance of certain bacteria taxa in infected compared to uninfected subjects. An increase in Paraprevotellaceae was seen in the Malaysian study, which seemed to be associated with Trichuris infection while an increase in Prevotella was reported in the study in Zimbabwe that was attributed to S. haematobium infection. Furthermore, in an interventional study carried out in another rural village in Malaysia [23], a significant change in order Bacteroidales and Clostridiales was observed after deworming while deworming of S. haematobium in an interventional study in Zimbabwe [22] did not seem to alter the microbiome.
The study designs which were used to investigate the human-gut microbiome in relation to helminth infections were either observational [20–22] or interventional without a control group [20–23] hampering the estimation of the true relationship between helminth infection and the microbiome composition. Motivated by the findings from previous studies of helminths on microbiome, we used samples from a larger randomized placebo-controlled trial of albendazole treatment in a population living in an area endemic for soil transmitted helminth infections [24] to further characterize the effect of helminth infection and treatment at before and 21 months after treatment. The study design allowed the investigation of the effect of helminths on the fecal microbial community through comparing helminth infected and uninfected at baseline and subsequently assessing the effect of treatment with albendazole. We also explored the effect of the interaction between treatment and infection status on the faecal microbiome. In addition, we used the opportunity to assess whether albendazole has a direct effect on the microbiome by analyzing those who received albendazole and were uninfected throughout the study. The placebo group enables the estimation of the effect of deworming on the microbiome composition in the absence of anthelminthic treatment which itself could affect the microbiome.
The analyses carried out in this study aim to characterize the joint effects of several predictors, such as helminth infection and treatment on each bacterial category. For comparing the gut microbiome of premature infants with different severities of necrotizing enterocolitis, a Dirichlet—multinomial model was used [25]. Here, we consider the same approach for modelling and hypothesis testing for the association between treatment and helminth infection on microbial composition at the phylum level. Our approach addresses the possible correlation between bacteria categories, the compositional feature of the microbiome data [26], and the multiple testing issue.
This study was nested within the ImmunoSPIN study, a double blind placebo-controlled trial conducted in Flores Island, Indonesia [24]. The ImmunoSPIN study has been approved by the Ethical Committee of Faculty of Medicine, Universitas Indonesia, ref:194/PT02.FK/Etik/2006 and has been filed by ethics committee of the Leiden University Medical Center. The clinical trial was registered with number: ISRCTN83830814 in which the protocol for the trial and supporting CONSORT checklist are available elsewhere [27]. The subjects gave their informed consent either by written signature or thumb print. Parental consent was obtained for children below 15 years old.
Households were randomized to receive either a single dose of 400 mg albendazole or placebo once every 3 months for 2 years. To assess the effect of treatment on the prevalence of soil transmitted helminth infection, yearly stool samples were collected on a voluntary basis. T. trichiura infection was detected by microscopy and a multiplex real time PCR was used for detection of hookworm (A. duodenale, N. americanus), A. lumbricoides and Strongyloides stercoralis DNA. For the current study, paired DNA samples before and at 21 months after treatment from 150 inhabitants in Nangapanda were selected based on the treatment allocation and infection status as well as the availability of complete stool data at pre and post-treatment (Fig 1). The procedure for sample collection and processing is already described elsewhere [24].
Briefly, prior to DNA isolations, approximately 100 mg unpreserved faeces (kept at -20°C) were suspended in 200μl PBS containing 2% polyvinylpolypyrolidone (PVPP;Sigma, Steinheim, Germany). Suspensions were heated at 100°C for 10 min and were treated subsequently with sodium dodecylsulphate-proteinase K at 55°C for 2 h. DNA was isolated using QIAamp DNeasy Tissue Kit spin columns (QIAgen, Venlo, The Netherlands). The whole procedure of DNA isolations and setup of PCR plates were performed using a custom-made automatic liquid handling station (Hamilton, Bonaduz, Switzerland).
As published already, sequences of the A. lumbricoides and N. americanus-specific primers and probes as well as the A. duodenale specific XS-probes were used to accommodate the specific fluorophor combinations of the CFX real-time PCR system (S1 Table) [24, 28]. The real-time PCRs were optimized first as monoplex assays with 10-fold dilution series of A. duodenale, N. americanus and A. lumbricoides DNA, respectively. The monoplex realtime PCRs were thereafter compared with the multiplex PCR with the PhHV internal control. The cycle threshold (Ct) values obtained from testing the dilution series of each pathogen in both the individual assay and the multiplex assay were similar, and the same analytical sensitivity was achieved.
Amplification reactions were performed in white PCR plates in a volume of 25μl with PCR buffer. Amplification consisted of 15 min at 95°C followed by 50 cycles of 15 s at 95°C, 30 s at 60°C, and 30 s at 72°C. Amplification, detection, and analysis were performed with the CFX real-time detection system (Bio-Rad laboratories). The PCR output from this system consists of a cyclethreshold (Ct) value, representing the amplification cycle in which the level of fluorescent signal exceeds the background fluorescence and reflecting the parasite-specific DNA load in the sample tested. In this manuscript, we set the ct value 30 as a threshold for the infection status i.e. subjects with PCR lower than 30 was identified as clearly infected and PCR above 30 as uninfected or very low infection. The analyses were carried with regard to the infection status and we do not consider the analysis in the level of infection.
Genomic DNA samples were isolated from 100 mg of fresh stool, which were also used for detection of helminth infection by real time PCR. The DNA amplification and pyrosequencing followed the protocols developed by the Human Microbiome Project (HMP) [29] at the McDonnell Genome Institute, Washington University School of Medicine in St. Louis. Briefly, The V1-V3 hypervariable region of the 16S rRNA gene was amplified by PCR and the PCR products were purified and sequenced on the Genome Sequencer Titanium FLX (Roche Diagnostics, Indianapolis, Indiana), generating on average 6,000 reads per sample. The filtering and analytical processing of 16S rRNA data for this cohort has been previously described in details [30]. The assembled contigs count data as a result of RDP classification was organized in matrix format with taxa in columns and subjects in row. The entries in the table represent the number of reads for each phyla for each subject. Rarefaction to 2000 reads was performed using an R package (vegan) [31]. We obtained the count data of 609 bacterial genera and 18 bacteria phyla. In the analysis at phylum level, we retained the 5 most prevalent phyla (Actinobacteria, Bacteroidetes, Firmicutes, Proteobacteria, and Unclassified Bacteria) and pooled the remaining phyla into a pooled category such that there are only 6 phyla categories. The Unclassified bacteria represents the category where all the sequences cannot be assigned into a phylum. We conducted further analyses by decomposing the statistically significant phylum (Bacteroidetes) into the two most prevalent genera (Bacteroides and Prevotella) and the remaining genera into a pooled Bacteroidetes category and combining the Proteobacteria, Unclassified Bacteria and Pooled in a Pooled Phyla category. In total we have six categories since we also selected Actinobacteria and Firmicutes at phylum level.
The within sample diversity (Shannon and richness diversity) indices as well as the between sample diversity (Bray-Curtis distance) were computed at baseline and follow-up using the dataset at genera level. Clustering of samples and bacteria was studied by plotting a heat map of bacteria genera which were present in at least one sample and which had an average relative abundance of more than 1%. This cutoff was chosen to exclude rare genera. Unless stated otherwise, the rest of the analyses were done at the phylum level. A Pearson’s chi-squared test statistic was used to test for differences of infection prevalence between the two treatment groups at pre and at post-treatment.
Although the study design allows for the pairwise analysis, unfortunately no method is available for multivariate categorical count data. For this reason, we used the Dirichlet-multinomial regression where the characterization of infection and treatment are similar to the interpretation in loglinear model. Each count outcome within a category was assumed to follow the negative binomial distribution. This distribution is the result of a Poisson distribution for counts with the additional assumption that the underlying parameter is a random variable which follows the conjugate distribution (Gamma). By assuming that the underlying parameter was random, the presence of overdispersion due to multiple counts observed within a sample was modelled. To incorporate the fact that the total count is fixed per sample, we conditioned the probability of the multivariate count outcome on the total count per sample. This model is equivalent to the approach of Guimaraes and Lindrooth [32], i.e. the Dirichlet-multinomial regression model. The model parameters are log of odds ratios which compare the prevalence rate of each bacteria phyla associated with the covariates with the reference category. In all analyses, Firmicutes was used as reference since it has the highest abundance among the phyla. The covariates were infection status and treatment allocation which are both binary variables.
The likelihood ratio statistic was used to test the null hypothesis of no effect of the covariate on the microbiome composition. The test statistic follows asymptotically a χ2 distribution with J degrees of freedom, representing the J − 1 bacterial comparison with the reference and one overdispersion parameter.
As the Dirichlet—multinomial regression is available for cross-sectional setting, we modelled the association between microbiome composition and covariates including treatment at 21 months after treatment. First, we modelled the association between treatment and microbiome composition by including all study participants. Next, we selected subjects who were infected with at least one single helminth at baseline and included a categorical variable representing the four combinations of treatment allocation and infection status at post-treatment in the model. The R package MGLM [33] was used for analyses. The results were reported in terms of odds ratios, 95% confidence intervals and p-values.
To confirm our finding with this method, we used the univariate pairwise analysis for single bacterial categories of interest in albendazole arm. For this purpose, the inverted beta binomial test was applied to test the null hypothesis that the relative abundance of certain bacteria category at pre-treatment is similar to the relative abundance at post-treatment. Note that the inverted beta-binomial regression model is only defined for two categories and is equivalent to the Dirichlet-multinomial. The R package ibb [34, 35] was used for this test. All computations were conducted in R version 3.1.0 [36].
At baseline, 94 out of 150 (62.7%) individuals were infected with one or more helminth species, and hookworm was the most dominant species (52.1%) followed by T. trichiura (44.7%) and A. lumbricoides (37.2%). The baseline characteristics such as age, gender, and helminth prevalence were similar between the two treatment arms although the prevalence of N. americanus was slightly higher in albendazole group, but not statistically significant (Table 1). The additional relevant characteristics of the participants are listed in S2 Table. With regard to the microbiome composition, the proportions of each bacterial phyla were also similar between two treatment arms with the highest abundance at the phylum level being Firmicutes followed by Actinobacteria, Proteobacteria and Bacteroidetes.
At 21 months after treatment, the prevalence of STH infection was 21.7% in the albendazole arm and 54.3% in placebo arm (p-value < 0.001). Albendazole had the greatest effect on hookworm (24.7% (placebo) vs 4.3% (albendazole)) followed by A. lumbricoides (28.4% (placebo) vs 4.3% (albendazole)) and lastly T. trichiura (28.4% (placebo) vs 15.9% (albendazole)). These percentages are similar to what was seen in the whole ImmunoSPIN trial [24]. These data show that while infections with A. lumbricoides and with hookworms decrease at post-treatment, the infections with T. trichiura was not affected much by albendazole and therefore the proportion of individuals infected with T. trichiura increased when considering those that remained infected at post-treatment (Fig 2). In the placebo group, there was no such difference in the composition of helminth species at post-treatment. It was noted that 12 (2 from albendazole and 10 from placebo) out of 56 uninfected subjects at baseline (21.4%) gained helminth infection over the study time period.
Using bacterial data at the genus level (a total of 609 genera), we calculated the within sample diversity (richness and Shannon index) and between sample diversity (Bray-Curtis dissimilarity). We observed a similar within-sample diversity at pre and post-treatment as evident from the Shannon diversity index (2.99 vs 2.96) and the richness index (66.17 vs 62.16). The Bray-Curtis dissimilarity measures the percentage of similarities between two samples in a community and the values range from 0 (completely similar) to 1 (completely dissimilar). As reported earlier [30], the Bray-Curtis dissimilarities calculated from 150 subjects at pre-treatment was 0.61 and the same average was obtained when calculating the Bray-Curtis dissimilarities at post-treatment, indicating that in average there was 61% dissimilatory percentages between each pairs of samples. When stratifying all samples based on infection status at pre-treatment and on randomization arm at post-treatment, again we observed similar beta-diversities, indicating that neither infection nor treatment induced a shift in diversity. When analyzing the genera in relation to infection status rather than treatment, the average Shannon diversity index as well as the average richness was similar between the infected and the uninfected group at pre-treatment and post-treatment (S1 Fig).
The average relative abundances of all bacterial genera at both time-points were below 10%, with the highest being in the phylum Firmicutes, specifically the genus Catenibacterium (6.7% at pre-treatment) and the unclassified genus belonging to the family Ruminococcaceae (5.6% at post-treatment). The relative abundance at the genus level as well as the dominant genera vary between populations as observed in studies where samples in rural Ecuador [20] or Malaysia were compared with the US [21] or in studies where samples of healthy European and American adults were analysed [37]. To illustrate the bacterial genera profile in relation to infection and treatment status, we selected the 29 genera (at pre and post-treatment) with an average of relative abundance across all samples larger than 1%. Genera from phylum Firmicutes are the most dominant (21 of 29 genera belongs to Firmicutes). As shown in heatmaps based on composition of the most prevalent genera, no significant clustering could be seen, neither at the level of bacteria nor at the level of individuals (Fig 3A and 3B) in relation to helminth infection or treatment, which indicates that neither helminths nor treatment affected the predominant genera in the gut.
Using the Dirichlet-multinomial regression model, we observed that there was no difference on the microbiome composition at the phylum level when subjects with any helminth infection were compared with uninfected ones either at pre (Fig 4A) or at post treatment (Fig 4B) time points. The same was the case when infection with a specific helminth species was considered (Fig 4A and 4B).
The Dirichlet-multiomial regression model was also used to discern the effect of helminths and treatment on the microbiome data at post treatment. Six bacterial categories were considered in the analyses with Firmicutes used as a reference. The effect of treatment on microbiome composition in all individuals irrespective of whether they were infected or not at post-treatment was not significant. No differences were observed between placebo and albendazole at post-treatment (p-value = 0.305, Table 2A, likelihood ratio test).
We further selected subjects who were infected at baseline (N = 94) and characterized their microbiome composition at post-treatment with regard to their infection status and treatment arm, namely: subjects who lost their infection either in the albendazole (group 1, N = 34) or placebo arm (group 2, N = 13), and subjects who remained infected in either the albendazole (group 3,N = 13) or placebo arm (group 4, N = 34). We compared the microbiome composition of the first three groups to the group of remained infected in the placebo arm (group 4) as the latter group were neither influenced by treatment nor the changing of infection status. When subjects who were infected at pre-treatment and lost their infection in the albendazole arm were compared to subjects who remained-infected in placebo group, no differences were observed (p-value of 0.371, Table 2B), indicating that removing helminths with albendazole did not change the microbiome profile at a phylum level. Furthermore, in subjects who lost their infection in the placebo arm, there was a trend for decrease in Bacteroidetes and pooled category (OR 0.49, 95% CI:(0.27,0.91) and OR 0.47, 95% CI:(0.23,0.96), respectively, Table 2B), moreover, the whole composition in this group did not differ significantly from that in the group of remained infected in the placebo arm (p-value of 0.069). These two comparisons suggest that removing helminths regardless of treatment did not alter the microbiome composition when analysed at a phylum level. Interestingly, the comparison of microbiome composition between subjects who remained infected in the albendazole group was significantly different from the microbial composition in subjects who remained infected in the placebo group (p-value of 0.004,Table 2B). This difference was driven by the increasing odds of having Actinobacteria (OR 1.57, 95% CI of (1.05, 2.35)) and the decreasing odds of having Bacteroidetes (OR 0.35, 95% CI: (0.18,0.70)). To further analyse the direct treatment effect without the influence of helminth infection, we selected subjects who were uninfected at baseline and remained-uninfected at post-treatment (N = 44). For these subjects, we compared the microbial composition at post-treatment of subjects who received albendazole versus those who received placebo. No difference was observed (the estimate odds ratios range from 0.88, 95% CI: (0.56, 1.39) to 1.42, 95% CI: (0.88, 2.29), p-value = 0.666, illustrated in Fig 5), indicating that albendazole alone does not seem to affect the microbiome composition in uninfected subjects when compared at a phylum level.
As neither treatment alone nor the infection affected the microbial composition, we further hypothesized that the significant difference in microbiome composition in subjects who remained infected and received albendazole compared to the group that remained infected in the placebo arm was caused by the alteration of the abundance of Actinobacteria and Bacteroidetes during the treatment period. To test this hypothesis, we used the inverted beta-binomial test to compare the relative abundance of Actinobacteria and Bacteroidetes in subjects who remained infected in albendazole group at pre-treatment to the relative abundances of these bacterial phyla at post-treatment. While the relative abundance of Actinobacteria did not change significantly between pre and post-treatment (p-value of 0.155, inverted beta binomial test), the relative abundance of Bacteroidetes was estimated to be 1.88 fold higher at pre-treatment compared to post-treatment (p-value of 0.012, inverted beta binomial test). This result indicates that there is a complex interaction between helminths and treatment, which induces a change in bacterial composition during the treatment period. Using the same analysis, the direct effect of albendazole was assessed by comparing subjects who were uninfected but received albendazole at pre treatment and remained uninfected at post treatment. Although some differences were seen in the microbiome composition between pre and post-treatment, specifically in the phyla Actinobacteria, Bacteroidetes and Proteobacteria, these differences were not statistically significant (p-values of 0.149, 0.267 and 0.064, respectively). This is in line with the finding when we used the Dirichlet-multinomial regression model where no direct effect of albendazole on the microbiome composition was found. In addition, similar microbiome composition was seen in subjects free of helminth infection at baseline who received placebo and remained uninfected at post-treatment, which suggests that the microbiome was stable over time.
In the Dirichlet—multinomial regression analysis carried out at the phylum level, Bacteroidetes was the phyla that showed significant differences in subjects who remained infected in the albendazole arm compared to those who remained infected in the placebo arm. We dissected this further to assess which Bacteroidetes genera accounted for this difference using the Dirichlet-multinomial regression model on 6 bacterial categories which were obtained as follows. The phylum Bacteroidetes was divided into three categories, namely the Bacteroides, Prevotella and pooled Bacteroidetes. The first two genera were chosen as they were the two most abundant in the phylum Bacteroidetes. In the analyses, as 6 categories are needed, we included another three phyla, i.e., Actinobacteria, Firmicutes and pooled remaining phyla (pooled Phyla). As for the modelling at the phylum level, Firmicutes was used as a reference. Similar to the analyses at the phylum level, we characterized the association of infection and treatment on these 6 bacterial categories that comprised the genera belonging to Bacteroidetes.
When considering the whole study subjects irrespective of infection status, there was no difference between albendazole and placebo (Table 3A). When 94 infected subjects at pre-treatment were selected and 6 bacterial categories as above were analysed with regard to infection and treatment, we observed a decrease in odds of having Prevotella in subjects who lost their helminth infection in placebo group (OR 0.44, 95% CI: (0.21,0.90)) compared to subjects who remained infected in placebo group although this fell short of statistically significant (p-value of 0.086, Table 3B). Furthermore, in line with the finding at the phylum level, we also observed a significant difference in microbial composition of subjects who remained infected with albendazole compared to the microbial composition of subjects who remained infected in the placebo group (p-value of 0.016). This alteration was mainly due to the increase in odds of having Actinobacteria (OR 1.54, 95% CI: (1.00, 2.35)) and a decrease in odds of having Prevotella (OR 0.44, 95% CI: (0.21, 0.94)), suggesting that the decrease in Bacteroidetes at the phylum level observed in Table 2B was driven by Prevotella.
There are two unique aspects to the current study on the effect of helminths on the gut microbiome in subjects living in rural areas of Indonesia, namely the combination of the study design and the statistical approach. The statistical parametric or nonparametric approaches are typically used to test the hypothesis whether the microbiome compositions are significantly different between groups [20–23]. While the nonparametric approach suffers from lack of statistical power when the sample size is small [39], available parametric approaches consider the abundance of each bacterial categories separately, hence requiring multiple testing corrections. The previous studies in Zimbabwe, Malaysia and Ecuador relating microbiome and helminths compared the difference of abundance of certain bacteria category between groups by using the standard or paired t-test and addressed multiple testing by Bonferonni corrections or False Discovery Rate [20–22]. The clustering of bacteria has been investigated before using descriptive nonparametric approaches such as PCA or NMDS. When we applied these method to our genera data, no clustering was observed; neither by infection status nor by randomization arm. This might be an indication that PCA or NMDS were unable to capture the correlation between genera. We further analysed the multivariate data composed of 6 phyla (Firmicutes, Actinobacteria, Bacteroidetes, Proteobacteria, Unclassified bacteria and pooled category) simultaneously in relation to helminth infection status and treatment using a parametric approach. This multivariate approach takes into account the nature of metagenomics data, such as the abundance of all phyla forming the compositional structure and that these abundances are known to vary highly between subjects [40]. Our method is able to quantify the relationship between the whole bacteria community with regard to the presence/absence of helminths or antihelminthic treatment while taking into account the correlational structure between bacterial categories imposed by the compositional nature. As bacterial categories are correlated, the decrease of one category should cause the increase of other categories and vice versa [7, 41]. Several microbiome studies have reported the change of the ratio Firmicutes to Bacteroidetes [42, 43]. Thus, inference with regard to the decrease or increase of certain bacteria only makes sense when all bacterial categories are considered.
The reparameterization of Dirichlet—multinomial in the data analyses provided an interpretation in terms of odds ratios on how bacterial categories were affected by the helminth infection or treatment allocation. To obtain odds ratios, a reference category needs to be selected. In this study, we used Firmicutes as a reference due to its high abundance among bacterial categories as well as its presence in all samples. The high abundance of Firmicutes remained relatively stable, which had the advantage of allowing us to reveal subtle differences in other bacterial categories.
One potential limitation of our multivariate method is that the number of bacterial categories to be modelled was limited. As a consequence, taxa had to be pooled. Such a procedure assumes that the effect of the underlying taxa are captured in one single parameter. On the other hand, pooling can be viewed as a practical way to deal with sequencing error by providing a more robust model [26]. Instead of pooling, one might use a shrinkage method as proposed by Chen and Li [26] to deal with multiple rare taxa. As an alternative to biostatistical regression methods, machine learning methods are typically used for analysis of microbiome data. However, such methods require larger samples to allow the split into a training and a validation set. Our dataset is too small for such a method. Moreover, this method ignores the correlation structure, such as overdispersion.
It should be noted that the coverage depth in our study is relatively low (in average of 6000 reads per sample) as a result of using pyrosequencing platform (454) compared to more recent deep sequencing technologies (Illumina). We noted that two microbiome studies have reported similar average reads per samples as in our study [20, 44]. As a consequence, rare taxa or taxa with low abundance might not be detected [45], and it is also possible that the similar diversity that we observed could be caused by the use of this platform. However, a direct comparison between Illumina MiSeq and the 454 platform has revealed that the limitation of the 454 is at the genus and family level, while at the higher taxonomic level (such as order, class and phylum level), the 454 platform is able to detect the same number of bacterial categories as the Illumina platform [30]. This could be considered as an advantage of this approach allowing the analysis at the phylum level.
Another unique aspect regarding our study was that a placebo-controlled anthelminthic trial design was used, while other studies were either observational or used an intervention without a placebo group. A control group that did not get the anthelminthic treatment (received placebo) has the advantage of controlling for confounders and estimating a direct treatment effect [46, 47].
There were no significant differences in the microbiome composition, analyzed at the phylum level, of subjects with and without helminth infection at baseline, nor at the 21 months time point. One possibility is the low resolution of the bacterial data at phylum level. It is also possible that the similarity in microbiome composition between infected and uninfected subjects is due to infection history[21]. Surprisingly, we observed a significant difference in the microbiome composition between placebo and albendazole-treated subjects at post-treatment in those who remained infected (Table 2B). This difference seemed to be represented by an increase in relative abundance of Actinobacteria and a decrease in relative abundance of Bacteroidetes. This difference in the relative abundance of Bacteroidetes was confirmed by comparing paired samples at pre and post-treatment in the albendazole group who were infected at baseline and remained infected at post-treatment. No significant difference in microbiome composition was found when comparing the albendazole and placebo arms in subjects who remained uninfected, or when comparing pre and post-treatment in those who received albendazole but remained uninfected. These data indicate that first of all, microbiome composition is stable over time and second, albendazole has no direct effect on microbiome composition. Together, our results suggest that the interplay between anthelminthic treatment and helminths in the gut has a complex effect on the microbiome composition. We observed that deworming is more effective against certain helminth species but not others. Indeed, T. trichiura infection was dominant after treatment in our study. This means that infected subjects who had received placebo harboured different helminth species than those who had received albendazole. However, at pre-treatment, there was no difference between the microbiome associated specifically with T. trichiura, A. lumbricoides or hookworm and therefore the effect of albendazole on the microbiome at post-treatment in infected subjects can not only be due to the dominance of T. trichiura but possibly the result of a combination of Trichuris and albendazole on the microbiome composition. It should be noted that in a recent study taking a different approach from us by using machine learning techniques, considering all taxonomic levels, and large sample size from not only Indonesia but also Liberia, differences in certain taxa were found to be worm-specific [30]. Therefore, to confirm whether T. trichiura has a different effect on microbiome composition after albendazole treatment compared to other helminth species, further and larger studies are needed.
With regard to the treatment effect, a study in Malaysia reported the increasing abundance of Bacteroidales (an order of Bacteroidetes) and the decreasing abundance of Clostridiales (an order of Firmicutes) after treatment [23]. This result might be confounded as there was no control group to assess the treatment effect. Another interventional study was carried out in Zimbabwe, but it did not provide information on the effect of treatment in those who remained infected since the microbiome composition was only measured in subjects who completely cleared their helminths.
A longitudinal setting in microbiome studies has the advantage of analysing the microbiome composition at different time points in the same population. However, the studies using longitudinal approach differed in the length of follow-up time. The studies in Malaysia [23] had a follow-up time of 21 days, the study in Zimbabwe [22] examined the microbiome composition at 12 weeks after treatment while our study had the longest follow-up time of 21 months (with treatment given every three months). Thus, so far the previous studies have examined the effect of short term removal of helminths on microbiota [23], while in our study, we used a longer follow-up time to ensure succesful and long lasting deworming of the subjects. Differences in study design and techniques used for collection and analysis of samples hamper comparison across studies.
The regression model used in this study is only applicable in a cross-sectional manner and assumes a simple correlation structure between bacterial categories. Such a method could be extended to more complex correlation structures. One is the correlation between bacterial categories or between the microbiome composition of the same subject measured at different time points. A statistical test for paired two categorical counts is available, however to model the change in microbiome composition over time we would need to extend our model.
To conclude, the microbiome composition is likely to change due to interactions between helminth and anthelminthic treatment, but a direct impact of treatment on microbiome composition has not been observed. Larger studies are needed to dissect these effects of treatment and also to take into account the history of helminth infection. Furthermore, new statistical methods that allow longitudinal analysis of changes in the microbiome composition need to be developed.
The 16S rDNA assembled sequences, annotation and abundances from all the Indonesia samples are available for download from Nematode.net (nematode.net/Indonesia_Microbiome.html) [48].”
|
10.1371/journal.pgen.1001228 | Competitive Repair by Naturally Dispersed Repetitive DNA during Non-Allelic Homologous Recombination | Genome rearrangements often result from non-allelic homologous recombination (NAHR) between repetitive DNA elements dispersed throughout the genome. Here we systematically analyze NAHR between Ty retrotransposons using a genome-wide approach that exploits unique features of Saccharomyces cerevisiae purebred and Saccharomyces cerevisiae/Saccharomyces bayanus hybrid diploids. We find that DNA double-strand breaks (DSBs) induce NAHR–dependent rearrangements using Ty elements located 12 to 48 kilobases distal to the break site. This break-distal recombination (BDR) occurs frequently, even when allelic recombination can repair the break using the homolog. Robust BDR–dependent NAHR demonstrates that sequences very distal to DSBs can effectively compete with proximal sequences for repair of the break. In addition, our analysis of NAHR partner choice between Ty repeats shows that intrachromosomal Ty partners are preferred despite the abundance of potential interchromosomal Ty partners that share higher sequence identity. This competitive advantage of intrachromosomal Tys results from the relative efficiencies of different NAHR repair pathways. Finally, NAHR generates deleterious rearrangements more frequently when DSBs occur outside rather than within a Ty repeat. These findings yield insights into mechanisms of repeat-mediated genome rearrangements associated with evolution and cancer.
| The human genome is structurally dynamic, frequently undergoing loss, duplication, and rearrangement of large chromosome segments. These structural changes occur both in normal and in cancerous cells and are thought to cause both benign and deleterious changes in cell function. Many of these structural alterations are generated when two dispersed repeated DNA sequences at non-allelic sites recombine during non-allelic homologous recombination (NAHR). Here we study NAHR on a genome-wide scale using the experimentally tractable budding yeast as a eukaryotic model genome with its fully sequenced family of repeated DNA elements, the Ty retrotransposons. With our novel system, we simultaneously measure the effects of known recombination parameters on the frequency of NAHR to understand which parameters most influence the occurrence of rearrangements between repetitive sequences. These findings provide a basic framework for interpreting how structural changes observed in the human genome may have arisen.
| Human structural variation contributes to phenotypic differences and susceptibility to disease [1]. Recent studies suggest that many structural variants are mediated by non-allelic homologous recombination (NAHR) between dispersed repetitive DNA elements [2]–[5]. While the importance of NAHR in shaping genome structure is becoming more apparent, the mechanism of NAHR remains poorly understood.
NAHR (also known as ectopic recombination) utilizes the molecular pathways that mediate allelic homologous recombination (AHR) between sister chromatids or homologs. AHR and NAHR are both initiated by a double-strand break (DSB) that is processed by 5′-3′ DNA resection to generate 3′-OH tailed single-stranded DNA (ssDNA) intermediates [6]. The resected ssDNA, called the recipient, is activated to search for homologous sequences, called the donor, to be used as a template for repair. If the recipient is unique DNA, then the donor will be the homolog or sister chromatid, and AHR ensues. However, if the recipient is repetitive DNA, it may choose a non-allelic repeat as a donor, leading to NAHR and potentially a chromosome rearrangement. The establishment of this basic recipient-donor partnership during homologous recombination (HR) defines four fundamental parameters for NAHR that we address here.
The first parameter is the position of a DSB relative to repetitive and unique sequences. DNA resection starts from the DSB ends and is thought to activate break-proximal sequences before break-distal sequences [7]. Based on this model, break-proximal recipients (sequences at or near the break site) direct homology searches before break-distal recipients (sequences distal from the break site). Therefore, a DSB near or in a repetitive element should activate that repeat as a recipient, which may search for a non-allelic donor repeat to promote NAHR. Alternatively, a DSB in a large track of unique sequences should preferentially activate break-proximal unique sequences as recipients. In a diploid, these break-proximal recipients can repair efficiently using allelic donors on the sister chromatid or homolog. Therefore it has been assumed, but never tested directly, that a DSB in unique sequences in a diploid will rarely induce NAHR. However, a few studies in haploid yeast have observed a preference for recombination using more distal sequences over break-proximal recipients, suggesting that break-distal recipients can participate in homology searches [8]–[10].
The second important parameter of NAHR is the percent and length of identity shared between a recipient and potential donors. Introduction of ∼1% sequence divergence between model repeats decreases recombination rates 9- to 25-fold [11], [12], suggesting that even very limited divergence may significantly affect NAHR rates. The minimum length of uninterrupted identity between two sequences needed for efficient recombination is called the minimal effective processing segment (MEPS) [13]. Using model repeats, the MEPS necessary for efficient NAHR is about 250 bp [14], [15]. This suggests that small retroelements, such as Alus (∼300 bp) and long terminal repeats (LTRs; ∼330 bp), are potentially sufficient to promote efficient NAHR. However, how homology between natural repeats relates to usage for NAHR has never been assessed at a genome-wide scale.
The third important parameter of NAHR is genomic position of a recipient and potential donors. Recipients and donors are more likely to recombine when they are on the same chromosome than when they are on different chromosomes [16]–[18]. Interchromosomal recombination between model repeats can also be influenced by their proximity to centromeres and telomeres [19], [20]. However, these NAHR position preferences have not been tested with natural repeats in an unbiased system, where the unrestricted choice of repair partners and pathways is allowed.
Finally, which HR pathway acts upon a recipient and donor may impact whether NAHR occurs. Single-strand annealing (SSA) can occur when resection from a DSB proceeds through flanking direct repeats, exposing complementary sequences that anneal to generate a deletion product [6]. In contrast, Rad51-dependent HR pathways involve strand invasion events where Rad51 polymerizes onto resected recipient DNA to mediate invasion into a homologous duplex donor. When recipient sequences on both sides of the DSB invade the same donor, repair can occur by gene conversion (GC). However, if the recipient shares identity with the donor on just one side of a DSB, then one-ended strand invasion events can repair through break-induced replication (BIR). GC is faster and more efficient at repairing DSBs than BIR [21]. In addition, GC competes effectively with SSA [22], [23]. While the competition between SSA, GC, and BIR can influence NAHR outcomes, little is known about the relative usage of these pathways during NAHR with natural repeats.
Thus the efficiency and outcome of NAHR are potentially influenced by its ability to compete with AHR, the sequence identity between recipients and donors and their genomic position, and the usage of HR pathways. Yet these potential influences remain untested or unresolved, particularly in the context of a family of naturally repeated sequences. To address these fundamental issues, we developed a new genome-wide system to study NAHR between the dispersed and divergent families of Ty retrotransposons in purebred and hybrid diploids of budding yeasts. We exploit this system to provide insight into the most important parameters that control NAHR in a eukaryotic diploid genome.
Ty1 and Ty2 represent the most abundant families of dispersed repetitive elements in S. cerevisiae. Our system to study Ty-mediated NAHR relies on three components: (1) knowledge of the sequence and position of all Ty1/Ty2 elements in the genome, (2) strains with genetic features for the recovery of Ty-mediated NAHR events, and (3) a protocol to measure these events out of all possible outcomes. Below we provide a brief description of each component.
As a first step, we completed the sequence of the S. cerevisiae unannotated chromosome III Ty elements (Figure S1). With the completed sequence, we generated a map of the distribution of 37 full length Ty1s and 13 full length Ty2s [which includes 98 Ty-associated 5′ and 3′ long terminal repeats (LTRs)], and 208 solo LTRs (Figure 1A). The sequence and positional information is critical since it defines all potential Ty1/Ty2 recipients and donors in the S. cerevisiae genome, allowing us to determine whether some repeats are used and others are not in NAHR.
The potential for Ty elements to act as recipients and donors in NAHR depends in part on their sequence identity. The average percent sequence identity is 95.7±2.4% between Ty1s, 95.9±4.8% between Ty2s, and 73.9±3.4% between Ty1 and Ty2 (Table S3). Previous work has determined that recombination between model repeats decreases 9-fold with 99% identity and 50-fold with 91–94% identity relative to identical model repeats [11]. Thus the sequence divergence of the Ty1/Ty2 family could dramatically reduce the pool of potential Ty recipients and donors, limiting the number of elements that participate in NAHR.
However, if the mismatches are clustered, rather than distributed evenly within the full length of Ty1/Ty2 (5.9 kb), then long stretches of identity may allow efficient NAHR. With this in mind, we analyzed the longest block of uninterrupted identity between all pairwise alignments of Ty1/Ty2, a parameter that has not been previously assessed for Ty elements. To evaluate the significance of these blocks, we categorized them according to the previously determined MEPS value of about 250 bp for NAHR [14], [15]. Recombination rates are predicted to significantly drop when lengths are below MEPS and proportionally increase when lengths are above MEPS [13].
Using our binning analysis, 73% of all Ty1/Ty2 alignments (891 out of 1225) have blocks of identity ≥250 bp (Figure 1B and Table S4). All pairwise comparisons between repeats within either the Ty1 or Ty2 family are above the MEPS value while 31% of pairwise comparisons between Ty1 and Ty2 repeats have a block of identity ≥250 bp. Thus, for the full length Ty1s and Ty2s, the shared blocks of uninterrupted identity strongly predict that a given Ty1/Ty2 recipient can undergo NAHR with many potential Ty1/Ty2 donors, thereby establishing a competition among donors. In contrast, only 1% of all LTR pairwise comparisons (544 out of 46,665) have a block of uninterrupted identity ≥250 bp (Figure 1C and Table S5). This limited length of uninterrupted identity between the LTRs predicts that they may be inefficient substrates for NAHR. In addition, sequence identity amongst pairwise comparisons of the 306 LTR elements widely range between 3%–100%, with an average of 59.6%±22.7% (Table S6). Thus the poor sequence identity between LTRs suggests that solo LTRs will be unfavorable substrates for NAHR.
The second component of our system is the use of specific strains to optimize the recovery of Ty-mediated NAHR events. In order to recover all possible NAHR events, we use diploid yeast where loss of genetic material can be complemented by homologs. In contrast, Ty-mediated rearrangements that occur in haploids may delete genes necessary for viability. Along with S. cerevisiae diploids (referred hereafter as “purebred”), we generated synthetic hybrid diploids by mating S. cerevisiae with a sequenced relative, S. bayanus (referred hereafter as “hybrid”) (Figure 2), which is largely devoid of Ty1/Ty2 elements [24], [25]. The diploids are genetically marked to allow identification of all cells that suffer an I-SceI site-specific DSB as well as the subset of cells in which the broken chromosome is repaired or lost (Figure 2 and see below). Like the purebreds, viability remains high after induction of an I-SceI-induced DSB in the hybrid diploids (Figure 2). In addition, the hybrid diploids grow well and are competent in DNA maintenance and repair like the purebred diploids (Figure S2). Since S. bayanus complements almost all the genes in S. cerevisiae [26], S. bayanus can also balance S. cerevisiae by suppressing any loss of gene function due to NAHR of the S. cerevisiae genome. However, in contrast to the purebred diploids, the hybrid diploids have three important advantages. The significant sequence divergence between the two genomes (62% intergenic, 80% genic) [27] suppresses AHR, favoring NAHR between the more homologous Ty1/Ty2 elements and thus enhancing the recovery of Ty-mediated NAHR events. The sequence divergence also facilitates the analysis of S. cerevisiae rearrangements by array comparative genomic hybridization (aCGH) and PCR. Finally, the comparison of NAHR between the purebred and hybrid diploids allows the assessment of NAHR with and without AHR competition (Figure 2).
The third component of our system is an unbiased clone-based assay to determine the frequencies of NAHR events among all possible outcomes (Figure 3A). An I-SceI recognition sequence [referred to as the I-SceI cut site (cs)], along with a Hygromycin-resistance gene (HYG), is integrated at different positions on the S. cerevisiae chromosome III homolog. We choose to initiate a DSB on the S. cerevisiae chromosome III since this chromosome has the highest density of Ty1/Ty2 elements relative to all other chromosomes (see Figure 1A), making it a good model for the repetitive-rich chromosomes of higher eukaryotes. We initiate the DSB with the addition of galactose to the media for two hours in exponentially growing cultures to induce expression of the I-SceI endonuclease fused to the galactose promoter. Galactose induction of I-SceI expression leads to formation of a DSB at the 163cs position on one S. cerevisiae chromosome III homolog (Figure 3B), which activates recipient sequences adjacent to the break site to undergo a homology search. The cells are then plated onto nonselective YEPD media for individual colonies (referred to as clones). These clones are then phenotyped to determine whether the I-SceI-induced DSB occurred (HygS, see Figure S3) followed by chromosome repair (Leu+Ura+ or Leu+Ura−) or loss (Leu−Ura−). We find that the majority of I-SceI-induced DSBs are repaired in both the purebred (99±2%) and hybrid (79±5%) diploids, although the hybrid diploids exhibit a significant increase in chromosome loss (20±5%) compared to the purebred diploids (1±2%) (Figure 3C). HR mediates almost all of this DSB repair in both diploids since repair is nearly abolished when the essential HR protein Rad52 is absent (Figure 3C).
To assess the structure of the repaired chromosome in the two genetic repair classes, a random subset of clones in each class are further analyzed by pulse-field gel electrophoresis (PFGE)/Southern analysis (Figure 3D). An I-SceI-induced DSB at the 163cs position that is repaired by AHR results in an unchanged chromosome III size whereas repair by NAHR results in a rearrangement with a changed chromosome III size (Figure 3D). Further aCGH and PCR characterization of the genetic repair classes reveals four types of chromosome III rearrangement structures with Ty elements localized to the recombination junctions (Figure S4 and see Materials and Methods). The Leu+HygSUra+ repair class I contains internal deletions, and the Leu+HygSUra− repair class II includes isochromosomes, rings, and translocations (see schematics in Figure 3A). The recovery of these distinct Ty-mediated NAHR rearrangements from one site-specific DSB reveals a competition between recipient and donor Ty elements for NAHR, validating our system as a means to study NAHR between complex families of natural repeats.
A site-specific DSB in unique DNA allows us to assess the likelihood that break-distal repeats are activated as recipients in a homology search to facilitate NAHR. HR events that use a break-distal recipient for recombination are termed here as break-distal recombination (BDR). With 163cs positioned inside 18.1 kb of unique DNA on chromosome III (see map in Figure 3A), we tested the possibility for BDR by monitoring three potential Ty recipient loci (YCRCdelta6, YCRCdelta7, RAHS) at various distances distal from the break site. Because our assay employs no selection, we are able to calculate the frequencies of I-SceI-induced Ty-mediated rearrangements among all possible outcomes after the DSB (see Materials and Methods). Below we highlight the major points from the data compiled in Table 1 and Table 2.
In purebred diploids, 17% of cells after DSB at 163cs undergo NAHR through BDR to mediate rearrangements. Despite a sufficient length of unique sequences that can facilitate AHR with the identical homolog after the DSB, 15±6% of cells use the RAHS recipient, 0.3±0.3% of cells use YCRCdelta7, and 2±0.7% of cells use the YCRCdelta6 recipient located 11.7 kb, 28.9 kb, and 47.5 kb distal from the DSB, respectively (Figure 4A). To test the robustness of BDR, we changed a number of parameters. We eliminated the nonhomology immediately at the DSB ends (1.6 kb I-SceIcs/HYG construct) to test whether BDR is due to the presence of nonhomologous ends, which may inhibit the coordination of two-ended strand invasion events during GC [28]. However, with identity at the DSB ends, BDR is still observed, generating rearrangements (Figure S5). We further tested if BDR was specific to the 163cs position by moving the position of the DSB more centromere-proximal. With the I-SceI-induced DSB at 147cs, BDR-mediated Ty rearrangements occur in 3±3% of cells after DSB (Figure 4A). Interestingly, the frequency of YCRCdelta6/YCRCdelta7 usage is similar to when a DSB initiates at 163cs, suggesting that the usage of these LTR recipients is not determined by their distance from the break site. Lastly, we tested if BDR occurs when the I-SceI-induced DSB initiates on a different chromosome. BDR still occurs in 8±4% of cells after formation of a DSB on S. cerevisiae chromosome V to generate Ty-mediated rearrangements (Figure S6). Thus distal repeats mediate BDR despite the presence of break-proximal unique DNA that can effectively facilitate AHR. This result suggests that unique and repetitive recipient sequences at least 47.5 kb distal to a DSB can participate in recombination.
To test whether AHR competes with BDR, we analyzed BDR in the hybrid diploids. In the hybrid diploids, AHR is mostly suppressed compared to purebred diploids (3±4% of cells after DSB in hybrid compared to 82±6% of cells after DSB in purebred, Figure 4B), as expected from the extent of divergence between S. cerevisiae and S. bayanus genomes. Under these conditions of suppressed AHR, the frequency of BDR increases 4.5-fold compared to purebred diploids (increasing from 17% to 76%, Figure 4B), indicating that BDR competes with AHR. Furthermore, the distribution of different BDR-mediated rearrangements remains the same between hybrid and purebred diploids (compare Figure 4C to Figure 4A, and Table 1). Thus the presence of a divergent homolog at the break site enhances BDR-mediated rearrangements but does not alter preferences of Ty recipient and donors on chromosome III. This aspect of hybrid diploids makes them an excellent model to investigate the features of the recipients and donors that give rise to their preferred use.
To begin to define the parameters that influence the preferred use of recipient sequences to repair a DSB, we determined the largest block of uninterrupted identity between the recipient and its donor. The DSB at 163cs is positioned in the right arm of chromosome III distanced 57.4 kb from the centromere and 165.6 kb from the right telomere. Thus for AHR in purebred diploids, there is >50 kb of identity with the homolog on both sides of the DSB. In contrast, among the BDR events, the largest block of uninterrupted identity with the donors is 1,877 bp for the RAHS recipient, 29 bp for YCRCdelta7 recipient, and 98 bp for the YCRCdelta6 recipient. This reveals that the homology search in purebred diploids can be efficiently directed by 0.1%, 0.2%, or 3% (29, 98 or 1,877 bp out of 57,453 bp) of the potential recipient sequences activated by the DSB, and that this small fraction very distal to the break site generates rearrangements in a total of 17% of cells after DSB. In addition, the smaller and more break-distal solo LTRs, YCRCdelta6 and YCRCdelta7, compete effectively with the larger and more break-proximal RAHS cluster in both purebred and hybrid diploids (see Figure 4A and Figure 4C). These data are consistent with our analysis of AHR in hybrid diploids, where the recombinant junctions occur both proximal and distal to the break site (data not shown). Moreover, these hybrid allelic junctions do not coincide with the longest length of uninterrupted identity (138 bp) found between potential recipients through S. cerevisiae and S. bayanus chromosome III alignments. Thus the relative effectiveness of repetitive and unique recipient sequences competing next to the DSB is not solely predicted by length of uninterrupted identity or distance from the DSB.
Our characterization of Ty-mediated NAHR events also allowed us to investigate the preferred usage of Ty donors with a DSB at 163cs. Intrachromosomal Ty sequences are used as donors in 75±4% of hybrid and 17±6% of purebred cells after DSB at 163cs, generating internal deletions, isochromosomes, or chromosome rings (intra-NAHR in Figure 5A and Table 1). In contrast, only 1±0.7% and 0.3±0.3% of cells after DSB at 163cs produce Ty-mediated interchromosomal translocations in hybrid and purebred diploids, respectively (inter-NAHR in Figure 5A and Table 1). Thus despite the greater number of potential inter- than intrachromosomal Ty donors (see Figure 1A), Ty donors on the same chromosome are preferred approximately 50 times more than Ty donors on a different chromosome.
Again as a first assessment, we wondered whether the NAHR biases for intra- over interchromosomal donors and amongst the two intrachromosomal donors (LAHS and FRAHS) are dictated by sequence identity between the donors with its Ty recipient. We generated a ranked list of sequence homology, comparing the three Ty recipient elements distal to 163cs (YCRCdelta6, YCRCdelta7, RAHS) with all potential Ty donor elements in the genome. We find that the intrachromosomal Ty donors (LAHS and FRAHS) are not among the most identical by either percent sequence identity or the longest block of uninterrupted identity (Figure 5B and Table S7, Table S8). Of the intra-NAHR Ty partners, we also find no correlation with the extent of sequence homology between the chosen Ty donors and their frequency of usage. For example, in the hybrid diploids, 61±3% of cells after DSB generate internal deletions between RAHS and YCRWTy1-5 at FRAHS (97% identity, 1,635 bp largest block of uninterrupted identity) whereas only 3±1% of cells after DSB generate a chromosome ring between the same RAHS recipient and the LAHS donor (97% identity, 1,877 bp largest block of uninterrupted identity). Furthermore, relaxing the stringency for sequence identity in NAHR using msh2Δ/msh2Δ, msh6Δ/msh6Δ, and sgs1Δ/sgs1Δ mutants in hybrid diploids does not abolish the intrachromosomal donor preference (Figure 5A), further suggesting that the preferred usage of donors is not due to sequence identity [29], but donor position. Similar to the findings for the usage of recipient sequences for NAHR, the preferred usage of Ty donors is neither dictated nor can be predicted by sequence homology. Thus the primary determinant of Ty donor choice during NAHR is genomic position, with ∼50-fold preference for intrachromosomal over interchromosomal donors.
Sequence homology between the Ty1/Ty2 families failed to dictate the recipient and donor competition during NAHR. One explanation is that each Ty-mediated rearrangement requires different genetic factors (Table 1), suggesting that they are generated through distinct NAHR pathways. Since HR pathways are known to compete after a DSB, we examined how this competition affected recipient and donor choice. In the hybrid diploids with the I-SceI-induced DSB in unique sequences at 163cs, 61±3% of cells form internal deletions between the RAHS recipient and the FRAHS donor (Table 1). These deletions form independent of RAD51 suggesting they occur through SSA (Table 1). RAHS also mediates isochromosomes (3±1%) and rings (3±1%) with the LAHS donor, and translocations with interchromosomal Ty donors (1±0.7%), all of which have Rad51-dependencies (Table 1). Thus the same RAHS recipient mediates internal deletions 20–40 fold higher than isochromosomes, rings, or translocations, suggesting that SSA dominates the NAHR pathway choice to generate Ty-mediated rearrangements when a DSB occurs in unique sequences.
With at least four NAHR pathways operating after the DSB at 163cs (suggested by the different genetic dependencies of the Ty-mediated BDR rearrangements, see Table 1), we then asked if these NAHR pathways were in competition with one another. To address pathway competition, we attempted to abolish or enhance particular NAHR pathways by removing their intrachromosomal donors and/or repositioning the I-SceIcs in the hybrid diploids. We then compared changes in the frequencies of the Ty-mediated rearrangement product as a readout of their NAHR pathway, where compensatory effects indicate competing pathways. In addition, since Rad51-independent SSA and Rad51-dependent pathways have been shown to compensate for each other after a DSB and hence compete [22], [30], our analysis groups the NAHR pathways into these two distinct HR mechanisms.
We first eliminated the dominant SSA pathway by deleting the FRAHS donor (FRAHSΔ, B in Figure 6) and looked for compensation through the remaining rearrangements. These rearrangements are grouped as Rad51-dependent NAHR since rings show full Rad51-dependency while isochromosomes and translocations have partial Rad51-dependency (Table 1). While some Rad51-dependent rearrangements show a modest increase (rings increase 3±1% to 11±3%, Table 2), the majority of cells cannot repair the DSB at 163cs without SSA, resulting in chromosome loss (71±3% loss, Figure 6). One possibility for this repair inefficiency is that the DSB is too far from the Ty recipients (at least 11.7 kb from the break site) to effectively activate the recipients in Rad51-dependent NAHR pathways. This would be consistent with evidence that Rad51 binding is limited to about 5 kb on either side of a DSB [31]. We then repositioned the I-SceIcs at 151cs, within 0.1 kb of the RAHS recipient in the FRAHSΔ strain (C in Figure 6), in order to enhance Rad51 presynaptic filament assembly onto RAHS. Although a modest increase in Rad51-dependent rearrangements was observed, the majority of cells after the DSB at 151cs with FRAHSΔ cannot efficiently repair the chromosome in the absence of SSA (58±2% loss, Figure 6). These data reveal that Rad51-dependent NAHR pathways induced by a DSB in unique sequences (163cs or 151cs) are inherently inept at repairing the DSB using Ty1/Ty2 elements. Taken together, for a DSB in unique DNA, the efficiency of the SSA pathway coupled with the inefficiency of Rad51-dependent NAHR pathways generates the intrachromosomal position bias and preferential usage of Ty recipients and donors.
Our findings show that the I-SceI-induced DSB in unique DNA (147cs, 151cs, or 163cs) generates substantial NAHR between Ty repeats, giving rise to a broad spectrum of rearrangements through BDR in the purebred diploids. This is in contrast to current models that propose that break-proximal sequences determine the outcome, where DSBs in unique DNA lead to AHR (between sisters or homologs) and DSBs in repetitive DNA can lead to NAHR [32]. To assess the relative consequence of DSBs in unique versus repetitive DNA, we repositioned the I-SceIcs into the RAHS locus (called RAHScs, Figure 7) and used our nonselective assay to measure all possible outcomes after the DSB at RAHScs in hybrid and purebred diploids. From the repair clones generated in our assay, we further characterized two Ty-mediated products that exclusively arise with the DSB at RAHScs, intra-Ty deletions and Ty GC. These Leu+HygSUra+ repair clones are distinguished from each other by assaying RAHS locus size using PFGE/Southern analysis (Figure S7). In comparison to the wild-type RAHS size, we observe a smaller RAHS size for intra-Ty deletion events and a similar RAHS size (with only the removal of the small nonhomologous 1.6 kb I-Scecs/HYG ends) for Ty GC events.
Similar to results with the DSB at 163cs, SSA dominates the NAHR pathway competition, with 66% and 61% of cells after DSB at RAHScs generating Ty-mediated deletions in hybrid and purebred diploids, respectively (Table 2). SSA again imposes a strong intrachromosomal position bias, dictating recipient and donor preferences. The internal deletions from RAHScs, however, can be generated between the RAHS recipient and two different Ty donors, sequences within RAHS itself (referred to as intra-Ty) and FRAHS (now referred to as inter-Ty). All of the internal deletions in purebred diploids are intra-Ty events (61±9%) whereas in hybrid diploids, 59±9% are intra-Ty and 7±5% are inter-Ty (Figure 7 and Table 2). This is consistent with previous work describing a proximity effect during SSA using model repeat donors, with break-proximal donors preferred over break-distal donors [7].
In addition to the events observed with a DSB at 163cs, we find that the second most frequent event after DSB at RAHScs is Ty GC. 22±8% and 33±10% of cells after DSB at RAHScs lead to Ty GC events in hybrid and purebred diploids, respectively (Figure 7). The lower frequency of Ty GC relative to intra-Ty deletions measured in our diploids are in agreement with those events measured using an HO-induced DSB inside Ty1 in S. cerevisiae haploids [33]. Ty GC occurs through the coordination of a two-ended strand invasion event into a Ty donor, which is not a possibility when the DSB initiates in unique DNA (as for 163cs). These GC events in the hybrid diploids must be mediated by a non-allelic Ty donor from the S. cerevisiae genome (since S. bayanus lacks Ty1/Ty2), which likely occurs in purebred diploids as well [16]. Thus, paradoxically, NAHR efficiently mediates conservative repair when a DSB occurs in repetitive DNA.
Having completed our analyses of a DSB within a Ty1 repeat, we can now compare its impact to a DSB in unique DNA on genome integrity. We categorized the outcomes of the I-SceI-induced DSB at RAHScs and at 163cs into two groups: (1) change in gene copy number (inter-Ty deletion, isochromosome, ring, translocation, and chromosome loss) and (2) no change in gene copy number (intra-Ty deletion, Ty GC, and allelic). This comparison reveals that the DSB in unique DNA is 3 to 5-fold more likely to cause a change in gene copy number than the DSB in repetitive DNA (increases from 19% to 97% in hybrid diploids and 6% to 19% in purebred diploids, Figure 8). Thus, distinct from models that highlight the role of DSBs inside repeats in mediating genome rearrangements, our results suggest that the relative mutagenic potential of a DSB in the genome actually decreases when the break occurs within repetitive DNA. Furthermore, this finding suggests that DSBs in unique DNA are more likely to lead to mutagenic rearrangements than DSBs in repetitive DNA.
We report a novel genome-wide system in budding yeast to study non-allelic homologous recombination (NAHR) between natural repeats. While previous assays isolate aspects of competitive repair addressed here, our system gauges the competition between all parameters concurrently, as what naturally transpires in a cell. The value of this new approach is evidenced by the surprising features of NAHR our system reveals. Remarkably, in purebred diploids, DSBs within a long stretch of unique sequences are not always repaired by allelic homologous recombination (AHR) as previously assumed. Rather, 17% of these DSBs repair by NAHR. This NAHR arises because the DSB activates Ty recipients 12 to 48 kb distal from the break site to recombine with non-allelic Ty donor sequences. Robust NAHR through break-distal recombination (BDR) is supported by a previous study of bridge-breakage-fusion in diploid budding yeast by Malkova and colleagues [34].
In this and the previous study, competition between BDR-dependent NAHR and AHR occurs after an endonuclease-induced DSB. In diploids, endonucleases can cleave one homolog prior to DNA replication and both its sister chromatids after DNA replication, thereby eliminating the sister chromatid as a donor for AHR. Therefore, the only AHR donor is the uncut homolog. However, a homolog is also the only AHR donor for repair of spontaneous DSBs that occur on unreplicated DNA in G1 or S. Indeed, recent evidence suggests that spontaneous DSBs occur on unreplicated DNA [35]. We suggest that spontaneous DSBs in unique unreplicated DNA are also likely to induce robust BDR-dependent NAHR.
The fact that break-distal Ty sequences undergoes frequent NAHR reveals two surprising features of recombination that have important mechanistic implications for current models of recipient activation and choice. The first surprise is that distal Ty repeats are activated as recipients at all (presumably by becoming single-stranded) when break-proximal ssDNA can undergo AHR. Indeed, a recent study in diploid yeast suggests that ssDNA is generated at least 10 kb from a DSB before its repair is complete [36]. To explain this extensive break-distal resection, we suggest that a step after resection must be slow, such as the homology search for donor sequences. A slow homology search would provide time for break-distal sequences to be resected and compete with previously resected break-proximal sequences. Such a slow homology search is consistent with studies suggesting the slow diffusion of chromosomal sequences [37].
The second surprise is the disproportionate use of very small break-distal Ty sequences as recipients for NAHR. They would represent only a very small proportion of the entire block of resected DNA, which can all act as a recipient for AHR. We suggest that the smaller Ty recipients encounter their potential Ty donors first because chromosome territories [38] generate a high local concentration of potential intrachromosomal Ty donors. In contrast, the larger allelic recipients must travel further to partner with allelic donors on the homolog. Consistent with this model, almost all NAHR rearrangements through break-distal Ty recipients result from pairing with intrachromosomal Ty donors.
Along with recipient usage, our genome-wide system reveals the role sequence homology and genomic position play in NAHR donor choice. We find that the Ty donors chosen by a recipient are not among the most homologous in the genome by the criteria of either percent identity or longest block of uninterrupted identity. Rather the primary determinant of NAHR donor choice is local proximity. We observe a ∼50-fold preference for Ty repeat donors on the same chromosome over different chromosomes. This intrachromosomal NAHR preference is consistent with previous studies [16]–[19], although the magnitude of this preference differs, possibly due to specific configurations of repeats relative to a break site, as observed in our studies. However, in contrast to previous work, our study shows this intrachromosomal bias occurs under conditions that allow unrestricted choice of repair pathways and partners amongst a natural repetitive family. Interestingly, Ty1/Ty2 elements are preferentially inserted within 750 bp upstream of tRNA genes [39], and dispersed tRNA genes cluster together [40]. Our results suggest that possible Ty interchromosomal contacts mediated by tRNA clustering is not sufficient to overcome an intrachromosomal bias. It will be interesting to see whether higher-order chromosome organization may influence donor repair choice of natural repeats when only interchromosomal donors are available for NAHR.
Our system also provides insights into the preferred repair pathways that act on a family of natural repeats. We show that NAHR occurs mostly by the SSA pathway whether DSBs occur in unique sequences or a Ty repeat. The robustness of SSA is consistent with previous studies using model repeats [18], [23], [30], [41], [42]. Since repair of a single DSB by SSA will occur through an intrachromosomal donor, the predominance of SSA helps explain the preferential usage of intrachromosomal donors and the resulting preference for intrachromosomal NAHR.
Importantly, our pathway analysis of NAHR also helps explain one of the most surprising and striking observations of this study: DSBs that occur outside repeat clusters are more mutagenic than DSBs that occur inside repeat clusters. This seemingly counterintuitive observation arises because DSBs that occur inside a Ty have better options for repair, both in efficiency of pathways and favorably positioned donors. DSBs within the Ty predominately repair through two highly efficient pathways, SSA within the Ty locus or GC with preferred intrachromosomal Ty donors [16]. These types of repair preserve gene copy number since neighboring unique genes are unaffected. Since SSA and GC are compensatory pathways [22], it is possible that DSBs inside repetitive elements that cannot undergo SSA (i.e. solo insertion of LINE-1) efficiently repair through GC events [43]. A recombination execution checkpoint has been suggested to maintain genome integrity by ensuring the coordination of two-ended strand invasion events during GC for conservative repair [28]. Consistent with this, our results suggest that NAHR through GC between natural repeats is a major mechanism that limits changes in genome structure.
In contrast, DSBs in unique sequences that repair predominately through GC with the homolog is not as effective in limiting detrimental rearrangements. As the search for the interchromosomal homolog allows for more time to activate a break-distal Ty as a recipient, BDR occurs more frequently through SSA between distinct Ty loci or one-ended events through the BIR pathway. In this situation, SSA always, and BIR often times, change the copy number of neighboring unique genes. Hence, this opens up the possibility that DSBs in unique sequences, rather than repeats, may generate spontaneous or irradiation-induced NAHR-dependent rearrangements observed in yeast [32], [44]. Similarly, NAHR-dependent rearrangements in the human genome may also occur by a DSB in the surrounding unique DNA followed by BDR-dependent NAHR. If so, then the recombinant junction would not coincide with the site of the initiating lesion. Therefore, analysis of NAHR junctions alone may miss underlying mechanisms for genome rearrangements. Examining broad regions around NAHR junctions could potentially identify fragile sites that predispose a locus to recurrent instability, contributing to genetic diversity and disease.
Standard yeast genetic and molecular biology methods were used [45]. All S. cerevisiae strains were derived from BY4700 (MATa ura3Δ0), BY4716 (MATα lys2Δ0), or BY4704 (MATa ade2Δ::hisG his3Δ200 leu2Δ0 lys2Δ0 met15Δ0 trp1Δ63) [46]. All S. bayanus strains were derived from a S. bayanus prototroph received as a gift from Ed Louis. Deletion of the HO gene and auxotrophic markers were introduced by transformation to generate a number of haploid S. bayanus strains for laboratory use, including MH3399 (MATa hoΔ::hisG ura3Δ::NAT leu2Δ::NAT ade2Δ::hisG), YZB9-4B1 (MATa hoΔ::KAN ura3Δ::NAT leu2Δ::NAT), YZB5-102 (MATα hoΔ::KAN lys2-1) (this study, [47]). Since S. bayanus is sensitive to high temperatures, the following modifications were made to the high efficiency yeast transformation protocol [48] for S. bayanus and hybrid diploids strains: room temperature incubation of transformation mix for 30 minutes, 5 minute heat shock at 42°C, and 5 minute rest at room temperature following heat shock.
Except for some noted below, insertion/knockout constructs were generated through one-step transformation of a PCR amplified linear construct. Each primer for these constructs included ∼50 bp of homology to target for genomic integration and ∼20 bp that anneal to a plasmid template for the amplification of a selectable marker [pAG32-hphMX4 (Hygromycin B), pAG25-ClonatMX4 (Clonat), pFA6a-kanMX4 (Kanamycin), or pMPY-ZAP (hisG-URA3-hisG pop-in/pop-out construct)]. One primer of each of the I-SceI cut site primer pairs also included the 30 bp I-SceI recognition sequence from [49]. For RAHScs, the primers included linkers to amplify an AgeI-I-SceIcs/HYG-ClaI fragment, which was digested and ligated into AgeI-ClaI site of pFT1 (derived from p150Ty, this study). The resulting plasmid, called pFT1-SceIcs, was double-digested with NotI and KpnI and a 10.2 kb purified NotI-KpnI fragment was used for transformation to create RAHScs. For FRAHSΔ::hisG, three primer pairs (FRAHSΔ-left, FRAHSΔ-middle, FRAHSΔ-right) were used to generate three overlapping fragments that were co-transformed. Sequences for gene knockout primers are available upon request. All other strain construction primers included in Table S2. All genome manipulations were performed in haploid strains, and all constructs were verified by Southern blot analysis. Pairs of S. cerevisiae and S. bayanus haploids were mated to generate the desired purebred and hybrid diploids, and then transformed with the I-SceI expression plasmid (see below). All experiments in this study were performed at 23°C unless noted otherwise.
Yeast strains were grown in YEP, SC-ADE, SC-ADE-URA media supplemented with 2% dextrose (D), 2% lactic acid 3% glycerol (LAG), 0.3 mg/ml Hygromycin B (HYG), as indicated. YEPD media was supplemented with 10 µg/ml adenine. Glucose and glycerol was purchased from EMD Biosciences, lactic acid (40% v/v stock, [pH 5.7]) from Fisher Scientific, and Hygromycin B (HYG) from Roche. SC dropout powders were homemade from amino acids purchased from Sigma-Aldrich.
The GALp-I-SceI construct was from pWJ1320 [49], a gift from Rodney Rothstein. pMH5 was derived from pWJ1320 (2 micron-based) by deleting a 2.0 kb EcoO109I fragment containing URA3 marker. pMH6 (2 micron-based) and pMH7 (CEN-based) were created by ligating the 2.0 kb SalI fragment from pWJ1320 (containing the GALp-I-SceI expression construct) into the unique SalI site of pRS422 and pRS412, respectively. pMH6 and pMH7 were generated to include a larger promoter sequence for the ADE2 marker, however, all plasmids yielded similar results.
A single colony from SC-ADE-URA+D+HYG plates [to select for GALp-I-SceI expression plasmid (Ade+) and no DSB (HygRUra+)] was used to inoculate SC-ADE-URA+D for a 5 ml starter culture that was grown to saturation. A small volume of the starter was used to inoculate SC-ADE+LAG cultures and these cultures were grown for more than two doubling to exponential phase [OD(600) ∼1.0]. For the uninduced control, immediately before DSB induction, an aliquot was appropriately diluted in water and plated onto YEPD for individual colonies (uninduced frequencies are subtracted out of induced frequencies, see below). To induce the DSB, galactose (20% v/v stock) was added to a final of 2% and after two hours, the cultures were diluted in water and plated onto YEPD for individual colonies (referred to as clones). Plates were incubated at 23°C for 3–5 days.
YEPD platings from uninduced and induced were first replica plated onto YEPD or 2% agar plates. This replica plate was then immediately used on a fresh velvet to replica onto YEPD+HYG, SC-URA+D, and SC-LEU+D plates. These marker plates were incubated at 23°C for 2–4 days. Each colony from the original YEPD plate was scored for the presence or absence of chromosome III markers (LEU2, HYG, URA3) by growth or no growth on marker plates. Assessment of the heterozygous markers (present on the S. cerevisiae homolog with the I-SceIcs) determines whether the founding cell had experienced an I-SceI-induced DSB (leading to the HygS phenotype) followed by chromosome repair [HygS and Leu+Ura+ (class I) or HygS and Leu+Ura− (class II)] or chromosome loss [HygS and Leu−Ura− (class III)]. The HygS phenotype most likely occurs through the removal of the nonhomologous ends (1.6 kb I-SceIcs/HYG construct), which is a natural and efficient step during HR repair [50], [51].
The following three steps were used to calculate frequencies of repair and loss events. First, the numbers of clones that fell into each genetic class (I, II, III) out of the total number of clones scored were calculated as percentages for both uninduced and induced cultures. Second, uninduced percentages were subtracted from induced percentages to eliminate events that occurred before galactose addition. Occasionally, cultures with high background frequencies (>50% of clones were HygS in uninduced cultures) were observed and not used. HygS phenotypes before galactose induction are due to leakiness of the galactose promoter during nonrepressive growth (see Figure S3). Third, the total percentage (class I + class II + class III) was normalized to 100%. A third potential repair class, HygS and Leu−Ura+, arose so infrequently (<1% in wild-type purebred and hybrid diploids) that it was omitted from these calculations.
Single repair clones (class I and II) from SC-LEU+D marker plates were restruck for individual isolates onto fresh SC-LEU+D plates to ensure clonality (i.e. possible mixing during replica plating process). One isolate from this restreak was used to inoculated YEPD media and grown to saturation for the subsequent isolation of genomic DNA for PFGE/Southern analysis using a LEU2 probe (see below). Hybridization that resulted in wild-type chromosome III size (purebred diploids at 341 kb, hybrid diploids at 320 kb) was identified as AHR and those with an altered chromosome III size, indicative of a rearrangement, were classified as potential NAHR. The structures of the chromosome III rearrangement structures were first determined in wild-type hybrid diploids (MH3360) due to the advantage of no signal from an uncut homolog.
Frequencies were calculated in three steps. 1) Frequencies of genetic classes (I, II, III) of uninduced cultures were subtracted from frequencies of induced cultures to eliminate events that occurred prior to galactose addition (described in more detail above, frequency of chromosome loss determined here). 2) For the repair events, the fraction of each type of repair (i.e. allelic, internal deletion, etc) among the total PFG plugs analyzed from its corresponding genetic class (I or II) was calculated. 3) For the repair events, the genetic class frequency (step one) was multiplied by the fraction of each repair type in that genetic class (step two). For example, in wild-type purebred diploids (MH3359), 85.7% of HygS clones (n = 1062) were class I (Leu+HygSUra+). 5 out of 32 random repair clones of class I were classified as internal deletions by PFGE/Southern analysis, so the frequency of internal deletions in MH3359 is 5/32(85.7%) = 13.4%.
Yeast genomic DNA was prepared in 1% low-melting agarose plugs (SeaPlaque 50100) as previously described [52] and resolved on 1% agarose gel (Bio-Rad 162-0138) in 0.5XTBE using Bio-Rad CHEF-DR III System. To optimize resolution between S. cerevisiae and S. bayanus chromosome III the following parameters were used: 6 V/cm, 120° angle, 1–25 s switch times, 24 hours at 14°C. To assess yeast whole genome karyotypes (i.e. for translocations), the parameters were the same except for 60–120 s switch times. Gels were blotted using GeneScreen Plus membrane (Perkin Elmer NEF988) and probed with a 1.3 kb fragment from the S. cerevisiae LEU2 locus amplified using the U2-FOR/U2-REV primer pair (Table S2).
To calculate SEMs for the repair outcomes, the following numbers were used: (a) average frequency of Leu+HygSUra+ genetic class I, (b) average frequency of Leu+HygSUra− genetic class II, (c) total number of Leu+HygSUra+ (class I) plugs analyzed by PFGE/Southern analysis, (d) total number of Leu+HygSUra− (class II) plugs analyzed using PFGE/Southern analysis, (e) number of Leu+HygSUra+ (class I) plugs of a particular repair outcome (i.e. allelic, internal deletion), (f) number of Leu+HygSUra− (class II) plugs of a particular outcome (i.e. ring, translocation, isochromosome). SEM was calculated in two steps. First, the initial SEM was calculated using the formula SQRT(pq/n), where p = fraction of a particular repair outcome observed by PFGE/Southern analysis over total analyzed from that class (e or f divided by c or d, respectively), q = 1-p, and n = total number of repair clones analyzed by PFGE/Southern analysis from that corresponding class (c or d). Second, the final SEM was calculated by weighting the SEM with the corresponding genetic class frequency (initial SEM multiplied by a or b).
The rationale for this method was to be most stringent by using the smallest n (d or e). In the following cases e or f was assigned the number 1: (1) when all Leu+HygSUra+ plugs were deletions (i.e. in hybrid diploids), (2) no products appear in any plugs analyzed (i.e. rings in rad51Δ/rad51Δ mutant), (3) genetic class is 0 (i.e. Leu+HygSUra− class II in rad52Δ/rad52Δ hybrid diploids), (4) when no plugs analyzed (i.e. Leu+HygSUra− class II in rad52Δ/rad52Δ purebred diploids). For case 1, the error was estimated by assuming the next plug would not be that particular outcome. For case 2, 3, and 4, the upper bound was estimated by assuming the next plug would be that particular outcome. In the case where repair outcomes came from both the Leu+HygSUra+ and Leu+HygSUra− genetic classes (i.e. other, allelic in purebred diploids), “final SEMs” were calculated as described above and then “final SEMs” from each class was added together for the reported SEM. To calculate SEMs for chromosome loss, the formula SD/SQRT(n) was used where SD (standard deviation) = SD of the frequency of Leu−HygSUra− clones from different isolates and/or DSB-inductions (same experiment used to generate numbers for a and b above) and n = total number of different DSB-inductions performed for that particular strain (ranging between 2 to 8).
Exponential cultures in –ade +2% lactic acid +3% glycerol were appropriately diluted in water and the same volume was plated on –ade +2% galactose and –ade +2% glucose. Plates were incubated at 23°C. Percent viability was calculated as the number of colony forming units on galactose divided by the number of colony forming units on glucose.
aCGH methods were performed as previously described [53]. S. cerevisiae/S. bayanus hybrid microarrays were custom designed and printed by Lewis-Sigler Institute Microarray Facility at Princeton University.
Numerous studies have brought to light unannotated Ty elements on chromosome III [34], [44], [54]–[56], with a few studies publishing a limited restriction digest map of the Ty structure in these regions [44], [54], [55]. These unannotated Ty clusters were sequenced here. Each cluster was cloned from strain MH3303 (MATa lys2Δ0 ura3Δ0, derived from BY4716 [46]) by gap repair to create p85Ty, p150Ty, and p169Ty (see Figure S1). Each plasmid was subjected to transposon bombing using the Finnzymes Template Generation System (TGS). For each plasmid, 192 clones with different random transposon insertions were picked and sequenced with a pair of primers located at the edges of the TGS transposon to produce pairs of oppositely directed reads. 384 attempted reads were performed per yeast clone. Sequence data were processed, assembled and edited using the Phred/Phrap/Consed suite of programs [57]. Each assembly was reviewed and edited to ensure there were no discrepancies due to misplaced reads or low quality regions. The automated assembler resulted in collapses of repeats, and these were manually resolved. 16.8 kb of sequence at LAHS, 14.5 kb at RAHS, and 14.7 kb at FRAHS were deposited into GenBank with accession number GU224294, GU220389, and GU220390, respectively. The sequence included five additional full length Ty1s and a solo LTR, complementing the LAHS reference sequence in SGD and almost entirely replacing the RAHS and FRAHS reference sequence. The new sequence changes chromosome III size from 316,617 bp (in SGD) to 341,823 bp.
Sequences for all previously described Ty1, Ty2 and LTRs (delta) elements were obtained from the SGD “Non-ORF dataset” (http://downloads.yeastgenome.org/, timestamp January 5, 2010). Several corrections were made based on our resequencing and analysis: (1) addition of five Ty1 elements on chrIII (Ty1–1 through Ty1–5) (2) addition of nine delta elements on chrIII (delta16 through delta24) (3) removal of three delta elements on chrIII (YCRWdelta8, YCRWdelta9, and YCRWdelta10) (4) addition of one unannotated Ty1 element on chrXII (encompassing YLR035C-A) (5) addition of two unannotated delta elements on chrIV (LTRs for YDRCTy1-2).
The “Overall Identity (%)” between two sequences was determined by creating a global sequence alignment using the Needleman-Wunsch algorithm (gapopen = 10, gapextend = 0.5) as implemented in needleall v6.2.0 [58].
The “Longest Block of 100% Identity (nt)” was determined by first creating a local sequence alignment using the NCBI BLAST algorithm (match = 1, mismatch = −3, gapopen = −1, gapextend = −1) as implemented in bl2seq v2.2.18 [59]. Custom Perl scripts using BioPerl v1.6.1 iterated through each set of hits to identify the longest contiguous block of matching nucleotides [60].
Finally, the contribution of sequence similarity to donor usage is likely more complex than either overall identity or longest block of perfect identity. We therefore calculated bit scores using the BLAST heuristic, which attempts to balance length and perfect identity when searching for a shared region between two sequences that has the “most” similarity. This “Local Identity (bitscore)” was determined using blastall.
Source code and data files can be found at: http://dl.getdropbox.com/u/547386/code.zip
|
10.1371/journal.pntd.0003214 | Hyperferritinaemia in Dengue Virus Infected Patients Is Associated with Immune Activation and Coagulation Disturbances | During a dengue outbreak on the Caribbean island Aruba, highly elevated levels of ferritin were detected in dengue virus infected patients. Ferritin is an acute-phase reactant and hyperferritinaemia is a hallmark of diseases caused by extensive immune activation, such as haemophagocytic lymphohistiocytosis. The aim of this study was to investigate whether hyperferritinaemia in dengue patients was associated with clinical markers of extensive immune activation and coagulation disturbances.
Levels of ferritin, standard laboratory markers, sIL-2R, IL-18 and coagulation and fibrinolytic markers were determined in samples from patients with uncomplicated dengue in Aruba. Levels of ferritin were significantly increased in dengue patients compared to patients with other febrile illnesses. Moreover, levels of ferritin associated significantly with the occurrence of viraemia. Hyperferritinaemia was also significantly associated with thrombocytopenia, elevated liver enzymes and coagulation disturbances. The results were validated in a cohort of dengue virus infected patients in Brazil. In this cohort levels of ferritin and cytokine profiles were determined. Increased levels of ferritin in dengue virus infected patients in Brazil were associated with disease severity and a pro-inflammatory cytokine profile.
Altogether, we provide evidence that ferritin can be used as a clinical marker to discriminate between dengue and other febrile illnesses. The occurrence of hyperferritinaemia in dengue virus infected patients is indicative for highly active disease resulting in immune activation and coagulation disturbances. Therefore, we recommend that patients with hyperferritinaemia are monitored carefully.
| Ferritin is an acute-phase reactant and produced by reticulo-endothelial cells in response to inflammation and infection. In general, ferritin levels are increased in inflammatory conditions, but in this study we found that ferritin levels were much higher in dengue virus infected patients than in patients with other febrile illnesses. This indicates that ferritin could be used as a marker to discriminate between dengue and other febrile diseases. Moreover, the presence of hyperferritinaemia (ferritin levels≥500 µg/L) was associated with markers of immune activation and coagulation disturbances and clinical disease severity, suggesting that it could serve as a marker of activity of disease. Clinical markers to determine the presence and severity of dengue virus infection are important for diagnostic and treatment purposes. Our results indicate that increased ferritin levels could be used to increase the likelihood on a positive dengue diagnosis. Moreover, patients with hyperferritinaemia should be monitored carefully, because they are at risk to develop severe disease due to extensive immune activation.
| Outbreaks of dengue virus (DENV) infection have become more frequent in the American and Caribbean region, even threatening to spread in the United States [1]. DENV is a flavivirus, which is transmitted by the bite of an Aedes mosquito. Brazil is the country with most reported dengue cases in the Americas. A large DENV-2 outbreak in 2010 caused more than 34.000 cases and 64 deaths in the State of São Paulo, Brazil [2]. On the Caribbean island Aruba, there was an epidemic from September 2011 till April 2012, in which DENV-1 and DENV-4 were both co-circulating.
The symptoms of DENV infection are mild and self-limiting in the majority of cases, consisting of fever, headache, retro-orbital pain, myalgia, arthralgia, thrombocytopenia, minor mucosal bleeding and skin manifestations. Some patients develop severe symptoms, such as shock, severe bleeding or organ impairment. These symptoms usually develop three to five days after the onset of disease around the time of defervescence. It has been hypothesized that severe dengue is caused by a cytokine storm inducing systemic inflammatory effects (Reviewed in [3]). The pathophysiological mechanisms that cause this cytokine storm are not fully unravelled and represent an important focus for dengue research.
In addition to the current laboratory markers for dengue, ferritin levels were described to be associated with clinical disease severity in children [4]. In many cases ferritin levels higher than 500 µg/L were detected, defined as hyperferritinaemia [5]. Ferritin is an acute-phase reactant and highly expressed by cells of the reticulo-endothelial system in response to infection and inflammation. Ferritin binds iron, limiting its availability in the circulation. Because many pathogenic microorganisms need iron for their proliferation, this mechanism is favourable for the host. Moreover, iron deficiency enhances the immunological performance of lymphocytes, neutrophils and macrophages (reviewed in [6]).
Hyperferritinaemia is a hallmark of diseases, characterized by extensive immune activation, including haemophagocytic lymphohistiocytosis (HLH) and macrophage activation syndrome (MAS). HLH can be congenital or triggered by an external stimulus, such as malignancy or viral infection, including dengue [7]. NK cells and CD8+ T lymphocytes are impaired in their cytotoxic function in patients with HLH, which results in reduced clearance of infected and antigen-presenting cells from the circulation. This may lead to an exaggerated immune response with proliferation of dendritic cells, tissue macrophages and T-cells, contributing to a cytokine storm (reviewed in [8]). The symptoms of HLH consist of ongoing fever, hepatosplenomegaly, cytopenia (affecting more than 2 cell lineages), hypofibrinogenaemia, hypertriglyceridaemia, hyperferritinaemia, increased levels of sIL-2R and coagulopathy (reviewed in [5]).
Although clinical symptoms of DENV infection are usually rather mild compared to HLH, some patients develop severe symptoms. These are most probably caused by extensive immune activation and show similarities with the clinical hallmarks of HLH and MAS, suggesting similar pathophysiology.
The aim of this study was to investigate the association between hyperferritinaemia, immune activation and coagulation disturbances in DENV infected patients. We showed that the presence of hyperferritinaemia could discriminate between dengue and other febrile diseases. Moreover, we found an association between increased ferritin levels and severe clinical disease, thrombocytopenia, liver enzyme and coagulation disturbances and a pro-inflammatory cytokine profile.
In order to determine the infecting serotype a semi-quantitative RT-PCR (Taqman) was performed. Primers and probes directed against the capsid were derived from Sadon et al. [13]. Briefly, 4× TaqMan Fast Virus 1-step Master Mix (Invitrogen) was used with 20 pmol of primers and 10 pmol of probes. The cycling program consisted of 5 minutes at 50°C, then 20 seconds at 95°C followed by 40 cycles of 3 seconds at 95°C and 30 seconds at 60°C.
Another quantitative RT-PCR was performed to determine the viral copy number. The primers and probes directed against the 3′UTR were derived from Drosten et al. [14]. Briefly, 4× TaqMan Fast Virus 1-step Master Mix was used with 15 pmol of primers and 10 pmol of probes and an additional 25 mM of MgCl2 was added [15]. The cycling program was similar to the serotype quantitative RT-PCR.
Plasma ferritin concentrations were determined at the Landslaboratorium in Aruba within a few hours after blood sampling. The assay was performed using the ‘Access’ (Beckman Coultier, USA) under standardized conditions.
Serum sIL-2Rα and IL-18 levels were determined at the department of experimental internal medicine from the Radboud University. sIL-2Rα was measured using a commercially available luminex kit (‘Milliplex’, Merck Millipore, Germany). Samples were diluted 1∶5 and the assay was performed according to the manufacturer's instructions and run on a Luminex 200 dual laser detection system. The sensitivity limit was 15 pg/ml. Levels of IL-18 were measured using a commercially available ELISA kit (MBL, Japan) according to the manufacturer's instructions.
All markers of coagulation were determined in citrate samples.
Activated partial thromboplastin time (APTT) and prothrombin time (PT) were determined at the Landslaboratorium in Aruba within a few hours after blood withdrawal. PT (Dade Innovin) and APTT (Dade Actin FSL) were determined on a Sysmex CA-1500 System (Siemens Healthcare Diagnostics, USA).
All other coagulation parameters were determined at the department of Experimental Vascular Medicine from the Academic Medical Centre. Von Willebrand Factor (vWF) was measured using a home-made ELISA with antibodies from DAKO (Glostrup, Denmark). In vitro thrombin generation was assayed by measuring peak thrombin levels with the Calibrated Automated Thrombography (CAT) as described previously [16]. In vivo thrombin generation was determined by detecting thrombin-antithrombin complexes (TAT) using a commercially available ELISA (Enzygnost). Levels of the fibrinolytic markers plasminogen activator inhibitor type 1 (PAI-1) and plasmin-α2-antiplasmin (PAP) complexes were measured with commercially available ELISAs according to the manufacturer's instructions (PAI-1, Hyphen BioMed; PAP complexes, DRG Diagnostics). D-dimer levels were determined with a particle-enhanced immunoturbidimetric assay (Innovance D-dimer, Siemens Healthcare Diagnostics).
Serum ferritin levels were determined with a commercially available ELISA (Biolisa ferritina, Bioclin, Brazil) performed according to the manufacturer's instructions.
The measurement of cytokines and the cluster analysis have been described in a previous publication [12]. Briefly, levels of thirty cytokines were measured using a multiplex immunoassay kit with spectrally encoded antibody-conjugated beads (Human Cytokine 30-plex panel, Invitrogen, USA). Twenty-three cytokines were used in a cluster analysis procedure, which was adapted from van den Ham et al. [17]. Briefly, cytokine values were log-transformed and subjected to hierarchical correlation clustering (i.e., with distance measure 1 – pearson's pairwise correlation value) using Ward's method.
IBM SPSS Statistics v.20 was used to calculate statistical significance. The Mann Whitney U test was used to compare the difference between two groups. The Spearman's correlation coefficient was applied to calculate correlations. The Chi-Squared test was used to calculate differences in proportions between groups and the Fisher's exact to determine whether one distribution was unequally distributed over the groups. Using bonferroni correction the p-value was adjusted for multiple testing.
Seventy-three patients were included in Aruba between September 2011 and April 2012. The clinical diagnosis of forty-four patients could be confirmed by serology and/or RT-PCR (Table 1). Seventeen patients tested negative for DENV infection and were included in the OFI group. Twelve patients were excluded with an inconclusive diagnosis. In that particular season, both DENV-1 and DENV-4 were circulating. One patient tested positive for DENV-2, but this patient was probably infected in Suriname. Moreover, 11 patients suffered from a primary and 30 patients from a secondary infection. In three patients the infection status could not be determined.
The epidemic was rather mild and only one case of severe dengue was recorded according to the 2009 WHO dengue case classification. This patient presented with melaena and was included in the WS+ group. From the other dengue positive patients, seventeen were classified as WS− and twenty-six as WS+ dengue. The most common warning sign was abdominal pain (20/44, 45%) followed by vomiting (10/44, 23%) and in a few patients epistaxis (2/44, 5%) and hepatomegaly (1/44, 2%) was reported. Pleural effusion and ascites were not reported, probably because ultrasound and/or X-ray examination were performed on a limited basis. Fifteen dengue patients were admitted to the hospital.
A total of 191 sequential samples from the forty-four patients from Aruba were included in this analysis collected at day 2–8 after the onset of fever (sample size in Table S1). Follow-up samples collected at day 28 from dengue and OFI patients served as an autologous control group.
Ferritin levels were determined in patients with dengue and OFI to identify any association with disease severity. Using the 2009 WHO dengue case classification, ferritin levels were significantly increased in WS+ patients compared to OFI at each time point and in WS− patients compared to OFI at day 4–5 (Figure 1A). At day 4–5 and 6–8 the highest ferritin levels were observed and a tendency was shown towards higher ferritin levels in WS+ patients compared to WS−, although these differences were not statistically significant.
In clinical practice and according to the official HLH-criteria, ferritin levels ≥500 µg/L are considered hyperferritinaemia [5]. A larger proportion of males showed hyperferritinaemia compared to females in this cohort, which approached statistical significance (Table 1). It is known that baseline ferritin levels are higher in the male than in the female population [18]. We calculated a fold change by dividing the absolute values of ferritin by the median ferritin levels for males and females from the autologous control group (Females:/37 µg/L and males:/154 µg/L), which were similar to levels previously described [18]. The ferritin fold change was significantly increased in WS+ dengue patients compared to OFI at each time point (Figure 1B).
Another marker of disease severity is the hospitalization rate. Absolute levels and the ferritin fold change were significantly increased in hospitalized and outpatients compared to OFI at almost all time points (Figure 1C and 1D). The absolute ferritin levels as well as the fold change showed a tendency of increased values in hospitalized patients compared to outpatients.
Because the difference in ferritin levels between patients with dengue and OFI was significant, we calculated an odds ratio for the occurrence of hyperferritinaemia and a confirmed diagnosis of DENV infection. In dengue patients, 19 out of 43 had hyperferritinaemia compared to two out of 17 patients with OFI. This resulted in a sensitivity of 44%, a specificity of 88% and an odds ratio of 6. The high values of the specificity and odds ratio suggest that the occurrence of hyperferritinaemia may serve as a discriminatory marker between dengue and OFI.
The presence or absence of viraemia in the early phase was linked to ferritin levels during the course of disease. Patients were considered viraemic if they had detectable virus titres at day 2–3 and day 4–5. Patients with undetectable levels at these days were considered non-viraemic.
The absolute ferritin levels were significantly elevated in viraemic patients compared to non-viraemic patients at day 6–8 (Figure 2A). The ferritin fold change was significantly elevated in viraemic patients at all time points (Figure 2B). There were no strong correlations between the viral load and the levels of ferritin at the same day of disease. However, absolute levels of ferritin at day 6–8 correlated significantly with the viral copy number at day 2–3 (ρ = 0.5; P = 0.008) and day 4–5 (ρ = 0.5; P = 0.002) (Figure S1). The ferritin fold change at day 6–8 also showed a significant correlation with the viral load at day 2–3 (ρ = 0.5; P = 0.003) and day 4–5 (ρ = 0.6; P<0.0001). This suggests that viral replication in the early phase of disease may cause an increase in ferritin levels in the convalescent phase.
Hyperferritinaemia is a prominent symptom of patients with HLH. To investigate whether the clinical picture of DENV infection shows more similarities, the official diagnostic criteria for HLH [5] were linked to hyperferritinaemia in dengue patients. In each patient the occurrence of hyperferritinaemia was evaluated at each time point.
Severe cytopenia in at least two cell lineages is a prominent feature of HLH due to the increased phagocytic activity of macrophages. In our cohort the platelet count was significantly decreased in patients with hyperferritinaemia compared to patients with no hyperferritinaemia and OFI at each time point (Figure 3A, Table S2). No significant differences in the leukocyte count were detected (Data not shown).
Another criterium is the presence of hypertriglyceridaemia and/or hypofibrinogenaemia. Levels of fibrinogen were significantly decreased in patients with hyperferritinaemia compared to patients without hyperferritinaemia at day 6–8, but levels were still in the range of the autologous control group (Figure 3B). The triglyceride levels were in the normal range of the autologous control group in both dengue as well as OFI patients (data not shown).
MAS is characterized by hepatosplenomegaly and liver dysfunction. The liver also plays an important role in the pathogenesis of DENV infection. Levels of the liver enzyme ASAT were significantly increased in patients with hyperferritinaemia compared to patients with no hyperferritinaemia and OFI at each time point (Figure 3C). ALAT levels were also significantly increased in patients with hyperferritinaemia at day 4–5 and 6–8 (Figure 3D).
sIL-2R is a marker of T-cell activation and IL-18 of macrophage activation. sIL-2R was significantly increased in patients with hyperferritinaemia compared to OFI at day 2–3 (Figure 3E). Levels of IL-18 were significantly elevated in patients with no hyperferritinaemia compared to OFI at day 2–3 and in patients with hyperferritinaemia compared to OFI at day 4–5 (Figure 3E). Altogether, we can conclude that hyperferritinaemia in uncomplicated dengue patients is strongly associated with thrombocytopenia and elevated liver enzymes, but these patients had no hypertriglyceridaemia, hypofibrinogenaemia or cytopenia in another lineage than the platelets.
Hyperferritinaemia was investigated in association with parameters, indicating the activation of coagulation and fibrinolysis. The APTT and PT showed no significant differences between any of the groups (data not shown). vWF is released upon endothelial cell activation and plays an important role in the formation of the thrombus. Significantly increased levels were found in patients with hyperferritinaemia compared to OFI at day 2–3 and in both dengue groups compared to OFI at day 4–5 and 6–8 (Figure 4A). Activation of the coagulation cascade starts with thrombin generation after which it is bound by antithrombin. Thrombin-antithrombin (TAT) complexes are a marker for activation of the coagulation cascade in vivo. Levels were significantly elevated in dengue patients with hyperferritinaemia compared to OFI at day 2–3 and 4–5 and also in patients without hyperferritinaemia compared to OFI at day 4–5 (Figure 4B). The ability of plasma to generate thrombin in vitro can be investigated by the calibrated automated thrombrogram measuring peak thrombin levels. Interestingly, while the levels of TAT were significantly increased, the peak thrombin levels were significantly decreased in patients with hyperferritinaemia compared to OFI at day 2–3 and 4–5 (Figure 4C). Thrombin generation will lead to fibrin formation and eventually fibrinolysis, resulting in the formation of plasminogen-α2-antiplasmin (PAP) complexes. PAP showed increased levels in patients with hyperferritinaemia compared to patients with OFI at each time point (Figure 4D). Plasminogen activator inhibitor-1 (PAI-1) can counteract the fibrinolytic system. Levels of PAI-1 were significantly elevated in patients with hyperferritinaemia compared to OFI at day 4–5 (Figure 4E). Activation of the coagulation and fibrinolytic systems eventually result in the production of D-dimers (Figure 4F). Levels of D-dimers were significantly increased in patients with hyperferritinaemia compared to patients with no hyperferritinaemia and OFI at day 2–3 and levels were significantly elevated in patients with hyperferritinaemia compared to OFI at day 4–5 and in patients without hyperferritinaemia compared to OFI at day 6–8.
The coagulation and fibrinolytic systems are highly activated in dengue patients and dengue patients with hyperferritinaemia in particular. The strongest activation was shown at day 2–3 and 4–5 after onset of fever with increased levels of vWF, TAT, PAP and D-dimer.
To confirm our findings concerning ferritin levels in the cohort from Aruba, we studied ferritin in a previously published dengue cohort obtained during the 2010 DENV outbreak in Brazil. This cohort consisted of 50 WS−, 49 WS+ and 33 severe dengue patients (More clinical details about this cohort are described in the previous publication and in Tables S1,S3 and S4 [12]).
In this cohort the ferritin fold change was calculated with the same formula as described for the cohort of Aruba, because the autologous control group in Aruba was much larger (N = 45) than the healthy control group in Brazil (N = 14). The ferritin fold change was significantly elevated in patients with severe dengue according to the 2009 WHO classification, as well as in patients with shock and severe haemorrhage compared to patients with uncomplicated dengue (Figure 5A, B and C). In non-survivors levels were significantly elevated compared to survivors (Figure 5D). The absolute values of the ferritin fold change were on average higher in the Brazilian than in the Aruba cohort. This could be due to the presence of more severe disease in the cohort from Brazil and the use of a different assay.
Patients were clustered based on the expression of the determined cytokines as has been previously described (Figure 5E and heatmap) [12]. Cluster A contained mainly healthy controls, cluster B mild to moderately ill dengue patients and cluster C contained severely ill dengue patients. Severe dengue (P = 2.2×10−16), shock (3.4×10−5), severe haemorrhage (P = 0.007) and death (P = 0.03) occurred significantly more often in cluster C than the other two clusters. Cluster C showed a pro-inflammatory cytokine profile with increased expression of IL-6, IL-8, IL-10, IL-15, IL-1RA, sIL-2R, HGF, VEGF, G-CSF, MCP-1, IP-10, and MIG. Levels of ferritin were significantly increased in cluster C compared to the other two clusters and levels were also significantly elevated in the ‘dengue’ clusters B and C compared to healthy control cluster A (Figure 5E). In summary, we can conclude that levels of ferritin were significantly associated with clinical disease severity and a pro-inflammatory cytokine profile.
In the cohort from Aruba, increased concentrations of ferritin were significantly associated with a confirmed dengue diagnosis and viraemia. Moreover, hyperferritinaemia in dengue was strongly associated with thrombocytopenia and increased levels of liver enzymes and both activation of the coagulation and the fibrinolytic systems. The findings were confirmed in a cohort from Brazil, in which increased levels of ferritin were associated with severe disease and a pro-inflammatory cytokine profile.
Ferritin is an acute-phase reactant and a significant amount is produced by monocytes, macrophages and hepatic cells. It has been shown that synthesis of ferritin can be induced by cytokines and iron [19], [20]. We showed that increased levels of ferritin were associated with a pro-inflammatory cytokine profile. Lipopolysaccharide (LPS) was shown to induce iron retention in human monocytic cells, which may subsequently induce ferritin expression [21]. Interestingly, increased levels of LPS have been reported in patients with dengue and were also associated with a pro-inflammatory cytokine profile, suggesting that cytokines, LPS and ferritin all play a role in immune activation in severe dengue [12], [22]. Interestingly, it has been shown in vitro that ferritin can bind to high-molecular-weight kininogen and block the release of bradykinin [23]. Bradykinin is a potent vasoactive agent and plays an important role in the induction of vascular permeability and even hypotension (reviewed in [24], [25]). This may suggest that ferritin during DENV infection is induced in an effort to protect the host.
It is well known that infectious diseases in general cause hyperferritinaemia (reviewed in [6]). We showed that even in mild dengue, the occurrence of hyperferritinaemia could serve as a discriminatory marker between dengue and other febrile illnesses. Increased levels of cytokines and LPS have been reported in several infectious diseases and therefore these mechanisms cannot solely explain these extremely high ferritin levels. Macrophages, monocytes and lymphocytes in the peripheral blood are the major target cells of DENV replication in vivo [26], [27]. Monocytes and macrophages are also important producers of ferritin and therefore direct infection and subsequent viral replication in these cells may activate them and increase the ferritin production. In agreement with this, ferritin levels in the convalescent phase correlated strongly with the viral load in the early phase. Interestingly, a high viral load in the early phase of DENV infection has previously been associated with the development of severe symptoms around the time of defervescence [28].
Hepatocytes can also synthesize ferritin and in our study liver enzymes were significantly elevated in patients with hyperferritinaemia. DENV replicates very well in hepatic cell lines in vitro, but whether DENV replicates well in the liver in vivo is still a matter of debate [26]. However, it is likely that liver cells are also indirectly activated by cytokines and/or activated immune cells to produce high amounts of ferritin. It has been shown that DENV infection in mice resulted in NK and CD8+ T cell infiltration of the liver [29].
HLH is characterized by extensive activation and proliferation of NK and CD8+ T cells. CD8+ T-cells can be infected by DENV in vitro [30]. Moreover, apoptosis of CD8+ T cells plays an important role in immune modulation during DENV infection [31]–[33]. sIL-2R is a marker of T-cell activation and increased levels of sIL-2R have been detected in dengue patients with severe disease [28], [34], [35]. In a previous study with patients from the Brazilian cohort, increased levels of sIL-2R were associated with mortality [12]. In this study, using the same cohort, levels of ferritin were also significantly associated with mortality, suggesting that extensive activation of monocytes and macrophages with subsequent T-cell activation may be detrimental for the host during DENV infection.
Certain HLH-criteria, such as hypertriglyceridaemia, hypofibrinogenaemia and cytopenia in at least two cell lineages were not found in this study, most probably because the patients in this cohort only suffered from uncomplicated dengue. Increased triglyceride levels and hypofibrinogenaemia have been reported in patients with dengue shock syndrome and non-survivors [36], [37]. Therefore, we cannot exclude that HLH-like disease occurs in dengue patients with severe symptoms.
In our study thrombocytopenia was strongly associated with hyperferritinaemia. Thrombocytopenia is a hallmark of DENV infection and it is hypothesized that it can be caused by binding of platelets to activated endothelial cells [38]. Platelets are most probably bound by vWF multimers, which were increased in patients with hyperferritinaemia in this study. Because the cytopenia was limited to the platelet count in DENV infection, it is not very likely that phagocytosis by highly activated macrophages is the cause of thrombocytopenia as in the case of HLH.
Coagulopathy is one of the criteria of HLH and also described in severe dengue (reviewed in [39]). In our cohort, the coagulation and fibrinolytic systems were highly activated in patients with hyperferritinaemia at day 2–3 and 4–5 after the onset of fever resulting in increased levels of vWF, TAT, PAP and D-dimer. Levels of TAT were increased in patients with hyperferritinaemia, while levels of peak thrombin were decreased. TAT is a marker of thrombin generation in vivo, while peak thrombin is a marker for the potential of plasma to generate thrombin in vitro. From these results we may conclude that coagulation activation, thrombin formation and the consumption of coagulation factors decrease the ex vivo capacity for clotting during DENV infection, which may result in clinical bleeding symptoms. In addition to activation of the coagulation cascade, increased levels of PAP, PAI-1 and D-dimer showed that the fibrinolytic system was also highly activated in patients with hyperferritinaemia.
Based on the collective results presented in the manuscript, hyperferritinaemia can be considered as a clinical marker for DENV infection, which can discriminate between dengue and other febrile illness. Moreover, ferritin can also serve as a marker for highly active disease resulting in extensive immune activation, coagulation disturbances and severe clinical symptoms. Therefore, we suggest that patients with hyperferritinaemia are monitored carefully, as they are at higher odds to develop severe disease.
|
10.1371/journal.pcbi.1004599 | Direct Measurements of Local Coupling between Myosin Molecules Are Consistent with a Model of Muscle Activation | Muscle contracts due to ATP-dependent interactions of myosin motors with thin filaments composed of the proteins actin, troponin, and tropomyosin. Contraction is initiated when calcium binds to troponin, which changes conformation and displaces tropomyosin, a filamentous protein that wraps around the actin filament, thereby exposing myosin binding sites on actin. Myosin motors interact with each other indirectly via tropomyosin, since myosin binding to actin locally displaces tropomyosin and thereby facilitates binding of nearby myosin. Defining and modeling this local coupling between myosin motors is an open problem in muscle modeling and, more broadly, a requirement to understanding the connection between muscle contraction at the molecular and macro scale. It is challenging to directly observe this coupling, and such measurements have only recently been made. Analysis of these data suggests that two myosin heads are required to activate the thin filament. This result contrasts with a theoretical model, which reproduces several indirect measurements of coupling between myosin, that assumes a single myosin head can activate the thin filament. To understand this apparent discrepancy, we incorporated the model into stochastic simulations of the experiments, which generated simulated data that were then analyzed identically to the experimental measurements. By varying a single parameter, good agreement between simulation and experiment was established. The conclusion that two myosin molecules are required to activate the thin filament arises from an assumption, made during data analysis, that the intensity of the fluorescent tags attached to myosin varies depending on experimental condition. We provide an alternative explanation that reconciles theory and experiment without assuming that the intensity of the fluorescent tags varies.
| Despite decades of study, there is no clear connection between muscle contraction at the molecular and the macroscopic scale. For example, we cannot yet predict how a genetic defect in a muscle protein will result in a physiological change in the heart. This multi-scale understanding is difficult, in part, because molecules cooperate during muscle contraction; that is, one molecule’s behavior is influenced by the behavior of its neighbors. It is difficult to make direct measurements from such coupled molecular systems and also difficult to describe them quantitatively. Despite these obstacles, we recently published experimental measurements and theoretical models of this coupling, but there were apparent discrepancies between the two. Here, we use detailed computer simulations of these experiments to show that, in fact, the measurements agree with the model to a remarkable extent. This agreement suggests that the model captures the essential molecular events that underlie the coupling between muscle molecules. This removes a major obstacle to a multi-scale understanding of muscle contraction and, while more work is necessary, suggests that a connection between the molecular and macroscopic scale is within reach.
| Muscle contraction underlies most voluntary and involuntary motion of multicellular animals. The last 100 years have revolutionized our understanding of this process. Advances in light microscopy in the early 20th century culminated in the sliding filament theory, the discovery that muscle contracts via the relative sliding of two sets of protein filaments (Fig 1A) [1, 2]. To achieve this motion, it was hypothesized that myosin molecules in thick filaments form transient bonds with specific binding sites on actin in thin filaments [3]. Subsequent experiments characterized the biochemical reactions between myosin and actin, culminating in the determination of how these reactions couple to the hydrolysis of ATP [4]. More recently, single molecule techniques have allowed direct observation of myosin binding to actin [5] and measurement of the effects of external force on this chemical interaction [6, 7]. However, despite this wealth of knowledge, a clear understanding of the connection between muscle contraction at the molecular and cell or organ scale remains unclear. Besides integrating a century of research on muscle physiology, such an understanding might allow, for example, novel treatments of genetic heart disease [8].
Connecting the molecular and macro scales of muscle contraction is a complex problem. Forming this connection requires 1) understanding how an isolated myosin molecule interacts with a thin filament; 2) understanding how this interaction changes when multiple myosin molecules work together; 3) understanding how the collective behavior of multiple myosin motors changes upon the addition of accessory proteins like titin and myosin binding protein-C (MyBP-C); and 4) understanding how collections of myosin with accessory proteins work together when arranged as sarcomeres in series, myofibrils in parallel and so on up to whole muscle (Fig 1A). While recent single molecule experiments have given a relatively clear picture of how isolated myosin molecules interact with actin filaments e.g. [5–7], the remaining three ingredients of a molecular to macro scale understanding remain much less clear.
Thus, one obstacle to connecting muscle at the molecular and macro scale is that myosin motors interact with each other via a common actin filament. This coupling occurs in two ways. The first, which we call mechanochemical coupling, occurs because myosin motors apply forces on each other when they bind to the same actin filament. These forces can accelerate chemical reaction rates, and lead to emergent ensemble behavior [9, 10]. If the actin filament is relatively stiff, then this coupling affects every molecule equally—that is, it doesn’t matter where on the actin filament the bound myosin molecules are located.
The second type of coupling, herein termed local coupling, occurs when a myosin molecule binds to the actin filament and affects the reaction rates of nearby myosin molecules, but has no effect on distant myosin. Local coupling happens when myosin binds to actin at low calcium in the presence of the regulatory proteins troponin and tropomyosin. Under these conditions, the protein tropomyosin, which wraps around the actin filament, sterically hinders myosin binding by covering the binding sites on actin (Fig 1A) [11]. If a myosin molecule then binds, it deforms tropomyosin. This deformation, being local, makes it easier for neighboring myosin molecules to bind to actin, but has no effect on distant myosin molecules (Fig 1B–1D) [12–17]. Direct observation of either type of coupling presents a major challenge, since traditional single molecule techniques cannot be used. With the lack of direct measurements, mathematical models can be used to interpret indirect observations. However, it is unclear whether mechanochemical and/or local coupling can be captured by current muscle models, since many muscle models incorporate the assumption that molecules act independently.
Mathematical modeling has played an important role in our understanding of the connection between muscle contraction at the micro- and macro-scale. Shortly after the sliding filament and cross-bridge theories of muscle contraction were proposed [1, 2], they were implemented in a mathematical model [3]. Since then, as biochemical and biophysical experiments have uncovered more about the molecular scale, increasingly detailed and increasingly accurate mathematical muscle models have been developed [18–30]. The majority of these models build on the ideas and assumptions of Huxley’s original model [3]. One of these assumptions is that each myosin molecule acts independently of neighboring myosin molecules, the independent force generator assumption [31].
One might expect that any type of coupling between myosin molecules in an ensemble would violate the independent force generator assumption. However, mechanochemical coupling does not. This is because, assuming that the thin and thick filaments are stiff, mechanochemical coupling affects each actin-bound myosin molecule equally, regardless of its position on actin. Thus, the future behavior of each myosin molecule in an ensemble can be determined from the average properties of the ensemble, an alternate statement of the independent force generator assumption [10].
In contrast, local coupling does violate the independent force generator assumption [32]. This is because the future behavior of a myosin ensemble cannot be predicted simply by average ensemble properties, but also depends on the spatial distribution of those properties. For example, two ensembles with, say, 25% of the myosin molecules bound to actin will behave differently if all of the bound molecules are tightly clustered together or if the bound molecules are distributed evenly [13]. Thus, muscle models that incorporate Huxley’s independent force generator assumption cannot capture aspects of muscle activation where local coupling plays an important role.
Since local coupling violates the independent force generator assumption, modeling locally coupled myosin ensembles is challenging. Some of the earliest studies [13] recognized that local coupling occurs in some physical models. In particular, the 1D Ising model considers the behavior of an ensemble of molecules, each of whom is coupled to its nearest neighbor. For this model, analytic expressions can be derived for the equilibrium distribution of the ensemble [13, 33]. But this approach is also limited, in that it is unclear how to model long-distance coupling or non-equilibrium effects, both of which are of central interest in muscle modeling.
An alternate approach is to explicitly model each molecule in a myosin ensemble [14, 34–40]. To capture stochastic effects, Monte-Carlo methods are used [41]. One advantage of this approach is that, since each molecule is considered explicitly, local coupling can be easily included. Thus, effects like filament elasticity, which can contribute to local coupling, can be incorporated into models [35, 37]. These models require significant computational expense, and so have only recently become tractable. But, even with modern computational power, optimizing fits of such models to data is difficult. As a result, most muscle models that include local coupling cannot describe a large suite of experiments.
With the aim of developing a muscle model that describes the largest number of experimental results, a model has recently been proposed that includes both mechanochemical and local coupling in a realistic way and yet can be implemented with minimal computational expense [32, 42, 43]. This model describes a series of experiments that indirectly measure local coupling between myosin molecules. In particular, the local coupling model (summarized in more detail in the Methods section), which has only two parameters, when incorporated into a model of mechanochemical coupling [10], describes measurements of in vitro motility of thin filaments at low ATP [14, 42], at more physiological ATP and in the presence of MyBP-C [43–45] and at physiological ATP and variable calcium [14, 32, 46]. Importantly, the same model with consistent parameters describes these in vitro motility experiments and then reasonably predicts fiber level experiments [32, 47–51]. It is therefore plausible that this model describes how myosin motors work together, and how that interaction changes with the addition of the accessory protein MyBP-C. Given that these are requirements to a multi-scale understanding of muscle contraction, the model represents progress in that direction. However, as yet, the connection of this local coupling model to the molecular scale is speculative, as the model has not been compared to direct molecular-scale measurements.
Recently, the first direct observations of myosin binding to thin filaments were made across a range of activating conditions [52]. Analysis of these data, which clearly demonstrate local coupling, revealed a discrepancy between experiment and theory: the experimental data point towards two heads minimally being required to activate the thin filament whereas the model predicts only one head is required. To understand this discrepancy, we simulated the experimental data using a stochastic method and analyzed the outputs using the same algorithms used for the experimental data. Remarkably, by assuming that fluorescently-tagged myosin emits a constant amount of light per unit time, we found that the model is consistent with the experimental measurements. This assumption, for which we find indirect support, contrasts with an assumption made by Desai et al. [52] that emission depends on experimental condition. Thus, here we propose an alternative scenario where a single myosin head is sufficient to activate a thin filament, thereby reconciling experimental data and theoretical predictions.
We adapted a local coupling model [32, 42, 43] of myosin’s interaction with thin filaments to simulate recent single molecule experiments [52]. Here, we provide a brief description of the model, the experiments, and the simulations.
To describe the local coupling between myosin molecules, we use a simplified mechanochemical model of the thin filament, based on the continuous flexible chain model [15–17]. In this model, troponin/tropomyosin is modeled as a continuous elastic beam, which contrasts with more structurally detailed models of local coupling that explicitly model each troponin and tropomyosin molecule, e.g. [36, 37]. Although this simplifying assumption makes some effects more difficult to include (e.g. potential variations in troponin density [37]), one advantage of this approach is that model parameters are related to physical properties of the system.
The local coupling model used here starts with three assumptions based on the continuous flexible chain model [15–17] 1. troponin/tropomyosin (Tn-Tm) is a slender, infinite, linear-elastic beam constrained to a plane; 2. the interaction of Tn-Tm with the actin filament, described by potential energy density, W(z), varies azimuthally along actin but does not vary (or the variations can be averaged out) longitudinally along actin; 3. myosin binding induces a local displacement of the Tn-Tm beam. These assumptions are shown in Fig 1B.
A few simplifications differentiate this model from the continuous flexible chain model [42]. The first simplification is based on the observation that the energy of the Tn-Tm beam (E) contains two quantities: an elastic component that is minimized when the beam is straight, and a potential energy component that depends on the details of W(z). Importantly, the elastic component dominates over short distances, while the potential energy component dominates over long distances. Therefore, when two nearby myosin molecules bind to actin, the Tn-Tm beam between the two myosin remains relatively straight, minimizing elastic energy; conversely, when two distant myosin molecules bind to actin, they each induce independent deformations, minimizing potential energy (Fig 1C).
These ideas lead to specific predictions of the energy change of the actin-Tn-Tm system when myosin binds to actin (ΔE). Consider, for example, the energy change that occurs when a myosin molecule binds to actin when another myosin molecule, a distance L away, has previously bound to actin. If L is small, the Tn-Tm beam should be displaced uniformly, minimizing elastic energy. And, since W(z) does not vary longitudinally along actin, the energy change should increase linearly with L. Conversely, if L is large, then myosin binding should induce an independent deformation in the Tn-Tm beam, minimizing potential energy, and the energy change should not depend on L (see Fig 1D). It is convenient to assume that the transition between these two regimes is abrupt, and occurs at some critical separation L = LC. Then, the final curve is defined by two parameters, C and ε. The first is the critical length scaled by the separation between myosin molecules, C = L C / Δ L. The second is the reduction in attachment rate due to the regulatory proteins—i.e. if myosin binds to unregulated actin at a rate k0, then myosin binds to regulated actin at a rate εk0 when L ≫ LC (see Fig 1D).
These results are independent of the details of Tn-Tm’s interaction with actin (i.e. the exact shape of W(z) [42]). These ideas define any pair-wise interaction between myosin molecules, and therefore we can calculate the energy change that would occur with the binding of any myosin molecule. Assuming that attachment rate depends exponentially on this energy, we can then incorporate regulation into cross-bridge muscle models [10].
More specifically, given a cross-bridge muscle model, regulation is incorporated by modifying the attachment rate and leaving all other rate constants unchanged. Thus, if myosin binds to unregulated actin at a rate k0 in a particular model, and supposing that a given molecule’s ith neighbor to the left and jth neighbor to the right are bound to actin, then that given molecule has an attachment rate k from the following equation [42]
k ( i , j , C , ε ) = { ε k 0 : i , j ≥ C ε i / C k 0 : i < C , j ≥ C ε j / C k 0 : i ≥ C , j < C ε ( i + j - C ) / C k 0 : i , j < C , i + j > C k 0 : i + j ≤ C (1)
As this attachment rate depends on the state of nearby molecules, it introduces local coupling to the model.
Differential equation methods can be used to simulate the behavior of myosin ensembles with both this kind of local coupling and also mechanochemical coupling [32, 42]. These differential equation methods are much more computationally efficient than Monte-Carlo methods, allowing optimization of fits and parameter estimation. Thus, from fits to in vitro motility data, we can obtain values for the parameters C = 11 and ε = 0.003 at pCa 8 [42], where pCa is the negative log of the calcium concentration. With these parameters, the model fits motility experiments in the presence of MyBP-C, suggesting that MyBP-C both activates the thin filament and specifically binds to actin, thereby competing with myosin and creating a viscous drag [43]. By assuming that calcium only affects ε, and that C remains constant, force-pCa experiments from muscle fibers and fiber twitch experiments can be reproduced and then motility-pCa curves and twitch summation experiments are successfully predicted [32]. The reasonable agreement between model and experiments provides support for its assumptions.
If this model is correct, then at low calcium, the binding of one myosin molecule can locally activate the thin filament. Desai et al. [52] interpret their experiments to support the view that at least two myosin molecules are necessary to activate the thin filament, contradicting the model. To understand this difference we used stochastic methods to simulate the experimental observations.
In the previously published experiments, the head domain of myosin (S1) was directly observed binding to thin filaments (actin with troponin and tropomyosin) [52]. The experiment starts by using fluid flow to suspend a thin filament between two silica beads affixed to a glass surface (Fig 2A). Then, a solution is added that includes variable concentrations of 1) GFP-labeled S1 domains of myosin (GFP-S1); 2) ATP; and 3) calcium. The thin filaments are then imaged using Oblique Angle Fluorescence microscopy to detect fluorescent GFP-S1s bound to the thin filament.
With appropriate concentrations of GFP-S1, ATP and calcium, individual fluorescent spots are observed (Fig 2A). These spots can vary in brightness, indicating the binding of multiple GFP-S1s, can appear/disappear or can randomly diffuse along the thin filament. Movies of these spots were visualized in two ways, as kymographs (Fig 2B) and as histograms (Fig 2D).
To generate both a kymograph and a histogram from a movie, Desai et al. [52] performed the following steps. Pixels lying along the thin filament were isolated (Fig 2A, red box). For the kymograph, the intensity along this line of pixels was plotted for each frame of the movie, generating a 2D image with position in the ordinate and time on the abscissa (Fig 2B). Any GFP-S1 interactions with the thin filament appear as streaks that start when the GFP-S1 binds and end when the GFP-S1 detaches.
For a histogram, the intensity of the line of pixels was plotted as a function of pixel position (Fig 2C). These data were then fit with custom code [52] that determines the best-fit Gaussians of fixed standard deviation, but variable amplitude. The amplitudes of these best-fit Gaussians were determined for every frame of the movie and then plotted as a histogram (Fig 2D). As concentrations of GFP-S1, ATP and calcium were varied, clear differences are observed in both the kymographs and the histograms [52].
In order to compare our model most directly to experimental measurements, we made the following assumptions:
A single excited GFP emits a constant average amount of light per unit time, e. In the movies from Desai et al.[52], the recorded fluorescence intensity of a single GFP (I1) is inversely proportional to frame rate (f). Thus, a single GFP, imaged at f = 2 Hz, will be half as bright as a single GFP, imaged at f = 1 Hz. There are two sources of signal noise. The first is background noise (e.g. electrical noise, unbound GFP-S1s, etc.) that is uniform across the image. This noise varies, depending on experimental condition. We assume this noise is Gaussian with mean zero and standard deviation σN (see Fig 3A). The second source of signal noise is due to temporal variations in GFP intensity (e.g. flexing of the thin filament causing the center of the GFPs to deviate from the nominal average position of the thin filament, Fig 2A, red box). This noise is constant across all conditions. We assume this noise is Gaussian with mean zero and standard deviation σF (see Fig 3A). Since GFP-S1s are coming out of solution, they can bind to either side of the actin filament. We assume each side is regulated independently [53]. Thus, each thin filament contains two independent systems of the sort shown in Fig 1C. From fits to in vitro motility, we estimated C = 11 [42]. Since C = L C / Δ L, where ΔL is the spacing between myosin molecules, this gives a value of LC = 385nm, using ΔL = 35nm for myosin molecules attached to a glass surface [54]. Here, the GFP-S1s are not constrained by being bound to a surface so they can bind to each actin monomer, spaced 5.5 nm apart. Thus, ΔL = 5.5 nm, and so C = 385 / 5 . 5 = 70. We assume that this value is independent of calcium concentration. The value of ε is a function of calcium, and should lie between 0.003 (at very low calcium) and 1 (at high calcium, or in the absence of regulation) [42]. GFP-S1 myosin obeys the kinetic model described in Walcott et al. [10], but here it is coming out of solution and the rate limiting step is no longer the weak to strong transition, but rather the initial formation of the weak-binding complex (Fig 3B). Thus, the attachment rate per μm of actin is κ b 0 [ Myo ], where [Myo] is the concentration of GFP-S1 and κ b 0 = 0 . 2 nM - 1 μ m - 1 s - 1 [52]. Since binding sites are spaced 5.5 nm apart along both sides of an actin filament, this corresponds to a binding rate of
k b 0 = ( 0 . 2 nM - 1 μ m - 1 s - 1 ) ( 1 μ m 1000 nm ) ( 5 . 5 nm 2 binding sites ) = 0 . 00055 nM - 1 s - 1 per binding site Since myosin is coming out of solution, the myosin molecules do not apply forces on each other and so there is no mechanochemical coupling.
Assumptions 1 and 2, which posit that a single excited GFP emits, on average, a constant amount of light per unit time (e) and that the measured signal intensity (I(t)) is inversely proportional to frame rate (f), differ from the assumptions made by Desai et al. [52] that the recorded intensity of a GFP varies depending on experimental condition. We estimate e = 450 fluorescent units/s, giving a GFP intensity of I1 = 45 at f = 10 Hz and I1 = 118 at f = 3.8 Hz (see S1 Supplementary Material). To eliminate differences in measured signal intensity due to frame rate, we introduce a dimensionless quantity I F = I ( t ) f / e, the scaled intensity. With this definition, a cluster of n GFP-S1s generates a scaled intensity of I F = n.
To model the experiments, we estimated the background noise, defined by standard deviation σN, and the variability in GFP intensity, defined by standard deviation σF. For all simulations, we used σF = 0.22 (scaled intensity). The background noise varied depending on experimental condition, so we estimated σN for each condition. The values vary from σN = 0.21–0.36 (scaled intensity), and the exact values are given in Table 1. We can estimate overall signal noise by the root mean squared error between the measurement and the Gaussian fits. A comparison between the fitting error from a simulated experiment and from a measurement, consisting of 500 frames, suggests that our simulations have reasonably captured experimental noise (Fig 3C).
With these assumptions, and given a value for the local coupling parameter ε, we can perform simulations of the experiments performed by Desai et al. [52]. To do so, we used the following algorithm:
For each experimental condition, we performed five simulations.
To perform the simulations, we must specify the local coupling parameter ε. From assumption 7, we have 0.003 < ε < 1; we can further refine this estimate by comparison to measurements in the absence of regulation (ε = 1) and at pCa 4 and pCa 8 [52]. Even at high calcium (pCa 4), Desai et al. [52] observed a significant decrease in binding events in the presence of regulatory proteins compared to in their absence. We can reasonably replicate their data if we assume that 0.01 < ε < 0.5 in the presence of regulatory proteins, with the maximum and minimum values being obtained at pCa 4 and pCa 8, respectively (Fig 3D). This is consistent with the observation that, even at saturating calcium, myosin strong binding to actin induces a shift in the position of tropomyosin [55].
For each calcium concentration (pCa 5, 6 and 7), we estimated ε by trial-and-error, with the goal being to replicate the experimental measurements (see S1 Supplementary Material). Our model therefore has a single fitting parameter at each calcium concentration. The values used in our simulations are ε = 0.4 at pCa 5, ε = 0.06 at pCa 6 and ε = 0.02 at pCa 7. All model parameters are summarized in Table 2.
In the model, local coupling is determined by the parameters ε and C. These affect the rate at which a myosin molecule binds to actin, kb, according to Eq 2. Importantly, the value of kb depends on where neighboring myosin molecules are bound, thereby defining the local coupling in the system. Besides defining this reaction rate, the parameters ε and C also relate to mechanochemical properties of the thin filament. This connection is made explicitly, and mathematical expressions are provided, in Walcott 2013 (where the parameter ε is called δ) [42]; here we provide a more intuitive interpretation.
The parameter ε defines how the Tn-Tm beam decreases the attachment rate of an isolated myosin molecule. That is, if ε = 0.1, then the regulatory proteins decrease the binding rate of an isolated myosin 10-fold. Since calcium affects how easily myosin can bind to actin, ε is a function of calcium concentration. The value of ε can be related to physical properties of the thin filament, given the assumption that attachment rate depends exponentially on the energy required to displace the Tn-Tm beam, ΔE. In particular, since ε is defined as the reduction in an isolated myosin’s attachment rate due to the regulatory proteins, the exponential dependence of attachment rate implies that limL→∞ exp(−ΔE) = ε (where ΔE is measured in kB T, Boltzmann’s constant time temperature). Thus, ΔE asymptotes to −ln(ε) as L becomes large (Fig 1D).
The parameter C is defined as the ratio of two lengths: C ≡ L C / Δ L. The length ΔL is the spacing between neighboring myosin molecules interacting with a given actin filament, and depends only on how the myosin molecules are arranged in a given experiment. The length LC is the critical separation between myosin molecules. If two myosin molecules are bound to actin and the separation between them is greater than LC, then they induce independent deformations of the Tn-Tm beam. That is, they are uncoupled. Conversely, if two myosin molecules are bound to actin and the separation between them is less than LC, then they completely activate the intervening length of thin filament.
The value of LC depends on details of how the Tn-Tm beam interacts with myosin (W(z)), upon its mechanical properties (i.e. its flexural rigidity [42]), and upon the distance that the Tn-Tm beam is deformed upon myosin binding. These likely do not change upon changes in, say, ATP concentration or the geometry of the assay. Thus, we have assumed that LC is constant between the experiments considered here and the motility assay [32, 42, 43]. In contrast, the interaction between the Tn-Tm beam and actin, W(z), might depend on calcium, and so it is possible that C varies with calcium. However, the local coupling model reasonably reproduces experimental measurements if C is assumed to be calcium independent, and only ε varies with calcium [32]. We therefore make that assumption here, fixing C (see Table 2) and allowing ε to vary in order to give the best fit between model and measurement.
We simulated nine different experimental conditions to follow Desai et al. [52], varying the amount of GFP-S1 ([Myo]), the amount of calcium, and the amount of ATP (pCa 6, ATP = 0.1μM, [Myo] = 1, 5, 10, 15nM, 1000 frames collected at 10Hz; pCa 5, 6, 7, ATP = 0.5μM, [Myo] = 15nM, 500 frames collected at 3.8 Hz; pCa 6, ATP = 0.1, 0.5, 1μM, [Myo] = 15nM, 500 frames collected at 3.8Hz). Representative kymographs from measurements and simulation are shown in Fig 4. In all cases, agreement between simulation and experiments is reasonable.
We can make a more quantitative comparison between simulation and experiment by calculating the mean fluorescence per pixel in the kymographs. The agreement between simulation and experiments is again good (Fig 4D shows variable [Myo], the remaining plots are in the S1 Supplementary Material). Importantly, both the simulation and measurements show a non-linear dependence of mean fluorescence on [Myo]. Mean fluorescence (in units of scaled intensity) corresponds to binding probability and, since increasing [Myo] increases binding rate linearly, in the absence of local coupling mean fluorescence should increase, at most, linearly in [Myo]. The observation, in both experiment and simulation, that this curve has a faster-than-linear increase therefore implies the presence of local coupling in the experimental system and that the model accurately describes this local coupling.
The data contain more information than just mean fluorescence; each kymograph shows local clusters of fluorescence of varying brightness. The brightness of each spot can be quantified and plotted as a histogram, using custom code developed by Desai et al. [52]. We analyzed our simulated data with this code, as shown in Fig 2D, and the results are shown in Fig 5. With the exception of one measurement at pCa 6, [Myo] = 15 nM, and [ATP] = 0.1μM (rightmost panel in Fig 5C), the agreement between simulation and measurement was good. Indeed, this agreement is remarkable since, at each calcium concentration, we had only one parameter (the binding probability: ε) with which to fit the data.
The reason that the simulations do not fit all of the experimental measurements becomes apparent upon closer inspection. There were two measurements performed at pCa 6, [Myo] = 15 nM, and [ATP] = 0.1μM, one that was imaged at 10Hz (Fig 5A, rightmost panel) and one that was imaged at 3.8 Hz (Fig 5C, rightmost panel). While the simulations agree with the former, they do not agree with the latter. There are several possible reasons for this inconsistency, including variability in GFP fluorescence between the two experiments, but we believe the most likely explanation is that under these conditions where GFP-S1 binding is frequent and local coupling is strong, the system is highly sensitive to parameters. For example, with only a 50 nM increase in ATP concentration, the simulations are in reasonable agreement with the measurements (Fig 5D).
Accurate and efficient models of muscle contraction that are consistent with measurements across size scales could transform research in several fields. The study of genetic muscle defects, for example, would benefit from being able to predict how changes in molecular parameters due to mutation affect macroscale function [8, 56]. The study of human movement, and prosthetic design [57, 58], would benefit from being able to predict ATP hydrolysis in muscle as a function of muscle force and motion [59]. Most current muscle models are not able to achieve these tasks, in part, because they cannot accurately and efficiently describe the local coupling between myosin motors that occurs during activation. A recently proposed muscle model considers molecular mechanics and can be implemented using differential equations, thereby having the potential to describe this local coupling with unprecedented accuracy and efficiency [32, 42, 43]. However, until recent experimental advances [52], this model could only be tested against indirect measurements.
Here, we have tested this local coupling model [32, 42, 43] against these direct measurements of fluorescent myosin binding to thin filaments [52]. Not only does the model successfully describe myosin’s average binding probability, where we observe local coupling (Fig 4D), but it also describes the clustering of myosin molecules (Fig 5A–5C). Differences between measurement and simulation, seen in only one of the nine experiments, can be explained by the system being highly coupled and therefore sensitive to parameters (Fig 5D). This agreement between simulation and experiments suggests that the model captures the essential molecular events that underlie local coupling through the thin filament. While more work is necessary to understand how sarcomeres are coupled and the role of accessory proteins, this success at the molecular scale, in conjunction with the success of the model at the ensemble and fiber scale under more physiologically realistic conditions [32, 42, 43], represents progress in understanding the role of local coupling in muscle contraction and, more broadly, a multi-scale understanding of muscle contraction.
Although the simulations and measurements agree, the model requires only a single myosin head to activate the thin filament; however, Desai et al [52] inferred that two heads are required. This difference is explained by alternate assumptions. Here, we assume that an excited GFP emits the same amount of light per unit time (e), regardless of condition, and that the intensity of a GFP recorded by the camera is inversely proportional to frame rate (I1 = e/f). Desai et al. [52] did not explicitly define the fluorescence of single GFPs (I1); instead, the intensity distributions (see Fig 2D) were fit with a series of Gaussians of fixed standard deviation (see S1 Supplementary Material for details of this method). Further experimental measurements are necessary to ultimately resolve which assumption is correct and determine whether or not a single myosin can activate a thin filament (see the Proposed experiments section); but below we present evidence in favor of constant GFP emission.
To support the assumption of constant GFP-S1 fluorescence intensity, at e = 450 fluorescent units/s, we reanalyzed the data from Desai et al [52]. If GFP-S1 fluorescence varied significantly, then the fluorescent spots from single GFP-S1s between conditions could not be rationally rescaled. Using I F = I f / e we rescaled the histograms of fluorescent intensity observed by Desai et al. [52] under three conditions where isolated single GFP-S1s are expected to dominate binding. These conditions are: high ATP (pCa 6, ATP = 1μM, [Myo] = 15nM, f = 3.8Hz), low myosin (pCa 6, ATP = 0.1μM, [Myo] = 1nM, f = 10Hz), and low calcium (pCa 7, ATP = 0.5μM, [Myo] = 15nM, f = 3.8Hz). The resulting histograms are similar, supporting the assumption of equal intensity (Fig 6).
From this study it is clear that the local coupling model agrees well with data recently collected using single molecule fluorescence. The wider agreement of this model with bulk and physiological data suggests this model has the potential to describe and predict systems behavior at multiple levels. Further refinement of this model provided by single molecule data requires challenging one outcome from the experiments: that two heads are required to activate the thin filament. To experimentally validate or refute this outcome, the transition from one to two bound molecules must be observed. However, Desai et al. [52] used a steady-state approach to analyze their data, leaving the time-domain unexplored. Determining if two heads are required for activation can be most effectively done using fast imaging techniques with high spatial resolution. Such experiments can directly determine if the binding of a second head is required to generate activation. This would manifest as a change in the apparent second order binding rate constant once two heads are bound, with no (or little) change if a single head is bound. As an additional benefit, these measurements could also define model parameters. For example, determining the spatial separation between the first two heads to bind and the degree of collaboration provides a direct measure of LC, the critical length over which two myosin molecules communicate.
Desai et al’s [52] conclusion that few isolated GFP-S1s are bound to an activated thin filament not only requires that binding be coordinated, but also requires that unbinding be coordinated. In support of this view, they do not clearly observe stepwise dissociation [52]. The model described in this study predicts that unbinding should occur stepwise and that single GFP-S1s should exist in activated conditions. Coordinated detachment requires that the regulatory proteins act to remove low numbers of myosin molecules, which would challenge the view that regulation affects only myosin’s attachment rate [11]. Again, high spatial and temporal imaging in these challenging conditions would provide a measurement of the process of thin filament relaxation—a process of high relevance to disease [60].
Recent experimental observations [52] of thin filament activation by single myosins were well predicted by a local coupling model described in this study. The importance of this result is that a relatively simple model of local coupling, which can be incorporated into efficient differential equation models, now fits in vitro motility data at high and low calcium [42], in the presence of MyBP-C [43], fiber data [32] and, as we have just shown, direct molecular-scale measurements. This validation of the model across size scales suggests that it captures the essential phenomenology of the local coupling between myosin molecules that occurs upon muscle activation. As it captures this coupling across scales, the model has the potential to provide insight into precisely how molecular level parameters affect macroscopic function—a particularly important problem, since mutations that affect local coupling have been implicated in some genetic cardiomyopathies [61].
|
10.1371/journal.pcbi.1002663 | Dependencies among Editing Sites in Serotonin 2C Receptor mRNA | The serotonin 2C receptor (5-HT2CR)–a key regulator of diverse neurological processes–exhibits functional variability derived from editing of its pre-mRNA by site-specific adenosine deamination (A-to-I pre-mRNA editing) in five distinct sites. Here we describe a statistical technique that was developed for analysis of the dependencies among the editing states of the five sites. The statistical significance of the observed correlations was estimated by comparing editing patterns in multiple individuals. For both human and rat 5-HT2CR, the editing states of the physically proximal sites A and B were found to be strongly dependent. In contrast, the editing states of sites C and D, which are also physically close, seem not to be directly dependent but instead are linked through the dependencies on sites A and B, respectively. We observed pronounced differences between the editing patterns in humans and rats: in humans site A is the key determinant of the editing state of the other sites, whereas in rats this role belongs to site B. The structure of the dependencies among the editing sites is notably simpler in rats than it is in humans implying more complex regulation of 5-HT2CR editing and, by inference, function in the human brain. Thus, exhaustive statistical analysis of the 5-HT2CR editing patterns indicates that the editing state of sites A and B is the primary determinant of the editing states of the other three sites, and hence the overall editing pattern. Taken together, these findings allow us to propose a mechanistic model of concerted action of ADAR1 and ADAR2 in 5-HT2CR editing. Statistical approach developed here can be applied to other cases of interdependencies among modification sites in RNA and proteins.
| The serotonin receptor 2C is a key regulator of diverse neurological processes that affect feeding behavior, sleep, sexual behavior, anxiety and depression. The function of the receptor itself is regulated via so-called pre-mRNA editing, i.e. site-specific adenosine deamination in five distinct sites. The greater the number of edited sites in the serotonin receptor mRNA, the lower the activity of the receptor it encodes. Here we used the results of extensive massively parallel sequencing from human and rat brains to elucidate the dependencies among the editing states of the five sites. Despite the apparent simplicity of the problem, disambiguation of these dependencies is a difficult task that required development of a new statistical technique. We employed this method to analyse the dependencies among editing in the 5 susceptible sites of the receptor mRNA and found that the proximal, juxtaposed sites A and B are strongly interdependent, and that the editing state of these two sites is a major determinant of the editing states of the other three sites, and hence the overall editing pattern. The statistical approach we developed for the analysis of mRNA editing can be applied to other cases of multiple site modification in RNA and proteins.
| The serotonin receptor 2C (5-HT2CR) is widely distributed within the central nervous system [1], [2], where it mediates diverse neurological processes that affect feeding behavior, sleep, sexual activity, anxiety and depression [reviewed in [3], [4]]. The 5-HT2CR protein belongs to the G-protein-coupled receptor (GPCR) superfamily and potentiates multiple signal transduction pathways via several different G proteins (Gαq/11, Gα12/13 and Gαi) to modulate effector molecules such as phospholipases C, D and A2, as well as the extracellular signal-regulated kinases 1 and 2 [reviewed in [5], [6]].
The 5-HT2CR protein exhibits functional variability that is derived from editing of its pre-mRNA by site-specific adenosine deamination (A-to-I pre-mRNA editing) [6]. Editing of 5-HT2CR can produce inosine from adenine at up to five closely-spaced (within a 15 nucleotide segment) position that have been named A, B, E (also known as C'), C, and D sites. Because inosine is read as guanosine by the translational machinery, editing can alter codons for three amino acids in the second intracellular loop of the receptor [7], [8], a region involved in coupling with G-proteins [9]. Combinatorial editing at the five positions can generate up to 32 mRNA variants encoding 24 different receptor isoforms (sites A and B as well as sites E and C are situated in the same codons). The extent of editing is inversely correlated with 5-HT2CR functional activity such that the more highly edited isoforms are less active than less extensively edited ones [reviewed in [6]]. The unedited Ile156-Asn158-Ile160 (INI) isoform possesses considerable constitutive and agonist-stimulated activity. In contrast, when the 5-HT2CR is edited, its coupling to G-proteins and its affinity for serotonin are drastically reduced. Specifically, experiments in heterologous expression systems have shown that, compared to the INI, the fully-edited Val156-Gly158-Val160 (VGV) 5HT2CR isoform (which is edited at all five editing sits) has a 40-fold decreased serotonergic capability to stimulate phosphoinositide hydrolysis due to reduced Gq/11-protein coupling efficiency and decreased coupling to other signaling pathways [7], [10]. In addition, cells expressing more highly edited 5HT2CR isoforms such as VGV demonstrate considerably reduced (or absent) constitutive activity compared to cells expressing the non-edited INI isoform [10]. This reduction in coupling efficiency and constitutive activity derives from a difference in the ability of edited 5-HT2CR isoforms to spontaneously isomerize to the active conformation (R*), a form of the receptor that efficiently interacts with G-proteins in the absence of agonist [11].
A-to-I editing is catalyzed by specific editing enzymes, RNA-specific adenosine deaminases ADAR1 and ADAR2 [reviewed in [12], [13]]. A-to-I editing most frequently occurs in repetitive RNA sequences (e.g., Alu sequences) located within introns and 5′ or 3′ untranslated regions (UTRs). Although the biological significance of non-coding A-to-I RNA editing remains uncertain, the overall editing levels are higher in human compared to primate brains, thus suggesting a possible contribution of editing to the development of higher brain function [14]–[17]. Site-specific edited substrates have been identified in only a few transcripts, including 5-HT2CR mRNA, most of which are expressed in the central nervous system (CNS) and encode proteins involved in neurotransmission [6]. In these protein-coding transcripts, several adenosines are targeted within an imperfect RNA fold-back structure. The features that make RNA prone to site-specific editing are not fully understood, but it is thought that internal mismatches and bulges within double-stranded RNA (dsRNA) are important for the specificity of the ADARs [18]–[20].
Although the specificities of ADAR1 and ADAR2 toward different editing sites often overlap, some sites are edited entirely by one enzyme or the other, and the two enzymes display somewhat different preferences for nearest neighbors of the specific editing sites [19], [21]. Experiments on mouse models with null mutations in one or both ADARs suggest that, within 5-HT2CR mRNA, the A site is predominantly edited by ADAR1 and the D site is mostly edited by ADAR2 [22]–[24]. The other sites have the potential to be edited by both ADAR1 and ADAR2. In addition, it has been proposed that there is crosstalk between ADAR1 and ADAR2, and therefore the relative expression of the different ADARs might ultimately influence the pattern of editing [reviewed in [6]]. The mechanism underlying the putative crosstalk is unclear, but because the five 5-HT2CR editing sites are closely spaced, editing at one site might lead to perturbation of the dsRNA structure that, in turn, would facilitate further editing at other site(s). Indeed, apparent interdependence of editing among the sites has been previously reported for rodent brain [25], [26].
Serotonin signaling, including 5-HT2CR, has been implicated in the etiology of behavioral and psychiatric disorders, and 5-HT2CR is considered an important target for pharmacologic intervention [4]. Several groups have recently reported an association between 5-HT2CR editing and suicide [27]–[31]. Specifically, our studies suggest that in the three major psychiatric diseases (schizophrenia, bipolar disorder, and major depression) that comprise ∼75% of suicides, suicide is associated with enhanced levels of editing (and by inference, with lower activity) of 5-HT2CR in the prefrontal cortex independent of the contributions of the underlying disease [30], [31]. The biological mechanisms that contribute to higher 5-HT2CR editing (and therefore, hypoactive receptors) in suicide compared to non-suicide psychiatric patients remain unclear. However, it seems likely that because enhanced editing decreases 5-HT2CR activity, the resulting reduction in the receptor function might predispose some individuals to suicide by altering 5-HT2CR-dependent signal transduction in critical brain regions. Thus, altered editing mechanisms might be linked to liability for suicide, and detailed understanding of these mechanisms could facilitate the development of unique pharmacological strategies that target suicidal behavior.
Alteration of the 5HT2CR function via editing has also been reported in response to spinal cord injury (SCI) in rats [32]. Muscle paralysis after SCI is partly caused by a loss of all brainstem-derived neurotransmitters (including serotonin), which normally modulate motoneuron excitability. Murray et al. examined how motoneurons in the spinal cord of the SCI rats compensated for lost brain-derived neurotransmitters to regain excitability and found that changes in 5-HT2CR mRNA editing led to increased expression of the 5-HT2CR isoforms that are active without serotonin n [32]. Such constitutive receptor activity restored excitability of the motoneurons in the SCI rats in the absence of serotonin, helping motoneurons recover their ability to produce sustained tail muscle contractions. Accordingly, blocking constitutively active 5-HT2CR with specific drugs (SB206553 or cyproheptadine) largely eliminated these calcium currents and muscle spasms, providing a new rationale for antispastic drug therapy.
Recently, we applied the Massively Parallel Sequencing (MPS) technology to quantify 5-HT2CR editing in the postmortem human brain and the rat spinal cord specimens [31], [33]. The traditional cloning and sequencing approach [7], [34] relies on sampling a limited population of cloned transcripts (∼20–100), thus producing significant sampling errors that can obscure differences between experimental groups. The use of MPS, which analyzes several hundred thousand 5-HT2CR transcripts per specimen, not only allowed us to detect all 32 mRNA variants of 5-HT2CR in both species, but substantially increased precision and sensitivity in measuring 5-HT2CR editing frequencies for all these mRNA variants. Specifically, a comparison between MPS (over 730,000 reads per subject) and the traditional method (46 clones per subject), performed for the same human subjects and the same brain region, has shown that the mean coefficient of variation of the editing frequencies of all variants in the NGS analysis was approximately one-third that of the traditional method [31].
Here we use the MPS data generated in these recent studies on 5-HT2CR editing in the human and rat CNS specimens to comprehensively characterize the dependencies among the 5 different editing sites in the 5-HT2CR mRNA. The extremely high number of sequenced transcripts combined with the use of a newly developed rigorous statistical procedure allowed us to elucidate the fine structure of these interactions and compare them between the two species as well as among individuals.
5-HT2CR mRNA editing was measured in the specimens obtained from the human dorsolateral prefrontal cortex and rat spinal cord. The 101 human subjects comprised 45 individuals diagnosed with major depression, and 56 normal controls [31]. The 19 rats comprised 7 controls and 12 rats whose spinal cord was transacted six weeks prior to the data collection [33]. In these rat specimens, the mRNA levels are assumed to be unaffected by the transaction, being collected from a region above it. Overall, the analysed data included 56,690,398 human reads (an average of 561,291 per subject) and 5,659,108 rat reads (an average of 297,848 per rat) (Supplementary Table S1).
Each measurement (mRNA molecule) is represented by a binary vector indicating the editing states of the five sites A, B, E, C, and D. For example, a measurement in which sites A, B, and D are edited but E and C are not is represented by the binary pattern 11001. For a collection of measurements, we denote the editing pattern as the vector , where is the number of binary vectors whose decimal representation is .
First, we tested whether the editing patterns of all human normal controls were statistically indistinguishable from the editing patterns of all subjects with major depression. To this end, we conducted a conservative randomization test, whereby the -test statistic was repeatedly computed on modified data. In each repetition, we randomly assigned subjects as normal or as depressed, keeping the total number of normal controls and the total number of subjects with major depression fixed. For each repetition, we computed the test statistic of the -test,where and are the editing patterns of normal and depressed samples, respectively, and and are the total number of measurements from normal controls and from subjects with major depression, respectively. This procedure was repeated 106 times, and the p-value of the test was computed as the number of random test statistics that were larger than the true test statistic. A similar procedure was used to compare normal rats with transacted ones. We found that the editing pattern in normal human controls was indistinguishable from the editing pattern in subjects with major depression (P = 0.80), and that the editing pattern in normal rats was indistinguishable from that in transacted rats (P = 0.65). This result justifies pooling together all human subjects and all rats for further analysis. Using a similar randomization procedure, we found that the editing pattern in humans is very different from that in rats (P<10−6) (Supplementary Figure S1).
Next, we tested for each pair of sites whether their editing patterns were correlated. To this end, we computed the φ-coefficient (which, for binary data, is simply the correlation coefficient; see Methods), and found that all pairs of sites are correlated, either positively or negatively, except for the pair (D,E) in human, and the pair (A,E) in rat (Supplementary Table S2).
In order to obtain more detail on the level of dependence between different sites, we followed Ensterö et al. [26] and clustered the editing sites (Figure 1). We used the Jaccard distance coupled to single linkage hierarchical clustering (see Methods), but using Dice distance following Ensterö et al. [26] had no significant effect on the clustering (Supplementary Figure S2). In order to assign confidence level to the clusters, we repeated the clustering for each individual and measured the fraction of cases in which the cluster was supported (see Methods). In both human and rat the strongest association was observed to exist between sites A and B, to which site D joins next. Sites C and E were more weakly associated with the rest of the editing sites, at least in human, and the order by which they join the dendrogram changed from human to rat.
Clustering, by nature, identifies groups of associated sites. However, to obtain finer resolution of the relationship between the sites, we resorted to more elaborate methods. The ultimate description of the dependency between the editing sites would be their joint probability distribution. For five editing sites, there are 8,782 possible joint distribution functions. We enumerated all the 8,782 functions, and ranked them according to how well they fit the data using both maximum-likelihood and Bayesian inference (see Methods).
In both human and rat, and for both maximum-likelihood and Bayesian inference, the best model was the maximally-dependent joint probability distribution, . We graphically represent probability models by a pDAG (partial Directed Acyclic Graph), which is a Bayesian network containing a mixture of directed and undirected edges (see Methods). The pDAG of this maximally-dependent model is simply the fully connected undirected graph (Figures 2, 3). This result is consistent with our previous finding that all pairs of sites are significantly dependent.
Such result is expected given the large size of the data. In order to find which edges in the graph are more strongly supported by the data, we divided all the 8,782 probability models into 11 groups according to the number of edges in the corresponding pDAG. The first group consists of all models with zero edges (which is simply the single model ), the second group consists of all (ten) models with one edge, etc. Then, we computed the best-fitting probabilistic model within each group. Hereinafter, denotes the best-fitting model from within the group of models with edges. The results for the maximum-likelihood Bayesian Information Criterion (BIC) score (see Methods) in human are shown in Figure 2. Adding an edge to a model always improves its score, , but the improvement becomes smaller as increases. The estimated parameters of each of the best-fitting models are given in Supplementary Table S3.
To further explore the relative impact of the different edges, we ranked the edges by the order in which they first appear in the sequence of models . Specifically, the rank of an edge is the smallest integer for which contains this edge (Table 1). Importantly, an edge in need not necessarily be included in , which is the reason why two edges – (B,E) and (E,C) – have the same rank, and why no new edge appeared in . To account for the possibility that edges may disappear or reappear as the number of edges grows, we define for each edge its support. The support of an edge with rank is the fraction of the models that contain this edge (Table 1). Clearly, the higher the support, the more confident we are that the edge has a unique contribution to making the respective model better fitting the data. The edge (A,B) is the first to appear (), and has a full support (it appears in all the models through ), indicating that the dependence between A and B is obviously the strongest among all pairs of sites, in accord with the findings described above. Next appear edges (B,D) and (A,C) that both also have full support. This observation is consistent with the clustering analysis results (Figure 1) but provides more detail on the interdependencies among A, B, and C, D. The next edge to appear is (A,E), but it does not have full support which lowers our confidence in its unique contribution to the score of the best model. The edges that appear in models and – (C,D) and (E,D) –make (at least qualitatively) only marginal contributions to the score. Repeating the analysis with the maximum-likelihood Akaike Information Criterion (AIC) scores, or with Bayesian scores, gave the same series of best-fitting models to (Supplementary Figure S3).
We conducted the same analysis for the rat data. Here, too, adding edges kept improving the BIC score of the model (Figure 3). The estimated parameters of the best-fitting models are given in Supplementary Table S4. For rat, all edges have full support, which means that an edge with rank appears in all the models to (Table 2). In rat, the two edges with the lowest rank – (A,B) and (B,D) – have the same rank as in human. However, the edge with in rat is (B,C) as opposed to (A,C) in the equivalent human model. Similarly, the edge with is (A,E) in human, but it is (B,E) in rat. This suggests that the central role of site A in governing the editing state of sites E, C, and D in human is taken by site B in rat. Indeed, referring collectively to site A or site B as F, human and rat show very similar edge rankings (compare Tables 1 and 2). Repeating the analysis with Bayes scores yielded identical series of best-fitting models for rat (Supplementary Figure S4b). Using AIC scores produced only a single difference, in model . The edge (C,D), which is present in the BIC and Bayes scores, was replaced by the edge (E,C) for the AIC score (Supplementary Figures S4a and S5). However, as we have seen, these edges anyway have marginal contribution to the best-fitting model. On the whole, the information contribution of additional edges dropped much faster for the rat data than it did for the human data (compare Figures 2 and 3), suggestive of a more complex pattern of dependencies among editing sites and accordingly more subtle regulation of the editing process in human brain.
The above analysis lacks measure of score variance, thus hindering quantitative evaluation of the significance of each edge to the total score. To overcome this, we repeated the analysis for each individual separately, for both human and rat. In this way, each individual provides its own sequence of best fitting models , and for each number of edges (), there is now a sequence of best models, where is the total number of individuals. For a certain , let individuals support different best-models, , such that is supported by individuals. Let us further assume that we have sorted the sequence according to the level of support, such that . Next, we define a set of models that are equally supported by the different individuals (). To this end we make a Bonferroni-corrected series of proportion tests, asking whether is supported significantly more than the other best-fitting models . is the first model whose support is significantly lower than that of . The results for the BIC scores in human at significance level 0.05 are given in Table 3. The results for the AIC and Bayes scores are similar, and are given in Supplementary Tables S5 and S6. As an example, in the BIC score analysis, out of 101 individuals 78 (77.2%) support the single-edge best-fitting model (A→B) (Table 3). The second-supported model is supported by 21 individuals (20.8%), which is significantly lower than the support for and so in this case and the set of best fitting models is simply (). As another example, is supported by 14 individuals (13.9%), but this level of support is not statistically different from the support by 3 individuals (3.0%) of the model , and in this case . Overall, there is a good agreement between this individual-based analysis and the pooled analysis. The pooled best-fitting model for each is marked by asterisk in Table 3 and Supplementary Tables S5 and S6, and it is always within the group of models that are equally supported by the different individuals. Very similar results had been obtained for rat (Supplementary Tables S7, S8, S9). Here too, the pooled best-fitting model for each is always within the group of models that are equally supported by the different individuals.
The individual-based approach can be used not only to re-evaluate the support for the different graphical models but also to perform an edge-by-edge analysis. To this end, we can look at each edge, and count how many times (in either direction) it appears in the sequence . These counts are binomial random variables, so if an edge appears, overall, in the best-fitting model of individuals out of a total of individuals, its variance is . The support of each edge for any in human is given in Figure 4. Consider, for example, . Overall, the 101 individuals support different best-fitting models. Yet, the edge (A,B) appears in all of them, and thus is supported by all the individuals. The edge (B,D) is supported by 98 individuals, or by 97% of the best-fitting models. For each , we can take the first most-supported edges as the basic set of edges in the model. Then, we can check how unique is this set of edges by testing (using proportion test) whether the e'th supported edge is significantly more supported than the next edges (Table 4). From Table 4 and Figure 4 we see that the first three edges (A,B), (B,D), and (A,C) are clearly more supported than all other edges, in this order. However, the next edges (B,E), (E,C), and (A,D) all have approximately the same support and no one is more significant than the others. The results are almost identical when using AIC or Bayes scores (Supplementary Figures S6 and S7).
Similar analysis for rat shows, in accord with our previous results, a more hierarchical relationship between the edges (Figure 5 and Table 5). Here, the order of importance is clear for the first six edges: (A,B), (B,D), (B,C), (B,E), (A,C), and (A,D). The edges (C,D), (E,D), and (E,C) all have approximately the same support and no one is more significant than the others. The results are almost identical when using AIC or Bayes scores (Supplementary Figures S8 and S9).
Here we analysed interdependencies among editing sites within mRNA of 5-HT2CR. The studies were performed using available data sets for the human dorsolateral prefrontal cortex and rat spinal cord tissues. Alterations in 5-HT2CR editing in these particular species and CNS regions were reported in connection to completed suicide and in response to SCI, respectively [30]–[32]. Thus, detailed understanding of editing mechanisms in these particular areas of the human and mouse CNS are expected to aid in the development of unique pharmacological strategies that target suicidal behavior as well as SCI-related spasticity.
The dependencies among editing sites described here allow us to propose a hypothetical mechanistic model for the concerted action of ADAR1 and ADAR2 in 5-HT2CR editing. Given that the dependence between sites A and B was by far the strongest revealed (see Figures 1–5) and that these sites are adjacent in 5-HT2CR mRNA, we speculate that ADAR1 that is known to be responsible for editing at A [22] also edits B. Moreover, the strong connection between sites A and B mechanistically might stem from editing of both sites by the same ADAR1 molecule without dissociation of the enzyme from the mRNA (Figure 6). Given that site D, known to be edited by ADAR2 [22], is next after sites A and B in terms of the strength of the dependency, followed by site C, we further speculate that editing of sites A and B by ADAR1 affects the RNA structure such that binding of ADAR2 followed by editing at site D and possibly the two remaining sites is enhanced (Figure 6). A more far reaching implication is that the apparent primary role of ADAR1 in 5-HT2CR editing makes it the most attractive target for pharmacological intervention in the associated psychiatric disorders. It is worth noting that such intervention would not interfere with the essential editing of the GluR2 subunit of the AMPA receptor that is primarily dependent on ADAR2 [35].
The results reported here show that for both human and rat 5-HT2CR, the editing states of the physically proximal sites A and B are highly dependent. In contrast, the editing states of sites C and D, which are also physically close, seem not to be directly dependent, but rather indirectly linked through the dependencies of C and D on sites A and B, respectively. The results also reveal pronounced differences between the editing patterns in humans and rats: in humans site A has the key role in determining the editing state of the other sites whereas in rats this role belongs to site B. Although not detected by the simple analysis of the dependencies among the editing sites, computing the best-fitting probabilistic models shows that the editing state of site E is strongly dependent on the state of site A in human and on the state of site B in rat (Tables 1 and 2). Furthermore, the structure of the dependences between the editing sites is simpler in rats than it is in human implying more complex regulation of 5-HT2CR editing and by inference function in human brain. Mechanistically, the differences between the emerging patterns of editing regulation in humans and rats could be underpinned by the notable differences in the predicted secondary structures of the respective pre-mRNA regions [6].
To conclude, the results of the exhaustive analysis of 5-HT2CR editing patterns described here indicate that sites A and B strongly depend on each other in both human and rat, and that the editing state of these two sites is a key determinant of the editing state of the other three sites, and hence the overall editing pattern. The direct dependencies among the editing states of sites E, C, and D are much weaker, and the observed dependencies are probably an indirect effect of the dependency of those three sites on editing in sites A and B. Taken together, these findings allowed us to propose a mechanistic model of concerted action of ADAR1 and ADAR2 in 5-HT2CR editing. Methods of statistical inference developed here can be applied to other cases of interdependencies among multiple modification sites in RNA and proteins.
The editing state of the five editing sites (A,B,E,C,D) in a single mRNA molecule is represented by a 5-digit binary vector, with one designating an edited site and zero designating an unedited site. The data comprise measurements of the editing state of all five editing sites in mRNA molecules.
We tested whether the editing state of a pair of sites and is correlated by computing the contingency tablefor each individual, where is the number of mRNA molecules in which and () in that individual. Then, the φ-coefficient was computed.This computation was repeated for all possible pairs in all individuals. Grouping the values from all individuals, the mean and standard deviation were computed for the φ-coefficient for each pair of sites, and the z-test was used to test for significance.
We defined the distance between sites and as the Jaccard distance between their binary patterns,This distance was computed for all pairs of editing sites, and the distance matrix served as input for a single linkage (shortest-distance) hierarchical clustering. Using the Dice distance, as in [26],had a negligible effect on the results (Figure 1, Supplementary Figure S2). Edges were given support between 0 and 1 according to the number of individuals in which they are supported.
The editing state of a site is a random variable. Thus, the joint probability distribution is the ultimate description of the dependencies among the five serotonin receptor editing sites. Any joint probability distribution can be decomposed in many different ways as a product of conditional and marginal probabilities, where each decomposition may represent different dependencies among the sites. For example, the joint probability distribution of two random variables can be decomposed in three ways: , , and . The first two models represent dependency between and , whereas the third model represents independence between and . There is a recursive formula for computing the number of possible decomposition for any given number of random variables [36]. In our case, the joint probability distribution of five random variables can be decomposed in 29,281 different ways. Importantly, many of these decompositions are redundant in the sense that several decompositions can describe essentially the same probabilistic model. In the two-random variable example above, Bayes law renders equivalence between the first two decompositions, . A set of equivalent decompositions is denoted equivalence class. There is no known general formula to compute the number of equivalence classes for a given number of random variables. However, there is an algorithm allowing one to tell, given two decompositions, whether they belong to the same equivalence class or not [37].
Here we propose a technique to find the joint probability distribution that fits best to the data. This technique, being exponential with the number of editing sites, is useful when there is a small number of editing sites, as in the present case and in several other functionally important cases of mRNA editing (e.g., kainate 2 glutamate receptor or CaV1.3 channel) [38], [39]. In a nutshell, we scanned through the entire set of 29,281 possible decompositions, and constructed the full set of equivalence classes. Then, we tested which of the equivalence classes fits the data best (see details below).
In order to enumerate all the possible decompositions of , we used Steinsky's ranking algorithm, that allows for a one-to-one mapping between the set of all decompositions and the integers 0,1,2,…, [36]. Then, we scanned through the list of decompositions by a series of pairwise comparisons, and kept only a single decomposition from each equivalence class. In this way, we found that the joint probability distribution of five random variables can be decomposed into 8,782 equivalence classes (Supplementary Table S1).
A Bayesian network provides a compact graphical representation of a decomposition. It is a directed acyclic graph (DAG) in which the nodes are the random variables, and an edge leading from a node to each of its children (a parent of a node is a node upon which is conditionally dependent in the decomposition). In the context of Bayesian networks, the collection of DAGs that represent equivalence class is called Markov equivalence class. For convenience, we shall hereinafter use probabilistic model as a synonym to equivalence class or to Markov equivalence class. Bayesian networks have been proved as a very efficient tool to facilitate calculations on probabilistic models.
In order to score how well each probability model fits the observed data, we used two alternative scoring methods. The first is based on a maximum-likelihood (ML) procedure, and the second is based on Bayesian inference. Below, we describe both methods.
The Bayesian learning formalism requires assumptions about the prior probability of the parameters . We used the Dirichlet priors, which is the standard choice of priors in this kind of problems because it bears desirable properties such as global and local parameter independence [40]. For each node , and for each editing state of its parents , the Dirichlet priors are specified by two parameters that we denote and . The use of these priors can be conceived as adding another pseudo-measurements to the observed measurements, where is the number of pseudo-measurements in which and the editing state of is , and is the number of pseudo-measurements in which and the editing state of is . We denote by the number of pseudo-measurements in which the editing state of is , . The Bayesian Score (BS) of a probabilistic model is given bywhere the summation is over all the nodes, andwhere the product is over all possible editing statees of , and is the gamma function [40].
We generated the set of pseudo-measurements to consist exactly one of each of the possible editing statees of the five editing sites. That is, the pseudo-measurements consist a single measurement 00000, a single measurement 00001, etc. If we denote the number of editing sites by (), then the set of pseudo-measurements consists of measurements. If we denote the number of parents of node by , then , and . This giveswhich is justIf node has no parents, then , , , and the formula further simplifies to
A whole Markov equivalence class can be described by a partial-directed acyclic graph (pDAG) [40], [41], which is a graph made of both directed and undirected edges. If an edge can be oriented differently in DAGs belonging to the same Markov equivalence class, it would be undirected. In this work, whenever a probabilistic model is visualized as a Bayesian network, the pDAG representation is employed.
|
10.1371/journal.ppat.1008040 | HLA-B locus products resist degradation by the human cytomegalovirus immunoevasin US11 | To escape CD8+ T-cell immunity, human cytomegalovirus (HCMV) US11 redirects MHC-I for rapid ER-associated proteolytic degradation (ERAD). In humans, classical MHC-I molecules are encoded by the highly polymorphic HLA-A, -B and -C gene loci. While HLA-C resists US11 degradation, the specificity for HLA-A and HLA-B products has not been systematically studied. In this study we analyzed the MHC-I peptide ligands in HCMV-infected cells. A US11-dependent loss of HLA-A ligands was observed, but not of HLA-B. We revealed a general ability of HLA-B to assemble with β2m and exit from the ER in the presence of US11. Surprisingly, a low-complexity region between the signal peptide sequence and the Ig-like domain of US11, was necessary to form a stable interaction with assembled MHC-I and, moreover, this region was also responsible for changing the pool of HLA-B ligands. Our data suggest a two-pronged strategy by US11 to escape CD8+ T-cell immunity, firstly, by degrading HLA-A molecules, and secondly, by manipulating the HLA-B ligandome.
| The human immune system can cover the presentation of a wide array of pathogen derived antigens owing to the three extraordinary polymorphic MHC class I (MHC-I) gene loci, called HLA-A, -B and -C in humans. Studying the HLA peptide ligands of human cytomegalovirus (HCMV) infected cells, we realized that the HCMV encoded glycoprotein US11 targeted different HLA gene products in distinct manners. More than 20 years ago the first HCMV encoded MHC-I inhibitors were identified, including US11, targeting MHC-I for proteasomal degradation. Here, we describe that the prime target for US11-mediated degradation is HLA-A, whereas HLA-B can resist degradation. Our further mechanistic analysis revealed that US11 uses various domains for distinct functions. Remarkably, the ability of US11 to interact with assembled MHC-I and modify peptide loading of degradation-resistant HLA-B was dependent on a low-complexity region (LCR) located between the signal peptide and the immunoglobulin-like domain of US11. To redirect MHC-I for proteasomal degradation the LCR was dispensable. These findings now raise the intriguing question why US11 has evolved to target HLA-A and -B differentially. Possibly, HLA-B molecules are spared in order to dampen NK cell attack against infected cells.
| Human cytomegalovirus (HCMV) represents a prototypic β-herpesvirus persisting throughout life in its host with periodic phases of latency and reactivation of productive infection. Despite rare cases of clinical disease in healthy individuals, HCMV has a permanent impact on immune cells, e.g. resulting in the extraordinary expansion of both CD8+ memory T-cells and memory-like NK cells [1, 2]. As demonstrated in CMV animal models, protection from CMV disease is strongly dependent on MHC class I (MHC-I) restricted CD8+ T-cell responses [3].
MHC-I molecules are dimers formed by the membrane attached heavy chain (HC) and the soluble beta-2-microglobulin (β2m). Upon assembly in the ER the heterodimeric MHC-I molecule is recruited to the peptide loading complex (PLC), composed of the transporter associated with antigen processing (TAP) and the chaperones tapasin, ERp57 and calreticulin [4, 5]. In the course of loading an optimal peptide, the trimolecular MHC-I complex is released for transport through the secretory pathway and surface expression.
The US6 gene family region of HCMV encodes for several immunoevasins that target the MHC-I antigen presentation pathway at different stages of the protracted HCMV replication cycle [6]. US3 blocks maturation of MHC-I and interferes with tapasin function, US2 and US11 target MHC-I for rapid proteasomal degradation, and US6 inhibits peptide translocation by TAP [7–13].
Of the reported HCMV encoded MHC-I inhibitors, so far a crystal structure exists only for US2 in complex with HLA-A*02:01 [14]. Similar to US2, US11 binds to MHC-I with its ectodomain, but the contact site on MHC-I has not been defined [15, 16]. The lack of insight into the contacts sites is mirrored by the poor understanding of the HLA allotype specificity of US11. Whereas varying effects of US11 on HLA-B and -C allotypes have been reported [17–20], consistent downregulation of HLA-A allotypes was observed [19–21].
HLA-A*02:01 has been an instrumental substrate to elucidate ER associated degradation (ERAD) pathways activated by US11. Upon binding to MHC-I, a glutamine in the transmembrane segment of US11 recruits Derlin-1 [15, 22]. Derlin-1 is a crucial component of an ERAD pathway also including the recently identified E3 ligase TMEM129 and the E2 ligase Ube2J2, required for ubiquitination and subsequent degradation of MHC-I [23, 24]. Whereas US11 itself is not degraded by this pathway, low MHC-I expression exposes US11 to an alternative degradation pathway including HRD1 and SEL1L [24, 25].
The genes of the US6 family most likely evolved in a cytomegalovirus ancestor prior to the split of Old World monkeys and hominoids, as homologs of these genes are found also in chimpanzee and rhesus CMV (rhCMV) [26, 27]. In the context of a rhCMV vaccine vector [28], in addition to degradation of MHC-I, it was observed that the rhCMV US11 homolog Rh189 is able to suppress the CD8+ T-cell responses to canonical epitopes [29]. Although the data are not formally published, Früh and Picker mention in a recent overview publication that substitution of Rh189 with HCMV US11 in a rhCMV mutant maintains the ability of the virus to block canonical CD8+ T-cell responses [28], suggesting that in addition to MHC-I degradation, US11 proteins execute further manipulation of MHC-I antigen presentation.
When studying the MHC-I ligandome in HCMV-infected cells, we noticed a striking differential effect of US11 on HLA class I locus products. Ligands from HLA-B and -C molecules remained largely unaltered in the presence of US11, while HLA-A ligands were efficiently eliminated. Further investigations reported here revealed that US11 targets HLA-B by manipulating the quality of the peptide ligands. This illustrates the specific role of a single MHC-I immunoevasin that not randomly degrades MHC-I molecules, but has evolved to exert HLA locus-specific functions.
To gain deeper insight into the identity and amount of HCMV peptides presented by MHC-I we undertook MHC-I ligandome analysis. MRC-5 fibroblasts were infected with HCMV AD169VarL strain derived mutants lacking either the MHC-I inhibitor region US2-US6 (ΔUS2-6) or US2-US6 plus US11 (ΔUS2-6/US11), leaving the US7-US10 region intact as demonstrated by the expression of the neighboring gene US10 in ΔUS2-6/US11 infected cells (S1 Fig). At 48 h post-infection cells were harvested and MHC-I complexes were isolated by the pan-MHC-I-reactive monoclonal antibody W6/32. Identification and relative quantification of HLA ligands was performed by LC-MS/MS. Much to our surprise, a clear difference in the ability of US11 to target HLA-A and HLA-B molecules was observed. About 90% of all peptide ligands significantly down-regulated by US11 (ca 22% of total pool; Fig 1A, blue dots) was derived from HLA-A*02:01 and A*29:02 molecules (Fig 1B, left panel). Unexpectedly, some peptides (ca 11% of the total pool) were found to be increased in cells infected with the ΔUS2-6 mutant (Fig 1A, red dots). The majority (ca 80%) of these peptides could be attributed to HLA-B*44:02 (Fig 1B, right panel). No major changes were observed for the ligandome of HLA-B*07:02. To reassure that our observations were not distorted by ligands from uninfected cells possibly present in the culture, we analyzed the anchor residues of identified HCMV-derived peptides to predict their HLA-I specificities. The distribution of these viral peptides showed the same HLA-I pattern as the total ligandome, but more pronounced; US11 almost completely abolished viral ligands of HLA-A, but not of HLA-B. Again, HLA-B*44:02 ligands increased in the presence of US11 (S2B Fig).
Since the protocol for ligandome analysis does not differentiate between surface and intracellular MHC-I molecules we next conducted flow cytometry analysis of MRC-5 fibroblasts mock treated or infected with the ΔUS2-6 or ΔUS2-6/US11 deletion viruses, to determine the surface levels of HLA-A and HLA-B using available specific antibodies for HLA-A*02:01, B*07:02 and B*44:02. Whereas less than two-fold reduction for HLA-B surface expression was observed (Fig 1C and 1D), HLA-A*02:01 was reduced 3-fold in the presence of US11 as compared to infection with the virus lacking US11. Therefore, also on the surface on HCMV-infected fibroblasts, a stronger effect of US11 was measured for the HLA-A allotype A*02:01 than for the HLA-B allotypes B*07:02 and B*44:02. Stronger regulation of HLA-A*02 was also observed on ARPE19 epithelial cells infected with the TB40 derived ΔUS2-6 BAC virus. Compared to mock-treated cells HLA-A*02 was downregulated 5-fold and HLA-B*07 1.5-fold (S1E Fig), indicating that different regulation of HLA-A*02:01 and HLA-B*07:02 by US11 is not a fibroblast adapted function of HCMV.
We wondered whether the strong resistance of HLA-B*07:02 would also hold true in the context of non-infected cells and therefore N-terminally HA-tagged (HA-tagged molecules are indicated with ~ in all following figures) HLA-A*02:01, A*03:01, B*07:02 and CD99 were cloned into a lentiviral vector. HeLa cells were first transduced and selected to express the MHC-I molecules or the control protein CD99 and subsequently transduced with a lentivirus encoding US11 in front of an IRES and EGFP sequence. Cell surface expression of HA-tagged MHC-I and CD99 was analyzed on EGFP positive cells in comparison to EGFP negative cells at 0, 24 and 48 h after transduction. Remarkably, in this context, similarly to the control protein CD99, HLA-B*07:02 appeared completely unaffected by US11, but not HLA-A*02:01 and A*03:01 (Fig 1E), demonstrating that resistance of HLA-B*07:02 against US11 is independent of other viral or virally induced proteins. In infected cells, US11 may work in concert with other proteins and could explain the stronger effect of US11 on HLA-B in infected MRC-5 cells. HCMV also reorganizes the secretory pathway in infected cells [30], which could further influence trafficking and surface expression of MHC-I allotypes in the presence of US11.
To gain insight into the breadth of the resistance of HLA-B locus products against US11, we cloned various HLA-A and -B sequences with an N-terminal HA-tag und analyzed the cell surface level in the absence and presence of US11 after transient transfection of HeLa cells. Usage of the CMV major IE promoter for MHC-I expression leads to only low level of MHC-I regulation (probably due to overload of the ER and insufficient levels of ERAD-associated proteins required for US11 function) and we therefore proceeded with a different vector harboring a less potent promoter (spleen focus-forming virus (SFFV) U3 promoter) to be able to measure an adequate level of US11-mediated downregulation. Using this approach, all HLA-A allotypes were strongly downregulated by US11 (Fig 2A and 2B), while no significant reduction of HLA-B cell surface expression was noted. However, HLA-B*44:02, from which more diverse peptide ligands were eluted in the presence of US11 in infected cells (Fig 1B, right panel), displayed an induced level of surface expression in US11-expressing cells. Altogether, this suggests that the observed resistance towards US11-mediated downregulation is a general feature of HLA-B locus products.
Next, pulse-chase experiments were conducted to measure the stability of MHC-I molecules in the absence and presence of US11. As expected, a clear destabilization of HLA-A*02:01 and A*03:01 could be observed. Surprisingly, also the stability of HLA-B HCs was strongly reduced in the presence of US11 (Fig 2C and S4 Fig, using shorter labeling and chase times). However, in contrast to HLA-A, HLA-B allotypes were able to dimerize with β2m in the presence of US11 and maturation of these molecules could be observed as a slightly slower migrating HC band (likely due to glycan modifications) at 45 min of chase (Fig 2C). This observation suggests, that different from HLA-A, HLA-B molecules are able to exit the ER and accumulate at the cell surface in the presence of US11. We therefore analyzed the steady-state level of EndoH (Endoglycosidase H; digests Asn-linked glycans that have not been subjected to modification in the Golgi apparatus) resistant molecules by Western blot and indeed observed that although the total level of HLA-B was reduced in the presence of US11, the EndoH resistant molecules were largely unchanged compared to control cells. This was not the case for HLA-A, as a complete loss of EndoH resistant molecules was observed (Fig 2D). We asked whether HLA-B surface expression in the presence of US11 could be more readily recognized by an inhibitory MHC-I receptor such as LIR1. Indeed, LIR1-Fc incubated with HeLa cells treated as described above showed a clear binding to HLA-B*07:02 expressing cells despite co-transfection of US11, whereas this was not the case for HLA*02:01 and -A*03:01 expressing cells (Fig 2E).
In conclusion, HLA-A and also HLA-B molecules are targets for US11-mediated degradation. However, different from HLA-A, a fraction of the HLA-B molecules can escape degradation and be expressed on the cell surface.
In the pulse-chase experiment we observed an apparent co-immunoprecipitation of US11 with all studied types of MHC-I molecules (S3 and S4 Figs). This suggested that although HLA-B molecules escape downregulation by US11, they are still bound by US11 efficiently. To depict this more clearly, we took advantage of the fact that MHC-I molecules expressed transiently from a strong promoter (CMV IE promoter) remain largely stable in the presence of US11. Under these conditions we assessed binding of US11 to HLA-A*02:01, A*03:01, B*07:02 and B*15:03 (Fig 3A). With US11 and MHC-I being strongly overexpressed at saturating levels, US11 bound similarly to all MHC-I molecules.
We next assessed the possibility that the cytosolic tail of HLA-A allotypes, which is three residues (CysLysVal) longer than that of HLA-B allotypes, could be decisive for differential US11 regulation. The C-terminal Val has previously been described to be important for US11-mediated degradation [31]. To this end, we constructed an HLA-A*03:01 mutant without these residues (A3ΔCKV) and compared it to the reciprocal HLA-B*07:02 mutant (B7+CKV). The residues CysLysVal had only small and non-significant effects on surface expression in US11 expressing cells as measured by flow cytometry (Fig 3B and 3C). Even though HLA-B*07 was more efficiently downregulated, when expressed with C-terminal CysLysVal residues, downregulation was significantly different from HLA-A*03:01. Therefore, the C-terminal CysLysVal residues are not the major determinant for the difference in US11-mediated regulation.
We next scrutinized the role of β2m in the process of US11-mediated degradation, since the HLA-B molecules HLA-B8 and -B5 were reported to possess a higher affinity for β2m compared to HLA-A1 and -A2 [32]. In FO-1 cells MHC-I is not expressed on the surface due to lack of β2m. Co-transfection of a plasmid encoding β2m rescued cell surface expression of MHC-I (Fig 3D). Similarly to HeLa cells, in FO-1 cells transfected with a β2m encoding plasmid, HLA-A*03:01 was downregulated by US11, whereas HLA-B*07:02 was not (Fig 3D). Also the stability of HLA-B*07:02 was higher compared to A*03:01, and this was not influenced by β2m (Fig 3E), implying that the resistance of HLA-B against US11 is conferred at the stage of unassembled HC.
In conclusion, although US11 binding to HLA-B is preserved, downregulation of HLA-B is much less efficient and this is only partly due to the shorter cytosolic tail of HLA-B alloforms.
In our previous studies of the PLC composition in HCMV-infected cells [33], co-immunoprecipitation experiments suggested that US11 interacts with MHC-I as part of the the PLC. This was unexpected, as the concept for MHC-I targeting by US11 has been a rapid degradation facilitated by ERAD. However, in light of our observation, that HLA-B molecules display intrinsic resistance against US11-mediated degradation, we supposed that US11 pursues two different strategies when interacting with HLA-A and HLA-B, respectively. To re-investigate our earlier findings, MRC-5 fibroblasts were treated with siRNA targeting US11 or control siRNA and subsequently infected with the ΔUS2-6 HCMV deletion mutant. At 24 h post-infection a co-immunoprecipitation experiment was performed using W6/32 or anti-ERp57 antibodies. As expected, a band corresponding to the size of US11 was found in complex with both MHC-I and ERp57, which was not present in US11 siRNA treated cells (Fig 4A). To confirm the identity of the co-immunoprecipitated protein, MRC5 cells were mock treated or infected with the HCMV deletion mutants ΔUS2-6 and ΔUS2-6/US11 and a re-immunoprecipitation was performed. After dissociation of proteins immunoprecipitated with an anti-ERp57 antibody, an anti-US11 antiserum was applied. A protein, the size of US11, was co-immunoprecipitated from the ΔUS2-6 sample, but not from the ΔUS2-6/US11 sample (Fig 4B), strongly indicating that US11 interacts with the PLC in HCMV infected cells. Therefore, our data imply, that US11 binds both unassembled MHC-I HCs and, in addition, MHC-I assembled in the PLC. Possibly, US11 binds to a structure of MHC-I that is not changed during the transition from unassembled to assembled form, e.g. in the alpha-3 domain. Alternatively, US11 interacts with MHC-I in different manners, one that can target the unassembled MHC-I for degradation and another that leads to a prolonged interaction with assembled MHC-I in the PLC.
In this regard we found it interesting, that US11 has a predicted low-complexity region (LCR; amino acids 28–42 [34]) N-terminal of the Ig-like domain. Since LCRs tend to be engaged in protein-protein interactions [35], we set out to investigate the importance of this domain for US11 function and interaction with MHC-I. To this end, we deleted the N-terminus (amino acids 20–44 of the full-length protein) with the LCR from US11 wild-type and from a US11 Gln192Ala (US11Q/A) mutant (schematic illustration in Fig 5A). The Gln192Ala mutation prevents the recruitment of Derlin-1 and subsequent MHC-I degradation, which leads to retention of MHC-I in the ER [22]. Therefore, this modification of US11 allows for interaction studies, circumventing the issue of losing substrates due to degradation. The HA-tagged US11 versions were stably transduced into HeLa cells and the cells were first checked for steady-state level expression of US11 and MHC-I. In the analysis we included HeLa cells expressing US6 and US3 to control for strong block of MHC-I peptide loading and retention, respectively. Lysates from these cells were subjected to Western blot analysis. The total level of MHC-I was strongly reduced in both US11 and ΔLCRUS11 expressing cells, suggesting that the US11 LCR is dispensable for degradation of MHC-I (Fig 5B). The expected rescue of MHC-I expression was detected in cells expressing the US11 Q192A mutants. Flow cytometry analysis revealed that surface expression of MHC-I was downregulated both by US11 and ΔLCRUS11 to the same extent as by US6 (Fig 5C). Expression of the US11 Q192A mutants lead to lower level of MHC-I downregulation, also when compared to US3 expressing cells, indicating that the retention is not as strong as for US3. To analyze interactions between US11, MHC-I and the PLC in more detail, we next performed co-immunoprecipitation experiments using metabolically labeled cells. (Fig 5D; immunoprecipitation using an anti-transferrin receptor control antibody is shown in S8 Fig). Whereas the LCR was dispensable for US11-dependent degradation of MHC-I molecules in the context of these stable cell lines (compare MHC-I HC levels in Fig 5D, lanes 11, 12, and 13), MHC-I ER retention caused by the Q192A mutation (compare the level of EndoH sensitive MHC-I HC in lanes 11, 14 and 15) and stabilization of MHC-I in the PLC (compare the level of MHC-I HC in lanes 6, 9 and 10 as well as in lanes 16, 19, and 20) was dependent on the LCR, as the mentioned effects was only observed by the full-lengh US11Q/A mutant, but not by ΔLCRUS11Q/A. This difference in function correlated with co-immunoprecipitation of the US11 mutants: US11Q/A co-immunoprecipitated with W6/32 (lane 14), anti-tapasin (lane 9) and anti-ERp57 antibodies (lane 19). Most convincingly, a weak co-immunoprecipitation of wildtype US11 was observed with anti-tapasin (lane 7) and anti-ERp57 antibodies (lane 17), despite very low MHC-I level in these samples, while no co-immunoprecipitation of ΔLCRUS11Q/A was observed by any of the PLC or MHC-I reactive antibodies.
In contrast, ΔLCRUS11Q/A could be detected in association with unassembled HCs when using the mAb HC10 for immunoprecipitation, which predominantly binds to free HCs (Fig 5E, lane 9; of note, HC10 also immunoprecipitates HLA-A*68:02 HC [36, 37]). This is consistent with the finding that US11 lacking the LCR is still able to mediate degradation of MHC-I HCs. Moreover, we observed that the affinity of US11 for fully assembled MHC-I, which are recognized by the mAb W6/32, was strongly reduced (compare US11 co-immunoprecipitation in Fig 5E, lanes 5 and 8; long exposure in S9 Fig), and this was even more pronounced for the ΔLCRUS11 mutant, demonstrating that US11 interacts with unassembled and assembled MHC-I in different manners. In conclusion, in stably transduced cells US11 interacts with unassembled MHC-I HCs and redirect them for degradation independently of the N-terminal LCR. If, however, the recruitment of ERAD is prohibited by the US11 Q192A mutation, US11 remains in a complex with assembled MHC-I molecules that are bound to the PLC and retained in the ER. This ability of US11 is dependent on the LCR, since in ΔLCRUS11Q/A expressing cells, MHC-I molecules matured as in control cells. The findings are summarized in a schematic Table in the supplementary material (S10 Fig).
An additional observation from this experiment that appeared contradictory, was the lack of resistant MHC-I molecules in cells stably expressing US11. However, in contrast to low MHC-I expression levels in these HeLa cells, MHC-I is massively induced upon HCMV infection [33]. To obtain expression levels comparable to infected cells, we treated the US11-expressing cells with IFNγ. Under these conditions MHC-I was more readily detectable even in the presence of US11 (Fig 6A, lane 4). In control HeLa cells, the lower MHC-I HC band (red asterisk) appeared stronger after IFNγ induction than the upper band (blue asterisk), whereas in US11-expressing cells the lower band was weaker than the upper one. This strongly suggests that the lower MHC-I molecule is more sensitive to degradation. To determine the identity of the MHC-I HC bands, HA-tagged versions of the single HLA-A, -B and -C molecules expressed in HeLa cells (HLA-A*68:02, B*15:03, and C*12:03) [38], were expressed by transient transfection and subsequently their SDS-PAGE separation properties were determined after immunoprecipitation (Fig 6B). The obtained pattern suggests that the lower US11 sensitive MHC-I HC corresponds to HLA-A*68:02. The upper band that was more resistant to US11 appeared as a smear and could comprise both HLA-B*15:03 and C*12:03. In conclusion, under conditions of a high US11/MHC-I ratio, US11 selectivity is less pronounced. Elevated levels of MHC-I leads to a distinct preference of US11 for degradation of HLA-A.
Our data shows that US11 binds to MHC-I HCs and redirects them for proteasomal degradation independently of the LCR. We asked what could be the purpose of the LCR in complex with assembled MHC-I. The PLC not only maintains empty or suboptimally loaded assembled MHC-I molecules in a peptide-receptive state, but also selects peptides with stabilizing properties [39–41]. To clarify the role of US11 in the PLC, we next investigated whether US11 can influence peptide selection. To this end, the MHC-I ligandome of HeLa cells overexpressing US11Q/A, ΔLCRUS11Q/A, or US3 was compared to non-transduced control cells. US3 was included as a control because it retains MHC-I in the ER and was reported to interact with the PLC [10, 13, 42]. The ligandome was determined by LC-MS/MS after isolation of MHC-I by the mAb W6/32 and elution of peptides [43]. The stable cell lines were analyzed in two replicates, the results of which clustered tightly. For better binding prediction, only 9-mer peptides were used for further analysis and assigned as ligands of HLA-A*68:02 or B*15:03 if their affinities were predicted to be <500 nM by NetMHC3.4 [44]. If a ligand was classified as a binder to both MHC-I molecules, it was assigned as a ligand to the one for which a higher affinity was predicted.
Very similar amounts of ligands were found for HLA-A*68:02 and HLA-B*15:03 in control cells (42–45% each of all 9-mers, Table 1 and Fig 7A). In transduced cells (US11Q/A, ΔLCRUS11Q/A, US3), however, the percentage of HLA-B*15:03 derived ligands decreased (28–33% of all 9-mers), pointing to an advantage for HLA-A*68:02 expression or loading in the transduced cells. Changes in MHC-I antigen presentation upon lentiviral transduction has been observed also by others [45]. In US11Q/A expressing cells the overall amount of 9-mer ligands was strongly reduced compared to the other samples (Table 1), suggesting that US11Q/A impaired proper peptide loading. Of note, this was not the case for HeLa cells expressing ~ΔLCRUS11Q/A or US3.
To gain a more detailed view of the effect of ~US11Q/A on MHC-I peptide loading, the usage of ligand anchor residues was compared between HeLa cell lines. HLA-B*15:03 prefers ligands with a lysine or glutamine at position 2 (P2) and a tyrosine or phenylalanine at the C-terminal position (P9). While no change in the usage of the P9 anchor residues was observed between cells (S11 Fig, right panel), the usage frequency of lysine and glutamine at P2 was inversed in US11Q/A expressing HeLa cells (Figs 7B and S11, right panel). In these cells lysine was found at P2 in 15–20% of the HLA-B*15:03 ligands and glutamine in 37–39%, whereas in wild-type HeLa and in the other transduced cell lines, including the ~ΔLCRUS11Q/A expressing cells, lysine was the most common residue and glutamine was less frequently used, 33–41% and 23–28%, respectively. We did not detect changes in the ligandome of HLA-A*68:02 (S11 Fig, left panel). These data suggest that the LCR of US11 interferes with peptide loading of distinct MHC-I molecules.
To analyze whether US11 affects the MHC-I ligandome also in HCMV infected cells, we used the data sets from Fig 1. However, no clear changes in the HLA-B*07:02 and B*44:02 ligandomes were observed when we compared the cells infected with ΔUS2-6 (US11pos) and ΔUS2-6/US11 (US11neg) (S12 Fig). These HLA-B allotypes are very strict in their usage of P2 anchor residues with a high frequency of proline at P2 in B*07:02 peptides and glutamic acid at P2 in B*44:02 peptides and might therefore be more resistant to US11-mediated peptide modification. To better visualize possible changes, the ligandomes were divided into pools of common and unique peptides in the ΔUS2-6 (US11pos) and ΔUS2-6/US11 (US11neg) samples (S13A Fig). Using this setting, we found that unique HLA-B*07:02 peptides in the ΔUS2-6 (US11pos) pool varied at P1, but not much at P3 (Figs 7C and S13B) compared to the common pool. Interestingly, unique HLA-B*07:02 peptides in the ΔUS2-6/US11 (US11neg) pool showed an opposite effect at P1, emphasizing that the effect is US11-specific. Regarding HLA-B*44:02, only three unique peptides could be defined in the ΔUS2-6/US11 (US11neg) pool (S13A Fig) and therefore changes in the ligandome could not be strengthened by this group of peptides. However, we observed that the commonly used proline at P4 [46] was reduced by US11 and instead usage of glutamate was increased. As only a few HLA-A peptides were detected in the ΔUS2-6 (US11pos) unique pool (18 and 15 for A*02:01 and A*29:02, respectively) this data set could not be applied conclusively for analysis of peptide modification. The increased frequency of glutamate at P2 in HLA-A*29:02 ligands of the ΔUS2-6 (US11pos) unique pool is, nonetheless, an interesting observation (S14 Fig, right panel).
The human cytomegalovirus is the largest of all human herpesviruses, both with respect to its genome length and numbers of transcribed ORFs [47]. Whereas all herpesviruses interfere with the host immune defense at some level, the unique infection biology of cytomegaloviruses has come along with the evolution of a large array of highly specialized immune evasive genes and reprogramming of multiple immune functions without compromising the human host [48]. HCMV controls distinct checkpoints of the MHC-I antigen presentation pathway by at least five gene products [6, 49] and most likely even more genes are involved in a less defined manner [33, 50–52]. The human immune system can cover the presentation of a wide breadth of pathogen derived antigens owing to the three extraordinary polymorphic MHC-I genes each individual possesses. This unique and continuing diversification of human HLA may have promoted the emergence of multiple HCMV genes blocking MHC-I antigen presentation. Obviously, it would be less strategic to attack the MHC-I pathway only at a checkpoint shared by all HLA gene products since they display differential immunological impacts. Nevertheless, so far, specificities for representative alloforms of HCMV encoded inhibitors have not been comprehensively studied. Here, we show that a prime target for US11-mediated degradation is HLA-A locus products, whereas HLA-B resists this effect. Instead, US11 has acquired an independent function in its N-terminal LCR to manipulate peptide loading of HLA-B molecules (model in Fig 8).
Advancements in mass spectrometry analysis [53] encouraged us to analyze the MHC-I ligandome of HCMV-infected MRC-5 fibroblasts. The primary aim was to identify novel CD8+ T-cell restricted epitopes (Lübke et al., manuscript submitted). During these studies we came across a remarkable phenomenon regarding US11: whereas the quantity of HLA-A allotype (HLA-A*02:01, A*29:02)-derived peptides was reduced by US11 as expected, this was not the case for the HLA-B (HLA-B*07:02, B*44:02) and -C (HLA-C*05:01, C*07:02) ligandomes. This effect was even more pronounced when only HCMV derived MHC-I ligands were considered. Flow cytometry measurements of infected fibroblast and epithelial cells demonstrated that US11 slightly reduced the level of HLA-B*07:02 surface expression, however, this was much less pronounced as compared with HLA-A*02:01. Altogether, this indicates that in HCMV-infected cells US11 is not able to degrade HLA-B efficiently. The observed higher US11-resistance by HLA-B is in agreement with recent observations of HCMV-infected cells [54, 55] and plasma membrane profiling studies of THP-1 cells expressing US11 [19].
To assess the effect of US11 on a larger panel of various MHC-I molecules, we elaborated a convenient and fast flow cytometry based assay system measuring MHC-I cell surface disposition after transient transfection of HeLa cells (Fig 2A and 2B). MHC-I molecules were designed to express an inert HA-epitope tag at the N-terminus to overcome the need for allotype-specific antibodies. In this way, we measured unrestricted expression of HLA-B allotypes on the cell surface, whereas HLA-A allotypes were strongly reduced.
Unexpectedly, we observed in pulse-chase experiments that US11 substantially affected HLA-B expression in the early secretory pathway (Fig 4C), despite the unchanged density on the cell surface. However, at variance with HLA-A locus products, HLA-B alloforms were able to dimerize with β2m and mature in the presence of US11. HLA-B molecules thus escaped US11 and accumulated on the surface to the same extent as in US11-negative cells. The functional expression of HLA-B*07:02 in the presence of US11 could be further demonstrated by binding to LIR1.
Furthermore, we found that the level of polymorphic MHC-I synthesized in the ER strongly affects the efficacy and selectivity of US11-mediated degradation. In HCMV-infected cells transcription and biosynthesis of MHC-I is highly upregulated [33]. This could explain why US11 does not reach the critical level required to degrade HLA-B, while HLA-A is still efficiently recruited to ERAD, as we observed also in IFNγ-induced HeLa cells stably expressing US11. This underlines the relevance of the strict regulation of US11 expresssion via the HRD1-dependent autoregulatory loop [24, 25]; in the absence of MHC-I substrates US11 itself is targeted to ERAD degradation, controlled by HRD1.
We have begun to address the molecular basis of HLA-B resistance. The shorter cytosolic tail of HLA-B alleles confers some level of resistance, as described previously for US11 [31] and also for HIV-1 encoded Nef [56], but was not a pivotal factor for US11 in our experimental setup. The results obtained with β2m-deficient FO-1 cells showed that β2m is not critically involved in resistance against degradation. Indeed, this demonstrated that this resistance should be an intrinsic property of HLA-B HCs, which is now an object of further investigation.
US11 possesses an LCR sequence between its signal peptide and the Ig-like luminal domain. This region is 15 amino acids long (residues 28–42) and contains seven proline residues. Such structurally undefined regions often function as multiprotein interaction hubs, e.g. found in chaperons [35]. Furthermore, LCRs may have advantages for faster adaptation and evolution [57]. Our analysis revealed that the LCR of US11 is required for several features of US11 that are possibly interconnected. Firstly, the ability of US11 to interact with folded heterodimers of MHC-I HC and β2m, as defined by recognition by the mAb W6/32 [58], is dependent on the LCR. Secondly, we found that US11 interacts with the PLC most likely via binding to MHC-I, since low MHC-I expression levels resulted in low US11 co-immunoprecipitation with the PLC. Again, this interaction was dependent on the US11 LCR, confirming that only folded heterodimeric MHC-I molecules interact with the PLC [5, 59]. US11 with a Q192A mutation is not able to forward MHC-I to the ERAD pathway. As a consequence MHC-I is retained in the ER [15, 22]. This requires a stable interaction between US11 and assembled MHC-I heterodimers that involves the LCR of US11, because deletion of the LCR rescued MHC-I transport through the secretory pathway in the context of the US11Q/A mutant. However, the ability of US11 to forward MHC-I HC for ERAD degradation is not affected by the deletion of the LCR. In cells stably expressing ΔLCRUS11 a strong reduction of MHC-I was observed, in accordance with the finding that ΔLCRUS11Q/A is still able to stably interact with unassembled MHC-I HCs. This indicates that US11 can initiate retrotranslocation and degradation of MHC-I without its LCR at a stage before heterodimeric MHC-I molecules assemble. The targeting of unassembled MHC-I in β2m deficient cells has been observed previously [60].
The most remarkable feature of the US11 LCR, however, was its ability to manipulate HLA-B*15:03 peptide ligands. The usage of the N-terminal P2 position of the ligand anchor residue was changed in a way that the frequently appearing lysine was strongly reduced and the generally less used glutamine emerged most frequently in the presence of US11. Control cells, ΔLCRUS11Q/A cells or cells expressing US3, which also binds to and retains MHC-I heterodimers, did not exhibit this effect on HLA-B*15:03, indicating that it is a specific feature of the US11 LCR that interferes with peptide loading. We observed this change only for the N-terminal anchor residues and not for the C-terminal. We were not able to define any similar changes in the ligandome of HLA-A*68:02.
The HLA-B molecules HLA-B*07:02 and B*44:02 present in MRC-5 fibroblast do not allow for measurable modifications at P2, since the P2 residue is strongly fixed for these molecules (by proline and glutamate, respectively). However, unique HLA-B*07:02 and HLA-B*44:02 ligands in MRC-5 cells infected with the ΔUS2-6 HCMV mutant virus expressing US11, displayed changes in the neighboring P1 and P4, respectively, suggesting that US11 influences the peptide selection for a broad range of HLA-B allotypes.
The recently resolved structure of the PLC [5] provides a molecular basis to model manipulation of HLA-B by US11. The MHC-I peptide binding groove is deeply buried in the PLC with the F-pocked that binds the C-terminal peptide anchor residue pointing inwards into the center of the PLC. The opposite side of the MHC-I HC is the only region of MHC-I still accessible for further protein interactions. This interface is also used by HCMV encoded US2 and adenovirus encoded E3-19K as demonstrated in resolved crystal structures [14, 61]. If US11 also binds to this particular surface, which is likely, since US11 interacts with MHC-I during its processing through the PLC, the LCR could be in the vicinity of MHC-I residues contributing to the formation of pockets that fix the N-terminal part of the peptide.
Whereas HLA-B allotypes are at the forefront in studies determining protective and sensitizing MHC-I in HIV and HCV infections [62, 63], such observations have not been made for HCMV or other herpes viruses. This goes well together with the suggestion that HLA-A is more important to control co-evolving DNA viruses [64]. The differential targeting of HLA-A and -B by US11 underlines this view and implies that complete block of antigen presentation by HLA-A is crucial for the virus to cope with highly specific CD8+ T-cells. Unlike HLA-A, a large fraction of HLA-B allotypes contains the Bw4 motif recognized by inhibitory KIRs (Killer cell immunoglobulin-like receptors) on NK cells [65]. Therefore, the costs for allowing a reduced level of HLA-B surface expression, yet, with a modified peptide repertoire, might be tolerated by HCMV, in order to dampen NK cell activation.
Future studies will provide more insight into the mechanism how the US11 LCR alters the quality of MHC-I peptide ligands and the functional ramifications of this alteration. It is conceivable that this feature of US11 could confound CD8+ T-cell recognition of HCMV infected target cells. The initial priming of CD8+ T precursors is believed to occur via cross-presentation [66, 67], i.e. by non-infected dendritic cells in the absence of US11. Thus, the quality of the MHC-I presented peptides might differ significantly between productively HCMV-infected cells, in which US11 is actively expressed and professional APC priming the CD8+ T-cells. Whether US11 will impact the formation of memory cells and memory inflation is not predictable. However, it will be of great interest to learn whether Rh189 (RhCMV US11 homolog)-induced non-canonical CD8+ T-cell restricted epitopes [29] are also dependent on the N-terminal part of the protein and could be a result of manipulation of peptide loading. Although an LCR is not predicted in the N-terminus of Rh189, it still contains some conserved residues (S15 Fig), possibly important for interaction with assembled MHC-I.
MRC-5 fibroblasts (ECACC 05090501; HLA-A*02:01, A*29:02, B*07:02, B*44:02, C*05:01, C*07:02), HeLa (ATCC CCL-2; HLA-A*68:02, B*15:03, C*12:03; ATCC CCL-2), and US6-HA-HeLa cells [68], ARPE-19 (ATCC CRL-2302), the melanoma cell line FO-1 [69] and HEK293T (ATCC CRL‐11268) cells were grown in DMEM supplemented with 10% FCS, penicillin and streptomycin. HeLa cells were tranfected with Superfect (Qiagen) and FO-1 cells with Jetprime (Polyplus Transfection). Small interfering RNA (siRNA) targeting US11 (ACACUUGAAUCACUGCCACCCCC) was purchased from Riboxx. Knock-down experiments were performed using Lipofectamin RNAiMax Reagent (Invitrogen).
The recombinant HCMV mutants ΔUS2-6/US11 and ΔUS2-US11 was generated according to a previously published procedure [70] using the BAC-cloned AD169varL genome pAD169 [71] as parental BAC. Briefly, PCR fragments was generated using the primer pair KL-DeltaUS11-Kana1 CAAAAAGTCTGGTGAGTCGTTTCCGAGCGACTCGAGATGCACTCCGCTTCAGTCTATATACCAGTGAATTCGAGCTCGGTAC and KL-DeltaUS11-Kana2 TAAGACAGCCTTACAGCTTTTGAGTCTAGACAGGGTAACAGCCTTCCCTTGTAAGACAGAGACCATGATTACGCCAAGCTCC for the ΔUS2-6/US11 mutant and the primer pair KL-DeltaUS7-Kana1 ACCTTTTGTGCATACGGTTTATATATGACCATCCACGCTTATAACGAACCTAACAGTTTACCAGTGAATTCGAGCTCGGTAC and KL-DeltaUS11-Kana2 TAAGACAGCCTTACAGCTTTTGAGTCTAGACAGGGTAACAGCCTTCCCTTGTAAGACAGAGACCATGATTACGCCAAGCTCC for the ΔUS2-US11 mutant and the plasmid pSLFRTKn [72] as template DNA. The PCR fragment containing a kanamycin resistance gene was inserted into the parental BAC by homologous recombination in E. coli. Correct mutagenesis was confirmed by Southern blot and PCR analysis. Recombinant HCMVs including TB40/E ΔUS2-6 [73] were reconstituted from HCMV BAC DNA by Superfect (Qiagen) transfection into permissive MRC-5 fibroblasts. Virus titers were determined by standard plaque assay.
Production of lentiviruses was performed as described previously [38]. At 48 h post transfection the supernatant was collected and filtered through a 45 μm filter prior to transduction of HeLa cells by centrifugal enhancement. When selected, the cells were cultivated in normal medium for 3–4 days before treatment with 5 μg/ml puromycin (Sigma).
The following antibodies were applied: W6/32 (anti-pan-HLA-A,B,C assembled with β2m and peptide, [58]), BB7.2 (anti-HLA-A2 [74]), BB7.1 (anti-HLA-B7 [74]), TT4-A20 (anti-HLA-B44 [75]), HC10 and HCA2 recognizing free HLA-B/C and HLA-A heavy chains, respectively [76], anti-CD71 (immunotech), anti-CD85j (LIR1; Miltenyi) mouse and rabbit anti-HA antibodies (Sigma), anti-ERp57 (Millipore), APC-coupled anti-mouse antibodies (BD Pharmingen). Polyclonal anti-tapasin and anti-US11 anti-sera were raised by immunization of rabbits (Genscript) with synthetic peptides (aa 418–428 and 90–103, respectively).
The tapasin signal peptide sequence was amplified in front of a human influenza hemagglutinin (HA)–tag and cloned into XhoI and PstI of pIRES-EGFP (Tpn-SP-pIRES-EGFP; CMV IE promoter). HLA-A*02:01, HLA-B*07:02, HLA-C*07:02, CD99 were amplified from cDNA prepared from MRC-5 cells and HLA-A*68:02 and HLA-B*15:03 from cDNA from HeLa cells. HLA-B*44:02 and HLA-B*44:05 were described previously [38]. The cDNA clones for HLA-A*01:01 (NM_001242758, BC003069), HLA-A*03:01 (NM_002116) and HLA-B*08:01 (AK292226, BC091497) have been purchased from Source Bioscience, Nottingham, UK. HLA-A*29:02 (IMGT/HLA database) was synthesized as a gBlock (Integrated DNA Technologies, Inc.) gene fragment. Irrespective of their source, all MHC-I sequences were used as a template for further amplification using specific primers (Table 2). PCR products were digested with PstI or NsiI and BamHI restriction enzymes and cloned into Tpn-SP-pIRES-EGFP. Sequenced inserts were subsequently subcloned into the puc2CL6IP (pUC-IP, with spleen focus-forming virus U3 promoter) lentiviral vector [38] using the restriction sites NheI and BamHI. HA-HLA-C*12:03 was purchased in pcDNA3.1 from Biocat and subcloned into puc2CL6IP. US11 and US3 cDNA was amplified from AD169 HCMV DNA and cloned into pIRES-EGFP via NheI and BamH or into puc2CL6IP and puc2CL6-IRES-EGFP (pUC-EGFP) lentiviral vector using the same enzymes. Point mutation in US11 was inserted using the QuickChange II XL Site-Directed Mutagenesis Kit (Agilent) following the protocol described by the manufacturer.
MHC-I ligands were isolated by standard immunoaffinity purification using the mAb W6/32 as described previously [43]. LC-MS/MS analysis of MHC-I ligand extracts using nanoflow HPLC (RSLCnano, Thermo Fisher) on a 50 μm × 25 cm PepMap RSLC column (Thermo Fisher) with a gradient ranging from 2.4 to 32.0% acetonitrile over the course of 90 min. Eluted peptides were analyzed in an online-coupled LTQ Orbitrap XL mass spectrometer (Thermo Fisher) using a top 5 CID (collision-induced dissociation) method. The procedure for label-free quantification (LFQ) of relative HLA ligand abundances was performed as follows: total injected peptide amounts of paired samples were normalized and LC-MS/MS analysis was performed in five technical replicates for each sample. For normalization, the relative amounts of substance in paired samples were determined by calculating the summed area of peptide identifications in dose-finding LC-MS/MS runs and the samples were adjusted accordingly by dilution. Relative quantification of HLA ligands was performed by calculating the area under the curve of the corresponding precursor extracted ion chromatograms (XIC) using ProteomeDiscoverer 1.4 (Thermo Fisher). For Volcano plots, the ratios of the mean areas of the individual peptides in the five LFQ-MS runs of each sample were calculated and two-tailed t-tests implementing Benjamini-Hochberg correction were performed using an in-house R script (v3.2). Data processing and spectral annotation was performed as described previously [77]. In brief, the Mascot search engine (Mascot 2.2.04; Matrix Science, London, UK) was used to search the human proteome as comprised in the Swiss-Prot database (20,279 reviewed protein sequences, September 27th 2013) without enzymatic restriction. Oxidized methionine was allowed as a dynamic modification. The false discovery rate was estimated using the Percolator algorithm [78] and set to 5%. Peptide lengths were limited to 8–12 amino acids for HLA class I. Protein inference was disabled, allowing for multiple protein annotations of peptides. HLA annotation was performed using NetMHC [44] (v3.4), annotating peptides with IC50 scores below 500 nM as ligands of the corresponding HLA allotype. In cases of multiple possible annotations, the HLA allotype yielding the lowest IC50 score was selected.
MRC-5 and ARPE-19 cells were detached with Accutase (Sigma), treated with FcR blocking reagent as recommended by manufacturer (Miltenyi) and stained with antibodies diluted in 3% FCS/PBS. Cells were washed in 3% FCS/PBS supplemented with DAPI and fixed in 4% paraformaldehyde. Cells were analyzed by FACS Canto II (Becton Dickinson).
For analysis of MHC-I expression after transient transfection of HeLa cells US11 variants or a control pIRES-EGFP plasmid together with HA-tagged HLA-alleles or control molecules in puc2CL6IP were co-transfected using SuperFect (Qiagen). At 20 h post-transfection, cells were detached with trypsin and measured as described above. LIR-1 binding was analyzed by incubating the cells with recombinant protein [79] and subsequently incubating the cells with an APC-coupled anti-CD85j (LIR-1) antibody.
Acquired data was analyzed by FlowJo (v10.1, Tree Star Inc.). For statistical analyses Mann-Whitney U-test or one-way ANOVA followed by Tukey’s or Dunnett´s multiple comparison test were performed using the GraphPad Prism 6 Software. A p-value <0.05 was considered significant (*, p<0,05; **, p<0,005; ***, p<0,0005).
Immunoprecipitation was performed as described previously [38]. Briefly, cells grown in 6-well plates were washed with PBS and metabolically labeled (Easytag Express [35S]-Met/Cys protein labeling, Perkin Elmer) with 100 Ci/ml for various times. Cells were lysed in digitonin lysis buffer (140 mM NaCl, 20 mM Tris [pH 7.6], 5 mM MgCl2, and 1% digitonin (Calbiochem)) and cleared from membrane debris at 13,000 rpm for 30 min at 4°C. For analysis of lysates with several antibodies, identical lysates were pooled then split up into equal aliquots. Lysates were incubated with antibodies for 2 h at 4°C in an overhead tumbler before immune complexes were retrieved by protein A- or G-sepharose (GE Healthcare). Sepharose pellets were washed four times with increasing NaCl concentrations (0.15 to 0.5 M in lysis buffer containing 0.2% detergent). For a re-immunoprecipitation the washed beads were subsequently incubated with a lysisbuffer supplemented with 1% Igepal (Sigma) and 1% SDS at 95° C for five minutes. The lysisbuffer was diluted to reach a final concentration of 0.1% SDS and a subsequent immunoprecipitation was performed. Endoglycosidase H (New England Biolabs) treatment was performed as recommended by the manufacturer. Prior to loading onto a SDS-PAGE iImmune complexes were dissociated at 95°C for 5 min in a DTT (40 mM) containing sample buffer. Fixed and dried gels were exposed overnight to a phosphor screen, scanned by Typhoon FLA 7000 (GE Healthcare). For better visualization of the results contrast and light were adjusted. Where mentioned in the figure legend a short or long exposure to x-ray film was used for autoradiography.
For Western blot analysis, equal amount of cells were washed in PBS and lysed in lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Igepal, and Complete protease inhibitor (Roche)). The proteins were separated by SDS-PAGE and transferred to nitrocellulose filter. After incubation with primary antibody a peroxidase coupled secondary antibody was used and chemiluminescence was detected using a LI-COR Blot Scanner.
Mock treated or infected MRC5 cells were grown in technical replicates in a 6-well format. Cells were lysed at 24 h post-infection and total RNA was extracted using NucleoSpin kit RNA II (Macherey-Nagel) and 600ng was reverse transcribed using a QuantiTect reverse transcription kit (Qiagen) and further used for both semi-quantitative and quantitative RT-PCRs. Quantitative RT-PCRs were performed with a QuantiTect SYBR green PCR kit (Qiagen). CT values were normalized to actin (ΔCT) and plotted relative to the ΔCT values of the mock treated control cells. For the semi-quantitave analysis the following primer pairs were used: US11-ctrl3’ tggtccgaaaacatccaggg and US11-ctrl5’ ttcgatgaacctccgccctt; US10-ctrl’3 aaccgcatatcaggaggaggga and US10-ctrl’5 tcacgtgcggctgtgttattca, UL40-1 gcagctagcgccgccaccatgaacaaat and UL40-2 cgaggatcctcaagcctttttcaaggcg. For the qRT-PCR we used the primers: qUS10-1 acgacggggaaaatcacgaa and qUS10-2 cagagtagtttcggggtcgg; actin beta primers (Qiagen, Hs_ACTB_1_SG QuantiTect Primer).
|
10.1371/journal.pgen.1002540 | A Downstream CpG Island Controls Transcript Initiation and Elongation and the Methylation State of the Imprinted Airn Macro ncRNA Promoter | A CpG island (CGI) lies at the 5′ end of the Airn macro non-protein-coding (nc) RNA that represses the flanking Igf2r promoter in cis on paternally inherited chromosomes. In addition to being modified on maternally inherited chromosomes by a DNA methylation imprint, the Airn CGI shows two unusual organization features: its position immediately downstream of the Airn promoter and transcription start site and a series of tandem direct repeats (TDRs) occupying its second half. The physical separation of the Airn promoter from the CGI provides a model to investigate if the CGI plays distinct transcriptional and epigenetic roles. We used homologous recombination to generate embryonic stem cells carrying deletions at the endogenous locus of the entire CGI or just the TDRs. The deleted Airn alleles were analyzed by using an ES cell imprinting model that recapitulates the onset of Igf2r imprinted expression in embryonic development or by using knock-out mice. The results show that the CGI is required for efficient Airn initiation and to maintain the unmethylated state of the Airn promoter, which are both necessary for Igf2r repression on the paternal chromosome. The TDRs occupying the second half of the CGI play a minor role in Airn transcriptional elongation or processivity, but are essential for methylation on the maternal Airn promoter that is necessary for Igf2r to be expressed from this chromosome. Together the data indicate the existence of a class of regulatory CGIs in the mammalian genome that act downstream of the promoter and transcription start.
| CpG islands are CG-rich regions associated with the majority of mammalian promoters. Although widely considered to be necessary for promoter activity, their exact function is unknown. CpG islands are mostly unmethylated during development and differentiate with notable exceptions including imprinted genes and genes on the inactive X chromosome. Here we analysed how the imprinted Airn ncRNA CpG island, that is normally methylated on the maternal and unmethylated on the paternal chromosome, regulates its associated upstream promoter. We used embryonic stem cells or mice carrying a deletion of either the whole CpG island or of the second half that contains an unusual series of tandem direct repeats. Our results show that the CpG island is needed for efficient transcription of the Airn promoter on the paternal chromosome and to keep it free from DNA methylation. The series of tandem direct repeats plays a minor role in regulating the length of the Airn transcript on the paternal chromosome but is essential for DNA methylation of the Airn promoter on the maternal chromosome. These results show that CpG islands do not only function as classical promoters to bind RNA polymerase II and initiate transcription, but can also play other roles in regulating transcription processivity as well as the epigenetic state of their associated promoter.
| Atypical CpG-rich regions known as CpG islands (CGIs) overlap 60–70% of mammalian transcription start sites [1]. Although most CGIs extend downstream of the transcription start and are therefore partly transcribed, they are considered to have promoter regulatory functions and are often described as ‘CGI promoters’. A recent study used a biochemical purification strategy to identify a large number of novel CGIs not associated with annotated promoters, in the body of coding genes or in intergenic regions [2]. While this could indicate the mammalian genome has many transcripts still to be identified, it is also possible that CGIs have additional functions in addition to promoter regulation.
The best examples of CGIs with additional regulatory functions are those that lie inside imprint control elements (ICE, also known as an imprint control region) [3]. An ICE is a genetically defined region whose epigenetic state controls parental-specific expression of small clusters of genes [4]–[6]. CGIs within an ICE are similar to classic promoter-associated CGIs as they show a CpG density higher than the genome average and lack sequence conservation, even between homologous mouse and human elements [7]. However they differ in several ways [8], [9]. First, their CpG density is less than that of classic promoter-associated CGIs. Second, whereas most promoter-associated CGIs are free of DNA methylation [1], CGIs within ICEs gain DNA methylation during gametogenesis but only in one of the two parental gametes. These modified regions are also known as gametic ‘differentially-methylated-regions’ (gDMRs), since once established in a gamete they are maintained in all somatic cells on the same parental chromosome, while the other parental allele remains methylation-free. In six imprinted clusters the ICE has been shown by deletion experiments that include the CGI, to control repression of all imprinted genes [10]–[15]. Thus, the third distinguishing feature is that an unmethylated ICE can act as a cis-acting long-range repressor of multiple flanking genes. This indicates that CGIs residing in an ICE may be a prototype for a class of cis-regulatory CGIs that may differ from classic promoter-associated CGIs.
Remarkably, the silencing ability of the unmethylated ICE correlates with its action as a promoter or cis-activator of a macro non-protein-coding (nc) RNA (provisionally defined as a ncRNA >200 bp whose function does not depend on processing to smaller RNAs) [16], [17]. Three imprinted macro ncRNAs that play a direct role in imprinted gene silencing i.e., Airn, Kcnq1ot1, and Nespas, have their promoter in the ICE [18]–[20]. These cis-repressor macro ncRNAs therefore contain CGI sequences at their 5′ end that could contribute to their repressor function. The Airn, Kcnq1ot1 and Nespas macro ncRNAs are expressed only from the paternal chromosome and induce paternal-specific silencing of flanking protein-coding genes. Imprinted expression of the flanking protein-coding genes arises because these repressor macro ncRNAs are repressed on the maternal chromosome by an ICE gametic methylation imprint [9], [21], [22]. Maternal gametic methylation imprints depend on expression in growing oocytes of the DNMT3A/B de novo methyltransferases and the DNMT3L cofactor [23], [24]. It has also been shown that transcription across the ICE controlling Nespas ncRNA expression is required for methylation in oocytes [25]. In addition, recent high-throughput analyses show a general link between overlapping transcription and CGI methylation in oocytes [26]. However, there is little information on the relative contribution of DNA elements within the ICE for the methylation state. Tandem direct repeats (TDRs) that show organizational but not sequence conservation, are frequently found in or adjacent to the ICE and have been suggested to guide epigenetic modifications [9], [27], [28]. The TDRs are present on both parental chromosomes but methylation of the ICE restricts expression of the macro ncRNA to one parental chromosome. Thus, it is possible that the TDRs play a role in ICE methylation on one parental chromosome and in the repressor function of the macro ncRNA expressed from the other parental chromosome. However, to date various experiments analysing their function either at the endogenous locus or in a transgene context, have not yet identified a general function for TDRs in imprinted clusters [28].
The Airn macro ncRNA promoter that is embedded in the ICE, lies in intron 2 of the Igf2r gene. Airn overlaps and represses the Igf2r promoter (Figure 1A); in extra-embryonic lineages Airn also represses the non-overlapped Slc22a2 and Slc22a3 genes that lie more than 100 kb upstream of the Airn promoter [19], [29]. Airn is an unusually long 118 kb ncRNA that is transcribed by RNA polymerase II (RNAPII). The majority of nascent Airn transcripts are unspliced and nuclear-localized while the minority that are spliced are exported to the cytoplasm [21]. Splicing suppression, unusual length and gene silencing ability are also features shared with the Kcnq1ot1 macro ncRNA [30]. Previously we have established an ES cell imprinting model that recapitulates the onset of imprinted Igf2r expression in early mouse embryonic development [31] (Figure 1A). Undifferentiated ES cells show bi-allelic but low-level Igf2r expression and Airn is not expressed [31]. Airn expression is initiated during ES cell differentiation and induces imprinted Igf2r expression by blocking up-regulation of the overlapped paternal promoter between days 3–5. The Igf2r promoter, which is also associated with a CGI, gains DNA methylation on the paternal allele after the onset of imprinted expression between days 5–14. However, this somatic methylation mark is not required for repression, as Airn silences Igf2r in mouse embryos lacking DNA methylation [21], [22]. We previously used a deletion/replacement approach in this ES cell imprinting model to identify a 959 bp promoter region immediately upstream of the Airn main transcription start [32]. These experiments demonstrated not only that the promoter lies upstream of the annotated CGI (Figure 1B), but also, that the endogenous Airn promoter does not control the unusual features of the macro ncRNA, as Airn driven by the mouse Pgk1 promoter is indistinguishable from Airn driven by its endogenous promoter [32]. Thus control of the unusual biology of the Airn macro ncRNA lies outside its promoter.
Here we use the ES cell imprinting model and also mouse models, to test if the Airn downstream CGI plays a role either on the paternal allele in regulating Airn expression and function and the unmethylated state of the ICE or, on the maternal allele in regulating ICE methylation (Figure 1A, 1B). The Airn downstream CGI contains in its distal half two classes of imperfect TDRs that are each repeated three times, one with a 172–180 bp monomer length and one with a 30–32 bp monomer length (Figure 1B). We used homologous recombination in ES cells to delete a 1129 bp fragment containing the entire CGI and also to delete a 692 bp fragment containing just the TDRs. Both deletions left the Airn promoter and transcription start site (TSS) intact. Analysis of the effects of the deletion on the paternal chromosome that expresses Airn shows that the CGI deletion decreased Airn transcription initiation and strongly reduced transcript elongation, which as predicted from previous analyses [19], led to a loss of its ability to repress Igf2r in cis. The TDR deletion on the paternal chromosome led to a minor defect in transcript processivity that progressively affected the 3′ end of Airn, combined with a minor effect on its repressor function in differentiated ES cells and in mouse tissues. In contrast to the minor role on the paternal chromosome, analysis in mouse embryos of maternal chromosomes carrying the TDR deleted allele shows this element is essential for the DNA methylation imprint. Together these data show the Airn CGI has a dual parental-specific function and is necessary both for Airn biology and function as well as the critical epigenetic modifications that control its imprinted expression.
We first examined if the position of the Airn CGI that lies downstream of the TSS, represents a rare exception or a common occurrence in the mouse genome (Figure 1C). We asked, for each CGI annotated by the UCSC genome browser (http://genome.ucsc.edu/), if a known protein-coding or non-protein-coding gene taken from the NCBI RNA reference sequences collection (RefSeq), has its TSS within the CGI or in the DNA region representing half the length of the CGI up- or downstream. 57% of all RefSeq genes were associated with a CGI, of which 88.5% including Igf2r, have their TSSs within the body of the CGI (grey shaded area Figure 1C), with the majority lying in the first half. 11.5% of CGI-associated RefSeq genes have their TSS located outside of the annotated CGI, 8.7% of these have their TSS upstream and 2.8% have a TSS downstream of the CGI. Airn represents one of those with their TSS located upstream of the CGI while the Kcnq1ot1 macro ncRNA has its TSS on the 5′ border of the CGI. As the Airn macro ncRNA has its TSS located upstream of the CGI it is possible to distinguish separate functions for the promoter (previously mapped to a 959 bp fragment lying upstream of the Airn-TSS [32]) and the CGI.
We first generated a deletion of the TDRs that occupy the second half of the Airn downstream CGI (Figure 1B). We used feeder-dependent D3 ES cells previously modified to contain a single nucleotide polymorphism (SNP) in Igf2r exon 12 that is used to distinguish maternal and paternal Igf2r expression [31]. The SNP modifies the maternal allele, thus this cell line is called S12/+ (note the maternal allele is always written on the left i.e., Mat/Pat). S12/+ cells were used as the wildtype control for all differentiation experiments. We targeted S12/+ cells by homologous recombination to delete a 692 bp region containing all TDRs starting 614 bp downstream of the major Airn-TSS (T1). The selection cassette was inserted 2 kb upstream of the deletion to avoid leaving a loxP site at the deletion, which has been reported to attract DNA methylation [33] and also to minimise potential effects on the Airn promoter region from the transient presence of a selection cassette (Figure S1A). Two independent homologously-targeted clones (named S12/TDRΔ+cas-1 and -2) were verified by Southern blot. The selection cassette was removed by transient transfection with a CRE-recombinase expressing plasmid (Figure S1A, S1B) and cells were subcloned to obtain four cell lines (named S12/TDRΔ-1A/-1B/-2A/-2B). Southern blot analysis showed that all cells were targeted on the paternal allele that carries the unmethylated active ICE and expresses the Airn ncRNA (Figure S1C, S1D). Preferential paternal targeting of the region between the Airn and Igf2r promoters is a feature of this cluster (data not shown). Initial analysis of the deletion was performed in the ES cell imprinting model that we have shown recapitulates the onset of imprinted Igf2r expression in the early embryo [31], [32]. To further validate the ES cell imprinting model and to observe a possible role of the TDRs in ICE methylation on the maternal allele, we deleted the same region in intraspecies 129/B6 A9-ES cells to generate knock-out mice (Figure S2A). The homologously-targeted A9 clone (+/TDRΔ+cas) was injected into blastocysts and the selection cassette removed by mating to a MORE CRE-deleter strain (Figure S2B, S2C) [34].
It is technically challenging to determine the exact length of macro ncRNAs as they are too long to be resolved on RNA blots. Therefore to test if the TDR deletion changed the length of the Airn transcript, we performed an RNA hybridisation to a genome tiling array. Genome tiling arrays allow the detection of changes in transcript length using approximately 50 bp long probes spaced every 100 bp of single copy genomic DNA. As Airn is only expressed upon differentiation, we differentiated wildtype (S12/+) and two independently derived TDRΔ ES cells (S12/TDRΔ-2A and +/TDRΔ). cDNA prepared from total RNA labelled with one fluorochrome and sonicated genomic DNA labelled with a second fluorochrome, were cohybridized to the tiling array and the relative signal intensity plotted (Figure 2). The displayed 180 kb region contains the overlapping Igf2r and Airn transcripts and can be divided into three parts: the first part is specific to the 3′ end of the Igf2r transcript, the second part is the region of Igf2r/Airn sense/antisense transcriptional overlap and the third part is specific to the 3′ end of the Airn transcript from 28–118 kb. In the ‘Igf2r-specific’ region, comparison of signal intensities does not indicate a difference in Igf2r levels between the wildtype and two TDRΔ ES cell lines, as the overlapping error bars show the technical variation is larger than the biological difference. All three relative signal intensities then show an abrupt increase in the ‘overlap’ region due to the combined signals of Igf2r and Airn. In the ‘Airn-specific’ region, all three signal intensities decline from left to right as previously shown for wildtype Airn [32]. However, in both TDRΔ cell lines compared to wildtype cells, although Airn relative intensity was unchanged from 28–68 kb downstream of the Airn-TSS, it was reduced after 68 kb and absent from 90 kb onwards while wildtype Airn extends 20 kb further (Figure 2 single and double dashed arrows). Thus the TDR deletion on the paternal chromosome has no effect on the first half of Airn but reduces its overall length progressively towards the 3′ end.
While 95% of Airn transcripts are unspliced, spliced transcripts comprise 23–44% of the steady-state population due to their increased stability [21]. To confirm the shortening of Airn on unspliced and spliced transcripts, we quantified steady-state levels of spliced and unspliced Airn in undifferentiated and differentiated TDRΔ and control ES cells using qPCR assays spread throughout the Airn transcript (Figure 2 map). These qPCR assays allowed us to specifically test if splicing suppression of Airn is affected by the TDR deletion. As previously reported, neither spliced nor unspliced Airn is expressed in undifferentiated (d0) wildtype (S12/+) ES cells but Airn is strongly upregulated during differentiation (d5–d14) [31] (Figure 3A, 3B). Similar kinetic behaviour with some biological variation was found for all four S12/TDRΔ ES cells using four qPCR assays spaced over the first 73 kb of Airn. These were an assay within the first exon of Airn that detects both unspliced and spliced Airn (START, Figure 3A black bars), two assays that detect only unspliced Airn (RP11 at 154 bp and Airn-middle at 53 kb, Figure 3A light and dark grey bars) and, two assays that only detect spliced Airn (RP6 detecting the SV1a splice variant at 38 kb and RP21 detecting the SV1 variant at 73 kb that are the most abundant spliced products, Figure 3B black and light grey bars). At d14, RP5 detecting the SV2 splice variant at 88 kb showed a significant reduction in 2 of 4 TDRΔ clones compared to wildtype (Figure 3B dark grey bars). Two qPCR assays in the 3′ part of Airn showed a significant reduction in all four S12/TDRΔ cells compared to wildtype cells. These were one assay that detects unspliced Airn (Airn-end at 99 kb, Figure 3A white bars) and one assay that detects spliced Airn (RP4 detecting the SV3 splice variant at 118 kb; Figure 3B white bars). Although splice variants that used an exon 2 located after 73 kb were reduced, the TDR deletion did not induce a major shift in spliced versus unspliced Airn transcripts for SV1a and SV1 that end before 73 kb. Thus, the inefficient splicing of Airn is not dependent on sequences in the TDRs. In addition, neither the absence of Airn transcription in undifferentiated ES cells, nor its ability to be upregulated during differentiation depends on TDR sequences.
We also analysed steady-state levels of Airn with or without the TDRs in 12.5–13.5 dpc mouse embryos. We used +/+ and +/TDRΔ embryos and as additional controls, Thp/+ and Thp/TDRΔ embryos. Thp is a hemizygous deletion of the Igf2r cluster allowing the specific analysis of one parental allele [35]. We analysed unspliced Airn with three different qPCR assays at the beginning, middle and end of Airn and found that the mouse data largely recapitulate those from the ES cell imprinting model by showing a progressive length reduction towards the 3′ end of Airn (Figure 3C). The 5′ assay (RP11 at 154 bp, grey triangles) detected similar amounts of Airn from the wildtype and the TDRΔ allele. In contrast to the results obtained from the ES cell imprinting model, one assay in the middle of Airn (Airn-middle at 53 kb, black rectangles) showed a significant reduction to approximately 55% of wildtype levels of Airn from the TDRΔ allele. The end assay, in agreement with the ES cell imprinting model (Airn-end at 99 kb, black circles), detected significant reduction to approximately 5% from the TDRΔ allele compared to the wildtype allele. A similar progressive loss of Airn towards the 3′ end was detected in the extra-embryonic visceral yolk sac (VYS) (Figure S3A). Together, the quantitative analysis of Airn expression supports the conclusion drawn from the genome tiling array, that sequences in the TDR deletion are necessary for full-length Airn.
In addition to the unusual biology that results in a very long unspliced RNA, the other key property of the Airn ncRNA is its cis-silencing ability. We have previously shown that shortening Airn from 118 kb to 3 kb by targeted insertion of a polyA-signal, leads to a loss of repression of the overlapped Igf2r gene and the upstream flanking Slc22a2 and Slc22a3 genes [19]. As TDR deletion led to a 3′ shortening of Airn we first asked if this affected its ability to repress Igf2r. This can be monitored indirectly by the gain of DNA methylation on the paternal Igf2r promoter-associated CGI or, directly by assaying allelic Igf2r expression. We first analysed gain of DNA methylation by methyl-sensitive restriction digestion of genomic DNA from undifferentiated and differentiated wildtype (S12/+) and TDRΔ (S12/TDRΔ) ES cells (Figure 4A). In undifferentiated wildtype ES cells both Igf2r parental alleles are methylation-free as demonstrated by the single 4 kb band. Upon differentiation a 5 kb band indicating a gain of DNA methylation on the paternal allele appears, which increases in strength during differentiation. At d14, quantification from three independent differentiation sets (Figure 4A, Figure S4A) revealed methylated∶unmethylated ratios of 0.75∶1 in wildtype ES cells and a similar ratio in four TDRΔ ES cells. This indicates the gain of DNA methylation on the repressed paternal Igf2r promoter is not dependent on the TDRs.
We next directly analysed Igf2r imprinted expression using the SNP that lies in Igf2r exon 12, 20 kb upstream to the ICE, which can be distinguished by PstI digestion (Figure 4B, Figure S4B). In undifferentiated (d0) wildtype ES cells, PstI digested cDNA results in an undigested maternal (Mat) fragment and two restriction fragments representing the paternal (Pat) allele, indicating biallelic Igf2r expression. As previously described, the paternal-specific fragments are gradually lost during differentiation, which indicates maternally-biased Igf2r imprinted expression [31]. Undifferentiated ES cells with a paternal TDRΔ allele (S12/TDRΔ) also express Igf2r biallelically. However, they differ from wildtype cells by showing reduced paternal Igf2r repression during differentiation, as the two paternal-specific restriction fragments are more visible at d5 and in some cases, remain visible at d14. We quantified this effect on paternal Igf2r repression using a qPCR assay that uses forward primers specific for the two SNP alleles in combination with a common reverse primer. The ratio of the maternal to the paternal allele in undifferentiated wildtype ES cells was set to 1 (Figure 4C), as they were shown previously to express Igf2r biallelically [31]. Wildtype ES cells show a consistent increase in the maternal to paternal Igf2r ratio during differentiation, representing specific upregulation of the maternal allele, with constant low-level paternal expression. Undifferentiated ES cells with a paternal TDRΔ allele (S12/TDRΔ) also show ratios close to 1, indicating biallelic Igf2r expression. S12/TDRΔ cells show an increased maternal∶paternal Igf2r ratio during differentiation, however, this only reached 44–55% at d5 and 49–65% at d14 of the ratio seen in wildtype cells, representing an approximate 2-fold upregulation of the paternal Igf2r allele in TDRΔ cells compared to wildtype (Figure 4C). However, neither the maternal∶paternal Igf2r ratio, nor total Igf2r levels (data not shown) were statistically different in TDRΔ cells compared to wildtype cells. Total Igf2r levels were then analyzed in 12.5–13.5 dpc mouse embryos carrying a paternal TDRΔ (+/TDRΔ, Thp/TDRΔ) or paternal wildtype (+/+, Thp/+) chromosome (Figure 4D). Mean total Igf2r levels in +/+ embryos were set to 100% and +/TDRΔ embryos showed an average of 106%. As the majority of Igf2r transcripts are produced from the maternal wildtype allele that could mask changes on the paternal allele, we analysed embryos carrying a maternal Thp deletion allele that only have the paternal Igf2r allele. The wildtype chromosome in Thp/+ embryos showed 4.5% of levels in +/+ embryos while the TDRΔ chromosome in Thp/TDRΔ embryos showed 7.6% of wild type levels, representing a 1.7-fold upregulation of Igf2r from the paternal TDRΔ allele that was however, not statistically significant. In extra-embryonic tissues, in addition to Igf2r, the Slc22a2 and Slc22a3 genes show Airn-dependent imprinted expression [19]. Analysis of all three genes in VYS shows a similar trend for a modest but not consistently significant loss, of paternal repression upon paternal transmission of the TDRΔ allele (Figure S3B–S3D). Together this indicates a similar trend for a minor loss of imprinted repression of protein-coding genes in both ES cells and mid-late gestation embryos and extra-embryonic tissues, indicating that deletion of the TDRs slightly reduces the repressor efficiency of Airn.
The Airn CGI is contained within the ICE that carries a gametic DNA methylation imprint on the maternal allele while the paternal allele is free of methylation. This gametic methylation imprint is present in undifferentiated ES cells as they are derived from the inner cell mass of the 3.5 dpc blastocyst [31]. To test if TDR deletion from the paternal allele compromised the methylation-free state of the paternal ICE, we analysed genomic DNA from undifferentiated (d0) and differentiated (d5 and/or d14) S12/+ and S12/TDRΔ ES cells by methyl-sensitive restriction digestion of genomic DNA (Figure 5A). Wildtype (S12/+) undifferentiated and differentiated ES cells both show a 6.2 kb band originating from the methylated maternal allele and a 5.0 kb band from the unmethylated paternal allele. In S12/TDRΔ ES cells, two fragments were similarly present, the 6.2 kb fragment from the wildtype maternal allele and a 4.3 kb fragment from the unmethylated paternal allele that is shortened by the TDR deletion. In addition, a faint but reproducible 5.5 kb band that must originate from a methylated TDRΔ paternal allele (Figure S1) was detected in undifferentiated but not in differentiated S12/TDRΔ ES cells (*Figure 5A). This indicates a transient gain of DNA methylation in undifferentiated S12/TDRΔ ES cells that is lost during differentiation. Figure S5A shows two additional differentiation sets with similar behaviour. To test if low-level DNA methylation on the paternal ICE represents a property of undifferentiated ES cells, rather than a consequence of the TDR deletion, we performed bisulfite sequencing, specifically analysing the paternal allele, in undifferentiated S12/TDRΔ-1A and -2A ES cells and a control ES cell line with the ICE deleted from the maternal allele (R2Δ/+) [36]. Figure 5B, 5C and Figure S5C–S5E show that low-level DNA methylation from 11–14% with extremes ranging from 0–64%, is a general feature of the paternal ICE in undifferentiated ES cells and not a consequence of the TDR deletion. This low-level DNA methylation is however transient as the ICE becomes methylation-free in differentiated ES cells and in differentiated primary embryonic cells (Figure 5A, Figure S5A, S5D, S5E).
A possible effect of the TDR deletion on the methylated state of the maternal allele cannot be analysed in the ES cell imprinting model, as homologous recombination with the targeting vector replaces the CpG methylated genomic DNA with unmethylated DNA grown in bacteria. We therefore used TDRΔ embryos to analyse DNA methylation of the ICE by methyl-sensitive restriction digestion of genomic DNA. Paternal transmission of the TDRΔ did not affect the maintenance of the unmethylated state, confirming the data obtained from the ES cell imprinting model (data not shown). In contrast, maternal transmission of the TDRΔ led to almost complete loss of the methylated 5.5 kb fragment (detected as 6.2 kb in wildtype mice, Figure 5D). Bisulfite sequencing of mouse embryos carrying a maternal TDRΔ (TDRΔ/+ and TDRΔ/Thp) or a wildtype maternal allele (+/+) confirms that maternal transmission of the TDRΔ allele led to near complete loss of the maternal methylation imprint (Figure 5E, 5F, Figure S5F). The TDRΔ allele showed mean methylation levels of only 4% with extremes ranging from 0–29%. Together these results demonstrate that the TDRs do not play a role in the maintenance of the methylation-free state of the paternal ICE but are essential for methylation on the maternal ICE. Whether the TDRs play a role in the acquisition of the maternal ICE methylation mark in oocytes or its maintenance at later developmental stages was not determined. The loss of maternal ICE methylation in TDRΔ/+ embryos and VYS, resulted in maternal expression of the same progressively shorter Airn transcript with similar ability to repress Igf2r, Slc22a2 and Slc22a3 (Figure S3E–S3J), as shown above for a paternal TDRΔ allele. Neither paternal nor maternal TDR inheritance has an effect on viability or fertility, examining respectively 29 and 65 offspring. In addition TDR homozygotes are obtained in the expected ratio from double heterozygote crosses (i.e., 19 wildtype, 34 heterozygotes and 12 homozygotes were found in 65 offspring). However, although male TDR homozygotes are fertile and produce viable young (16 offspring from 4 litters), female TDR homozygotes show reduced fertility and do not produce viable offspring (10 offspring were obtained from 4 litters but all died within 2 days, indicating a role for Igf2r in the female reproductive tract as noted earlier [37]. We have previously shown that low levels of Igf2r that continue to be expressed from the paternal allele in wildtype embryos (approximately 5%, see Thp/+ in Figure 4D) are not sufficient for viability in the absence of a maternal Igf2r allele [37]. Live born fertile TDRΔ offspring are obtained with Igf2r levels that average 16% (ranging from 11–21%) of wildtype at 12.5–13.5 dpc (see Figure S3F). These crosses contain 129Sv and C57BL/6J genotypes and additional contribution to the survival of TDRΔ/+ mice could come from a mixed genetic background, which was previously shown to influence viability upon loss of maternal Igf2r contribution [37], [38].
The above data shows that the TDRs that lie in the 3′ half of the CGI (Figure 1B), act on the paternal chromosome to control the full-length of Airn and on the maternal chromosome to regulate ICE DNA methylation. To determine if the CGI contains additional elements regulating Airn expression we next removed the entire CGI. The same wildtype parental (S12/+) ES cells were used to delete a 1129 bp fragment, starting 177 bp downstream of the Airn-TSS and ending at the same position as for the TDR deletion (Figure S6A). This deletion left behind 106 bp of the 5′ part of the CGI including the diagnostic MluI site, however the remnant is too small to fit conventional CGI definition criteria [39]. The selection cassette was inserted at the same position as for the TDR deletion to avoid leaving a loxP site at the deletion site. We obtained two homologously-targeted clones that were both targeted on the paternal allele (S12/CGIΔ+cas-1, -2) (Figure S6B, S6C). The selection cassette was removed by transient CRE expression and four ES cell subclones (S12/CGIΔ-1A/-1B/-2A/-2B) were used for analysis in the ES cell imprinting model (Figure 1A, Figure S6B).
To test if the CGI deletion enhanced the shortening of the Airn transcript observed after the TDR deletion, we differentiated S12/+ and S12/CGIΔ-1A ES cells and analysed them by RNA hybridization to genome tiling arrays (Figure 6A). In contrast to the S12/+ cells (and S12/TDRΔ cells in Figure 2), the relative signal intensities from S12/CGIΔ cells did not increase at the transition from the ‘Igf2r-specific’ to the ‘overlap’ region but instead were similar. Since signals in the overlap region are derived from both Igf2r and Airn, this indicated an absence of Airn transcription in this region or a shortening of Airn not resolved on the array. In addition, S12/CGIΔ cells showed a sharp drop in signal intensity at the start of the ‘Airn-specific’ region and signals were not detected after 73 kb downstream of the Airn transcription start (respectively single and double dashed arrows, Figure 6A). This was in contrast to Airn in TDRΔ cells that showed a drop in signal intensity from 68 kb onwards and no signal only after 90 kb (Figure 2). Lastly, higher signal intensities in the Igf2r-specific region in S12/CGIΔ cells compared to wildtype, indicate a gain of bi-allelic Igf2r expression. We also performed strand-specific RNA-Seq and plotted the log2 ratio of the number of reads originating from the forward and reverse strand to obtain an estimate for strand-specific expression in the analysed region (Figure 6A, dotted lines). In S12/+ cells, reads in the ‘Igf2r-specific’ region show specific expression of Igf2r. In the ‘overlap’ region, the ratio then shifts towards the Airn-expressing forward strand, which is even more pronounced in the ‘Airn-specific’ region. In S12/CGIΔ cells, a similar albeit less pronounced shift in the ratio is seen upon transition from the ‘Igf2r-specific’ to the ‘overlap’ region, with a further shift occurring at the transition from the ‘overlap’ to the ‘Airn-specific’ region that is not detected after 73 kb and is reduced compared to S12/+ cells. This confirms in a strand-specific manner, a low level but persistent Airn-expression upon deletion of the CGI. Together, these data indicate that high Airn expression and production of full-length Airn transcripts and as a consequence, the ability to repress Igf2r in cis, are dependent on the CGI.
To validate the genome tiling array data we analysed cDNA from undifferentiated and differentiated wildtype (S12/+) and CGIΔ (S12/CGIΔ) ES cells using five qPCR assays spaced along the length of Airn. In undifferentiated ES cells with and without the CGI, Airn expression was mostly absent consistent with the previously observed lack of Airn expression in undifferentiated ES cells [31]. The low level of Airn expression seen in undifferentiated ES cells in Figure 6B represents a small amount of spontaneous differentiation that was similar in wildtype and CGIΔ ES cells. During differentiation Airn was upregulated in wildtype ES cells using all five qPCR assays. In contrast, the CGIΔ ES cells showed consistently less Airn at all analysed positions (Figure 6B). However, the extent of the loss of steady-state levels differed along the length of the transcript. The START assay showed that total unspliced and spliced Airn is reduced to an average of 24–30%. Unspliced Airn detected with RP11 at 154 bp downstream of the TSS showed an average reduction to 50–65%, which was statistically significant in 3 of 4 clones at d14. This difference between total and unspliced Airn is explained by the major loss of the Airn splice variants that require transcription elongation to at least 72 kb (Figure 2), and represent up to 44% of steady-state Airn levels [21]. Unspliced Airn detected with the AirnT3 assay at 2.8 kb from the TSS on the CGIΔ allele showed an average reduction to 8–14% and detection by the Airn-middle assay at 52 kb from the TSS showed reduction to 6–8% (note that distances from the TSS on the CGIΔ allele are reduced by 1129 bp). Finally at Airn-end (98 kb from the Airn-TSS) the reduction was to 0–0.4% of wildtype levels. This indicates that the CGI deletion induced successive loss of Airn with increasing distance from the 5′ end.
To further map the observed shortening of Airn we used two more assays at the 5′ end (Figure 6C). In CGIΔ cells, the Airn-124 assay at 570 bp from the TSS showed an average reduction of Airn steady-state levels to 17–21%, while the Airn-117 assay at 7.3 kb from the TSS showed an average reduction to 13–14%. We also analysed steady-state levels of two Airn splice variants to see if the splicing suppression was altered by the CGI deletion (Figure 6C). The RP6 assay showed the SV1a splice variant is reduced on average to 5–7% of wildtype levels, while the RP21 assay showed the SV1 splice variant is reduced on average to 4–7%. Both these splice variants require transcription elongation to 72 kb (Figure 2). Splicing suppression was therefore not altered after the CGI deletion as the abundance of splice variants decreased in a similar manner as unspliced Airn. Furthermore, the abundance of splice variants from the CGIΔ allele is at most 7% of wildtype levels. As splice variants represent up to 44% of the Airn steady-state population, this indicates that the 24–30% of steady-state levels observed with the START assay (Figure 2B) represent more than 60% of initiating transcripts, confirming the result obtained for the RP11 assay. Together the data shows that Airn full-length elongation is significantly affected by deletion of the CGI with a successive loss of Airn with increasing distance from the 5′ end. However, by analysing RNA steady-state levels by qPCR a more moderate change is seen in transcription initiation such that 50–65% of Airn transcripts elongate at least to 154 bp.
To test if the observed decrease of Airn ncRNA expression was reflected by altered recruitment of RNA polymerase II (RNAPII) we performed chromatin immunoprecipitation using antibodies specifically recognising initiating RNAPII phosphorylated at the Serine 5 (Ser5P) residue of its carboxy-terminal domain (CTD) and elongating RNAPII phosphorylated at Serine 2 (Ser2P) of its CTD. RNAPII occupancy in differentiated S12/+, S12/TDRΔ-1A and S12/CGIΔ-1A ES cells was analysed at five positions along the gene body of Airn as well as in intron 5 of Igf2r to control for overlapping Igf2r transcription (Figure 6D map). Whereas equal amounts of RNAPII Ser5P were found in S12/+ and S12/TDRΔ-1A cells, it was strongly reduced at the Airn 5′ region in S12/CGIΔ-1A cells, indicating reduced Airn transcriptional initiation on the CGIΔ allele (Figure 6D left). For RNAPII Ser2P, S12/+ and S12/TDRΔ-1A showed similar enrichment except for Airn-end, where S12/TDRΔ-1A showed reduced levels. In S12/CGIΔ-1A cells, RNAPII Ser2P levels were increased in intron5 of Igf2r consistent with the increase in Igf2r levels observed in Figure 6A. RNAPII Ser2P levels within the Igf2r/Airn transcriptional overlap were lower compared to the other two cell lines indicating that Airn transcriptional elongation on the CGIΔ allele is strongly reduced (Figure 6D right). An independent RNAPII ChIP experiment showed a similar result (data not shown). Together, the analysis of RNA levels and RNAPII occupancy indicate that the CGI which is localised downstream of the Airn promoter, controls Airn initiation and elongation.
As the majority of Airn transcripts were only between 154–570 bp long (Figure 6C) we tested if Airn produced from the CGIΔ allele was unable to silence Igf2r, as expected from previous experiments that truncated Airn to 3 kb from the TSS [19]. We first analysed the DNA methylation status on the paternal Igf2r promoter-associated CGI, as described in Figure 4A for the TDRΔ. Figure 6E and Figure S7 show that in contrast to wildtype ES cells, all four CGIΔ ES cell lines fail to gain DNA methylation on the paternal Igf2r promoter during differentiation, indicative of biallelic Igf2r expression in these cells. Next, we performed allelic expression analysis of Igf2r using the qPCR assay described above for the TDR deletion in Figure 4C. The results (Figure 6F) show that all differentiated CGIΔ cells displayed an unchanging maternal∶paternal expression ratio during differentiation indicative of biallelic Igf2r expression. This contrasts to wildtype cells that show an increasing ratio of maternal∶paternal Igf2r expression during differentiation indicative of maternally-biased imprinted expression. An absence of Igf2r imprinted expression can also be inferred from the tiling array analysis in Figure 6A where increased Igf2r hybridization signals are seen in the Igf2r-specific region and from the increased RNAPII occupancy in Igf2r intron 5 in Figure 6D. Thus, these results show that Airn transcripts in S12/CGIΔ ES cells, the majority of which had a length of between 154–570 bp are as expected, defective in their ability to silence Igf2r.
Finally, we tested if deletion of the CGI affected the methylation-free state of the Airn promoter region on the normally unmethylated paternal allele. The CGI deletion left behind 106 bp of the 5′ part of the CGI including the diagnostic MluI site analysed for the TDR deletion in Figure 5A. Undifferentiated ES cells with a wildtype paternal allele (S12/+) showed a 6.2 kb maternally methylated and an equally strong 1.1 kb paternally unmethylated band (Figure 7A). Cells with a paternal CGI deletion (S12/CGIΔ) showed a wildtype 6.2 kb maternally-methylated fragment and a 5 kb paternally-methylated band. The size of the paternal fragment is reduced to 1.1 kb when unmethylated. In Figure 7A we also examined CGIΔ cells with (S12/CGIΔ+cas) and without (S12/CGIΔ) the selection cassette, to obtain information from cells that had experienced a short and long culture period since the loss of the CGI. Compared to S12/CGIΔ+cas cells, S12/CGIΔ cells that lack the selection cassette have been an additional 8 passages in culture and show an increased intensity of the 5.0 kb band indicating that DNA methylation increases with passage number. However, in contrast to the TDR deletion shown in Figure 5A, DNA methylation was not lost upon differentiation as indicated by the similar intensity of the 5 kb band in d0 and d14 S12/CGIΔ cells (Figure 7B, S8). Bisulfite sequencing was used to determine the extent of DNA methylation on the CGIΔ allele in undifferentiated ES cells using primers spanning the deletion that specifically amplify the paternal CGIΔ allele (Figure S5B). In two S12/CGIΔ ES cell lines the results show a high level of DNA methylation (Figure 7C left). The % methylation levels were 69–76%, with extremes ranging from 9–100% (Figure 7C right). Taken together, this analysis shows that deletion of the CGI leads to a strong non-reversible gain of DNA methylation in cis, indicating that one major function of the CGI on the paternal allele is to block DNA methylation on the paternal ICE.
This study assessed the possible role played by the Airn CpG island (CGI) and a region of tandem direct repeats (TDRs) on the Airn transcript and the allelic methylation state of the ICE that controls Airn expression. Since these elements lie immediately downstream of the Airn transcription start and thus are present on the Airn transcript, they may also play a role in Airn biology. Our results using two targeted deletions analyzed in an ES cell imprinting model and in knockout mice, show that the CGI regulates both initiation and elongation efficiency of the Airn promoter and is also necessary to maintain the unmethylated state of the paternal ICE. This indicates the existence of a new transcriptional role for CGIs in the mammalian genome acting downstream of the promoter and transcription start site. In contrast, the TDRs that occupy the second half of the CGI play a minor role in Airn transcriptional processivity but are essential for methylation of the maternal ICE that represses the Airn promoter on this chromosome.
The 1129 bp deletion of the complete Airn downstream CGI had a moderate effect on the most 5′ RNA levels such that two qPCR assays within the first 154 bp downstream of the Airn-TSS detected approximately 50–65% of wildtype levels of the normally 118 kb long Airn ncRNA. Airn transcripts were reduced to ∼14% between 1.7–8.5 kb and to ∼6% between 53–73 kb, while no transcripts were detected at 99 kb. In addition, both initiating and elongating forms of RNAPII were reduced compared to control cells. In contrast the TDR deletion had a minor effect on the length of Airn with transcripts at normal levels for the first two-thirds, but progressively reduced after 68 kb and absent at 99 kb, with the elongating form of RNAPII reduced at the 3′ end of Airn. This indicates that the efficiency of RNAPII to elongate Airn over 118 kb is regulated by the CGI and at least in part, also by the TDRs. Notably, in view of the splicing suppression of wildtype Airn that results in splicing of only 5% of transcripts [21], the production of all four splice variants was decreased in proportion to the unspliced transcripts indicating that neither the CGI nor TDRs cause splicing suppression.
We have previously shown that Airn must be longer than 3 kb and be expressed from a strong promoter, to induce silencing of the overlapped Igf2r promoter [19], [32]. A loss of paternal Igf2r repression after the CGI deletion that shortened the majority of Airn transcripts to less than 0.5% of its normal length is therefore expected, and this deletion was only analysed in the ES cell imprinting model. The TDR deletion although producing normal levels of the Airn transcript that overlap the 28 kb distant Igf2r promoter and elongate up to 90 kb from the Airn-TSS, nevertheless showed a minor loss of paternal Igf2r repression. This was seen as a 1.7–2.0 fold upregulation of the paternal Igf2r allele in differentiated ES cells and in mid-late gestation embryos and VYS, which was not statistically significantly different from paternal repression on wildtype chromosomes. A similar minor increase in paternal steady-state levels was observed for Slc22a2 and Slc22a3 in the VYS. We lack an explanation for this minor effect. It appears not to arise from changed developmental kinetics of Airn expression that were similar in wildtype and TDRΔ differentiating ES cells, but it may reflect changes in RNAPII post-translational modification not detected with current antibodies. Currently it is unknown if the Airn ncRNA or the act of its transcription induce imprinted expression of Igf2r [16], [40]. For the Slc22a3 gene that lies 275 kb downstream of Igf2r and is repressed by Airn only in placenta, Airn was shown to localize to the Slc22a3 promoter and to induce imprinted expression by interacting with the G9A histone methyltransferase. However, imprinted expression of Igf2r was not affected in these studies [41] or in studies eliminating PRC2 activity [42]. The results obtained here do not distinguish between a role for the Airn ncRNA or its transcription, but are in agreement with previous analyses that demonstrated a role for high Airn expression and a length longer than 3 kb to repress Igf2r in cis [21], [32].
The Xist macro ncRNA that induces whole chromosome silencing in female XX mammals has been suggested to share similarities with imprinted repressor ncRNAs such as Airn and Kcnq1ot1 [43], [44]. Notably Xist contains a set of 5′ direct ‘A’ repeats that are essential for Xist to induce chromosome silencing [45]. The Airn TDRs may have served a similar purpose. Here we show that the TDRs are not required for Airn to repress its target genes as despite the minor loss of paternal repression, imprinted expression is present in TDRΔ cells and mice. However, since the TDRs are required for maternal ICE methylation, they are necessary to ensure expression of the maternal Igf2r allele as it has been previously shown that mouse embryos lacking maintenance DNA methylation, repress both parental Igf2r alleles [22]. The imprinted Kcnq1ot1 macro ncRNA shares many features with the Airn ncRNA and its TSS lies on the 5′ border of the CGI (Figure 1C), which contains a series of TDRs that lack sequence conservation with those in Airn [30], [46]. Two overlapping deletions have been used to test the function of this region in the Kcnq1ot1 ncRNA. The first is a 657 bp deletion starting just downstream of the Kcnq1ot1-TSS [18], while the second deletion removed 890 bp and overlapped 40% of the 657 bp deletion [47]. The ability of the deleted Kcnq1ot1 ncRNA to repress flanking genes on the paternal chromosome was found to be unchanged in midgestation embryos for the 657 bp deletion. However, the 890 bp deleted Kcnq1ot1 allele showed a failure to repress some genes in this cluster in a lineage-specific manner that correlated with failure to gain DNA methylation on the derepressed genes. Although the failure to repress flanking genes was attributed to a failure in recruiting DNMT1 due to the lack of the 890 bp region in the ncRNA [47], both Kcnq1ot1 and Airn are able to repress genes in mouse embryos lacking the Dnmt1 gene that are deficient in genomic methylation [29], [48]. The TDR deletion described here resulted in loss of the 3′ part of Airn and a minor loss of paternal repression of protein-coding genes with the paternal Igf2r promoter showing a normal gain of DNA methylation (the Slc22a2 and Slc22a3 genes are repressed in the absence of promoter methylation [49]). A direct comparison between the two imprinted clusters is not possible since although Kcnq1ot1 steady-state levels were unchanged in both the deletion experiments [18], [47], measurements were only made in the first half of the transcript and it is not known if these deletions affected the full-length of Kcnq1ot1.
Classic mammalian promoter-associated CGIs extend upstream and downstream of the transcription start of the majority of mouse and human genes and these CGIs are considered to have promoter regulatory functions [1]. The promoter region of a CGI is perceived as the region between the 5′ boundary of the CGI and the TSS [50], although none have been subject to deletion at the endogenous locus and analyzed as described here. Recently, evidence has been accumulating that gene regulation acts not only at the step of RNAPII recruitment by the promoter, but also at later steps of transcription elongation and processing [51]–[53]. The data here show that elements located downstream of the transcription start site are required for RNAPII transcription initiation and elongation and also indicate that CGIs can play a different role to that of the upstream promoter.
Reduced Airn transcript length could be explained by alternative polyadenylation site choice that is often seen in mammalian genes [54]. The Airn ncRNA produces four splice variants, three of which have alternative polyadenylation sites spread over 45 kb (Figure 2) [21]. Although premature polyadenylation could explain progressive Airn shortening in TDRΔ and CGIΔ alleles, we think this unlikely for two reasons. First, the genome tiling array analysis shows Airn shortening is gradual and not stepwise, which would be expected from use of alternative polyadenylation sites. Second, the RT-qPCR data indicate that Airn shortening on CGIΔ alleles occurs within the first 570 bp, which does not contain a known polyadenylation site (http://rulai.cshl.edu/tools/polyadq/polyadq_form.html). Cells with a paternal TDRΔ allele showed similar occupancy of the initiating and elongating forms of RNAPII to wildtype cells, except for the 3′ end of Airn where elongating RNAPII was reduced. As Airn transcription initiation is unchanged in TDRΔ cells with the majority of transcripts longer than 68 kb, this indicates the length of Airn is subject to regulation after the switch between paused and elongated transcription. In cells with a CGIΔ allele however, both initiating and elongating RNAPII were decreased compared to wildtype and TDRΔ cells, although ∼60% of wildtype RNA levels were found at the very 5′ end. This indicates that the deletion of the whole CGI affected the ability of the upstream promoter region not only to elongate but also to efficiently initiate Airn transcription. The finding that both the TDR and CGI deletions induced progressive Airn shortening indicates that cumulative elements distributed throughout the CGI play distinct roles in regulating Airn transcription elongation and processivity.
An obvious feature involved in regulating expression of a CGI associated gene is DNA methylation. Gain of methylation was not seen on the paternal TDRΔ allele, but up to 70% of DNA methylation was gained on the flanking sequences after paternal CGI deletion. However, Airn 5′ levels only showed a moderate change on the CGIΔ allele. As methylation levels showed a high variability between different alleles, ranging from 9–100%, it could be possible that hypomethylated alleles are still able to initiate Airn transcription as detected by RT-qPCR which specifically analyses Airn transcripts, but not by the RNAPII ChIP which might be relatively less sensitive and also suffers from background problems due to increased Igf2r levels in the overlap region. We therefore suggest that this gain of methylation and not loss of the CGI, explains the reduction in Airn transcription initiation shown by the CGIΔ allele. Since most CpG dinucleotides including those in the body of genes, are methylated when they lie outside CGIs [55], it is clear that DNA methylation downstream of promoters does not block transcript elongation of endogenous mammalian genes. Furthermore, as no increase in DNA methylation was observed upon TDR deletion on the paternal allele that also induced shortening of Airn, we can exclude DNA methylation as the cause of the length phenotype. Thus, loss of sequences within the CGI and not gain of DNA methylation correlate with loss of full-length Airn.
Deletion of the TDRs removed the 3′ half of the CGI from the paternal ICE but did not change its unmethylated status. Notably, controls used in these experiments allowed us to observe for the first time, a low level of DNA methylation on the wildtype paternal ICE in two different undifferentiated ES cell lines, that was fully reversible upon differentiation and was also absent in differentiated primary embryonic fibroblasts. Although Airn is not expressed in undifferentiated ES cells, the paternal ICE is marked by H3K4me3 [31], [56], which has been shown to block DNMT3L, an essential cofactor for the de novo methylation complex, from binding histone H3 [57]. The existence of low-level DNA methylation at the Airn promoter in undifferentiated cells despite the presence of H3K4me3 indicates either, that high Airn expression induced during differentiation is required in addition to H3K4me3 to fully block DNA methylation or, that DNA methylation modifies a small number of chromosomes in the population that lack H3K4me3 [58]. Deletion of the whole CGI led to a substantial gain of DNA methylation on the paternal allele that was not reversible upon differentiation but was enhanced after removal of the selection cassette. We attribute this enhancement to the longer period in cell culture required to remove the cassette. Thus the CGI deletion shows that one role located in the first half of the island, is to block DNA methylation on the paternal Airn promoter that is 177 bp upstream from the deleted sequences. Transgene reporter experiments have been used to show that SP1 transcription factor binding sites protect a CGI from DNA methylation [59]–[62]. Furthermore a high CpG density also correlates with protection from DNA methylation by recruitment of the CpG-binding protein CFP1, which in turn leads to H3K4me3 via recruitment of the SETD1 histone methyltransferase [58]. As the CGI deletion reduces CpG density considerably and removes three predicted SP1 binding sites [21], this may explain the gain of DNA methylation upon deletion of the CGI.
Deletion of the 3′ half of the CGI that included the TDRs, led to loss of ICE methylation following maternal transmission of the deleted allele. The Airn-TDRs are conserved in human and mouse at an organizational level and in their ability to be methylated on the maternal chromosome only [27], [63]. The conservation of TDRs in the ICE may be explained by the preference of the DNMT3A de novo methyltransferase for an 8–10 bp periodicity in CpG frequency, that is seen in the 12 known maternally-methylated ICE [64]. Previous experiments using multicopy transgenes randomly inserted in the genome have also identified the Airn-TDRs, in particular the three long 172–180 bp monomer repeats, as important for maternal-specific methylation of a hybrid RSVIgmyc imprinted transgene [65]. These experiments also demonstrated a role for the TDRs in maintaining the unmethylated state on paternal transmission. The data reported here that deleted the TDRs from the endogenous Airn CGI, confirm a role for the TDRs in the methylation of the maternal ICE, but do not demonstrate a role in maintaining the unmethylated state of the paternal ICE. The two overlapping 657 and 890 bp deletions cited above for the Kcnq1ot1 downstream CGI [18], [47], were not directly tested for their role in the methylation of the maternal ICE. Indirect evidence that indicates no role for these deleted regions comes from the finding that the maternal transmission of the 890 bp deletion did not lead to derepression of Kcnq1ot1. Together this would indicate that the Airn-TDRs but not the Kcnq1ot1 TDRs, have a function in methylation of the maternal ICE. However two minor caveats could be considered. First, the two overlapping deletions reported from the Kcnq1ot1 downstream CGI might not have removed all necessary sequences and second, these two overlapping deletions left a single loxP site at the site of the deletion, which has been reported to attract DNA methylation [33]. In contrast, the Airn-TDR deletions reported here placed the remaining single loxP site 2 kb upstream from the deletion and we are now able to assign a specific role for the Airn-TDRs in the methylation of the maternal ICE at the endogenous locus.
Together the data presented here show that the CGI lying immediately downstream of the Airn transcription start regulates both the epigenetic and transcription state of its upstream promoter. Classically, with the exception of retrotransposons, RNA polymerase II promoters are viewed as lying upstream of the transcription start [66], [67]. In contrast, the majority of CGIs can be seen in Figure 1C to extend downstream of the transcription start, with some located entirely downstream of the transcription start. The importance of CGIs as regulators of gene expression has been emphasised with the advent of genome-wide studies showing CGIs are not only associated with genes showing tissue-specific and inducible expression but are also present in large numbers as orphan CGIs not associated with annotated promoters [1], [2], [68]. The data here identify a role for the downstream Airn CGI to regulate its epigenetic state and the production of transcripts expressed at sufficiently high levels and of sufficient length to silence flanking target genes. Future work will determine how this regulation is achieved and if these features are shared by CGIs regulating non-imprinted gene expression.
Mice were bred and housed at the Forschungsinstitut für Molekulare Pathologie GmbH, Dr. Bohr-Gasse 7, 1030 Vienna, Austria in strict accordance with national recommendations described in the “IMP/IMBA Common Institutional policy concerning the care and use of live animals” with the permission of the national authorities under Laboratory Animal Facility Permit MA58-0375/2007/4. Blastocyst injections and chimeric mice were prepared under the permit M58/003079/2009/8: Production of Chimeras, Examination of Germline, Examination of Gene Effects in Parents and Successor Generations (Model B). Mouse embryos were obtained after humane killing of pregnant female mice by cervical dislocation by skilled qualified personnel.
All targeting vectors were generated from a plasmid with a 6.4 kb 129Sv homology region (chr17:12931344–12937792/NCBI37-mm9). In the TDRΔ construct a 692 bp SacII-NsiI fragment (chr17:12934848–12935543) was deleted. The selection cassette (loxP)-(HSVTk-Neomyocin-SV40polyA)-(HSVTk-ThymidineKinase+polyA)-(loxP) for the ES cell imprinting model and (loxP)-(Pgk1-Neomycin-Pgk1polyA)-(loxP) for blastocyst injection was subcloned into the NheI site at chr17:12932836. In the CGIΔ construct the 1129 bp deletion (chr17: 12934414–12935543) was created by PCR (primers: TGGAACCCTTCCTTTGCGGAATC - TGCATGAGGGTGCCACACTCCT). The selection cassette: (loxP)-(Pgk1-Neomycin-Pgk1polyA)-(loxP) was inserted at the same position as for TDRΔ. Electroporation and neomycin-selection were performed using standard conditions into S12/+ cells (a D3 feeder-dependent 129 ES line previously modified to carry a SNP in Igf2r exon 12 [31]) for the ES cell imprinting model experiments and into the feeder-dependent BL6/129 intraspecies A9 ES cell line for blastocyst injection. The selection cassette in the ES cells used for the ES cell imprinting model was removed by electroporation of the pMC-Cre plasmid leaving a single loxP site 2 kb upstream of each deletion. One A9 ES cell clone carrying the TDRΔ+cas allele was injected into C57BL/6J blastocysts and transferred into pseudo-pregnant recipient mice and one chimeric male mouse was obtained who transmitted the TDRΔ+cas allele. The selection cassette was removed by crossing TDRΔ+cas males with MORE-Cre females. Heterozygous TDRΔ mice were mated with wildtype FVB or FVB with a Thp allele and embryos were isolated at 12.5 dpc or 13.5 dpc. Visceral yolk sacs were isolated as described in [29].
ES cells were grown on irradiated primary mouse embryonic fibroblasts using standard conditions and differentiation induced by feeder-depletion, LIF-withdrawal and 0.27 µM retinoic acid.
RNA was isolated using TRIreagent and was DNaseI treated prior to reverse transcription. Realtime qPCR for Taqman assays was as described [21]. SybrGreen assays used 100 nM primers and cycling conditions: 5 min 95°C, 40 cycles: 15 sec 95°C+1 min 60–65°C. Allele-specific qPCR was as described [31] with 5 mM MgCl2. The assay specificity was improved by a mismatch in the primer body. See for primers and probes. All assays were normalised to CyclophilinA. DNA isolation and blots were performed using standard techniques and ImageJ quantified signal intensities. For some blots the contrast was linearly enhanced with Adobe Photoshop. RNA hybridization to genome tiling array was performed by Source BioScience LifeSciences, Berlin, as described [69]. The data were Tukey bi-weight normalised before analysis. Relative signal intensities (normalised to the average signal in the region) of overlapping windows of 9 tiles were averaged and each displayed data point is the average of 20 windows, the standard deviation is displayed as error bars. Two pseudogenes in the region, Au76 and LA41, were removed from the analysis. RNA sequencing:1 µg of total RNA was treated with the RiboZero kit (Epicentre) and two strand-specific RNA-Seq libraries prepared using the ScripSeq kit and two compatible barcodes (Epicentre) according to the manufacturer's protocol. Sequencing and read alignment to the mouse genome (mm9) was as described [70]. The region shown in Figure 6A was divided into non-overlapping 3.2 kb windows, reads mapping to the forward or reverse strand in these windows were counted and the log2 ratio of these counts was calculated and plotted. Windows overlapping Igf2r exons and the Au76 and the LA41 pseudogenes were removed from the plot.
Preparation of soluble chromatin and chromatin immunoprecipitation assays were carried out as described [71]. 25 µg of sonicated chromatin were diluted 10-fold and precipitated overnight with the following antibodies: Phospho RNA Pol II (S5) (Bethyl Laboratories A300-655A), Phospho RNA Pol II (S2) (Bethyl Laboratories A300-654A) or rabbit IgG (Invitrogen 10500C) as control. Chromatin antibody complexes were isolated using Protein A magnetic beads (Dynabeads). The extracted DNA was then used for qPCR as described above. A 1∶20 dilution of input DNA was assayed.
1 µg genomic DNA from undifferentiated ES cells grown on feeder cells carrying a homozygous ICE deletion or 1 µg genomic DNA from 12.5–13.5 dpc embryos was RNaseA treated, EcoRI digested and Bisulfite converted using the EpiTect Bisulfite Kit (Qiagen). PCR amplification used JumpStart Taq DNA Polymerase (Sigma), primers: DMR2-F4 (GGGGAATTGAGGTAAGTTAGGGTTTT) with DMR2-R4 (TCTTATAACCCAAAAATCTTCACCCTAAC) for wt alleles or DMR2-R9 (AACACCTTCATATACCCCTAAACAC) for TDRΔ and CGIΔ alleles [8], cycle conditions: 94°C 1 min, 40 cycles of 94°C 1 min, 60°C 1 min, 72°C 1 min then 72°C 5 min. PCR fragments were gel-purified, subcloned and plasmid DNA from single colonies sequenced using standard primers. Analysis and sequence quality control used BiQAnalyzer and standard settings [72].
Each CGI (UCSC Genome Browser, mm9) with flanking regions (50% of the CGI length upstream and downstream) was divided into 100 equal-sized bins (i.e., parts) and the number of RefSeq genes (USCS mm9) was calculated with a transcription start site in each bin. Bins were summed for all CGIs and plotted using Microsoft Excel.
For qPCRs an unpaired t-test was performed using www.graphpad.com/quickcalcs/.
|
10.1371/journal.pgen.1004487 | Comparative Phylogenomics Uncovers the Impact of Symbiotic Associations on Host Genome Evolution | Mutualistic symbioses between eukaryotes and beneficial microorganisms of their microbiome play an essential role in nutrition, protection against disease, and development of the host. However, the impact of beneficial symbionts on the evolution of host genomes remains poorly characterized. Here we used the independent loss of the most widespread plant–microbe symbiosis, arbuscular mycorrhization (AM), as a model to address this question. Using a large phenotypic approach and phylogenetic analyses, we present evidence that loss of AM symbiosis correlates with the loss of many symbiotic genes in the Arabidopsis lineage (Brassicales). Then, by analyzing the genome and/or transcriptomes of nine other phylogenetically divergent non-host plants, we show that this correlation occurred in a convergent manner in four additional plant lineages, demonstrating the existence of an evolutionary pattern specific to symbiotic genes. Finally, we use a global comparative phylogenomic approach to track this evolutionary pattern among land plants. Based on this approach, we identify a set of 174 highly conserved genes and demonstrate enrichment in symbiosis-related genes. Our findings are consistent with the hypothesis that beneficial symbionts maintain purifying selection on host gene networks during the evolution of entire lineages.
| Symbiotic associations between eukaryotes and microbes play essential roles in the nutrition, health and behavior of both partners. It is well accepted that hosts control and shape their associated microbiome. In this study, we provide evidence that symbiotic microbes also participate in the evolution of host genomes. In particular, we show that the independent loss of a symbiosis in several plant lineages results in a convergent modification of non-host genomes. Interestingly, a significant fraction of genes lost in non-hosts play an important role in this symbiosis, supporting the use of comparative genomics as a powerful approach to identify undiscovered gene networks.
| Eukaryotes interact with microbes in a dynamic network of symbiotic associations. These associations represent a continuum from parasitic, where one partner takes advantage of the other one, to mutualistic, where both partners benefit from the interaction. Mutualistic symbioses between eukaryotes and a subset of their microbiome are essential to their nutrition, protection against diseases and development, as exemplified by the gut microbiome in humans or the arbuscular mycorrhizal (AM) symbiosis in plants [1], [2]. During the lifetime of a single individual or at the scale of an entire population, hosts are known to select and shape their associated microbiome [3], [4]. Reciprocally, recent studies shed light on the effect of the microbiome on plant and animal development by modifying gene expression [5]–[7]. However the impact of associated microorganisms on the evolution of host organisms remains poorly characterized.
AM symbiosis is an almost ubiquitous interaction between land plants and AM fungi that has been playing a tremendous role in plant evolution and is proposed to have allowed the colonization of land by plants [8], 9. Nutrient exchanges occur at specialized interfaces, the arbuscules, formed in root cortical cells. Establishment of an efficient symbiosis relies on a set of highly conserved genes characterized in legumes, the so called “symbiotic toolkit” [10]. This toolkit is required for the perception of AM fungi signals, root colonization, arbuscule development and to control the level of root colonization [11]. Interestingly, several angiosperm species, including the model plant Arabidopsis thaliana (Arabidopsis), have lost the ability to form this symbiosis and are non-hosts for AM fungi [12]. Loss of traits is a common feature of eukaryote evolution. It can result from or be the result of modification in gene expression pattern or of gene loss [13], [14]. Targeted phylogenetic analyses in Arabidopsis led to the broad classification of the “symbiotic toolkit” genes into two subsets: 1. a subset called ‘conserved’ genes that is conserved in Arabidopsis thaliana despite the loss of AM symbiosis and 2. a subset of ‘symbiosis-specific’ genes that are absent in this non-host species [10]. Most of the ‘conserved’ genes have been demonstrated to play non-symbiotic roles [15], [16]. In contrast, only symbiotic functions are known for the “symbiosis-specific” group. Thus, it seems that the loss of a symbiotic association might result in the loss of genes specifically required for its establishment and maintenance. A reciprocal hypothesis would be that associated microbes constrain host genomes to maintain symbiotic genes. To test this hypothesis, we developed several approaches using the AM symbiosis as a model. First, focusing on the Arabidopsis lineage (order Brassicales), we tested if the absence of symbiotic ability and the absence of ‘symbiosis-specific’ genes are the result of independent or correlated events. To this end we conducted a large phenotypic screen on Brassicales species. In parallel we analyzed the genomes and/or transcriptomes of Brassicales to determine the absence/presence of symbiosis-specific and conserved genes. Then we performed a similar analysis on four additional non-host lineages. We hypothesized that if symbiotic associations affect the evolution of host gene networks, the loss of symbiotic ability could be correlated with the loss of specific genes. We used a comparative phylogenomic pipeline to determine the global impact of symbiosis loss on non-host plant genomes and potentially identify new genes involved in AM symbiosis.
The eudicot order of Brassicales encompasses many non-host species for AM fungi, such as the model plant Arabidopsis thaliana (Brassicaceae), and hosts such as papaya (Carica papaya, Caricaceae) [17]. To investigate the distribution of non-host species across the Brassicales, we tested the symbiotic status of eighteen Brassicaceae species, including Aethionema arabicum that belongs to the earliest diverging lineage in the family, and fourteen other species distributed across more basal Brassicales families, including Cleomaceae, Resedaceae, Limnanthaceae and Moringaceae (Figure 1A). Among the tested species only Moringa oleifera was well colonized by AM fungi (Figure 1A, B). Then we used ancestral trait reconstruction and the published phylogeny of Brassicales [18] to determine the number of transitions between host and non-host states. This analysis predicted a single transition in the Brassicales, before the divergence of the Limnanthaceae (Figure 1A). Most of the symbiotic toolkit is absent in Arabidopsis but its conservation in other Brassicales species was unknown. In order to determine when the ‘symbiosis-specific’ genes have been lost in Brassicales and test if this loss correlates with the loss of the symbiotic ability, we assessed the presence of these genes in five sequenced Brassicaceae genomes, in the transcriptomes of four other Brassicaceae, including Aethionema arabicum, and in thirteen other taxa belonging to more basal Brassicales families. We also included the genomes of cacao (Theobroma cacao, Malvaceae), cotton (Gossypium raymondii, Malvaceae) and papaya which are three well-characterized host species [17]. The ‘conserved’ genes were present in all tested taxa (Figure 2). In contrast, ‘symbiosis-specific’ genes were only found in the genomes or transcriptomes of host species (Figure 2).
To further assess the absence of these genes, we conducted comparative whole-genome synteny analyses of hosts (Grape, Poplar, Peach, and Papaya) and non-hosts (Arabidopsis thaliana, Tarenaya hassleriana, Brassica rapa, and Aethionema arabicum). We identified genomic blocks containing ‘symbiosis-specific’ genes and ‘conserved’ genes in the host genomes, and localized the syntenic blocks in the genomes of the four non-host Brassicales (Text S1, Table S1). The ‘conserved’ genes were present in the corresponding syntenic block, whereas “symbiosis-specific” genes where missing from these syntenic blocks confirming their likely absence in non-host genomes (Text S1, Table S1). The absence of detectable transcript in transcriptome data could be a sampling bias due to the lack or low levels of gene expression or due to actual gene loss by pseudogenization or deletion. In order to test if low expression levels or lack of expression could explain our transcriptome observations, we applied a generalized linear model to evaluate the probability for each gene to be detected in the transcriptome of each species if this gene is actually present (see Methods). Our model predicts that at least five ‘symbiosis-specific’ genes should be detected if present, hence strongly supporting their absence in each of the non-host Brassicales species where we did not detect them (Figure S1). For the six other genes, we calculated the probability to detect them in at least one non-host species if present in all of them and confirmed their likely absence for four of them (Table S2). Our data strongly support that the loss of AM symbiosis in Brassicales correlates with the large-scale deletion or pseudogenization of ‘symbiosis-specific’ genes.
Besides Brassicales, the AM symbiosis has been lost independently in several lineages of flowering plants [19]. Using publicly available genomic and transcriptomic data, we investigated the presence of genes from the symbiotic toolkit in these non-host lineages. We first tested the presence of these genes, either ‘conserved’ or ‘symbiosis-specific’, in the genomes of sugar beet and spinach (Beta vulgaris and Spinacia oleracea, Amaranthaceae, Caryophyllales [20]), in the genome of a carnivorous plant Utricularia gibba (Lentibuliaraceae, Lamiales, [21]), and in the transcriptome of three obligate parasitic plants Cuscuta sativa (Convolvulaceae, Solanales [22]), Striga hermontica, and Orobanche aegyptiana (Orobanchaceae, Lamiales, [23]) that are all well-characterized non-hosts for AM fungi. As controls, we used transcriptome data from close relatives: Sesamum indicum (Pedaliaceae, Lamiales [24]), Capsicum anuum (Solanaceae, Solanales [25]), Ipomoea batatas (Convolvulaceae, Solanales, [26]), and Lindenbergia philippensis, a basal and non-parasitic Orobanchaceae. We also included as outgroups the sequenced genomes of monkey-flower (Mimulus guttatus, Scrophulariaceae, Lamiales) as well as the genomes of tomato and potato (Solanum lycopersicum and Solanum tuberosum, Solanaceae [27], [28]). All control and outgroup species are able to develop bona fide associations with AM fungi [29]–[31] (Figure 3B). ‘Conserved’ genes, but no ‘symbiosis-specific’ genes, were found in the genome and/or transcriptome data of non-hosts (Figure 3, S2). In contrast, both groups of genes were present in host species (Figure 3). In addition, by applying the probabilistic analysis described above, we predicted the likely absence for several of the ‘symbiosis-specific’ genes in Striga hermontica and Orobanche aegyptiana using their transcriptomes (Figure S2 and Table S3).
Legume species in the genus Lupinus (lupines) are also well-known non-hosts for AM fungi [32]. Despite the absence of AM symbiosis, Lupinus species are able to associate with nitrogen-fixing rhizobia, leading to the development of root nodules [33]. This rhizobium–legume symbiosis requires part of the symbiotic toolkit, called the ‘common symbiotic pathway’ (CSP) [2]. Therefore, we looked for the presence of ‘symbiosis-specific’ genes and ‘conserved’ genes in the transcriptome of Lupinus albus, in the draft genome of Lupinus angustifolius [34], in the transcriptome of Arachis hypogea [35], in the genome and transcriptome of Medicago truncatula (Medicago [36]), and in the genome of four other legumes. We also included poplar as an outgroup (Populus trichocarpa, Salicaceae [37]). ‘Conserved’ genes and CSP genes were present in all these datasets (Figure 4). In contrast, AM-specific genes were not detected in the Lupinus albus transcriptome and were absent from the Lupinus angustifolius genome (Figure 4). According to our probabilistic analysis, at least two of these five genes should have been detected in the transcriptomes of Lupinus albus if present (Figure S3 and Table S4). To confirm their absence experimentally, we used a PCR approach on one of them, RAM2. Medicago ram2 mutants are defective in AM symbiosis, but not in the rhizobium–legume symbiosis [38]. In addition, RAM2 is very well conserved at the DNA sequence level across legumes, making it a good candidate for this approach. We experimentally tested fifteen species within the Papilionoidae legume subfamily, including three Lupinus species, three species closely related to the Lupinus genus (Laburnum alpinum, a Cytisus sp., and Genista tinctoria), and a Prosopis sp. which belongs to subfamily Mimosoideae [39]. We were able to amplify RAM2 from the genomic DNA of all the tested legumes except the three Lupinus species (Figure 4B, Table S5). As a control, we amplified the ‘conserved’ gene DMI1 in all the legumes tested including the three Lupinus species (Figure 4B, Table S5). Therefore, Lupinus seems to have lost genes required for AM symbiosis, but retained those also required to associate with rhizobia. Taken together, our results show that the loss of known symbiotic genes occurred in a convergent manner in at least five non-symbiotic lineages, at the order, family, and genus levels.
Based on the strong correlation observed between the loss of AM symbiosis and the loss of ‘symbiosis-specific’ genes, we hypothesized that, in addition to the small set of genes identified so far through genetics in legumes, other genes could have been lost in non-host lineages and thus could be identified through a comparative phylogenomic approach. To test this hypothesis, we reconstructed the evolutionary history of 33 fully sequenced plant genomes using BigPlant, a phylogenomic pipeline originally developed to analyze genomes and transcriptomes of seed plants [40]. Using this phylogenomic framework to analyze the genomes of 33 fully sequenced species (see Methods), we identified a set of 395 ortholog groups, corresponding to 305 and 409 genes in Medicago and rice (Oryza sativa), respectively (Table 1, Tables S6 and Figure S4), that are highly conserved across land plants, but missing in the genomes of the five Brassicaceae sequenced to date (Table S6). To test the biological relevance of this list, we used the list of annotated Medicago genes (because this model has been used extensively to study symbiotic associations) and estimated its enrichment in symbiosis-related genes (i.e. ‘symbiosis-specific’ genes and genes known to be expressed during AM symbiosis according to a previous study [41]) compared to ten lists of 305 randomly selected genes from Medicago. We found that the list generated using our phylogenomic pipeline is strongly enriched in symbiosis-related genes compared to the random lists, as determined by χ2 test of independence (p-value<0.001, Table 1). To refine this analysis, and to remove genes possibly resulting from lineage-specific loss (i.e. Brassicales-specific), we then removed from the list genes present in other non-symbiotic taxa in a stepwise manner. Removing orthologs present in the sugar beet genome reduced the list down to 250 genes, and sequential refinement with the genome of Utricularia gibba (one gene) and the transcriptome of the parasitic plants Striga hermontica and Orobanche aegyptiana (75 genes) resulted in a list of 174 Medicago genes. The same approach with rice as reference resulted in a list refined of 167 genes (Table S10). Among these genes 65 are shared between Medicago and rice (Table S15, S16). The presence of non-overlapping genes between the lists can be explained by three main factors: non-completion of genome sequences, lineage-specific gene duplications, and divergence time between rice and Medicago. The refined Medicago gene lists systematically showed a very significant enrichment in symbiosis-related genes compared to randomly-generated lists (p-value<0.001, Table 1). Moreover, none of the symbiosis-related genes identified in the first list was removed after refinement (Table 1, Table S6, S7, S8, S9, S10). Thus a significant proportion of the genes identified using this approach is very likely involved in symbiotic processes. For instance, we found two members of the LysM-domain containing receptor-like kinase family, which could be part of the so-far uncharacterized Myc-factor receptor complex. At later stages the secretion machinery is reoriented to shape the symbiotic interface required for nutrient exchange [42]. At least five proteins associated with cellular trafficking have been identified through this phylogenetic analysis and are potentially playing a role in this process.
A subset of already characterized symbiotic-genes, called CSP genes, is involved in both AM and root nodule symbioses. Part of the newly identified genes could also be CSP genes. To identify such genes, we compared the refined list and the Lupinus albus transcriptome. Given that Lupinus retains CSP genes but has lost genes specifically required for AM symbiosis, genes absent in Lupinus (Medicago Table S11, rice Table S12, overlapping Table S15) are strong candidates for ‘AM-symbiosis’ genes. By contrast, genes still present in Lupinus (Medicago Table S13, rice Table S14, overlapping Table S16) are potential CSP genes. Most of the already-characterized CSP genes are present in this list and the missing ones were not identified in the pipeline because of their absence in the used Medicago or rice gene models (CASTOR and VAPYRIN). Among the other genes identified as potential common symbiosis genes, we found, for instance, MtCbf3, which has been recently found strongly up-regulated in response to Nod factors [36]. Another interesting candidate is MtDXS2 that is known to play a role during AM symbiosis [43]. Conservation of MtDXS2 in Lupinus albus suggests its potential involvement during root nodule symbiosis too. Alternatively these genes might be the only relict of AM-specific genes in Lupinus.
Interestingly, the expression pattern of many genes that came out of the comparative phylogenomic approach, including the already characterized ‘symbiosis-specific’ genes, is not affected during symbiosis and thus these candidates could not be detected by conventional transcriptomic or proteomic approaches. Further reverse genetic and biochemical studies will be necessary to determine the role played by these putative new components in symbiotic plant–microbe associations.
The AM symbiosis and the symbiotic toolkit required for its establishment are highly conserved among land plants [10]. Previous studies have found that some of these genes are missing in the non-host model plant Arabidopsis [11], [44], [45]. We discovered that many of these genes are also missing in the genome of seven other phylogenetically divergent non-host species. However, two biases could explain why we did not find these genes in non-host plants. First, genome sequences are never absolutely complete, so we cannot rule out the possibility that symbiosis-specific genes might be present in not yet sequenced regions of non-host genomes. However, the sequencing completion of host and non-host genomes is comparable (Table S17A) making this hypothesis very unlikely. Secondly, neo- or sub-functionalization acting on ‘symbiosis-specific’ genes in non-host plants might have affected our ability to detect them using homology-based searches. For instance, NSP1 a ‘conserved’ gene is under less constrained selection in non-hosts compared to hosts [46]. However, using comparative whole genome synteny analyses, we found that ‘symbiosis-specific’ genes are well anchored in conserved syntenic blocks in host species whereas they are absent in corresponding blocks in non-host species (Table S1). In addition to genomic data, we took advantage of transcriptomic data available for non-host species and their closely related host species. The ability to detect a gene in a transcriptome dataset is dependent of two main factors: sampled tissues and transcriptome depth. Both host and non-host transcriptomes have been generated from various tissues (Table S17B) and the average transcriptome depths are comparable (Figure S5). Moreover, some ‘symbiosis-specific genes that are almost exclusively expressed in plant cells colonized by the AM fungi, such as PT4 (Javot et al. 2007), have been detected in several host species with deep transcriptomes data (i.e. Sesamum and Capsicum) whereas we did not detect them in the transcriptome of non-host species with similar or even deeper coverage (Table S17B). Our analysis integrating genomic and transcriptomic data strongly supports that the loss of AM symbiosis repeatedly lead to the loss of an entire set of genes required for this symbiosis.
This finding supports the unifying hypothesis that extant non-host lineages cannot interact with AM fungi because they lack key genes required for this association. However, the mechanisms leading to the transition from host to non-host status are still unclear. Emergence of a new trait allowing efficient nutrient uptake has been proposed to decrease selection pressure for symbiotic nutrient acquisition leading to the loss of AM symbiosis [19]. In support of this hypothesis, Lupinus albus adapts its root system very efficiently under nutrient-limiting conditions by forming highly branched cluster-roots and releasing organic acids into the soil in order to solubilize phosphorus [47]. However, such mechanisms are absent in early diverging, non-host Lupinus species [47]. Thus loss of AM symbiosis in this genus likely predated the appearance of cluster roots and represents a compensatory adaptation. In addition, some species with an alternative nutrient-uptake mechanism are still able to form an efficient symbiosis with AM fungi. For instance, the carnivorous plants Drosera [48] and a Nepenthes sp. (N. Séjalon-Delmas, personal communication), the facultative hemi-parasite Pedicularis sp. [49], and the cluster-root forming species Casuarina glauca [50] can still associate very well with AM fungi. Our results support the reverse hypothesis: the loss of gene(s) from the symbiotic toolkit was the primary cause for the loss of AM symbiosis, and was followed by the emergence of alternative nutrient uptake strategies. Under such a hypothesis, a strong selection pressure against one or more genes from the symbiotic toolkit would be required. Interestingly, mutations in RAM2 in Medicago confer resistance to the broad host-range pathogen Phytophtora palmivora [38]. Thus, at least in legumes, loss of this gene could come under purifying selection, leading to the loss of AM symbiosis followed by the loss of other genes from the symbiotic toolkit. It has been hypothesized that besides RAM2 other symbiotic mechanisms might have been hijacked by pathogens [38], [51], [52]. Thus under pathogenic pressure loss of a single symbiotic gene could have been selected for, followed by the loss of others, and eventually, through a highly reproducible domino effect, to the loss of all the other ‘symbiosis-specific’ genes.
Such correlated loss of a trait and the associated genes is not unique to symbiosis [13]. With the increasing number of genome and transcriptome sequences available, tracking convergent gene losses by comparative phylogenomic frameworks such as BigPlant opens the way to discover new gene networks and pathways toward a better understanding of plant biodiversity, development and evolution.
The specific and convergent gene loss in five independent non-host lineages that we have demonstrated also supports and is consistent with the hypothesis that AM fungi maintain purifying selection on host gene networks during the evolution of entire lineages. This phenomenon is likely to be conserved in other symbiotic associations. For instance, the mammalian gut microbiome is significantly influenced by the phylogenic position of the host, with omnivorous primates sharing a large proportion of their microbiome [53]. Because of its critical role, natural loss of the entire microbiome is very unlikely. The development of gnotobiotic organisms has already demonstrated the importance of the associated microbiome in many processes [54]–[56]. Experimental evolution experiments where different microbial symbionts or microbiome assemblies would be associated to specific host lineages could be the next step towards confirming the impact of associated microbiota on host genomes.
See Table S18.
For each species, ten to forty individuals were tested, except for Aethionema arabicum were eight plants were used. Germinated seedlings were transferred to pots filled with metro-mix and incubated for two weeks (24°C, 16 h light/8 h day). Then plants were transplanted to pots containing Turface (Moltan Company or Profile). Each pot was inoculated either with Mighty Myco Soluble, a commercial mix of eight AM fungal species (Glomus aggregatum, Glomus brazillanum, Globus clarum, Glomus deserticola, Glomus intraradices, Globus monosporum, Glomus mosseae, and Gigaspora margarita), with 400 spores of Rhizophagus irregularis, or suspended in water. For each experiment Zea mays B73 and Medicago truncatula Jemalong A17 were used as positive controls. Plants were watered three times per week with a Long-Ashton solution with low phosphate concentration [57] and with water as needed. After 8 weeks plants were harvested, stained as previously described [57], and fungal colonization monitored by microscope.
Protein sequences of Medicago truncatula symbiotic genes (NFP, DMI2, DMI1, CASTOR, NUP85, NUP133, NENA, DMI3, IPD3, NSP1, NSP2, RAM1, RAM2, VAPYRIN, CCD7, CCD8, MAX2, STR, STR2, and PT4, Table S5) were used as queries for BLASTp or tBLASTn searches manually performed on GenBank (http://blast.ncbi.nlm.nih.gov/Blast.cgi), Phytozome (http://www.phytozome.net/), or species-specific databases, as indicated in Table S5. For all the genes in each species, the best hits, based on E-values, were selected as well as the ones displaying the highest identity (if coverage >20%).
To amplify RAM2 and DMI1 from legumes, genomic DNA was extracted from the leaves of at least two different plants per species using the GenCatch Plant Genomic DNA Purification Kit (Epoch Life Science). DMI1 was amplified using primers described previously [58] and RAM2 was amplified using primer RAM2-Fwd: 5′-CTCCCAAAACCCATCGTCTTCCA and RAM2-Rev: 5′-GGACTAGGGTTCATGAAGAAGTA. PCR products were gel purified using the QIAquick Gel Extraction Kit (Qiagen) and sequenced at the UW–Madison DNA sequencing facility (http://www.biotech.wisc.edu/facilities/dnaseq/home). All the candidates obtained either by PCR and sequencing or by BLAST searches were then tested by reciprocal BLAST analysis on the Medicago truncatula genome (http://blast.jcvi.org/er-blast/index.cgi?project=mtbe). For genes belonging to large gene families (DMI2, STR, STR2, PT4, RAM2) or with closely related homologs (CASTOR), a phylogenetic approach was also performed to confirm the absence or presence. For this purpose, each candidate gene was aligned with the targeted gene in Medicago truncatula, Populus trichocarpa, and Oryza sativa and the closest homologs of this gene in these species. Alignments were performed using MAFFT and manually edited with BioEdit. Gaps were systematically removed. Phylogenetic trees were constructed with MEGA5 [59] by Maximum-Likelihood with 500 bootstraps. Accession numbers of sequences used or generated in this study are indicated in Supplementary Table S5.
The symbiosis-specific and core set of conserved genes were screened for their presence across the Arabidopsis thaliana (At), Brassica rapa (Br), Aethionema arabicum (Aa), Tarenaya hassleriana (Th), Carica papaya (Cp), Prunus persica (Pp), Populus trichocarpa (Pt), and Vitis vinifera (Vv) genomes using comparative genomic analyses (http://www.genomevolution.org/CoGe/, Table S1, [60]). The supplemental file includes hyperlinks to regenerate all species comparisons, showing all the parameters utilized for synteny analysis. Due to multiple lineage-specific, ancient, whole-genome duplication events at this phylogenetic scale, this file represents only the analysis of the most syntenic region between these species. However, the entire genome was analyzed across all species (i.e. comparison of all homoeologous genomic regions). Due to the age of these duplications, the majority of the duplicated regions have returned back to a single copy state. Following the most recent event, which occurred over 30 MYA, only ∼21% of all genes are still retained in duplicate by the entire Brassicaceae family. These have been shown to encode a very specific set of highly dosage sensitive set of genes (e.g. transcription factors and highly connected signaling molecules). Nonetheless, since the symbiosis specific genes are absent in the sister family Cleomaceae which does not share the most recent whole genome duplication, the most parsimonious explanation is that the gene was lost prior to the duplication (consistent with Figures 1 & 2). For example, there are up to twelve homoeologous regions in Brassica rapa to each syntenic region in Vitis vinifera. We screened all Br∶Vv regions, and are reporting the results for the most syntenic with the target gene (if present in the genome). We also report genome-wide significant BLAST results for the target gene, which are consistent with our syntenic analyses (Rows 5 and 13). The syntenic analyses for symbiosis-specific genes were split into two separate analyses: A) the first showing the presence across outgroups Pt, Pp, Vv, and Cp (Row 4) and B) the second showing absence across At, Aa, Br, and Th (but presence of various flanking genes)(Row 6). The syntenic analyses for core conserved genes show largely the presence across all species (Row 12), both in the Brassicaceae and outgroup species.
Detail about the analysis and corresponding figures are provided in Text S1.
To determine the probability for a gene to be detected in the transcriptome of a given species if the gene is present, we used a logistic model. This approach used the detection/non-detection data in situations when gene presence is strongly supported, that is, for conserved genes in host and non-host species, and for ‘symbiotic specific’ genes in AM-hosts. We estimated the probability of detection based on two factors: a gene-specific effect αi for gene i (as explained by its expression level) and a species-specific effect βj for species j (as explained by its transcriptome coverage). With our logistic model, the probability of detecting gene i in species j is given by:(1)In other words, αi+βj is the log of the odds of detecting the gene's presence. Model parameters (α's and β's) were estimated with maximum likelihood using function ‘glm’ in R [61]. Intuitively, the transcriptome coverage effect of a given species reflects the percentage of conserved genes detected in the transcriptome, and the expression level effect of a gene reflects the ability to detect this gene in species where it is supposed to be present. For instance, for the Brassicales, PT4 was not detected in either Moringa or Akania which are AM hosts. Thus, it was impossible to reject the presence of PT4 in the other Brassicales transcriptomes (with the notations above, pij = 0 for gene I = PT4). After determination of model parameters, a prediction was performed using equation (1) again through the function ‘predict’ in R, but for the symbiotic genes in the non-host species (see Text S2). Next, for each gene we calculated the probability to be detected in at least one non-host species if present in all of them. For gene i, this is one minus the product of (1−pij) values over all non-host species j: 1−Πnon-host species j (1−pij). Transcriptomes of Fabaceae and Lamiales were combined because of the limited number of transcriptomes available. In order to experimentally validate prediction analysis, we used the genome and transcriptomes of Amborella trichopoda. Amborella is an early diverging lineage among angiosperms [62]. Because of this, the determination of prediction parameters can be performed using the transcriptomes of all the host and non-host studied species. All the symbiotic genes were found in the Amborella genome (Figure S6). Most of them were also found in the transcriptome data with the exception of two ‘conserved’ genes (NSP1 and NENA) and four ‘symbiosis-specific’ genes (NFP, STR, RAM2 and PT4). These genes are present but not detected. We then determined the probability for each ‘symbiosis-specific’ gene to be detected if absent using the GLM. As shown Figure S6, only the absence of NFP is supported whereas absence of STR, RAM2, and PT4 is not predicted. Based on this experiment we can estimate the false discovery rate of the GLM at 25%.
The BigPlant pipeline [37], which was previously built to incorporate complete and partial genomes in a single phylogenetic analysis, was used for the phylogenomic analysis. BigPlant simultaneously reconstructs the evolutionary history of the species included and the sets of genes supporting this history [37]. The initial stages of this BigPlant pipeline performs an all-to-all BLAST comparison followed by an OrthoMCL clustering, to group genes into gene families that span across species. For the current application, a BigPlant phylogenomic pipeline analysis was initiated using 31 fully-sequenced Angiosperm genomes and two outgroups (Table S19). A gene family tree is then constructed for each gene family. We determined sets of orthologs from these gene family trees by extracting the largest non-overlapping subset of genes that are orthologous according to the tree topology. This partitioning of the gene families generates ortholog groups (OGs) that contain zero to one representative gene per species. These OGs were then analyzed to identify those entirely absent from Brassicaceae. A confounding factor for this analysis is that any given gene family has members missing in one or more species, owing to the incompleteness of genome assemblies, gene models, etc. The set of genes missing in Brassicaceae includes many such families. Therefore, to increase the likelihood of identifying genes truly missing in Brassicaceae a global distribution of “apparent” gene loss was computed for any gene missing in a random set of 5 species but present in n other species. This distribution was used as the background rate of gene loss (Table S20). Based on this distribution, the size of the set of genes missing in Brassicaceae but present in 13 or more species lies outside two standard deviations from the mean. This threshold was chosen to identify genes as missing in Brassicaceae with a chance greater than random. An additional requirement was to find the members of this OG in at least one of the monocots included in this analysis since they exhibit AM symbiosis despite the large evolutionary distance. Further filters of presence/absence (using BLAST E-Value cutoff 1E-10) in the relevant transcriptomes, from other non-host species, were applied to generate the putative symbiosis-related gene list (Figure S5). Medicago was used as the reference AM-host genome because of its importance as model plant to study beneficial plant–microbe associations. A parallel analysis using Rice as the reference AM-host genome identified a very similar set of 138 genes. There is a 48% overlap between the gene set identified using Medicago as reference and the set using Rice as reference. Ortholog identification is more reliable in Medicago since it is phylogenetically closer to the other non-host species and hence we use the gene set derived from Medicago to draw the list of putative AM symbiosis genes.
To determine the enrichment in symbiosis-related genes of generated lists, each accession number of the list was searched against a list composed by the genes up-regulated in arbuscules [41] and the ‘symbiosis-specific’ genes included in the current Medicago truncatula gene model (Table S5). To test for the significance of this enrichment, lists of random genes containing 305, 250, 249, 174 or 110 Medicago truncatula genes were also compared to the symbiosis-related genes. A χ2 test was then performed to determine if the number of symbiosis-related genes present in the generated lists was significantly higher than in each of the randomly generated lists.
|
10.1371/journal.pgen.1007383 | Analysis of motor dysfunction in Down Syndrome reveals motor neuron degeneration | Down Syndrome (DS) is caused by trisomy of chromosome 21 (Hsa21) and results in a spectrum of phenotypes including learning and memory deficits, and motor dysfunction. It has been hypothesized that an additional copy of a few Hsa21 dosage-sensitive genes causes these phenotypes, but this has been challenged by observations that aneuploidy can cause phenotypes by the mass action of large numbers of genes, with undetectable contributions from individual sequences. The motor abnormalities in DS are relatively understudied—the identity of causative dosage-sensitive genes and the mechanism underpinning the phenotypes are unknown. Using a panel of mouse strains with duplications of regions of mouse chromosomes orthologous to Hsa21 we show that increased dosage of small numbers of genes causes locomotor dysfunction and, moreover, that the Dyrk1a gene is required in three copies to cause the phenotype. Furthermore, we show for the first time a new DS phenotype: loss of motor neurons both in mouse models and, importantly, in humans with DS, that may contribute to locomotor dysfunction.
| Down Syndrome is caused by an extra copy of chromosome 21 and results in many different phenotypes including learning difficulties, Alzheimer’s disease and problems with motor function such as abnormal gait and poor fine motor skills. These different phenotypes are thought to result from an increased copy of one more of the genes on chromosome 21, but it is not known which gene or genes cause the phenotypes. Using a panel of mouse strains with an extra copy of different sets of mouse genes that are equivalent to the human genes on chromosome 21 we were able to show that an extra copy of a small number of genes was sufficient to cause motor abnormalities in the mice, and that one of these genes, Dyrk1a, was required in three copies for this phenotype. Furthermore, we found that these mouse models of Down Syndrome showed loss of motor neurons, which could account in part for the motor dysfunction. This result led us to look in spinal cords from humans with DS, where we also found decreased numbers of motor neurons, a phenotype that has never been previously reported.
| Down Syndrome (DS), trisomy 21, is characterized by a wide range of phenotypes including cognitive deficits, early-onset Alzheimer’s disease, locomotor dysfunction and congenital heart defects [1, 2]. These diverse phenotypes could be caused by small numbers of dosage-sensitive genes on Hsa21 whose increased copy number leads to increased expression and hence phenotypic effects [3]. Alternatively, phenotypes may result from increased dosage and thus expression of large numbers of sequences on Hsa21. Such a possibility was recently reported for aneuploidy in yeast, where the deleterious effects of an extra chromosome on cell proliferation were shown to be due to the combined action of large numbers of genes rather than small numbers of dosage-sensitive genes, potentially resulting in proteotoxic stress [4, 5]. Finally, it is possible that aneuploidy per se, rather than additional copies of genes may cause phenotypes; and any combination of these phenomena may be important for individual phenotypes.
To understand the genetic basis of DS, we and others have used mouse genetics to model the syndrome by generating a series of strains containing either Hsa21 or duplications of mouse chromosome regions orthologous to Hsa21 located on mouse chromosome 16 (Mmu16), Mmu17 and Mmu10 (Fig 1) [1, 6–13]. Some of these strains are also aneuploid–the Tc1 mouse strain carries a freely segregating Hsa21 and has increased dosage of genes on Hsa21 [9]. While Tc1 mice have a number of phenotypes that model aspects of DS, the mice are mosaic for Hsa21 and this chromosome is rearranged such that only ~200 Hsa21 protein-coding genes are present (~75% of all Hsa21 genes) [14]. The Ts65Dn strain is also aneuploid, containing an extra chromosome carrying a portion of Mmu16 orthologous to Hsa21 and has been used extensively for the study of DS phenotypes [7]. However, this additional chromosome also includes ~10 Mb of Mmu17 containing 60 mouse genes that are not orthologous to Hsa21, thus limiting the utility of this model [15]. Chromosome engineering has allowed the generation of strains with carefully designed gene dosage increases. In particular, three strains known as Dp(16)1Yey, Dp(17)1Yey and Dp(10)1Yey, have been generated carrying tandem duplications of the entire Hsa21-orthologous regions of Mmu16, Mmu17 and Mmu10 respectively (Fig 1) [10, 11]. The intercross of these three mutations (triple trisomic mouse) gives the most complete mouse model of DS to date because it carries an extra copy of every mouse gene orthologous to Hsa21. We have also generated the Dp1Tyb strain with a tandem duplication of the entire region of Mmu16 orthologous to Hsa21, the largest of the three syntenic regions [16]. Furthermore, to allow an unbiased mapping of dosage-sensitive genes that may cause DS phenotypes we constructed another 6 strains (Dp2Tyb, Dp3Tyb, Dp4Tyb, Dp5Tyb, Dp6Tyb and Dp9Tyb) that have duplications of shorter regions of Mmu16 nested within the region duplicated in Dp1Tyb (Fig 1) [16]. Thus, between them, these strains can be used to establish if phenotypes are caused by small or large numbers of genes, or they if they require aneuploidy.
People with DS have deficits in motor function, showing alterations in balance, postural control and fine motor skills [17–22]. This understudied DS phenotype has been suggested to arise in part because of defects in cerebellar anatomy [23, 24]. The Ts65Dn mouse model has reduced cerebellar size and decreased numbers of granule cells, and the same defects were found in cerebella of humans with DS [24]. The defect in this model is likely caused by decreased Sonic Hedgehog-induced proliferation of granule cell precursors [25]. However, we do not know the identity of any dosage-sensitive genes whose increased copy number is required for motor defects, or indeed if these defects are caused by small numbers of dosage-sensitive genes, or by the mass action of increased dosage of large numbers of genes on Hsa21, or by aneuploidy.
In this study, we aimed to identify dosage-sensitive genes that cause motor defects and to examine possible pathological changes that might underpin them. In particular, we investigated changes in cerebellar anatomy as well as sensory and motor neuron function. Using our genetic mapping panel of mouse strains, we show that motor dysfunction can be caused by increased dosage of a region with a small number of genes, and within these we demonstrate that the Dyrk1a gene is required in three copies to cause the phenotype. Furthermore, we show that, surprisingly, there is no alteration in cerebellar anatomy in mice that have increased dosage only of genes orthologous to Hsa21. However, we identified an entirely novel form of neurodegeneration in DS, the progressive loss of motor neurons, a phenotype that, importantly, is recapitulated in human samples with DS and may contribute to the locomotor dysfunction. Our results support the hypothesis that some DS phenotypes are caused by increased copy number of small numbers of dosage-sensitive genes, and broaden the neurodegenerative phenotypes in DS.
Previously, we showed that Tc1 mice have defects in locomotor function using both a static rod and a rotating rod (Rotarod) test [26]. To identify whether locomotor defects can be modeled by increased dosage of mouse genes orthologous to Hsa21, we examined locomotor function in Dp(16)1Yey, Dp(17)1Yey and Dp(10)1Yey mice that between them carry duplications of all three regions of mouse chromosomes orthologous to Hsa21 (Fig 1). We chose to use a Rotarod paradigm in which mice are placed onto an accelerating rod, recording the speed of the rod at which the mouse falls. Each mouse was tested 3 times during one day, and then a further 3 times on the second and third day of testing–a protocol in which control mice usually show improved performance over the 3 days, demonstrating motor learning. We found Dp(16)1Yey mice performed significantly less well than wild-type littermates, whereas Dp(17)1Yey mice showed no defects, and Dp(10)1Yey mice had improved performance, demonstrating that duplication of the orthologous region on Mmu16 was sufficient to cause locomotor defects (Fig 2A). To evaluate whether the orthologous regions on Mmu17 and Mmu10 may contribute to the phenotype when combined with the duplication on Mmu16, we intercrossed the three mutant strains and analyzed the progeny. We found that the Dp(17)1Yey and Dp(10)1Yey mutations alone or together did not exacerbate the phenotype of the Dp(16)1Yey mice, with the triple trisomic mice performing no worse than Dp(16)1Yey (Fig 2A). Thus, the region of Hsa21 orthology on Mmu16 is both required and sufficient to cause locomotor defects.
It is possible that the mice performed poorly in the Rotarod test because of reduced motivation rather than locomotor dysfunction. To address this, we tested Dp1Tyb mice (which bear a duplication of the same Mmu16 genes as Dp(16)1Yey mice) using the Locotronic apparatus (Intellibio). In this assay, mice traverse a horizontal ladder with evenly spaced rungs, and the number of errors in foot placement (missed rungs) was recorded. Mice were tested 2–3 times, and trials where mice took > 60 s to traverse the ladder were excluded from the analysis in order to eliminate trials where the mice were insufficiently motivated. Dp1Tyb mice showed a significant increase in errors, supporting the conclusion that duplication of the Hsa21-orthologous region of Mmu16 results in locomotor defects (Fig 2B).
To narrow down the location of potential dosage-sensitive genes causing this defect, we examined Dp2Tyb, Dp3Tyb and Dp9Tyb mice which contain duplications that between them cover the entire region duplicated in Dp(16)1Yey (Fig 1). We found that only Dp3Tyb mice had a significant defect in the Rotarod assay, though we noted that the extent of the defect was smaller than that seen in Dp(16)1Yey (Fig 2A). A similar defect was also seen in Ts1Rhr mice that contain a duplication of 33 genes that is entirely contained within the Dp3Tyb region, but is smaller by 8 genes [27]. Lastly, we examined Dp4Tyb, Dp5Tyb and Dp6Tyb mice that break down the region duplicated in Dp3Tyb into three smaller regions. We found both Dp4Tyb and Dp5Tyb showed defects in the Rotarod assay. Thus, the regions duplicated in Dp4Tyb and Dp5Tyb, spanning a total of 3.3Mb and containing 15 and 12 genes respectively are each sufficient to cause some locomotor dysfunction, although genes in other areas also contribute to the full phenotype.
DYRK1A is protein kinase encoded on Hsa21, whose overexpression has been implicated in neuronal phenotypes in DS, including brain development and synaptic plasticity [28]. Transgenic mice overexpressing DYRK1A have motor defects [29–32], and since Dyrk1a is located within the region duplicated in Dp(16)1Yey, Dp3Tyb, Ts1Rhr and Dp5Tyb mice, we tested whether three copies of Dyrk1a are required for the locomotor defects. We crossed Dp(16)1Yey and Ts1Rhr mice to mice heterozygous for a null allele of Dyrk1a (Dyrk1a+/-). The phenotype was rescued in both the Dp(16)1Yey/Dyrk1a+/- and Ts1Rhr/Dyrk1a+/- progeny (2 copies of Dyrk1a), which showed no defect in the Rotarod assay, thus demonstrating that three copies of the Dyrk1a gene are required for the locomotor deficit (Fig 2A).
If an increased gene dosage of Dyrk1a is required for the locomotor defects, there should be increased Dyrk1a mRNA expression in mice with a duplication that includes this gene. We found significantly increased Dyrk1a mRNA in the cerebellum of Dp(16)1Yey mice at both 6 days and 10 weeks of age and in 10 week old Dp3Tyb mice (Fig 3A–3C). We also saw a trend for increased Dyrk1a expression in Dp5Tyb mice (Fig 3D). Interestingly, the upregulation of Dyrk1a was larger in young Dp(16)1Yey mice at 6 days of age (1.64-fold) compared to adult 10 week old mice (1.25-fold). This is similar to a previous report on cerebellar DYRK1A levels in Ts1Cje mice that have an additional copy of 87 Mmu16 genes including Dyrk1a [33]. This study shows that the increase in cerebellar DYRK1A was larger in young compared to old Ts1Cje mice.
It has been previously described that increased Dyrk1a expression in transgenic mBACtgDyrk1a mice leads to increased levels of Gad1 and Gad2 in several brain regions including the cerebellum [34]. GAD1 and GAD2 are isoforms of glutamate decarboxylase, an enzyme that synthesizes the gamma-aminobutyric acid (GABA) neurotransmitter, and are expressed in inhibitory interneurons. The increased levels of GAD1 and GAD2 in mice overexpressing Dyrk1a have been postulated to indicate elevated numbers of inhibitory neurons, a phenotype that could contribute directly to behavioral and cognitive changes [34]. Analysis of cerebellar mRNA in Dp(16)1Yey mice at 6 days and 10 weeks of age and in 10-week old Dp3Tyb and Dp5Tyb mice showed no significant change in the expression of either Gad1 or Gad2 (Fig 3A–3D). This result shows that the upregulation of Dyrk1a in these strains (1.25- to 1.64-fold) is insufficient to perturb expression of Gad1 or Gad2, and suggests that numbers of interneurons may not be altered in the cerebellum of these mice.
Since defects in cerebellar anatomy have been described in both humans with DS and in the aneuploid Ts65Dn and Tc1 mouse models [9, 24], and these have been proposed to contribute to motor defects in DS [23], we investigated the anatomy of the cerebellum in Dp(16)1Yey mice, which have motor defects but are not aneuploid. We analyzed cerebella at two ages, 6 days after birth (P6) and in adult mice at 6 months of age, since previous studies had documented changes at these ages in Ts65Dn mice [24, 25]. At P6 we found no significant changes in cerebellar area, width of the external granule cell layer, or in the density of Purkinje cells or granule cells in Dp(16)1Yey mice, either in individual lobules or averaged over the whole cerebellum (Fig 4A–4E). Similarly, in adult mice there was no change in the width of the granule cell and molecular layers or in the density of Purkinje cells, granule cells or interneurons, again over individual lobules or averaged over the whole structure (Fig 4F–4L). Thus, in contrast to Ts65Dn mice, Dp(16)1Yey mice have no obvious defect in cerebellar anatomy, and this cannot contribute to their locomotor dysfunction. These results do not rule out that there may be functional abnormalities in the cerebellum despite the absence of anatomical changes.
Given the changes in motor function we also screened for sensory deficits across a range of modalities, since these can result in locomotor defects [35, 36], and have not been examined in DS. Using assays measuring nocifensive responses to cold, heat, mechanical stimulation and formalin injection, we compared Tc1 mice with controls and found no differences between the groups (Fig 5A–5D). We also analyzed the dorsal root ganglia (DRG) by histology. Sensory neuron cell bodies located in the DRG can be classified based on their neurochemical characteristics. Large and medium diameter sensory neurons with myelinated axons were identified by expression of NF200. Peptidergic or non-peptidergic small diameter sensory neurons with unmyelinated axons (C-fibers) were identified by expression of calcitonin gene related peptide (CGRP) or by binding the isolectin Griffonia simplicifolia I-B4 (IB4) respectively [37]. We found no change in the fraction of neurons expressing NF200 or CGRP, but a significant decrease in the fraction of neurons binding IB4 in Tc1 mice (Fig 5E). We extended these studies to Dp1Tyb mice that, similar to Dp(16)1Yey mice, contain a duplication of the entire Hsa21 orthologous region on Mmu16 (Fig 1). Once again we found no changes in nocifensive responses to heat, mechanical stimulation and formalin injection, but the response to cold was partially impaired (Fig 5F–5I). Importantly, Dp1Tyb mice showed no defect in a beam walk test, a measure of proprioception (Fig 5J), but, interestingly, we again found a decrease in IB4-binding neurons in the DRGs of Dp1Tyb mice, similar to that seen in Tc1 mice (Fig 5K). These IB4+ neurons respond to high threshold noxious mechanical stimuli and thus their reduction is unlikely to contribute to the locomotor defects [38]. In summary, we found little evidence of a broad sensory defect in mouse models of DS that could account for the motor defects, but discovered a specific reduction in one class of nociceptive sensory neurons, IB4+ afferents, in Tc1 and Dp1Tyb mice, a phenotype that would merit further investigation in both mice and humans.
We undertook a series of studies to investigate if defects in muscle function or its innervation contribute to the locomotor defects. We previously showed that Tc1 mice have no defect in muscle strength as measured by a grip test [26]. To extend these studies we measured the maximum force produced by the tibialis anterior (TA), extensor digitorum longus (EDL) and soleus muscles of the mouse hindlimb in response to a tetanic stimulation of the sciatic nerve in live anaesthetized mice. We found no change in muscle force in Tc1 mice at either 4 or 19 months of age (Fig 6A–6C). However, analysis of the number of physiological motor units innervating the EDL muscle showed a significant 8% decrease in Tc1 mice at 4 months of age, rising to a 11% decrease at 19 months of age (Fig 6D and 6E). In agreement with this, analysis of motor neuron numbers in the sciatic motor pool showed an 18% and a 23% decrease in Tc1 mice at 6 and 19 months of age respectively (Fig 6F and 6G and S1 Table). Histology of the TA muscles confirmed this observation, showing changes characteristic of denervation and subsequent re-innervation, resulting in characteristic fiber type grouping of oxidative fiber types (Fig 6H). Interestingly, there was no decrease in motor neuron numbers in young Tc1 mice aged 22 days, indicating that there is no developmental deficit in the generation of these cells (Fig 6I). Thus, Tc1 mice show a progressive loss of motor neurons and motor unit function, which could contribute to the locomotor dysfunction.
To investigate the genetic basis of the motor neuron loss, we counted motor neurons in models with duplications of mouse regions orthologous to Hsa21. Analysis of Dp(16)1Yey, Dp(17)1Yey and Dp(10)1Yey mice at 6 months of age showed that only Dp(16)1Yey mice had decreased numbers of motor neurons (20% reduction), similar to that seen in Tc1 mice (Fig 6J and 6K and S1 Table). Furthermore, mice with duplications of all three regions (triple trisomic) showed a loss of motor neurons that was no greater than that seen in the single mutant Dp(16)1Yey mice (Fig 6J). Thus, duplication of the orthologous region of Mmu16 is both necessary and sufficient to cause a reduction of motor neurons, and the Hsa21 orthologous regions on Mmu10 and Mmu17 do not contribute to the phenotype. In contrast we found no change in motor neuron numbers in Dp(16)1Yey mice at P6 (Fig 6L), once again showing that the loss of motor neurons is progressive neurodegeneration and not a developmental deficit.
To map the location of potential dosage-sensitive genes that cause this neurodegeneration, we analyzed motor neuron numbers in Dp2Tyb, Dp3Tyb, and Dp9Tyb mice at 6 months of age (Fig 1). We saw no change in motor neuron numbers in any of these three strains (Fig 6M and S1 Table). Thus, the motor neuron loss is caused by an additional copy of at least 2 genes and these are located in 2 or more of the Mmu16 regions duplicated in Dp2Tyb, Dp3Tyb or Dp9Tyb.
The loss of motor neurons observed in Tc1 and Dp(16)1Yey mice led us to investigate spinal cord motor neuron numbers in humans with DS, since these have not been previously reported. Strikingly, we found decreased numbers of motor neurons in humans with DS compared to non-DS controls (Fig 7A and 7B and S2 Table). We compared our results to motor neuron counts in spinal cord sections of people with the motor neuron disease amyotrophic lateral sclerosis (ALS) and found the decrease in DS was less than in ALS. Thus, observation of reduced numbers of motor neurons in a mouse model of DS has led us to discover a novel phenotype in humans with DS.
Using a combination of mouse strains that contain Hsa21 genes or additional copies of their mouse orthologues that are also aneuploid (Tc1) or not (Dp strains) we were able to show that locomotor defects resulted from an additional copy of small numbers of genes, and that aneuploidy was not required for this phenotype. This supports the hypothesis that at least some DS phenotypes are due to increased dosage of small numbers of dosage-sensitive genes, rather than mass action of large numbers of additional genes or aneuploidy. This has important implications for driving forward future investigations into mitigating the effects of a few or single dosage-sensitive genes in DS. Nonetheless we noted that the locomotor defects became weaker as we reduced the size of the duplications, implying that in addition to the small regions containing dosage-sensitive genes that are sufficient on their own to cause phenotypes, there are additional genes outside these regions that also contribute.
The presence of defects in two different locomotor assays (Rotarod and Locotronic) supports our conclusion that the mutant mice performed poorly because they have locomotor defects rather than lacking motivation. In particular, in the Locotronic test we eliminated trials in which the mice took >60 s to traverse the ladder, thereby excluding trials where the mice were insufficiently motivated. However, it is possible that the poorer performance in these tests was due to other causes. For example Dp(16)1Yey have been shown to have disrupted sleep, which could impair performance [39].
The locomotor defect was evident in both Dp4Tyb and Dp5Tyb mice, showing that increased dosage of two separate regions is sufficient to cause this phenotype, and implies that at least two different dosage-sensitive genes contribute to this defect. It is likely that the more severe phenotype in Dp(16)1Yey mice is caused by additive effects of the Dp4Tyb and Dp5Tyb regions, together with genes outside these regions. Furthermore, since Ts1Rhr mice also have defects, the dosage-sensitive genes are most likely to be within the 25 genes that are duplicated in both Ts1Rhr and Dp4Tyb and Dp5Tyb (Fig 8). One of these is Dyrk1a and we have shown that three copies of this gene are required for the locomotor defect in Dp(16)1Yey mice, in agreement with previous studies showing that overexpression of DYRK1A leads to motor defects [29–32]. However, our results also show that the situation is complex, since Dp4Tyb mice show a phenotype whereas Dp(16)1Yey/Dyrk1a+/- do not, despite having an extra copy of all the genes that are also duplicated in Dp4Tyb. Thus, there must be genes outside the Dp4Tyb region that suppress the effects of increased dosage of gene(s) in Dp4Tyb, and suggests that DS phenotypes may result from the interplay of dosage-sensitive genes that both enhance and suppress phenotypes. DYRK1A is a protein kinase whose overexpression has been proposed to lead to defects in brain development, synaptic plasticity and in learning and memory [28], however, the mechanism by which this happens is not understood.
Recently published studies proposed that increased dosage of regions between Hspa13 and App on Mmu16 and Abcg1 and U2af1 on Mmu17 contribute to locomotor defects [13, 40]. These regions are duplicated in Dp9Tyb and Dp(17)1Yey mice respectively and we showed that in three copies they are not sufficient to cause locomotor defects, though we cannot rule out that they could contribute to the phenotype when combined with duplication of the region in Ts1Rhr. Indeed the stronger phenotype in Dp(16)1Yey mice compared to Ts1Rhr may be due to an increased dosage of genes in the Dp9Tyb region.
Previous studies had shown that both Ts65Dn and Tc1 mice have defects in cerebellar anatomy, but despite a more extensive analysis than was carried out in these strains, defects were not observed in Dp(16)1Yey mice. There are a number of differences between these models, including a different complement of duplicated genes, but we note that both Ts65Dn and Tc1 are aneuploid and this may possibly contribute to the phenotype. A recent study reported that Dp(16)1Yey mice have a reduced density of Purkinje cells and granule cells in the cerebellum [41]. It is unclear why our findings are different, but the two studies were carried out on different genetic backgrounds, which may have an effect.
Our results show that both Tc1 and Dp(16)1Yey mice have progressive degenerative loss of motor neurons in the spinal cord. This unexpected and novel phenotype led us to examine spinal cords from humans with DS; importantly, we found that humans also show reduced numbers of motor neurons. Since we only analyzed older adults, we are unable to distinguish if the loss is degenerative or a consequence of developmental abnormalities. To establish this would require analysis of spinal cords from young people with DS; to our knowledge such samples are not currently available. A previous study showed that people with DS have defective peripheral and central nervous system conduction parameters, consistent with axonal degeneration [42]. Interestingly, genetic mapping showed that the Hsa21 orthologous region of Mmu16 was both required and sufficient to cause the motor neuron loss, that the orthologous regions on Mmu10 and Mmu17 did not contribute, and that aneuploidy was not required. However, breaking up the Mmu16 region into three smaller regions caused the phenotype to disappear. Thus, the neuronal loss is caused by at least 2 genes and these must reside in 2 or more of these 3 regions. It is possible that this phenotype is caused by the mass action by an increased dosage of a large number of genes–there are 148 duplicated genes in Dp(16)1Yey–or by a small number of dosage-sensitive genes. Further mapping studies would be needed to distinguish these possibilities.
The extent of motor neuron loss (around 20%) is not large enough to explain the locomotor defects. Furthermore, genetic mapping showed that the locomotor defects and motor neuron loss do not map to the same region, again suggesting that the locomotor defects are not caused by motor neuron loss at the ages we tested. Nonetheless it is possible that the loss of these neurons may contribute to the defects in combination with other pathological mechanisms, and may account for why the locomotor phenotype of Dp(16)1Yey mice is stronger than that of Ts1Rhr mice.
In summary, we have been able to map the genomic location of causative genes underlying DS phenotypes. We have identified that three copies of the Dyrk1a gene are required for locomotor dysfunction, a relatively understudied phenotype, and using mouse models of DS we have discovered a decreased number of motor neurons, a novel phenotype which we have shown is also present in humans with DS.
Mice carrying the Dp(16Lipi-Zbtb21)1TybEmcf (Dp1Tyb), Dp(16Mis18a-Runx1)2TybEmcf (Dp2Tyb), Dp(16Mir802-Zbtb21)3TybEmcf (Dp3Tyb), Dp(16Mir802-Dscr3)4TybEmcf (Dp4Tyb), Dp(16Dyrk1a-B3galt5)5TybEmcf (Dp5Tyb), Dp(16Igsf5-Zbtb21)6TybEmcf (Dp6Tyb), Dp(16Lipi-Hunk)9TybEmcf (Dp9Tyb), Dp(16Lipi-Zbtb21)1Yey (Dp(16)1Yey), Dp(17Abcg1-Rrp1b)1Yey (Dp(17)1Yey), Dp(10Prmt2-Pdxk)1Yey (Dp(10)1Yey), Dp(16Cbr1-Fam3b)1Rhr (Ts1Rhr), Tc(HSA21)1TybEmcf (Tc1) and Dyrk1atm1Mla (Dyrk1a+/-) alleles have been described [9–11, 16, 27, 43]. Triple trisomic mice were generated by intercrossing Dp(16)1Yey, Dp(10)1Yey and Dp(17)1Yey mice to generate Dp(16)1Yey/Dp(10)1Yey/ Dp(17)1Yey mice with all three duplications. All strains were maintained by backcrossing to C57BL/6JNimr, except for mice bearing Tc1, Ts1Rhr and Dyrk1a+/- alleles, which were maintained by crossing to (C57BL/6J x 129S8)F1. All mice on the C57BL/6JNimr background that were used for experiments had been backcrossed for at least 5 generations. The intercross of Dp(16)1Yey and Dyrk1a+/- mice was backcrossed to C57BL/6JNimr for two generations. All mice were bred and maintained at the MRC National Institute for Medical Research (now The Francis Crick Institute), except for Dp1Tyb mice used in the Locotronic assay which were bred at the MRC Harwell Institute, and 10-week old Dp(16)1Yey mice whose cerebella were used for Q-RTPCR studies which were bred at Institut de Génétique et de Biologie Moléculaire et Cellulaire, Illkirch, France and provided by Veronique Brault. All experiments were carried out on males, using age-matched littermate controls, except for analysis of Dp(16)1Yey mice at P6 where both genders were used. No randomization was used, but in all cases analyses were carried out by experimenters who were blind to genotype.
The Rotarod test was used to evaluate locomotor function. 12-week old male mice were habituated to the RotaRod Advanced apparatus (Letica Scientific Instruments) the day before testing commenced by placing them on the rotating rod at a constant speed of 4 rpm. Motor performance was tested by placing the mouse onto the accelerating rod (4 to 40 rpm over 5 min) and then recording the speed of the rod at which the mouse fell off. The trial was repeated three times per day with an hour inter-trial interval. The average RPM of the three trials for each day for each animal was then calculated. Trials were then repeated on a second and third day to evaluate learning.
The Locotronic test (Intellibio, France) is a test of locomotor function. Mice traverse a horizontal ladder with evenly spaced rungs, along a narrow corridor to reach the exit. The number of rungs that the mouse stepped on and/or missed was recorded automatically, thereby determining how many rungs were missed. Each animal (n = 13 WT, 12 Dp1Tyb) was tested 2–3 times. Trials where mice took more than 60 s to traverse the ladder were excluded from the analysis; a total of 10 out of 69 trials were excluded, 5 from WT mice and 5 from Dp1Tyb mice, indicating no difference in motivation between genotypes.
Cerebella from 10-week or 6-day old mice were dissected and snap frozen. Each sample was homogenized in Buffer RLT (Qiagen) with 143mM 2-mercaptoethanol using a Pellet Pestle cordless motor (Kimble). Tissue lysates were loaded onto the QIAshredder (Qiagen) according to manufacturer’s instructions. Total RNA was isolated using RNeasy Mini Kit (Qiagen) and quantified using NanoDrop1000 (Thermo Scientific). RNA was reverse transcribed into cDNA using a MEGAscript T7 Kit (Invitrogen), and analysed by quantitative real-time PCR on a Quant Studio3 Real-Time PCR Machine (Thermo Fisher Scientific) using TaqMan gene expression assays (Thermo Fisher Scientific). Expression of test genes (Dyrk1a, Gad1 and Gad2) was normalised to the expression Gapdh and then to expression in the wild-type control samples.
P6 and 6-month old mice were sacrificed and brains were extracted and immersion fixed in 10% Formalin (VWR). After fixation, brains were embedded in paraffin and 5 μm sections were taken at the midline. Sections were stained with hematoxylin and eosin and images were acquired using an Olympus VS120 slide scanner. Measurements were performed in FIJI software (ImageJ) and, except where indicated, all enumeration was performed manually using the cell counter tool. Purkinje cell counts were performed along the whole length of the indicated lobule and density derived by dividing cell counts by the length of the Purkinje cell layer. Cerebellar granule cells were counted in identical locations at the tip of each lobule and density calculated by dividing cell counts by the area. Width measurements of the granule cell layer and molecular layer were all performed at identical locations within each lobule. Interneuron numbers were analyzed by drawing regions of interest around the molecular layer of lobule IX and dividing this region into three further regions of interest, then using the automated cell counter with the spot detector plugin in Icy (http://icy.bioimageanalysis.org/). Similarly the cerebellar granule cell numbers at P6 were counted using the automated cell counter in Icy. Tissue samples in this and all other histological analyses were excluded if quality or integrity of the sample was diminished or damaged.
Mice were sacrificed and the L3-L5 DRG were dissected and fixed for 90 min in 4% paraformaldehyde in 0.1 M phosphate buffer and mounted in OCT embedding compound. DRG blocks were sectioned on a cryostat at 10 μm thickness. Sections were washed in Phosphate Buffered Saline (PBS) and blocked using 10% normal goat serum (Vector Laboratories) for 30 min. Primary antibodies were incubated overnight and secondary antibodies were incubated for 2 h at room temperature in the dark. After each incubation step, the sections were washed three times with PBS. All reagents were diluted in PBS containing 0.2% Triton X-100 and 0.1% sodium azide. Slides were sealed with coverslips mounted with Vectashield medium. Immunofluorescence was visualized using a Zeiss Imager.ZI fluorescence microscope and pictures acquired with Axiovision software. Primary antibodies or lectins used: polyclonal rabbit anti-mouse CGRP (Biomolecular, CA1137, 1:500), biotinylated isolectin B4 (Sigma, L2140, 1:100), mouse anti-mouse NF200 (Millipore, MAB5266, 1:200), polyclonal rabbit anti-human PGP9.5 (Ultraclone, RA95101, 1:800). Secondary antibodies used: anti-rabbit-Cy3 (Stratech Scientific Ltd. 711-166-152, 1:500), extra-avidin FITC (Invitrogen, 1:400), anti-mouse FITC (Jackson Immunoresearch, 1:500). All antibodies listed here have been validated by their suppliers and references can be found on their website or on the online validation databases Antibodypedia and 1DegreeBio. Quantification was performed by manually counting the total number of DRG cell profiles identified by positive staining for PGP9.5 as well as the numbers of profiles positive for NF200, CGRP or IB4 in a representative region of interest in a DRG, from which a percentage was calculated. For each DRG at least three sections were quantified.
For the assessment of muscle function in vivo, mice were examined as previously described [48].
Following the physiological assessment of muscle function, the TA, EDL and soleus muscles were dissected and snap frozen in isopentane cooled in liquid nitrogen and stored at -80°C until processing. Frozen muscle samples were cut on a cryostat at 12 μm. Serial cross sections were collected on glass slides and stained for succinate dehydrogenase (SDH) activity to determine the oxidative capacity of the muscle fibres, as previously described[48].
In all mouse strains except Tc1 mice at 6 and 19 months, motor neurons were counted using a rapid dissection method. Animals were sacrificed and the L4 and L5 spinal segment was identified externally by distance from the most caudal rib and the iliac crest bone. The whole spinal column was extracted and placed in 10% Formalin before being decalcified by immersion in 5% formic acid (Immunocal, Decal Chemical Corporation, NY, USA) overnight, processed for paraffin embedding and sectioned at 5 μm thickness. Every eighth section was collected, thus giving a 35 μm interval between representative sections and 40 sections in total were collected per animal. Sections were stained with Cresyl Violet for Nissl bodies and images acquired with an Olympus VS120 Slide Scanner. All sections were assessed for correct anatomical location before counting. Motor neurons in the sciatic motor pool were located and counted based on the following criteria: a visible nucleolus, a soma rich in Nissl bodies, a diameter >15 μm and multipolar morphology with visible dendritic branching. Spinal cords of Tc1 mice at 6 and 19 months were removed, fixed in 4% paraformaldehyde and cryopreserved in 30% sucrose overnight. The lumbar L2-L6 region was sectioned on a cryostat and 20 μm transverse sections were collected serially onto glass slides and stained for Nissl bodies with gallocyanin. The number of Nissl-stained large motor neurons in every third L3-L6 lumbar section was established, using the same criteria as described above, with counted sections being at least 60 μm apart. When counting motor neurons at different levels of the spinal cord, the number of sections included in the counts was standardized. Thus, a total of 5 sections from L3, 10 sections from L4, 20 sections from L5 and 5 sections from L6 region were analyzed. 40 sections were counted for each animal.
Cervical spinal cord autopsy samples from people with Down syndrome (n = 3, 58–70 years old), amyotrophic lateral sclerosis (n = 3, 33–70 years old) and controls with no neurological disease (n = 4, 59–71 years old) were obtained from the Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam, Netherlands (8 samples), and the Thomas Willis Oxford Brain Collection, John Radcliffe Hospital, Oxford UK (3 samples). From the samples, 3 were paraffin-embedded tissues (2 controls and one DS), which were serially cut at 6–14 μm, whereas 8 specimens were frozen tissues which were cut at 20 μm serially. Sections were stained with Luxol Fast Blue, which stains the myelin sheaths blue and combined with Cresyl Violet staining that stains Nissl granules (RNA in rough endoplasmic reticulum) pink. For each sample a total of 20 sections each 60 μm apart, were assessed. The total number of large motor neurons with intense Nissl staining was counted in each section. Results are shown as number of motor neurons per hemisection for each disease group.
The work on human tissues was approved by a UK Research Ethics Committee, specifically the NRES Committee London—Queen Square (REC approval number 09/H0716/57). This ethics approval covers the procurement and use of linked-anonymised and anonymised human tissue and samples and data from banks, established collections and collaborators in the UK and internationally. All patient consent, consent from a legal representative, an opinion from a consultee or consent from an individual in a qualifying relationship to the deceased was obtained directly by the providing bank in line with their national legislation and institution policy. In order to maintain donor or consentee privacy, details of personal identifiable information about the patient/donor is held confidentially by the providing bank(s) and was not shared with the research team.
Mouse experiments were approved by the Animal Welfare and Ethical Review Panel (AWERP) of the Francis Crick Institute and were carried out under the authority of a Project Licence (PPL70/8843) granted by the UK Home Office under the regulations of the Animals (Scientific Procedures) Act 1986.
|
10.1371/journal.pntd.0001039 | Microarray-Based Analysis of Differential Gene Expression between
Infective and Noninfective Larvae of Strongyloides
stercoralis | Differences between noninfective first-stage (L1) and infective third-stage (L3i)
larvae of parasitic nematode Strongyloides stercoralis at the
molecular level are relatively uncharacterized. DNA microarrays were developed and
utilized for this purpose.
Oligonucleotide hybridization probes for the array were designed to bind 3,571
putative mRNA transcripts predicted by analysis of 11,335 expressed sequence tags
(ESTs) obtained as part of the Nematode EST project. RNA obtained from S.
stercoralis L3i and L1 was co-hybridized to each array after labeling
the individual samples with different fluorescent tags. Bioinformatic predictions
of gene function were developed using a novel cDNA Annotation System software. We
identified 935 differentially expressed genes (469 L3i-biased; 466 L1-biased)
having two-fold expression differences or greater and microarray signals with a p
value<0.01. Based on a functional analysis, L1 larvae have a larger number of
genes putatively involved in transcription (p = 0.004), and
L3i larvae have biased expression of putative heat shock proteins (such as
hsp-90). Genes with products known to be immunoreactive in
S. stercoralis-infected humans (such as SsIR
and NIE) had L3i biased expression. Abundantly expressed L3i
contigs of interest included S. stercoralis orthologs of
cytochrome oxidase ucr 2.1 and hsp-90, which may
be potential chemotherapeutic targets. The S. stercoralis
ortholog of fatty acid and retinol binding protein-1, successfully used in a
vaccine against Ancylostoma ceylanicum, was identified among the
25 most highly expressed L3i genes. The sperm-containing glycoprotein domain,
utilized in a vaccine against the nematode Cooperia punctata, was
exclusively found in L3i biased genes and may be a valuable S.
stercoralis target of interest.
A new DNA microarray tool for the examination of S. stercoralis
biology has been developed and provides new and valuable insights regarding
differences between infective and noninfective S. stercoralis
larvae. Potential therapeutic and vaccine targets were identified for further
study.
| Strongyloides stercoralis is a soil-transmitted helminth that
affects an estimated 30–100 million people worldwide. Chronically infected
persons who are exposed to corticosteroids can develop disseminated disease, which
carries a high mortality (87–100%) if untreated. Despite this, little is
known about the fundamental biology of this parasite, including the features that
enable infection. We developed the first DNA microarray for this parasite and used it
to compare infective third-stage larvae (L3i) with non-infective first stage larvae
(L1). Using this method, we identified 935 differentially expressed genes. Functional
characterization of these genes revealed L3i biased expression of heat shock proteins
and genes with products that have previously been shown to be immunoreactive in
infected humans. Genes putatively involved in transcription were found to have L1
biased expression. Potential chemotherapeutic and vaccine targets such as
far-1, ucr 2.1 and hsp-90 were
identified for further study.
| Strongyloides stercoralis is a parasitic nematode endemic to the
tropics and subtropics that infects an estimated 30–100 million people worldwide.
Chronically infected individuals have the potential to develop hyperinfection syndrome
or disseminated disease, clinical entities that carry a very high (87–100%)
mortality if unrecognized [1].
Free-living S. stercoralis infective third stage (L3i) larvae residing
in the soil penetrate intact skin and blood vessels, ultimately developing to adults in
the small intestine. Adult females, typically residing in the duodenum of the host,
produce eggs by mitotic parthenogenesis that develop into first-stage (L1) larvae that
are excreted into the stool. L1 larval progeny of parasitic females develop into
free-living adults unless triggered by genetic, environmental, or host factors to
develop directly into L3i larvae [2], [3]. Despite sharing many characteristics, L1 and L3i larvae can be
distinguished by their behavior and morphology. L1 larvae have a short, trilobed pharynx
and expend much of their energy on feeding and growth [3]. L3i larvae, by contrast, can survive
in harsh environmental conditions, enabled by a comparatively thickened cuticle,
constricted gastrointestinal tract, and closed mouth. These larvae are developmentally
arrested, non-feeding, stress resistant, and long lived [3]–[5].
A high degree of specificity between these stages has been suggested by expressed
sequence tag (EST) based analysis of free living L1 and L3i larvae for S.
stercoralis
[6]–[8]. These comparisons,
however, are based on short reads of cDNA libraries and assumptions about abundance.
There remain many unanswered questions about the basic molecular features underlying the
apparent morphologic and behavioral differences between these larval stages. An improved
understanding of these differences can provide insights into what defines infectivity
and may ultimately prove useful in defining targets for the development of vaccines and
therapeutics against this parasite.
In order to answer these questions, a DNA microarray tool for S.
stercoralis – the species causing the vast majority of human infection
worldwide - is needed. Although a DNA microarray has recently been developed for
Strongyloides ratti, the natural parasite of brown rats (Rattus
norvegicus) [9],
previous work has suggested little conservation of gene expression profiles between
these two species [10], underscoring the need for a DNA microarray specific to this
species.
The availability of a S. stercoralis DNA microarray enables comparative
analyses across nematodes, which can be utilized to further our understanding of the
biologic determinants of parasitism. The free-living, non-parasitic, nematode C.
elegans has been used as a model species for comparison with S.
stercoralis. C. elegans dauer stage larvae and S.
stercoralis L3i larvae share many morphologic and physiologic
characteristics. The ‘dauer hypothesis’ recognizes these similarities and
suggests that the same molecular genetic mechanisms control the morphogenesis of these
stages [11].
Comparative genomics of gene expression based on EST abundance data for S.
stercoralis suggests a higher degree of similarity between S.
stercoralis L1 and C. elegans non-dauer expressed genes
[6]. By contrast,
a robust ‘dauer-L3i expression signature’ has not been found [6]. A comparative
analysis based on microarray expression data for these species could prove useful not
only in identifying a ‘dauer-L3i expression signature’ should it exist, but
also in uncovering potentially significant determinants of S.
stercoralis L3i infectivity.
The purpose of this study was to: 1) develop and optimize a DNA microarray tool for
S. stercoralis, 2) utilize this microarray to examine differences in
gene expression between L3i and L1 larvae and 3) perform a comparative microarray
analysis between parasitic S. stercoralis and non-parasitic C.
elegans in order to develop further insights into the biologic determinants
of parasitism.
Animal handling and experimental procedures were undertaken in compliance with the
University of Pennsylvania's Institutional Animal Care and Use Committee (IACUC)
guidelines. Ethical approval was obtained for the study (protocol number 702342) from
IACUC (University of Pennsylvania, Philadelphia, PA).
All larvae used in this analysis were obtained from laboratory dogs infected with
S. stercoralis, UPD strain [12]. Fecal samples from dogs were
processed using the charcoal coproculture followed by Baermann funnel technique, as
outlined elsewhere [13]. Post parasitic L1 larvae were recovered from freshly
deposited stool samples; L3i larvae were recovered after 7 days of stool
incubation at 25°C. L3i larvae underwent surface decontamination by migration
through low-melting-point agarose. L1 larvae were decontaminated by 3 washes with
phosphate buffered saline (PBS) containing an antibiotic cocktail. Decontaminated
parasites were subsequently stored in Trizol reagent (Invitrogen, San Diego, CA) at
−80°C. Using this method, 30,700 post-parasitic L1 and 50,000 L3i
larvae were collected.
Total RNA was extracted by thawing pooled samples of L1 and L3i larvae at 37°C in
a warm water bath and centrifuging the samples at 4°C (805× g) for 10
minutes to obtain a pellet. The pellet was frozen in liquid nitrogen, ground
thoroughly with an autoclaved mortar and pestle and then purified using an RNeasy
mini kit (Qiagen, Valencia, CA) following the manufacturer's protocol. A Nano
Drop-1000 spectrophotometer (NanoDrop Products, Wilmington DE) was used to determine
the RNA concentration in each sample. RNA was more precisely quantified and quality
assessed using the 2100 Bioanalyzer (Agilent, Santa Clara, CA).
RNA samples from L1 and L3i stage larvae were co-hybridized using Cy3 and Cy5 labels
to discriminate the relative level of target bound to the microarray probe.
Fluorescent-labeled cDNA targets were prepared from total RNA using the Ovations
amino-allyl kit (NuGEN, San Carlos, CA) according to the manufacturer's
protocol. The kit utilizes an oligo dT primer for selective amplification of mRNA
transcripts.
Labeled samples were combined with blocking components poly(dA), yeast tRNA, and
human Cot-1, in hybridization buffer composed of 25% formamide/5×
saline-sodium citrate (SSC)/0.2% (w/v) sodium dodecyl sulfate (SDS) to a total
volume of 60 µl. After heating the sample (95°C for 3 minutes), it was
centrifuged (20,000× g) for 3 minutes. Fifty eight µl of the sample (1.6
µg of labeled cDNA) was loaded onto the microarray chip. The microarray chips
were hybridized overnight at 45°C using the MicroArray User Interface (MAUI)
hybridization system (BioMicro Systems, Inc., Salt Lake City, UT). The following day,
the chips were washed twice in 1× SSC/0.05% (w/v) SDS buffer (3 minutes
each wash) and twice in 0.1× SSC buffer (5 minutes each wash).
For the present study, four technical replicate experiments using pooled L1 and L3i
larvae were performed, including one dye swap. The microarray chips were imaged using
a GenePix 4000 B scanner (Molecular Devices, Sunnyvale, CA). Agilent Feature
Extraction software was used for image analysis, protocol GE2-v5 10 Apr08. The data
discussed in this publication have been deposited in the National Center for
Biotechnical Information (NCBI) Gene Expression Omnibus (GEO) and are accessible
through GEO Series accession number GSE24735 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE24735).
ESTs (11,335) were identified from L1 and L3i cDNA libraries created as part of the
nematode EST project [6], [7]. ESTs were organized into 3,571 contigs by bioinformatics
analysis [14].
Oligonucleotide probes designed to hybridize with these contigs were used to develop
early versions (V1 and V2) of chips manufactured by Combimatrix (Irvine, CA) based on
a variety of algorithms for oligonucleotide design. Versions 1 and 2 were assessed
for performance using RNA from L1 and L3i larvae. After testing the performance of
these two versions of the arrays, an optimized version (V3) was developed. The best
probe for each target was selected based on the average signal intensity for all
arrays and the number of arrays with detectable signal. The spot density was 22K
spots per array. Of the six oligonucleotides designed per target, one was designed
using the Array Designer program (Premier Biosoft International, Palo Alto, CA), two
were designed using E-Array (Agilent, Santa Clara, CA) using the “base
composition” method (replicated twice), two were designed using E-array
“best Tm” method, and the last was a 40-mer designed using Array
Designer. Probes were selected to avoid cross-hybridization to other sequences in the
target (contig) dataset manufactured by Agilent SurePrint. The probes designed to
make the V3 microarray are found in Table S1 in Supporting Information Text S1.
All data were exported into the cDNA Annotation System (dCAS) [14], [15]. This tool enabled annotation of
each S. stercoralis contig based on Basic Local Alignment Search
Tool (BLAST) alignments against multiple databases (NCBI nr protein database (NR),
Gene Ontology (GO), euKaryotic Orthologous Groups (KOG), Pfam protein families
database (PFAM), Simple Modular Architecture Research Tool (SMART), Wormbase (CELEG),
and Saccharomyces genome database (YEAST) and provided the corresponding E-values.
The database was also annotated manually with a composite categorization that
summarized the findings across databases. The entire annotated database, with
hyperlinks to the NIAID exon website, is accessible for download at: http://exon.niaid.nih.gov/transcriptome/S_stercoralis/SS-Supp-Web.zip.
A stand-alone version can also be accessed and downloaded at: http://exon.niaid.nih.gov/transcriptome/S_stercoralis/SS-Supp-StandAlone.zip.
Extract the excel file and the links directory to your own computer for browsing the
hyperlinks locally.
Spot values were calculated using a linear lowess dye normalization. Further, the
50th percentile of a set containing all the ribosomal genes in the
array was applied to all spot values. In cases of multiple spots for the same
S. stercoralis contig, the average of the log2 signal
was calculated for each array. The mean signal ratio (log2 L3i/L1) was
calculated from the signals for all 4 arrays. No surrogate values were applied. A
single group t-test analysis was calculated on the data set.
Variance shrinkage was not used when calculating p-values for differential
expression. Differentially expressed genes were identified using a
‘cutoff’ of 2 fold expression difference or greater for log2
L3i/L1 signal ratios, and p<0.01 for microarray signal data (false discovery rate
(FDR) = 2.5%).
A functional analysis was performed based on annotations provided by each database
(Pfam, SMART, KOG, etc.). The number of genes per functional category (e.g.
transcription, cytoskeleton, metabolism, etc.) was compared between L1 and L3i
differentially expressed genes (as defined by the above cutoff). To ascertain whether
genes belonging to certain functional classes were more likely to be highly expressed
in one stage or another, we used a statistical test for one proportion using Normal
approximation. Assuming a null proportion of 0.5 (i.e., that there is no difference
in the number of genes of that category for the two classes), p values were
calculated for deviation from 0.5 using Normal approximation. P values were adjusted
for multiple comparisons using the Bonferroni criterion.
Gene Set Enrichment Analysis (GSEA) is a robust method for analyzing molecular
profiling data examines the clustering of a pre-defined group of genes (gene set)
across the entire microarray database (all 3,571 contigs) in order to determine
whether the gene set has biased expression in one larval stage versus another [16]. GSEA was
used in this study to complement our use of single gene methods and determine whether
S. stercoralis gene sets grouped according to various putative
categories (for example, putative extracellular matrix genes) showed biased
expression in either larval stage. For this analysis, the entire list of contigs on
the microarray was sorted by mean log2 L3i/L1 signal ratios. The
distribution of genes from an a priori defined gene set throughout this ranked list
was then determined using GSEA. Based on this distribution, the expression difference
for each gene in the set is aggregated and a p-value for significance of the gene set
as a whole is calculated using the Kolmogorov-Smirnoff test.
Gene sets were compiled by first downloading GO categories from Wormbase (www.wormbase.org) for C. elegans genes. Definitions
for each GO category used can be found at http://www.wormbase.org/db/ontology/gene. S.
stercoralis orthologs for C. elegans genes were
determined by dCAS based on BLAST alignments to the C. elegans gene.
BLAST matches with E values>0.05 were excluded. Gene sets with fewer than 5
S. stercoralis contig matches were excluded from GSEA analysis.
Using these criteria, 18 S. stercoralis gene sets were created (see
Figure 1A). Additional
manually compiled gene sets included the group of S. stercoralis
genes whose products have been shown to be immunoreactive in humans infected with S.
stercoralis
[17]–[19], and a group
of putatively identified heat shock proteins.
Microarray expression data for S. stercoralis L3i and C.
elegans dauer larvae were compared using several methods as follows: 1)
We defined three gene sets comprising the S. stercoralis orthologs
of “dauer-enriched” C. elegans genes derived from either
C. elegans microarray expression data alone, both serial analysis
of gene expression (SAGE) and microarray expression data or from the Gene Ontology
category dauer larval development (Figure 1A) [20], [21]. We then used GSEA to determine whether these gene sets
showed significant L3i enrichment. 2) We examined whether a correlation exists
between C. elegans dauer/L1 microarray expression data obtained by
Wang and colleagues [20] with our S. stercoralis L3i/L1 microarray
expression data. The previously obtained C. elegans microarray
expression data can be found at http://cmgm.stanford.edu/~kimlab/dauer/ExtraData.htm, Table S1 in
Supporting Information Text S1, column “AdjD/L1_Ratio”
which corresponds to the average log2 expression values for C.
elegans dauer larvae at time 0 relative to L1 larvae [20]. 3) Using these
data, we calculated the absolute value of the difference between fold change values
for C. elegans genes and their S. stercoralis
orthologs (C. elegans dauer/L1 fold change - S.
stercoralis L3i/L1 fold change). Only those genes with robust microarray
expression data (p values<0.01) were included. In order to identify those genes
that are expressed differently by S. stercoralis L3i and C.
elegans dauer larvae, a list was generated of all S.
stercoralis-C. elegans orthologs with the greatest differences in fold
change values (absolute value >2). The list was further narrowed to include only
those S. stercoralis-C. elegans gene pairs where gene expression was
regulated in opposite directions between the two nematodes (Table 1).
The sequences of L3i biased genes (contigs 24, 25, 65, 243, 2136) and L1 biased genes
(contigs 55, 222, 387, 2328) were used to create primer-probe sets designed and
manufactured by Applied Biosystems (Foster City, CA). The sequences for these primer
probes are listed in Table S2 in Supporting Information Text S1. The
S. stercoralis control genes for qPCR analysis was S.
stercoralis glyceraldehyde 3 phosphate dehydrogenase (GAPDH; GenBank
accession number BI773092; contig_90;
log2L3i/L1 = −0.28179). Post-parasitic L1 and
L3i larvae (distinct from those hybridized onto the microarray) were collected and
total RNA made as described above. Total RNA (1 µg) from L1 and L3i larvae was
used to synthesize cDNA. qPCR was performed using all 9 primer probe sets in separate
reactions with L1 cDNA and also with L3i cDNA. The reaction was performed using
10× RT buffer (10 µl), 25 mM MgCl2 (22 µl), dNTP (20 µl),
random hexamers (5 µl), RNase inhibitor (2 µl), and multiscribe reverse
transcriptase (50 U/µl; 6.25 µl) in a microamp 96-well reaction
plate (Applied Biosystems). De-ionized, distilled water was added to total volume of
65.25 µl. Cycling conditions were: 25°C for 10 minutes, 37°C for 60
minutes, 95°C for 5 minutes, then 4.0°C. Each experiment was performed in
triplicate. The mean negative delta threshold cycle (delta CT) was
calculated for each sample. The data generated by performing qPCR using primer probes
for 9 contigs on L1 and L3i cDNA (n = 18) was plotted against
the average L1 and L3i intensity signals for each gene (Figure 2).
A total of 3,571 distinct contigs were studied by this microarray analysis (Table S3
in Supplemental Information Text S1). Using pre-defined cutoffs, 935 contigs
were identified as differentially expressed as shown in the volcano plot (Figure 3). Of these, 466 genes were
L1 biased (Table S4 in Supporting Information Text S1) and 469 genes were L3i biased (Table S5 in
Supporting Information Text S1). Among the 25 most highly expressed L3i
genes were the S. stercoralis orthologs of fatty acid/retinol
binding protein-1 (contig 1151; 11 fold expression difference), a ferritin chain
homolog (contig 94; 14 fold expression difference), and one of four putative
trehalases (contig 68; 14-fold expression difference). Among the 25 most highly
expressed L1 genes were electron transport chain proteins such as NADH dehydrogenase
(contig 371; 0.13-fold change); cytochrome b (contig 2328; 0.19 fold
change) and cytochrome c oxidase subunit 1 (contig 55; 0.29 fold change). The 25
most highly expressed L1 or L3i genes are listed in Table S6 in Supporting
Information Text
S1.
A greater number of L1 (n = 40) than L3i biased
(n = 18) genes were putatively involved in transcription
(p = 0.004, not Bonferroni adjusted; see Figure 4A,B). A complete listing of
these genes is shown in Figure
4B. This finding was also noted in an analysis of classifications based on GO
categories (p = 0.01 for ‘transcription’), and
manual annotations (p = 0.007 for ‘transcription
machinery’), although p values were not <0.05 when Bonferroni-adjusted for
multiple comparisons. BLAST matches to SMART and Pfam databases both indicated that
the sperm-containing glycoprotein (SCP) domain was found exclusively in the L3i-group
(n = 13 genes; see Table S7 in Supporting Information Text S1 for the
complete list; p value based on matches to Pfam = 0.003,
Bonferroni-adjusted for multiple comparisons).
Of the entire 3,571 contigs, 1,351 S. stercoralis genes
(37.8%) were of unknown function (manual annotation).
S. stercoralis orthologs were matched to 35 sets of C.
elegans genes grouped by various categories (e.g. negative regulation of
vulval induction, oviposition, heat shock proteins, etc.). Eighteen of 35 gene sets
queried met criteria for inclusion into the GSEA analysis (based on minimum size of 5
genes; see Figure 1A). Of
these 18 gene sets, only 2 gene sets were significantly enriched in the L3i phenotype
at nominal p value<5%. The most significantly enriched genes were those
with immunoreactive gene products recognized by sera from infected individuals (Figure 1B; nominal
p-value<0.0001; FDR<0.0001). Heat shock proteins were the next most highly
enriched (nominal p value = 0.034,
FDR = 0.56). For an annotated list of the individual genes
enriched in each of these categories, refer to Tables S8 and S9 in Supporting
Information Text
S1. None of the 18 gene sets were enriched in the L1 phenotype.
Four hundred and twenty two of 3,571 S. stercoralis contigs had
C. elegans orthologs for which robust microarray signal data were
available. When C. elegans and S. stercoralis
microarray signals were plotted against each other, a poor and non-significant
correlation was found (Spearman rank = 0.06;
p = 0.2444, graph not shown). No significant L3i enrichment of
S. stercoralis orthologs of C. elegans
‘dauer enriched’ genes was found by GSEA (nominal
p-value = 0.10). On the contrary, 25 orthologs expressed in
opposite directions by dauer and L3i larvae relative to their respective L1 stage
larvae were identified (see Table
1).
A statistically significant positive correlation was found between microarray
expression data and EST abundance data (p<0.0001; max
R2 = 0.26; graph not shown).
A positive correlation was found (Spearman rank = 0.4778;
p = 0.0449) between average L1 or L3i microarray intensity
signals and mean negative delta CT of qPCR (Figure 2).
In this microarray based analysis of differential gene expression between infective and
noninfective S. stercoralis larvae, we uncovered differences in the
expression of genes putatively encoding transcription factors, heat shock proteins and
antigens known to be immunoreactive in sera from infected humans. A comparative
microarray analysis of our data revealed several differences between S.
stercoralis L3i and C. elegans dauer stage larvae, such as
in the expression of genes putatively encoding collagen and myosin. Potential
therapeutic and vaccine targets were identified for further study.
Analogous to their non-dauer C. elegans counterparts, actively
growing S. stercoralis L1 larvae are thought to have higher rates of
transcription relative to L3i-stage larvae. This supposition is based on comparisons
between C. elegans non-dauer biased genes and S.
stercoralis L1-biased genes that suggest transcriptional conservation of
genes involved in early larval growth [6]. Consistent with this finding, we
found L1 biased expression of genes putatively involved in transcription. Among the
S. stercoralis L1-biased genes involved in transcription were
transcription initiation factors (contigs 3245, 1037, 686), transcription factors
(contigs 1905, 1277, 891, 2023, 2446, 1036, 1794, 592, 2210), and subunits of RNA
polymerase (contigs 1505, 3218, 1020, 2917). By contrast, the L3i-biased genes
involved in transcription though fewer, included transcriptional regulators (contigs
446, 445, 156) as well as transcription factors (contigs 1521, 519, 836, 167, 1478),
implying that L3i larvae are not transcriptionally inactive and may regulate
transcription differently. This would be consistent with what is known of C.
elegans dauer larvae, which express distinct sets of dauer-specific genes
at certain time points (dauer exit, for example) [20], [21].
Not surprisingly, genes encoding S. stercoralis antigens known to
produce robust antibody responses in infected humans were found to have L3i biased
expression by GSEA [17]–[19]. Two of these genes, IgG immunoreactive antigen (SsIR) and
NIE antigen, have been recently employed in serodiagnostic assays with some advantage
over crude antigen [19]. The finding that genes with products capable of inducing
protective immunity demonstrate stage-biased gene expression supports the further
investigation of these genes as vaccine candidates.
Heat shock proteins have been shown to play a critical role in determining parasite
survival during stressful conditions because they can bind denatured or misfolded
proteins [22],
[23]. Biased
expression of genes encoding heat shock proteins in the S.
stercoralis L3i relative to L1 larvae, as suggested by GSEA, is
consistent with this role. Hsp-90 in particular has been identified
as a parasitism-central gene based on changes in S. ratti gene
expression during high immune pressure [22] and is similarly abundantly
expressed by S. stercoralis L3i larvae.
The SCP domain, found exclusively in L3i biased genes, is a conserved domain of
unknown function present in a wide range of organisms [24]. Interestingly, it has been
found to be present in activation-associated secreted proteins that have been studied
as potential vaccine targets in other nematodes [24], [25]. Whether overrepresentation of
the SCP domain in the L3i group is related to the presence of these secreted proteins
is unclear, but activation-associated secreted proteins have been found to be
important in many parasitic nematodes in which they have been studied to date.
Consistent with previous findings, a striking L3i-C. elegans
‘dauer expression signature’ was not uncovered in this comparative
microarray analysis [6]. We instead identified genes that are regulated in
apparently opposite manners by C. elegans dauer and S.
stercoralis L3i larvae which offer useful clues about the biology of
S. stercoralis parasitism. L3i biased expression of the putative
nmy-2 gene (encoding the myosin heavy chain) is consistent with
the highly motile nature of L3i larvae which, unlike their dauer counterparts, seek
out and initiate infection in a host. Although dauer and L3i larvae both contain a
cuticle that enables survival in the environment, the parasitic cuticle has been
associated with the ability of infective stages to evade the immune response of the
host, and its structure varies from one species to another [26]. Biased expression of genes
putatively encoding particular collagens (col-37,
col-119) in the L3i but not the C. elegans
dauer, points to differences in the composition of the parasitic cuticle that could
potentially have a role in this regard. In fact, a recent microarray based analysis
of the response of the S. ratti transcriptome to host immunologic
environment notes upregulation of collagen genes by S. ratti which
is believed to play a protective role for the parasite [27].
C. elegans dauer and S. stercoralis L3i larvae can
survive in the environment even in the absence of a steady source of food. One way by
which this occurs is by the development of electron-dense intestinal granules that
store non-lipid products [11]. The gene lmp-1 plays an essential role in
this regard for dauer larvae as suggested by RNA interference studies [28]. It is likely
that L3i larvae similarly utilize these granules while in the environment. The
presence of these granules may additionally explain the darkened color of the
radially constricted intestines of L3i larvae, an appearance shared by its dauer
counterpart.
A key feature shared by dauer and L3i larvae is the ability to extend the lifespan
while in the free-living state. In both C. elegans and S.
stercoralis, the forkhead transcription factor DAF-16 plays a role in
regulating dauer diapause, longevity and metabolism [11], [29], [30]. A downstream target of DAF-16,
egl-10, is known to be negatively regulated by DAF-16 in
C. elegans
[29]. By contrast, this
gene was found to have biased L3i larval expression in S.
stercoralis. Such discordance is consistent with findings from a prior
study that failed to detect a transcriptional profile typical of down-regulated
insulin-like signaling in long-lived parasitic females of S. ratti
[31]. Although
the downstream targets of insulin-like signaling have not been fully elucidated in
Strongyloides species, the apparent upregulation of
Ss-egl-10 in the L3i potentially highlights adaptations at a
molecular level that likely underlie the evolution to parasitism. Such adaptations
could include alterations in genes controlling metabolic and developmental functions,
adaptations of pre-existing genes to encode new functions, and gene duplication and
diversification [32]. The apparent lack of a C. elegans
dauer-like transcriptional profile in S. stercoralis L3i is also
consistent with published findings on the expression of transcripts encoding the
orthologs of DAF-7 in this parasite [33] and in S. ratti and
Parastrongyloides trichosuri
[34]. DAF-7 is the
ligand that activates TGF-β-like signaling and thereby promotes continuous (i.e.
non-dauer) development in C. elegans. Its expression is biased
towards C. elegans first-stage larvae fated for continuous
development rather than dauer third-stage larvae [34], [35]. By contrast, messages encoding
DAF-7 orthologs in S. stercoralis, S. ratti and
P. trichosuri all show biased expression in the L3i, which has
been characterized heretofore as dauer-like [33], [34]. These facts notwithstanding,
outright rejection of the ‘dauer hypothesis’ of developmental regulation
in the L3i of parasitic nematodes on the basic of transcriptional data alone is
likely to be premature [36]. It is particularly noteworthy in this regard that key
signal transducing elements such as DAF-16 that directly regulate C.
elegans dauer development are constitutively transcribed and their
functions governed not at the transcriptional level but rather by posttranslational
modifications such as phosphorylation [37], [38].
The true value in identifying these and other genetic determinants of S.
stercoralis parasitism lies in whether the products of these genes can
induce protective immunity. Indeed, one of the genes identified in our list, the
S. stercoralis ortholog of eat-6
Na+k+ATPase, has already been identified as a potential vaccine
candidate based on animal experiments [39].
Contig 1872, a gene with L3i biased expression, encodes an ortholog of C.
elegans core subunit of the cytochrome bc1 complex, UCR 2.1
(E-value = 1E-014). This subunit has been shown to be a
potential target for antiparasitic drugs based on the finding that in C.
elegans, UCR 2.1 is essential for viability and is less related to
mammalian UCR-1 than to mitochondrial processing peptidases from other organisms
[40]. S.
stercoralis transgenesis experiments [41] may prove useful in investigating
the question of whether this gene is similarly essential for S.
stercoralis larval survival.
In our microarray analysis of S. stercoralis, we found abundant L3i
expression of the S. stercoralis ortholog of
hsp-90, contig_77 (3 fold expression difference).
Interestingly, the hsp-90 inhibitor geldanamycin has been shown to
have a macrofilaricidal effect on filarial nematode Brugia pahangi
[42].
Hsp-90 has been identified among S. ratti
parasitism central genes critical for survival and further studies investigating it
as a chemotherapeutic target are warranted.
Contig 1151, which was among the 25 most highly biased L3i genes (11-fold expression
difference), corresponds to fatty acid and retinol binding protein-1 (FAR-1;
E-value = 1E-016). FAR-like proteins are major secreted products
of parasitic nematodes that allow the parasite to scavenge essential nutrients from
its host [43].
Depletion of host lipids is thought to be necessary for parasite survival and may
additionally impair the host immune response [44]. These proteins have
additionally demonstrated stage and gender specificity in other nematodes, most
notably in the hookworm Ancylostoma ceylanicum
[45]. The
immunodiagnostic potential of FAR-like proteins has been assessed in other nematodes,
such as Onchocerca volvulus, in a serologic assay based on
Ov-20 (FAR-1) [45], [46], [47]. FAR-1 proteins have been successfully used in a vaccine in
animals infected with A.ceylanicum
[45]. These
microarray data identify S. stercoralis far-1 as an L3i-biased
target that may be a potential vaccine candidate or immunodiagnostic antigen.
Approximately one-third of S. stercoralis genes are of unknown
function. This finding is consistent with a previous EST analysis that revealed a
similar percentage (25%) of S. stercoralis clusters with no
significant BLAST alignments [8]. This finding is also consistent with functional genomics
analyses of the C. elegans and human genomes where significant
numbers of genes of unknown function were identified [48], [49]. Some of these unknown sequences
may derive from 3′ untranslated mRNA regions, which are common in
polydT-primed libraries [50]. The complete genome sequence of S.
stercoralis is not available to date. Inferred functional annotations of
an analogous nematode C. elegans, while useful, may not be directly
applicable to S. stercoralis, as suggested by interspecies
differences uncovered in the present comparative microarray analysis. Because a
number of C. elegans genes did not have S.
stercoralis orthologs that were also differentially expressed according
to our predefined ‘cutoffs,’ it was difficult to formulate gene lists
organized into functional categories with at least 5 contigs. This limited our
ability to analyze biochemical or metabolic pathways of potential importance. As our
knowledge of the S. stercoralis genome increases, these microarray
analyses will likely gain in usefulness and a more direct approach using annotation
based on known S. stercoralis gene functions would be even more
informative.
DNA microarrays allow for simultaneous analysis of large numbers of genes from two or
more biologic conditions. This powerful method of analysis has revolutionized our
understanding of the immunopathogenesis of schistosomiasis [51], for example, and has advanced the
development of vaccine discovery and therapeutics in parasitology [52], [53]. Until now,
studies of S. stercoralis have been limited to the analysis of ESTs
rather than the full genome sequence. Development of a novel DNA microarray tool for
the study of S. stercoralis represents an exciting step forward in
our understanding of this parasite.
|
10.1371/journal.pcbi.1004882 | Estimating Time of Infection Using Prior Serological and Individual Information Can Greatly Improve Incidence Estimation of Human and Wildlife Infections | Diseases of humans and wildlife are typically tracked and studied through incidence, the number of new infections per time unit. Estimating incidence is not without difficulties, as asymptomatic infections, low sampling intervals and low sample sizes can introduce large estimation errors. After infection, biomarkers such as antibodies or pathogens often change predictably over time, and this temporal pattern can contain information about the time since infection that could improve incidence estimation. Antibody level and avidity have been used to estimate time since infection and to recreate incidence, but the errors on these estimates using currently existing methods are generally large. Using a semi-parametric model in a Bayesian framework, we introduce a method that allows the use of multiple sources of information (such as antibody level, pathogen presence in different organs, individual age, season) for estimating individual time since infection. When sufficient background data are available, this method can greatly improve incidence estimation, which we show using arenavirus infection in multimammate mice as a test case. The method performs well, especially compared to the situation in which seroconversion events between sampling sessions are the main data source. The possibility to implement several sources of information allows the use of data that are in many cases already available, which means that existing incidence data can be improved without the need for additional sampling efforts or laboratory assays.
| Human and wildlife diseases can be tracked by looking at incidence, which is the number of new infections per time unit (typically day, week or month). While theoretically this would only be a matter of counting the number of newly infected individuals, in reality these data are difficult to acquire due to limited sampling possibilities and undetectable cases. This means that a method must be used to estimate the real incidence using a limited amount of data. For many infections, the concentration and quality of antibodies changes predictably over time, which means that one could use the antibody level at any point in time to back-calculate how much time passed since the infection entered the body. Other information, such as the age of the individual, or the presence of the pathogen, can also help to estimate when an individual became infected. Improving on existing methods, we developed a method that allows the use of a wide range of information sources for estimating individual time since infection. Using arenavirus infection in mice, we show that this method works well when sufficient background data are available, and that it can greatly improve the estimation of incidence patterns.
| Infection incidence (the number of new infections per time unit) is a basic epidemiological measure that describes the transmission of an infection through time. Because the exact time at which an individual acquired an infection is difficult to assess, time of symptom onset is often used as a proxy (e.g. [1]). When the time between the moment of infection and symptom onset (the incubation period) is predictable, this proxy will not bias results, but incidence estimation does become problematic with asymptomatic infection or when incubation periods vary unpredictably [2].
Another common problem for measuring incidence is the time resolution of data, as the temporal precision of incidence is directly related to that of data “sampling”. Ideally, each new infection is detected and recorded immediately, but in reality this is rarely possible and new cases are often recorded at irregular intervals and a low number of time points, resulting in suboptimal resolution incidence data [3, 4]. Even more importantly, when sampling intervals are larger than the duration of symptoms, a proportion of cases will be missed. This problem is especially common in the case of wildlife diseases, as natural populations are often sampled incompletely and at relatively large intervals [5]. In such cases, indirect measures of incidence that rely on evidence of past infection are needed.
The presence of specific antibodies indicates whether an individual has previously been infected, and the distribution of different antibody (Ab) types (e.g. IgG, IgM, IgA) can give a rough indication of how recently the individual was infected [6–9]. If individuals in a population are sampled repeatedly, a seroconversion event in between two sampling events provides further information about the time since infection. Aside from being present or not, Abs vary over time in quantity (titer) and quality (avidity). On the condition that this temporal variation is sufficiently constant and predictable within and between individuals, these antibody dynamic properties can be used for a more accurate estimation of the time since infection.
Avidity (Ab-antigen bond strength) tends to increase with time since infection, which means that it can in some cases be used to back-calculate the time since infection. But although this method is used routinely, e.g. for human cytomegalovirus [10, 11], its sensitivity is low, and it can only differentiate between “recent” or “old” (e.g. less or more than 90 days since infection for cytomegalovirus) infection events [6, 12].
Temporal dynamics of Ab levels can be another source of information about time since infection. In such cases a model must be created that describes the course of Ab levels (titers) over time since infection using known serological response data. This model is then used to back-calculate, given an Ab titer, the time since infection, which in turn can be used for incidence estimation. This has been done for pertussis [13, 14], HIV [15, 16] and Salmonella [17, 18].
While this method is promising, significant improvements are still possible in two main ways. A common, important limitation for developing good time since infection models is the lack of detailed information about individual Ab dynamics, which limits the explanatory power of such models as they must in that case be estimated using cross-sectional instead of individual data (e.g. [18]). Experimental challenge studies, in which the exact time since infection is known, would be needed to describe and model the within-individual Ab dynamics needed to calculate time since infection, but these are notoriously difficult to conduct [19]. A perhaps more feasible approach to improving time since infection models would be to make optimal use of all available sources of information on the course of infection. While changes in Ab presence/titer over time can contain much information on time since infection and are the most obvious input data, additional information is contained in parameters such as the presence/quantity of the pathogen (or of other immune response markers), individual age (e.g. for typical childhood infections, young individuals are more likely to have been infected recently than older ones) or season (e.g. for seasonal infections, individuals are more likely to have been infected recently during or short after the peak transmission season).
Here, we present a novel method that allows the integration of multiple serological biomarkers (Ab presence/absence/titer, pathogen presence/absence) as well as additional prior knowledge (e.g. age, season, capture probability) to inform a semi-parametric mixed model that back-calculates the time since infection of each individual, in a Bayesian framework. The integration of multiple sources of information ensures the optimal use of data that are often already available but not yet taken into account.
We apply this method to estimate the incidence of Morogoro virus (MORV) infection in Natal multimammate mice (Mastomys natalensis). This model system is used because the epidemiological and demographic parameters necessary for testing this method are well known for this infection. MORV is a member of the arenaviruses, a family of zoonotic viruses that includes viruses able to cause hemorrhagic fever in humans after acquiring infection from wild rodents (e.g. Lassa virus (LASV), Junin virus, Machupo virus) [20]. It is restricted to East-Africa, and while it does not seem to cause disease in humans it is closely related to Lassa virus which causes Lassa hemorrhagic fever in West-Africa, and with which it shares the same host species. Because both the population ecology of the rodent host M. natalensis and the infection ecology of MORV have been studied thoroughly (driven by the host’s status as an agricultural pest species and the virus’ close resemblance to LASV) [21, 22], MORV infection provides a good model system for testing the current method.
As is the case for other time since infection methods, two types of datasets are needed to estimate incidence. A first dataset, consisting of any type of data that contains information on the temporal course of infection (e.g. Ab titer dynamics in an infected individual), is used once in order to create an integrated model of individual time since infection. Once created, this model can be used to estimate incidence from cross-sectional sampling data that ideally (but not necessarily) includes repeated measures of individuals.
We use a wildlife disease model system to develop and test the method because detailed individual-level infection/antibody dynamics are available, but also to show that the method is applicable to both human and wildlife infections. Because it is usually difficult to monitor infections at a high time-resolution, this method can provide a way to improve the quality of longitudinal data without having to increase sampling efforts.
In the following, we show how different types of data (e.g. levels, presence/absence) can be used to estimate the time of infection, and as a proof of principle we apply the method to MORV transmission in the multimammate mouse M. natalensis. For each type of data we present a generalised method and immediately apply it to MORV, and we show how to use individual estimates of the time of infection to estimate incidence in the population. Finally, through the use of simulated MORV transmission data we investigate method performance under different conditions.
MORV Ab level dynamics and virus presence in blood and excretions (urine, feces, saliva) have been quantified previously in a challenge study, described in [23], where multimammate mice from a breeding colony were injected with cultured MORV and sampled frequently for 210 days, which is more than their average lifetime in natural conditions (Fig 1 and [23]).
The estimation of the time of infection θ can be based on different dimensions of the immune response that each require a slightly different approach. In the following we consider two different sources of information.
Because an individual can of course only have been infected when it was alive and present in the population, the estimation of θ can be improved by incorporating prior information about the probability of an individual being alive/present, i.e. by modeling P(θ|T). Here, we show how to implement information on mortality rate and maximum life span, age at the time of sampling, and encounter probability, but note that any source of information can be used in a similar way as long as it results in a realistic prior distribution.
Knowledge about the maximum life span can be informative because it sets an upper boundary to the possible time since infection, and is especially useful in situations where the maximum life span is short relative to the possible time since infection. If an individual was last tested at time tn and the maximum life span is known, then the prior distribution P(θ|T) can be reduced to
P ( θ | T ) ∼ 1 life span θ > ( t n - life span) θ < t n ,
with [. < .] is a boolean operator that returns 1 or 0 when the equality is true or false, as shown in Fig 3a.
Similarly, one could make use of the mortality rate, as this is directly associated with the possible age of an individual. If an individual was first encountered at time t1 and we assume a mortality rate γ as inferred from data, we arrive at prior distribution
P ( θ | T ) ∼ max exp ( γ ( θ - t 1 ) ) , 1 θ < t n ,
as shown in Fig 3b. This figure clearly shows that, due to mortality, it becomes increasingly unlikely for individuals to have been alive, and therefore infected, further in the past.
When more precise information exists on the age of an individual focus individual (which is trivial for humans, while for wild animals this can be based on physiological or morphological features such as weight), this can be taken into account explicitly by including
P ( θ | T ) ∼ θ > ( t 1 - age ( t 1 ) ) θ < t n ,
if the individual was first encountered at time t1, see Fig 3c.
More applicable to wildlife infections is the use of encounter probability (typically termed trapping or capture probability, but for consistency and human application we will here refer to it as encounter probability). In a typical capture-mark-recapture study, only a proportion of individual is captured during each session, and well-developed methods exist for estimating encounter probability [27, 28]. This encounter probability can be used to estimate the likelihood of an individual being alive at a certain point in time, assuming a closed population during that time (no migration). If an individual is first encountered at time t1, the probability of it being born at time θ decreases with t1 − θ, as it becomes increasingly unlikely that it was not encountered during (t1 − θ) / Δt trapping sessions.
If we estimate encounter probability penc for every trapping session, this information can be used to further improve the prior time distribution:
P(θ|T) ∼ max[ (1−penc)(t1−θ)/Δt,1 ][ θ<tn ] ≈ max{ exp[ penc(θ−t1)Δt ],1 }[ θ<tn ],
where Δt is the sampling or trapping interval time, and with the latter approximation valid only when penc < <1. This approach only holds if one can assume a closed population where every individual was in the population during its lifetime and the effects of migration are negligible.
One could also use seasonal information or cross-sectional data to inform the prior P(θ|T), or in fact any other data source that contains any type of information about the time since infection.
Given the resulting posterior probability P(θ|X, T), the observer still has to use a decision criterion to decide which time of infection θ is most likely. Probably the most obvious decision criterion is the mean squared error (MSE) of the time since infection by selecting the θ ^ i for which
M S E = 1 N i n d ∑ 1 = 1 N i n d θ ^ ( i ) - θ ( i ) 2 ,
with i running over a population of Nind individuals, is minimal. It can be shown that this is the case for θ ^ = ∫ d θ P ( θ | T , X ) θ [29].
In order to assess the quality of the estimates, the remaining uncertainty on the time since infection can be inspected conditional on the observed data (X, T), which can be quantified using the conditional entropy E(θ|X, T) [29], i.e.,
E ( θ | X , T ) = ∫ θ ′ d θ ′ p ( θ ′ | X , T ) log 2 p ( θ ′ | X , T ) ,
where θ′ runs over all possible time since infection values. Conditional entropy is a commonly used measure in information theory that quantifies (in bits) the remaining amount of uncertainty about the actual value of the quantity of interest (here: time since infection). The highest entropy is attained for a uniform posterior probability distribution (maximum uncertainty), whereas the minimum (zero) entropy is obtained when there is no uncertainty left about the actual value [29]. In an epidemiological context, the entropy value can be used to improve the reliability of estimated incidence (see next paragraph) by removing all estimates of θ for which the entropy value is larger than a threshold value. The choice of this threshold value will mostly depend on the trade-off between sample size and estimation error: a low threshold value will generally result in a higher quality of the remaining θ estimates, but at the cost of reducing the final size of the dataset, and will therefore be dataset-specific.
One of the main purposes of knowing the time of infection of an individual is to analyse and model infection incidence on a population level. To this end, we need to estimate the time of infection θi for all sampled individuals i in the population and count the number of newly infecteds on a regular (usually daily) basis. Because in most situations only a proportion of individuals will be encountered and sampled, the “real” proportion of new infections needs to be estimated. This can be done by dividing the number of infecteds by an estimate of the proportion of encountered individuals. Given a certain sampling interval Δt and an encounter probability at each session (penc), this proportion can be approximated by
proportion encountered = γ ∫ d t exp ( - γ t ) 1 - ( 1 - p e n c ) t / Δ t ,
where the integral runs over all the survival times t following an exponential distribution with 1/γ (the average lifespan of an individual in our simulation), t / Δt is the approximate number of sampling sessions during lifetime t, and (1 − penc)t/ Δt is the approximate probability that an individual is never encountered during these sessions.
Next, in order to test the back-calculation scheme, we need a dataset of individuals in a population, with full knowledge of their infectious status at each moment. Also, to test the efficacy of the method as a function of sample size (with regard to intervals between sampling sessions as well as the sampling effort), we need datasets collected under different trapping regimes. We therefore simulate MORV transmission in a population of multimammate mice, “sampled” in different trapping sessions, with each individual given simulated infection attribute data based on the experimentally-derived [23] course of Ab levels and probability of virus presence in blood and excretions. These simulated data are equivalent to epidemiological data obtained through surveys with repeated sampling, but now of course with the difference that our simulated data are completely known for testing purposes. All simulated data, as well as the Matlab code used to apply the time of infection estimation method, can be found in S1 Data.
As input for the model, we use simulated data from an existing individual-based spatially-explicit SEIR model, which models the population dynamics and the transmission of Morogoro virus in M. natalensis [30]. In this model, individuals are born in the susceptible (S) state and can become infected through contact with infectious (I—infectious state) individuals. When infected, they enter a latent stage (E—exposed state) during which they cannot transmit the virus, until they become infectious (I) after around 6 days. After around 45 days they stop being infectious, recover from the infection (R—recovered state) and remain immune against re-infection for the remainder of their life. Latent and infectious periods were simulated assuming an exponential distribution. The simulation is run over a total area of 10ha, but in order to recreate a realistic situation in which individuals can move freely in and out of the study site, only the individuals that are encountered within a central 5ha area were available for “trapping”. Realistic population densities and fluctuations are used, ranging between around 10 and 150 per ha. After a simulation burn-in period, two years of data are considered (from day 1000 until 1730).
Throughout the simulation we keep track of each individual’s age, time since infection t, and we simulate trapping sessions with a time interval Δt, in which every individual present in the 5ha area has a probability ptrap to be trapped. Whether an individual is trapped or not is determined using pseudo random numbers. This way, for every individual we can generate an artificial set of measurements (T, Xk) that we can then use to estimate the time of infection θ ^. Xab are random realisations according to the multivariate distribution shown in Eq 8 at times T. Xvb and Xve are random draws with respective probabilities pvb and pve at times T. We vary the time intervals between capture sessions using Δt = 1, 7, 14, 28, 56 days, as well as the probability for each of the individuals to be captured using ptrap ∈ (0, 1).
We implement a maximum life span of M. natalensis of 450 days based on [31]. The average mortality rate (averaged across the year) is calculated from the simulation data, and estimated to be μ = 0.008537 mice/day (average life span of 117 days). Both maximum and average life span are used as prior information for all time of infection estimates.
The estimation of the time since infection is much improved by the use of Ab levels, as opposed to when only using Ab presence/absence data (Figs 4 and 5). The use of Ab levels also results in a much better reconstruction of incidence dynamics, even without including additional information such as virus presence or individual age (Fig 6). When using Ab presence/absence data, incidence can only be estimated with a low temporal resolution, the main consequence being that the peaks and troughs of the incidence dynamics were estimated badly (Fig 6). Although the incidence peaks are estimated quite well when using Ab levels, the periods of low incidence are still often over-estimated (Fig 6).
The inclusion of additional information (Vb, Ve, individual age) greatly improves the estimation of time since infection and incidence (Figs 4–6). Interestingly, this effect is more pronounced when using Ab presence/absence than when using Ab levels. The combined use of Ab levels and other available information results in the highest quality reconstruction of incidence dynamics, where the inclusion of additional information mainly reduces the previously observed over-estimation of low incidence levels between peaks.
Nevertheless, even when using Ab presence/absence instead of Ab level data, incidence can be reconstructed well when including Vb, Ve and individual age. This is encouraging, given the fact that many datasets, especially for wildlife infections, already contain some or all of this information; it means that by applying the back-calculation method, many existing incidence estimations can be improved significantly without additional laboratory or sampling efforts.
The quality of the estimates strongly depends on sampling frequency (or trapping interval) and the proportion of individuals that is encountered (or trapped) and sampled. While more additional prior information always results in a better estimation of the time since infection, we see that, at low (realistic) encounter probabilities, this effect is strongest (Fig 4). We also observe that a higher sampling frequency results in better estimates (Fig 5), and this is largely an effect of increased sample sizes: when adjusting the trapping probability to equalise sample sizes of different sampling frequencies, this effect mostly disappears (S1 Fig). This means that, in theory, similar results can be reached for any sampling frequency or trapping interval, but only if the sampling effort is increased so that a sufficient number of individuals can be sampled. Nevertheless, we observe that long sampling intervals (28–56 days) generally result in lower quality estimates (S1 Fig), indicating that a shorter interval would still be preferred.
In the model, we introduce the use of entropy (which is inversely related to information) as an indicator of the amount of uncertainty contained by an estimate. Fig 7 shows how estimates of the time since infection with a higher deviation from the real time since infection generally also contain less information (i.e. have a higher entropy). Similarly, we observe a strongly positive correlation between the MSE of the estimated time of infection and the entropy level (S2 Fig). Therefore, by removing estimates above a critical entropy value, the MSE can be lowered, albeit at the cost of a lower sample size. Because of this trade-off it is not possible to suggest an optimal critical entropy cut-off value, which should rather be chosen depending on the specific situation, sample size and quality of available information.
Although the model performs well and seems promising for a wide range of situations, there are a number of important assumptions and prerequisites that must be met before it is possible to apply the model to data. First, of course, empirical data on the dynamics of biomarkers (e.g. antibodies, viral RNA, etc) within individuals must be available. These can be relatively straightforward data such as knowledge about when after infection individuals seroconvert and how long antibodies remain detectable, or more elaborate information such as the temporal variation of antibody and virus levels after infection.
Then, these data can only be used if they are sufficiently consistent across individuals. If there is too much inter-individual variation in the shape of biomarker dynamics, it will not be possible to predict individual patterns. This does not however mean that there can not be individual variation in the magnitude of the response, as this would in fact be easy to implement into the model.
Further care must be taken if biomarker data have been obtained through laboratory experiments. Because laboratory conditions are often controlled and limited, natural variation in factors such as individual differences in immune response, stress, secondary infection, initial dose, boosting, etc. may result in different biomarker dynamics that could invalidate a time since infection model if they can not be incorporated into the model [32]. Ideally this is tested through a comparative study between laboratory and field patterns, but if such a study has not been done we must assume that the patterns observed in laboratory conditions apply to the natural situation.
Other factors that could render the use of a time since infection model difficult are the existence of maternal antibodies and the simultaneous presence of chronically and acutely infected individuals, as these factors would be difficult (but not necessarily impossible) to disentangle and take into account. On the other hand, under certain conditions these factors may even improve the model, as they provide additional information; for example, if maternal antibodies only occur for a certain period in newborn individuals, and if maternal antibodies can be distinguished from other antibodies (e.g. because of lower levels or using a different assay), this information can likely improve the estimation of the time since infection when incorporated into the model.
Under the conditions described here, the model is a significant improvement on existing models (e.g. [14, 17, 18, 33]). It provides a relatively simple probabilistic framework for the incorporation of any data source that can inform the estimation of time since infection, such as biomarker level/presence, age, season, sex, weight, etc., and thus allows for the use of individual-level data to interpret cross-sectional survey data and estimate population-level incidence. An important strength of the method is that it does not assume a certain form for the underlying models, which makes it possible to use a general spline method but also a more specific ordinary differential equation (ODE) method when a good ODE can be found (e.g. [17]).
More specifically for wildlife infections, the method has the potential to enhance existing long-term data. Often, large logistical efforts are necessary to collect longitudinal data on wildlife infections, and even the best datasets have a relatively low temporal resolution, typically consisting of monthly (but often less frequent) capture sessions [5, 34–37]. Prevalence or incidence patterns resulting from such data are usually also limited to this capture frequency, and to our knowledge the only efforts for improving these data have been the rough estimation of seroconversion events between two capture sessions (e.g. [38, 39]). We have shown however that by integrating multiple sources of information (that have often already been collected or analysed), the quality of incidence data can be greatly improved, especially (but not uniquely) when predictable antibody level dynamics are available.
Due to its flexibility, the model presented here allows the integration of multiple sources of information, thus making optimal use of all available data for estimating individual times of infection and population incidence. It provides a conceptually simple, flexible framework for estimating the time since infection and incidence of human as well as wildlife infections, and can potentially be used to significantly improve incidence estimation based on already existing data.
|
10.1371/journal.ppat.1005017 | A Structural and Functional Comparison Between Infectious and Non-Infectious Autocatalytic Recombinant PrP Conformers | Infectious prions contain a self-propagating, misfolded conformer of the prion protein termed PrPSc. A critical prediction of the protein-only hypothesis is that autocatalytic PrPSc molecules should be infectious. However, some autocatalytic recombinant PrPSc molecules have low or undetectable levels of specific infectivity in bioassays, and the essential determinants of recombinant prion infectivity remain obscure. To identify structural and functional features specifically associated with infectivity, we compared the properties of two autocatalytic recombinant PrP conformers derived from the same original template, which differ by >105-fold in specific infectivity for wild-type mice. Structurally, hydrogen/deuterium exchange mass spectrometry (DXMS) studies revealed that solvent accessibility profiles of infectious and non-infectious autocatalytic recombinant PrP conformers are remarkably similar throughout their protease-resistant cores, except for two domains encompassing residues 91-115 and 144-163. Raman spectroscopy and immunoprecipitation studies confirm that these domains adopt distinct conformations within infectious versus non-infectious autocatalytic recombinant PrP conformers. Functionally, in vitro prion propagation experiments show that the non-infectious conformer is unable to seed mouse PrPC substrates containing a glycosylphosphatidylinositol (GPI) anchor, including native PrPC. Taken together, these results indicate that having a conformation that can be specifically adopted by post-translationally modified PrPC molecules is an essential determinant of biological infectivity for recombinant prions, and suggest that this ability is associated with discrete features of PrPSc structure.
| A key prediction of the prion hypothesis is that autocatalytic, misfolded PrPSc molecules should be highly infectious. Various recombinant PrPSc conformers are able to self-propagate in vitro, yet paradoxically only some of these conformers possess significant levels of specific infectivity in bioassays. Here we use two closely-matched autocatalytic recombinant PrP conformers that share the same origin but differ by >105-fold in specific infectivity to study the molecular basis of prion infectivity. We show that infectious and non-infectious autocatalytic recombinant PrP conformers have subtle structural differences, and that GPI-anchored PrP substrate molecules can only adopt the infectious PrPSc conformation. We conclude that post-translational modifications of host PrPC molecules play a critical role in restricting the range of recombinant PrPSc conformers that are biologically infectious.
| The conformational conversion of the host-encoded prion protein (PrP) is a central pathogenic event in the prion diseases [1]. In healthy individuals, PrP adopts a fold that is rich in α-helix, termed PrPC, and is post-translationally modified by the incorporation of N-linked glycans and a C-terminal glycosylphosphatidylinositol (GPI) anchor. In individuals suffering from prion disease, PrPC is misfolded into a β-sheet rich conformation, termed PrPSc, which is capable of acting as a template for the conformational conversion of additional PrPC molecules into PrPSc. This self-propagating activity of PrPSc is referred to as autocatalysis and is thought to underlie the infectious nature of the prion diseases. A critical prediction of the protein-only hypothesis is that autocatalytic PrPSc molecules should be infectious.
A number of in vitro techniques for generating misfolded, autocatalytic PrPSc conformers have been developed and refined, including the cell-free conversion assay [2] and the serial protein misfolding cyclic amplification (sPMCA) technique [3,4]. With few exceptions, PrPSc conformers derived from post-translationally modified, native PrPC substrates have been highly infectious when bioassayed in wild-type animals [4–8]. In contrast, various autocatalytic PrPSc conformers derived from recombinant PrP substrates lacking post-translational modifications have displayed large variations in specific infectivity levels as determined by bioassay in wild-type animals [9–15]. The structural and functional basis of this striking variability in specific infectivity between different autocatalytic recombinant PrPSc molecules remains unknown.
Recently, using only bacterially expressed recombinant PrP and a single endogenous phospholipid cofactor molecule, phosphatidylethanolamine (PE), as substrates, Deleault et al. successfully produced high titer (2.2 x 106 LD50 U/μg PrP), chemically defined mouse prions in vitro [10]. Interestingly, it was observed that the prions produced from these minimal components always formed into a single infectious strain with unique, novel biological properties regardless of the seed originally used to template the in vitro reactions. Importantly, when this novel prion strain was subsequently propagated in the absence of PE cofactor, a new misfolded recombinant PrP conformer was produced, which could also self-propagate in sPMCA reactions, but which surprisingly failed to cause disease upon injection into wild-type mice [10]. This new autocatalytic conformer, which we refer to as protein-only PrPSc, therefore had a >105-fold lower level of specific infectivity as compared to the PrPSc conformer produced in the presence of PE cofactor, which we refer to as cofactor PrPSc.
We saw an opportunity to identify structural and functional properties associated with recombinant PrPSc infectivity by directly comparing these two related PrPSc conformers, which share the same origin and autocatalytic behavior, but differ strikingly in biological infectivity.
Cofactor and protein-only PrPSc are distinct misfolded recombinant PrP conformers that differ >105-fold in their specific infectivity for wild-type mice [10]. While both of these conformers demonstrate autocatalytic activity when used to seed sPMCA reactions containing recombinant PrP substrate (Fig 1A and [10]), only cofactor PrPSc also demonstrates autocatalysis when used to seed sPMCA reactions containing normal brain homogenate as the substrate (Fig 1B and [10]). The complete failure of protein-only PrPSc to function as a seed for conversion reactions containing native PrPC substrate (Fig 1B, left sample group) provides a logical explanation for this conformer’s lack of infectious activity in vivo, and may apply more generally to other recombinant PrPSc conformers which demonstrate low levels of specific infectivity in bioassays. Using cofactor PrPSc as a well-matched control, we therefore sought to gain structural and mechanistic insight into the substrate-dependence of protein-only PrPSc autocatalytic activity as a means to understand the structural and functional determinants of recombinant PrPSc infectivity.
To help focus the structural comparison between cofactor and protein-only PrPSc molecules, we first tested whether the protease-resistant cores of both conformers, which contain approximately two thirds of the residues of mature full-length PrP, have the same substrate-specific activity as their respective parent PrPSc molecules in in vitro propagation experiments (Fig 1). We found that in sPMCA reactions containing recombinant PrP substrate and defined cofactors, both full-length and truncated cofactor and protein-only PrPSc molecules function as competent seeds which faithfully propagate the characteristic PK-resistant bands associated with their parent PrPSc molecules (Fig 1A) [10]. Moreover, in in vitro conversion reactions containing native, brain-derived PrPC substrate we found that full-length and PK-digested cofactor PrPSc drive the conversion of native PrPC (Fig 1B, third and fourth panels), while full-length and truncated protein-only PrPSc do not (Fig 1B, first and second panels), indicating that functional differences in the in vitro activity of cofactor and protein-only PrPSc molecules can be localized to the PK-resistant core. Epitope mapping of the cofactor and protein-only PrPSc PK-resistant cores (S1 Fig), revealed that both are C-terminal PrP fragments that include the 6D11 epitope (residues 93–109, with 97–100 as the major determinants of binding [16]) (S1 Fig, top panel) and the extreme C-terminus (S1 Fig, bottom panel). Based on these results, we focused our subsequent structural analyses on C-terminal residues beginning at glycine 89 (G89), the primary PK-cleavage site for PrPSc 27–30 [17].
To compare the structures of cofactor PrPSc and protein-only PrPSc molecules, we performed hydrogen/deuterium exchange MS (DXMS) on these two conformers generated in parallel from the same OSU prion strain seed (Fig 2A). The OSU prion strain was originally synthesized by Wang et al. using recombinant PrP, total liver RNA and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) [15], and it has previously been referred to as the OSU strain by Deleault et al. [10]. OSU-seeded cofactor and protein-only PrPSc samples used for structural analysis in this study were generated with high conversion efficiency in sPMCA (S2 Fig) and then purified prior to DXMS analysis by a series of ultracentrifugation steps described in Materials and Methods. Co-sedimentation of significant quantities of non-specifically aggregated, protease-sensitive PrP was ruled out in the OSU-seeded cofactor PrPSc DXMS sample due to its near complete conversion to PK-resistant PrPSc (96% PK-resistant conversion efficiency, S2 Fig). To assess the potential contribution of non-specifically aggregated, protease-sensitive PrP to the OSU-seeded protein-only PrPSc sample analyzed by DXMS (77% PK-resistant conversion efficiency, S2 Fig), we mock-seeded protein-only PMCA reactions and subjected the resulting material to the DXMS purification protocol (S3 Fig, top panel). We did not detect any non-specifically aggregated PrP in these mock-seeded PMCA reactions (S3 Fig, top panel, samples S0 vs P3), indicating that the protein-only PrPSc DXMS sample does not contain appreciable quantities of non-specifically aggregated, protease sensitive PrP.
Regional solvent accessibility of cofactor and protein-only PrPSc was determined by incorporating deuteration data from 188 overlapping C-terminal peptides (Fig 2C) recovered in a representative experiment from each deuterium-labeled PrPSc sample. This high density of overlapping deuterated peptides provides PrPSc solvent accessibility measurements with resolution down to segments of ~5 amino acids. In examining and discussing the results of this study, specific regions of the PrP primary sequence are referred to either by explicit residue numbering, based on the mouse PrP sequence, or with reference to the location of known secondary structural elements in monomeric, α-helical recombinant PrP (α-PrP). For example, the domain corresponding to residues 178–216 could also be described as α2-α3 because it encompasses the second and third α-helices in α-PrP.
The C-terminal cores of cofactor and protein-only PrPSc are both substantially protected from solvent exchange (Fig 2A) as compared to α-PrP (S4 Fig). This solvent protection is consistent with widespread conversion to β-sheet secondary structure and/or the formation of large solvent-excluding aggregates [20]. Within the misfolded, solvent-protected PrPSc core there are large regions in which cofactor PrPSc and protein-only PrPSc have remarkably similar solvent accessibility profiles—in particular, the regions containing residues 118–143 and 165–230. However, there are also specific domains in which the cofactor and protein-only conformations can be distinguished by solvent accessibility. Most clearly, the domain encompassing residues 144–163, corresponding to α1 and β2 in α-PrP [21], is more solvent-exposed in protein-only PrPSc than in cofactor PrPSc (Fig 2A). The relatively exposed structure of the α1-β2 domain was preserved, and in fact accentuated, in an independently prepared protein-only PrPSc sample (S5 Fig). In addition, residues 91–110 appear to be slightly more exposed in protein-only PrPSc than in cofactor PrPSc while residues in the palindromic region (amino acids 111–120) are more obviously protected in protein-only PrPSc. Differences in these relatively N-terminal regions may contribute to the differing susceptibility to PK cleavage observed for the cofactor and protein-only PrPSc conformations (Fig 1 and S1 and S2 Figs).
Deleault et al. [10] originally used three prion strains with distinct infectious phenotypes to seed the chemically-defined, recombinant PrP conversion system used to produce cofactor and protein-only PrPSc molecules. Interestingly, the three input strains converged into a single strain with a novel biological phenotype upon propagation in sPMCA reactions containing only recombinant PrP substrate and a single cofactor. We used DXMS to examine the structures of the two additional cofactor PrPSc samples generated by Deleault et al. in sPMCA reactions that were initially seeded with mouse prion strains distinct from the OSU strain (301C- and ME7-seeded cofactor PrPSc) (Fig 2B). The results revealed that the PK-resistant cores of all three cofactor PrPSc molecules have nearly identical solvent accessibility profiles (Fig 2A and 2B), consistent with convergence into a single cofactor PrPSc conformation. As was the case with OSU-seeded cofactor and protein-only PrPSc, we assessed the contribution of non-specifically aggregated, protease-sensitive PrP to the DXMS data for 301C- and ME7-seeded cofactor PrPSc by determining sample conversion efficiency (S2 Fig) and performing a mock-seeding experiment (S3 Fig, bottom panel). 301C-seeded cofactor PrPSc is almost entirely converted to PK-resistant PrPSc (99% PK-resistant conversion efficiency, S2 Fig), ruling out any significant contribution of non-specific PrP aggregation to the presented DXMS data. Mock-seeding of cofactor-supplemented PMCA reactions resulted in the recovery of approximately 8% of the starting material as non-specifically aggregated, protease-sensitive PrP (S3 Fig, bottom panel, samples S0 vs P3). For ME7-seeded cofactor PrPSc (82% PK-resistant conversion efficiency, S2 Fig), we therefore estimate that such non-specifically aggregated PrP accounts for no more than 2% of the sample analyzed by DXMS (ie. 8% of the unconverted, PK-sensitive material yields the estimate for non-specifically aggregated, PK-sensitive PrP, or ~1.4% of the total input PrP, which is then divided by the sum of the PK-resistant and PK-sensitive insoluble PrP, or ~83.4% of the total input PrP).
The data from cofactor PrPSc and protein-only PrPSc sample replicates were aggregated, yielding 63 shared peptides for which individual deuteration curves could be plotted (S6 and S7 Figs). Interestingly, peptides that include residues N-terminal to G89 were less frequently recovered and showed irregular deuteration profiles specifically in cofactor PrPSc samples (S6 and S7 Figs).
Having identified the α1-β2 domain by DXMS as a region of conformational divergence in our cofactor and protein-only PrPSc samples, we sought to confirm this finding using additional biochemical and biophysical approaches. The α1-β2 region contains a portion of the epitope for 15B3, a well-characterized PrPSc-specific conformational antibody [18]. The regions of PrP primary structure that comprise the discontinuous 15B3 epitope are shown schematically in Fig 2A. In a single immunoprecipitation experiment, 15B3 efficiently pulled down all three of our cofactor PrPSc samples (Fig 3, top three panels), but only weakly bound protein-only PrPSc (Fig 3, bottom panel), indicating a disruption of the 15B3 conformational epitope, consistent with our DXMS results.
We further sought to confirm a conformational difference between cofactor and protein-only PrPSc in the α1-β2 domain using Raman spectroscopy (Fig 4). Analysis of the Raman spectra acquired from these two conformers identified multiple Raman shifts that could be assigned to tyrosine residues, which are plentiful in the PrP C-terminus and specifically enriched in the α1-β2 domain (6 of 11 total C-terminal tyrosines). By Raman spectroscopy, protein-only PrPSc appears to contain more exposed tyrosine residues than cofactor PrPSc as evidenced from the increased ring ν(C = C) intensity at ~1620 cm-1 (Fig 4, left panel), the 850 cm-1/830 cm-1 ratio being greater than 1 (Fig 4, right panel) [22], and the increased ring ν(CH) intensity at ~3075 cm-1 (S8 Fig, left panel) [23,24], consistent with our DXMS results. In addition, and also consistent with our DXMS data, protein-only PrPSc appears to contain more exposed CNH groups than cofactor PrPSc as indicated by the increased intensity in the 1530–1580 cm-1 Amide II region, corresponding to ν(CN) and δ(CNH) Raman shifts (S8 Fig, right panel), as well as an increased ν(CN) intensity at ~3300 cm-1 (S8 Fig, left panel). These exposed CNH groups likely originate from the 4 exposed arginine (R) and single exposed glutamine (Q), asparagine (N) and tryptophan (W) residues in the α1-β2 domain, or from CNH-containing side chains in the N-terminal portion of the PK-resistant PrPSc core (residues ~91–115).
Our DXMS data suggests that structural differences between cofactor and protein-only PrPSc are limited to specific domains (Fig 2A) and that these structural differences affect the ability of recombinant PrPSc to convert native PrPC (Fig 1B). As conversion substrates, α-PrP and native PrPC share the same primary sequence, but PrPC also contains bulky N-linked glycans and a GPI anchor as post-translational modifications. Therefore, we hypothesized that limited conformational differences might dramatically alter a recombinant conformer’s infectious activity by impinging on spatial regions that would be occupied by N-linked glycans or a GPI anchor should PrPC adopt the same conformation. To test this hypothesis, we partially purified native PrPC and performed enzymatic deglycosylation with PNGase F. The resulting PrPC molecules, which uniformly contain a GPI anchor as the sole post-translational modification, were used as the substrate in sPMCA experiments seeded with cofactor and protein-only PrPSc (Fig 5 and repeated in S9 Fig). Like brain-derived prions, recombinant cofactor PrPSc was able to template the conversion of unglycosylated PrPC (Fig 5 and S9 Fig, middle and right sample groups), whereas protein-only PrPSc failed to function as an autocatalytic seed in the same substrate (Fig 5 and S9 Fig, left sample group). Note that initial conversion of unglycosylated PrPC substrate molecules to a PK-resistant form was detected after seeding with both cofactor and protein-only PrPSc and 24 h of intermittent sonication, as indicated by the appearance of PK digestion products of slightly higher apparent molecular weight than the respective input seeds (Fig 5 and S9 Fig, left and middle sample groups, round 1 vs 0). In the case of protein-only PrPSc, this higher molecular weight product is diluted out in proportion to the initial seed (Fig 5 and S9 Fig, left sample group, round 2 vs 1), suggesting stoichiometric, as opposed to autocatalytic, PrP conversion. In contrast, the higher molecular weight band generated from infectious PrPSc seed was able to propagate independently of the initial PrPSc seed, suggesting autocatalytic PrP conversion (Fig 5 and S9 Fig, left and middle sample groups, rounds 1–3). This result indicates that the existence of bulky N-linked substrate glycans is not solely responsible for the failure of recombinant protein-only PrPSc to function as a seed for native PrPC. Moreover, by isolating the effect of a single substrate post-translational modification on recombinant PrPSc function, this result provides proof of principle that such modifications may impair the efficient replication of certain recombinant PrPSc conformers in vivo.
A misfolded conformer of the prion protein, PrPSc, is an essential, and possibly the sole, component of infectious prions [1]. Although PrPSc is known to be rich in β-sheet [25–27], a high-resolution structure of this PrP conformer is lacking, and the structural features of PrPSc that determine its infectious activity remain obscure.
Recently, Deleault et al. used a chemically-defined, minimal PrP conversion system and identical seeding material to generate two distinct, autocatalytic recombinant PrPSc conformers that differ >105-fold in their specific infectivity [10]. One recombinant conformer, produced in the presence of PE cofactor molecules and termed cofactor PrPSc, had a titer in normal C57BL mice nearly equivalent to that of brain-derived PrPSc (2.2 x 106 LD50 U/μg PrP), while the other conformer, produced in the absence of cofactor molecules and termed protein-only PrPSc, failed to cause disease in the same host, even at the most concentrated dose tested. In the present study, we have taken advantage of the uniquely controlled opportunity presented by these two PrPSc conformers, which share the same origin and autocatalytic behavior but differ strikingly in their biological activity, in order to investigate the structural and functional determinants of recombinant PrPSc infectivity.
By comparing the structures of these two conformers using DXMS, we find that cofactor and protein-only PrPSc molecules have remarkably similar solvent accessibility profiles within the PK-resistant PrPSc core (Fig 2A). Indeed, they are virtually indistinguishable by this measure in the domains that correspond roughly to α2-α3 in α-helical PrP (~ residues 165–230), and to the stretch of residues between the hydrophobic domain and the start of α1 (~118–143) (Fig 2A), both displaying significant protection from solvent exchange, consistent with widespread conversion to β-sheet secondary structure. The degree of solvent protection, especially N-terminal to β2, distinguishes cofactor and protein-only PrPSc from synthetic PrP amyloids [27–31], and is most consistent with previous DXMS studies using brain-derived and recombinant prions [27,32,33]. Like these recent DXMS studies, the data from the present study are consistent only with those models of PrPSc structure that involve a complete refolding of the PrPC C-terminal α-helices to β-sheet—for example, the parallel in-register β-sheet architectures proposed by Cobb et al. [34] and more recently by Groveman et al.[35]. Interestingly, the modest increase in solvent accessibility seen at residues 187–196 in all four PrPSc conformers studied here (Fig 2) corresponds well with the proposed loop of the native disulfide hairpin predicted by both of these models.
In addition to broad similarities between cofactor and protein-only PrPSc in solvent accessibility, we have identified two specific domains in which the cofactor and protein-only PrPSc conformers can be conformationally distinguished: most clearly within the domain that corresponds to α1 through the C-terminus of β2 in PrPC (~ residues 144–163), but also within a domain at the N-terminus of the PK-reistant core, comprising residues ~91–115 (Fig 2A). In both of these domains, protein-only PrPSc appears to be more exposed to solvent exchange than cofactor PrPSc. Given the role PE cofactor molecules play in the formation of cofactor, but not protein-only, PrPSc [10], it is likely that the relatively solvent-protected structural features selectively associated with cofactor PrPSc are cofactor-induced. Moreover, the fact that the PK-resistant cores of all three cofactor PrPSc samples, derived from distinct prion strains but propagated in the same chemically-defined system, are highly similar (Fig 2A and 2B) suggests that the convergence of biological strain properties observed by Deleault et al. [10] is associated with a convergence of PrPSc structure.
The α1-β2 domain, which appears to adopt different conformational states in cofactor and protein-only PrPSc (Figs 2–4), has previously been identified as a region of PrP that has important implications for PrP misfolding. For example, several PrPSc-selective conformational antibodies are known to have epitopes that reside within this domain [18,36], and small deletions towards the C-terminus of this domain produce PrP molecules that do not readily form PrPSc and, in fact, function as dominant-negative inhibitors of PrPSc replication in full-length PrP substrates [37]. Moreover, in the complete absence of the α1-β2 domain, a redacted ‘miniprion’ is capable misfolding to form PrPSc, but does not cause disease when inoculated into animals expressing full-length PrP [38,39]. Interestingly, the α1-β2 domain lies adjacent the β2-α2 loop, a region of PrP known to play an important role in prion formation and interspecies prion transmission [40–43].
Less is known about the second domain (amino acids ~91–115) in which cofactor and protein-only PrPSc appear to adopt different conformational states (Fig 2 and S5 Fig). This domain includes the so-called ‘fifth site’ for Cu2+ binding [44–46].
It is important to acknowledge that any comparisons drawn between DXMS results in the present study and the activities of cofactor and protein-only PrPSc are correlations only, and that it is possible the structural features underlying PrPSc-associated activities such as autocatalysis and infectivity may be subtle and/or beyond the resolution of the DXMS approach. Similarly, it should be acknowledged that DXMS provides a measure of the average deuterium incorporation at a given amide proton position over a population of PrPSc molecules. It has been proposed that PrPSc exists as a heterogeneous conformational mixture of so-called quasi-species [47], but it is not known—for the recombinant PrPSc conformers studied here, or any other PrPSc preparation—what proportion of PrPSc molecules exhibit autocatalytic or infectious activity. Therefore, the solvent accessibility profiles obtained in this study are representative of conformational differences at a PrPSc population level, and may not be representative of rare, and potentially biochemically/biologically active, components within a given PrPSc sample. In addition, it should be made clear that when interpreting ribbon diagrams, as in Fig 2, similar solvent accessibility profiles do not guarantee similar tertiary and/or quaternary structures, although by using a set of matched peptides with dense, overlapping coverage of the region of interest (Fig 2C) to compare the solvent accessibility of the cofactor and protein-only PrPSc conformers, we have increased confidence in such an interpretation of the data. Finally, we do not claim that a relatively solvent inaccessible conformation of either of the two structurally divergent domains identified in the present study is required for PrPSc infectivity generally. The data presented here are specific to the situation in which seed and substrate are both of the mouse PrP sequence, and it is possible that different PrP sequences may have different infectivity-associated conformational spaces. Consistent with a sequence-specific interpretation of our results, it has previously been shown that native mouse prions also appear to have a relatively solvent-protected structure in the α1-β2 domain [27], while a recent study showed that this same domain is relatively solvent exposed in native human prions [33].
The dissociation of in vitro autocatalytic activity and infectivity seen in some, but not all, recombinant PrPSc conformers [10,14,15] presents an interesting functional question: why is it that a PrPSc conformer capable of self-replication in recombinant PrP substrate fails to function as a template for native PrPC conversion in vivo? One obvious possibility relates to substrate complexity: although α-PrP and native PrPC share similar secondary structures [21,48], native PrPC is a more complex conversion substrate due to the post-translational addition of N-linked glycans and a GPI anchor, and due its location within the membrane environment of cells. Using our cofactor and protein-only recombinant PrPSc seeds as a controlled pair, we performed in vitro conversion experiments in which we, in a step-wise manner, modified native PrPC substrate to make it more and more like α-PrP in an effort to identify the factor(s) that prevent protein-only PrPSc from converting PrPC in vivo. Remarkably, the functional difference between cofactor and protein-only PrPSc persisted even after extracting native PrPC from the membrane environment (Fig 1B and [10]) and removing all N-linked glycans (Fig 5 and S9 Fig), suggesting that neither of these factors is responsible for preventing protein-only PrPSc from converting native PrPC in vivo. Unfortunately, the complementary experiment, in which the GPI anchor of PrPC is selectively removed and that modified PrPC substrate used in conversion reactions is not possible for two reasons: 1. Delipidated PrPC is a poor conversion substrate, even for sPMCA experiments seeded with native prions [49], and 2. Detection of delipidated GPI-anchored proteins by Western blotting is technically challenging [50]. Nevertheless, we infer from the available data that the functional difference between cofactor and protein-only PrPSc is mostly likely attributable to differing abilities of these two conformers to template the conversion of wild-type mouse PrP substrates containing a GPI anchor. Interestingly, Kim et al. have previously demonstrated that the GPI anchor plays an important role in the in vitro formation of native PrPSc conformers [49].
To integrate the results of our structural and functional comparison of cofactor and protein-only PrPSc, we propose a model to account for the striking variation in specific infectivity observed for recombinant PrPSc conformers [9–15] (Fig 6). In this model, the PrP polypeptide backbone, represented by recombinant PrP, is capable of adopting a wide variety of autocatalytic conformations. However, post-translationally modified, native PrPC can only adopt a subset of these conformations, and this subset represents the infectious recombinant PrPSc conformers. From previous studies of PrP and other proteins, there is evidence to suggest that post-translational modifications can alter or restrict protein folding pathways. Indeed, N-linked glycans are known broadly to have chaperone-like effects and to contribute to protein stability [51]. In the case of PrP, it has been observed that the presence of glycans can alter the rate of amyloid fibril formation [52,53], and that by restricting available PrP glycoforms it is possible to control prion strain susceptibility [54]. Whether GPI anchors play a role in protein folding or misfolding is less clear. The loss of the GPI anchor of PrP (which is concomitant with a significant reduction in PrPC glycosylation in vivo [55]) appears to have only a modest effect on PrPSc structure [56], but substantially alters the biochemical properties of PrPSc and promotes the formation of fibrillar aggregates [57,58] and interferes with PrPSc replication in vitro [49]. Interestingly, the co-expresssion of anchorless and wild-type PrPC molecules in vivo appears to enhance host susceptibility to recombinant PrP amyloid fibrils [59], while the experimental addition of a GPI anchor to the amyloidogenic yeast protein, Sup35p, prevents the formation of fibrillar structures, leading instead to the formation of PrPSc-like, non-fibrillar aggregates [60]. While the data from the present study specifically point to a role for the GPI anchor of native mouse PrPC in restricting the range of recombinant PrPSc conformers that possess infectious activity, we cannot exclude the possibility that N-linked glycans also influence the infectivity-associated recombinant PrPSc conformational space. In fact, we speculate that the boundaries of this subset of the conformational space are likely to be highly context-dependent, determined by a complex interplay between the polypeptide sequence of a given PrPC substrate molecule and all of its associated post-translational modifications.
In conclusion, we report for the first time a structural and functional comparison between two autocatalytic recombinant PrPSc conformers that share the same origin and biochemical behavior, but differ >105-fold in infectious titer for wild-type mice. Based on our findings, we suggest that those autocatalytic recombinant PrPSc conformers which are also highly infectious contain specific structural features in a limited number of PrP domains, and that these features may be required in order to accommodate specific substrate post-translational modifications during native PrPC misfolding in vivo.
Cofactor and protein-only PrPSc [10]were generated by sPMCA as described [10]. Briefly, 100–200 ul reactions containing 6 μg/ml recombinant mouse PrP 23–230 (recombinant PrP) in conversion buffer [20 mM Tris (pH 7.5), 135 mM NaCl, 5 mM EDTA (pH 7.5), 0.15% Triton X-100] were supplemented with either brain-derived cofactor [10] for cofactor PrPSc propagation, or water for protein-only PrPSc propagation. Reactions were seeded with 1/10th volume of converted cofactor or protein-only PrPSc PMCA product and sonicated with 15-s pulses every 30 min for 24 h at 37°C. After 24 h of PMCA, 1/10th volume of the reaction was used to seed fresh substrate cocktail and the 24 h sonication program was repeated. PMCA reactions were sonicated in microplate horns using a Misonix S-4000 power supply (Qsonica) set to amplitude 50–60. Sample tubes were sealed with Parafilm (Bemis Company) and the sonicator horn was soaked in 100% bleach prior to switching to propagation of a different PrPSc species in order to prevent cross-contamination.
Formation of cofactor and protein-only PrPSc was monitored by digestion of PMCA samples with proteinase K (PK) and Western blotting. Samples were treated with 25 μg/ ml PK (Roche) for 30 min at 37°C, and digestion reactions quenched by the addition of SDS-PAGE loading buffer and heating to 95°C for 10 min. SDS-PAGE and Western blotting were performed as described previously [10] using mAb 27/33, unless otherwise specified.
All centrifugation was done at 4°C. Cofactor and protein-only PrPSc PMCA products were treated or mock-treated with PK as described above and the reaction quenched by the addition of PMSF to 5 mM final concentration. To remove PK, excess lipids, and any soluble recombinant PrP digestion products, digested PrPSc was washed twice with nOG wash buffer (1% n-octyl-beta-D-glucopyranoside (nOG), 150 mM NaCl, 8.3 mM Tris pH 7.2) by centrifugation at 100,000 rcf for 1 h, with resuspension by sonication (60-s pulse, 70 amplitude) followed by brief vortexing. After the second wash, samples were pelleted by centrifugation at 100,000 rcf for 1 hr and resuspended in conversion buffer by sonication and vortexing to a recombinant PrP concentration of 6 μg/ml for use in epitope mapping and sPMCA experiments.
Epitope mapping of the cofactor and protein-only PrPSc PK-resistant cores was performed using mAbs 6D11 [61] and R2 [62]. Aliquots of identical purified samples were run on SDS-PAGE, transferred to PVDF membrane, and processed independently by Western blotting using trays and containers that had never been in contact with anti-PrP antibodies.
For normal brain homogenate sPMCA, brain homogenates were prepared at 10% (w/v) in conversion buffer (PBS containing 1% (v/v) Triton X-100 and cOmplete protease inhibitor cocktail (Roche)) from healthy C57BL/6 mice, as described by Castilla et al. [5]. Homogenates were clarified by brief sonication (F60 Sonic Dismembrator (Fisher Scientific), three 5-s pulses at ~1.8 amplitude), followed by centrifugation at 500 rcf for 15 min. Clarified substrates were then seeded with 1/10th volume of purified PrPSc samples, and sPMCA carried out as described above, with 20-s sonication pulses every 30 min.
For deglycosylated PrPC sPMCA, diglycosylated native PrPC was first purified from normal mouse brain and then fully deglycosylated by treatment with PNGase F (New England Biolabs) as described [63]. Deglycosylated PrPC sPMCA reactions [63] were seeded with 5% (v/v) of recombinant or brain-derived PrPSc and sonicated as described above for normal brain homogenate sPMCA.
All centrifugation was done at 4°C. Samples were prepared for deuterium exchange as described previously [32] with the following modifications. For each PrPSc species, converted PMCA cocktail was washed twice with nOG wash buffer (1% n-octyl-beta-D-glucopyranoside (Anatrace), 150 mM NaCl, 8.3 mM Tris pH 7.2) by centrifugation at 100,000 rcf for 1 h, with resuspension by 60 s of sonication at 70 amplitude followed by brief vortexing. After the second wash, samples were pelleted by centrifugation at 100,000 rcf for 1 hr and resuspended in a volume of mock labeling buffer (150 mM NaCl, 8.3 mM Tris, pH 7.2) by sonication and vortexing immediately prior to the initiation of deuterium exchange. Deuterium exchange was initiated by the addition an equal volume of labeling buffer (D2O containing 150 mM NaCl, 8.3 mM Tris, pH* 7.2, where pH* is the pH meter reading without taking into account the hydrogen isotope effect) and samples were incubated at room temperature (22°C). During the last 30 min of labeling, samples were pelleted by centrifugation at 100,000 rcf. Labeling buffer was removed and the samples were quenched on ice for 2 min with ice-cold quench buffer [0.8% formic acid, 6.4 M guanidine hydrocholride, 150 mM tris(2-carboxyethyl)phosphine (TCEP) (Pierce)]. Quenched samples were diluted with 3 volumes of ice-cold acid diluent (0.8% formic acid, 16.6% glycerol), transferred to chilled autosampler microvials, frozen on crushed dry ice, sealed and stored at -70°C until analysis. For all PrP samples, including those described below, Western blotting of quenched material was performed and confirmed the presence of 1–2 μg recombinant PrP per sample vial.
Deuterium exchange and quenching of normally folded, α-helical recombinant PrP (α-PrP) was performed as described above for cofactor and protein-only PrPSc, with the following modifications. Recombinant PrP was resuspended to 1.0 mg/ml in water and an equal volume of 2x labeling buffer (D2O containing 300 mM NaCl, 16.6 mM Tris, pH* 7.2) was added to initiate deuterium exchange. Thirty minutes prior to quenching, an aliquot of the labeling reaction was placed at 4°C to replicate the temperature change experienced by PrPSc samples during centrifugation. Deuterium-labeled α-PrP was then quenched on ice for 2 min by the addition of 1.25 volumes of ice-cold quench buffer containing 700 mM TCEP. To the quenched samples was added 1.30 volumes of ice-cold acid diluent prior to aliquoting into autosampler microvials and freezing on dry ice, as described above.
Equilibrium-deuterated samples were prepared by resuspension of recombinant PrP in 2.5 or 6.0 M guanidine hydrochloride solution containing a 1:1 molar ratio of protons:deuterons by mixing appropriate quantities of H2O, D2O, guanidine HCl and guanidine (D6) DCl (Cambridge Isotope Laboratories, Andover, MA). Deuterium exchange was allowed to proceed for 72 hours at room temperature (22°C) prior to quenching as described above for α-PrP.
To estimate the fraction of non-specifically aggregated PrP that could potentially co-sediment during the purification of PrPSc samples for DXMS, mock-seeded PMCA reactions were performed using PMCA cocktail supplemented with brain-derived cofactor or water. After 24 h of PMCA, the mock-seeded PMCA reactions were purified by ultracentrifugation as described above for DXMS samples and the fraction of the input PrP that was recovered as non-specifically aggregated, insoluble PrP was quantified by Western blot.
Measurement of deuterium incorporation by LC-MS was performed as described previously [32], with the following modifications. Samples were loaded onto the in-line immobilized fungal protease XIII column at a rate of 60 μl/min, allowed to digest for 3 min, and then pushed onto the in-line immobilized pepsin column at a rate of 20 μl/min. Peptides were collected during pepsin digestion on a C18 trap column (Michrom MAGIC C18AQ, 0.2x2) preceding the C18 resolving column (Michrom MAGIC C18AQ, 0.2x50). All measurements were made on an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific), and data was analyzed as described previously [32].
Rat anti-mouse IgM-conjugated Dynabeads (Life Technologies) were washed according to the manufacturer’s protocol and coated with mAb 15B3 (Prionics) at 5 μg per 10 μl beads with gentle mixing at room temperature for 2 h. Coated beads were washed three times to remove unbound antibody and stored at 4°C for no more than one week prior to use. Converted cofactor or protein-only PrPSc PMCA cocktail was washed twice with nOG wash buffer as described above for the preparation of samples for DXMS. After the second wash, samples were collected by centrifugation at 100,000 rcf for 1 hr at 4°C and the pellet was gently washed with Prionics homogenization buffer (Prionics). Samples were then centrifuged at 100,000 rcf for 10 min at 4°C and the supernatant discarded. The pellet was resuspended in Prionics homogenization buffer by sonication (30-s pulse, 70 amplitude) and vortexing to a final concentration of ~60 ng/μl PrP. For each PrPSc sample, ~250 ng PrP was added to 0.5 ml Prionics IP buffer (Prionics) and to this was added 10 μl of beads that were either coated with 15B3 or uncoated. Samples were allowed to interact with the beads at room temperature for 4 h with gentle mixing, followed by two washes with Prionics IP buffer and resuspension of the beads in 2x SDS-PAGE loading buffer. Samples were incubated at 95°C for 10 min, briefly centrifuged to concentrate the beads and the supernatant was collected for analysis by Western blot.
Cofactor and protein-only PrPSc were generated by sPMCA supplemented with synthetic plasmalogen PE (Avanti Polar Lipids) as the sole cofactor, as described previously [64]. Converted PMCA cocktail was digested with PK as described above and quenched by the addition of PMSF (Sigma Aldrich) to 2 mM final concentration. Digested samples were washed twice with nOG wash buffer, as described above, and then twice with water to remove residual buffer components and detergent. Samples were then resuspended in water to a concentration of ~140 ng/μl PrP by vortexing and a 15-s sonication pulse. 10 μl of the resulting sample was spotted onto a glass slide and allowed to dry under a stream of nitrogen. Once dry, another 10 μl was spotted on top of the first and again allowed to dry under nitrogen. Spotted samples were scanned using a WITec CRM200 Raman confocal light microscope, equipped with a 100x lens and a 514 nm argon laser with 45 mW output. An f/4, 300 mm imaging spectrograph was employed with 2 exit ports and a 600 lines/mm grating, with a Peltier-cooled CCD, 1340 x 100 pixel format, and a 16-bit camera controller. The fiber optic connecting the microscope with the spectrograph was 50 μm in diameter. Spectra were acquired using an integration time of 8 s, with two hardware and two software accumulations per shot and a spectral resolution of 4 cm-1. Presented spectra are averages of 20–30 shots. In each figure, the baseline was adjusted to zero and data points were joined with a smoothed line in Microsoft Excel. Although spectral normalization was not possible, data were collected with the same instrument at the same time from highly concentrated films of protein, and it can be expected that intensity differences between samples originate in structural differences between conformers.
All experiments involving mice in this study were conducted in accordance with protocol supa.su.1 as reviewed and approved by Dartmouth College’s Institutional Animal Care and Use Committee, operating under the regulations/guidelines of the NIH Office of Laboratory Animal Welfare (assurance number A3259-01).
|
10.1371/journal.pntd.0005705 | Climate change and sugarcane expansion increase Hantavirus infection risk | Hantavirus Cardiopulmonary Syndrome (HCPS) is a disease caused by Hantavirus, which is highly virulent for humans. High temperatures and conversion of native vegetation to agriculture, particularly sugarcane cultivation can alter abundance of rodent generalist species that serve as the principal reservoir host for HCPS, but our understanding of the compound effects of land use and climate on HCPS incidence remains limited, particularly in tropical regions. Here we rely on a Bayesian model to fill this research gap and to predict the effects of sugarcane expansion and expected changes in temperature on Hantavirus infection risk in the state of São Paulo, Brazil. The sugarcane expansion scenario was based on historical data between 2000 and 2010 combined with an agro-environment zoning guideline for the sugar and ethanol industry. Future evolution of temperature anomalies was derived using 32 general circulation models from scenarios RCP4.5 and RCP8.5 (Representative greenhouse gases Concentration Pathways adopted by IPCC). Currently, the state of São Paulo has an average Hantavirus risk of 1.3%, with 6% of the 645 municipalities of the state being classified as high risk (HCPS risk ≥ 5%). Our results indicate that sugarcane expansion alone will increase average HCPS risk to 1.5%, placing 20% more people at HCPS risk. Temperature anomalies alone increase HCPS risk even more (1.6% for RCP4.5 and 1.7%, for RCP8.5), and place 31% and 34% more people at risk. Combined sugarcane and temperature increases led to the same predictions as scenarios that only included temperature. Our results demonstrate that climate change effects are likely to be more severe than those from sugarcane expansion. Forecasting disease is critical for the timely and efficient planning of operational control programs that can address the expected effects of sugarcane expansion and climate change on HCPS infection risk. The predicted spatial location of HCPS infection risks obtained here can be used to prioritize management actions and develop educational campaigns.
| Hantavirus, hosted by rodent species, causes HCPS, a disease with a 50% mortality rate in humans. The conversion of native vegetation to sugarcane increases the abundance of hantavirus reservoir rodent species, augmenting disease risk. Additionally, temperature also has positive effects on disease risk because it affects rodent population and the time the virus remains infectious in the environment. Here we evaluate the impacts of climate change and sugarcane expansion on HCPS risk. Expansion of sugarcane increases average HCPS risk placing 20% more people at risk for acquiring HCPS than under current conditions. Temperature anomalies increase HCPS risk even more and place 31% and 34% more people at risk (RCP4.5 and RCP8.5, respectively). Our results confirm the impacts of climate change and agriculture expansion on disease risk and highlight the need for timely and efficient planning of operational control programs in order to avoid disease propagation in the future.
| Global average temperatures are projected to increase between 1.7 and 4.8°C by the end of this century [1,2], with potential effects on human health, including mortality from extreme heat and cold, and changes in the ecology of infectious diseases [3–5]. Climatic variability and extreme weather events have profound impacts on infectious diseases since fluctuations in temperature and precipitation influence both infectious agents (such as protozoa, bacteria, and viruses) and population dynamics of their vectors (such as mosquitoes, ticks, and rodents) [3, 6–8]. Outbreaks of some diseases such as Ross River virus disease [9], malaria [10], meningitis [11] and Hantavirus Cardiopulmonary Syndrome (HCPS) [12] have been associated with climate anomalies.
At the same time, increasing evidence suggests that land cover and land use change affect disease incidence by altering the interactions, abundance, and movement patterns of hosts, vectors, and people [13,14]. For instance, outbreaks of Hantavirus, Lyme disease and tick-borne encephalitis have been associated not only with climate-related changes in the density of host rodent and tick populations [15–17], but also with shifts in the extent and type of land use [17–23].
Hantavirus (Bunyaviridae) is a virus transmitted by small mammals [17] which causes two syndromes in humans: Hantavirus Cardiopulmonary Syndrome (HCPS), restricted to the Americas, and hemorrhagic fever with renal syndrome (HFRS) present in Eurasia and Africa [24, 25]. HCPS was first identified in 1993 in both the United States and Brazil [26, 27] and exhibits lethality rates as high as 50% [26, 28]. Unlike HFRS, a vaccine is not available for HCPS. Transmission to humans occurs through inhalation of the aerosolized form present in the urine, saliva and feces of infected rodents [29–31].
Climate conditions can influence Hantavirus host population abundance and disease transmission dynamics [32]. A number of studies in arid and semi-arid region of the U.S. have uncovered a positive association between precipitation, population size of rodent hosts and prevalence of Hantavirus [33–36]. Anomalously high precipitation increases vegetation growth, boosting rodent densities and enhancing the probability of encounters between humans and infected rodents and consequently Hantavirus transmission [12, 37]. Temperature can influence rodent abundance and disease risk by altering vegetation growth [38], reproduction and survival rates of small rodents [38–40], as well as the time the virus remains infectious in the environment [41]. The capacity of Hantavirus to survive outside its host plays a critical role in transmission dynamics [41]. High temperatures have been associated with more frequent Hantavirus outbreaks [16, 37, 42, 43], most likely because high temperature leads to greater aerosolization of the virus and higher rates of inhalation by both humans and rodents [29, 31]. There is evidence that variation in temperature, but not precipitation, affect HCPS risk in Brazil [32].
Sugarcane plantations may be also associated with increases in Hantavirus infection risk [19, 32, 44]. Experimental studies have shown that small mammal populations are frequently food-limited [45]; thus the presence of an abundant, highly nutritious food resource, such as sugarcane, with yields as high as 120 tons·year·ha−1 [46], might allow the increase and maintenance of large populations of these species, relative to other land uses, either natural or agricultural [19, 47, 48]. Furthermore, sugarcane offers protective cover for feeding, burrowing and breeding activities, throughout the year [49].
Many developing countries are expanding sugarcane plantation areas to produce biofuel, as a strategy to reduce their dependence on petroleum, to increase opportunities for the agricultural sector, and to mitigate global warming [50]. In Brazil, the creation of the pro-alcohol program, developed to replace a significant percentage of fossil-fuel consumption with ethanol produced from sugarcane [51], was triggered by an increase of 428% in oil prices in 1973 [52]. This program and the recent interest in alternative energy sources have fostered an expansion in the extent of sugarcane cultivation, making the country the world’s leader in ethanol production [53] and sugarcane (~ 490 million tons per year) exports [54]. The majority of this production (~74%) comes from the southeastern region, with the state of São Paulo producing 60% of the total yield [55, 56].
The combined consequences of bio-energy expansion and climate variability and change on Hantavirus infection risk remain unexamined. Understanding how these factors impact infectious disease risk is essential to fully evaluate the actual costs of the biofuel programs and is critical for timely and efficient planning of operational control programs. In this paper we analyze how sugarcane expansion and temperature changes under two climate scenarios can potentially influence HCPS risk in the state of São Paulo by 2050. To do so, we relied on a baseline model we previously developed [32], which evaluated the effects of landscape, climate and social predictors, including historical climate and sugarcane as predictor variables, on Hantavirus risk between 1993 and 2012. In this paper, we used climate and sugarcane data derived from various scenarios (see methods section) to test the independent and combined effect of these two factors on HCPS risk. We hypothesize that HCPS incidence will show an increase under all scenarios because both changes in climate and sugarcane expansion are expected to increase HCPS risk through their positive effects on rodent abundance and virus survival and aerosolization, and that their combined effect can exacerbate their individual impacts.
We focused our analyses on the state of São Paulo, the wealthiest Brazilian state, where HCPS was first identified in 1993 and where the risk of disease increase is particularly high, due to both sugarcane expansion and climate change. São Paulo state is located in southeastern Brazil, in an area of approximately 248,210 km2 (Fig 1), and has a population of about 42 million, representing 21% of the Brazilian population [57].
The probability of Hantavirus infection risk in the state of São Paulo was calculated as a function of landscape, social and climatic factors using a Bayesian model, and is described in detail in [32]. Given that Hantavirus exhibits high host specificity, with each region having different reservoir host species and virus strains [73], Hantavirus transmission risk was modeled separately for Atlantic forest and Cerrado biomes. Although some geographic overlap occurs [74], Araraquara virus (ARAV) is the dominant Hantavirus in Cerrado [74], whereas Juquitiba (JUQV) is the dominant one in Atlantic forest [74]. Municipalities were assigned to Cerrado or Atlantic forest if >50% of their surface area fell inside one of the biome. The biome distribution map was obtained from IBGE (www.ibge.gov.br).
HCPS infection risk was predicted using a Bernoulli distribution and the model (baseline model) contained 7 predictor variables as fixed covariates: proportion of sugarcane, proportion of native vegetation cover, density of native vegetation patches, HDI, mean annual temperature (°C), total annual precipitation (mm), and rural male population >14 years old [32]. Risk was defined as the annual probability of HCPS infection. Municipality was included as a random effect to account for differences among these administrative units that are not captured in the fixed covariates [32]. To facilitate interpretation, all estimated parameters were standardized by centering them on their mean and dividing by two standard deviations [75].
We tested models of raw HPS incidence as well as model residuals for spatial autocorrelation using Moran’s I. We used the spatial contiguity matrix based on the Queen´s case neighborhood relation and treated each year separately. This test is commonly used and accepted as a fair evaluation of spatial autocorrelation and dependence [76], especially in disease studies [77, 78]. For all models and most years, we found no spatial autocorrelation, justifying the use of a non-spatial model.
To evaluate changes in Hantavirus infection risk, estimated probability of HCPS infection under current conditions (baseline model) was compared to the predicted probability under five scenarios: two possible future climate change scenarios (RCP4.5 and RCP8.5), one possible sugarcane expansion scenario, and the combinations of each climate scenarios and sugarcane expansion (RCP4.5 + sugarcane; RCP8.5+ sugarcane) (Table 1).
To estimate the predicted probability of HCPS infection under the five future scenarios we used the parameter estimates from the baseline model [32] and used sugarcane and climate data derived for each scenario. We made the simplifying assumption that the biological relationships governing disease transmission would remain largely unchanged over the estimation period. Uncertainty was measured using lower and upper limits of risk estimates for each scenario, derived from the 2.5% and 97.5% quantiles of the baseline model parameters for sugarcane and temperatures. Results are presented in S3 and S4 Figs. The covariates percent of native vegetation cover, number of patches, total annual precipitation, human development index, and people at risk were assumed to be the same as the covariates from the previous year, available from the baseline model (year of 2012), and were kept constant for the predictions. These are reasonable assumptions considering that trends of urban-rural migration in São Paulo are constant [79], deforestation has been drastically reduced in the state [62] and precipitation is not relevant for HCPS risk [32]. Despite the increase in sugarcane mechanization, manual harvest is still necessary and present in some parts of the process [80]. Additionally, skilled workers are replacing unskilled workers, while temporary workers are still being hired at the same rates as before [81]. The number of people employed in sugarcane areas is not diminishing with sugarcane mechanization.
To obtain a clear view of the probability of change in Hantavirus risk, we created a map with the change in infection risk for each scenario that was calculated using the difference between the current Hantavirus risk and the predicted risk for each scenario. We also used model simulations to generate a map of Hantavirus infection risk for the State of São Paulo for each scenario, where Hantavirus infection risk is classified as small (<5%), medium (≥5 and ≤10%), high (≥10 and ≤ 20%) and extremely high (≥20%). We considered that a municipality with a risk higher than 5% should be a target for preventive measures due to the high disease lethality (maps are shown in supplementary material—S5 Fig).
By associating the estimated probability of HCPS infection risk generated for each scenario (baseline model and the five future scenarios) with the at risk population for each municipality (rural men older than 14 years), we predicted current and future human exposure to HCPS. We also calculated the percent increase in the number of people that could be infected in each scenario, by comparing each scenario with the baseline (Table 2).
According to our sugarcane expansion scenario, this crop cover will increase to ~30% on average in the state of São Paulo by 2050. Sugarcane area will increase from 26% to 34% in the Cerrado region (11,200 to 14,500 ha) and from 23% to 31% in the Atlantic Forest region (8,200 to 11,100 ha) (S6 Fig).
Considering climate change scenarios, there is a general consensus among the 32 models evaluated, for both RCP4.5 and RCP8.5 scenarios, in the direction of the projected temperature change for the state of São Paulo, with both experiments presenting increases. Also, RCP8.5 presents a smaller variation and lower standard deviation between the 32 models analyzed than RCP4.5 models, especially from 2013 to 2050 (Fig 2). After 2050, the anomalies of RCP8.5 become larger than those from the RCP4.5 experiment, showing larger increases in temperature anomalies.
Under current conditions, the state has an average Hantavirus infection risk of 1.3%, with 6% of the municipalities classified as high risk (HCPS ≥ 5%). Hantavirus infection risk increases under all scenarios evaluated (0.25% to 0.37%) (Table 1). Sugarcane expansion is the scenario that predicted the smallest increase in Hantavirus risk, with a 0.25% increase on average. The most pronounced changes are expected to occur in the west and mid-west parts of the state where almost all municipalities exhibit an increase of 1.5% in HCPS infection risk (Fig 3B). Also, sugarcane scenario will lead to a risk greater than 5% for HCPS for about 6.6% of all municipalities (43 municipalities).
Projected temperature anomalies for both climate change scenarios predicted similar average increases of HCPS in the state (0.35% for RCP4.5 and 0.37% for RCP8.5), with larger increases concentrated in the northeast region, but with RCP4.5 predicting smaller increases than RCP8.5 (Fig 3C and 3D). Moreover, there is a significant increase in the risk of infection for some municipalities that already had a high risk, especially in the mid-west region, with HCPS risk reaching 52.3% in RCP4.5 and 52.7% in RCP8.5. Under the RCP4.5 scenario there were 42 municipalities (~6.5% of the state) with a HCPS risk greater than 5%, while under the RCP8.5 scenario there were 44 municipalities (~6.9% of the state). The HCPS risk simulated using the temperature anomalies ± 1 standard deviation showed that the uncertainty of models simulations is small, and that disease risk is similar for all three experiments (mean temperature anomalies, mean temperature anomalies + 1 standard deviation and mean temperature anomalies—1standard deviation). Therefore, our confidence interval of predictions due to temperature change analysis is narrow, showing similar and consistent trends (S2 Fig). Additionally, the confidence interval of risk estimates for most of municipalities is narrow, except for some municipalities, where upper limits are high (S3A and S3B Fig).
When combining climate change scenarios and sugarcane expansion, the average increase and the maximum HCPS risk for the state is the same as under the climate change scenarios alone (RCP4.5 and RCP8.5), showing that there is no additional effects between temperature and sugarcane. However, the increase in Hantavirus risk became more homogeneous throughout the state when considering the combined sugarcane-climate change (Fig 3E and 3F), with the inclusion of ~7% of the municipalities of the state with HCPS risk greater than 5%.
When we consider the number of people that can be infected by HCPS (based on the number of population at risk), the sugarcane expansion scenario alone presents an increase of 20%. For RCP4.5 and RCP8.5 temperature scenarios alone and combined with sugarcane expansion the number of people is the same, presenting an increase of 31% and 35% respectively (Table 2).
In accordance to our predictions, sugarcane expansion and rising temperature will lead to increases in HCPS risk but with relatively weak effects on average. Our results suggest that climate change effects will be more severe than those from sugarcane expansion, and surprisingly, there was no evidence of additive effects of sugarcane and climate on HCPS risk for the state.
The effects of sugarcane expansion and temperature anomalies on HCPS risk were smaller than initially expected, which may have occurred because transmission to humans is complex and involves a number of factors that are not yet fully understood, especially in the tropics. Hantavirus infection rates and prevalence in rodent populations are generally low [34]. Transmission increases in density-dependent fashion with greater intraspecific encounter rates and virus transmission at high rodent densities [7]. Consequently, the virus load in the environment and the human risk of acquiring HCPS also increases at high rodent densities [82]. However, high abundances of reservoirs alone do not guarantee that humans will become infected. To acquire HCPS, human exposure to infected rodents is also necessary, with disease transmission resulting from a combination of human behaviors (i.e., inadequate storage of grains and lack of protective measures) and, density and prevalence of reservoirs. Climate also affects virus survival and aerosolization in the environment [41]. Disease transmission to humans requires that these four main factors interact: an infected rodent; a certain abundance of reservoir rodents to proliferate the infection throughout the rodent population (in which prevalence is generally low); suitable climatic conditions in order to maintain the virus in the environment and allow its aerosolization, and a susceptible human population. Due to these complex dynamics, HCPS transmission to humans is difficult and can be considered as a rare event, with a low number of cases reported each year. However, this increase is extremely relevant given the high lethality rate of the HCPS, which is around 50% in the state.
Our results confirmed previous studies showing that increases in the amount of sugarcane can augment HCPS risk [32, 44, 83]. Our scenario predicted an increase in ~6,000 ha of land occupied with sugarcane on average, for both the Cerrado and Atlantic forest regions, until 2050, forecasting increases of ~15% in HCPS risk. This expansion can be considered small, since the area planted with sugarcane in the state has tripled from 1990 to 2010, increasing from 3,000 to 9,000 ha on average for the entire state [56, 84]. Over this same time period HCPS risk in the state has also increased almost four times (382%) from 0.34 to 1.3%. Therefore, the increase in disease risk, predicted by our model and according to the expansion of sugarcane, is concordant with the historical increase in risk experienced from 1993 to 2012.
This increase in disease risk, without any change in temperature, is on average low, but can be as high as 6.6% in some municipalities. Overall our sugarcane expansion scenario predicts an increase of 20% in the number of people that can acquire HCPS. The main underlying mechanism to explain this pattern is that sugarcane provides a highly nutritious food, leading to increased recruitment and a rapid population growth of rodents [85]. Sugarcane plantations are a suitable habitat for these generalist rodent species, as Hantavirus reservoirs, supporting greater abundances of rodents than other ecosystems, whether natural or agricultural [47], with sugarcane becoming a predominant part of their diets [86].
Land-use changes also indirectly influence local temperature [87] and alter albedo and evapotranspiration, which can directly influence local climate [88]. Sugarcane plantations have cooler temperatures and more moisture than pasture and other crops, being micro-climatically more similar to areas of natural vegetation near to the soil level [87]. This microclimate changes may contribute to the increase in HCPS risk, since it makes sugarcane an even better habitat for rodents. This climate aspect can also affect the indirect path of transmission, extending the time the virus remains infectious in this environment and augmenting HCPS risk, since virus inactivation happens only in dry conditions and above 37°C [41].
Climate change scenarios predict larger increases in HCPS risk when compared with sugarcane expansion alone. Increases in temperature may be more important than sugarcane expansion, because temperature interacts with disease transmission through multiple mechanisms. Temperature positively affects vegetation growth [36, 89], leading to increases in the abundance of reservoir rodent species, since small mammal populations are food-limited [45]. Temperature also affects reproduction and survival of small rodents [36, 40], which may have a positive or negative effect depending on the magnitude of temperature change [36, 90]. In addition, temperature directly affects process of HCPS transmission, determining virus survival and aerosolization in the environment [41].
There is a lack of studies involving reservoir rodent species and Hantavirus related to HCPS and climate variables, but for HFRS, mild temperatures (10–25°C) are most favorable for breeding of reservoirs rodents [91] and for the time the virus remains infectious in the environment [41]. Increases in temperature lead to greater aerosolization of the virus and higher rates of inhalation by both humans and rodents [92]. In this way, increases in temperature may have a positive effect on reservoir rodent abundance and virus survival and aerosolization until reaching a certain threshold (around 40°0°CC) from where temperature will exert a negative effect.
The relatively low magnitude of the effect of climate change on HCPS risk in our study maybe explained by the fact that temperature anomalies, until 2050, can be considered small and similar for both RCP4.5 and RCP8.5, with larger increases being observed only after 2050. Nevertheless, it is important to highlight that increases in temperature anomalies lead to increase in HCPS risk, though small, in all the 645 municipalities of the state of São Paulo. Therefore, higher increases for disease risk are expected after 2050 if carbon emissions are not controlled and climate change mitigation actions are not successful.
Individual evaluation of climate and sugarcane could have resulted in a better understanding of the individual contributions of each factor on disease risk. However, evaluating these scenarios together is a more realistic approach, given that they will occur and act together. Sugarcane expansion and temperature anomalies together showed no additionality, predicting the same average increase in HCPS risk when compared to climate change scenarios alone. This may have happened because there are multiple mechanisms through which temperature influences HCPS risk, some of which overlaps sugarcane mechanisms (i.e., effect on rodent densities). Particularly, even in conditions where rodent abundances and prevalence are high, if temperature conditions are not ideal for virus survival and aerosolization, transmission to humans will not occur. Therefore, the ability of the virus to survive outside the host is critical for the transmission within rodent populations and to humans, with temperature being one of the determining factors of this survival. This effect, may have contributed to the lack of additionality between temperature and sugarcane, since sugarcane effects will only occur when temperature conditions are also adequate.
Land cover and land use change are at the origin of the outbreaks of Hantavirus and can also be an important component to reduce or mitigate its spread. Given that temperature increase will lead to increases in HCPS risk, forest restoration can be an alternative to attenuate the effects of higher temperature on HCPS risk for three main reasons. First, forest regrowth, especially in tropical regions, can sequester atmospheric carbon, absorbing about 30% of all CO2 emissions from fossil fuel burning and net deforestation [88, 93], contributing to climate change mitigation. Second, forest regeneration can mitigate the creation of warmer and drier climates in agricultural systems [88], reducing the ideal conditions for hantavirus survival. Third, increasing forest cover could also reduce HCPS risk arising from sugarcane expansion, since it would lead to increased suitable habitat for habitat specialist species, leading to a more diverse community, with decreased abundance of habitat generalist species [94], such as hantavirus reservoir species.
We note that the use of ethanol from sugarcane as a gas substitute leads to a very important reduction in greenhouse gases emissions, which can reach up to 85% [95, 96]. This is mostly due to the fact that they replace fossil fuels [97], and sequester carbon through the growth of the feedstock [98, 99], especially when pastures are converted to sugarcane fields and managed without fire [95]. The use of ethanol from sugarcane, produced in landscapes with a large amount of connected forest cover, could ameliorate disease risk, since it would increase diversity community, diminishing reservoir rodent abundance, and would contribute to climate mitigation.
Public health costs will also increase under the expected increase in temperature and sugarcane expansion. At least part of these costs should be factored into sugarcane production, including expenditures associated with rodent control, education and preventive campaigns targeting how to avoid virus inhalation and contact with infected rodents excreta. These control measures are likely to yield additional benefits since rodents are considered major pests of this crop [86], leading to a loss of 825.000 tons of sugarcane in one year in India [100]. States and municipalities considering sugarcane expansion should also plan for costs involved with educational campaigns and preventive measures, for example, educating workers and residents from rural areas about how to avoid Hantavirus inhalation and contact with infect rodents excreta. This could be crucial to avoid disease propagation to places where HCPS risk is currently low or absent. This type of information should be incorporated into the costs of land use management. Sugarcane expansion can provide a solution to one specific problem, such as supplying the oil market, but can on the other hand create a human health problem by increasing risks of acquiring HCPS.
Our results reinforce the links between climate change and rises in incidence of diseases, such as Lyme, West Nile Virus and Echinococcus [101–103]. These findings should be considered as an additional argument to encourage governments, companies and citizens to sign agreements and start massive campaigns in order to mitigate climate change impacts.
Our scenarios of future sugarcane expansion and climate change RCP4.5 and RCP8.5 predicted a low but significant increase in HCPS risk in the state of São Paulo by 2050. Despite the lack of additive effects of sugarcane and climate in HCPS risk, we suggest that prevention and mitigation actions should focus on land use planning and forest restoration programs, and by concentrating healthcare effort in areas that are predicted to be at higher HCPS risk and have a high variation in the confidence interval.
To better explore the underlying mechanisms of the observed pattern, we suggest future studies should test the effects of sugarcane production, temperature, and moisture on reservoir rodent population dynamics and on virus survival and aerosolization. Understanding those relationships is crucial to better understand HCPS transmission dynamics in different environments and situations, which is important for the effective design of preventive health strategies.
|
10.1371/journal.ppat.1000538 | Salmonella Typhimurium Type III Secretion Effectors Stimulate Innate Immune Responses in Cultured Epithelial Cells | Recognition of conserved bacterial products by innate immune receptors leads to inflammatory responses that control pathogen spread but that can also result in pathology. Intestinal epithelial cells are exposed to bacterial products and therefore must prevent signaling through innate immune receptors to avoid pathology. However, enteric pathogens are able to stimulate intestinal inflammation. We show here that the enteric pathogen Salmonella Typhimurium can stimulate innate immune responses in cultured epithelial cells by mechanisms that do not involve receptors of the innate immune system. Instead, S. Typhimurium stimulates these responses by delivering through its type III secretion system the bacterial effector proteins SopE, SopE2, and SopB, which in a redundant fashion stimulate Rho-family GTPases leading to the activation of mitogen-activated protein (MAP) kinase and NF-κB signaling. These observations have implications for the understanding of the mechanisms by which Salmonella Typhimurium induces intestinal inflammation as well as other intestinal inflammatory pathologies.
| Salmonella Typhimurium is one of the most common causes of food-borne illness in the United States and a major cause of diarrheal diseases in developing countries. This pathogen induces diarrhea by stimulating inflammation in the intestinal tract. This study shows that S. Typhimurium delivers molecules into epithelial cells with the capacity to stimulate the production of pro-inflammatory substances. This mechanism may help the pathogen to initiate the inflammatory response in the intestinal epithelium. This study provides insight into the mechanisms by which Salmonella Typhimurium causes diarrhea.
| It is widely believed that one of the main triggers of host inflammation is the recognition of microbial products by receptors of the innate immune system [1]–[3]. Conserved microbial products, collectively known as “pathogen-associated molecular patterns” (PAMPS), stimulate specific host receptors triggering a variety of generic responses directed at controlling pathogen spread. In the case of bacterial pathogens, conserved bacterial products such as lipopolysacharide, flagella, or peptidoglycan, are recognized by transmembrane Toll-like receptors (TLRs), or intracellular nucleotide oligomerizaton domain-like receptors (NLRs), leading to conserved signaling cascades that culminate in the activation of NF-κB, mitogen-activated protein kinases (MAPK), and the production of pro-inflammatory cytokines [4]–[7]. These receptors are widely expressed in cells of the innate immune system such as macrophages, dendritic cells, and neutrophils, which consequently are well equipped to respond to virtually any invading pathogen.
Intestinal epithelial cells, however, are a special case in that they are exposed to massive amounts of bacterial products with the potential to activate innate immune receptors. Therefore, signaling through these receptors, in particular surface TLRs, must be prevented to avoid uncontrolled inflammation, which would be detrimental to the host. The mechanisms by which this negative regulation of innate immune receptor activation is exerted are poorly understood. It is believed that a combination of topological sequestration of the receptors and the activity of specific negative regulators that control the output through TLRs, are responsible for the control of intestinal cell homeostasis [7]–[12]. Misregulation of these negative regulatory control mechanisms leads to intestinal inflammatory pathology.
Certain enteropathogens are able to stimulate intestinal inflammation [13]. For example, Salmonella enterica serovar Typhimurium (S. Typhimurium), a major cause of food-borne illness, causes intestinal inflammation leading to diarrhea, which is essential for its replication and spread [14],[15]. The mechanisms by which these or other intestinal pathogens bypass the negative regulatory controls in the intestinal epithelium to induce inflammatory responses are poorly understood. Although limited gene expression studies have shown that S. Typhimurium reprograms gene expression in cultured cells leading to the production of proinflammatory cytokines [16]–[18], it is unclear to what extent these responses overlap with responses stimulated through the activation of innate immune receptors. The ability of S. Typhimurium to stimulate intestinal responses requires the activity of a type III protein secretion system (TTSS) encoded within its pathogenicity island 1 (SPI-1) [16]. However, it is unclear whether this system stimulates inflammatory responses directly, or whether it works in conjunction with surface innate immune receptors that may recognize components of this machine, or cytoplasmic receptors that may recognize conserved bacterial products, since this TTSS is also required for S. Typhimurium to enter epithelial cells [19]. Here we report that S. Typhimurium can stimulate innate immune responses in cultured epithelial cells through the activity of bacterial effector proteins delivered by its TTSS and in a manner that is independent of innate immune receptors. This mechanism has implications for the ability of S. typhimurium to induce intestinal inflammation.
In an effort to understand the mechanisms by which S. Typhimurium stimulates intestinal innate immune responses and inflammation, we first determined the global transcriptional response of cultured human Henle-407 epithelial cells infected with wild-type S. Typhimurium using DNA microarray analysis. Consistent with previous more limited studies [18], infection of epithelial cells with wild-type S. Typhimurium led to a very significant transcriptional reprogramming in comparison to uninfected cells (Fig. 1 and Tables S1 and S2). We identified 66 genes whose expression changed at least 4-fold in cells infected with wild-type S. Typhimurium relative to uninfected controls (Fig. 1A) and 290 genes whose expression changed at least 2-fold (Table S1). The majority (235 of 290) of these genes showed increased expression in response to infection and, in some cases, up to several hundred fold (Table S1). We validated the changes of gene expression observed by the microarray analysis by performing quantitative real-time PCR (qRT-PCR) determination of the mRNA (Fig. 2) or in some cases western blot analysis of protein levels (Fig. S1) of selected genes or gene products, respectively. The transcriptional program stimulated by wild-type S. Typhimurium infection included several genes whose products are pro-inflammatory such as several chemokines and cytokines and their receptors [e. g. CXCL2 (Mip2a), CXCL3 (Mip2b), IL-8, IL1a, IL11, IL1R1, COX-2, TNFRSF10D, IL4R, TNFRSF12A]. S. Typhimurium also induced the stimulation of expression of a number of transcription factors (e. g. FOS, FOSB, FOSL1, JUN, JUNB, EGR1, EGR3, ATF3, STAT3) that amplify the immune response and that most likely are constituents of a positive feed-back loop resulting in the stimulation of expression of themselves and other pro-inflammatory genes. In addition, bacterial infection stimulated the expression of several genes whose products limit the immune response, which most likely constitute a negative feed-back loop to preserve homeostasis and limit damage as a consequence of the inflammatory response. These included the mRNA stability regulator tristetraprolin (ZFP36), the suppressors of cytokine signaling SOCS2 and SOCS3, NAD(P)H dehydrogenase (NQO2), the NF-κB inhibitor IκB zeta, and several members of the DUSP family of tyrosine phosphatases (DUSP1, DUSP2, DUSP4, DUSP5, DUSP6, DUSP8).
The transcriptional response to wild-type S. Typhimurium exhibited striking similarities with reported transcription profiles of cells stimulated by agonists of receptors of the innate immune system such as TLRs, or the core response to other microbial pathogens [20]. A gene ontology analysis indicated that genes involved in the innate immune response were significantly over-represented among those induced 2-fold or more by wild-type S. Typhimurium infection (p<0.05). In fact, 54 of the 290 differentially-expressed genes belong to the cluster of genes that have been defined as the ‘common host response’ to pathogens (Fig. 1B), a group of genes that are induced in many different cell types in response to exposure to several different microbial pathogens [20]. The proportion is much higher if genes induced by other bacterial pathogens are considered (Table S3). These results indicate that this pathogen has the capacity to stimulate a pro-inflammatory response in cultured epithelial cells that exhibits a great deal of commonality with the innate immune responses triggered by agonists of innate immune receptors or by other pathogens.
To determine the extent to which the transcriptional changes induced by S. Typhimurium depend on the components and activities of its SPI-1 TTSS, we examined the transcriptional response of cultured Henle-407 epithelial cells infected with an isogenic ΔinvA mutant strain. InvA is an essential component of the SPI-1 TTSS and therefore the S. Typhimurium ΔinvA mutant is defective in all phenotypes dependent on this system, which include bacterial internalization into non-phagocytic cells and the induction of programmed cell death in macrophages [21],[22]. The transcriptional profile of cultured epithelial cells infected with this mutant strain closely resembled that of uninfected cells (Fig. 1 and Tables S1 and S2), demonstrating that the transcriptional reprogramming stimulated by wild-type S. Typhimurium requires a functional SPI-1 TTSS. The levels of lipopolysaccharide, flagellin, and other potential agonists of innate immune receptors are not affected by the invA mutation and are therefore identical to those of wild type (Fig. S2) [21]. Hence, these results also indicate that the responses observed in cells infected with wild-type S. Typhimurium cannot be the result of the stimulation of surface Toll-like receptors by these bacterial PAMPS.
The requirement of a functional SPI-1 TTSS for S. Typhimurium to trigger transcriptional reprogramming in cultured epithelial cells suggested the possibility that one or more effector proteins delivered by this system are responsible for the stimulation of those responses. Alternatively, such responses could be the result of other activities dependent on a functional TTSS. For example, the outermost component or “needle substructure” of the TTSS needle complex organelle is absent from the ΔinvA mutant strain [23],[24]. Therefore, it is possible that the needle substructure itself may be directly recognized by innate immune receptors in epithelial cells thus stimulating transcriptional responses. In addition, delivery of proteins by the TTSSs requires the deployment of a group of TTSS secreted protein translocases, which mediate the passage of effector proteins through the host cell plasma membrane [25],[26]. Therefore, it is possible that potential membrane perturbations caused by the deployment of the TTSS protein translocases, which does not occur in cells infected with the ΔinvA mutant, could be sensed by the cell triggering innate immune responses [25],[26]. Finally, stimulation of innate immune responses may be the result of the TTSS-mediate delivery of some agonist of intracellular immune receptors (e. g. flagellin).
To distinguish among these possibilities we examined the transcriptional responses of human epithelial cells infected with a S. Typhimurium strain lacking the protein effectors delivered by the SPI-1 TTSS (from here forth referred as “effectorless”). This strain displays a wild-type TTSS organelle, including the needle component, is fully capable of deploying the translocases SipB, SipC and SipD, and it is therefore competent for TTSS-mediated protein translocation [27]. Cells infected with the S. Typhimurium effectorless strain showed a transcriptional profile similar to that of uninfected cells or cells infected with the TTSS-defective invA mutant (Fig. 1, Fig. 2, and Table S1). These results ruled out the hypothesis that the innate immune responses stimulated by wild-type S. typhimurium are the result of the recognition of the needle component of the TTSS by innate immune receptors. In addition, these results excluded the hypothesis that the S. typhimurium-induced responses are the result of membrane perturbations induced by the deployment of the translocases, or the delivery of intracellular receptor agonists such as flagellin since all these activities or agonists are present in this mutant strain. Rather, these results unambiguously indicate that one or more effector proteins delivered by this TTSS must be responsible for the induction of the transcriptional reprogramming.
The guanidyl nucleotide exchange factors SopE and SopE2 and the phosphoinositide phosphatase SopB are good candidates to mediate these responses since, in a functionally redundant manner, they activate Rho-family GTPases [28]–[30], which can lead to MAPK and NF-κB activation [31]–[33]. Cells infected with a ΔsopE ΔsopE2 ΔsopB triple mutant strain showed a transcriptional profile largely similar to that of uninfected cells or cells infected with a strain lacking a functional SPI-1 TTSS (Fig. 1, Fig. 2, and Tables S1 and S2). Although strains carrying individual mutations in each one of these effectors retained their ability to activate MAPKs and NF-κB (Fig. 3), the ΔsopE ΔsopE2 ΔsopB S. Typhimurium triple mutant strain did not activate these signaling pathways (Fig. 3). These results are consistent with the hypothesis that the activation of these signaling pathways by the bacterially-encoded Rho-family GTPase activators SopE, SopE2, and SopB is central to S. Typhimurium's ability to stimulate innate immune responses.
In addition to its ability to activate Rho-family GTPases, SopB mediates activation of AKT by poorly understood mechanisms that require its phosphoinositide phosphatase activity [34]. However, we found that the transcriptome of cells infected with this strain was largely equivalent to that of cells infected with wild-type S. Typhimurium (Fig. 1, Fig. 2, and Tables S1 and S2) arguing against a prominent role for AKT in the stimulation of the transcriptional responses. Taken together, these results indicate that SopE, SopE2, and SopB, which operating in a functionally redundant manner activate Rho-family GTPases, MAPK and NF-κB signaling, are responsible for the transcriptional reprogramming induced by S. Typhimurium in cultured epithelial cells.
The SPI-1-TTSS S. Typhimurium mutant strains (i. e. ΔinvA, ΔsopE/ΔsopE2/ΔsopB, or “effectorless”), which were unable to stimulate transcriptional responses, posses “wild type” agonists of the innate immune system (i. e. LPS, flagellin, etc.) (Fig. S2). Consequently, these observations suggest that S. Typhimurium can trigger innate immune responses in a manner that does not require innate immune receptors such as TLRs. However, since the SPI-1 TTSS, and in particular the effector proteins SopE, SopE2, and SopB, are also required for bacterial internalization, the ΔsopE/ΔsopE2/ΔsopB or the “effectorless” mutant strains are unable to enter into host cells [28],[32]. Therefore, it is formally possible that the gene expression changes observed in cells infected with wild type S. Typhimurium are stimulated by intracellular bacteria through the intracellular cytoplasmic NLRs such as Nod1 or Nod2 [5], which can sense conserved bacterial products to stimulate innate immune responses [35]. To test this hypothesis we depleted cells of Rip2, a kinase essential for Nod1 and Nod2 signaling [36]. Depletion of Rip2 by RNAi (Fig. 4A), which had no effect on the ability of S. Typhimurium to enter epithelial cells (Fig. S3), did not prevent the transcriptional responses stimulated by S. typhimurium infection (in fact, the expression of some the reporter genes examined was actually higher in the Rip2-depleted cells) (Fig. 4B). Therefore, these results indicate that the Nod1 or Nod2 receptors are not required for the stimulation of the SPI-1 TTSS-dependent transcriptional responses.
To further address the potential contribution of cytoplasmic sensors to the S. Typhimurium SPI-1 TTSS-mediated stimulation of transcriptional responses in epithelial cells we used an alternative approach. We reasoned that if cytoplasmic sensors were responsible for the stimulation of cellular responses by intracellular S. Typhimurium, internalization of a SPI-1 TTSS-defective mutant strain by an alternative entry pathway should lead to the stimulation of the same transcriptional responses observed in cells infected with wild type S. typhimurium. We therefore expressed in the SPI-1 TTSS-defective ΔinvA mutant strain the Yersinia pseudotuberculosis invasin protein, which mediates bacterial uptake by interacting with α4-β1 integrin receptors [37],[38]. Infection of epithelial cells with the S. Typhimurium ΔinvA (p-invasin) strain did not stimulate the expression of 7 out of 9 reporter genes tested (Fig. 4C) despite the presence of similar levels of intracellular bacteria to that of cells infected with wild-type S. Typhimurium (Fig. 4D). The reduced but significant increase in the expression of IL-8 and COX2 is most likely due the effect of the stimulation of α4-β1 integrins by invasin, since these genes have been reported to be induced by the invasin protein [39]. These results indicate that the intracellular location of S. Typhimurium is not sufficient to stimulate the transcriptional responses observed after infection with wild-type S. Typhimurium. Furthermore, these observations are consistent with the hypothesis that specific signaling pathways triggered by the activity of the SPI-1 TTSS effectors SopE, SopE2, and SopB are responsible for the stimulation of the transcriptional reprogramming in epithelial cells.
SopE, SopE2, and SopB stimulate Rho-family GTPase signaling by different mechanisms. SopE and SopE2 are exchange factors for Rac1, Cdc42, and RhoG [32],[40], while SopB, through its phosphoinositide phosphatase activity, stimulates the RhoG exchange factor SGEF and an unknown exchange factor for Cdc42 [29]. Both Rac1 and RhoG are required for S. Typhimurium entry into cells [29],[32]. Cdc42, in contrast, is dispensable for bacterial entry although it is required for efficient stimulation of MAPK [29],[31]. In fact, S. Typhimurium can enter cells depleted of Cdc42 in a manner indistinguishable from non-depleted cells [29]. Furthermore, the vesicular traffic and intracellular location of S. typhimurium in Cdc42-depleted and non-depleted cells is indistinguishable (Fig. S4). Therefore, the lack of involvement of Cdc42 in the entry and intracellular fate of S. Typhimurium allowed us to specifically test its potential role in the transcriptional reprogramming induced by wild-type bacteria, independent from any secondary effect due to actin remodeling or the intracellular location of the internalized bacteria. We found that depletion of Cdc42 significantly prevented the stimulation of expression of the reporter genes (Fig. 5). These results indicate that this Rho-family GTPase plays a critical role in the SPI-1 TTSS-dependent transcriptional reprogramming induced by wild-type S. Typhimurium in epithelial cells. The observation that Cdc42 depletion did not completely block the stimulation of gene expression by S. Typhimurium suggests that signaling through the other Rho-family GTPases such as Rac1 and RhoG may also contribute to the stimulation of nuclear responses. Alternatively, incomplete depletion of Cdc42 could also account for this result since the transfection efficiency of the cell line used in this study is less than 100%.
We further explored the ability of the Salmonella encoded Rho-family GTPase activators to stimulate inflammatory responses by transiently expressing SopE in epithelial cells and examining its effect on IL-8 expression using a transcriptional reporter construct. Transient expression of SopE resulted in a significant increase in IL-8 transcription, and this effect was effectively prevented by the expression of dominant negative Cdc42 (Cdc42N17) (Fig. 5C). These results further demonstrate that bacterial effector proteins such as SopE can stimulate an inflammatory response, and that such a response requires Rho-family GTPases.
S. Typhimurium can induce colitis in streptomycin-treated Myd88-deficient mice [41] suggesting that signaling through most TLRs is not required to induce inflammation in this infection model. TLR4, the main Toll-like receptor involved in the control of S. Typhimurium infection in mice [42], can signal in a Myd88-independent fashion [6]. However, we found that S. Typhimurium can induce inflammation in Myd88/TLR4 deficient mice in the same manner as in wild-type animals (Fig. S5). Furthermore, S. Typhimurium can also induce inflammation in caspase-1 deficient mice [43]. These results indicate that TLR signaling or activation of the caspase-1 inflammasome [44] are not essential for the inflammatory response to S. Typhimurium. To assess the potential contribution of the NLRs Nod1 and Nod2 in S. Typhimurium-induced intestinal inflammation, we tested the ability of wild-type S. Typhimurium to induce colitis in Rip2-deficient animals. Using the streptomycin-treated mouse model, we found that wild-type S. typhimurium induced inflammation in a manner that was indistinguishable from the inflammation observed in wild type animals (Fig. 6). Furthermore, the ability of S. Typhimurium to induce inflammation in these animals also required the SPI-1 TTSS since a type III-deficient ΔinvA mutant did not cause observable pathology in this infection model system (Fig. 6). Taken together, these results indicate that in this model system, the ability of S. Typhimurium to cause inflammation through the activity of its TTSS effectors does not require known receptors of the innate immune system, and are consistent with the results obtained with cultured epithelial cells. Furthermore, these results are also consistent with the observation that SPI-1 TTSS effector proteins are required for the induction of intestinal inflammation in animal models of infection [45]–[48].
Intestinal pathogens such as S. Typhimurium must induce intestinal inflammation to secure the acquisition of scarce nutrients as well as their spread to other hosts through the induction of diarrhea [49]. An intact intestinal epithelium does not respond to the multitude of potential agonists of the innate immune system. Such responsiveness would be detrimental in the context of the presence of the abundant intestinal microbial flora containing a plethora of innate immune receptor agonists. Although the mechanisms for this unresponsiveness are not fully understood, it is thought that these mechanisms involve exclusion of innate immune receptors from the epithelial cells' apical side, and the expression of specific negative regulators of innate immune receptor signaling [7]–[12]. Therefore, to induce inflammation, enteropathogens must be able to circumvent these negative regulatory mechanisms [13]. We have shown here that S. Typhimurium has evolved a unique mechanism to stimulate innate immune responses in epithelial cells in a manner that is independent of the canonical innate immune receptors and conserved bacterial products or PAMPs. Rather, we have shown that S. Typhimurium uses a specific set of effector proteins, SopB, SopE2, and SopE, which are delivered into cells via its SPI-1 TTSS, to activate these responses. We have demonstrated that these responses are not the consequence of bacterial internalization and its subsequent detection by cytoplasmic sensors. Instead, S. Typhimurium stimulation of innate immune responses in epithelial cells are the result of the effector-mediated stimulation of Rho-family GTPases leading to the activation of MAPK and NF-κB signaling pathways. Nevertheless, these responses very closely mimic those induced by the stimulation of innate immune receptors such as TLRs or NLRs. It has been previously reported that TLR2 signaling requires Rac-1 [50], presumably at steps downstream from the receptor proximal components of the signaling cascade. Therefore, by targeting similar downstream components of the signaling cascade of innate immune receptors, S. Typhimurium effectors can trigger similar outputs as those triggered by the stimulation of those receptors. This mechanism may allow S. Typhimurium to bypass the negative regulatory mechanisms that prevent signaling through innate immune receptors in intestinal epithelial cells and thus initiate the inflammatory response at this site.
S. Typhimurium can induce intestinal inflammation in mice deficient in various innate immune pathways such as myd88−/− [41], caspase 1−/− [43], or as we have shown here, myd88/TLR4−/− and rip2−/− deficient mice. These observations indicate that our observations with cultured epithelial cells are likely to be relevant in vivo. Innate immune receptors, however, do play a major role in controlling S. Typhimurium infection and spread to systemic tissues, since mice with deficiencies in innate immune pathways are more susceptible to S. Typhimurium infection. Innate immune receptor pathways may also contribute to the amplification of the inflammatory response in the gut once initiated by S. Typimurium through the specific adaptations described here. However, our results as well as previous in-vivo studies [45]–[48] show that with an intact epithelium, S. Typhimurium needs its SPI-TTSS to initiate an intestinal inflammatory response presumably via the mechanism reported here.
It is well established that many pathogens have evolved specific adaptations to counteract their hosts' innate immune responses [51],[52]. However, we have shown here that S. Typhimurium has evolved specific mechanisms to stimulate innate immune responses, a remarkable adaptation presumably evolved to bypass mechanisms that are in place in intestinal cells to prevent inflammation. Although the mechanisms that prevent intestinal epithelial cells from responding to innate immune receptor agonists are poorly understood, it is well established that misregulation of such mechanisms can lead to chronic inflammatory conditions such as inflammatory bowel disease or Crohn's disease [7]–[11],[53]. The information gained from the study of a pathogen capable of inducing intestinal inflammation could lead to a better understanding of very important but poorly understood chronic inflammatory pathologies and may lead to novel therapeutic strategies.
The wild-type strain of S. enterica serovar Typhimurium (S. typhimurium) SL1344 and its isogenic derivatives used in this study, ΔinvA (SB136), ΔsopE (SB856), ΔsopE2 (SB1300), ΔsopB (SB1120), ΔsopE ΔsopE2 (SB1301), ΔsopE ΔsopB (SB925), ΔsopE ΔsopE2 ΔsopB (SB1302), have been previously described [21], [27]–[29]. The strain referred to as “effectorless” (SB1011) has the relevant genotype ΔsopA ΔsopB ΔsopD ΔsopE ΔsopE2 ΔavrA ΔsptP ΔslrP ΔsspH1, and has also been previously described [27]. The S. Typhimurium ΔinvA strain expressing the Yersinia pseudotuberculosis invasin protein was created by transforming the strain SB136 with plasmid pRI207 encoding the inv gene from Yersinia pseudotuberculosis [37]. All bacterial strains were cultured under conditions that stimulate the expression of the Salmonella Typhimurium pathogenicity island-1 encoded TTSS [54].
For bacterial infections, Henle-407 human epithelial cells (80% confluency) were washed twice with DMEM (without serum and antibiotics) and grown in 2 ml of DMEM (without serum and antibiotics) for 16–20 hs. Cells were then washed twice with Hank's buffered salt solution (HBSS) and allowed to equilibrate in HBSS at 37°C for 15 min, and infected at an MOI of 30. In experiments using the S. Typhimurium ΔinvA mutant strain expressing invasin, an MOI of 100 was used to obtain equal number of internalized bacteria as that of cells infected with wild type. One hour after infection, each well was washed twice with 2 ml of DMEM containing gentamicin (100 µg/ml) and then grown in DMEM containing gentamicin (100 µg/ml) for three additional hours after which the cells were harvested for RNA extraction. The infections for the microarray experiments were performed in 10 cm dishes. All other infections were performed in 6-well dishes.
Total RNA from infected cells and uninfected control cells was extracted using the RNeasy Midi kit (Qiagen) following manufacturer's instructions. Sample preparation and hybridization to Affymetrix Human Genome U133 Plus 2.0 gene arrays were performed at the Yale University W.M. Keck facility. Briefly, target cDNA generated from each sample was biotinylated, hybridized, and stained as per manufacturer's recommendation using an Affymetrix GeneChip Instrument System. Arrays were scanned on an Affymetrix GeneChip scanner 3000 according to Affymetrix standard protocols (GeneChip Expression Analysis Technical Manual, Affymetrix, 2004). Data was processed using Affymetrix Microarray Suite version 5.0, scaled to a target intensity of 500. Raw and normalized data have been submitted to the GEO database (http://www.ncbi.nlm.nih.gov/geo/), accession number (pending). Data were further analyzed using GeneSpring (Silicongenetics) and the BioConductor software package [55]. Fold change were calculated for each strain relative to the uninfected control in each experimental group. Genes with a MAS5 change call other than “No Change” were considered to be differentially expressed if they also met a minimum fold change requirement in all the experiments using that strain. Analyzed expression data are presented in Table S1. The gene ontology analysis was performed using the GOstats package by BioConductor [56] with the gene universe consisting of all genes with a MAS5 change call other than “No Change” in any experiment of any mutant strain.
Depletion of endogenous Cdc42 was performed using a short hairpin sequence targeting human Cdc42 as described previously [29]. Silencing of RIP2K gene expression was achieved using synthetic SMARTpools (Dharmacon), each comprising four proprietary siRNA sequences. Both the RIP2 SMARTpool and short hairpin RNA constructs were transfected into Henle-407 cells using Lipofectamine 2000 (Invitrogen), either alone or in combination with pLZRS-CFP (kindly provided by Walther Mothes) for detection of transfected cells during microscopy. Transfections were allowed to proceed for 48 hrs before serum starvation was begun for subsequent infection experiments. The silencing efficiency of the RNAi constructs was measured by qRT-PCR at the end of the infection experiments as described below, or by western blot analysis using a mouse anti-Cdc42 (BD Biosciences).
The acquisition of LAMP1 by the Salmonella-containing vacuole and its co-localization with the endocytic tracer Alexa-Fluor-488-labeled dextran were assayed in Cdc42-depleted and control cells by flurorescence microscopy as previously described [57]. Briefly, Cdc42-depleted and control cells grown on glass coverslips were incubated in the presence of 250 µg/ml dextran-Alexa-Fluor-488 MW 10,000 (Molecular Probes), which was chased into lysosomes as previously described [58]. Three hours before infection, cells were washed twice with PBS and subsequently incubated in cultured medium. Cells were infected for 1 h with a multiplicity of infection of 10 with wild-type S. Typhimurium expressing dsRed under the control of an arabinose-inducible promoter [59], which had been grown in the presence of 0.1% arabinose. After additional 3 h in medium containing 100 µg/ml Gentamicin, cells were fixed with 3% PFA and subsequently stained with mouse-anti-LAMP1 (clone H4A3) and goat-anti-mouse-Alexa-Fluor-488 (Molecular Probes). For quantification, the localization of bacteria relative to LAMP1 or dextran was determined in CFP positive cells (to identify transfected cells) using fluorescence microscopy.
Total RNA was isolated from infected Henle-407 cells using TRIzol reagent (Invitrogen) following the manufacturer's protocol. The final pellet was resuspended in RNAse-free water and further purified using the RNeasy Mini kit (Qiagen), digested with DNAse I (Invitrogen) and used as a template for cDNA synthesis using the iScript reverse transcriptase (Bio-Rad). The cDNA was subject to qRT-PCR using iQ SYBR Green Supermix (Bio-Rad). Reactions were run and measurements were obtained using an iCycler realt time PCR machine and iCycler iQ software (Bio-Rad).
Cells were lysed in lysis buffer containing 10 mM Tris-HCl, pH 7.5, 40 mM Na-Pyrophosphate, 5 mM EDTA, 150 mM NaCl, 1% NP-40, 0.5% Na-Deoxycholate, 0.025% SDS, 1 mM Na-orthovanadate and protease inhibitors (complete Protease Inhibitor Cocktail, Roche). Cell lysates were separated by SDS-PAGE, and examined by western immunoblotting using the following antibodies: rabbit-anti-Erk, rabbit-anti-IκBα, rabbit-anti-JNK/SAPK, mouse-anti-phospho-Erk [Thr 202, Tyr 204], mouse-anti-phospho-IκBα [Ser32, Ser 36], mouse-anti-phospho-JNK/SAPK [Thr 183, Tyr 185], rabbit anti-phospho HSP27 [Ser82] (to evaluate p38 activation), and mouse-anti-phospho-P38 [Thr 180, Tyr 182], all purchased from Cell Signaling Technology (Danvers, MA), rabbit-anti-P38, mouse anti SCCA-1, and rabbit-anti-actin purchased from Santa Cruz Biotechnology (Santa Cruz, CA), and rabbit anti Tristetraprolin (TTP) purchased from Abcom (Cambridge, MA).
Cells were transfected with a plasmid encoding SopE [32] along with the reporter plasmid PSB2805, which encodes a fusion between the IL-8 promoter and firefly luciferase [29]. To standardize transfection experiments, cells were also transfected with a plasmid pSB2806, which is a derivative of pCDN3.1 encoding Renilla luciferase [29]. Two days after transfection (including overnight serum starvation), cells were lysed and the levels of firefly and Renilla luciferase were determined using the Dual Luciferase Reporter assay (Promega) according to the manufacturer's instructions. Transfection efficiency was normalized by the comparison of IL-8–induced firefly luciferase levels with that of constitutively expressed Renilla luciferase. Induction of the IL-8 reporter by SopE was expressed relative to that of a vector control.
Levels of LPS and flagellin in whole cell bacterial lysates of the different strains was determined by standard western immunoblot analysis using specific antisera (Difco).
Bacterial infections of streptomycin-treated animals and histopathology analysis of tissues were performed essentially as described previously [43]. All animals were maintained and animal experiments conducted in accordance with the guidelines of the Yale Institutional Animal Use and Care Committee.
|
10.1371/journal.pntd.0002454 | Sequence Analysis and Serological Responses against Borrelia turicatae BipA, a Putative Species-Specific Antigen | Relapsing fever spirochetes are global yet neglected pathogens causing recurrent febrile episodes, chills, nausea, vomiting, and pregnancy complications. Given these nonspecific clinical manifestations, improving diagnostic assays for relapsing fever spirochetes will allow for identification of endemic foci and expedite proper treatment. Previously, an antigen designated the Borrelia immunogenic protein A (BipA) was identified in the North American species Borrelia hermsii. Thus far, BipA appears unique to relapsing fever spirochetes. The antigen remains unidentified outside of these pathogens, while interspecies amino acid identity for BipA in relapsing fever spirochetes is only 24–36%. The current study investigated the immunogenicity of BipA in Borrelia turicatae, a species distributed in the southern United States and Latin America.
bipA was amplified from six isolates of Borrelia turicatae, and sequence analysis demonstrated that the gene is conserved among isolates. A tick transmission system was developed for B. turicatae in mice and a canine, two likely vertebrate hosts, which enabled the evaluation of serological responses against recombinant BipA (rBipA). These studies indicated that BipA is antigenic in both animal systems after infection by tick bite, yet serum antibodies failed to bind to B. hermsii rBipA at a detectable level. Moreover, mice continued to generate an antibody response against BipA one year after the initial infection, further demonstrating the protein's potential toward identifying endemic foci for B. turicatae.
These initial studies support the hypothesis that BipA is a spirochete antigen unique to a relapsing fever Borrelia species, and could be used to improve efforts for identifying B. turicatae endemic regions.
| Undiagnosed febrile illnesses continue to afflict those in resource poor countries. Relapsing fever spirochetes are one such pathogen causing a significant health burden, yet the pathogenesis, ecology, and distribution of B. turicatae is understudied. To address these shortcomings, we analyzed the amino acid sequence of the Borrelia immunogenic protein A (BipA) in isolates of B. turicatae. Mice and a canine were also infected by tick bite and transmission and serological responses were evaluated in these two likely mammalian hosts. B. turicatae was visualized within the blood of both animals and antibody responses generated against recombinant BipA indicated that the antigen that may be unique to infections caused by B. turicatae. Moreover, mice continued to generate antibodies a year after tick bite, suggesting a persistent infection. Our results indicate that the immune responses generated against BipA could identify additional vertebrate hosts, define endemic foci for B. turicatae, and increase the awareness of the disease to improve healthcare.
| Three causative agents of tick-borne relapsing fever borreliosis in the New World are Borrelia hermsii, Borrelia parkeri, and Borrelia turicatae, with B. hermsii being the most epidemiologically and ecologically characterized species [1]. While B. hermsii is distributed in high elevation coniferous forests and maintained in enzootic cycles with rodents as the primary reservoir, less is known regarding the other two species. Moreover, few epidemiological studies have been performed and little molecular data exists for B. turicatae and its arthropod vectors Ornithodoros turicata. There are endemic foci for B. turicatae in Texas and Florida, where clinical isolates have been obtained from sick dogs [2], [3], which suggests a role for wild canids in the maintenance of the spirochetes in nature. Dr. Oscar Felsenfeld also reported the distribution of O. turicata into Mexico, Central, and South America [4], yet given the absence of Latin American isolates for B. turicatae the identification of endemic foci is unclear.
A limitation in defining the distribution of B. turicatae has been the absence of diagnostic antigens specific for the species. Previously, the Borrelia immunogenic protein A (BipA) of B. hermsii was demonstrated to discriminate infections caused by Lyme and relapsing fever borreliosis. Outside of relapsing fever spirochete spp. a homologue of BipA has not been cataloged in the GenBank database [5]. With 36% amino acid identity between B. hermsii and B. turicatae BipA [5], it is also unclear if the B. turicatae homologue induces a host antibody response during infection. This study investigated the sequence similarity of BipA between B. turicatae isolates, and we developed a tick transmission system for the spirochetes to determine the antigenicity of recombinant BipA (rBipA) during rodent and canine infections. Collectively, these results suggests that BipA can be used as a diagnostic antigen for B. turicatae.
All animal studies were in accordance with the Mississippi State University Institutional Animal Care and Use Committee (IACUC protocol #'s 11-091 and 12-067). Animal husbandry was provided by veterinary staff and technicians within the Association for Assessment and Accreditation of Laboratory Animal Care and the National Institutes of Health Office of Laboratory Animal Welfare assured program at Mississippi State University. All work was performed in adherence to the United States Public Health Service Policy on Humane Care and Use of Laboratory Animals and the Guide for the Care and Use of Laboratory Animals.
B. turicatae was cultivated in mBSK medium containing 12% rabbit serum [6], [7]. Amplification and sequencing primers for bipA were designed using the 91E135 isolate of B. turicatae (Table 1). Additional samples were water (negative control), 95PE-570, 95PE-1807, TCB-1, TCB-2, and FCB-1 isolates [2]. Polymerase chain reaction (PCR) was performed as previously described using the GoTaq Flexi DNA Polymerase (Promega Corporation, Madison, WI, USA). Amplicons were electrophoresed on a 1% agarose gel to visualize the DNA fragment and processed through the QIAquick PCR Purification kit (Qiagen, Germantown, MD, USA), and sequencing performed at Biodesign Institute (Arizona State University, Phoenix, AZ, USA). Nucleotide sequences were analyzed with the Vector NTI software (Life Technologies, Carlsbad, CA, USA), and deposited to GenBank under accession numbers KC845527-KC845531.
The ticks used in the study originated from 12 uninfected adults initially maintained at the Rocky Mountain Laboratories. A cohort of second nymphal stage O. turicata was infected by first needle inoculating a group of 3 three Swiss Webster mice with B. turicatae 91E135 [2], and uninfected ticks were allowed to engorge on the animals. After molting, vector colonization was confirmed by dissecting the midgut and salivary glands from five ticks and performing immunofluorescent assays (IFA) as previously described [8]. Chicken serum generated by Cocalico Biologicals Inc. against B. turicatae recombinant flagellin (rFlaB) was used to detect spirochetes and the secondary antibody was goat anti-chicken IgY Alexa Fluor 568 (Life Technologies, Grand Island, NY, USA).
Prior to the transmission studies, pre-infection serum samples were collected from all animals. Cohorts of five infected ticks per animal were allowed to feed to repletion. The animals used in the study were 10 outbred Swiss Webster mice (Harlan Laboratories Inc., Tampa, FL, USA) and a one-year-old Bluetick hound (Marshall Bioresources, North Rose, NY, USA). For 16 consecutive days, animals were monitored for clinical symptoms and a drop of blood was collected from the mice by tail nick and from the canine's cephalic vein to visualize spirochetes. The day following the first and second febrile episode, 3 ml of blood were collected from the canine for a complete blood count and serum chemistry profile.
Of the mice used in the initial transmission study, four were maintained for one year to assess long-term serological responses generated against BipA. Prior to serum collection for immunoblotting, 2.5 µl of blood was collected for 10 consecutive days and placed into 47.5 µl of SideStep Lysis and Stabilization Buffer (Agilent, Santa Clara, CA, USA). Quantitative PCR (qPCR) was performed as previously described [9] to determine the presence of circulating spirochetes. The probe (flaB probe) and primer set (flaB F and flaB R) used for qPCR were designed for B. turicatae flagellin (Table 1).
Two groups of three mice were also needle inoculated intraperitoneally with 1×105 TCB-1 or FCB-1 spirochetes. Infection was confirmed by dark field microscopy and serum samples were collected one month after inoculation to determine reactivity to rBipA expressed from the 91E135 isolate.
To express bipA from B. turicatae 91E135 as a 75 kDa thioredoxin and histidine tagged fusion protein in Escherichia coli, the gene was amplified as previously described [5] using bipA F Topo and bipA R Topo primers (Table 1). The amplicon was cloned into the pET 102/Directional TOPO expression vector following the manufacturer's instructions (Life Technologies). Top10 E. coli were transformed, plasmid DNA isolated, and sequence analysis using bipA F1, bipA F2, bipA R1, bipA R2 primers (Table 1) was performed as previously described [5] to determine if an error had been introduced during amplification. To produce recombinant protein, BL21 E. coli were transformed with the bipA expression vector following the manufacturer's instructions, and induction was performed with 1 mM IPTG. rBipA was purified using the Ni-MAC Purification system (Novagen, Durmstadt, Germany).
Immunoblotting was performed to evaluate the immunogenicity of rBipA during B. turicatae infections. Protein lysates from 1×108 spirochetes, 1 µg of B. hermsii rBipA , and 1 µg B. turicatae rBipA were electrophoresed and transferred to polyvinylidene fluoride (PVDF) membranes using TGX gels, the Mini-PROTEAN Tera cell, and the Mini Trans Blot system (BioRad, Hercules, CA, USA). Pre- and post-infection serum samples were evaluated by immunoblotting at a 1∶500 dilution and the secondary molecule used was Rec-protein G-HRP (Life Technologies) at a 1∶4,000 dilution. Immunoblots were also probed with the Anti-polyHistidine Peroxidase monoclonal antibody (Sigma-Aldrich, St. Louis, MO, USA) at a 1∶4,000 dilution. Titers against rBipA were determined by immunoblotting with serum dilutions ranging from 1∶500 to 1∶125,000.
Linear regression analysis was performed by calculating the density of rBipA protein bands from immunoblots probed with serum samples diluted from 1∶800 to 1∶12,800. Serum samples included Anti-polyHistidine Peroxidase monoclonal antibody (Sigma-Aldrich, St. Louis, MO, USA) and serum samples from mice and the canine infected by tick bite. ImageJ, http://imagej.nih.gov/ij (National Institutes of Health, Bethesda, Maryland, USA), was used to analyze digitally scanned immunoblots and the density of each protein band was calculated. The R software package, www.r-project.org, was used to calculate equations of regression, R2 values, and significance.
One year after tick bite, mice were euthanized by isoflurane inhalation followed by cervical dislocation and tissues were placed in 10% neutral buffered formalin for fixation. Sections of brain and synovial joints from the front and rear leg were processed and embedded in paraffin using standard histologic techniques. Paraffin embedded tissues were cut in 5 µm sections, deparaffinized, adhered to glass slides, and stained with hematoxylin and eosin (H&E), or silver stained using standard histologic techniques. The sections were examined by light microscopy for inflammatory and degenerative lesions in the cerebrum, cerebellum, brainstem, and synovial joints by a board certified veterinary pathologist at Mississippi State University.
BipA was originally identified by an immunoproteomic antigen discovery approach [5], [10]. The degree dissimilarity between B. hermsii and B. turicatae homologues prompted our investigation to compare sequences between B. turicatae isolates, and to evaluate the protein's immunogenicity in two mammals likely to be naturally exposed to the spirochetes. PCR amplification of bipA from three canine (FCB-1, TCB-1, and TCB-2) and three tick (91E135, 95PE-570, and 95PE-1807) isolates produced a product of the expected molecular mass, with TCB-1 producing a slightly larger amplicon (Figure 1). Sequence analysis identified an additional 135 nucleotides encoding 45 amino acids for TCB-1 (Figure 2). While there was an overall 89% amino acid identity of BipA between isolates, the sequences flanking the 45 additional amino acids of TCB-1 were 97% identical. A similar observation was reported for B. hermsii BipA [5]. The protein from genomic group II (GG II) isolates of B. hermsii contained five regions of 3–24 amino acid insertions when compared to genomic group I (GG I) isolates. Furthermore, the amino and carboxy terminus of BipA between GGI and GGII isolates shared the highest degree of conservation.
Currently, the only known isolates of B. turicatae originate from argasid soft ticks and sick dogs [2]. Furthermore, the mammalian hosts for most species of relapsing fever spirochetes include rodents and insectivores [11]. This knowledge directed us to evaluate the antigenicity of B. turicatae rBipA after tick bite using canine and murine animal models. To establish an infected tick colony, uninfected O. turicata engorged on a Swiss Webster mouse needle-inoculated with B. turicatae 91E135. Spirochete colonization was confirmed by performing IFA on the midgut and salivary glands after the ticks molted (data not shown). The remaining infected ticks fed to repletion on Swiss Webster mice and a Bluetick hound, and within four and eight days after tick bite spirochetes were visualized in murine and canine blood, respectively (Figure 3 A and B). While the mice remained active when B. turicatae were visualized in the blood, the canine became febrile, lethargic, and following a given spirochetemic episode, acutely thrombocytopenic (Table 2). B. turicatae repopulated the blood from both groups of animals within four days after the initial spirochetemia, after which bacteria were undetectable by microscopy.
Producing rBipA using the same expression vector as B. hermsii bipA [5] indicated that infected animals generated an immunological response that recognized B. turicatae rBipA, yet antibody binding against the recombinant B. hermsii homologue was undetectable (Figure 4 A and B). Probing the immunoblots with an anti-polyhistidine monoclonal antibody confirmed that similar protein loads of B. turicatae and B. hermsii rBipA were electrophoresed, while pre-infection serum samples failed to produce a detectable antibody response against B. turicatae protein lysates or rBipA (Figure 4 C–E). Canine and murine IgG titers using serum samples collected 8 weeks after tick bite ranged from 1∶12,800 to1∶28,800. Also, regression analysis indicated significant differences (P≤0.05) in slopes and correlation coefficients (R2) when immune serum samples were probed against B. turicatae and B. hermsii rBipA (Figure 5 A–C). These results indicate different affinity characteristics against rBipA from a given species when animals were infected with B. turicatae.
With BipA from TCB-1 and FCB-1 being the most divergent to the 91E135 homologue, we evaluated serological responses against rBipA from animals infected with TCB-1 and FCB-1. Inoculating mice with each isolate determined that antibodies generated against TCB-1 and FCB-1 BipA were cross reactive against rBipA from B. turicatae 91E135 (Figure 6 A–D). Similarly, in B. hermsii there was sufficient amino acid conservation between BipA from B. hermsii GGI and GGII isolates that mice infected with GG II isolates produced a detectable serological response to rBipA that was expressed from a GG I isolate [5]. Collectively, these results suggest that BipA may be a unique antigen for the given species of relapsing fever spirochete causing infection.
Previous studies by Cadavid et al. reported that BALB/c and SCID mice needle inoculated with the Ozona isolate of B. turicatae developed long-term infections of the brain and joints [12]. Given the persistent nature of the spirochetes within rodents, IgG responses in mice were evaluated one year after transmission by tick bite. Prior to serological analyses, qPCR performed on murine blood samples collected for 10 consecutive days indicated that the mice were no longer spirochetemic (data not shown). Immunoblotting demonstrated that three of four mice continued to generate an IgG response against rBipA one year after the initial exposure, while one animal produced a weakly detectable response (Figure 7 A–D). Interestingly, B. turicatae is no longer detected within the blood of Swiss Webster mice after approximately 14 days after tick bite (data not shown), and with a serum half-life of 20–30 days for IgG, these results suggest a persistent infection and antigen exposure to the host immune response.
Central nervous system (CNS) infections by relapsing fever spirochetes vary between species and genetic variants. Borrelia duttonii can reemerge in the blood from the brain after a period of quiescence [13]. Serotype A of B. turicatae Ozona were neurotropic in mice, while animals infected with serotype B spirochetes colonize the joints and heart [12], [14], [15]. Interestingly, CNS infection caused by B. duttonii and serotype A of B. turicatae Ozona failed to produce noticeable tissue damage [12], [13]. In our study, postmortem necropsies of mice one year after infection by tick bite did not identify inflammatory or degenerative changes within the cerebrum, cerebellum, brain stem, or diarthrodial joints of the fore or hind limbs (data not shown). Spirochetes were also undetectable in tissue sections, and it was unclear if the animals were still infected at the time of euthanasia. However, the persistent antibody responses generated against rBipA can be targeted to increase the likelihood of determining if an animal has been exposed to the spirochetes.
With results suggesting that BipA may be a species-specific antigen, additional studies should evaluate homologues from less characterized yet closely related species to B. turicatae. For example multilocus sequencing indicated that Borrelia johnsonii, a novel species of relapsing fever spirochete that colonizes Carios kelleyi [16], was closely related to B. turicatae and Borrelia parkeri. As additional sequence information is obtained from B. johnsonii and B. parkeri and animal models developed, the diagnostic potential of BipA can be further evaluated as an antigen unique to a given species of relapsing fever spirochete.
The maintenance and ecology of B. turicatae in the southern United States and Latin America is poorly understood, and given the nonspecific clinical symptoms, the disease is likely under reported. Historically, mapping endemic foci has been associated with capturing ticks at sites where human infection occurred and evaluating the arthropods for spirochete colonization, or by obtaining clinical isolates from sick dogs [2], [3], [17]. Pathogen surveillance based on identifying infected ticks can be difficult because O. turicata are nest-, den-, and cave-dwelling with a 5–60 minute bloodmeal [3], [11], [18], and the ticks are rarely identified on the host. We are also unaware of serological surveys for B. turicatae probably due to the degree of antibody cross-reactivity that occurs during spirochete infections [5], [19], [20]. Given the characterization of BipA, serological analyses to identify endemic foci for B. turicatae where rodents and wild canids are monitored as sentinels are possible.
|
10.1371/journal.pgen.1003210 | A Newly Uncovered Group of Distantly Related Lysine Methyltransferases Preferentially Interact with Molecular Chaperones to Regulate Their Activity | Methylation is a post-translational modification that can affect numerous features of proteins, notably cellular localization, turnover, activity, and molecular interactions. Recent genome-wide analyses have considerably extended the list of human genes encoding putative methyltransferases. Studies on protein methyltransferases have revealed that the regulatory function of methylation is not limited to epigenetics, with many non-histone substrates now being discovered. We present here our findings on a novel family of distantly related putative methyltransferases. Affinity purification coupled to mass spectrometry shows a marked preference for these proteins to associate with various chaperones. Based on the spectral data, we were able to identify methylation sites in substrates, notably trimethylation of K135 of KIN/Kin17, K561 of HSPA8/Hsc70 as well as corresponding lysine residues in other Hsp70 isoforms, and K315 of VCP/p97. All modification sites were subsequently confirmed in vitro. In the case of VCP, methylation by METTL21D was stimulated by the addition of the UBX cofactor ASPSCR1, which we show directly interacts with the methyltransferase. This stimulatory effect was lost when we used VCP mutants (R155H, R159G, and R191Q) known to cause Inclusion Body Myopathy with Paget's disease of bone and Fronto-temporal Dementia (IBMPFD) and/or familial Amyotrophic Lateral Sclerosis (ALS). Lysine 315 falls in proximity to the Walker B motif of VCP's first ATPase/D1 domain. Our results indicate that methylation of this site negatively impacts its ATPase activity. Overall, this report uncovers a new role for protein methylation as a regulatory pathway for molecular chaperones and defines a novel regulatory mechanism for the chaperone VCP, whose deregulation is causative of degenerative neuromuscular diseases.
| Methylation, or transfer of a single or multiple methyl groups (CH3), is one of many post-translational modifications that occur on proteins. Such modifications can, in turn, affect numerous aspects of a protein, notably cellular localization, turnover, activity, and molecular interactions. In addition to post-translational modifications, the structural organization of a protein or protein complex can also have a significant impact on its function and stability. A group of factors known as “molecular chaperones” aid newly synthesized proteins in reaching their native conformation or alternating between physiologically relevant states. We present here a new family of factors that promote methylation of chaperones and show that, at least in one case, this modification translates into a modulation in the activity of the substrate chaperone. Our results not only characterize the function of previously unknown gene products, uncover a new role for protein methylation as a regulatory pathway for chaperones, and define a novel regulatory mechanism for the chaperone VCP, whose deregulation is causative of neuromuscular diseases, but also suggest the existence of a post-translational modification code that regulates molecular chaperones. Further decrypting this “chaperone code” will help understanding how the functional organization of the proteome is orchestrated.
| Methyltransferases catalyze the transfer of a methyl group (CH3) from a donor, generally S-adenosyl-L-methionine (AdoMet), to various acceptor molecules such as proteins, DNA, RNA, lipids, and small metabolites [1]–[3]. Protein N-methylation predominantly targets the side chains of two amino acids, lysine and arginine, whereas the side chains of other residues, including histidine, glutamine, and asparagine represent minor targets for methylation [4]–[6]. Dicarboxylic amino acids (glutamate, aspartate) and cysteine are also known to be respectively O- and S-methylated on occasion [7], [8]. In addition, some proteins were shown to be methylated on their N-terminal and C-terminal ends [9]–[11]. The vast majority of methyltransferases are grouped into three large families based on their structure, namely seven-β-strand, SET and SPOUT domain-containing methyltransferases [2]. All protein R methyltransferases (PRMT) are part of the seven-β-strand superfamily, while protein K methyltransferases (PKMT) fall almost exclusively within the SET domain-containing group. Until recently, the only known seven-β-strand PKMT was Dot1 [12].
Efforts to characterize substrates of PKMT have mostly focused on nucleosome components. Methylation of histone H3 residues K4, K36, and K79 are associated with transcriptionally active euchromatin, whereas methylation of H3K9, H3K27 and H4K20 coincides with heterochromatin and transcriptional repression [13], [14]. Recent reports have furthermore shown that the type of lysine methylation (i.e., mono-, di- or trimethylation) should also be taken into consideration when assessing chromatin state [15]–[17]. Epigenetics has been paramount in demonstrating that a modification as seemingly insignificant as the addition of a methyl group can have a considerable impact on a biological process as crucial as gene expression. Evidence of lysine methylation-driven regulation has been documented for an ever-increasing number of non-histone proteins, including calmodulin, cytochrome C, Rubisco, ribosomal proteins, p53, and NF-κB [18]–[27].
As part of an effort to systematically map protein-protein interactions, we came across a previously uncharacterized protein sharing distant homology with PRMTs nestled within a network of molecular chaperones involved in protein complex assembly. Subsequent local alignement searches using that protein as seed uncovered a group of 10 distantly related putative methyltransferases. Characterization of the interaction network of this novel subgroup of methyltransferases was undertaken by Affinity Purification coupled to Mass Spectrometry (AP-MS) and then computationally assessed. Our results revealed that enzymes of this subgroup preferentially interact with molecular chaperones. Validation experiments using three of the identified interactors, Kin17, Hsc70, and VCP/p97, indicated that they represent bona fide substrates. In each case, trimethylated lysine residues were identified in vivo and confirmed in vitro using recombinant methyltransferase-substrate pairs. In addition, we have shown that methylation of one of these substrates, VCP/p97, by METTL21D can be modulated by ASPSCR1/UBXD9 and that this modification regulates ATPase activity of the VCP chaperone. The results presented here bring to light an entirely new cast of PKMTs of the seven-β-strand variety and expands our knowledge of non-histones substrates, most notably molecular chaperones. This finding points to a new role for protein methylation in regulating protein folding, quality control, and turnover.
The study of this group of previously uncharacterized methyltransferases was initiated when METTL22 was identified in the soluble fraction of a protein affinity purification that targeted the DNA/RNA binding protein Kin17/KIN. Local alignment searches were performed to ascertain the function of this protein (data not shown). It was discovered that METTL22 was part of a larger group of 10 proteins (if the diversity of FAM86 closely-related isoforms are considered as a single member) that shared distant homology with PRMTs. Phylogenetic analysis of the most conserved region of these two protein groups (Figure S1) confirmed this observation, suggesting that this family of uncharacterized methyltransferases is related to, but distinct from, PRMTs (Figure 1A; Figure S2). Computational structure prediction further demonstrated the similarity between the members of this family of putative methyltransferases and PRMTs (Figure 1B, 1C). The subsequent publication of the human “methyltransferasome” by Clarke and colleagues confirmed that these proteins are putative methyltransferases and that they form a distinct family [2]. Indeed, most of the methyltransferases described here fall within the so-called “Group J.”
Based on the observed homology with PRMTs, we hypothesized that these proteins were likely protein methyltransferases themselves. To identify possible substrates and cofactors, we elected to map the protein interaction network for each member of this novel family by AP-MS (main interactors are marked in Figure 2, additional targets are listed in Table S1) [28]–[35]. The main METTL22 interactor identified in the soluble fraction was KIN, which confirmed our initial observation and further strengthened the notion that these two factors interact. A common theme for most of these putative methyltransferases' interactors was chaperones, be they of the Hsp70 or Hsp90 variety (see METTL18, CAMKMT, METTL21C, METTL22, METTL23, METTL21A, and METTL21B), chaperonins HSPD1 and CCT (see METTL18, CAMKMT, METTL20, and METTL21B), and even the AAA ATPase VCP (see METTL21D) that is believed to act as a chaperone in various processes, most notably Endoplasmic Reticulum-Associated Protein Degradation (ERAD) [36]–. Recent publications have substantiated the accuracy of this interactome. Firstly, it was demonstrated in another report by Steven Clarke on the yeast homolog of METTL18, YIL110W, that this protein methylates the ribosomal protein RPL3 [6]. Our own purification of METTL18 identified RPL3 and its associated ribosome biogenesis factor GRWD1 [41] as the two main interactors. Secondly, CAMKMT has been shown to methylate calmodulin on a lysine residue [42]. We were likewise able to co-purify calmodulin protein CALM2 in our CAMKMT affinity purification. However, it should be noted that since our AP-MS protocol includes beads bearing the calmodulin-binding peptide (CBP), calmodulin often appears as a non-specific target, although usually with a weaker signal. Computational assessment showed that CALM2 was a high confidence interactor of CAMKMT (FDR<10%).
Protein database searches were repeated allowing for mono-, di-, and trimethylation of lysine residues (as a variable modification). Of note, lysine trimethylation and acetylation are sometimes mis-annotated due to the closeness in mass of these modifications (+42.0468 Dalton and +42.0105 Dalton, respectively) [43]. Fortunately, the high mass accuracy obtained with the LTQ Orbitrap mass spectrometer was sufficient to distinguish between these PTMs. The most promising hits were a trimethylated lysine on KIN at position 135 in the METTL22 purification, another trimethylated lysine at position 315 on VCP in the METTL21D purification, and a number of trimethylated lysines on multiple Hsp70 isoforms, which correspond to a homologous site, in the METTL21A purification (Figure 3A; see corresponding mass spectra in Figure S3). These methylation sites were highly conserved through evolution (Figure 3B). Conservation of the VCP methylation site K315 is not surprising considering the relative immutability of the overall protein (roughly 70% sequence identity from Homo sapiens to Saccharomyces cerevisiae). The target lysine 561 in HSPA8/Hsc70 was likewise conserved through evolution and orthologs are found in species as distant as S. cerevisiae. Moreover, and as mentioned previously, this site is also retained in a number human Hsp70 paralogs (HSPA1, HSPA1L, HSPA2, HSPA5, and HSPA6). In fact the only conserved residue in this region of loose homology is the target lysine, pointing to a possible important regulatory role for this modification. The target lysine in KIN, K135, is present in a number of species, including Arabidopsis thaliana and Drosophila melanogaster, but is absent in Saccharomyces cerevisiae and Plasmodium falciparum. Interestingly, METTL22 orthologs are concurrently absent in species where the corresponding lysine in KIN is not conserved, which further suggests a strong link between the two.
We then proceeded with in vitro methylation assays to confirm the identity of these methylation targets (Figure 3C–3E). A positive signal was observed for each reaction, confirming that these are in fact protein methyltransferases. Moreover, substitution of each identified lysine to an arginine, a relatively conserved substitution, led to the abrogation of the methylation signal. In the case of KIN, the K135R substitution greatly diminished the methylation signal, but did not completely abolish it as with other mutants tested. This could mean that there might be a second methylation site on KIN, but given that no other methylated peptide was ever observed by mass spectrometry, either in the original purification of METTL22 or in the in vitro methylation reaction itself (see corresponding mass spectra in Figure S4), we believe that the residual methylation is more likely to occur on a cryptic site, i.e., one that is not normally methylated in wild-type KIN. Of note, methylation by METTL21A was assayed on three Hsp70 homologs (HSPA1, HSPA5, and HSPA8), but it stands to reason that the modification would also apply to other isoforms where the lysine residue is conserved.
To further characterize the function of the methyltransferases, intracellular localization was determined by immunofluorescence. To this end, recombinant FLAG-tagged proteins were expressed in HeLa cells (Figure 4). For most methyltransferases, a marked preference for the cytoplasm was observed, although this trend is reversed in METTL21C and METT22, where the localization is predominantly nuclear. This nuclear distribution could hint at a nucleosomal methylation activity, since it is frequent with most other protein methyltransferases, but none of the four major histones appear to be methylated in vitro by the members of this family (see Figure S5). Localization of METTL18 was never determined, since no significant expression of the recombinant methyltransferase has ever been observed. It is tempting to speculate that this effect could be the result of impaired translation, since METTL18 interacts with, and probably methylates, ribosomal subunit RPL3. Whereas most methyltransferases display a diffuse distribution, METTL20 is concentrated in cytoplasmic granular foci and METTL23 displays internal membrane-like structures.
In the three instances where a methylation target has been identified, the substrate proteins (or associated protein ASPSCR1, in the case of METTL21D) bearing a GFP marker were co-expressed with the corresponding methyltransferase (Figure 4B). All methyltransferases and methyl acceptors more or less colocalize within the same cell compartments. In the case of METTL21A with HSPA8 and METTL21D with VCP or ASPSCR1, the colocalization is nearly perfect. METTL22 and KIN are both present in the nucleus, although METTL22 is clearly more concentrated in the periphery than KIN.
Methyltransferases often require cofactors to aid in the modification of their substrates. A good example of this is methylation of spliceosomal Sm proteins by the PRMT5/WD45 complex with the help of pICIn [44]. In the purification of METTL21D, we were able to identify two poorly documented VCP binding proteins, UBXN6/UBXD1 and ASPSCR1/UBXD9. This came as a surprise since VCP was shown to interact with an impressive number of cofactors including the entirety of the UBX (ubiquitin regulatory X) family [45]. Given this apparent specificity, we tested whether these proteins could act as cofactors in the methylation of VCP. As shown in Figure 5A, methylation experiments revealed that neither ASPSRC1 nor UBXN6 could be methylated directly by METTL21D but that only the addition of ASPSCR1, not UBXN6, could enhance methylation of VCP. N- and C-terminal fragments of ASPSCR1 were generated in an effort to determine which domain of the cofactor is responsible for this effect (Figure 5B). To our surprise, we observed that only the C-terminal fragment (residues 280–553), which was previously shown to interact weakly with VCP [46], could enhance VCP methylation in a similar manner as full-length ASPSCR1. An in vitro GST pull-down experiment (Figure 5C) confirms direct binding of the methyltransferase METTL21D to its substrate VCP, but also shows interaction with ASPSCR1, more specifically, to its C-terminal fragment. Furthermore, addition of VCP and ASPSCR1 or VCP and the C-terminal fragment of ASPSCR1 together appear to have a synergetic effect on binding to METTL21D, which could account for the concomitant increase in methylation signal.
Numerous mutations in the VCP gene have been linked with genetic disorders such as Inclusion Body Myopathy with Paget's disease of bone and Fronto-temporal Dementia (IBMPFD) and familial Amyotrophic Lateral Sclerosis (ALS) [47], [48]. Substitutions R155H and R191Q have been implicated in both IBMPFD and ALS. Furthermore, R159G was observed in patient with ALS, although other substitutions targeting arginine 159 were found in patients with IBMPFD (Figure 5D). In vitro methylation assay using recombinant VCP bearing these substitutions was done in order to test whether disease-causing mutations can also impact VCP methylation (Figure 5E). Although all mutant proteins appears to be methylated to a similar degree as wild-type VCP in vitro, the addition of the UBX protein no longer seems to enhance the methylation signal. These results can be explained by the notion the mutants used in this assay, as with most described VCP mutations, reside within the N-terminal domain believed to be involved in cofactor association [49]–[51]. In vitro GST pull-down experiment (Figure 5F) confirms that mutation of VCP has no impact on affinity of METTL21D for its substrate. However, when ASPSCR1 is added to the mix, the synergetic increase in binding is only observed with wild-type VCP.
Given VCP's involvement in disease, we decided to further scrutinize the functional implications of its methylation. This member of the AAA (ATPases Associated with various cellular Activities) family of ATPases contains dual ATPase domains. The methylation site falls in close proximity to the Walker B motif of VCP's first ATPase domain (Figure 6A). Knowing that Walker B motifs are usually involved in ATP hydrolysis, we hypothesized that trimethylation of lysine 315 might affect the ATPase activity of this domain. To test this idea, in vitro ATPase assays were performed with a fragment of VCP spanning its N-terminal and first ATPase domain. The reasoning behind this was that since most of VCP's ATPase activity stems from its second ATPase domain [52], if methylation of K315 only affects the activity of the first ATPase domain, we might not have detected a change in the overall activity of the full-length protein. Before going forward with in vitro ATPase assays, we first verified that the fragment could still be methylated and that methylation could be inhibited by S-adenosylhomocysteine (AdoHcy), a byproduct of methylation that also acts as a methylation inhibitor for most methyltransferases. A catalytically inactive mutated form of the methyltransferase was also created that targets the conserved acidic residue in METTL21D's AdoMet-binding motif (E73Q, see Figure 1C). The results show that the fragment is methylated as efficiently as full-length VCP (Figure 6B). Furthermore, substitution of the VCP fragment by an unmethylatable mutant (K315R), substitution of the methyltransferase by the catalytically inactive mutant (E73Q), or even addition of AdoHcy all resulted in nearly complete inhibition of methylation.
We then performed the in vitro ATPase assay and found that when a wild-type VCP fragment is pre-incubated with wild type METTL21D and the methyl donor, AdoMet, a significant decrease in ATPase activity was detected as compared to three separate control reactions where either the methyltransferase is replaced by its E73Q mutant; VCP is replaced with its unmethylatable mutant (K315R); or the methyl donor is replaced with AdoHcy (Figure 6C and 6D).
A possible interpretation of this finding is that methylation of VCP does not inhibit the ATPase activity per se, but that binding of the methyltransferase itself hinders the ATPase function. To eliminate this possibility, in vitro GST pulldowns were carried out (Figure 6E). Although we did observe a decreased binding between methyltransferase and substrate when METTL21D is replaced by the E73Q mutant, addition of AdoHcy and mutation of VCP lysine 315 do not appear to affect the interaction when compared to wild-type METTL21D and wild-type VCP in presence of AdoMet. This result confirms the conclusion that methylation of VCP directly modulates the ATPase activity of its first ATPase domain. Additionally, these experiments were repeated with a full-length form of VCP whose second ATPase domain has been inactivated by a mutation targeting a critical residue within the Walker B motif (E578Q; Figure S6) [53]. Again, a decrease in ATPase activity is observed when VCP_E578Q is preincubated with METTL21D and AdoMet.
The data presented here bring an entirely new group of protein methyltransferases into light and suggest a role for this post-translational modification in modulating chaperone function. Hsp70 isoforms have been known to be methylated both on arginine and lysine residues for quite some time [54], [55], but up until now the exact sites of these modifications and the enzymes responsible for them had remained elusive. The role of these modifications is also uncertain, but we speculate that they may help direct specificity of these chaperones towards substrates and cofactors. Evidence for this could be derived from the AP-MS data presented here. Indeed, METTL21A, the only known Hsp70 methyltransferase identified so far, copurified with a number of Hsp70 isoforms but few cofactors aside from Hsp110s. The closely related METTL21B also copurified with significant amounts of Hsp70 but this time appeared to be complexed with STIP1/Hsp90 or CCT chaperonin. That differential methylation by these enzymes drives Hsp70 specificity is a hypothesis that remains to be tested.
What is certain based on the results presented in this article is that the ATPase activity of another seemingly unrelated chaperone, VCP, can be modulated by METTL21D-dependent lysine trimethylation. As with Hsp70s, VCP has also been shown to be extensively modified, mostly by phosphorylation and acetylation [56]–[60]. In this report, we demonstrate that methylation of the VCP requires a novel, specific methyltransferase, which in turn seems to be highly conserved throughout evolution. Indeed, tandem-affinity purification of a yeast homolog of METTL21D, Nnt1p, led to the identification of the yeast homolog of VCP, Cdc48p (Figure S7 and Table S2), hinting at the importance of this interaction.
Strickingly, methylation of VCP is further enhanced by direct interaction of the methyltransferase with ASPSCR1, a poorly characterized VCP cofactor, and this effect appears to require the C-terminal half of ASPSCR1. Mutations in the VCP gene have been linked to autosomal dominant disorders Inclusion Body Myopathy with Paget's disease of bone and Fronto-temporal Dementia (IBMPFD) and familial Amyotrophic Lateral Sclerosis (ALS) [47], [48]. Most VCP mutations reside within the N-terminal domain, which has been proposed to be involved in cofactor association [49]–[51].
Substitutions R155H, R159G and R191Q have no impact on the in vitro methylation of the protein. However, addition of ASPSCR1 no longer appears to increase the methylation of mutant VCP as compared to the wild-type protein. This observation opens up a whole new area of investigation in understanding the molecular physiopathology of IBMPFD and familial ALS. It may therefore be of interest to assess the relative methylation of VCP in affected patients as compared to healthy individuals. Many studies were performed to define how these disease mutations affect the function of VCP. From a biochemical point of view, the most promising alteration concerned the increased ATPase activity that may reflect structural changes induced by ATP binding [61], [62]. Methylation of VCP by METTL21D is shown here to significantly decrease activity of the first ATPase domain of this chaperone. This modification could eventually help modulate enzymatic activity of VCP that has gone haywire due to mutation.
Our work on the KIN protein, which eventually led to the discovery of its associated methyltransferase METTL22, began when it was detected in the interactome of a number of prefoldins (see supporting data in [32]). Thus, even though KIN is not known to have chaperone activity, it still appears to interact with chaperones and potentially affect their activity. The exact function of KIN is still a matter of debate. This DNA and RNA binding protein has been assigned a role in DNA repair and/or replication [63]–[66] and possibly mRNA processing as suggested by its identification in a number of proteomic analyses of the spliceosome [67], [68]. The role of the herein identified methylation will likely go hand in hand with the function of the winged helix domain in which it resides. Interestingly, yet another winged helix-containing protein was detected in the METTL22 purification, FOXK1. In this case, the function of the winged helix is known since it is required for DNA binding of this transcription factor. If METTL22 is shown to methylate FOXK1 as it did with KIN, this may in turn point to a more complex regulation of winged helix factors.
Advances in proteomics have helped to catalog various post-translational modifications of the proteome, and it now seems evident that chaperones contain several occurrences of such modifications. Recent identification of Hsp90 methylation by lysine methyltransferase SMYD2 is further evidence of the significance of this modification in regulating chaperone function [69]. Just like post-translational modifications of histone tails were shown to modulate binding to multiple chromatin remodeling, transcription, and mRNA processing factors, we believe that chaperone modifications may compose a similar code to help define specificity of discrete subsets to their seemingly innumerable effectors. Further decrypting this “chaperone code” is now required to understand how the functional organization of the proteome is orchestrated.
Coding sequences for methyltransferases discussed in this article were obtained from the I.M.A.G.E. consortium clone library (Thermo Scientific). The corresponding cDNAs were cloned into the mammalian expression vector pMZI [70] carrying a TAP tag at its C-terminus [71], [72]. Stable human embryonic kidney cell lines (EcR-293; derived from HEK293) carrying these constructs were produced as described previously [28], [33].
Induction for 48 hours with 3 µM ponasterone A (Life Technologies) was used to express the TAP-tagged proteins. Whole cell extracts prepared from induced and non-induced stable EcR-293 cell lines were subjected to purification by the TAP procedure as described previously [28], [33].
TAP eluates were desalted and concentrated on Amicon Ultra 10K centrifugal filter units (Millipore) and resolved on NuPAGE 4–12% Bis-Tris Gel (Life Technologies). The gels were silver-stained, and the entirety of the tracks where the eluates had migrated were cut in about 20 slices that were subsequently digested with trypsin as described previously [28], [33]. The resulting tryptic peptides were purified and identified by LC-tandem mass spectrometry (MS/MS) using a microcapillary reversed-phase high pressure liquid chromatography coupled to a LTQ Orbitrap (ThermoElectron) mass spectrometer with a nanospray interface. The peak list files were generated with extract_msn.exe (version February 15, 2005) using the following parameters: minimum mass set to 600 Da, maximum mass set to 6000 Da, no grouping of MS/MS spectra, precursor charge set to auto, and minimum number of fragment ions set to 10. Protein database searching was performed with Mascot 2.3 (Matrix Science) against the human NCBInr protein database (version April 2, 2009). There are 10,427,007 sequences in this database. The mass tolerances for precursor and fragment ions were set to 10 ppm and 0.6 Da, respectively. Trypsin was used as the enzyme allowing for up to two missed cleavages. Carbamidomethyl, oxidation of methionine, mono-, di-, and trimethylation of lysine were allowed as variable modifications. In cases where multiple gene products were identified from the same peptide set, all were unambiguously removed from the data set. In the case of multiple isoforms stemming from a unique gene, the isoform with the best sequence coverage was reported. Proteins identified on the basis of a single peptide were also discarded.
Reliability assessment of protein-protein interactions was performed by our previously published software Decontaminator [73]. A set of 17 pairs of matched non-induced bait expression (control) and induced bait expression vectors was used for the algorithm training. Those were chosen on the basis of the absence of leaky expression of the tagged protein in the non-induced experiments. Decontaminator builds Bayesian probabilistic models of the Mascot scores [74] for each protein observed in the training set. It then assigns a p-value to each bait-prey interaction by computing the significance of the observed prey Mascot score compared to its corresponding control Mascot score model. A False Discovery Rate (FDR) is then calculated for each protein-protein interactions in the dataset using a leave-one-out scheme. All interactions with a FDR below 10% were reported as bait-specific, resulting in a dataset of 234 interactions. In other words, less than 24 interactions are expected to be the consequence of contamination.
The protocol was slightly modified from Inamitsu et al. [75]. Coding sequences of methyltransferases were cloned into pGEX-4-T1 vector (GE Healthcare). Coding sequences of putative methylation substrates were cloned into pET-23a(+) vector (EMD Chemicals). All vectors were transformed in One Shot BL21 Star (DE3) (Life Technologies), and protein synthesis was induced with IPTG. Bacteria were harvested by centrifugation, and pellets were lysed with the use of an IEC French Press (Thermo Scientific). The resulting GST- and 6xHis-fusion proteins were purified using Glutathione Sepharose 4B (GE Healthcare) and Ni-NTA Agarose (QIAGEN) beads, respectively, according to the manufacturers' specifications. In each reaction described in this article, 1 µg of GST-tagged methyltransferase was incubated with 2.5 µg His-tagged substrate and 5 µCi of S-[methyl-3H]-Adenosyl-L-methionine (81.7 Ci/mmol; PerkinElmer) in 50 µl of PBS for 90 min at 37°C. The samples were resolved on 10% acrylamide gels that were stained with Coomassie to show total amounts of proteins. Gels were then treated with EN3HANCE (PerkinElmer) according to the manufacturer's specifications. Tritium-based methylation signals were detected by autoradiography with four hours of exposure on Amersham Hyperfilm MP (GE Healthcare) at −80°C. Alternatively, an assay was produced with unlabeled S-adenosyl-L-methionine that was resolved on NuPAGE 4–12% Bis-Tris Gel (Life Technologies) and Coomassie stained. The bands corresponding to the His-tagged substrates were excised, trypsin digested, and analyzed as described above.
HeLa S3 (CCL-2.2 ATCC) cells were grown on Lab-Tek II chamber slides (Nalge Nunc) and co-transfected with FLAG and GFP expressing vectors (p3XFLAG-CMV-14 and pGFP2-N1; Sigma-Aldrich & PerkinElmer Life Sciences, respectively) using Lipofectamine 2000 according to the manufacturer's specifications (Life Technologies). Twenty-four hours following transfection, cells were fixed with 3.7% formaldehyde in Phosphate-Buffered Saline (PBS) and permeabilized with 0.3% Triton X-100 PBS. Slides were then fixed in 5% donkey serum PBS for 1 hour, incubated with 1∶200 monoclonal FLAG antibody (M2; Sigma-Aldrich) in 5% donkey serum PBS for 90 min, and then incubated for an additional hour with 1∶50 Cy3-conjugated donkey anti-mouse IgG secondary antibody (Jackson ImmunoResearch) in 5% donkey serum PBS. Slides were washed three times with PBS for 5 min after each step. DNA was stained with TO-PRO-3 (Molecular Probes). Slides were mounted using ProLong Gold antifade reagent (Life Technologies). Images were acquired with an LSM 700 confocal laser scanning microscope at 63× magnification and analyzed using ZEN 2010 software (Zeiss, Toronto, Canada).
The assay was based on the protocol described in Zwijsen et al. [76]. Briefly, 500 ng of GST-tagged METTL21D was incubated for 1 hour at 4°C with 100 ng of His-tagged VCP, ASPSCR1 or UBXN6, and 25 µl Glutathione Sepharose 4B beads (GE Healthcare) in 1 ml of binding buffer (50 mM NaCl, 50 mM HEPES-KOH pH 7.6, 0.1% NP-40, 0.5% charcoal-stripped FBS) complemented with complete EDTA-free protease inhibitor cocktail (Roche). The beads were washed three times by centrifugation with the same buffer and the bound proteins were eluted by boiling for 5 min in sample buffer, and separated on NuPAGE 4–12% Bis-Tris Gel (Life Technologies). Binding of His-tagged proteins was detected by Western blot analysis using mouse monoclonal anti-6X His tag-antibody (abcam). A second blot was made with rabbit polyclonal anti-GST antibody (abcam) to ensure that comparable amounts of GST-tagged baits were purified by the pull-down.
PiColorLock Gold Phosphate Detection System (Innova Biosciences) was used to quantify ATPase activity in vitro. Three micrograms of a His-tagged fragment corresponding to the first 481 residues of VCP were incubated beforehand with 2 µg of GST-METTL21D and 0.5 mM S-adenosyl-L-methionine for 30 min at 37°C in 100 µl of 0.1 M Tris pH 7.5. All assays were performed in triplicate. Independent experiments were carried out where the VCP fragment was replaced by an unmethylatable mutant (K315R); the methyltransferase was replaced by a catalytically inactive mutant (E73Q); or the methyl donor, S-adenosyl-L-methionine, was replaced by a methylation inhibitor, S-adenosyl-L-homocysteine. The rest of the protocol followed the guidelines provided by the manufacturer.
All available ortholog sequences of PRMTs and of the family of 10 putative methyltransferases in the UniProt database [77] were aligned using the ClustalW2 multiple sequence alignment software (version 2.0.12) [78]. The most conserved region of the alignment was then selected to build an unrooted phylogenetic tree through the Jalview software [79] using the neighbor-joining algorithm [80] with the BLOSUM62 substitution matrix [81]. Orthologs forming monophyletic groups with their respective human sequences were collapsed to a single node in the phylogenetic tree shown in Figure 1A.
|
10.1371/journal.pntd.0004082 | Altered Protein Expression in the Ileum of Mice Associated with the Development of Chronic Infections with Echinostoma caproni (Trematoda) | Echinostoma caproni (Trematoda: Echinostomatidae) is an intestinal trematode that has been extensively used as experimental model to investigate the factors determining the expulsion of intestinal helminths or, in contrast, the development of chronic infections. Herein, we analyze the changes in protein expression induced by E. caproni infection in ICR mice, a host of high compatibility in which the parasites develop chronic infections.
To determine the changes in protein expression, a two-dimensional DIGE approach using protein extracts from the intestine of naïve and infected mice was employed; and spots showing significant differential expression were analyzed by mass spectrometry. A total of 37 spots were identified differentially expressed in infected mice (10 were found to be over-expressed and 27 down-regulated). These proteins were related to the restoration of the intestinal epithelium and the control of homeostatic dysregulation, concomitantly with mitochondrial and cytoskeletal proteins among others.
Our results suggests that changes in these processes in the ileal epithelium of ICR mice may facilitate the establishment of the parasite and the development of chronic infections. These results may serve to explain the factors determining the development of chronicity in intestinal helminth infection.
| Intestinal helminth infections are among the most prevalent parasitic diseases and about 1 billion people are currently infected with intestinal helminths. Incidence of intestinal helminth infections is high due to both socio-economic factors that facilitates continuous re-infections and the lack of effective vaccines. In this context, further knowledge on the host-parasite relationships is required to elucidate the factors that determine the expulsion of the intestinal helminths or, in contrast, the chronic establishment of the infections. Echinostoma caproni (Trematoda) is an intestinal trematode that has been extensively used as experimental model to investigate these factors. Depending on the host species. E. caproni is rapidly rejected or develops chronic infections. Herein, we analyze the changes in protein expression induced by E. caproni infection in a host in which the parasites develop chronic infections. These data may serve to get a better understanding of the factors determining the development of chronic intestinal infections. A total of 37 spots were identified differentially expressed. These proteins were related to the restoration of the intestinal epithelium and the control of homeostatic dysregulation, mitochondrial and cytoskeletal proteins among others. This suggests that the changes in these processes in the intestinal mucosa may facilitate the development of chronic infections.
| Intestinal helminth infections are among the most prevalent parasitic diseases. Recent studies have estimated that about 1 billion people are currently infected with at least one species of intestinal helminth mainly in developing regions of Asia, Africa and Latin-America [1]. Although intestinal helminths rarely kill their human hosts, they commonly cause chronic or recurrent infections that have an important impact in health. The most common symptoms are related to effects on nutrition causing growth retardation, malabsorption syndrome and vitamin deficiencies or impaired cognitive function [2,3]. Additional abnormalities such as intestinal obstruction, chronic dysentery, rectal prolapse, respiratory complications, iron-deficiency anemia or debilitating disease can also appear [4–6]. Moreover, parasitic helminth infections in livestock are responsible for significant economic losses due to decreases in animal productivity and the cost of anthelminthic treatments of parasitized individuals [1].
About 40 million people are infected with food-borne trematodes, including members of the family Echinostomatidae, mainly in East and Southeast Asia [7]. Echinostomes are cosmopolitan parasites that infect a large number of different warm-blooded hosts, both in nature and in the laboratory. About 20 species belonging to nine genera of Echinostomatidae are known to cause human infections around the world [8,9]. They constitute an important group of food-borne trematodes of public health importance with prevalences that ranges from 3% in some areas of Asia [10,11]. Apart from their interest in human health echinostomes, and particularly Echinostoma caproni, have been used for decades as experimental models to the study of food-borne trematodes—vertebrate host relationships [12,13]. E. caproni is an intestinal trematode with no tissue phases in the definitive host [13]. After infection, the metacercariae excyst in the duodenum and the juvenile worms migrate to the ileum, where they attach to the mucosa [13]. E. caproni has a wide range of definitive hosts, although its compatibility differs considerably between rodent species on the basis of worm survival and development [12]. In mice and other hosts of high compatibility, the infection becomes chronic, while in hosts of low compatibility, (such as rats) the worms are expelled from the 2–4 weeks post-infection (wpi) [14,15]. Moreover, the consequences of the infection in each host class are markedly different. The establishment of chronic infections in CD1 mice is dependent upon a local Th1 response with elevated production of IFN-γ [16]. The infection induces important inflammatory responses, a marked epithelial injury and a rapid increase of iNOS expression [15–17]. Concomitantly with these events, chronic infections impair the processes of renewal of the intestinal epithelium inducing elevated levels of crypt-cell proliferation and tissue hyperplasia at the site of the infection [18]. In contrast, the early rejection of E. caproni is associated with the development of a local Th2/Th17 phenotype and changes in the tissue structure are not observed [16,19]. Because of these characteristics, the E. caproni-rodent model is extensively used to elucidate several aspects of the host-parasite relationships in intestinal infections, such as the induction of distinct effector mechanisms and their effectiveness in parasite clearance.
Comparative proteomic studies allow to obtain a broad view of the changes induced by a particular process, as in this case, the establishment of intestinal infections. Herein, we analyze the alterations in the protein expression induced by the E. caproni infections in the ileum of a host of high compatibility in which chronic infections are developed. This information may be useful to gain a better understanding on the factors that facilitate the development of chronic infections with intestinal helminthes and the consequences of helminth infections in hosts chronically infected.
The present study was performed using male CD1 mice weighing 30–35 g. The strain of E. caproni employed and the infection procedures have been described previously [20]. Briefly, encysted metacercariae of E. caproni were removed from kidneys and periacardial cavities of experimentally infected Biomphalaria glabrata snails and used for infection. A total of 16 mice were each infected by gastric gavage with 75 metacercariae of E. caproni. Additionally, 16 mice were left uninfected and used as uninfected s. All the animals were sacrificed at 2 weeks post-infection (wpi) to obtain tissue samples.
The animals were maintained under conventional conditions with food and water ad libitum. This study has been approved by the Ethical Committee of Animal Welfare and Experimentation of the University of Valencia (Ref#A18348501775). Protocols adhered to Spanish (Real Decreto 53/2013) and European (2010/63/UE) regulations.
Ileal sections from uninfected and infected mice were removed at necropsy and intestinal epithelial cells (IEC) were isolated as described before [21]. In brief, the ileal sections were opened longitudinally and rinsed by gentle shaking in washing buffer: ice-cold Hank’s balance salt solution (HBSS) containing 2% of heat-inactivated fetal calf serum (FCS). Supernatant was then removed and fresh washing buffer was added to the ileal sections. This step was repeated at least 4 times, until the supernatant was clear. The tissue was then cut into small 1 cm-long segments and incubated for 20 min at 37°C in HBSS containing 10% FCS, 1nM EDTA, 1mM DTT, 100 U/ml penicillin and 100 μg/ml streptomycin (dissociation buffer). The supernatant was collected and maintained on ice and the incubation was repeated a second time with fresh dissociation buffer. Supernatants were combined and filtered through a 100 nm cell strainer before IEC were pelleted out by a centrifugation at 200 g for 10 min at 4°C and washed three times in PBS under the same centrifuge conditions to remove any residual medium.
Protein extraction was performed using the M-PER Mammalian Protein Extraction Reagent (Thermo Scientific) according to the manufacturer’s instructions. Shortly, M-PER Mammalian Protein Extraction Reagent was added to the IEC pellet (20:1, v/v), mixed by vortex and incubated at room temperature (RT) for 20 min with continuous gentle agitation. The lysate was then clarified by centrifugation at 18,000 g for 15 min at 4°C, transferred into a new tube and stored at -80°C until use.
In order to increase the biological significance and avoid erroneous conclusions due to individual variations, four biological replicates were performed for each experimental group (uninfected and infected). Each biological replicate was obtained by pooling the same amount (20 μg) of protein extracted from the IEC isolated from four different mice. Then, 50 μg of protein from each biological replicate were cleaned and precipitated with the 2D Clean-up Kit (GE Healthcare), pellets were resuspended in 18 μl of a proper buffer (25 mM Tris, 7 M urea, 2 M thiourea, 4% CHAPS, pH 8,5), and proteins were fluorescently tagged with CyDye DIGE Fluor minimal dyes (GE Healthcare), following manufacturer’s instructions. One microliter of dye (400 pmol) was added to each sample and maintained on ice for 30 min in the dark. The reaction was stopped by adding 1 μl of 10 mM lysine. To minimize any dye-specific labeling artefacts, two biological replicates of each experimental group (infected and uninfected) were labeled with Cy3 and the other two were labeled with Cy5. The internal standard, prepared by mixing the same amount of protein of each sample included in the experiment, was always labeled with Cy2.
Ileal protein extracts from E. caproni-infected and uninfected mice were compared across four 2D-DIGE gels to identify proteins significantly modulated by the presence of the parasite. The four pairs of Cy3- and Cy5-labeled biological replicates (50 μg of protein each) were combined with a 50 μg aliquot of the Cy2-labeled internal standard. The mixtures containing 150 μg of protein were then separated in the first dimension, i.e. isoelectric focusing as second dimension were run following previously described protocols using the isoelectric focusing protocol for 24 cm Immobiline Drystrips. The IPG strips (24 cm, nonlinear pH 3–11) where rehydrated overnight with rehydration buffer (8 M urea, 4% CHAPS, 1% ampholytes and 12 μl/ml of DeStreak™), and the labeled samples were then applied to the strips by anodic cup loading, after the addition of DTT and ampholytes up to a final concentration of 65 mM and 1%, respectively. The isoelectric focusing was carried out at 20°C in the Ettan IPGphor 3 System (GE Healthcare) as follows: (I) 300 V for 4 h; (II) gradient to 1,000 V for 6 h; (III) gradient to 8,000 V for 3 h; and (IV) 8,000 V up to 32,000 Vh. Prior to the second dimension the strips were equilibrated in two steps, 15 min each, in equilibration buffer (50 mM Tris, 6 M urea, 30% glycerol and 2% SDS) containing either 2% DTT or 2.5% iodoacetamide, respectively. The separation of proteins in the second dimension was performed on an Ettan DALTsix system (GE Healthcare) using 12.5% polyacrylamide gels. Electrophoresis was run at 1 W/gel for 1h followed by 5 h, approximately, at 15 W/gel.
Gels were scanned in a Typhoon 9400 Variable Mode Imager (GE Healthcare) at appropriate wavelengths for each fluorophore: Cy2 (488/520 nm), Cy3 (532/580 nm) and Cy5 (633/670 nm), and at 50 μm resolution. The non-essential information was removed using ImageQuant Tools software and DeCyder v7.0 software was employed for image analysis. The differential in gel analysis module was used for automatic spot detection and abundance measurements in each individual gel, comparing the normalized volume ratio of each spot from a Cy3- or Cy5-labeled sample to the corresponding Cy2 signal from the internal standard. The data sets were collectively analyzed using the biological variation analysis module of the same software, which allows inter-gel matching and calculation of standardized average volume ratios (AVRs) for each protein spot among the 4 gels of the study. Statistical analysis was assessed for each change in AVR using Student’s t test and false discovery rate (FDR). Statistical significance was considered when p<0.05 and q<0.05, respectively. Moreover, inter gels matching of statistically different spots was confirmed manually.
Unsupervised principal components analysis (PCA) and hierarchical clustering analysis (HCA) (Euclidean) were performed using the DeCyder extended data analysis module, both on all protein spots that were present in the 4 gels of the experiment (100% presence) and the group of spots identified as significantly modified as a consequence of the infection. These multivariate analyses clustered the individual biological replicates based on a collective comparison of expression patterns from the set of proteins chosen, with any a priori knowledge of the biological reasons for clustering.
The protein spots showing greater changes in their expression levels were manually excised from the gel and washed twice with double-distilled water. Thereafter, proteins were reduced in 100mM ammonium bicarbonate containing 10 mM DTT for 30 min at 56°C, alkylated with iodoacetamide 55 mM in 100 mM ammonium bicarbonate for 20 min at RT in the dark and, finally, digested in-gel with an excess of sequencing grade trypsin (Promega) overnight at 37°C, as described before [22]. Protein digestion was stopped with 1% trifluoroacetic acid (TFA) and peptides were dried in a vacuum centrifuge and resuspended in 7 μl of 0.1% TFA, pH 2. One microliter of peptide mixture was spotted onto a MALDI target plate and allowed to air dry at RT before adding 1 μl of matrix, a 5 mg/ml solution of α-cyano-4-hydroxy-transcinnamic acid (Sigma) in 0.1% TFA and 70% acetonitrile (ACN), and left to air dry again.
The samples were analyzed in a 5800 MALDI TOFTOF (AB Sciex) in positive reflectron mode using 3000 laser shots per position. Previously, the plate and the acquisition methods had been calibrated with 0.5 μl of CM5 calibration mixture (AB Sciex), in 13 positions. For the MS/MS analysis, 5 of the most intense precursors were selected for each position, according to the following threshold criteria: a minimum signal‐to‐noise of 10; a minimum cluster area of 500; a maximum precursor gap of 200 ppm and a maximum fraction gap of 4. MS/MS data was acquired using the default 1kV MS/MS method. Several spots could not be identified by MALDI TOFTOF, however, so liquid chromatography and tandem mass spectrometry (LC-MS/MS) was performed. Five microliters of each sample were loaded onto a trap column: NanoLC Column, 3 μ C18-CL, 350 μm x 0.5 mm (Eksigen) and desalted with 0.1% TFA at 3 μl/min for 5 min. The peptides were then loaded onto an analytical column: LC Column, 3 μ C18-CL, 75 μm x 12 cm (Nikkyo), equilibrated with 5% ACN, 0.1% formic acid (FA). Elution was carried out with gradient of 5 to 45% B in A for 15 min (A: 0.1% FA; B: ACN, 0.1% FA) at a constant flow rate of 300 nl/min. Peptides were analyzed in a mass spectrometer nanoESI qQTOF (5600 TripleTOF, AB Sciex). The tripleTOF was operated in information-dependent acquisition mode, in which a 0.25-s TOF MS scan from 350–1250 m/z was performed, followed by 0.05-s product ion scans from 100–1500 m/z on the 50 most intense 2–5 charged ions.
Both MS-MS/MS and LC-MS/MS data were sent to MASCOT 2.5 (Matrix Science) via ProteinPilot (AB Sciex) and database search was performed on NCBInr (non-redundant) database with taxonomy set to Metazoa. Searches were done with tryptic specificity, allowing one missed cleavage and a tolerance on mass measurement of 100 ppm in MS mode and 0.8 Da for MS/MS ions. Carbamidomethylation of cysteine was used as fixed modification and oxidation of methionine and deamination of asparagine and glutamine as variable modifications. A protein identification was considered accurate when at least three peptides were identified with an overall MASCOT score greater than 50.
Functional classification and intracellular localization of the identified proteins were assessed using the KEGG Pathway (http://www.genome.jp/kegg/pathway.html) and UniProtKB resource (http://www.uniprot.org/). A cytoscape plugin, the Biological Networks Gene Ontology (GO) tool (BiNGO 2.3) was used to identify overrepresented biological processes GO terms [23]. Settings for BiNGO included using a hypergeometric test with a significance threshold of 0.05. The P-values were corrected for multiple testing by the Benjamini & Hochberg correction.
A 2D-DIGE proteomic analysis was performed on whole ileal cell extracts from eight biological replicates corresponding to E. caproni-infected and uninfected mice (4 replicates each) and 2D-gel images were then subjected to computational analysis using the DeCyder software (S1 Fig). Both univariate and multivariate statistical analysis indicated that E. caproni infection induces an intense remodeling of protein expression pattern in the ileal mucosa of mice early after the establishment of the infection.
The inter-gel spot matching, carried out using biological variation analysis module, revealed a total of 1,698 well defined spots with a 100% of presence, i.e. found in each gel included in the 2D-DIGE experiment (Fig 1). The average abundance of each spot was then calculated and significant changes were evaluated. A total of 876 spots, representing a 51.6% of the total number of spots, showed significant variations within a 95% confidence interval in both Student’s t test (p<0.05) and FDR (q<0.05). In view of the large number of differentially expressed proteins, different selection criteria were sequentially applied in order to select for protein identification those spots whose expression was mostly affected as a consequence of the infection. In a first step, the confidence interval in the Student’s test was reduced to 99%, yielding a total of 361 spots showing a value of p<0.01. To guarantee the proper comparison of spots among gels, the correspondence of these 361 spots among all the gels were manually validated, and 148 were unambiguously confirmed (68 up-regulated in the ileum of infected mice and 80 down-regulated). Forty-seven of these spots showed an AVR greater than or equal to |2.00| in the four gels analyzed (17 overexpressed and 30 downregulated at 2 wpi). Finally, 37 of these spots (11 and 26 up and downregulated, respectively) could be extracted from the gel and successfully identified by MS. Fig 1 summarizes the results of applying the consecutive selection filters from the initial set of spots with 100% of presence to those that were eventually identified by MS.
In order to establish the biological significance of the infection-induced protein changes, multivariate statistical tests were performed on the proteins identified by 2D-DIGE. PCA and HCA were carried out on both the total number of spots with 100% presence in the experiment (1,698) and those displaying larger significant statistical differences between uninfected and infected mice (361 spots). As shown in Fig 2, both PCA and HCA applied to the set of 1,698 spots with 100% presence were able to separate graphically the two groups of samples. In the PCA the two groups were separated in the basis of the first principal component. Moreover, all biological replicates were within the range of normality (95% of confidence), discarding the existence of outliers among the samples (Fig 2). Similarly, HCA applied to the same set of protein spots grouped the replicates in two main categories according to their condition of infected or uninfected. The heat diagram shows that protein expression patterns displayed by uninfected and infected animals were clearly different, suggesting that E. caproni infection induces a significant change on IEC (S2 Fig). In this first analysis, however, the proteins could not be clustered according to their expression pattern indicating the existence of a wide variability when the spots are compared individually (S1 Fig). This is not strange since both, significantly and non-significantly differentially expressed spots were selected for the analysis and their correspondence among all the gels had not been manually validated.
As expected, when multivariate statistical tests were performed on the set of 361 spots differentially expressed between uninfected and infected mice (p<0.01 and q<0.05), both PCA and HCA also separated the biological replicates into two different groups. In the PCA the samples were separated by the first principal component, indicating that this set of proteins is enough to explicate the differences between the two groups (Fig 3). Similarly, HCA grouped replicates in two categories each including infected or uninfected samples (Fig 4). In this case, the protein spots were also classified in two main categories according to their expression pattern, i.e. up- or down-regulated in one group of samples respect the other one (Fig 4). This confirms that, E. caproni significantly alters the protein expression pattern of the IEC, affecting a large number of proteins 2 weeks after the infection.
In view of the large number of proteins significantly affected by the E. caproni infection, those spots displaying a greater difference between uninfected and infected animals were selected for identification by MS and database search. A total of 37 from 47 spots with p<0.01, q<0.05 and AVR≥|2| were accurately identified: 11 of them overexpressed in the ileum of infected mice and 26 down-regulated as a consequence of the infection (S1 Fig). These 37 spots corresponded to 31 different proteins (10 up-regulated and 21 down-regulated), since 6 redundancies were detected (Table 1). This can be attributed to different post-translational modifications, the existence of isoforms or to protein modifications during sample preparation [24]. Identified proteins are classified in Table 1 according to their function, with detailed information comprising accession number, 2D-DIGE-related data, cellular role, localization and identification parameters.
An analysis of the GO biological process of the proteins presenting an up-regulated or down-regulated expression was performed using the plugin BiNGO with Cytoscape. Proteins overexpressed in the ileum of infected mice were related to three main processes: Lipid and fatty acid metabolism, lipid and fatty acid transport and digestion and intestinal absorption (Fig 5). In contrast, proteins with a down-regulated expression in the intestine of infected mice were related to different biological processes such as energy and cell respiration, regulation of inflammatory responses, oxidative stress and lipid and fatty acid metabolism among others (Fig 6).
After a manual annotation of the proteins using the Uniprot database, the group of proteins that became more altered as a consequence of the infection was related to the energy metabolism. Our proteomic data suggests that mitochondrial function (particularly energy and cell respiration processes) is markedly reduced in the ileum of E. caproni-infected mice (Fig 6). Significant down-regulation of a component of pyruvate dehydrogenase complex (PDH) and the subunit α of NAD+-dependent isocitrate dehydrogenase 3 (IDH3) was detected (AVRs: -2.03 and -2.15, respectively). The mitochondrial PDH complex catalyzes the conversion of pyruvate to acetyl coenzyme A (acetyl-CoA), linking glycolysis to Krebs cycle. IDH3 is a mitochondrial matrix enzyme that catalyzes the rate-limiting step of the Krebs cycle, the oxidation of isocitrate to oxalosuccinate. Alteration of these processes deteriorate mitochondrial ATP production causing energy depletion [25]. In the case of an inefficient oxidative phosphorylation, the mitochondrial fat oxidation pathway becomes important in providing an alternative source of energy. The β-oxidation of fatty acids appears to be also affected in the IEC isolated from infected mice since enoyl-CoA hydratase, a mitochondrial enzyme that catalyzes the second step of each cycle of β-oxidation, was down-regulated after infection (AVR: -2.00). Moreover, fatty acids metabolism and, consequently, energy production were affected as a result of a reduction in the carnitine biosynthetic pathway. Carnitine is required for energy metabolism since it enables activated fatty acids to enter the mitochondrial matrix. In the ileum of E. caproni-infected mice, reduced expression of 4‐trimethylaminobutyraldehyde dehydrogenase, an enzyme involved in carnitine biosynthesis that catalyzes the conversion of 4-trimethylaminobutirate to γ-butirobetaine, was also noted (AVR: -3.39). Although the last step of carnitine biosynthesis from γ-butirobetaine occurs in the liver, precursor metabolites are absorbed in intestine and kidneys and transformed into γ-butirobetaine that is converted into carnitine [26]. Thus, reduced intestinal biosynthesis of carnitine-precursor metabolites affects mitochondrial import of fatty acids, reducing β-oxidation and favoring their cytosolic accumulation. This is supported by the fact that fatty acid binding proteins (FABPs) and apolipoprotein (Apo) A-I are overexppressed in the ileum of infected mice.
In the small intestine, both liver and intestinal FABPs (LFABP and IFABP, respectively) are expressed in villus enterocytes. In the ileum of E. caproni-infected mice both IFABP (AVR: +3.20) and LFABP (AVR: +2.54) overexpression occurs concomitantly with mitochondrial dysfunction and down-regulation of enoyl-CoA hydratase. In IEC, FABPs have been also proposed to have a role in the regulation of intracellular levels of unbound fatty acids [27–29], which can be toxic for the cells [30]. In our study, their overexpression may be a collateral consequence of the reduced mitochondrial metabolism and the subsequent accumulation of fatty acids in the cytosol of enterocytes. Increased expression of Apo A-I, the major protein component of high-density lipoprotein (HDL), was also detected in the ileum of infected mice (AVR: +3.45). The intestine can also act as a source of Apo A-I [31], which is incorporated to form mature chylomicrons that transport the exceeding lipids to other tissues such as adipose, cardiac or skeletal muscle [32].
All these metabolic alterations suggest that in the ileum of infected mice, enterocytes display a limited respiration and ATP production through the oxidative phosphorylation system, which can induce a metabolic shift to obtain energy from alternative metabolic processes. Indeed, parallel to mitochondrial dysfunction, lactate dehydrogenase (LDH) overexpression was detected in the ileum of infected mice (AVR: +2.70). LDH up-regulation associated to PDH down-regulation has been previously described [33,34] and involves a crucial shift in cell metabolism to prevent pyruvate accumulation and the consequent stop of glycolysis. This shift in the cellular energy supply is a signature of oxidative stress-induced cellular senescence [35]. Since oxidative phosphorylation is a more efficient mechanism for ATP production than anaerobic metabolism, such change in energy metabolism is normally accompanied by an increase in the glycolytic flux [33–35]. Our results revealed a simultaneous increase in the expression of two different isoforms of the glycolytic enzyme enolase 1, also named α-enolase, (AVRs: +2.00 and +2.74, respectively).
Mitochondrial dysfunction is associated with a number of medical disorders and ageing and may be a major mechanism underlying the development of mitochondria-related diseases consisting in an increase in the intracellular oxidative stress [36]. Increased production of reactive oxygen and nitrogen species (ROS and RNS, respectively) lead to an elevation in nitroxidative stress, which can oxidatively damage mitochondrial DNA, lipids and, primarily, proteins and, ultimately, induce tissue injury and cell death [36]. The establishment of chronic E. caproni infections in mice is known to be associated with the development of early and strong local inflammatory responses and tissue damage concomitantly with elevated mRNA levels of IFN-γ and iNOS [15,16].
Increased ROS production is commonly associated with failings in the mitochondrial respiratory chain that reduce effective oxidative phosphorilation and increase the leakage of electrons and the formation of reactive species [36]. According to our proteomic data two different isoforms of both α and β subunits of the electron transfer flavoprotein (ETF) were found to be down-regulated after infection (AVRs from -2.16 to -2.91), which is likely to be related to the infection-induced impairment of mitochondrial metabolism.
Furthermore, decreased expression of the cytochrome c oxidase (CcO) subunit IV isoform 1 (IV-1) was observed in the ileum of infected mice. Although CcO activity can be regulated at several levels, the subunit IV has been shown to be a key regulatory subunit in response to ATP and O2 levels [37]. At high ATP demand CcO IV-1 can be replaced by isoform 2 (IV-2) at the expense of ROS production [38]. Thus, the down-regulation of CcO IV-1 suggests that CcO activity is regulated through the modification of the expression of subunit IV isoforms in response to the infection. The gene expression of IV-2 is induced under hypoxic and toxic condition, and is up-regulated via hypoxia inducible factor 1 alpha (HIF-1α) [38,39]. In ischemic and inflammatory diseases of the intestine, the activation of HIF-1α in epithelial cells plays a protective role through the regulation of genes involved in the maintenance of epithelial tight barrier and mucosal immune response [40–42]. HIF-1α have been shown to correlate with PDH dysfunction and have a major role in promoting the shift of cell metabolism to anaerobic glycolysis [43], which agrees with our proteomic data. Although the mechanisms leading to inflammation-mediated hypoxia are not fully understood, most likely it involves vasculitis and edema [44]. Additionally, neutrophil migration into the intestinal mucosa is critical in depleting local O2 and activating HIF-1α [42]. Neutrophil infiltration at the site of the infection is also characteristic in the high-compatible host [15,16]. Therefore, in the ileum of infected mice, inflammation and neutrophilia may lead to overexpression of HIF-1α that, in turn, contributes to shift the enterocyte metabolism and the exhibition of a senescent phenotype. All this suggests that HIF-1α can be a key mediator in regulating the metabolic changes and controlling the intestinal pathology in response to E. caproni infection in mice, which we consider to merit further attention in future studies.
Finally, ROS accumulation would be favored by the down-regulation of the antioxidant enzyme manganese superoxide dismutase (MnSOD), a mitochondrial matrix and intermembrane space protein that transforms the highly reactive O2·− into H2O2 and O2. Inactivation of MnSOD gene in mouse induces mitochondrial disease associated with ROS toxicity and apoptosis [45]. Although antioxidant enzymes are generally believed to be up-regulated in response to an oxidative stress [46], a positive role for reduced expression of MnSOD in uninfected ling the homeostatic dysregulation of the intestinal tissue during E. caproni chronic infection is discussed below.
Overall, the infection-induced alterations on the IEC metabolism suggest that E. caproni infection induces a rapid and intense mitochondrial dysfunction, mainly characterized by the shift of cell metabolism to an anaerobic use of glucose. These changes seem to be consequence of the oxidative stress induced by the overexpression of IFN-γ and iNOS in the intestinal mucosa of infected mice.
Intestinal tissue hyper-proliferation is a hallmark response to E. caproni chronic infection [18], and the results obtained herein strongly support this observation. E. caproni infection in mice is characterized by an intense tissue damage in the ileum caused by both the parasites and the local inflammatory response developed against the infection [15–17]. Moreover, villi tip erosion and gaps in the epithelial line are common at the site of infection [15,17]. Despite these facts, tissue necrosis is not developed suggesting that epithelial restitution mechanisms work actively in an attempt to restore the constant tissue damage. Herein, we have seen that E. caproni infections in mice induce alterations in several proteins implicated in the IEC proliferation and epithelial restitution (Fig 6).
Galectin 2 (Gal2) was found to be overexpressed in the ileum of infected mice (AVR: +2.20). Galectins are a family of lectins play a major role in re-epithelialization of wounded tissues [47–51]. In addition to Gal2, an EF-hand domain containing protein (EFhd2, also named swiprosin-1) was among the most up-regulated proteins in the ileum of infected mice (AVR: +2.36). EFhd2 is also up-regulated under inflammatory conditions [52]. EFhd2 is found together with actin and actin-binding proteins modulating bundling and cell spreading [53] and actin remodeling [54], respectively. During epithelial restitution, an extensive reorganization of the actin cytoskeleton is needed [55], suggesting that EFhd2 may play a role in this process. Both IFN-γ and nitric oxide (NO) have been shown to impair IEC migration through different mechanisms [56–58], thus Gal2 and EFhd2 appear as potential candidates to direct epithelial restitution under inflammatory conditions in the ileum of E. caproni-infected mice.
Once epithelial restitution has started, augmented cell proliferation is required to provide new enterocytes to restore the damaged area and the results obtained herein reveal that several pathways are involved in the regulation of tissue hyper-proliferation. The downregulation of the antioxidant enzyme MnSOD plays a role in this process (Fig 6). Apart from its function in controlling oxidative damage, MnSOD also plays a role on tissue renewal, since MnSOD genetic deficiency promotes cell turnover [59]. In our study, this downregulation may be one of the mechanisms responsible for infection-induced cell hyper-proliferation during the establishment of chronic infections. A striking feature is that changes in MnSOD expression are commonly accompanied by increased levels of ornithine decarboxylase (ODC). ODC was not found to be among the most altered proteins. However, three isoforms of the ornithine aminotransferase (OAT), which is involved in the catabolism of L-ornithine, the substrate of ODC, were found to be markedly down-regulated (AVR: -3.94, -2.64 and -2.14). Another enzyme involved in ornithine catabolism, ornithine carmaboyltransferase (OCT), was also down-regulated (AVR: -2.26). Ornithine is synthesized from arginine (Arg) by the cytosolic isoform of the enzyme arginase and is a necessary metabolite for the synthesis of polyamines and prolines. In addition to ornithine synthesis, Arg can be also metabolized by nitric oxide synthase (NOS) to generate NO and L-citrulline, so that arginase/NOS balance is determinant to displace Arg metabolism to one or another pathway [60]. As mentioned above, E. caproni infection in CD1 mice is characterized by increased mRNA expression of iNOS [16]. Hence, in the ileal epithelium of infected mice Arg metabolism can be expected to be displaced to the production of NO and L-citrulline at the expense of ornithine synthesis. Neither arginase nor iNOS appeared to be primarily affected during E. caproni infection in mice at proteomic level. The down-regulation of ornithine catabolic enzymes (i.e. OAT and OCT) may allow the increase in ornithine bioavailability for polyamine synthesis through ODC in milieu in which ornithine biosynthesis is diminished due to the displacement of Arg metabolism. Polyamines (putrescine, spermidine and spermine) are small, polycationic, organic molecules, synthesized from ornithine via ODC, which are mandatory to cell proliferation [61]. In inflammatory models, NO production is considered to be an early phase response, whereas the production of polyamines occur in the repair-phase response after iNOS inhibition by agmatine aldehyde [62]. However, in E. caproni chronic infections, both iNOS overexpression and crypt-cell hyper-proliferation occur early and exacerbate over the course of the infection [16,18]. Thus, the increase of ornithine bioavailability for polyamine synthesis through the down-regulation of enzymes responsible for its use in other metabolic pathways may represent a different route to guarantee tissue repair in the presence of sustained elevated levels of NO production during chronic infections.
We also have found increased expression of several proliferation markers, such as keratin (K19) (AVR: +2.89), aminoacylase 1 (ACY1) (AVR: +2.23) and sulfotransferase (SULT)1B1 (AVR: +2.34) in addition to two isoforms of α-enolase (AVR: +2.74 and +2.00). K19 is a marker of the gut morphogenesis, as it is mainly expressed in proliferative crypts [63]. Its overexpression in the ileum of infected mice is consistent with the crypt hyperplasia developed [18]. Similarly, the activity of the cytosolic enzyme ACY1, responsible for the deacylation of α-acylated amino acid residues during intracellular protein catabolism, is greater in the crypt areas than in the villous portion of small intestine [64]. Although the biological function of SULT1B is not well defined, high levels of its mRNA expression are detected in human fetal small intestine [65]. Its overexpression suggests that it play a role in cell proliferation, differentiation and/or tissue structural organization in the small intestine.
Apart from its role as a glycolytic enzyme, α-enolase serves as a plasminogen receptor on the surface of a variety of cells activing plasmin [66,67]. Plasmin-enolase interactions are involved in promoting cell migration in pathophysiological processes, such as the inflammatory response, cell invasion and cancer metastasis [68,69]. mRNA expression of α-enolase increases in growing cells, but remains almost at an undetectable level in the stationery/resting/quiescent phase [70] and, at protein level, it was found to be around 2-fold in proliferating versus differentiated human keratinocytes [71]. In the gastrointestinal tissue, α-enolase overexpression has been found in Helicobacter pylori-infected gastric mucosa, both at mRNA and protein levels [70,72], as well as in ulcerative colitis [73]. In E. caproni-infected mice, the elevated expression of α-enolase in the ileal enterocytes may well be a marker of intestinal inflammation and/or tissue overproliferation. This protein was also found to be overexpressed in the ileum of E. caproni-infected rats [21] in which increased epithelial cell renewal occurs in the absence of inflammatory responses [16–18], suggesting that α-enolase may have a key role in restoring homeostasis of injured intestine.
We have also found that several proteins implicated in the regulation of cell death became altered in the ileum of mice after E. caproni infection (Fig 6). In particular, proteins related to the mitochondrial-driven apoptotic pathway were affected, suggesting that the intrinsic pathway is activated because of the infection. This is of importance since in a context with increased cell proliferation, elevated levels of cell death are required to maintain tissue homeostasis and prevent massive dysregulation [74]. Moreover, prolonged inflammation and wound healing represent a high risk for DNA damage and malignant transformation and defective cells need to be rapidly eliminated [75].
Among the down-regulated proteins implicated in cell growth and apoptosis, we found MnSOD. It has been previously shown that MnSOD deficiency increases cell turnover via AP-1- p53-mediated pathways [59]. Moreover, the down-regulation of proteasome subunit alpha type 1 (PSMA1) may also affect the levels of p53. PSMA1 plays a role in gating the entry of proteins into the proteasome and its overexpression has been involved in tumor genesis [76,77]. PSMA1 is an important regulator of proteasome-mediated proteolysis, with a key role in cancer development and/or progression through modulation of p53 and nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) signaling, which can play a key role in the control of intestinal tissue hyperplasia and homeostatic dysregulation after E. caproni infection. Proteasome inhibition has been associated with increased intrinsic apoptosis by different mechanisms. The availability of p53 increases since this protein is degraded through the ubiquitin-proteasome pathway [78,79]. Moreover, proteasome dysfunction also affects NF-κB through the stabilization of the inhibitory subunit IκB-a [80]. PSMA1 down-regulation has been detected in colorectal cancer cells after treatment with caffeic acid phenethyl ester [81] and this has been shown to induce apoptosis of cancerous cells through the inhibition of NF-κB signaling [82,83].
As mentioned before, the decreased expression of MnSOD in a milieu of elevated oxidative stress is surprinsing. However, it has been shown that this enzyme is transcriptionally regulated by NF-κB [84]. Suppression of NF-κB translocation results in reduction of MnSOD expression leading to ROS accumulation and cell death [85]. Therefore, the down-regulation of this antioxidant enzyme observed herein may be consequence of NF-κB repression and is likely to play a positive role in the ileum of infected mice, promoting ROS-mediated programmed cell death to counteract homeostatic dysregulation.
Alterations in different structural proteins were also noted. Together with the augmented expression of crypt-specific K19, a decrease in the expression of type II cytokeratin 8 (K8) was detected (AVR: -2.87). Keratins are structural proteins that associate to form non-covalent tissue specific heteropolymers (i.e. type I and type II keratins) that build up the intermediate filament cytoskeleton of epithelial cells. K8 is the main type II keratin present in mature enterocytes from the opening of the crypts to the villi apices [86]. K8-/- mice showed a lack of intermediate filament cytoskeleton in small IEC, with differentiated enterocytes displaying progressive loss of apical membrane-associated proteins and alterations in microtubule organization [87]. Although tissue functional deficiencies were not observed in K8-/- mice, it was noted that mature IEC lacking intermediated filament cytoskeleton displayed shortened microvilli and they seem to be unable to fully recover from tissue injury [87]. Moreover, small intestine enterocytes from K8-/- mice appear to be more predisposed to apoptosis compared to those obtained from K8+/+ mice [88]. In the ileum of E. caproni-infected mice, this cytoskeletal deficiency may convert the infected epithelium even more sensitive to both parasite- and immune-mediated tissue damage, accelerating the induction of cell death.
A marked downregulation of the structural nuclear protein lamin B was also observed in mice after E. caproni infection (AVR: -2.16). Lamins, A- and B-types, are the major components of the nuclear lamina and, in addition to their role as structural proteins, these type V intermediate filament proteins contribute to nuclear envelope integrity [89]. A number of studies have linked B-type lamins to several aspects of cell physiology such as transcription, replication, spindle assembly, chromatin organization, resistance to oxidative stress or regulation of cell senenscence [90–93]. The decrease in lamin B1 expression occurs in response to stimulation of either p53 or pRB tumor suppressor pathways and induces inhibition of proliferation and premature senescence [92,93]. However, altered lamin expression is common in gastrointestinal neoplasms and reduced expression of either lamin A/C or lamin B1 is a marker of potential malignancy in the gastrointestinal tract in humans [94], thus their role in the infection with E. caproni needs to be further characterised. Finally, the expression of two isoforms of the mitochondrial elongation factor Tu (TUFM) was also found to be down-regulated in IEC from infected mice (AVR: -2.87 and -3.56, respectively). TUFM is one of the major mitochondrial biogenesis regulating proteins, which has been found to be down-regulated during ageing in muscle cells [95]. Moreover, its inhibition induces mitochondrial dysfunction and increased cell death in cancer cells [96], which is fully consistent with the results observed herein.
Despite the fact that proteomic data strongly support the idea of elevated levels of IEC death, tissue hyperplasia develops in the ileum of E. caproni infected mice [18], thereby suggesting that mitochondrial dysfunction and premature cellular senescence are not enough to equilibrate cell proliferation and death rates. In primary cultured hepatocytes, it has been shown that high amounts of IFN-γ-induced ROS are not sufficient to induce cell death, but a combination of ROS and proper endoplasmic reticulum (ER) stress responses is required to induce apoptosis [97]. In this sense, we have found protein disulfide isomerase A3 (PDIA3, also known as ERp57 or 1,25D3-MARRS) to be down-regulated in the ileum of infected mice (AVR: -2.27). PDIA3 is a stress-responsive protein, which is involved in protein folding, glycoprotein quality control and the assembly of the major compatibility complex class I in the ER [98]. Therefore, the lack of proper ER stress responses may be responsible for the low rate of IEC death and the development of tissue hyperplasia, despite premature senescence is induced in mature enterocytes in response to the infection. Nevertheless, in addition to the ER, PDIA3 is present in many other subcellular locations, which makes it difficult to predict the effects of its down-regulation over the course of the intestinal infection [98].
Constant wound repair represents an elevated risk for DNA damage and genomic instability in proliferating cells, promoting the development of a tumorigenic environment, with chronic inflammation being the most important risk factor [99]. Moreover, a continuous state of chronic inflammation and wound healing have been regarded as the key events for cancer development in other chronic helminth infections [100,101]. Our proteomic data suggest that both pro-tumorigenic (i.e. inflammation-mediated oxidative stress, cell hyper-proliferation) and anti-tumorigenic mechanisms (i.e. cellular senescence, apoptosis) are activated early after infection in E. caproni-infected mice. Malignant tumors are often developed at sites of chronic injury, re-epithelialization and inflammation. Thus, according to our results, persistent damage of the intestinal epithelium in long-lasting infections could represent a risk factor for cancer development.
The proteomic alterations described herein can be directly associated with the chronic establishment of the parasite in hosts of high compatibility. These changes are markedly different to those observed in the ileum of infected rats, in which the parasite is rejected a few weeks after infection. A previous study [21] showed that the effects of E. caproni infection on the IEC of rats are low in comparison with mice, mainly inducing the overexpression of proteins related with the cytoskeleton and the maintenance of the functional integrity of the epithelial barrier (e.g. actin, T-plastin, both 8 and 19 cytokeratins or annexin A4). Consequently, changes on the absorptive/secretory function of enterocytes and, especially, an increased regenerative capacity of the intestinal epithelium appear to be potentially IL-13-mediated effector mechanisms involved in the early rejection of worms in rats. In contrast to mice, a strict control of proliferation and programmed cell death seems to be essential to maintain the intestinal homeostasis in rats, hence protecting the host against the injurious effects of the infection. This is consistent with the overexpression of the intestinal proliferation marker K19 and chaperones such as a heat shock cognate 71 KDa and BiP, together with the down-regulation of peroxiredoxin 3, prohibitin and 14-3-3 zeta isoform [21].
Moreover, proteomic data indicate that cellular energy metabolism becomes differentially modified in the ileum of E. caproni-infected mice and rats. Whereas in mice the intestinal infection induces mitochondrial dysfunction and an increase in the anaerobic use of glucose to yield ATP, in rats the transition to a more aerobic and oxidative metabolism is suggested, leading to a reduced glycolytic flux and overall ATP production [21]. These alterations in energy metabolism could be of relevance for a better comprehension of the mechanisms involved in the control of infections on mucosal surfaces.
In summary, our results indicate that the presence of the parasite induces a rapid and profound remodeling in the protein expression pattern of IEC, associated with the development of inflammation and oxidative stress. The identification of those proteins whose expression was mainly altered indicates that the cellular processes that become primarily affected after E. caproni infection in CD1 mice are related to the restoration of the damaged intestinal epithelium and the control of homeostatic dysregulation. Wound healing and crypt-cell hyper-proliferation appear to be constitutively active processes from the early stages of the infection. Concomitantly, mitochondrial dysfunction and cytoskeletal changes indicate that cellular senescence is induced on mature enterocytes. These facts, together with the pro-apoptotic changes observed, suggest that programmed cell death is augmented in the ileal mucosa of infected mice. Although previous studies have shown that proliferation and cell death are not well balanced in the ileum of infected mice, and the IEC turnover is diminished after infection, augmented cell death may be essential to control the level of homeostatic dysregulation in the gut and eliminate potentially damaged cells, which may conduct to malignant transformation.
|
10.1371/journal.pgen.1004654 | Mechanism of Suppression of Chromosomal Instability by DNA Polymerase POLQ | Although a defect in the DNA polymerase POLQ leads to ionizing radiation sensitivity in mammalian cells, the relevant enzymatic pathway has not been identified. Here we define the specific mechanism by which POLQ restricts harmful DNA instability. Our experiments show that Polq-null murine cells are selectively hypersensitive to DNA strand breaking agents, and that damage resistance requires the DNA polymerase activity of POLQ. Using a DNA break end joining assay in cells, we monitored repair of DNA ends with long 3′ single-stranded overhangs. End joining events retaining much of the overhang were dependent on POLQ, and independent of Ku70. To analyze the repair function in more detail, we examined immunoglobulin class switch joining between DNA segments in antibody genes. POLQ participates in end joining of a DNA break during immunoglobulin class-switching, producing insertions of base pairs at the joins with homology to IgH switch-region sequences. Biochemical experiments with purified human POLQ protein revealed the mechanism generating the insertions during DNA end joining, relying on the unique ability of POLQ to extend DNA from minimally paired primers. DNA breaks at the IgH locus can sometimes join with breaks in Myc, creating a chromosome translocation. We found a marked increase in Myc/IgH translocations in Polq-defective mice, showing that POLQ suppresses genomic instability and genome rearrangements originating at DNA double-strand breaks. This work clearly defines a role and mechanism for mammalian POLQ in an alternative end joining pathway that suppresses the formation of chromosomal translocations. Our findings depart from the prevailing view that alternative end joining processes are generically translocation-prone.
| The reason for the hypersensitivity of POLQ-defective mammalian cells to ionizing radiation has been elusive. Here we show that POLQ-defective mammalian cells are selectively susceptible to double-strand breaks in DNA. We present experiments in mammalian cells showing that a specific double-strand break repair pathway is POLQ-dependent. To analyze the repair function in more detail, we examined class switch joining between DNA segments in antibody genes. Insertions of DNA bases are sometimes found at the joins between such segments, but the origin of these insertions has been mysterious. We show that this class of insertion joins during immunoglobulin class-switching is entirely POLQ-dependent. In experiments with purified human POLQ protein, we found a novel biochemical mechanism explaining the formation of the insertions. POLQ has a unique biochemical ability to extend DNA with minimal base pairing. Finally, we examined the biological consequences for chromosome stability. Unexpectedly, the Burkitt lymphoma translocation (a major cancer-associated genome instability) is enhanced in the absence of POLQ. This alters the current view about the action of DNA end joining in mammalian cells, revealing that a POLQ-dependent DNA repair pathway combats potentially damaging chromosome translocations.
| A diverse group of at least 16 DNA polymerases carry out DNA replication, repair, and damage tolerance in the mammalian genome [1], [2]. One of these is DNA polymerase theta (POLQ). POLQ homologs are found in multicellular eukaryotes including plants, but an equivalent enzyme is absent from yeast [3]. The large 290 kDa human POLQ protein has an unusual bipartite structure with an N-terminal helicase-like domain and a C-terminal DNA polymerase domain [4]. This domain arrangement and the POLQ protein sequence is highly conserved in vertebrates [3].
Several functions have been suggested for POLQ [3] including bypass of template DNA lesions such as abasic sites and thymine glycols [5], [6], a backup role in DNA base excision repair [7], [8], and influencing the timing of DNA replication origin firing [9]. Loss of POLQ homologs in Drosophila and C. elegans causes hypersensitivity to DNA interstrand crosslink (ICL)-forming agents [10], [11] such as nitrogen mustards or cisplatin. A consistent picture of hypersensitivity to DNA damage in mammalian cells lacking POLQ has not emerged from studies reported so far [3]. Mice devoid of or carrying mutant alleles of Polq display an elevated level of micronuclei (indicating chromosome breaks) in their peripheral erythrocytes [12]–[14]. A further increased frequency of micronuclei is observed after ionizing radiation exposure, and is much elevated in Polq mutant animals [12], [14]. The majority (∼90%) of mice with double homozygous deficiencies in Polq and Atm die during the neonatal period, with surviving double mutant mice showing severe growth retardation [13]. From this observation it was suggested that POLQ has a unique role in maintaining genomic stability that is distinct from the major homologous DNA recombination pathway regulated by ATM [13].
DNA double-strand breaks (DSBs) can be formed in cellular genomes by environmental agents such as ionizing radiation. DSBs also arise during normal cellular duplication cycles, when DNA replication stalls at naturally occurring structures or at sites of internally-generated DNA damage. In diversification steps of the mammalian immune system, DSBs are deliberately formed by regulated enzymatic action, to initiate rearrangement of antibody and receptor segments, and as a means to introduce local variation. Because DSBs are toxic and/or mutagenic if not repaired, organisms have multiple mechanisms for DSB repair [15], [16]. The primary strategies are end-joining mechanisms, which process and rejoin the ends of a DSB, and homologous recombination (HR) pathways which employ an undamaged copy of the DNA [17]. End-joining pathways appear to be the first line of defense again DSBs. The most studied pathway is “classical” non-homologous end-joining (cNHEJ), which relies on the DNA-binding Ku70 (XRCC6) and Ku80 (XRCC5) gene products, and the DNA protein kinase (DNA-PK, PRKDC). One or more “alternative” end-joining pathways (altEJ) also exist, which are independent of these factors [18], [19]. During immunoglobulin diversification, the regional end-joining process of class switch recombination (CSR) replaces one constant region coding exon for another. This CSR process is known to occur through both cNHEJ and alternative end joining pathways [20]. In mammalian cells, an alternative end-joining repair pathway repair of DSBs is thought to play a role in driving the formation of chromosomal translocations, although the specific enzymology is unclear [21], [22].
Here, we report experiments that define a specific function and mechanism of action for POLQ in a pathway for alternative end-joining of DNA double-strand breaks in mammalian cells.
To clarify the cellular role of POLQ in response to DNA damage, we measured the sensitivities of Polq-null and Polq-proficient bone marrow stromal cell (BMSC) lines to various DNA damaging agents. Cells lacking POLQ exhibit hypersensitivity to ionizing radiation (Figure 1) [12], [23], and to the double-strand break-inducing chemical bleomycin, as previously reported [12].
We found that Polq−/− cells were also hypersensitive to other agents which directly cause DNA breaks, including the topoisomerase II inhibitors etoposide and ICRF-193 [24] and camptothecin, a topoisomerase I inhibitor. In contrast, loss of Polq did not cause hypersensitivity to agents that largely form DNA replication-blocking adducts on one DNA strand including ultraviolet radiation and the methylating agent temozolomide. The Polq−/− cells were also not more sensitive than control cells to mitomycin C, cisplatin, and UVA-photoactivated psoralen plus UVA, all of which induce some interstrand DNA crosslinks (ICLs) (Figure 1).
These data indicate that POLQ is most important in a process conferring resistance to direct DNA strand-breaks, particularly double-strand breaks. Cells lacking Polq were not hypersensitive to the PARP inhibitor olaparib (Figure 1) while control RAD51D-defective cells were hypersensitive (Figure S1A). This suggests that POLQ does not function in the BRCA/homologous recombination pathway [25]. POLQ-proficient cells treated with both olaparib and camptothecin were significantly sensitized compared to camptothecin alone. However, addition of olaparib to Polq-null cells only modestly increased the sensitivity to camptothecin (Figure 1). Consequently, PARP and POLQ may operate within the same subpathway of DNA repair.
It is important to know whether the elevated level of micronuclei in Polq-defective cells extends to cell types other than peripheral erythrocytes. To answer this question, matched wild-type and Polq−/− BMSC lines were exposed to etoposide or x-rays, and the number of cells with micronuclei after 48 h were enumerated (Figure 2A and B). Polq-null cells exhibited a ∼3 fold increase in frequency of spontaneous micronuclei formation (Figure 2C). Upon exposure to DNA damaging agents, the percentage of cells with micronuclei increased about 1.5-fold more per unit of damage for Polq−/− cells in comparison to Polq+/+ cells (Figure 2A and B). This shows that the susceptibility to micronucleus formation in Polq−/− cells is not confined to cells of the hematopoietic lineage, but occurs also in cultured cells, including fibroblast-like BMSCs.
Cells lacking Polq were analyzed for their ability to proliferate in culture. Two independent BMSC lines devoid of Polq expression proliferated at a rate comparable to a pair of wild-type control cells, the Polq BMSCs showing only a 5% increase in population doubling times (Figure 2D and E). We extended this analysis to isogenic immortalized mouse embryonic fibroblast (MEF) cell lines (Figure 2F and G). Polq−/− cells divided at a rate comparable to Polq-proficient cells. These findings fit with our previous observations that hematopoietic cell counts in irradiated Polq-null mice recovered at rates comparable to wild-type mice [12]. We have observed no major alterations in growth or development in unchallenged Polq null or mutant mice, consistent with previous reports [13], [14], [26]. These observations indicate that despite some increased chromosomal instability, POLQ-defective cells originating from a variety of tissues can proliferate at near-normal rates.
We sought next to investigate which catalytic activities of POLQ are necessary to confer resistance to DNA damaging agents. Lentiviral-delivered expression vectors were constructed to express wild-type or mutant versions of POLQ in immortalized MEFs, in order to test for functional complementation (Figure 3A). A tandem (D2330A,Y2331A) mutation was introduced into the DNA polymerase domain (POL); mutation of the corresponding residues in other DNA polymerases completely inactivates polymerase activity [27]. In a separate construct, a mutation was introduced into the conserved ATP-binding site of the Walker A motif (K121M) in the helicase-like domain (HLD). An equivalent mutation eliminates DNA helicase activity in related enzymes, including HELQ [28]. A third construct (DM) was made harboring mutations in both domains. These vectors expressed full-length recombinant POLQ as tested in transfected 293T cells (Figure 3B and C).
The mutant cDNAs were tested for their ability to genetically complement the bleomycin sensitivity of Polq-null MEFs. Stable clones with each of the constructs were generated and analyzed for expression of POLQ (Figure 3D). Independent clones of knockout MEFs expressing wild-type recombinant POLQ (WT4 and WT6) were able to rescue bleomycin hypersensitivity (Figure 3E) as an antibody that recognizes endogenous POLQ does not yet exist. Neither the polymerase domain mutant (POL) nor the polymerase-helicase double mutant (DM) restored bleomycin sensitivity (Figure 3E, Figure S1B). Expression of a construct with a mutation only in the helicase-like domain (HLD) was, however, still able to restore resistance to bleomycin. These data indicate that POLQ polymerase activity is essential for conferring resistance to DNA damage, while the ATPase activity of the helicase-like domain is not necessary. Similarly reintroduction of polymerase activity of POLQ into Polq-deficient MEFs was able to rescue chromosomal instability (micronuclei and DNA DSBs, as measured by 53BP1 and γH2AX colocalization (Figure 3F and 3G, Figure S2).
Mice with an S1932P mutation in Polq (the “chaos1” allele) have an increased spontaneous frequency of micronuclei [13]. We generated a human POLQ cDNA mimicking the chaos1 mutation (S1977P), but attempted expression of POLQ with this mutation in 293T cells did not yield detectable protein (Figure S3). This suggests that the chaos1-encoded mutant protein is unstable, consistent with the finding that chaos1 mice have a phenotype essentially indistinguishable from Polq knockout mice [13].
Immunoglobulin class-switch recombination (CSR) uses DNA end joining to exchange one constant region of an antibody gene for another constant region. CSR can occur by both Ku-dependent classical non-homologous end joining and Ku-independent altEJ [20]. The overall frequencies of CSR are similar in Polq-defective mice [29] and cultured B cells [30]. To determine whether POLQ is involved in a mechanistically distinct subset of CSR joins, we isolated and analyzed DNA sequences at such joins. Naïve B cells were isolated from the spleens of wild-type and Polq-null mice and stimulated for IgM to IgG class switching, and then the fraction of IgG1-positive B cells was measured by flow cytometry. Parallel B-cell cultures were incubated with NU7026, a DNA-PKcs inhibitor that suppresses cNHEJ [31]. It has been shown that B cells incubated with NU7026 have an increased proportion of CSR junctions with >1 bp insertion at the junction [31]. This suggests that when a pathway of altEJ is used during CSR, it more frequently results in insertion of nucleotides.
We found that B cells from Polq-proficient and deficient mice had similar overall frequencies of CSR (Figure 4A), and inhibition of DNA-PKcs increased the frequency of CSR in both genotypes by 1.5 to 2 fold (Figure 4B). The Sμ-Sγ1 junction was then sequenced from 100 clones of each group of IgG1-positive B cells. These data revealed that in wild-type B cells, insertions of >1 bp at Sμ-Sγ1 junctions, that are thought to be altEJ-dependent, comprised about 9% of total events, and that this increased to ∼21% in cells incubated with NU7026 (Figure 4C, Table 1).
Strikingly, in cells lacking Polq, this class of insertions at CSR junctions was absent, even in the presence of NU7026 (Figure 4D, only one insertion of 2 bp observed). Insertion of >1 bp therefore requires POLQ. This class of Polq-dependent joining events included insertions of between 2 and 35 bp. For longer insertions (greater than ∼10 bp) homologous sequences were unambiguously detected up to 2–5 kbp away from the junction site (Table 1), as has been reported for long insertions at Sμ-Sγ1 junctions in ATM-defective B cells [31]. This suggests that most or all of such insertions are formed in a templated manner during altEJ by POLQ.
The most important factor in determining which double-strand break repair pathway is used is whether or not the 5′ termini of broken ends are resected [32]. Ends with little or no single stranded overhang are typically rejoined by Ku-dependent cNHEJ. In contrast, CtIP and MRN-dependent resection of 5′ termini generates ends with extended single stranded 3′ overhangs; resection is thought to block cNHEJ [33] and enable repair by altEJ [34], [35].
To analyze differing requirements for end joining, with or without end resection, we generated two linear DNA substrates with 3′ single stranded overhangs; one with a short overhang (6 nt), and one a long overhang (45 nt, a “pre-resected end”) (Figure 5A). Both can be aligned with the same 4 nt of terminal complementary sequence. These substrates were then introduced into wild-type mouse fibroblasts or fibroblasts harboring deficiencies in Ku70 or Polq. Repaired products were recovered from cells and quantified. Repair of the short overhang substrate was, as anticipated, over 10-fold less efficient in cells without Ku70 (Figure 5B) when compared to Ku70-complemented controls. The absence of Polq−/− had no consequence for repair of this substrate.
End joining with the 45 nt overhang substrate was assessed using qPCR primers located to ensure that at least 10 nt of overhang was included in joined products (Figure 5A). Recovery of these products was no longer dependent on Ku; instead, it was increased 2.8-fold in Ku70-deficient cells (Figure 5C). This is consistent with previous studies arguing Ku suppresses repair by altEJ. Strikingly, joining of the long overhang substrate in Polq−/− cells was reduced 10-fold, near background levels of signal observed using this assay. Complementation of the knockout cells with POLQ returned joining to wild-type levels (Figure 5C). These data demonstrate that POLQ participates in some form of alEJ, but cells lacking POLQ maintain proficiency for cNHEJ.
Our results demonstrate that POLQ is necessary to form the insertions found in CSR junctions in a process of altEJ. We next sought to determine the mechanism. Like other DNA polymerases, an active polymerase fragment of POLQ [36] can catalyze template-dependent DNA synthesis from an annealed primer (Figure 6A). As is common for family-A DNA polymerases, only a single nucleotide is added to the end of duplex DNA [5]. Unusually, however, POLQ can catalyze extension of single-stranded oligonucleotides [37]. It was unclear whether this reflects a robust terminal deoxynucleotide transferase activity of POLQ on single-stranded DNA, or some form of template-dependent synthesis. For example, POLQ can extend a single-stranded 16-mer oligonucleotide provided without a complementary template (products up to 35 nt long), while E. coli pol I Klenow fragment has no activity on this substrate (Figure 6B). The major 22 nt extension product produced by POLQ on the 16-mer used in Figure 6B may be accounted for by inter- or intra-oligonucleotide pairing (Figure S4C). Neither POLQ nor Klenow fragment could extend an oligonucleotide that was incapable of annealing to itself (Figure S4) [37].
To identify the mechanism of 3′ single-stranded DNA extension by POLQ, we used a different single-stranded oligonucleotide designed to be unable to form self-complementary base pairs longer than a single nucleotide [37], and sequenced the products of POLQ-mediated extension. Individual extension products of 1 to 30 nt were found (Table S1). Most of the sequenced extension products feature AAC or AAAC sequences that could arise from copying GTTT sequences in the template via inter- or intra-molecular priming and re-priming (Figure 6C) following minimal base pairing at the 3′-primer end. These data reveal that POLQ uniquely extends 3′ DNA tails through template-dependent DNA synthesis from a primer with minimal base pairing and that the polymerase lacks true TdT-like activity. POLQ indeed has unique biochemical properties compatible with these observations. Unlike other DNA polymerases, POLQ can efficiently extend a DNA chain with a nucleotide incorporated opposite an abasic site [5], or from a mismatched primer-terminus [38]. Further, there is evidence that primers slip on DNA templates with an increased frequency during POLQ-mediated synthesis, as shown by the high frequency of single base pair frameshift mutations generated by purified POLQ [39].
Double-strand breaks initiated by AID activity in the immunoglobulin heavy chain (IgH) locus of B cells are necessary to generate immunological diversity, but breaks are sometimes generated at other chromosomal sites, providing an opportunity for dangerous chromosome translocations [21], [22], [40], [41]. For instance the oncogenic Myc/IgH translocation that causes Burkitt lymphoma is AID-dependent and requires breaks at both loci, with breaks in the Myc gene rate-limiting [42]. An altEJ process is implicated in the formation of oncogenic translocations in lymphoid tissues, including the Myc/IgH translocation in murine B cells [21], [43], [44]. cNHEJ suppresses the formation of such chromosomal translocations [45]. To determine the role of POLQ in chromosomal translocations, Polq+/+ and Polq−/− naïve splenic B cells were stimulated in culture and assayed for the frequency of Myc/IgH translocations (Figure 7A). Notably, in the absence of Polq there was a 4-fold increase in translocation frequency (Figure 7B and C). This indicates that mammalian POLQ acts in a subset of altEJ events to suppress chromosomal translocations. Additionally, an increase in intramolecular IgH rearrangements was found in B cells lacking Polq (Figure 7B). Therefore, although POLQ is involved in an altEJ pathway, it prevents rather than promotes chromosomal instability, rearrangements and the formation of Myc/IgH translocations.
We show that in mammalian cells, POLQ has a specific role in defense against DNA damaging agents that cause direct DNA double-strand breaks, including ionizing radiation, bleomycin, and topoisomerase inhibitors. Our findings indicate that POLQ participates in a novel pathway of alternative-end joining of DSBs, a process that can occur throughout the cell cycle in mammalian cells [17]. The minimal additional sensitization to camptothecin by olaparib in Polq-defective cells suggests that one function of PARP is to participate in a Polq-dependent altEJ pathway. Our experiments indicate that POLQ is an important factor in DNA DSB repair in all cells, not just cells of the hematopoietic lineage. Indeed, Polq is broadly expressed in murine tissues (Figure S5).
Mutants of POLQ homologs in Arabidopsis (TEBICHI), C. elegans (polq-1), and Drosophila (Mus308) are hypersensitive to ICL-inducing agents [3], whereas Polq-defective mammalian cells are not appreciably hypersensitive to such agents (Figure 1). This difference may arise because of differences between organisms in the priority of DNA repair pathway engagement. In proliferating mammalian cells, ICLs are usually dealt with through the Fanconi anemia pathway, which produces enzymatically induced double-strand breaks that are channeled into homologous recombination repair [46]. In Drosophila and some other organisms, an altEJ-dependent pathway may be more important for resolving ICL-associated double-strand breaks. Although Drosophila Mus308 mutants are not hypersensitive to IR, pronounced IR sensitivity occurs in a double mutant when HR is also inactivated [47]. The phenotypic consequences of POLQ-dependent altEJ of double-strand breaks may thus depend on the relative dominance of HR which varies between organisms.
We show here that the DNA polymerase activity of POLQ is necessary to prevent cell death and chromosome breaks (micronuclei) caused by a double-strand break-inducing agent. Disruption of the ATPase activity in the helicase-like domain of POLQ did not, however, alter the correcting function of POLQ addition to knockout cells. A previous study with mouse cell lines suggested that disruption of the polymerase domain of the murine Polq gene is less severe than complete disruption of Polq [30], but the result is difficult to evaluate in the absence of quantitative measurements of expression of the partially deleted form. No activity has yet been shown for the helicase-like domain, other than DNA-dependent ATPase function [4]. It is likely that an additional role remains to be discovered that is dependent on the ATPase function of POLQ.
When double-strand breaks form in mammalian cells, a majority will be repaired through cNHEJ. However, a subset of these breaks will be handled by alternative end-joining pathways in situations where the DNA end is not compatible with processing by Ku-dependent cNHEJ, or if core components of the cNHEJ machinery are absent or unavailable (Figure 7D). In general, altEJ is defined as a means for repair of chromosome breaks that is exclusive of Ku-dependent, classically defined NHEJ [48], and dependent on factors (CtIP, MRN) that resect double-strand breaks to generate extended 3′ ssDNA tails [34], [35] (Figure 5A). Accordingly, we showed joining of a “pre-resected” extrachromosomal substrate (substrate with 45 nucleotide 3′-ssDNA tails) was stimulated in Ku-deficient cells, similar to results using chromosomal substrates [35]. Joining of this substrate was also dependent on Polq (Figure 5C). Our experiments thus define an altEJ subpathway in mammalian cells that involves POLQ (termed synthesis-dependent end joining, SD-EJ, in Figure 7D), Additional Polq-independent altEJ subpathways may also be operational (Figure 7D). To some extent, different end-joining pathways can be been distinguished from one another by the ligase employed in the pathway, with DNA ligase IV (LIG4) suggested as essential for cNHEJ, and DNA ligase III (LIG3) for altEJ in mammalian cells [21], [43], [49]. There are caveats, however. For example, some functional redundancy is apparent between LIG1 and LIG3 in altEJ [44], [50]–[52]. Ligase deficiencies may thus not be the best marker for distinguishing different end-joining pathways. For the altEJ subpathway under consideration here, dependence on POLQ is the best available definition.
The biochemical properties of POLQ provide a mechanistic explanation for its contribution to altEJ. POLQ has a unique ability to add nucleotides to the 3′ ends of single-stranded DNA [37], primed by minimal pairing with other available DNA molecules (Figure 6 and Figure S6). Synthesis by POLQ in this context is consistent with the unusually efficient ability of the polymerase to extend from mismatched DNA termini [5], [38], and its tendency towards primer-template slippage [39]. In further biochemical experiments it will be of interest to examine the action of POLQ and DNA ligases at double strand breaks with 3′-single-stranded overhangs that closely mimic the resected ends of a DNA double-strand break. In vivo studies with such substrates, including those that can form hairpins in the single-stranded region, would give insight as to the preferred structures for POLQ-catalyzed extension.
Unique to the POLQ-dependent altEJ process are frequent joins displaying templated DNA insertions. Some form of altEJ has been implicated in resolution of a subset of double-strand break intermediates in CSR, producing templated insertions [20]. Our data support a role for POLQ in generating the CSR products with these templated insertions. These events are consistent with the templated insertions that occur during Mus308-dependent repair of directed double-strand breaks in Drosophila [47], [53] and in C. elegans [54]. In the absence of POLQ, the lack of insertion-containing joins is observed, but the global CSR frequency is relatively unchanged (Figure 4). These insertions are best explained by repeated initiation of synthesis by POLQ (Figure 6C) on template sites, ultimately leading to a joined product.
In the absence of POLQ, we found a ∼4-fold increase in the formation of the oncogenic translocation Myc/IgH in mice. This increase is comparable to that seen in B cells that have lost Tdrd3, a regulator of R-loop formation during transcription [55] and miRNA-155 which regulates AID and suppresses oncogenic translocations [56]. In the absence of Polq there is also an apparent enhancement of rearrangement events in the IgH locus, consistent with the elevated level of chromosomal instability observed in cells lacking POLQ [57].
altEJ is typically associated with frequent annealing of the DNA ends at existing microhomologies (2–5 bp) and large deletions at repair junctions [19]. Since translocations commonly feature such microhomologies at their breakpoint junctions [58], [59] and occur more frequently in cNHEJ defective cells, altEJ is considered the primary mechanism by which translocations occur. Thus, a striking finding of the present work is that the formation of Myc/IgH translocations is suppressed when the POLQ-dependent altEJ subpathway is operational. It is possible that DNA DSBs persist for a longer time in the absence of POLQ, giving more opportunity for the formation of translocations. Alternatively, the POLQ-dependent pathway may be the most efficient at repairing a structurally distinct class of translocation-prone DNA breaks.
These studies clearly define a role for POLQ in the repair of DNA strand-breaking agents and provide a mechanism of template-dependent extension of DNA ends necessary to repair breaks in a subpathway of altEJ. This distinct altEJ pathway is necessary to prevent the formation of chromosomal translocations as shown by our in vivo experiments. It has been suggested that suppression of POLQ may be useful in increasing the efficacy of DNA damaging treatments in cancer [3], [23], [60]. This promising prospect should be tempered with the knowledge that loss of POLQ may also lead surviving cells to be prone to potentially oncogenic chromosome translocations.
Research mice were handled according to MD Anderson Cancer Center Institutional Animal Care and Use Committee policies and protocol 08-08-08732. Mice were euthanized by CO2 euthanasia followed by cervical dislocation.
Polq+/+ and Polq−/− bone marrow stromal cells and mouse embryonic fibroblasts were plated in triplicate (200,000 cells per 10 cm dish) with 15 mL of complete media (Dulbecco's Modified Eagle Medium+Glutamax, 10% FBS, 1% PennStrep). On the indicated days, cells were trypsinized and live cells were counted using trypan blue exclusion (Countess automated cell counter, Life Technologies). Experiments were repeated three times in order to generate standard deviations. Viability was consistently high for all cell lines examined (>95% trypan blue-excluding cells).
For X-irradiation 5×105 cells were plated on a 10 cm plastic culture dish, and exposed the following day at 2 Gy/min, 160 kV peak energy (Rad Source 2000 irradiator, Suwanee, GA). Cells were then trypsinized for replating. For UVC-irradiation (254 nm peak germicidal lamp) cells were irradiated in 500 µl PBSA (105 cells/ml) at 5 J m−2 min−1 and then plated. For psoralen-UVA treatment, 5×105 cells were plated on a 10 cm dish and incubated in medium with the indicated concentration of HMT-psoralen for 1 h, the dish was irradiated for with 0.9 kJ m−2 UVA (365 nm peak, 30 min, 0.5 mJ m−2 sec−1), the psoralen-containing medium was removed, and the dish UVA-irradiated in fresh medium for a further 30 min before replating. Chemicals were added at the indicated concentrations to dishes at the beginning of the experiment. Drugs were solubilized in ethanol (mitomycin c), DMSO (ICRF-193, etoposide, camptothecin, HMT-psoralen, temozolomide, olaparib), or 150 mM NaCl (cisplatin). All chemicals were from Sigma (St. Louis, MO) except ICRF-193 (Enzo LifeScience, Farmingdale, NY), olaparib (AZD2281, Selleck Chemicals, Houston, TX), and mitomycin c (Calbiochem, Darmstadt, Germany). Cells were plated in triplicate in 10 cm dishes and grown for 7–10 days before being fixed and stained with crystal violet. Colonies of 50 or more cells were quantified and experiments were repeated three times to generate standard deviations. A clonogenic assay was performed with Rad51D+/+ and Rad51D−/− Chinese hamster ovary (CHO) cell lines exposed to varying concentrations of olaparib.
BMSCs were plated at 1.5×104 cells per well in chambered slides and treated with the indicated amount of x-rays or etoposide the following day. 48 hr later, cells were fixed with 2% para-formaldehyde, stained with DAPI and coverslipped. Micronuclei were scored by immunofluorescence for 300 cells per group. Experiments were repeated three times to generate standard deviations.
293T cells (kindly provided by Dr. Christopher Bakkenist, University of Pittsburgh Medical School) were plated at 150,000 cells in six-well plates and transfected the following day with 2.5 µg of either pCDH (System Biosciences, Mountain View, CA) containing empty control, POLQ, POLQ-K121M, POLQ-D2330A,Y2331A, POLQ-S1977P, or POLQ-DM cDNA using Lipofectamine 2000 (Life Technologies) according to manufacturer's specifications. 48 hr after transfection, cells were harvested for RNA isolation (RNeasy, Valencia, CA) or immunoblotting.
For immunoblots, cells transfected in six-well dishes were resuspended in 200 µL of 2× SDS loading buffer (4% SDS, 0.2% bromophenol blue, 20% glycerol, 100 mM Tris HCl pH 6.8, 12% 2-mercaptoethanol) and heated at 95°C for 5 min. 20 µL of extract was separated on a 4–20% polyacrylamide gel, transferred to PVDF membrane, blocked, and blotted with anti-alpha-Tubulin (Abcam, Cambridge, UK) ab4074, 1∶10,000), anti-FLAG (Sigma F7425, 1∶5,000), anti-PCNA (Santa Cruz, Santa Cruz, CA, sc-56, 1∶1,000), anti-HA (RW, 1∶10,000), or anti-POLQ (MDACC POLQ20, 1∶250) antibodies and corresponding secondary antibodies (Sigma A0168, A0545; 1∶10,000) and visualized with ECL reagent (Pierce, Rockford, IL).
Polq-null (Polq−/−) mice [13] were obtained from Jackson Laboratories and maintained on a C57BL/6J background. Isogenic primary MEFs were generated from 13.5 day pregnant females and cultured in a 2% O2 atmosphere. MEFs were then transfected with 1 µg of pSV-Tag [61], [62] and grown in atmospheric oxygen for six population doublings to allow for immortalization. To generate lentivirus used for transduction, 293T cells were cotransfected with psPAX2 (6 µg), pMD2G (6 µg), and pCDH (12 µg) expression vector (See Text S1 for construction of expression vectors) using Lipofectamine 2000. One day prior to transduction Polq−/− MEFs were seeded into a 10 cm dish at 1.5×105 cells with 12 mL complete media. 48 hr post-transfection virus-containing media was harvested, filtered through a 0.45 µm syringe filter and used to replace the media on the plated MEFs. MEFs were incubated in the virus-containing media for 24 hr before being split into T-75 flasks and allowed to grow to 80% confluence before undergoing three weeks of puromycin selection (2.5 µg/mL). Following selection, pure clones were isolated and cultured with complete media containing puromycin (1 µg/mL).
RNA isolated from the complemented MEF lines were analyzed for quality and purity using RNA 6000 Nano kit (Agilent Technologies, Santa Clara, CA). 1 µg of total RNA was used to generate cDNA using the High Capacity cDNA RT kit (Life Technologies). qPCR analysis was performed in triplicate using the ABI Prism 7900 HT thermocycler and the following Taqman Probe set or primer set with iTAQ SYBR Green Supermix with ROX (Bio-Rad, Hercules, CA): MmPolQ_FWD 5′-GGCTCTGAAGAACTCTTTGCCTTT-3′, MmPolQ_REV 5′-GCTGCTTCCTCTTCTTCATCCA-3′, probe 5′-TCCGGGCACTTTTG-3′; HsPOLD1_FWD 5′-CGACCTTCCGTACCTCATCTCT-3′ HsPOLD1_R 5′-ACACGGCCCAGGAAAGG-3′, probe 5′-CCCTCAAGGTACAAACAT-3′; Qexon FWD 5′-TGCCTTTCAAAAGTGCCCGGAAGGC3′, Qexon REV 5′-TGCCAGTCACCCANATAGTTCNCAT-3′. Data were analyzed using the ΔΔCt method. For absolute quantification, titration of pCR-XL-TOPO/MmPolQ and pET/MmPold1 plasmids were used to generate standard curves for expression. Transcript abundance was determined by extrapolation from linear regression analysis of best fit lines from titration experiments. GAPDH was used as an internal control in all experiments.
Complemented MEF lines were plated in triplicate into white 96 well plates at 1250 cells per well and grown overnight using complete media containing puromycin (1 µg/mL). The following day, cells are cultured with complete media containing the indicated amounts of bleomycin (dissolved in 150 mM NaCl) for 24 hr before the media was replaced. Cells were allowed to recover for 72 hr before cellular viability was measured using the ATPlite 1Step kit (Perkin Elmer, Waltham, MA) using a Biotek plate reader. Experiments were repeated three times.
Complemented MEF lines were plated at a density of 1.5×104 cells per well in 4-well chamber slides and the following day were irradiated with either 0 or 6 Gy of x-rays. Media was changed and cells were allowed to recover for 48 hr after damage before fixation with 2% para-formaldehyde and permeabilized with Triton X-100. Samples were blocked with donkey serum for 30 minutes before being incubated overnight with primary antibodies against 53BP1 (Bethyl, Montgomery, TX, A300-272A, 1∶500) and γH2AX (EMD Millipore 05-636, 1∶400). Cells were later incubated with AlexaFluor-488 goat-anti-mouse or AlexaFluor-594 goat-anti-rabbit secondary (Life Technologies, 1∶1000) and then stained with DAPI before being coverslipped. Cells were scored for DSBs by enumerating the percentage of cells with >2 53BP1 foci and >2 γH2AX foci [61], [63]. The majority of cells that contained >2 foci for each of the DSB markers, exhibited colocalization of the foci. Cells with pan-staining of γH2AX were not included in the analysis as they are proposed to represent pre-apoptotic cells [64]. Many of the cells with 53BP1 foci, exhibited enlarged foci that are associated with nuclear OPT (Oct-1, PTF, transcription) domains that sequester damaged DNA in G1 [65], [66]. Thus, most of the MEFs that were foci positive contained DSBs [65]. DAPI-stained micronuclei were also scored. For each experiment 250 cells were scored for three independent experiments for a total of 750 cells.
POLQ was purified as described [36]. Klenow Fragment (3′→5′ exo-) was purchased from NEB. POLQ was diluted in buffer containing 30 mM Tris-HCl pH 8.0, 50 mM NaCl, 2 mM DTT, 10% glycerol, 0.01% Triton X-100, and 0.1% BSA. Klenow Fragment (3′→5′ exo-) was diluted in buffer containing 25 mM Tris-HCl pH 7.4, 1 mM DTT, and 0.1 mM EDTA. POLQ reaction mixtures (10 µl) contained 20 mM Tris-HCl pH 8.8, 4% glycerol, 2 mM dithiothreitol (DTT), 80 µg/ml bovine serum albumin (BSA), 8 mM MgCl2, 0.1 mM EDTA, 100 µM of each dNTP, 30 nM of the primer-template or primer (see Text S1). Klenow Fragment (3′→5′ exo-) reaction mixtures (10 µl) contained 10 mM Tris-HCl pH 7.9, 50 mM NaCl, 1 mM DTT, 10 mM MgCl2, 100 µM of each dNTP, and 30 nM of the primer-template or primer. After incubation at 37°C for 10 min for a 16-1+PA42 substrate or 20 min for 16-1, C20, C19THF substrates, reactions were terminated by adding 10 µl of formamide stop buffer (98% formamide, 10 mM EDTA pH 8.0, 0.025% xylene cyanol FF, 0.025% bromophenol blue) and boiling at 95°C for 3 min. Products were electrophoresed on a denaturing 20% polyacrylamide-7 M urea gel, exposed to BioMax MS film, and analyzed with a STORM 860 Phosphor Imager (Molecular Dynamics).
A dermal fibroblast line from Ku70 and p53 deficient mice (the gift of Dr. P. Hasty, University of Texas Health Sciences Center) was transduced with empty vector (pBABE-puro) retrovirus or a retrovirus expressing mouse Ku70. Substrates were generated by ligating short linkers to the head and tail of a 556 bp linear double-stranded DNA fragment. Linkers possessed 16–17 bp of double-stranded DNA and either 6 or 45 nt 3′ single-stranded overhangs. The linkers with 6 nt overhangs were made by annealing 5′- AGTCTGAGATGGGTGTGAGATCTGC-3′ to 5′-CACTCTCTCACACCCATCTTA-3′ (“head” linker), and 5′-TGACTATACAGCTAAGCGATGATGCAG-3′ to 5′-CATCGCTTAGCTGTATA-3′ (“tail” linker). The linkers with 45 nucleotide 3′ overhangs were generated by annealing 5′-AGTCTGAGATGGGTGTGAGAGTGAAGATCCTCACCTTCGGAGTACTCCTTCTTTTGAGATCTGC-3′ to 5-CTCACACCCATCTCA-3′ (“head” linker) and 5′-TGACTATACAGCTAAGCGATGCTCTCACCGAGCGTATCTGCTGTGTTGTGGATGAATTAGATGCAG-3′ to 5′-CATCGCTTAGCTGTATA-3′ (“tail” linker). Excess linker was removed by QiaQuick purification and substrate purity validated by polyacrylamide gel electrophoresis. 75 ng of substrate was mixed with 1.1 µg of supercoiled pMAX-GFP (Lonza) plasmid carrier and introduced into 2×105 cells in a 10 µl volume by electroporation with one 30 ms 1350 V pulse (Neon, Invitrogen). Cells were harvested after incubation for 1 hour at 37°C, washed, resuspended in Hank's buffered saline solution supplemented with 5 mM MgCl2, and extracellular DNA digested by incubation with 6.25 U Benzonase (Novagen) for 15 min at 37°C. Cells were pelleted and DNA purified with the Qiamp kit (Qiagen). Joining efficiency was determined by quantification of head-to-tail junctions by qPCR using primers that either anneal within double-stranded flanks (5′- CTTACGTTTGATTTCCCTGACTATACAG-3′ and 5′- GCAGGGTAGCCAGTCTGAGATG-3′; 6 nt overhang, Figure 5B) or, for the 45 nt overhang substrate only, which anneal to overhang sequence (5′- TAAGCGATGCTCTCACCGAG and 5′- GATGGGTGTGAGAGTGAAGATC; 45 nt overhang, Figure 5C). Results from electroporated samples were further corrected for differences in transfection and sample processing efficiency using a qPCR specific for substrate (5′- GGCACTCTCCAAGGCAAAGA and 5′- ACATGTCTAGCCTATTCCCGGCTT).
B cells were isolated from mouse spleens (n = 6 per genotype) and stimulated for class-switching in culture for 72 hr. Where indicated, cultures were incubated with DNA-PKcs inhibitor 20 µM NU7026 (Tocris, Bristol, UK) dissolved in DMSO, or mock-treated. The stimulation procedure and flow-sorting for CSR analysis was as described [31], [67]. Prior to this analysis, cells were counted; numbers and viability were similar for all groups. Sμ-Sγ1 CSR junctions were amplified by PCR using the following conditions for 25 cycles at 95°C (30 s), 55°C (30 s), 68°C (180 s) using the primers (FWD 5′-AATGGATACCTCAGTGGTTTTTAATGGTGGGTTTA-3′; REV 5′ CAATTAGCTCCTGCTCTTCTGTGG-3′) and Pfu Turbo (Stratagene, La Jolla, CA). To the PCR reaction, 5 U of Taq polymerase (Promega, Madison, WI) was added and incubated at 72°C for 10 min. The resulting product was TOPO TA cloned and transformed into Top10 E. coli cells (Life Technologies, Carlsbad, CA) and plasmids were purified and sent for sequencing using M13 FWD and REV primers in addition to the amplification primers for sequencing. 100 clones for each group were analyzed for mutations, deletions, insertions, and sequence overlaps at the junction and both 30 nt upstream and downstream of the junction. p-values were determined by using two-tailed Fisher's exact test.
Naïve B cells from three pairs of Polq+/+ and Polq−/− mice were harvested as above, cultured for 72 hr, and DNA was isolated. 32 separate PCR reactions, each containing the genome from 1×105 cells, was performed with primers to amplify Myc/IgH translocations and amplified translocations were verified by Southern blotting using internal probes to the Myc and IgH loci as described [68], [69]. Three independent experiments were performed and the p-value was determined using two-tailed Fisher's exact test. %IgG1 was also measured as an internal control to ensure the B cells from each genotype were switching at a comparable level.
|
10.1371/journal.pgen.1005638 | Tissue-Specific Effects of Reduced β-catenin Expression on Adenomatous Polyposis Coli Mutation-Instigated Tumorigenesis in Mouse Colon and Ovarian Epithelium | Adenomatous polyposis coli (APC) inactivating mutations are present in most human colorectal cancers and some other cancers. The APC protein regulates the β-catenin protein pool that functions as a co-activator of T cell factor (TCF)-regulated transcription in Wnt pathway signaling. We studied effects of reduced dosage of the Ctnnb1 gene encoding β-catenin in Apc-mutation-induced colon and ovarian mouse tumorigenesis and cell culture models. Concurrent somatic inactivation of one Ctnnb1 allele, dramatically inhibited Apc mutation-induced colon polyposis and greatly extended Apc-mutant mouse survival. Ctnnb1 hemizygous dose markedly inhibited increases in β-catenin levels in the cytoplasm and nucleus following Apc inactivation in colon epithelium, with attenuated expression of key β-catenin/TCF-regulated target genes, including those encoding the EphB2/B3 receptors, the stem cell marker Lgr5, and Myc, leading to maintenance of crypt compartmentalization and restriction of stem and proliferating cells to the crypt base. A critical threshold for β-catenin levels in TCF-regulated transcription was uncovered for Apc mutation-induced effects in colon epithelium, along with evidence of a feed-forward role for β-catenin in Ctnnb1 gene expression and CTNNB1 transcription. The active β-catenin protein pool was highly sensitive to CTNNB1 transcript levels in colon cancer cells. In mouse ovarian endometrioid adenocarcinomas (OEAs) arising from Apc- and Pten-inactivation, while Ctnnb1 hemizygous dose affected β-catenin levels and some β-catenin/TCF target genes, Myc induction was retained and OEAs arose in a fashion akin to that seen with intact Ctnnb1 gene dose. Our findings indicate Ctnnb1 gene dose exerts tissue-specific differences in Apc mutation-instigated tumorigenesis. Differential expression of selected β-catenin/TCF-regulated genes, such as Myc, likely underlies context-dependent effects of Ctnnb1 gene dosage in tumorigenesis.
| Enhanced Wnt signaling contributes to colorectal and other cancers. β-catenin functions in Wnt signaling as a T cell factor (TCF) transcriptional co-activator. Previous studies showed specific β-catenin dosage favors Wnt signaling-dependent tumorigenesis for some tumor types. However, earlier studies emphasized the role of constitutional Ctnnb1 and Apc gene variations, rather than somatic gene targeting, and the work focused on small intestine tumors and no effects on colon tumor phenotypes were described. Furthermore, definitive insights were lacking into how reduced Ctnnb1 gene dosage affected Apc mutation-dependent tumorigenesis. Here, we show somatic inactivation of one Ctnnb1 allele dramatically inhibits mouse colon adenomatous polyposis induced by somatic bi-allelic Apc inactivation. In contrast, Ctnnb1 hemizygous inactivation does not affect mouse ovarian endometrioid adenocarcinoma development arising from Apc- and Pten-inactivation. Ctnnb1 hemizygous gene dose dramatically reduces the active pool of β-catenin, leading to the significant inhibition of β-catenin/TCF-regulated target gene expression, including those encoding key stem cell regulatory and crypt compartmentalization factors in colon epithelium. Tissue-specific differences for expression of selected β-catenin/TCF-regulated genes, such as Myc, may contribute to the context-dependent effects of Ctnnb1 gene dosage in Apc mutation-driven colon and ovarian tumors.
| Colorectal cancers (CRCs) harbor accumulated mutations in tumor suppressor genes and oncogenes along with epigenetic alterations. Many CRCs arise from precursor lesions, such as adenomatous polyps or serrated epithelial lesions with dysplasia. Inactivating mutations in the APC (adenomatous polyposis coli) and TP53 tumor suppressor genes are found in roughly 80% and 60% of CRCs, respectively [1]. Oncogenic mutations activating the functions of the KRAS and PI3KCA (phosphoinositide-3-kinase, catalytic, alpha polypeptide) proteins are found in about 40% and 20% of CRCs, respectively [1]. Constitutional mutations inactivating one APC allele underlie the familial adenomatous polyposis (FAP) syndrome, where affected individuals often develop hundreds to thousands of colon adenomas during their second or third decades of life. The wild type APC allele is somatically inactivated in adenomas arising in those with FAP [1, 2]. Mice carrying certain heterozygous, constitutional mutations inactivating Apc, such as the ApcMin mutation, may develop 50–100 small intestinal tumors and occasional colon tumors by 140 days of age and nearly all of the tumors are adenomas. Similar to the situation in FAP tumors, intestinal tumors in ApcMin mice show somatic inactivation of the wild type Apc allele [3].
The best understood function of the roughly 300 kD APC protein is regulation of the pool of β-catenin protein that functions in the canonical (β-catenin-dependent) Wnt signaling pathway [4–6]. In the absence of an activating Wnt ligand signal, the β-catenin destruction complex—comprised by the APC, AXIN, casein kinase I, and glycogen synthase kinase-3β factors and other proteins—promotes phosphorylation of conserved serine/threonine residues in the β-catenin amino (N)-terminal region. The N-terminally phosphorylated β-catenin can then be β ubiquitinated and degraded by the proteasome. Activating Wnt ligands inhibit degradation of the “free” or Wnt signaling pool of β-catenin via binding at the cell surface to the frizzled and LRP5/6 (low density lipoprotein-related proteins 5 and 6) cognate receptor complex, resulting in inhibition of β-catenin phosphorylation and/or ubiquitination by the destruction complex [4, 6]. In colon adenomas and CRCs where both APC alíeles are defective, destruction of the free pool of β-catenin is impaired and active β-catenin accumulates in the cytoplasm and nucleus, where it can complex with DNA binding proteins of the TCF (T-cell factor family)/Lef (lymphoid enhancer family) family. β-catenin functions as a transcriptional co-activator for TCFs [7]. Normally, β-catenin/TCF transcriptional activation is restricted to the crypt base, especially in the so-called crypt base columnar stem cells characterized by expression of the Wnt-regulated Lgr5 presumptive stem cell marker protein [8]. Constitutive activation of β-catenin/TCF transcription in Wnt pathway-defective adenomas and CRCs may promote a stem or progenitor cell phenotype in epithelial cells independent of cell position in the crypt [9, 10]. Activation of β-catenin/TCF-dependent transcription also alters crypt compartmentalization and coordinated migration of cells, apparently through increased expression of the EphB2 and EphB3 receptors and via inhibition of the expression of their ligands ephrin B1 and B2 [11, 12]. The MYC gene has been highlighted as a potentially key target gene regulated by β-catenin/TCF in CRCs. Genes encoding negative-feedback inhibitor proteins functioning in the Wnt/β-catenin/TCF pathway, such as AXIN2, DKK1, and NKD1, are also activated by β-catenin/TCF (see http://www.stanford.edu/~rnusse/pathways/targets.html for a list of candidates). In APC-mutant neoplastic cells, the ability of these induced regulator proteins to inhibit the Wnt signaling pathway is abrogated because the factors function upstream of or at the level of the APC protein in the pathway [13].
Besides these findings, other evidence indicates that APC inactivation may promote cancer development through β-catenin dysregulation. For instance, while most CRCs harbor APC mutations, a subset of CRCs and other cancers lacking APC mutations have CTNNB1 gene mutations resulting in production of oncogenic β-catenin proteins that are resistant to regulation by the destruction complex and that activate β-catenin/TCF transcription [6, 13]. Also, some prior studies have used genetic approaches to study effects of Ctnnb1 gene dosage on liver, small intestine, and mammary gland tumor phenotypes in mouse models as well as effects of Ctnnb1 hemizygous inactivation state (Ctnnb1+/-) in Apc-mutation induced mouse embryonic development phenotypes [14, 15]. The prior studies indicated the Ctnnb1+/- constitutional state can inhibit intestinal and liver tumorigenesis in mice carrying mutations in the Apc gene (Apc1638N, ApcMin, or Apcfl) [14, 15]. In contrast, mammary gland tumorigenesis was enhanced in Apc1638N Ctnnb1+/- mice, perhaps because Ctnnb1 functions as a tumor suppressor gene in the mammary gland tumors via β-catenin’s role in E-cadherin-dependent tumor suppression [14]. Nonetheless, while the prior studies yielded evidence that β-catenin signaling dosage impacts Apc mutation-induced tumorigenesis in some tissues, the prior work did not assess the role of Ctnnb1 dosage in Apc mutation-induced colon tumorigenesis, the chief site of APC mutation-dependent tumorigenesis in humans. Moreover, the work used mice constitutionally deficient in β-catenin, not just in Apc-mutant epithelial cells, and the findings did not highlight specific factors and mechanisms that might account for effects of Ctnnb1 dosage in Apc mutation-instigated tumorigenesis in different contexts. We report here on studies of the effects of Ctnnb1 gene dosage on β-catenin protein expression and β-catenin/TCF transcription in Apc mutation-induced colon and ovarian mouse tumors and cell culture models. We provide evidence that Apc mutation-induced tumorigenesis in the colon is inhibited by Ctnnb1 hemizygous gene status through marked effects on the free pool of β-catenin in the cytoplasm and nucleus and its ability to activate key β-catenin/TCF-regulated target genes, including those encoding key stem factors, such as Lgr5, and regulators of crypt compartmentalization, such as the EphB2/B3 receptors. We also uncovered a novel feed-forward mechanism where β-catenin protein stabilization and β-catenin/TCF transcription appear critical in regulating Ctnnb1/CTNNB1 transcription in the setting of Apc inactivation in mouse colon and human colon cancer cells. Moreover, we found that differences in the ability to activate Myc expression may underlie colon versus ovary tissue-specific differences in Apc mutation-instigated tumorigenesis in the setting of Ctnnb1 hemizygous gene dosage.
We previously described CDX2P-G22Cre transgenic mice, in which human CDX2 regulatory sequences and an out-of-frame Cre transgene allele, carrying a 22-basepair guanine nucleotide repeat tract affecting the Cre open reading frame, manifest mosaic Cre recombinase expression in caudal embryonic tissues and in epithelium of the distal ileum, cecum, colon, and rectum during adult life [16]. We also previously described CDX2P-CreERT2 transgenic mice that express a tamoxifen (TAM)-regulated Cre protein (CreERT2) under control of human CDX2 regulatory sequences, allowing for TAM-inducible targeting of loxP-containing alleles in adult terminal ileum, cecum, colon, and rectal epithelium [17]. Using the CDX2P-G22Cre or CDX2P-CreERT2 transgenic mice, we have described the phenotypic consequences in colon epithelium of somatic, bi-allelic, inactivating mutations in Apc [16, 17]. Consistent with our prior studies, we found CDX2P-G22Cre Apcfl/fl mice lived only for 8–20 days after birth (median survival = 13 d; Fig 1A). After three daily doses of TAM to inactivate both Apc alleles in distal intestinal epithelial tissues, CDX2P-CreERT2 Apcfl/fl adult mice lived on average for 22 days (Fig 1B). In marked contrast, concurrent somatic inactivation of one Ctnnb1 allele along with both Apc alleles, using either the CDX2P-G22Cre or CDX2P-CreERT2 transgene for somatic gene targeting, led to a dramatically increased life span relative to that seen in mice with Apc bi-allelic targeting, with median survival of 168 d of age in CDX2P-G22Cre Apcfl/fl Ctnnb1fl/+ mice and for 134 d after TAM treatment in the CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice (Fig 1A and 1B).
Consistent with our prior reports [16, 17], the proximal colon and cecum of both CDX2P-G22Cre Apcfl/fl mice (when moribund at 8–20 d of age) and CDX2P-CreERT2 Apcfl/fl mice (only 20 days after TAM induction) were dramatically thickened and many polypoid lesions were seen (S1 Fig). Histological analysis of proximal colon epithelial tissues from these mice showed significant hyperplastic and dysplastic (adenomatous) changes along with frequent crypt fission and branching (Fig 1C). The dramatic polyposis in cecum and colon seen following bi-allelic Apc inactivation was significantly inhibited at both early and later time points by concurrent inactivation of one Ctnnb1 allele, with no grossly discernable epithelial phenotype seen in the proximal colon and only two to four polyps in the cecum per mouse as the CDX2P-G22Cre Apcfl/fl Ctnnb1fl/+ and CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice were aged (S1 Fig). The cecal polyps arising in CDX2P-G22Cre Apcfl/fl Ctnnb1fl/+ and CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice may contribute to their premature mortality relative to control mice, as no other grossly detectable intestinal lesions or pathology were noted in the mice. The Cre-mediated somatic inactivation of both Apc alleles and one Ctnnb1 allele in proximal colon epithelium was confirmed by genotyping. The rare cecal adenomas arising in CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice were found to have significant fractions of cells that escaped Cre-mediated Ctnnb1 targeting, even though Cre-mediated somatic inactivation of both Apc alleles occurred to the same extent in the rare adenomas and proximal colon mucosa (S1 Fig). Compared to the situation in CDX2P-CreERT2 Apcfl/fl mice, microscopic examination of proximal colon tissues of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice revealed modest hyperplastic changes and minimal crypt branching (Fig 1C). Our efforts to inactivate both Ctnnb1 alleles in colon epithelium via either CDX2P-G22Cre- or CDXP-CreERT2-mediated targeting with or without Apc inactivation indicated that colon epithelial cells completely lacking β-catenin expression and function could not be generated. This likely reflects a required role for β-catenin function in colon epithelium, perhaps not limited to Wnt signaling, but also in cadherin-mediated adhesion, centrosome assembly or other functions.
Immunohistological analysis of colon sections from CDX2P-CreERT2 Apcfl/fl mice showed strong cytoplasmic and nuclear β-catenin expression in many epithelial cells, compared to the nearly uniform membrane β-catenin staining in colon epithelium of wild-type mice (Fig 2A). In CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice, we observed infrequent cells with elevated cytoplasmic and/or nuclear β-catenin expression (Fig 2A). Paneth cells, a specialized secretory cell linage that expresses lysozyme and other markers, are found at the crypt base in normal mouse small intestinal epithelium, but are absent in normal mouse colon epithelium. Paneth cells have been proposed to have a key role in generation and/or maintenance of the intestinal crypt stem cell niche [18]. Bi-allelic Apc inactivation has been associated with the generation of many ectopic lysozyme-expressing Paneth-like cells throughout the crypts of small intestine and colon [12, 17, 19, 20]. We confirmed this finding in CDX2P-CreERT2 Apcfl/fl mice (Fig 2A). Whereas no Paneth-like cells were seen in normal mouse colon, modest numbers of lysozyme-expressing cells were seen in the colons of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice (Fig 2A). We also used a transgenic mouse line carrying a Cre-activated enhanced yellow fluorescence protein (EYFP) reporter gene at the ubiquitously expressed Rosa26 locus to monitor colon epithelial cells and glands where Cre-mediated targeting had occurred. Ectopic lysozyme-expressing cells were found in nearly all of the EYFP-positive crypts in CDX2P-G22Cre Apcfl/fl Ctnnb1fl/+ and CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice (Fig 2B and 2C). The rare occurrence of Paneth-like cells in crypts without EYFP expression likely reflects the possibility that Cre may more efficiently target the loxP sites at the Apc locus than at the Rosa26 locus.
Prior studies from other groups and ours have shown that Apc bi-allelic inactivation increases both cell proliferation and apoptosis in intestine and colon epithelium [12, 17, 19, 21]. Following TAM-induced Apc bi-allelic inactivation in proximal colon epithelium, we confirmed significantly elongated crypts and increased cell proliferation and apoptosis relative to control epithelial tissues (Fig 2D and 2E). In contrast, only modestly increased crypt height, cell proliferation and apoptosis relative to control epithelium were seen in epithelium of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice following combined Apc and Ctnnb1 gene inactivation (Fig 2D and 2E). Furthermore, although bi-allelic Apc inactivation induced extensive cell proliferation in the upper half of targeted colon crypts, cell proliferation following gene targeting in CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice was largely restricted to the bottom half of each crypt, with only a slight increase in cell proliferation compared to control colon epithelium (Fig 2D and 2E). The cell proliferation and apoptosis results were well correlated with the immunohistochemical studies of β-catenin levels and localization (Fig 2A), suggesting differences in the strength of β-catenin-dependent Wnt signaling in cells with bi-allelic Apc defects underlie the observed effects on colon epithelial morphology, cell fate and differentiation, and cell proliferation and apoptosis.
The orientation of the mitotic spindle axis may impact on cell fate decisions in intestinal epithelium. At cytokinesis, the orientation of the spindle axis in a planar fashion (i.e., parallel to the crypt axis) is thought to generate two daughter cells with equivalent luminal (apical) and basement (extracellular matrix) surfaces. If the spindle axis is not oriented parallel to the crypt axis, cytokinesis generates daughter cells with differences in luminal and basement membrane surfaces and the potential for resultant differences in the fates adopted by the two daughter cells. We previously reported significant increases in the percentage of epithelial cells where the mitotic spindle axis was oriented orthogonal to the planar axis in Apc-mutant mouse colon crypts relative to wild type crypts [17]. Consistent with our prior results, in colon epithelium of CDX2P-CreERT2 Apcfl/fl mice treated with TAM to inactivate both Apc alleles, roughly 50% of the cells in mitosis had mitotic spindle axes ≥30° degrees out of the planar axis, with nearly 20% showing spindle axes between 60° and 90° out of planar alignment. In contrast, in epithelium of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice and control mice, >75% of mitotic colon epithelial cells had their mitotic spindles aligned within 30° of the planar (crypt) axis (S2 Fig). The findings indicate β-catenin levels have a key role in the altered mitotic spindle axis phenotype of Apc-mutant colon epithelium.
As described above, bi-allelic Apc inactivation acutely induces hyperproliferation and dysplastic alterations in mouse proximal colon epithelium, with the altered epithelium arising in part from expansion of the crypt progenitor compartment at the expense of the differentiated compartment, along with frequent crypt fission/branching [12, 17, 21]. The EphB/ephrinB signaling axis has been implicated in control of intestinal epithelial cell compartmentalization along the crypt axis and in cell migration [11, 22]. The EphB2 and EphB3 receptors are two key effectors of compartmentalization and cell migration in the crypt, and EphB2 and EphB3 are each encoded by a gene activated in intestinal tissues by β-catenin/Tcf transcription. The EphB2/B3 receptor ligands, ephrinB1 and ephrinB2, show highest expression levels in differentiated cells at the crypt surface, and expression of ephrins B1 and B2 is negatively regulated by β-catenin/Tcf activity [11, 23]. Of note, EphB-ephrinB interactions generate repulsive forces that separate and compartmentalize the EphB- and ephrinB-expressing cells to maintain crypt architecture [11, 23]. In normal mouse colon epithelium, the EphB2 and EphB3 receptors were expressed only in progenitor cells at the crypt base (Fig 3A and 3B). We found bi-allelic Apc inactivation in colon epithelium not only increased EphB2 and EphB3 expression, but also perturbed the gradient of EphB2 and B3 receptor expression along the crypt axis, with EphB2/B3 expression seen even at the crypt surface in Apc-mutant crypts (Fig 3A and 3B). In the case of ephrin ligand expression, our studies demonstrated strong expression of ephrinB1 and B2 in normal colon surface epithelial cells and normal colon crypt cells other than the crypt base. The normal pattern of ephrinB1/B2 expression remained largely unaffected in Apc-mutant crypts with one Ctnnb1 allele inactivated (Fig 3C). In contrast, in Apc-mutant crypts where Ctnnb1 dosage was intact, ephrinB1/B2 expression was markedly down-regulated in colon surface epithelial cells and throughout the crypt (Fig 3C). Our findings are consistent with those in a prior study that showed increased expression of EphB2/B3 and loss of ephrinB1/B2 expression in colon adenomas of Apcmin/+ mice [23]. Although expression of EphB2/B3 was moderately elevated in some colon epithelial cells of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice compared to control mice, elevated EphB2/B3 expression remained restricted to the crypt base region, rather than spreading throughout the crypt as was seen in Apc-mutant crypts with intact Ctnnb1 gene dosage (Fig 3A and 3B). This observation suggests the reduced β-catenin levels in CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice leads to a failure to induce enough β-catenin/TCF-regulated EphB2/B3 expression to overcome the repulsive effects of the retained expression of ephrinB1/B2 ligands in Apc-mutant crypts with reduced Ctnnb1 dosage.
A similar expression pattern to that seen for EphB2 and EphB3 was also found for Sox9, a transcription factor encoded by a β-catenin/Tcf target gene. Sox9 expression is restricted to stem/progenitor cells at the normal colon crypt base (S3 Fig). Sox9 expression was only modestly increased and expanded following gene targeting in crypts of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice relative to the marked changes in Sox9 levels and the number of Sox9-expressing cells in crypts from CDX2P-CreERT2 Apcfl/fl mice (S3 Fig). In spite of the reduced increase in β-catenin levels in colon crypts of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice relative to CDX2P-CreERT2 Apcfl/fl mice, the resultant signaling was still sufficient to generate some ectopic Paneth-like cells (Fig 3A and 3B). In addition, the modest increase in β-catenin levels in targeted crypts of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice was sufficient to induce expression in targeted crypts of a β-galactosidase reporter gene integrated into the β-catenin/TCF-regulated Axin2 locus. However, β-galactosidase expression was reduced in crypts and few if any colon surface epithelial cells expressed β-galactosidase in Apc-mutant epithelium with hemizgyous Ctnnb1 dosage (S3 Fig). In contrast, uniformly strong β-galactosidase expression was seen throughout Apc-mutant colon crypts and surface epithelial cells with intact Ctnnb1 dosage (S3 Fig). Taken together, the findings indicate distinct β-catenin/Tcf target genes in colon epithelium display differing transcriptional responses to β-catenin levels, with Axin2 perhaps representing a target gene capable of being activated by modest to moderate levels of β-catenin in colon epithelium. The Sox9, EphB2, and EphB3 genes appear dependent on higher levels of β-catenin for transcriptional activation in colon epithelium.
To address mechanisms underlying suppression of crypt fission and branching in Apc-deficient colon epithelium when one Ctnnb1 allele was inactivated, we compared expression of presumptive stem cell markers in Apc-deficient colon crypts where both Ctnnb1 alleles were intact or where only one Ctnnb1 allele was active. Consistent with our prior work [17], 20 days after TAM-induced bi-allelic Apc inactivation, we detected strong induction of enhanced green fluorescent protein (EGFP) expressed from the Lgr5 locus (Lgr5-EGFP) (Fig 4A) in Apc-deficient colon epithelium generated by CDX2P-CreERT2 targeting. Lgr5 is a β-catenin/TCF-regulated gene and a marker of presumptive crypt base columnar stem cells in normal colon, and the Lgr5 allele that we used has a EGFP open reading frame integrated in the locus to allow for monitoring of endogenous Lgr5 expression [8]. In Apc-mutant epithelium, we also confirmed strong induction of the Msi1 RNA-binding protein (Fig 4B), another presumptive intestinal stem cell marker [24, 25]. In contrast to a prior study where it was reported that Lgr5-expressing cells were only expanded at the lower part of the crypts in colon epithelium following mutant β-catenin induction [21], we detected EYFP-positive and Msi-positive cells essentially throughout the Apc-mutant dysplastic colon crypts when both Ctnnb1 alleles were active, though expression of EYFP was more prominent near the crypt base region, including in the de novo crypts. While the net number of EYFP- and Msi1-expressing cells per crypt were slightly increased (e.g. from 3–4 to 5–8 Lgr5-positive cells per crypt) in colon epithelium of TAM-treated CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice compared to the control mice, paralleling the subtle increase in crypt fission/budding seen, the expanded population of EYFP-positive cells remained restricted to the crypt base region (Fig 4A), consistent with the EphB2 and EphB3 data described above. EYFP expression patterns in colon similar to those seen in TAM-treated CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice were also obtained when we used TAM-treatment to activate the Lgr5-driven CreERT2 transgene to target Apc and Ctnnb1 alleles and EYFP expression was used to mark Lgr5-expressing cells (Fig 4A).
Consistent with our studies of Lgr5 and Msi expression patterns in colon epithelium, the levels of transcripts encoding presumptive stem cell markers, including Lgr5, CD44, Msi1, and Hopx, were also found to increase dramatically in the colon tissues of CDX2P-CreERT2 Apcfl/fl mice (S4 Fig). The induction of genes encoding stem cell markers and other selective β-catenin/Tcf target genes (such as Axin2, Nkd1, Ccnd1 and Irs1) observed in Apc-deficient colon epithelium was significantly suppressed in colon epithelium from CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice (S4 Fig). Taken together, our data indicate that the robust induction of many β-catenin/Tcf-regulated genes that is seen response to Apc inactivation was variably inhibited in reduced Ctnnb1 gene dosage and β-catenin protein levels in mouse colon epithelium. In the setting of inactivation of one Ctnnb1 allele, the inability of Apc inactivation to substantially activate certain key β-catenin/TCF-regulated genes with functions in colon crypt compartmentalization and cell migration (e.g., EphB2 and EphB3) or stem cell fate (e.g., Lgr5 and Msi) is likely to underlie the dramatic abrogation of adenoma formation in CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice.
Myc is a well-known β-catenin/TCF-regulated target gene [10, 26], and we found that the strong induction of Myc gene expression in mouse colon epithelium seen following Apc bi-allelic inactivation was abrogated when Apc bi-allelic inactivation occurred concurrently with somatic inactivation of one Ctnnb1 allele (Fig 5A). Another interesting observation was that Ctnnb1 transcripts were increased roughly 2-fold in proximal colon tissues following Apc bi-allelic inactivation in colon epithelium with wild type Ctnnb1 gene dosage, compared to the levels of Ctnnb1 transcript in untargeted colon tissues of Apcfl/fl mice (Fig 5B). Hemizygous Ctnnb1 gene dosage was associated with an inability of Apc bi-allelic inactivation to activate Ctnnb1 transcript levels in proximal colon tissues (Fig 5B). The effects of Apc inactivation and Ctnnb1 gene dosage in mouse colon epithelium on Myc and Ctnnb1 transcript levels did not appear to simply reflect a change in the epithelial cell numbers in Apc-mutant colon epithelium, because transcripts for the epithelial markers epithelial cell adhesion molecule (Epcam) and E-cadherin (Cdh1) were similar in the mouse colon tissues independent of genotype (Fig 5C and 5D). The findings indicate that not only does Apc inactivation lead to increased β-catenin protein levels in murine colon tissues, but Ctnnb1 transcript levels in the colon tissues are also increased by Apc inactivation, consistent with an apparent feed-forward mechanism for up-regulation of Ctnnb1 transcripts following Apc inactivation. Because the Apc mutation-dependent induction of Ctnnb1 transcripts in mouse colon epithelium was not seen in the setting of reduced Ctnnb1 gene dosage, the findings imply that the feed-forward mechanism for Ctnnb1 induction may require sufficient levels of β-catenin and β-catenin/TCF-dependent transcription.
Of interest with regard to a role for β-catenin and β-catenin/TCF transcription in regulating Ctnnb1 transcription is that chromatin immunoprecipitation (ChIP) studies from the ENCODE project indicate that the TCF4 protein, encoded by the TCF7L2 gene, is bound in the proximal promoter and exon 1 region of the CTNNB1 gene in selected cell lines. The mouse Ctnnb1 and human CTNNB1 promoter regions lack known TCF family protein consensus binding elements. Nonetheless, to further explore the role of CTNNB1 transcript and β-catenin levels in regulating CTNNB1 transcription, we generated a reporter gene construct in which a 555 bp fragment of human CTNNB1 upstream and exon 1 sequences (-336 to +219 relative to the transcription start site) were cloned upstream of a firefly luciferase sequence (Fig 6A). This region of the CTNNB1 gene corresponds to the region that the ENCODE ChIP data indicates is occupied by the TCF family member TCF4 in some cell lines. We used two different doxycycline-induced shRNAs against CTNNB1 to study effects of antagonizing CTNNB1 endogenous transcript levels in the DLD1 human colon cancer cell line, which harbor APC defects leading to constitutive activation of β-catenin/TCF signaling (Fig 6B). The doxycycline-mediated shRNA-mediated inhibition of CTNNB1 transcript levels in DLD1 cells also led to marked inhibition of MYC expression (Fig 6B). In addition, we found CTNNB1 reporter luciferase activity was inhibited by about 30–40% by the reduction in CTNNB1 transcripts in the cells, whereas, the prototypical Wnt/β-catenin/TCF reporter gene construct TOPflash was more robustly inhibited by the reduction in CTNNB1 levels (Fig 6C). These studies and data complement the primary mouse colon tissue findings presented above by showing that reduction of CTNNB1 levels in colon cancer cells has a demonstrable effect on CTNNB1 transcriptional activity.
Canonical (β-catenin-dependent) Wnt signaling is dependent on increases in the levels and localization of a hypo- or un-phosphorylated pool of β-catenin, which is often termed the “free” or “active” pool of β-catenin in the cytoplasm and nucleus of cells with Wnt pathway activation [27, 28]. Active β-catenin can function as a co-activator for TCF-dependent transcription of endogenous Wnt/β-catenin/TCF target genes [27]. To further address how reduced CTNNB1 gene expression affects β-catenin protein levels and β-catenin/TCF-regulated target gene induction in colon epithelial cells, we used an shRNA approach to antagonize CTNNB1 transcript and β-catenin protein levels in colon cell lines. In the immortalized, non-neoplastic human colon epithelial cell (HCEC) line, through use of a doxycycline (DOX)-regulated shRNA against APC, we reduced endogenous APC gene and protein expression in the cells to less than 10% of control levels (S5 Fig). Following APC shRNA induction, the levels of active β-catenin, as detected with a previously described antibody against the hypo-phosphorylated or active form of β-catenin, were significantly increased, whereas only a minor increase in total β-catenin levels was seen (Fig 7A and S6 Fig). Concurrent DOX-mediated induction of the APC shRNA and either of the two independent CTNNB1 shRNAs, which reduced CTNNB1 transcript levels to about 20–30% of control levels in HCECs, led to dramatic inhibition of the APC inactivation-stimulated effects on active β-catenin protein levels, but only modest to moderate reduction in the levels of total β-catenin protein (Fig 7A and S6 Fig). The marked effects of the APC and CTNNB1 shRNA approaches on the active β-catenin pool, with only more modest to moderate effects on total β-catenin levels in HCEC cells were reproducible (S6 Fig). Following APC shRNA induction by DOX treatment, expression of multiple β-catenin/TCF-regulated target genes, such as AXIN2, BMP4, NKD1 and IRS1, was significantly induced in HCECs (Fig 7B–7E). These APC shRNA-mediated increases in β-catenin/TCF-regulated target gene expression were almost completely abolished by shRNA-mediated inhibition of β-catenin (Fig 7B–7E). We also studied sub-cellular localization of β-catenin in the HCEC cells following APC shRNA induction and combined APC and CTNNB1 shRNA induction by DOX. Consistent with the marked increase in active β-catenin levels following APC shRNA induction, we found β-catenin protein mainly accumulated in the cytosol and nucleus of HCECs (S7 Fig). Concurrent induction of both the APC and CTNNB1 shRNAs in HCECs dramatically reduced the levels of β-catenin protein in the nucleus and cytoplasm (S7 Fig), consistent with the notion that the active, signaling pool of β-catenin in the cytoplasm and nucleus is highly sensitive to changes in Ctnnb1 transcript levels.
The strong inhibitory effect on the active pool of β-catenin compared to that for total β-catenin when CTNNB1 transcript levels were reduced in HCECs was further studied in three human colon cancer cell lines stably transduced with the two DOX-inducible CTNNB1 shRNAs. These included a colon cancer cell line with a gain-of-function mutation in CTNNB1 (HCT116) and two colon cancer cell lines with APC loss-of-function mutations (DLD1 and SW480). At 7 days after DOX-induction of the CTNNB1 shRNAs, in the three colon cancer cell lines, we found moderate (HCT116) to dramatic (DLD1 and SW480) decreases in the active pool of β-catenin protein with only modest changes in total β-catenin protein levels (S8 Fig). Expression of the CTNNB1 shRNAs led to potent inhibition of the expression of Wnt/β-catenin/TCF-regulated target genes in the DLD1 and HCT116 cells, including AXIN2, BMP4, NKD1, LGR5, and CD44 (S9 Fig).
To assess the role of β-catenin function in another Apc mutation-dependent tumor model, we explored the role of Ctnnb1 gene dosage in a mouse model of ovarian endometrial adenocarcinoma (OEA) arising from bi-allelic inactivation of both the Apc and Pten genes [29]. Prior studies have shown that the Wnt/β-catenin/Tcf signaling pathway is deregulated by mutations in 16%–38% of human OEAs, and PTEN mutations are often seen in the OEAs with Wnt pathway mutations [29–32]. In the mouse OEA model, tumors are initiated by conditional inactivation of the Apc and Pten genes following injection of AdCre into the right ovarian bursa of Apcfl/fl Ptenfl/fl mice [29]. Interestingly, in both Apcfl/fl Ptenfl/fl mice and Apcfl/fl Ptenfl/fl Ctnnb1fl/+ mice, adenocarcinomas morphologically similar to human OEAs formed following AdCre injection, with 100% penetrance and no difference in tumor latency between mice with two wild type Ctnnb1 alleles or one wild type and one floxed Ctnnb1 allele (Table 1). In addition, no significant differences in survival rates, tumor volumes, and rates of liver metastasis were found between AdCre-injected Apcfl/fl Ptenfl/fl mice and Apcfl/fl Ptenfl/fl Ctnnb1fl/+ littermates (Table 1), and OEAs arising in both lines of mice shared similar histological features and immunohistochemical staining patterns for cytokeratin-8 (CK8), E-cadherin and α-inhibin (Fig 8A and Table 1). Efficient Cre-mediated deletion of Ctnnb1 and Apc was confirmed in tumors from these mice, and no OEAs arose in the AdCre-injected right ovaries in Apcfl/fl Ptenfl/fl Ctnnb1fl/fl mice, indicating OEAs could not arise from cells completely lacking β-catenin.
The findings on the lack of a demonstrable effect of Ctnnb1 hemizygous gene dosage in the mouse OEA model contrast with the findings above, where Apc-mutation-dependent polyposis in colon epithelium was dramatically suppressed by Ctnnb1 hemizygous inactivation. Nonetheless, similar to the situation in mouse colon, based on immunohistochemical staining, the presumptive Wnt pathway signaling-competent pool of β-catenin in the nucleus and cytoplasm was significantly reduced in the OEAs in Apcfl/fl Ptenfl/fl Ctnnb1fl/+ mice compared to OEAs in the Apcfl/fl Ptenfl/fl mice (Fig 8B). We also examined the β-catenin/TCF-mediated gene transcription in the OEAs arising in the Apcfl/fl Ptenfl/fl mice and Apcfl/fl Ptenfl/fl Ctnnb1fl/+ mice. Consistent with the β-catenin dosage-dependent effects of Ctnnb1 transcripts seen in mouse Apc-mutant colon tissues described above, Ctnnb1 transcripts were significantly reduced in the OEAs arising in Apcfl/fl Ptenfl/fl Ctnnb1fl/+ mice compared to the OEAs in Apcfl/fl Ptenfl/fl mice (Fig 8C). Interestingly, although the Ctnnb1 hemizygous state in the Apc- and Pten-mutant OEAs markedly suppressed the induction of some β-catenin/TCF-regulated target genes, such as Axin2 and Nkd1 (Fig 8C), hemizygous Ctnnb1 function did not abrogate induction of Myc transcripts in the OEAs (Fig 8C). Therefore, our findings showing that Ctnnb1 hemizgyous state did not prevent development of Apc- and Pten-mutant OEAs even though there was a reduction in β-catenin levels and expression of some β-catenin/TCF-regulated genes suggest that retention of Myc induction in OEAs with hemizygous Ctnnb1 function, but not in Apc-deficient colon epithelium with hemizygous Ctnnb1 function, may be a contributing factor in the observed differences in tumor development in the two tissues. We also studied the consequences of shRNA-mediated inhibition of CTNNB1 on MYC gene expression in the TOV112D human ovarian endometrioid carcinoma cell line that harbors a CTNNB1 oncogenic mutation leading to β-catenin/TCF dysregulation [33]. We found that doxycycline-mediated induction of the two CTNNB1 shRNAs in TOV112D cells reduced CTNNB1 levels to about 50% of baseline (Fig 8D), but no statistically significant effect on MYC transcript levels was seen.
Mutations inactivating the APC tumor suppressor gene are believed to be critical initiating lesions in the majority of colon adenomas and carcinomas [6, 34, 35]. APC mutations are likely key contributing factors in the development of some other cancer types, including a subset of human OEAs. The best understood function of the APC protein is to act as a component of a phosphorylation- and ubiquitination-dependent destruction complex that regulates the free or active pool of β-catenin. This pool of β-catenin functions as a regulator of TCF transcription in the Wnt pathway signaling [1]. In the studies described above, we assessed the effects of Ctnnb1 gene dosage on Apc mutation-instigated tumorigenesis in mouse genetically engineered colon and ovarian tumor models and in cultured cells. We found the florid polyposis phenotype resulting from somatic Apc bi-allelic inactivation in mouse colon epithelium is potently inhibited by concurrent somatic inactivation of one Ctnnb1 allele. The few polyps arising in CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice were found to have escaped Ctnnb1 targeting, though Cre-mediated somatic inactivation of both Apc alleles occurred in the lesions, likely reflecting strong positive selection for maintenance of wild type Ctnnb1 gene dosage for Apc-mutant colon adenomas to arise and persist. In contrast to the situation in colon epithelium, in a mouse model of the human OEAs that harbor inactivating mutations in the APC and PTEN genes, we found that, regardless of whether the mice had two wild type Ctnnb1 alleles or one wild type and one targeted Ctnnb1 allele, adenocarcinomas morphologically similar to human OEAs formed with 100% penetrance and no differences in latency, size, morphology, or metastatic potential of the lesions arising from AdCre-mediated targeting of the Apc and Pten genes.
In-depth mutational analyses of the germline and somatic mutations in adenomas arising in patients with FAP led to the proposal there was strong biological selection for a “just-right” level of β-catenin signaling that would be optimal for tumor formation [36]. Some prior studies have used genetic approaches to study experimentally the effects of Ctnnb1 gene dosage on Apc mutation-dependent tumorigenesis in the small intestine, liver, and mammary gland in mouse models [14, 15]. The earlier work indicated that the Ctnnb1+/- constitutional hemizygous state can inhibit intestinal and liver tumorigenesis in mice carrying Apc mutations. In mammary gland tumorigenesis, tumorigenesis was enhanced in Apc1638N Ctnnb1+/- mice relative to Apc1638N Ctnnb1+/+ mice, perhaps because Ctnnb1 functions as a tumor suppressor gene in Apc1638N mammary gland tumors via β-catenin’s role in E-cadherin-dependent tumor suppression [14]. While our findings in a mouse Apc mutation-dependent colon tumorigenesis model are consistent with the prior work on Apc mutation-instigated small intestine and liver tumorigenesis, some significant differences in the studies should be noted. The prior work emphasized models where the mice carried constitutional mutations in one Apc allele and tumors arose following stochastic loss or inactivation of the remaining wild type Apc allele. In addition, mice in the prior small intestine work were constitutionally hemizygous for Ctnnb1. In our Apc mutation-dependent colon tumorigenesis model, both Apc alleles are somatically inactivated in colon epithelium by Cre-mediated targeting, and the Ctnnb1 hemizygous deficiency state was also somatically generated only in the colon epithelial cells by Cre-mediated targeting. In addition, our OEA model work contrasts with the prior published work, as it also relies on somatic targeting of Apc and Ctnnb1. The OEA results also differ from our own colon tumorigenesis results, as the findings indicate Ctnnb1 hemizygous gene dosage had no demonstrable effect on cancer latency, size, or morphology or the metastatic potential of mouse OEAs arising from combined somatic inactivation of Apc and Pten.
Besides highlighting tissue-specific differences for Ctnnb1 gene dosage in Apc mutation-instigated colon and ovarian tumorigenesis, our studies and data have provided several unique and in-depth insights into cell and tissue mechanisms by which Ctnnb1 gene dosage likely contributes to Apc mutation-dependent phenotypes in mouse colon epithelium. Inactivation of one Ctnnb1 allele markedly inhibited the increases in β-catenin cytoplasmic and nuclear levels that result from bi-allelic Apc inactivation in mouse colon epithelium. In turn, there was strikingly attenuated expression of key β-catenin/TCF-regulated target genes, including those encoding the EphB2/B3 receptors, and the stem cell markers Lgr5, Msi1, and Hopx. Of significant interest in terms of a likely key mechanisms through which Ctnnb1 gene dosage inhibits adenoma formation, the inability of the Apc-mutant colon epithelial cells to up-regulate and alter the crypt (high)-surface (low) gradient of EphB2/B3 expression appears to restrict high levels of EphB2/3 expression to the crypt base. Activated β-catenin/TCF transcription has been implicated in repression of ephrinB expression [11, 23]. We found that the robust ephrinB1/B2 expression seen in the upper two-thirds of normal colon crypts as well as in the normal colon surface epithelium was maintained in Apc-mutant crypts when one Ctnnb1 allele was inactivated. In contrast, Apc-mutant crypts with intact Ctnnb1 dosage markedly down-regulated ephrinB1/B2 expression and dramatically upregulated and expanded EphB2/B3 expression throughout the colon crypts. Because the EphB and ephrin molecules mediate critical repulsive interactions in intestinal crypts, the maintenance of the normal inverse EphB/ephrinB gradient from crypt base to cell surface in Apc-mutant crypts where one Ctnnb1 allele is inactive restricts the expansion of the Lgr5-positive crypt stem cell pool and the crypt fission/branching that would result from unrestrained crypt stem cell expansion and altered migration [11, 23]. As a result of the preservation of the inverse gradient of EphB/ephrinB expression in Apc mutant crypts with reduced Ctnnb1 dosage, stem cell expansion and the dysplastic and adenomatous changes induced by Apc inactivation in colon epithelium are potently inhibited, even though Lgr5 and some other stem cell marker genes are modestly increased in expression in the targeted crypts of CDX2P-CreERT2 Apcfl/fl Ctnnb1fl/+ mice relative to crypts in normal mice.
Our findings showing that Ctnnb1 transcripts are up-regulated in Apc-mutant mouse colon epithelium as well as in Apc-mutant mouse OEAs, together with our findings that Ctnnb1 hemizygous gene dosage inhibited Apc mutation-dependent Ctnnb1 transcript induction in colon and ovarian tumor models imply that the Ctnnb1 gene is subject to feed-forward activation by β-catenin levels and β-catenin/TCF-regulated transcription. Of interest with regard to a role for β-catenin and β-catenin/TCF transcription in regulating Ctnnb1 transcription are chromatin immunoprecipitation (ChIP) studies from the ENCODE project reporting that the TCF4 protein, encoded by the TCF7L2 gene, is bound in the promoter region of the CTNNB1 gene in selected cell lines. Based on our mouse colon tissue studies and the ENCODE project findings, we generated a CTNNB1 reporter gene construct containing 555 bp of human CTNNB1 upstream and exon 1 sequences and found that shRNA-mediated inhibition of CTNNB1 endogenous gene expression in the APC-mutant DLD1 human colon cancer cell line led to inhibition of the activity of the CTNNB1 reporter gene. These data demonstrate that CTNNB1 transcript levels affected CTNNB1 transcription in colon cancer cells. The lack of known TCF consensus element binding sites in the mouse Ctnnb1 and human CTNNB1 promoter regions currently limits support for the argument that β-catenin/TCF transcription directly regulates Ctnnb1 transcription, though further studies to address the point will need to be pursued.
In contrast to the near complete abrogation of Myc induction in Apc-mutant colon epithelium with one Ctnnb1 allele, Myc induction was retained in the Apc-mutant mouse OEAs with one functional Ctnnb1 allele. Of note, in prior studies, it has been shown that hemizyous inactivation of Myc dramatically inhibited Apc mutation-induced small intestine tumor phenotypes, but not Apc mutation-induced effects on liver cell proliferation and size [37, 38]. Hence, the findings from our work and these prior studies [37–39] highlight Myc as perhaps one of the key β-catenin/TCF-regulated genes with tissue-specific differences in its regulation by β-catenin/TCF that may account for why Ctnnb1 hemizygous state abrogates Apc mutation-induced effects in some tissues (e.g., small intestine and colon epithelium) but not in other tissues (e.g., liver and ovarian epithelium). The identification of a possible feed-forward mechanism for β-catenin and β-catenin/TCF transcription in regulating Ctnnb1 transcript levels following Apc inactivation are also potentially interesting with regard to Myc, because the ENCODE project work also indicates that the Myc protein is bound in the promoter and intron one regions of the Ctnnb1 gene in selected cell lines. As such, β-catenin/TCF transcription may cooperate in some fashion with Myc, itself encoded by a β-catenin/TCF target gene, in a more complex feed-forward loop to activate Ctnnb1 transcription in certain cell types when Apc is inactivated.
Further studies are needed to better understand the details of the apparent feed-forward mechanisms through which β-catenin and β-catenin/TCF transcription may regulate Ctnnb1/CTNNB1 transcription in the setting of Apc/APC inactivation. Besides Myc, other β-catenin/TCF target genes may also be differentially regulated in a tissue- and context-dependent fashion, perhaps contributing in some fashion to the tissue-specific differences of Ctnnb1 hemizygous gene dosage on Apc mutation-instigated tumorigenesis observed. In addition, the basis for the dramatic changes in the free or active pool of β-catenin protein relative to the more modest effects on total β-catenin protein levels when Ctnnb1/CTNNB1 transcript levels are reduced in colon epithelial cells with Wnt pathway dysregulation remains to be elucidated. Nonetheless, our findings highlight the possibility that novel approaches and/or agents that can reduce CTNNB1 transcript levels and/or the free pool of β-catenin protein might have quite dramatic effects on the development and perhaps persistence of neoplastic cells with Wnt pathway defects.
To target Apc and/or Ctnnb1 alleles in colon tissues, CDX2P-G22Cre transgenic mice [16], or CDX2P-CreERT2 transgenic mice [17], or Lgr5-EGFP-IRES-CreERT2 (B6.129P2-Lgr5tm1(cre/ERT2)Cle/J) transgenic mice [8] (The Jackson Laboratory, Bar Harbor, ME), were first intercrossed with mice homozygous for Apc targeted alleles (Apcfl/fl, 580S) [40] and Ctnnb1-targeted alleles (Ctnnb fl/fl, B6.129-Ctnnb1tm2Kem/KnwJ) [41]. The resulting Cre positive Apcfl/+ Ctnnb1 fl/+ mice were then crossed to Apcfl/fl mice in order to target two alleles of Apc and one allele of Ctnnb1 (Apcfl/fl Ctnnb1 fl/+) or only alleles of Apc (Apcfl/fl), respectively. The Cre positive Apcfl/fl Ctnnb1 fl/+ and Apcfl/fl littermates were compared and the Cre negative littermates served as normal control. The CDX2P-CreERT2 Lgr5-EGFP-IRES-CreERT2 Apcfl/fl compound mice or CDX2P-CreERT2 Lgr5-EGFP-IRES-CreERT2 Apcfl/fl Ctnnb1 fl/+ compound mice were constructed by crossing Lgr5-EGFP-IRES-CreERT2 Apcfl/fl mice to CDX2P-CreERT2 Apcfl/fl mice and CDX2P-CreERT2 Apcfl/fl Ctnnb fl/fl littermates, respectively. To assess Cre-mediated recombination or Wnt signaling in colon epithelium, mice carrying the Gt(ROSA)26Sor tm1(EYFP)Cos/J reporter allele (EYFP) [42] or the B6.129P2-Axin2tm1Wbm/J allele (Axin2-LacZ) [43] (The Jackson Laboratory) were bred into CDX2P-CreERT2 Apcfl/fl mice or CDX2P-CreERT2 Apcfl/fl Ctnnb1 fl/+ mice. To assess the role of β-catenin function in another Apc mutation-dependent mouse tumor model, we used the previously described mouse model of ovarian endometrioid adenocarcinoma (OEA), arising from bi-allelic inactivation of both the Apc and Pten genes (Apcfl/fl Ptenfl/fl) [29]. To introduce the floxed Ctnnb1 allele, Ctnnb fl/fl mice were first crossed to Apcfl/fl Ptenfl/fl mice to generate Apcfl/+ Ptenfl/+ Ctnnb fl/+ mice, and then Apcfl/+ Ptenfl/+ Ctnnb fl/+ mice were bred to Apcfl/fl Ptenfl/fl mice to generate Apcfl/fl Ptenfl/fl, and Apcfl/fl Ptenfl/fl Ctnnb fl/+ mice. All mice were on a mixed C57BL/6 and 129 background, which were backcrossed to C57BL/6 mice for at least 10 generations, except the EYFP reporter mice and the CDX2P-CreERT2 transgenic mice, which were backcrossed for 7 and 3 generations, respectively. All experimental compound mice were on a mixed C57BL/6 and 129 background, and littermates with similar genetic background and different genotypes were used for comparison (see breeding scheme above). Animal husbandry and experimental procedures were carried out under approval from the University Committee on Use and Care of Animals, University of Michigan and according to Michigan state and US federal regulations. All the mice were housed in specific-pathogen free (SPF) conditions. After weaning, rodent 5001 chow and automatically supplied water were provided ad libitum to mice. Animals were euthanized and analyzed at the specified time points, based on particular study design parameters or defined humane treatment and euthanasia guidelines.
Human colonic epithelial cells (HCEC) [44] were kindly provided by Dr. Jerry Shay (UT Southwestern Medical School, Dallas, TX) and routinely grown on media made up with Dulbecco's modified Eagle's medium (DMEM; Life Technologies, Grand Island, NY) and medium 199 (Thermo Scientific HyClone, Waltham, MA) at the ratio of 4:1, supplemented with EGF (25 ng/mL) (PeproTech, Inc, Rocky Hill, NJ), insulin (10 μg/mL, Life Technologies), hydrocortisone (1 μg/mL), transferrin (2 μg/mL), sodium selenite (5 nm) (all from Sigma-Aldrich, St Louis, MO), and 2% cosmic calf serum (Thermo Scientific HyClone). Cells were cultured on Primaria dishes (BD Biosciences, San Jose, CA) or chamber slides (Lab-Tek II, Vernon Hills, IL) and grown in 2% oxygen and 7% carbon dioxide. HCEC cells were infected with a TRIPZ inducible lentiviral vector (GE Dhamacon, Lafayette, CO) carrying a shRNA against APC (targeting sequence: 5’-CAAATCATATGGATGATAA-3’) or a non-silencing scramble shRNA (Scrmbl). Cells were selected with 1μg/mL of puromycin (Sigma-Aldrich) for 5 days. The resulting stable cell lines (HCEC/APC shRNA or HCEC/Scrmble) were further transduced with TRIPZ lentiviruses driving expression of two different shRNAs targeting CTNNB1 (CTNNB1-1 and CTNNB1-2; targeting sequence for CTNNB1-1: 5’-TGGGTGGTATAGAGGCTCT-3’; and targeting sequence for CTNNB1-2: 5’- AGCTGATATTGATGGACAG-3’) or a non-silencing scramble shRNA (Scrmbl). Human colon cancer cell lines, HCT116, SW480, and DLD1, and human OEA-derived cell line, TOV-112D, were grown in 5% CO2 with DMEM containing 10% fetal bovine serum and penicillin/streptomycin. HCT116, SW480, DLD1 and TOV-112D cells stably expressing the shRNAs targeting CTNNB1 (CTNNB1-1 and CTNNB1-2) or a non-silencing scramble shRNA (Scrmbl) were made in the same way as HCEC cells. Expression of shRNAs was induced by incubation of cells with doxycycline (DOX; Sigma-Aldrich) at 2 μg/ml or a solvent control for 3 days (for HCEC cells) or 7 days (for HCT116, SW480, and DLD1 cells). The degree of inhibition of the shRNAs on APC transcripts and protein and CTNNB1 transcripts and the respective β-catenin protein was assessed by qRT-PCR and Western blotting assays.
DNA fragment containing human CTNNB1 sequences from −336 to +219 relative to the transcription start site was obtained by PCR amplification of genomic DNA, and was subcloned upstream from the luciferase reporter gene in the pGL3Basic reporter vector (Promega, Madison, WI), using the MluI and XhoI sites. The forward primer for generating the CTNNB1 reporter construct was 5′-ACGCGTGCTGCTCTCCCGGTTCG -3′; the reverse primer for generating the CTNNB1 reporter construct was 5′- CTCGAGCAGGGGAACAGGCTCCTC-3′.
Mice with the CDX2P-CreERT2 transgene or Lgr5-EGFP-IRES-CreERT2 were injected intraperitoneally with TAM (Sigma-Aldrich) dissolved in corn oil (Sigma-Aldrich). For two TAM daily dosing, we used 150mg/kg weight per dose; for three consecutive daily doses, we administered TAM at 100mg/kg weight per dose. Mice were injected with TAM at 2- to 3-months of age. For OEA induction, 5 x 107 plaque-forming units of replication-incompetent recombinant adenovirus expressing Cre recombinase (AdCre, from the University of Michigan’s Vector Core) were injected into the right ovarian bursal cavities of 6–10 week old female mice as previously described [29].
Mouse tissues were prepared for paraffin-embedding or cryosectioning as described previously [16]. For assessment of cell proliferation, mice were pulsed with 5-bromo-2-deoxyuridine (BrdU; Sigma-Aldrich) for 1 hr before euthanasia. Sections of paraffin-embedded human or mouse tissues were subjected to immunohistochemical analysis as previously described [45]. The following primary antibodies were used for immunohistochemical analysis with sections of paraffin-embedded tissues: mouse anti-BrdU (1:500; BD Biosciences); rabbit anti-lysozyme (1:2000; Dako, Carpinteria, CA); mouse anti-β-catenin (1:800; BD Biosciences); rat anti-CK8 (1:100, The Developmental Studies Hybridoma Bank, Iowa City, IA); goat anti-E-cadherin (1:100, R&D Systems, Minneapolis, MN); mouse anti-α-inhibin (1:200, Bio-Rad Laboratories, Inc., Raleigh, NC). For BrdU staining, tissue sections were treated with 2N HCl at 37°C for 30 min after performing antigen retrieval with citrate buffer (pH 6.0, Biogenex, San Ramon, CA). For immunofluorescence using frozen sectioned tissues, mouse colon and intestinal tissues were fixed in 4% paraformaldehyde (PFA) overnight, cryo-protected and frozen in O.C.T. (Fisher HealthCare, Houston, TX 77038). Standard immunofluorescence staining was performed on 6-μm frozen sections with rabbit anti-lysozyme antibody (1:1000; Dako). For immunofluorescence using paraffin-embedded tissues, the following primary antibodies were used: rabbit anti-lysozyme (1:1000; Dako), rabbit anti-Sox9 (1:200; Millipore, Temecula, CA), rat anti-Msi1 (1:500; a gift from Dr. Hideyuki Okano [46, 47]), goat anti-EphB2 (1:100; R&D Systems), goat anti-EphB3 (1:100; R&D Systems), goat anti-ephrinB1 (1:200; R&D Systems), goat anti-ephrinB2 (1:100; R&D Systems), mouse anti-α-tublin (1:1000; Sigma-Aldrich), and rabbit anti-Crb3 (1:1000; kindly provided by Dr. Benjamin Margolis at University of Michigan). The secondary antibodies used were Alexa fluor 488-conjugated donkey anti-goat, Alexa fluor 488-conjugated donkey anti-rabbit, Alexa fluor 488-conjugated goat anti-rabbit, Alexa fluor 594-conjugated goat anti-mouse, Alexa fluor 488-conjugated goat anti-mouse, Alexa fluor 594-conjugated goat anti-rabbit, and Alexa fluor 488-conjugated goat anti-rat (Molecular Probes, Life Technologies, Carlsbad, CA), diluted at 1:1000. DNA was labeled by Hoechst 33342 (Molecular Probes, Life Technologies) by adding to the washing buffer at 5 μg/ml. β-gal analysis for mouse with Axin2-LacZ reporter was performed as described previously [16].
To assess apoptosis, TUNEL assays were undertaken using 4-μm sections of formalin-fixed, paraffin-embedded mouse colon tissues, after the tissue sections were deparaffinized, rehydrated and treated with 20 μg/ml protease K (Roche Applied Sciences, Indianapolis, IN) at 37°C for 15 min. The nicked DNA was labeled by using terminal transferase (TdT) (New England Biolabs, Ipswich, MA) and Biotin-16-UTP (Roche Applied Sciences) according to the manufacturer’s recommendation. The signal was detected by using the Vectastain ABC kit (Vector Laboratories, Burlingame, CA) according to the manufacturer’s suggestion.
The spindle angles were defined by the orientation of mitotic spindles, based on α-tubulin staining, relative to the most adjacent apical membrane, as indicated by Crb3 staining. The mitotic spindle axis angle relative to the planar axis of the cells (defined by the most adjacent apical membrane) was measured by ImageJ (NIH).
Western blot analyses on lysates from HCEC, HCT116, SW480, DLD1 and TOV-112D cells were performed as described [45]. The following antibodies were used: mouse anti-active β-catenin (1:2000; Millipore, Temecula, CA), mouse anti-total β-catenin (1:10,000; BD Biosciences), rabbit anti-APC (clone C-20, 1:1000; Santa Cruz Biotechnology, Santa Cruz, CA), mouse anti-APC (clone Ab-5, 1:1000; Millipore), and mouse anti-β-actin (1:10,000; Sigma). The density of Western blotting bands was quantified using AlphaImager HP system (ProteinSimple, San Jose, CA).
cDNA was synthesized using a high capacity cDNA reverse transcription kit (Applied Biosystems, Foster City, CA). qRT-PCR was performed with an ABI Prism 7300 Sequence Analyzer using a SYBR green fluorescence protocol (Applied Biosystems). See S1 Table for primer sequences used in qRT-PCR.
DLD1 cells, stably expressing two different doxycycline-inducible shRNAs targeting CTNNB1 (CTNNB1-1 and CTNNB1-2) or a non-silencing scramble shRNA (Scrmbl), were treated for 4 days with DOX at 2 μg/ml or a solvent control. At second day during DOX treatment, cells were plated in 35-mm six-well plates. After 12—24h cells were then transfected with 0.5 μg of CTNNB1 reporter or TOPflash, 1 μg of pCDNA3 (Invitrogen) and 0.05 μg of PRL-CMV Renilla luciferase reporter vector (Promega) using Mirus TransIT-LT1 transfection reagent (Mirus Bio, Madison, WI) according to the manufacturer's protocol. Cells were harvested 45h later and luciferase activities were measured using a Dual-luciferase kit and GloMax-Multi Detection System from Promega.
All data for qRT-PCR were evaluated by Student's t test and asterisks denote significance with P < 0.05. Error bars denote standard deviations (S.D.). Kaplan-Meier survival curves were compared by log-rank (Mantel-Cox) test. Chi-Square test was used to determine significance when mitotic spindle angles were compared among mice with different genotypes. P < 0.05 is considered statistically significant.
Animal husbandry and experimental procedures were carried out under approval from the University Committee on Use and Care of Animals, University of Michigan (PRO00005075) and according to Michigan state and US federal regulations. All the mice were housed in specific-pathogen free (SPF) conditions. After weaning, rodent 5001 chow and automatically supplied water were provided ad libitum to mice. Animals were euthanized and analyzed at the specified time points, based on particular study design parameters or defined humane treatment and euthanasia guidelines.
|
10.1371/journal.pgen.1008136 | Mck1 defines a key S-phase checkpoint effector in response to various degrees of replication threats | The S-phase checkpoint plays an essential role in regulation of the ribonucleotide reductase (RNR) activity to maintain the dNTP pools. How eukaryotic cells respond appropriately to different levels of replication threats remains elusive. Here, we have identified that a conserved GSK-3 kinase Mck1 cooperates with Dun1 in regulating this process. Deleting MCK1 sensitizes dun1Δ to hydroxyurea (HU) reminiscent of mec1Δ or rad53Δ. While Mck1 is downstream of Rad53, it does not participate in the post-translational regulation of RNR as Dun1 does. Mck1 phosphorylates and releases the Crt1 repressor from the promoters of DNA damage-inducible genes as RNR2-4 and HUG1. Hug1, an Rnr2 inhibitor normally silenced, is induced as a counterweight to excessive RNR. When cells suffer a more severe threat, Mck1 inhibits HUG1 transcription. Consistently, only a combined deletion of HUG1 and CRT1, confers a dramatic boost of dNTP levels and the survival of mck1Δdun1Δ or mec1Δ cells assaulted by a lethal dose of HU. These findings reveal the division-of-labor between Mck1 and Dun1 at the S-phase checkpoint pathway to fine-tune dNTP homeostasis.
| The appropriate amount and balance of four dNTPs are crucial for all cells correctly copying and passing on their genetic material generation by generation. Eukaryotes have developed an alert and response system to deal with the disturbance. Here, we uncovered a second-level effector branch. It is activated by the upstream surveillance kinase cascade, which can induce the expression of dNTP-producing enzymes. It can also reduce the inhibitor of these enzymes to further boost their activity according to the degrees of threats. These findings suggest a multi-level response system to guarantee the appropriate dNTP supply, which is essential to maintain genetic stability under various environmental challenges.
| To ensure the genome stability, DNA replication is under strict surveillance by the S-phase checkpoint (also known as the intra-S or replication checkpoint) in all eukaryotes [1–5]. The main kinases of the cascade, Mec1ATR and Rad53CHK2, are activated in response to aberrations in DNA replication [6–10]. Among all the various downstream effects, the essential role of Mec1-Rad53 has been demonstrated to be in regulation of the RNR activity in Saccharomyces cerevisiae [3, 11, 12].
RNR catalyzes the reduction of ribonucleotide diphosphates to their deoxy forms, which is the rate-limiting step in the de novo synthesis of deoxyribonucleotide triphosphates (dNTPs), the building blocks of DNA [13]. RNR is normally composed of two large subunits R1 (Rnr1 homodimer) and two small subunits R2 (Rnr2 and Rnr4 heterodimer in budding yeast). Proper and balanced cellular dNTP pools are essential for genome integrity [14, 15]. Therefore, several RNR inhibitors, such as hydroxyurea (HU), clofarabine and gemcitabine, have been exploited for the chemotherapy of several types of cancers [16]. The RNR activity is strictly controlled by multi-layer mechanisms in cells [15, 17]. First, RNR is allosterically regulated through the binding of different forms of effector nucleotides, for example, ATP or dATP. Second, the expression of RNR1-4 genes is controlled at both transcriptional and post-transcriptional levels. For instance, RNR1 gene is activated during G1/S transition by the MBF transcription factor, while the excessive expression of RNR2-4 is repressed by Crt1 (Constitutive RNR Transcription 1, also called Rfx1) through recruiting the Ssn6-Tup1 co-repressor complex to the promoter. Furthermore, RNR3, as an RNR1 paralog, is generally silenced until the release of Crt1 under stressed condition [18]. Third, the RNR enzyme activity is post-translationally inhibited by several small intrinsically disordered proteins such as Sml1, Dif1, and Hug1 in S. cerevisiae and Spd1 in Schizosaccharomyces pombe [15]. Sml1 binds to cytosolic Rnr1 and disrupts the regeneration of the Rnr1 catalytic site [19, 20]. Dif1 promotes the nuclear import of the Rnr2/Rnr4 heterodimer, which is anchored by Wtm1 in the nucleus [21–23], precluding Rnr2/Rnr4 from associating with Rnr1 or Rnr3 to form the RNR holo-enzyme in the cytoplasm. Hug1, like Dif1, also contains a HUG domain, which can inhibit RNR through binding Rnr2 [24, 25].
When cells encounter genotoxic agents, RNR is stimulated by the Mec1-Rad53-Dun1 kinase cascade at both transcriptional and post-translational levels to provide adequate dNTPs for DNA replication/repair [13, 15, 26–29]. One of the key Mec1-Rad53-Dun1 targets is Crt1, which becomes phosphorylated and therefore leaves the promoter of damage-inducible genes such as RNR2-4, HUG1, and CRT1 itself [18]. Apart from the Crt1 repressor, Dun1 also targets RNRs’ protein inhibitors including Sml1 and Dif1 [21–23, 30, 31], both of which are hyperphosphorylated and degraded [22, 32]. Spd1 is degraded in S phase and after DNA damage via the ubiquitin-proteasome pathway as well [33]. Unlike Sml1 and Dif1, Hug1 is induced together with Rnr2-4 due to the removal of the Crt1 repression from its promoter in the presence of genotoxic agents. As a result, Hug1 acts in a distinct but undefined manner compared with its paralogs Sml1 and Dif1 [25]. Intriguingly, the lethality of mec1Δ or rad53Δ can be suppressed by deleting any of the negative regulators of RNR mentioned above (CRT1, SML1, DIF1 or HUG1) [18]. All these findings highlight the importance of RNR regulation by Mec1-Rad53-Dun1.
In this study, we identify that a combinational deletion of MCK1 and DUN1 displays a synergistic effect, reminiscent of the extreme sensitivity of mec1Δsml1Δ or rad53Δsml1Δ to HU. Rad53 kinase is able to phosphorylate Mck1 in vitro. Moreover, CRT1 deletion suppresses the HU sensitivity of dun1Δmck1Δ. Crt1 phosphorylation is significantly compromised in mck1Δ accompanied with the persistence of Crt1 at the RNR promoters and decreased RNR3 induction. Apart from Crt1, Mck1 also negatively regulates HUG1 transcription. Taken together with previous findings, these data suggest that Mck1 and Dun1 define two non-redundant and cooperative branches of the Mec1-Rad53 kinase cascade in fine-tuning RNR activity when cells encounter replication stress.
The deletion of SML1, encoding an Rnr1 inhibitor, is known to suppress the lethality of mec1Δ or rad53Δ cells [34] (Fig 1A). Nevertheless, mec1Δsml1Δ or rad53Δsml1Δ are extremely sensitive to HU (Fig 1B). On the other hand, deletion of the only known downstream kinase of Mec1-Rad53 [27], DUN1, resulted in a much lower HU sensitivity than that of mec1Δsml1Δ or rad53Δsml1Δ. These data raise the possibility that there might be Dun1-independent players downstream the Mec1-Rad53 pathway working in parallel with Dun1 in response to the RNR inhibitor [18, 27] (Fig 1A).
We identified candidates of the novel Mec1-Rad53 downstream players using the synthetic genetic array approach [35, 36]. A DUN1 null mutant was crossed with the single deletion library of the cell-cycle-regulated non-essential genes. After acquiring the final double mutants through pinning on a series of selective media, the growth of the double mutants was analyzed in the presence or absence of HU. Gene mutants, when in combination with DUN1 deletion, that showed a synthetically sick or lethal phenotype were selected as the potential candidates of the Mec1-Rad53 downstream players. Among them, the mck1Δdun1Δ double mutant grew normally in the absence of HU (Fig 1B). However, it showed a remarkable HU sensitivity similar to that of rad53Δsml1Δ or mec1Δsml1Δ. The mck1Δ single mutant alone displayed mild sensitivity to 200 mM HU. These results indicate that Mck1 might work in parallel with Dun1 and has a Dun1-independent function in the cell’s survival in the presence of HU.
Given that MCK1 encodes one of the GSK-3 family serine/threonine kinases in budding yeast, we deleted its paralog YGK3 and orthologs MRK1 and RIM11. None of them exhibited HU sensitivity (S1A Fig) or synthetic interaction with DUN1 (S1B Fig). Interestingly, among all GSK-3 family kinases, only RIM11 deletion showed a synthetic HU sensitivity with mck1Δ (S1C Fig, line 3). Further deletion of MRK1 and YGK3 had no effects on mck1Δrim11Δ (lines 13–15). These results demonstrate that Mck1, partially redundant with Rim11, plays a major role in the replication checkpoint. GSK-3 kinases are known to regulate the general stress (e.g., glucose starvation, oxidative, heat shock, and low pH) responsive genes through transcription activators Msn2 and Msn4 [37]. However, the deletion of both MSN2 and MSN4 showed no additive effect with dun1Δ (S1B Fig), implying that Msn2/Msn4 is unlikely the major effectors of Mck1 in response to HU.
In addition to DUN1, MCK1 showed genetic interactions with checkpoint activators and mediators as well. MCK1 deletion markedly exacerbated the HU sensitivity of mre11Δ, ddc1Δ, mrc1Δ or rad9Δ (S2A and S2B Fig), further arguing for a critical role of Mck1 in the S-phase checkpoint pathway.
The synergistic HU sensitivity caused by the combined deletion of MCK1 and DUN1 raises the possibility that they may function cooperatively in the Mec1-Rad53 pathway.
We first tested whether MCK1 is a dosage suppressor of the mec1Δ or rad53Δ lethality. We constructed a high-copy number plasmid with a URA3 marker (pRS426) expressing MCK1 and introduced it into diploid strains wherein one copy of MEC1, RAD53 and SML1 was deleted. After sporulation, the tetrads were analyzed by microscopic dissection. The mec1Δ and rad53Δ spores hardly grew unless carrying the pRS426-MCK1 plasmid or in the absence of SML1 (S2C Fig). To verify it, we induced the loss of this plasmid on a plate containing 5-floroorotic acid (5-FOA). Without the MCK1 overexpression plasmid, neither mec1Δ nor rad53Δ was able to survive (Fig 1C). These results indicate that MCK1 overexpression is able to bypass the essential function of MEC1 and RAD53, validating a previous result of large scale screening [38].
We then examined whether Mck1 physically interacts with Mec1 or Rad53 using yeast two-hybrid assays and found that Mck1 shows positive interaction with Rad53 (Fig 1D). To determine which part of Rad53 is required for the interaction with Mck1, we expressed the forkhead homology-associated domains (FHA1 and FHA2) of Rad53. We found that FHA1 is sufficient to interact with Mck1. FHA1 preferentially binds the phosphothreonine (pThr) peptides bearing a pThr-x-x-D/E/I/L motif (x stands for any amino acid) [29]. Therefore, we mutated all six threonine residues to alanines in Mck1 (mck1-T6A), resulting in abolished interaction with Rad53 (Fig 1D). Among these six threonine residues, the T218A mutation dramatically reduced the Mck1-Rad53 interaction. These results suggest that Mck1 interacts with the FHA1 domain of Rad53 through a canonical phosphorylation-mediated mechanism.
Given the physical association of Mck1 and Rad53, we tested whether Mck1 is a substrate of Rad53. We expressed and purified Rad53 or rad53-KD (a kinase-dead mutant, rad53-K227A) for in vitro kinase assays. Rad53 wild-type (WT), but not the KD mutant, showed robust auto-phosphorylation as indicated by the incorporation of 32P and by the electrophoretic shift (Fig 1E, upper panel, compare lane 1 with 8), indicating that the robust kinase activity is Rad53-specific. Next, we isolated the endogenous FLAG-tagged Mck1 as a substrate through immunoprecipitation via anti-FLAG beads followed by FLAG peptide elution. With the increasing amounts of Rad53 added in the reactions, more 32P was transferred to Mck1 (Fig 1E, middle panel, lanes 2–6). On the other hand, excessive rad53-KD completely failed to phosphorylate the Mck1 substrate (lane 7). A Coomassie brilliant blue (CBB) stained gel revealed nearly equal loading of Mck1 in each reaction (Fig 1E, lower panel). These results suggest that Rad53 is able to phosphorylate Mck1 in vitro. Taken together with the synthetic genetic interaction between MCK1 and DUN1, these data also suggest that Mck1 defines a new downstream branch of Rad53 in parallel with Dun1.
To investigate the exact role of Mck1 in the Mec1-Rad53 pathway, we first asked whether SML1 is a potent suppressor of mck1Δ as well. Surprisingly, SML1 deletion was not able to suppress the checkpoint defect in mck1Δ, in stark contrast to its capability to bypass the essentiality of MEC1 or RAD53 (Fig 2A). Consistently, in the absence of Mck1, Sml1 was not affected at either mRNA or protein level upon HU treatment compared to WT (Fig 2B and 2C). Interestingly, other known effectors of Dun1, e.g., Dif1 and Wtm1, were not the suppressors of mck1Δ as well (Fig 2A). These results suggest that Sml1/Dif1/Wtm1 are not the downstream effector of Mck1, which are mainly targeted by the Dun1 branch.
Besides these RNR sequesters, RNR is also controlled at the transcriptional level mainly through the repressor Crt1 [18]. Indeed, CRT1 deletion significantly suppressed the sensitivity of mck1Δ and dun1Δ mutants to 200 mM and 50 mM HU, respectively (Fig 2D). Intriguingly, CRT1 deletion conferred a better growth of mck1Δdun1Δ double mutant than that of dun1Δ in the presence of up to 50 mM HU (compare lines 5 and 8). These results suggest that Crt1 is controlled by Dun1 and Mck1 in a non-redundant manner. Mck1 is required for efficient RNR induction through antagonizing Crt1, particularly in the presence of a high concentration of HU (e.g., 200 mM).
The low RNR level causes insufficient dNTP supply, which leads to genome instability such as the copy number change of ribosomal DNA (rDNA) located at chromosome XII in S. cerevisiae [39]. Therefore, we examined the rDNA copy number by pulsed-field gel electrophoresis (PFGE) followed by Southern blotting. Without HU treatment, mck1Δ exhibited an rDNA copy number loss phenotype less than dun1Δ (Fig 2E, compare lanes 2 and 3). However, the mck1Δdun1Δ double mutant showed a more severe rDNA repeat loss than every single mutant (lane 4). Moreover, CRT1 deletion prominently ameliorated the rDNA instability phenotype of dun1Δmck1Δ (lane 6). The degrees of rDNA copy number loss in these mutants correlated well with their growth defects in the presence of HU (compare Fig 2D and 2E). These data suggest that Mck1 and Dun1 contribute independently to rDNA/genome stability through regulating Crt1 and thus RNR expression.
To address whether the kinase activity of Mck1 is required for RNR regulation, we mutated two conserved residues within the catalytic core (D164) and activation-loop (Y199) of Mck1 [40]. Both mck1-D164A and mck1-Y199F showed synthetic lethality with dun1Δ in the presence of 50 mM HU (S3A Fig), demonstrating that Mck1’s kinase activity is indispensable in response to replication stress.
It is known that the repression function of Crt1 is relieved through phosphorylation by Dun1 [18]. Since CRT1 is a common suppressor for both mck1Δ and dun1Δ as mentioned above, we then hypothesized that Mck1 kinase may function through Crt1 phosphorylation as Dun1. To test this, we assessed the Crt1 phosphorylation level through western blotting. As reported previously [18], Crt1 displayed a slower mobility shift (Crt1-P) after separation in a high-resolution polyacrylamide gel (S3B Fig). There was a basal level of Crt1 phosphorylation, which was largely dependent on Mck1 (Compare lanes 2–4). The Crt1 phosphorylation level increased significantly following 200 mM HU treatment for 3 h in WT. The deletion of RAD53 or MCK1 caused a relatively lower level of Crt1 phosphorylation than DUN1 deletion, indicating the contribution of the Rad53-Mck1 branch in targeting Crt1.
As Crt1 phosphorylation is cell-cycle-regulated, we next examined its level in the synchronized cell samples. Cells were synchronized by α-factor in G1 and released into the fresh media for the indicated time. Cell cycle progression was monitored by fluorescence-activated cell sorting (FACS). Under normal condition, Crt1 phosphorylation occurred at the beginning of S phase and reached a peak at the end of S phase (60 min) in WT (Fig 3A and 3B). MCK1 deletion caused a decrease in Crt1 phosphorylation, whereas the combined deletion of MCK1 and DUN1 nearly abolished Crt1 phosphorylation. These results allow us to conclude that Mck1 and Dun1 function non-redundantly in Crt1 phosphorylation during normal S phase progression.
In the presence of 200 mM HU, the cell cycle progression of all alleles was almost completely halted for at least 150 min (Fig 3D). This indicates an intact S phase arrest function of the replication checkpoint in these mutants, consistent with the role of Mck1 and Dun1 downstream of Rad53. Importantly, Crt1 phosphorylation occurred more slowly with a significantly lower level in mck1Δ than in dun1Δ and WT (Fig 3C and 3E and S3C Fig). To examine whether Crt1 is a direct target of Mck1, we then performed in vitro kinase analysis basically as described in Fig 1E. As shown in Fig 4A, recombinant Crt1 was phosphorylated by WT Mck1, but barely by a kinase-dead mutant (Mck1-KD, D164A). These data suggest a critical and direct role of Mck1 in Crt1 phosphorylation in both normal and perturbed conditions.
To address the physiological significance of Mck1-mediated Crt1 phosphorylation, we reasoned that Mck1 may target Crt1 to antagonize its repressor function.
We first tested whether Crt1 phosphorylation can suppress the ultra-sensitivity of dun1Δmck1Δ to HU as crt1Δ. Crt1 comprises eight putative Mck1 recognition motifs (S1-S2, S4-S9, Fig 4B), (S/T)-x-x-x-(pS/T)*, where * stands for the priming phosphorylated residue and x for any amino acid [41]. Indeed, Crt1-S299D, a mutant mimicking primed phosphorylation, was more likely targeted by Mck1 in vitro (Fig 4A, compare lanes 3 and 4). We next mutated these serine or threonine residues to aspartic acids to mimic the phosphorylation state. To examine the suppression effect of crt1 mutations, the plasmids expressing various crt1 alleles were transformed into the dun1Δmck1Δcrt1Δ triple mutant. Consistently, a plasmid expressing WT CRT1 prominently sensitized dun1Δmck1Δcrt1Δ to 50 mM HU (Fig 4B, compare lines 9 and 10). Through a series of different combinations, we found that phospho-mimetics of many Mck1 sites (e.g., S222/T226 and S295/S299) and a putative Mec1 site (S323) are capable to suppress the HU sensitivity of the triple mutant to various extents (Fig 4B, lines 3 and 13). These results suggest that these putative Mck1 sites play partially redundant roles in the S-phase checkpoint. Among them, crt1-S5D (S295DS299D) rescued the growth of the dun1Δmck1Δcrt1Δ (Fig 4B, compare line 13 to 9) triple and mck1Δcrt1Δ double mutants (S4 Fig) to an extent comparable to the empty vector, indicating that we have isolated a complete loss-of-function phospho-mimetic mutant of the Crt1 repressor. These results suggest that Mck1 kinase abrogates the repressor function of Crt1 through phosphorylation (predominantly at the Mck1 kinase consensus sites S295/S299).
Interestingly, the dominant Mck1 sites S295 and S299 are located within the DNA binding domain of Crt1 (Fig 4B), raising the possibility that phosphorylation of these sites may regulate its DNA binding capability. Therefore, we next examined the binding of Crt1 on the RNR promoters through chromatin immunoprecipitation (ChIP). Crt1 was significantly enriched at the promoter regions of both RNR2 (Fig 4C) and RNR3 (Fig 4D), which was dramatically reduced after 200 mM HU treatment. These results indicate that Crt1 dissociates from the promoters of RNR2 and RNR3 in response to HU. Nevertheless, the phospho-mimetic mutant proteins (crt1-S5D) retained only approximately 20% enrichment of that of WT even in the absence of HU. Notably, Crt1-S5A still maintained the response to HU, indicating that Crt1 bears other Mck1 sites (e.g., S222 and S226) and/or phosphorylation sites of kinases other than Mck1 (e.g., Dun1 and Mec1). Taken together, these data indicate that Mck1-mediated Crt1 phosphorylation compromises the DNA binding activity and thereby regulates the repressor function of Crt1.
To directly test this, we next checked the expression of the Crt1-controlled genes. In crt1-S5D, RNR3 was constitutively expressed in a significantly higher level than in WT and crt1-S5A even in the absence of HU (Fig 4E). This result is consistent with its suppression effect on mck1Δ shown in Fig 4B and S4, indicating that phospho-mimetic mutation of these Mck1 sites is sufficient to abrogate the repression by Crt1. Similarly, the induction of RNR3 upon HU treatment was significantly compromised in mck1Δ though to a lesser extent than in dun1Δ at both mRNA (Fig 4E and 4F) and protein levels (Fig 4G), which is congruent with their relative HU sensitivity (Fig 2D). To further address the contribution of Mck1 and Dun1 in RNR3 induction, we performed time-course analysis of RNR3 transcription. In the absence of HU, RNR3 was barely expressed in G1-arrested cells (Fig 4H). After release into 200 mM HU for 30 min, the RNR3 transcripts were gradually elevated, which was prominently impaired in dun1Δ. Intriguingly, there was a stark rise in the RNR3 mRNA levels around 90 min in WT cells, which was significantly compromised when MCK1 was deleted. However, unlike dun1Δ, mck1Δ did not show an apparent effect during the initial induction stage (0–60 min), indicating that Mck1 likely functions kinetically later than Dun1 or through an indirect effect in response to HU (Fig 4H). This is also in good agreement with the observations that mck1Δ does not display sick growth in the presence of the moderate concentrations of HU (e.g., 7–50 mM, Figs 1 and 2). Taken together, these results argue for a critical role of Mck1 in antagonizing the Crt1 repression of RNR genes. These data also implicate that Dun1 acts as a primary kinase in initiating RNR3 induction, while Mck1 might be required for the additional augment when more RNR expression is needed (e.g., severe and/or persistent stress).
As shown in Fig 2D, CRT1 deletion only shows a partial suppression of the mck1Δ phenotype. This raises the possibility that there are additional Mck1 targets besides Crt1. To test this, we carefully compared the suppression effects of all known negative regulators of RNR.
Consistent with the results shown in Fig 2A, if we removed only one of the RNR-hijacking proteins including Sml1, Hug1, Dif1 and Wtm1, there was no detectable effects in both mck1Δ (Fig 5A and S5A Fig) and mck1Δdun1Δ mutants (S5B Fig). The possible reasons are: 1) Mck1 does not mainly contribute to regulating the RNR protein localization and/or nuclear-cytoplasmic trafficking; 2) the effects of Mck1 in RNR post-translational regulation may be masked by its dominant effects on the RNR expression level.
To test these possibilities, we next eliminated each RNR-binding protein together with the major suppressor Crt1. Consistently, CRT1 deletion showed suppression in either mck1Δ (Fig 5A, line 4), dun1Δ (S5C Fig, line 3) or mck1Δdun1Δ (S5B Fig, line 8). Further removal of SML1, DIF1 or WTM1 had no additive effects with crt1Δ on either mck1Δ (Fig 5A, lines 5, 7 and 8; S5D Fig) or dun1Δ (S5C Fig, lines 4–7). When we deleted CRT1 and HUG1 in combination, we found a synergistic rescue on mck1Δ (Fig 5A, compare line 6 to 4) and mck1Δdun1Δ (Fig 5B, compare line 8 to 7). On the contrary, hug1Δcrt1Δ exhibited no more suppression for dun1Δ than crt1Δ alone (Fig 5B, compare line 6 to 5). Although the HUG1 gene is adjacent to SML1, HUG1 deletion did not reduce SML1 transcription (Fig 5C). Moreover, despite the short length of the HUG1 gene (207 bp), it indeed acted as a protein because a nonsense mutation of the sole start codon (ATG replaced by TAG) led to the same suppression as hug1Δ (Fig 5D). These results suggest that besides Crt1, Hug1 protein is an additional key downstream effector of Mck1.
To further confirm this, we then examined the Hug1 protein levels in these mutants. Under normal condition, Hug1 protein was barely detectable in WT (Fig 6A, lane 1). Knockout of MCK1, but not DUN1, elevated the Hug1 levels, though to a lesser extent than crt1Δ (compare lanes 2, 3 and 9). Apart from the previously reported repression by Crt1 at the transcriptional level [18], these results indicate that HUG1 is also repressed by Mck1 kinase in normal condition. Under 200 mM HU treatment, mck1Δ caused a prominent increase comparable to crt1Δ, whereas the mck1Δcrt1Δ double mutant resulted in a synergistic augment of the Hug1 level (Fig 6A, compare lanes 6, 13 and 14). On the contrary, DUN1 deletion antagonized the induction of HUG1 caused by crt1Δ (compare lanes 13 and 15) or mck1Δ (compare lanes 6, 8, 14 and 16). These data suggest that Mck1 regulates HUG1 expression in a Crt1-independent manner.
Next, we asked how Mck1 regulates the cellular levels of Hug1. Hug1 is unlikely a substrate of Mck1 kinase due to lack of Mck1 consensus recognition sites. Thus, we examined the HUG1 mRNA levels after 200 mM HU treatment for 3 h. Consistently, the deletion of MCK1 and CRT1 individually led to an increase of nearly 100% and 150% in the HUG1 mRNA levels compared to WT, respectively, whereas deletion of MCK1 and CRT1 in combination resulted in an approximately 400% increase (Fig 6B). On the other hand, dun1Δ eliminated induced HUG1 transcription in crt1Δ and mck1Δ, but not in the crt1Δmck1Δ double mutant. Further removal of CRT1 led to a maximum HUG1 induction, confirming that HUG1 is repressed by Crt1 which is relieved by Dun1. In good agreement with the protein levels of Hug1 mentioned above, these data allow us to conclude that Mck1 inhibits the HUG1 induction at the transcriptional level.
Next, we quantitated the HUG1 mRNA levels at 30-min intervals after 200 mM HU treatment. Transcription of HUG1 was elevated more than 1000-fold within 2 h after HU treatment (Fig 6C). In the absence of MCK1, the induction of HUG1 nearly doubled than WT at each time point. Putting together, these data suggest that HUG1 transcription is exquisitely controlled by a pair of antagonistic mechanisms of the S-phase checkpoint, induction by eliminating the Crt1 repressor function (mainly through Dun1 kinase) and direct inhibition by Mck1 kinase.
Since mck1Δdun1Δ was identified to mimic mec1Δ or rad53Δ in response to HU (Fig 1B), we then tested whether deletion of the main targets of Mck1 and Dun1 is able to suppress the HU sensitivity of their upstream kinase mutants as well. CRT1 deletion alone was sufficient to afford mec1Δ to resist 7 mM HU, whereas SML1 deletion could not (Fig 6D, compare lines 2, 3 and 4). The combined deletion of CRT1 and SML1 slightly facilitated HU resistance of mec1Δ (line 5). We further deleted HUG1 and found significant enhanced HU resistance in mec1Δ (Fig 6D, lines 6 and7) as well as mck1Δdun1Δ double mutant (line 8). These data suggest that Crt1, Hug1 and Sml1 represent the major effectors of the Mec1-Rad53-Dun1/Mck1 kinase cascade in RNR regulation.
We next directly compared intracellular dNTP pools in WT or mck1Δ-related mutants. Because the dNTP level is cell-cycle controlled, cells were first arrested in G1 before releasing into the S phase in the presence of 200 mM HU. Considering that Mck1 may function kinetically later as shown in Fig 4H, we collected the cells after HU treatment for 0, 3 and 6 h for dNTP measurement. In the G1 cells before HU assault (0 h), all strains carrying mck1Δ had a moderate increase in dNTP pools compared with WT (Fig 7A), suggesting a role of Mck1 in regulating the dNTPs levels and balance in normal condition. In all tested strains, chronic HU treatment elicited a dramatic decrease of dNTP pools, with the highest decrease in dATP, which thus became the most limiting of the four dNTPs instead of dGTP (Fig 7B and 7C). These indicate that HU causes dNTP imbalance as well as depletion. In WT and dun1Δ cells, dNTP levels were partially restored in 6 h. However, dNTP restoration was abolished in mck1Δ and mck1Δdun1Δ. Strikingly, dATP remained extremely low in both mutants. These results indicate that the recovery of dNTP homeostasis (including both levels and balance) is dependent on Mck1 in the presence of a high concentration HU.
CRT1 deletion alone led to only a mild elevation of the dNTP levels in the presence of 200 mM HU (Fig 7B). Strikingly, when HUG1 was further deleted, we observed a dramatic expansion of dNTP pools, congruent with the percentages of HUG1 induction in the mck1Δdun1Δcrt1Δ triple mutant shown in Fig 6. Among them, dATP was augmented most significantly, suggesting that Hug1 is also able to restore the balance of four dNTPs impaired by HU. These results indicate that Hug1 is a potent suppressor of the RNR activity, particularly when the repressor function of Crt1 is abrogated. Importantly, the recovery of dNTP levels correlated with the growth of these mutants in the presence of HU (Figs 5A, 5B and 6D). Putting together, we propose that Mck1 defines a secondary effector branch of the Mec1-Rad53 cascade and plays a crucial role in coping with a more severe and/or long-lasting replication insult (Fig 7D).
Dun1-independent RNR induction in response to either exogenous or endogenous replication stress has been observed by different groups [18, 42]. Nearly two decades after the initial report, here we have identified that Mck1 is a new downstream kinase of Rad53 and functions in the Dun1-independent pathway in dNTP regulation.
Cells need to maintain the appropriate amount and balance of all four dNTPs, an even more challenging task when they suffer exogenous or endogenous replication stress. Moreover, cells should have multi-layer response systems to deal with various degrees of stress. Here we prove that Mck1 and Dun1 kinases cooperate to achieve this. Under the unperturbed condition, Crt1 represses the expression of RNR genes to avoid overproducing dNTPs. Under the moderate perturbed condition, the Mec1-Rad53 cascade activates Mck1 and Dun1. At the post-translational level, Dun1 is responsible for releasing the caged Rnr1 (by Sml1) and Rnr2/4 (by Dif1 and Wtm1), which allows more RNR holo-enzyme formation. At the transcriptional level, Dun1 and Mck1 alleviate the repressor function of Crt1 through phosphorylation at different sites with different kinetics, allowing a wide range adjustment of RNR2/3/4. Meanwhile, Crt1-controlled HUG1 is also induced, which very likely prevents overproducing dNTPs under this condition. Excessive dNTPs have been demonstrated to increase mutation risk and thus impair cell growth [42, 43]. However, the higher levels of RNR activity may be required to produce enough dNTPs if cells suffer a more severe and/or persistent assault (i.e. more than 150 mM HU). Mck1 operates under this circumstance by inhibiting the induction of HUG1 in a Crt1-independent manner. Apart from the dNTP levels, the Mck1-Hug1 pathway also regulates the dNTP balance under replication stress induced by HU. Although molecular details regarding how Mck1 and Hug1 achieve these need further investigation, our findings reveal a multi-level response system to a wide range of replication threats.
It is also noteworthy to point out that the low dNTP levels are unlikely the sole reason underlying the high HU sensitivity of mck1Δdun1Δ. Therefore, it will be interesting to search for additional roles of Mck1 in maintaining genome stability other than the mechanism reported here. Intriguingly, apart from the RNR regulation function reported here, Mck1/GSK-3 has been well-established as phosphodegrons of an array of vast substrates including cell cycle proteins like Cdc6 [44], Sld2 [45], Hst3 [46], Eco1 [47] in yeasts, and Bcl3, c-JUN, Mdm2, c-Myc, Rb and PTEN in mammals [48], which are all important for cell growth and proliferation.
Although there are no apparent orthologs of Hug1 in higher eukaryotes, several other studies have provided hints that the role of Mck1/GSK-3 in the S phase checkpoint might be conserved. An unusual feature of GSK-3 is that it is generally active under the unperturbed condition and primarily regulated by inhibition in response to extracellular signals (e.g. growth factors, insulin) through signaling pathways like Akt and mTOR (Target Of Rapamycin)[49]. The TOR kinase, which belongs to the highly conserved family of phosphatidylinositol-3-kinase-related kinases (PIKKs) as Mec1ATR, is involved in DNA damage-induced expression of RNR1 and RNR3 in yeasts [50, 51]. In mammalian cells, the translation of large and small RNR subunits RRM1 and RRM2 is cap-dependent, which is regulated by phosphorylation of eukaryotic translation initiation factor 4E (EIF4E)-binding protein 1 (4E-BP1) by mTORC1 [52].
Thus, further investigation of the role of GSK-3 in the S-phase checkpoint and RNR regulation in vertebrates may help to establish crosstalk among glucose metabolism, DNA metabolism and cell proliferation. In consideration of the clinical usage of HU and pharmaceutical interest in the inhibitors of the cell cycle checkpoint proteins including Gsk-3 kinases for neoplastic and non-neoplastic disease treatments [7, 53–55], the studies based on our results reported here may have potential implications for drug design.
S. cerevisiae strains congenic with BY4741/4742 and plasmids constructed in this study are listed in S1 and S2 Tables, respectively.
The dun1Δ (MATα) single mutant was crossed with a non-essential deletion collection of cell cycle-related genes for synthetic genetic screens as previously described [56, 57]. The obtained double mutant colonies were then examined for their growth in the presence or absence of 15 mM HU.
Fivefold serial dilution of log-phase growing cells (initial OD600 = 0.4) were spotted on YPD (yeast extract/peptone/dextrose) or synthetic media plates in the presence of the indicated concentrations of HU. Plates were incubated at 30°C for 48 h before photography.
To expressed His6-Rad53 and His6-rad53-KD, the pET-15b-RAD53 (a kind gift from Dr. John Diffley) and pET-15b-rad53-KD (K227A) plasmids were introduced in BL21-codon-plus (DE3)-RIL E. coli strain (Stratagene). His6-Crt1-(201–453) or His6-Crt1-(203–453)-S299D were cloned into a pET28a plasmid. Early log phase culture was treated with 0.2 mM IPTG to induce protein expression. After 3 h of incubation at 25°C, cells were harvested. The proteins were purified using Ni2+-beads (GE Healthcare) and eluted by 250 mM imidazole. pRS-313-pADH1-MCK1-5FLAG plasmid was transformed into an mck1Δ strain. Mck1-5FLAG was purified by 20 μl anti-FLAG M2 beads (Sigma) and eluted by 150 μl of 1 μg/μl FLAG peptide. In a typical kinase assay, 50 mM Tris, pH 7.5, 150 mM NaCl, 0.1% Tween-20, 10 mM MgCl2, 5 μCi of ɣ-32P-ATP were used. Each kinase reaction contained His6-Rad53 (0–10 μg) or His6-Rad53-KD (10 μg) and Mck1-5FLAG (10 μl), or His6-Crt1-(201–453) (8 μg) in 40 μl reaction volume and incubated for 30 min at 30°C. Kinase assay was stopped by heating at 100°C for 5 min in SDS sample buffer. Samples were then subject to SDS–PAGE. Phosphorylation was detected by 32P autoradiography. The amount of protein loaded was detected by CBB staining.
For immunoblot analysis, 5 ml of culture was grown in YPD to an OD600 of 1 and harvested. The indicated culture was treated with 200 mM HU for 3h before being harvested. Yeast extracts were prepared using the trichloroacetic acid (TCA) precipitation for analysis in SDS-gels. The Crt1-13Myc, Rnr3-13Myc, Hug1-13Myc protein levels were detected with mouse anti-Myc antibody (1:1000, ORIGENE) and HRP-conjugated anti-mouse IgG as the secondary antibody (1:10000, Sigma). Tubulin as loading control was detected with anti-tubulin (1:10000, MBL) and HRP-conjugated anti-rabbit IgG as the secondary antibody (1:10000, Sigma). For detecting the hyperphosphorylation of Crt1-13Myc, the special 30% Acrylamide Solution (acrylamide: N’N’-bis-methylene-acrylamide = 149:1) was used.
Stationary phase cells (2.5 × 107) were washed and re-suspended in 50 μl of Lyticase buffer (10 mM Phosphate buffer pH 7.0, 50 mM EDTA), and then solidified in blocks with 50 μl 1% low melting temperature agarose (Sigma). These were digested with 75 U/ml lyticase in Lyticase buffer for 24 h at 37°C, then with 2 mg/ml Proteinase K (Amresco) in 100 mM EDTA, 1% sodium lauryl sarcosine for 48 h at 42°C. After four washes with TE50 (10 mM Tris, pH7.0, 50 mM EDTA), plugs were run on 1% agarose gels on in 1× TBE at 3 V/cm, 300–900 s switch time, for 68 h. PFGE was carried out in a CHEF-MAPPER system (BioRad) for 68 h at 14°C. Chromosomes were visualized with ethidium bromide before treatment with 0.25 M HCl for 20 min, water for 5 min twice. DNA was transferred to HyBond N+ in transfer buffer (0.4 M NaOH, 1 M NaCl) and UV cross-linked before hybridization with a random primed probe (Takara) overnight at 42°C and washed twice for 20 min with 0.5× SSC 0.1% SDS at 65°C.
Total RNA extraction was performed using a commercial TRIzol Reagent (CoWin Biosciences) and the manufacturer’s instructions with slight modifications. After centrifugation, cells were added to 100 μl TRIzol Reagent together with 100 μl of sterile glass beads (0.5 mm in diameter). The cells were then disrupted by vortexing for 60 s followed by cooling on ice for 60 s. This step was repeated four times. The extraction was then continued according to the manufacturer’s instructions (CoWin Biosciences). For reverse transcription-PCR (RT-PCR) analysis, reverse transcription with Oligo (dT) Primer was performed with 2 μg of total RNA, 1 mM dNTPs, 1μl RT and 0.5 μl RNasin for 60 min at 42°C, which was followed by a 15 min heat inactivation at 95°C. For each gene, real-time quantitative PCR amplification (95°C for 10 min followed by 95°C for 15 s and 60°C for 1 min for 40 cycles) was performed using SYPR-Green on a QuantStudio 6 Flex system (Life).
Logarithmically growing cells were treated with formaldehyde prior to lysis.
ChIP was carried out according to the methods used in previous studies with slight modifications. In brief, 100 ml stationary phase cells were treated with or without 200 mM HU for 1 h at 30°C. 1% formaldehyde was used for crosslinking for 20 min at room temperature. Cells were lysed and sonicated. Endogenous Crt1 proteins carrying a 13Myc tag were precipitated by an anti-Myc antibody (9E10) overnight at 4°C. The immune complexes were harvested by the addition of 50 μl of protein G dynabeads. Formaldehyde crosslinks were reversed by incubation at 65°C for 5 h, followed by protease K treatment at 42°C for 2 h. Then co-precipitated genomic DNA was purified using phenol-chloroform extraction and subjected to quantitative real-time PCR SYPR-Green on a QuantStudio 6 Flex system (Life).
dNTP extraction and quantification were carried out as described [58].
|
10.1371/journal.pntd.0006802 | Histoplasma capsulatum antigen detection tests as an essential diagnostic tool for patients with advanced HIV disease in low and middle income countries: A systematic review of diagnostic accuracy studies | Disseminated histoplasmosis, a disease that often resembles and is mistaken for tuberculosis, is a major cause of death in patients with advanced HIV disease. Histoplasma antigen detection tests are an important addition to the diagnostic arsenal for patients with advanced HIV disease and should be considered for inclusion on the World Health Organization Essential Diagnostics List.
Our objective was to systematically review the literature to evaluate the diagnostic accuracy of Histoplasma antigen tests in the context of advanced HIV disease, with a focus on low- and middle-income countries.
A systematic review of the published literature extracted data on comparator groups, type of histoplasmosis, HIV status, performance results, patient numbers, whether patients were consecutively enrolled or if the study used biobank samples. PubMed, Scopus, Lilacs and Scielo databases were searched for published articles between 1981 and 2018. There was no language restriction.
Of 1327 screened abstracts we included a total of 16 studies in humans for further analysis. Most studies included used a heterogeneousgroup of patients, often without HIV or mixing HIV and non HIV patients, with disseminated or non-disseminated forms of histoplasmosis. Six studies did not systematically use mycologically confirmed cases as a gold standard but compared antigen detection tests against another antigen detection test. Patient numbers were generally small (19–65) in individual studies and, in most (7/10), no confidence intervals were given. The post test probability of a positive or negative test were good suggesting that this non invasive diagnostic tool would be very useful for HIV care givers at the level of reference hospitals or hospitals with the infrastructure to perform ELISA tests. The first results evaluating point of care antigen detection tests using a lateral flow assay were promising with high sensitivity and specificity.
Antigen detection tests are promising tools to improve detection of and ultimately reduce the burden of histoplasmosis mortality in patients with advanced HIV disease.
| Disseminated histoplasmosis, a disease that often resembles and is mistaken for tuberculosis, is a major cause of death in patients with advanced HIV disease. Histoplasma antigen detection tests are an important addition to the diagnostic arsenal for patients with advanced HIV disease and should be considered for inclusion on the World Health Organization Essential Diagnostics List. Our objective was to systematically review the literature to evaluate the diagnostic accuracy of Histoplasma antigen tests in the context of advanced HIV disease, with a focus on low- and middle-income countries. Systematic review of the published literature extracted data on comparator groups, type of histoplasmosis, HIV status, performance results, patient numbers, whether patients were consecutively enrolled or if the study used biobank samples. At the end of the screening process we included a total of 16 studies in humans for further analysis. Most studies included used a heterogeneous group of patients, often without HIV or mixing HIV and non HIV patients, with disseminated or non-disseminated forms of histoplasmosis. Patient numbers were generally small in individual studies and, in most of these, no confidence intervals were given. When considering the diagnostic accuracy of Histoplasma antigen detection tests evaluated among consecutive HIV-infected patients with confirmed histoplasmosis, the performance of the tests was good. These non invasive diagnostic tools would be very useful for HIV care givers at the level of reference hospitals or hospitals with the infrastructure to perform ELISA tests. Antigen detection tests are promising tools to improve detection of and ultimately reduce the burden of histoplasmosis mortality in patients with advanced HIV disease.
| Histoplasmosis was first described in Panama in 1906 in a patient who appeared to have miliary tuberculosis, a confusion that is still very much present today.[1] Disseminated histoplasmosis has been an AIDS-defining infection since 1987.[2,3] Recent estimates using different methods converge on the conclusion that, in Latin America, each year, over 22000 HIV-infected patients get disseminated histoplasmosis and that between 5000 and 10000 HIV-infected persons die from it, mostly for lack of diagnosis.[4] A multicentre study in Latin America reported a 79% increase in the hazard of dying among culture negative “tuberculosis” cases, the author concluded that the patients had another diagnosis, presumably a good illustration of the confusion that is so frequent between tuberculosis and histoplasmosis.[5] In French Guiana, among consecutive HIV patients hospitalized for infectious symptoms (fever, isolated or with other symptoms), 42% of those with CD4<200 had culture-confirmed histoplasmosis, and 85% of those with CD4<50 had histoplasmosis[6]; In Fortaleza, of 378 consecutively admitted HIV patients, 164 (43%) had microscopically confirmed histoplasmosis[7]; In Panama, 7.65% of patients with an HIV infection had culture-positive H. capsulatum[8], in Guatemala, and in Colombia histoplamosis and TB are the main opportunistic infections;[9,10] In Venezuela, in patients with AIDS, histoplasmosis was documented in 29 of 66 (44%) autopsies performed[11,12]. Most histoplasmin prevalence studies took place before the AIDS epidemic was identified, because histoplasmin testing is no longer performed. They however are a good marker for the current endemicity of Histoplasma in the countries where they were performed.[13] In all the above regions, these past studies showed that approximately 30% of the population was reactive to a histoplasmin skin test, thereby illustrating the ubiquity of the fungus[14]. These few studies illustrate the importance of histoplasmosis among HIV-infected patients in Latin America, however much of the continent has no data and often no diagnostic capacity.[4,15,16] But the data gap is even worse in other parts of the world. Cases have been reported in many parts of Africa.[17] In Cameroun a prospective study found that 13% had microscopically confirmed histoplasmosis[18], preliminary unpublished results in South Africa using antigen detection showed that 10–15% of hospitalized HIV-infected patients with a CD4 count <100 had positive antigenemia. In South Asia, South East Asia, China histoplasmosis has also been increasingly reported in HIV patients[19,20]. Overall, there are enormous data gaps regarding the global burden of histoplasmosis, and the lack of simple diagnostic tools creates a vicious circle where the absence of data perpetuates the low awareness of the importance of the problem and the lack of research on the topic.[21] Diagnostic tests are thus crucial to improve public health and to improve the care of patients with advanced HIV disease who are most at risk of developing disseminated histoplasmosis. Cohort studies have shown that incidence increases rapidly as CD4 counts fall below 200 cells per mm3.[22] The isolation of the pathogen may be performed using direct examination of tissue biopsy (identification of yeast-like forms in tissue from bone marrow, lymph nodes, liver, intestine and other organs…), allowing strong suspicion of the diagnosis ultimately confirmed by culture, which often takes over a month and requires expertise and a BSL2/3 laboratory because of the hazard of inhaling the fungal microconidia.[23] Treatment should be presumptive (Amphotericin B or itraconazole depending on the severity) given the delays of fungal culture, and waiting for results to start treatment may delay care for weeks and lead to the patient’s death. [24,25] The invasive procedures, the mycological expertise required and the important delays have led to search for alternative diagnostic methods using non invasive, diagnostic tests giving rapid results to the physician. Histoplasma antigen detection tests have been used in the USA for over 30 years[26], but rarely in tropical regions[27–29]. Antigen detection tests are the simplest diagnostic method that could be implemented in low and middle income countries in order to diagnose HIV patients with advanced disease and/or severe illness including a TB-like presentation.
There have been 2 systematic reviews on antigen detection tests.[30,31] One included a meta-analysis but with some methodological issues. First, this metaanalysis included a study on antibody detection with the antigen detection studies. Secondly, the meta-analysis pooled sensitivity and specificity of different Histoplasma antigen detection tests, which is debatable. Overall, both systematic reviews concluded that antigen detection tests in general were sensitive and specific diagnostic tools. However, the publication of new studies and concerns for some study limitations and the variation they introduce led us to review the literature in the specific context of advanced HIV disease, with a special focus on low and middle income countries. The systematic review aimed at providing evidence-based arguments for the essentiality of Histoplasma antigen detection test in persons with advanced HIV in low and middle income countries.
This analysis did not deal with individual patient data but with published data, which does not require regulatory approval.
A systematic review of published articles and conference abstracts evaluating Histoplasma antigen detection was conducted. PubMed, Scopus, Lilacs and Scielo databases were searched for published articles in English, Spanish and Portuguese between 1981 and 2018. The search terms usedwere: “histoplasmosis” and “antigen detection”, or “histoplasmosis” and “antigenuria”, or “histoplasmosis” and “antigenaemia”. Articles reporting studies in humans and the diagnostic accuracy of antigen detection tests were retained. To ensure completeness, we crosschecked with 2 recent systematic reviews to determine whether we had not missed important studies[30,31]. Finally we looked at Food and Drug administration reports of the Immuno-Mycologics (IMMY) Alpha study.[32] Then, because we were evaluating rigorous studies in HIV patients with disseminated histoplasmosis, articles with culture as a gold standard comparator, and including consecutive severely ill HIV-infected patients with disseminated histoplasmosis were retained. This decision was based on our focus on patients with advanced HIV disease and our assumption that among these patients disseminated histoplasmosis is a major killer. Thus, ideally, all patients (cases and controls) should be HIV-infected and all histoplasmosis patients should be microbiologically confirmed histoplasmosis (gold standard), because from a clinician’s point of view this would reflect the diagnostic challenge when facing a patients with unspecific signs of infection and advanced HIV disease. The studies retained used different diagnostic tests therefore we decided not to synthesize their results in a meta-analysis. Supporting information files show the PRISMA checklist and the PRISMA flow diagram.
From the reported sensitivities and specificities we computed pre and posttest odds, positive and negative likelihood ratios, and posttest probabilities for different hypothesized proportions of histoplasmosis prevalence and for both positive and negative test results. The posttest odds was the multiplication of the pretest odds (hypothesized prevalence) by the positive likelihood ratio; the posttest probability was posttest odds/(1+posttest odds). Pretest and posttest odds, positive and negative likelihood ratios, and posttest probabilities were calculated for scenarios of Histoplasma prevalence of 1%, 10%, 20%, and 40% in patients with advanced HIV disease. When missing in the published article, we calculated 95% confidence intervals. We computed the point estimates but also the lowest and highest values derived from the 95% confidence interval. Data were analyzed using STATA 13 (STATA Corporation, college station, Texas).
Table 1 shows the 16 studies retained and their characteristics. Studies on Histoplasma antigen detection tests included heterogeneous groups of patients, some (n = 2) without HIV or others (n = 3) mixing HIV and non HIV patients [32–34], with disseminated or non-disseminated forms of histoplasmosis, which may introduce great variation in measuring the diagnostic accuracy of a test. In 4 studies it was not clear whether patients were HIV-infected or not (Table 1). The control groups, when present, were also different: to calculate specificity some studies used non fungal controls, while others did not, thus potentially introducing variability. Four studies did not use mycologically confirmed cases as a gold standard but compared antigen detection tests against another antigen detection test, calculating sensitivity and specificity in the absence of a gold standard[35–38], instead of looking for agreement using split samples. Patient numbers were generally small (usually ranging from 19 to 65 histoplasmosis cases, the largest compiling 158 cases) and, in most individual studies, no confidence intervals were given. Finally, many studies used stored samples and failed to include consecutive patients. The first study in AIDS patients by Wheat et al. performed a comparison with a gold standard of 61 AIDS cases with microscopically proven disseminated histoplasmosis and 30 AIDS controls (total obtained by adding controls with different opportunistic infections). The assay had a very high sensitivity (96.7% (95% CI = 88.6–99.6)) and specificity (100% (95% CI = 88.4–100)).[26,39] Receiver operator characteristic (ROC) curves or likelihood ratios were not calculated in the paper. When calculating the post-test probabilities with the lower bounds of the 95% confidence intervals for sensitivity and specificity, a worst case scenario, the posterior probability of a positive test was 5% if prevalence was 1%, 38% for a 10% prevalence, 58% for a 20% prevalence and 78% for a 40% prevalence. The calculations using 96.7% and 100% point estimates yielded a 98.9% posterior probability for prevalence at 1%. The patients, however, were apparently not included consecutively. The second generation test from MiraVista was apparently tested using consecutive AIDS patients with disseminated histoplasmosis and 100 controls without fungal infections, who did not have HIV or AIDS[40]. We thus excluded the study from our analysis. Regarding the IMMY Alpha, although the samples (in data submitted to the FDA) were tested using microscopically confirmed histoplasmosis as a reference standard[32], it is not clear what proportion of patients were HIV-positive, or what proportion were disseminated histoplasmosis. The study was thus not further analyzed.
After retaining studies which consecutively enrolled HIV-seropositive patients with a comparison against the gold standard of culture only 3 studies remained. These studies compared 2 different antigen tests (the CDC polyclonal antigen test and the Immy monoclonal antigen test) to culture, all 3 studies having taken place in 2 populations in Latin America.[27–29] The CDC test in urine had a sensitivity of 81% (95%CI = 67–91%) and a specificity of 95%(95%CI = 91–98%). The area under the ROC curve was 0.87 (95% CI, 0.80 to 0.95), and positive and negative likelihood ratios were 16.1 (7.4–45.5)) and 0.2 (0.09–0.36), respectively. Thus for areas of low prevalence, there was a low post test probability when the CDC test was positive, but the post test probability increased rapidly with prevalence (Fig 1). On the contrary in areas with high prevalence a negative CDC test was still associated with a 0.11 probability of having the disease. Overall the IMMY monoclonal test performed better than the CDC test in a similar Latin American context (Fig 2).
For the IMMY monoclonal antigen detection when using the manufacturer instructions test sensitivity was 98% (95%CI = 95–100%) and specificity was 97% (95%CI = 96–99%). The areas under the ROC curve were 0.99 for the quantitative determination and 0.97 for a semi quantitative adaptation. The results in the patients from Guatemala and from the patients in Colombia showed similar performances. The positive likelihood ratiowas 32.6 (95% CI = 19–100) and the negative likelihood ratio was 0.02 (95%CI = 0–0.05).
Finally, an abstract presented at the 2018 ISHAM conference compared a new lateral flow assay from MiraVista diagnostics used on culture confirmed HIV-associated histoplasmosis and non-HIV infected controls. The point estimates for this new point of care test were 95% sensitivity and 82% specificity.
The first antigen detection test was developed in the USA in 1986, using polyclonal antibodies against Histoplasma galactomannan.[26] It was initially a radio immune assay and was subsequently modified as an EIA.[34,40] This test has very good reported sensitivity and specificity but it is only performed in Indianapolis at MiraVista Diagnostics and it is not FDA approved. In the context of low and middle income countries this test is thus apparently not a viable option.[25] In 2007, a polyclonal FDA approved EIA was commercialized (IMMY Alpha).There were conflicting results after comparisons with the MiraVista test with a variety of methodological issues ranging from the types of patients selected in the study, their HIV status, the test procedure, and other biases. If we focus on item 18 of the recent PRISMA DTA guidelines, it is noteworthy that there was a conflict of interest (many of the authors pointing the IMMY alpha’s lack of sensitivity were linked to MiraVista a company with a large share of the United States’ market of histoplamosis diagnosis).[44] Despite these methodological issues, the early controversy has stood in the way of a widespread use of the IMMY alpha.[32,35,36] The IMMY test has been recently modified using monoclonal antibodies which have greatly improved its sensitivity.[29] Given the lack of diagnostic test for low and middle income countries the CDC’s mycotic diseases branch developed a polyclonal EIA that was evaluated in consecutive HIV patients in Colombia and Guatemala, and was successfully implemented in Brazil, Suriname, and French Guiana.[28] Although there is an agreement that antigen detection was a very sensitive and specific non-invasive test,[23] the published studies have greatly varied in design and in comparison groups. The burden of histoplasmosis in patients with advanced HIV-disease has been greatly underestimated,[4,20] in this specific context looking at consecutive HIV-infected patients with advanced disease, the antigen tests seem to have great value. The CDC test and the MiraVista EIA are not likely to be submitted for FDA approval, and the IMMY monoclonal EIA is apparently in the process of FDA approval for commercialization. There is currently 1 test with FDA approval: IMMY alpha.[32] We did not retain the analysis from the FDA data because it was not clear that all cases only had disseminated histoplasmosis or a mix of clinical forms, and if any or some cases had HIV.
Beyond diagnostic accuracy, the epidemiological context is important to keep in mind when interpreting the results of a test. We do not precisely know the burden of histoplasmosis in HIV-positive patients in many parts of the world. Until now, antigen detection tests have been used successfully for research in a number of tropical countries including, Colombia, Guatemala, Panama, Brazil, French Guiana, and Suriname. This suggests that antigen detection is appropriate for different regional clades.[27–29,42] Costing will be an important aspect since diagnosis is mostly a problem in low and middle income countries with high HIV prevalence.[25] Although it has been argued that doing the antigen test on urine and serum increased sensitivity,[45] recent studies have shown there was no significant benefit to repeat the test.[46] In resource limited countries, this would unnecessarily increase costs. There are still discussions on whether urine is more sensitive than serum, this should be further evaluated using a proper study design in consecutive patients with advanced HIV disease. The test is an EIA format and therefore requires a minimum infrastructure with an EIA reader, electricity, refrigeration, and organization with sample transport, conservation, batching EIA runs … Therefore, reference hospitals and large hospitals may be equipped to perform such a test. The cost of an EIA for low and middle income countries is not available yet. However, costs often increase dramatically with multilayered distribution networks that are necessary for a manufacturer in a high income country to reach the end users in countries where they have no presence. In this situation, each additional intermediary level will amplify costs, whatever the manufacturers’ initial cost. The fact that reference laboratories are targeted would possibly make it easier for manufacturers to streamline distribution and avoid unnecessary cost hikes. When point of care antigen detection format becomes available this would allow further scale up of diagnosis to the most remote health care facilities. The recent communication by Caceres et al. on the miravista lateral flow assay seems very promising and could radically change things if the test confirms its diagnostic performance and if it becomes available in all endemic countries at an affordable price.
The WHO has published an essential diagnostics list[47]. We believe that Histoplasma antigen detection tests should be included because this would greatly raise awareness of clinicians and public health authorities, a crucial first step to reduce unnecessary AIDS deaths for lack of a proper diagnosis.[4] While the performance measurements of the test are a first step, the most important question would be will these tests make a difference in terms of saving lives? Thus, as recommended by WHO, a PICO (Population, Intervention, Comparison, Outcome) question regarding antigen detection tests could be: Among patients with advanced HIV-disease, are Histoplasma antigen detection tests better than the present standard of diagnosis to improve the diagnosis of histoplasmosis and reduce mortality? Since the most frequent standard of diagnosis is no diagnosis for histoplasmosis, the answer to this question may seem straightforward: it is better to make a diagnosis than to miss a diagnosis. In terms of outcome, for diagnosis and mortality, data from Colombia showed that the availability of antigen testsand training dramatically increased the number of diagnoses.[48] In French Guiana, increased awareness and diagnostic progress led to very important increases in the number of diagnoses and a 4 fold reduction of case fatality at one month.[49] Increased awareness alone will have huge benefits on the number of patients diagnosed and treated, an antigen detection test may even show more of the hidden part of the Histoplasmosis “iceberg”.
In conclusion, the studies on Histoplasma antigen detection methods have suffered from great heterogeneity, partly because it is challenging to get sufficient numbers of consecutive HIV-infected culture proven disseminated histoplasmosis cases. Excellence centers in low and middle income countries seem better positioned to perform these studies. As shown by the pioneering Colombian and Guatemalan collaboration studies, the evaluation of future antigen test upgrades (lateral flow assays) should rest on well-designed studies in consecutive patients with HIV and confirmed histoplasmosis test that can best be achieved through a North-South collaboration. Furthermore, 30 years after the first test, Histoplasma antigen detection tests manufacturers should go through the regulatory approval process in order to make tests available in low and middle income countries which have been an underappreciated potential market. Meanwhile, the available antigen detection tests, should be included in the essential diagnostics list to start mapping the global burden of disseminated histoplasmosis. This would greatly accelerate the goal of having diagnostic tests and effective drugs for disseminated histoplasmosis in most hospitals in endemic countries be achievable.[50]
|
10.1371/journal.pbio.1000419 | Active RNA Polymerases: Mobile or Immobile Molecular Machines? | It is widely assumed that active RNA polymerases track along their templates to produce a transcript. We test this using chromosome conformation capture and human genes switched on rapidly and synchronously by tumour necrosis factor alpha (TNFα); one is 221 kbp SAMD4A, which a polymerase takes more than 1 h to transcribe. Ten minutes after stimulation, the SAMD4A promoter comes together with other TNFα-responsive promoters. Subsequently, these contacts are lost as new downstream ones appear; contacts are invariably between sequences being transcribed. Super-resolution microscopy confirms that nascent transcripts (detected by RNA fluorescence in situ hybridization) co-localize at relevant times. Results are consistent with an alternative view of transcription: polymerases fixed in factories reel in their respective templates, so different parts of the templates transiently lie together.
| We were all taught that an RNA polymerase becomes active by diffusing to a promoter, initiating transcription, and then tracking like a locomotive down the DNA template. We test this using tumour necrosis factor alpha (TNFα) to switch on transcription of two human genes which lie far apart on the genetic map and then measure how close the two are in 3D nuclear space. If what we were taught were true, there is no reason to expect the two genes to lie together. What we find—using two different techniques (cutting/ligating nearby sequences, and super-resolution microscopy)—is that the two genes are initially apart; then the parts of the genes being transcribed at a particular moment transiently come into close proximity. Our results are consistent with a model in which genes diffuse to a cluster of polymerases—a transcription factory—with transcripts being made as immobile polymerases reel in their respective templates. The DNA moves, not the polymerase.
| It is widely assumed that an RNA polymerase transcribes by diffusing to a promoter, binding, and then tracking down the template as it makes its transcript [1]. Accumulating evidence, however, is consistent with an alternative: a promoter diffuses to a transcription factory where it binds to a transiently immobilized polymerase, which then reels in its template as it extrudes a transcript [2]–[6]. Here, we address the question: Are transcribing enzymes mobile or immobile?
Our strategy involves switching on transcription of two genes rapidly and synchronously using tumour necrosis factor alpha (TNFα). This cytokine orchestrates the inflammatory response in human umbilical vein endothelial cells (HUVECs) by signalling through nuclear factor kappa B (NF-κB) to activate a sub-set of genes [7]–[8]. SAMD4A—a 221 kbp-long gene that encodes a regulator of this pathway—is amongst the first few to respond. Microarray analysis reveals that a synchronous wave of transcription initiates within 15 min, before sweeping down the gene (at ∼3 kbp/min) to reach the terminus ∼70 min later (Figure S1); no transcripts from the non-coding strand are detected [9]. RNA FISH using intronic probes confirms that almost half the cells in the population respond; essentially no nascent RNA can be detected prior to stimulation, no transcription occurs from the antisense strand, and probes targeting successive introns only yield signal as the wave passes by [9].
TNFAIP2—a short 11 kbp gene that lies ∼50 Mbp away from SAMD4A on chromosome 14—encodes another regulator. It is switched on as rapidly and then repeatedly transcribed over the next 90 min. We use it as an external reference point (or “anchor”) and analyze the contacts it makes with different parts of SAMD4A using chromosome conformation capture (3C)—a powerful tool for detecting proximity of two DNA sequences in 3D nuclear space [10]–[12]. If the conventional model for transcription applies, we would not expect the anchor to lie close to any part of SAMD4A either before or after adding TNFα, as it lies so far away on the chromosome (Figure 1, left). Even if polymerases on the two genes happened to lie together (for whatever reason), tracking of one down the long gene should increase the distance between transcribed parts of the two genes. But if both genes were transcribed by polymerases transiently immobilized in one factory, the short gene—which would repeatedly attach to (and detach from) the factory as it initiates (and terminates)—should always lie close to just the part of SAMD4A being transcribed at that particular moment (Figure 1, right). Thus, as the polymerase reels in SAMD4A, introns 1, 2, 3, etc. should successively be brought into the factory to lie transiently next to the anchor. Results using TNFAIP2 (and other anchors) are impossible to reconcile with the widely held assumption that polymerases track; rather they are consistent with active polymerases being immobilized in factories.
As our strategy requires one gene to be used as an anchor, we applied 3C and a variant of “associated chromosome trap” (ACT) [13]–[14] to search for genes that interacted with SAMD4A. A number were found, and we chose four that were detected in independent experiments and which were relatively short (<60 kbp): TNFAIP2, GCH1, PTRF, and SLC6A5 (Figure S2).
We initially verified that all five genes responded to TNFα by reverse-transcriptase PCR (RT-PCR). No intronic RNA (or only low levels in the case of PTRF) copied from the five genes was detected before induction, but higher levels were seen within 10 min of TNFα treatment (Figures 2A and S3F). Intronic RNA copied from further downstream in SAMD4A then appeared consistent with pioneering polymerases transcribing its 221 kbp at ∼3 kbp/min. Thus, RNA copied immediately downstream of the transcription start site (tss) appeared after 10 min, from ∼34 kbp into intron 1 after 30 min, from intron 3 after 60 min, and from the terminus after 85 min. In contrast, signal from each end of TNFAIP2 is seen by 10 min. This 11 kbp gene is so short, and synchrony sufficiently poor, that some polymerases in the population are initiating as others are terminating (Figure 2A). GCH1 and SLC6A5—both genes of ∼60 kbp—present intermediate patterns; pioneering polymerases reach termini after ∼30 min, before a second (reasonably synchronous) transcription cycle begins (Figures 2A and S3F). Such cycling has now been seen on various mammalian genes (e.g., [15]). Chromatin immunoprecipitation (ChIP) showed an enrichment of RNA polymerase II bound to the tss of all five genes within 10 min (Figures 2B and S3G). It also showed that NF-κB bound to promoters within 10 min (Figure S4), as might be expected [16]. RNA fluorescence in situ hybridization (FISH) also shows that intronic RNA copied from the relevant parts of the genes is present at the appropriate times (Figure S5). Therefore, results obtained with four independent methods (i.e., microarrays, RT-PCR, ChIP, RNA FISH) are in agreement and provide data on when polymerizing complexes are actively transcribing the sequences to be analyzed. These data are summarized in cartoons that accompany the results.
Contacts between selected regions of SAMD4A and TNFAIP2 were monitored by 3C, where the presence of a band after 34 PCR cycles reflects a high contact frequency (Figure 3). Essentially no contacts are seen between the tss of TNFAIP2 (the anchor) and regions ∼25 kbp upstream or downstream of SAMD4A (a, h) at any time, or between the anchor and any region of SAMD4A (b–g) at 0 min—when no polymerases are engaged on either gene (Figure 3B, cartoon). By 10 min (when polymerases are first found on both genes; cartoon), contacts appear between the anchor and SAMD4A regions b, c (Figure 3B). Such contacts are soon lost, as new ones appear further 3′ on SAMD4A; they seem to steadily “slide” down the long gene. Thus, by 30 min, contacts are with regions c and d, by 60 min with region e, and by 85 min with regions e, f, and g. (The presence of more than one contact at certain times is consistent with imperfect synchrony amongst the ∼106 cells assayed.) Treatment with DRB (5,6-dichloro-1-β-D-ribofuranosylbenzimidazole)—a reagent that inhibits transcription and releases polymerases from the template (Figure S6; [17]–[18])—reduces contacts (Figure 3B, grey box). Similar changing contacts were seen using (i) real-time PCR to quantify selected interactions (Figure S7), (ii) the 3′ end of TNFAIP2 as an anchor (Figure 3C, D; the gene is short enough for polymerases to be found at the same times on promoter and terminus in different cells in the population), and (iii) if HindIII replaced SacI as the restriction enzyme used for 3C (Figure S8A, B). In every case, contacts are only seen at times when active polymerases are transcribing contacting sequences. Note that several genes lying within 50 Mbp on either side of SAMD4A do not interact with it (e.g., responsive NFKBIA, SAV1, IRF9, GPR68, and PAPLN; non-responsive GMFB, YY1, HIF1A, and C14orf2; and constitutive RCOR1; Figure S9A). As a whole, these results are inconsistent with the model involving tracking polymerases (Figure 1, left) but are simply explained if the two contacting templates are transiently tethered to polymerases fixed in one factory (Figure 1, right).
PTRF is a 21 kbp gene that lies on a different chromosome (i.e., 17) from SAMD4A (on 14). The pattern of interactions between the two is much the same as those seen between SAMD4A and TNFAIP2 (Figure S3D, E), which is again consistent with the model involving fixed polymerases (Figure 1, right).
A more complex pattern of changing contacts is seen between SAMD4A and a 60 kbp gene on chromosome 11, SLC6A5 (Figure 4); this pattern suggests that polymerases must be present on both contacting sequences. Thus, as before, no contacts are seen between the tss of SLC6A5 (the anchor) and regions upstream or downstream of SAMD4A (a, h) at any time, or between the anchor and any region of SAMD4A at 0 min—when no polymerases are engaged on either gene (Figure 4B, cartoon). Again as before, contacts appear between the anchor and SAMD4A region c (which includes the tss and the beginning of intron 1) after 10 min (Figure 4B), when polymerases are first found on both. But after 30 min (when contacts with region d were seen in Figure 3B), essentially no contacts are found (Figure 4B). This is consistent with pioneering polymerases leaving the tss of the anchor so that they are now transcribing the 3′ end of this ∼60 kbp gene, as data in Figure 2 indicate. By 60 min (when a second polymerase is just initiating on the tss of SLC6A5; Figure 2), we see a strong (second) contact with the region on SAMD4A that its pioneering polymerase is now transcribing (i.e., e in Figure 4B). This interaction is DRB-sensitive (Figure 4B, grey box), and so depends on continuing transcription. No prominent interactions are seen at 85 min (Figure 4B) even though we know SAMD4A is still being transcribed. Moreover, the contact seen with region f in Figure 3B is missing, presumably because the second polymerase on SLC6A5 has left the tss used as the anchor and is now transcribing the 3′ end (Figure 2). An almost identical pattern with analogous missing contacts is seen if HindIII replaces SacI during preparation of the 3C template (Figure S8A, C).
If the above explanation is correct, with contacts only being seen if active polymerases are present on both contacting partners, then use of the 3′ end of SLC6A5 as an anchor should change the pattern as follows. The two bands seen in Figure 4B should disappear (as polymerases at the relevant times are on the tss and not the 3′ end now used as the anchor), while the two “missing” bands should reappear (as polymerases have now reached the 3′ end); they do. For example, comparison of Figure 4B and C shows that the first missing band/contact (with d at 30 min in Figure 3B) reappears in Figure 4C, as does the second (with f at 85 min). Bands/contacts are also sensitive to DRB (Figures 4B,C, grey boxes).
This interpretation is reinforced by an analysis involving 5′ and 3′ anchors on another gene (of similar length as SLC6A5) that lie on the same chromosome as SAMD4A. Thus, GCH1 is ∼0.8 Mbp away from SAMD4A and responds as rapidly to TNFα (Figure S3F, G). When its 5′ and 3′ ends are used as anchors, a complex set of changing contacts (and missing bands) is again seen (Figure S3A–C).
We also confirmed that the tss of GCH1 lay next to the tss of TNFAIP2 at 10 min but not at 0 min (Figure S9A). This is consistent with responding promoters coming together to the same factory when active. As all other contacts analyzed involve SAMD4A, these results also indicate that such reorganization is not peculiar to one long gene.
If responding regions only lie together when transcribed, their nascent transcripts should also only be together at the appropriate times. To test this we used RNA FISH with pairs of probes each able to detect an intron within a single nascent transcript copied RNA transcript at its transcription site; colocalization of nascent transcripts copied from the two different genes then yields a yellow focus [9],[19]. Yellow foci were given by the TNFAIP2 probe (red) and SAMD4A probes c, d, and e/f (green) at 10, 30, and 60 min post-induction (Figure 5A–C). No such colocalization was seen at other times (Figure S5), when relevant regions were not being transcribed. As a control, we analyzed nascent transcripts copied from a non-responsive (constitutively-active) gene—RCOR1—that lies between SAMD4A and TNFAIP2 (Figure S9A); no yellow foci were detected (Figure 5D). Just as 3C showed the templates lie together (Figure 2), RNA FISH confirms their transcripts also colocalize.
We also investigated inter-chromosomal contacts 30 min post-induction, using probes targeting (green) SAMD4A region d and (red) SLC6A5 intron 1 (close to the tss) or intron 10 (close to the 3′ end). When no 3C contacts between SAMD4A region d and the tss of SLC6A5 were seen (Figure 4B), no yellow foci were detected (Figure 5E; Figure S5C). But the “missing” 3C band was seen at 30 min using the 3′ terminus as anchor (Figure 4C), and then yellow foci are seen (Figure 5F). As a control, we analyzed nascent transcripts copied from another non-responsive (constitutively-active) gene—EDN1—that lies on a different chromosome; again, no yellow foci were seen (Figure 5G).
Electron microscopy reveals that nascent nucleoplasmic transcripts typically lie on the surface of ∼87 nm (protein-rich) factories [20]. To see if colocalizing transcripts encoded by the SAMD4A d:TNFAIP2 and SAMD4A d:SLC6A5 pairs lie this close together, we used a new approach that allows resolution beyond the diffraction limit of the light microscope [21]–[23]. We assume the red and green signals that yield a yellow focus (e.g., Figure S5B) mark two sub-diffraction spots, fit Gaussian curves to their intensities, and measure the distance (with 15 nm precision) between peaks [23]; the distance between the two transcripts ranges from 7 to 102 nm, with a mean separation of 62 nm (Figure 5H). This distribution is much like that seen when a pair of red and green points are repeatedly and randomly distributed in a 35 nm shell surrounding an 87 nm diameter sphere (Figure 5H, orange line). [Subdiffraction-sized red/green fluorescent beads of 110 nm serve as a truly co-localizing control (Figure S5B, left); then, the distance between their red and green peaks is within the uncertainty of our measurements (n = 8; not shown).] These results are consistent with nascent transcripts copied from the two different genes lying on the surface of the same transcription factory.
We tested the two models illustrated in Figure 1 to address one fundamental assumption of modern molecular biology, namely that a transcribing polymerase tracks along its template as it makes its transcript. SAMD4A has a unique set of properties that make it particularly useful for this analysis; it can be switched on rapidly and synchronously by TNFα (with approximately half the cells in the population responding), its length provides sufficient temporal and spatial resolution (it takes ∼70 min to transcribe, and contains many restriction sites that facilitate the use of 3C to discriminate between contacts produced by different parts of the gene), and neither its sense or anti-sense strands encode other transcription units that might complicate analysis. 3C reveals that just the parts of SAMD4A being transcribed at a particular moment lie close to just the parts of three other genes being transcribed at that moment (Figures 3, 4, S3, and S8). These inter-genic contacts occur infrequently, as expected [24]–[26]. RNA FISH confirmed that the relevant nascent RNAs lie together at the appropriate times (Figures 5 and S5), while “super-resolution” microscopy (allowing measurements below the diffraction limit) showed that the distance between the two transcripts is consistent with them lying within 35 nm of the surface of an 87 nm sphere (Figure 5H). Such results are difficult—if not impossible—to explain if polymerases track. Rather, they are consistent with an alternative where two responding genes diffuse to an 87 nm factory to be transcribed by immobilized enzymes. Then, as the two genes are reeled in, only parts being transcribed at a given moment will lie transiently together [5].
These results beg many questions. For example, we were able to detect interacting sequences at a reasonable frequency simply by assuming the existence of factories dedicated to transcribing genes that respond rapidly to TNFα (Figures S2 and S9). If such specialized factories exist [27],[28], how many might there be in a nucleus, and how many are accessible to a gene like SAMD4A? Fortunately, these questions will soon be answered, as techniques for analyzing all contacts made by any gene in a nucleus have been developed [29]. We also note that our results are consistent with others obtained from a recent genome-wide study; after stimulating human cells with estrogen and mapping contacts made by bound estrogen receptor-α (using ChIP, 3C, and “deep” sequencing), contacting partners were often associated with bound RNA polymerase II [30].
A detailed description of the experimental procedures is given in Text S1.
HUVECs from pooled donors (Lonza) were grown to 80%–90% confluency in Endothelial Basal Medium 2-MV with supplements (EBM; Lonza), starved (18 h) in EBM+0.5% FBS, and treated with TNFα (10 ng/ml; Peprotech) for up to 85 min. In some cases, 50 µM 5,6-dichloro-1-β-D-ribofuranosylbenzimidazole (DRB; Sigma-Aldrich) was added 20 min before harvesting cells.
3C was performed as described [10]. In brief, 107 cells were fixed (10 min; room temperature) in 1% paraformaldehyde (Electron Microscopy Sciences), “Dounce”-homogenized, and membranes lyzed (30 min; 4°C) using 0.2% Igepal (Sigma-Aldrich). Nuclei were pelleted and resuspended in the appropriate restriction buffer, incubated (16 h; 37°C) with SacI or HindIII (800 units/106 cells; New England Biolabs), diluted to 8 ml in ligation buffer, T4 DNA ligase added (4,000 units/106 cells; New England Biolabs), and incubated (48 h at 4°C, then 20 min at room temperature). After reversing cross-links (16 h; 65°C), DNA was purified by phenol extraction and ethanol precipitation, cut with BglII to reduce fragment length, and repurified. 71%–78% restriction sites in the template were cut by SacI or HindIII (determined as in [31]). PCR conditions were adjusted so that reactions were within the linear range of amplification (i.e., ∼175 ng template/reaction; 1.75 mM MgCl2, 1% dimethylsulphoxide, 10 pmoles of each primer, and GoTaq polymerase (Promega); 95°C for 2 min, then 34 cycles at 95°C for 55 s, 59°C for 45 s, and 72°C for 20 s, followed by one cycle at 72°C for 2 min); amplimers were resolved on 2.5% agarose gels, stained with SYBR Green (Invitrogen), and scanned using an FLA-5000 scanner (Fuji). Identities of all 3C products were confirmed by DNA sequencing (Geneservices, Oxford), except for those in Figure S8 (where identities were confirmed by restriction digestion). Amplification efficiencies were examined using a control template generated by SacI or HindIII digestion of BAC clones covering GAPDH on HSA12 (RP5-940J5; ImaGenes), SAMD4A, GCH1 (RP11-170J16, CTC-775N1, CTD-2586I5, CTD-2378G4; CHORI, Invitrogen), and TNFAIP2 (CTD-2594N9; Invitrogen) on HSA14, SLC6A5 on HSA11 (RP11-120F6; CHORI), and PTRF on HSA17 (RP11-194N12; CHORI) followed by ligation. This synthetic template was spiked (to reach 175 ng/µl) with HUVEC DNA cut with the relevant restriction enzyme and ligated. Other control templates included non-digested/ligated DNA and digested/non-ligated DNA (both from 106 cells). Results shown were reproduced using at least two independently obtained templates.
|
10.1371/journal.pgen.1002169 | NatF Contributes to an Evolutionary Shift in Protein N-Terminal Acetylation and Is Important for Normal Chromosome Segregation | N-terminal acetylation (N-Ac) is a highly abundant eukaryotic protein modification. Proteomics revealed a significant increase in the occurrence of N-Ac from lower to higher eukaryotes, but evidence explaining the underlying molecular mechanism(s) is currently lacking. We first analysed protein N-termini and their acetylation degrees, suggesting that evolution of substrates is not a major cause for the evolutionary shift in N-Ac. Further, we investigated the presence of putative N-terminal acetyltransferases (NATs) in higher eukaryotes. The purified recombinant human and Drosophila homologues of a novel NAT candidate was subjected to in vitro peptide library acetylation assays. This provided evidence for its NAT activity targeting Met-Lys- and other Met-starting protein N-termini, and the enzyme was termed Naa60p and its activity NatF. Its in vivo activity was investigated by ectopically expressing human Naa60p in yeast followed by N-terminal COFRADIC analyses. hNaa60p acetylated distinct Met-starting yeast protein N-termini and increased general acetylation levels, thereby altering yeast in vivo acetylation patterns towards those of higher eukaryotes. Further, its activity in human cells was verified by overexpression and knockdown of hNAA60 followed by N-terminal COFRADIC. NatF's cellular impact was demonstrated in Drosophila cells where NAA60 knockdown induced chromosomal segregation defects. In summary, our study revealed a novel major protein modifier contributing to the evolution of N-Ac, redundancy among NATs, and an essential regulator of normal chromosome segregation. With the characterization of NatF, the co-translational N-Ac machinery appears complete since all the major substrate groups in eukaryotes are accounted for.
| Small chemical groups are commonly attached to proteins in order to control their activity, localization, and stability. An abundant protein modification is N-terminal acetylation, in which an N-terminal acetyltransferase (NAT) catalyzes the transfer of an acetyl group to the very N-terminal amino acid of the protein. When going from lower to higher eukaryotes there is a significant increase in the occurrence of N-terminal acetylation. We demonstrate here that this is partly because higher eukaryotes uniquely express NatF, an enzyme capable of acetylating a large group of protein N-termini including those previously found to display an increased N-acetylation potential in higher eukaryotes. Thus, the current study has possibly identified the last major component of the eukaryotic machinery responsible for co-translational N-acetylation of proteins. All eukaryotic proteins start with methionine, which is co-translationally cleaved when the second amino acid is small. Thereafter, NatA may acetylate these newly exposed N-termini. Interestingly, NatF also has the potential to act on these types of N-termini where the methionine was not cleaved. At the cellular level, we further found that NatF is essential for normal chromosome segregation during cell division.
| N-terminal acetylation (N-Ac) is a common modification of proteins, but its general role has remained rather enigmatic. For specific proteins, N-Ac is recognized as an important regulator of function and localization [1]–[4]. Recently, it was suggested that it may act as a general destabilization signal for some yeast proteins, [5] while other reports imply that it might serve as a stabilizer, for instance by blocking N-terminal ubiquitination mediated degradation [6]. N-Ac in eukaryotes mainly occurs co-translationally when 25–50 amino acids protrude from the ribosome, by the action of ribosome associated N-terminal acetyltransferases (NATs) [7]–[12]. N-Ac may occur on the initiator Met (iMet) or on the first residue after iMet excision by methionine aminopeptidases (MAPs) [13], [14]. Three major NAT complexes conserved from yeast to humans are thought to be responsible for the majority of N-terminal acetylation events: NatA, NatB and NatC [15]. Each complex is composed of specific catalytic and auxiliary subunits. NatA, the first NAT defined by Sternglanz and co-workers [16], potentially acetylates Ser-, Ala-, Thr-, Val-, Gly-, and Cys- N-termini after iMet-cleavage [17]–[19]. NatB and NatC potentially acetylate Met- N-termini when the second residue is either acidic or hydrophobic respectively [19]–[21]. In yeast, NatD was described to acetylate the Ser- N-termini of histones 2A and 4 in vitro and in vivo [22], while no such activity has yet been presented for higher eukaryotes. NatE is another NAT of which the in vitro activity was described for the human hNaa50p towards some Met-Leu- N-termini [23], but direct evidence of in vivo activity is still lacking. Thus, each hitherto in vivo characterized NAT appears to acetylate a distinct subset of substrates defined by the very first N-terminal amino acids. Phenotypes induced by loss or reduction of NATs suggest that these enzymes, and thus probably N-Ac, are implicated in a number of cellular processes. In higher eukaryotes, depletion of NatA, NatB or NatC is associated with cell cycle arrest or apoptosis [20], [21], [24]–[28] while sister chromatid cohesion defects are observed upon NatE depletion [29]–[31].
N-Ac occurs on more than 50% and 80% of cytosolic yeast and human proteins, respectively [18]. The reason for the major difference in occurrence of N-Ac between yeast and humans to date is not known. Furthermore, the fact that specific subsets of protein N-termini, like those initiated by Met-Lys-, are often acetylated in humans and fruit fly while rarely being acetylated in yeast, is also an unsolved issue [18], [32]. Further, such substrates do not match the predicted substrate specificity of any of the known NATs. Potential explanations for this evolutionary shift from lower to higher eukaryotes include: i) evolution towards more acetylation-prone N-termini in higher eukaryotes, ii) a shift in the substrate specificity between species-specific NATs, iii) the presence of novel, yet uncharacterized NATs in higher eukaryotes, and iv) the presence of species-specific co-factors or chaperones such as HYPK [33]. However, so far, no evidence for any of these hypotheses was presented.
In the current investigation, we sought to elucidate the mechanistic explanations for the evolutionary shift in N-terminal acetylation from lower to higher eukaryotes. To this end we investigated the potential evolution of acetylation prone N-termini, but found this to be a trivial contributing factor. We further explored the presence of novel NATs in higher eukaryotes as a possible explanation. In silico analysis revealed the existence of an uncharacterized human protein with a significant sequence similarity to known catalytic NAT subunits. Indeed, multiple lines of in vitro and in vivo evidence clearly demonstrate that this candidate protein conserved among animals is a major NAT displaying distinct substrate specificity, denoted Naa60p (NatF). Our data collectively suggest that Naa60p contributes to the increased occurrence of N-terminal acetylation in higher versus lower eukaryotes, and additionally revealed a novel regulator of chromosome segregation.
We first investigated whether an evolution towards more acetylation-prone N-termini in higher eukaryotes could help explain the higher acetylation levels observed. Upon comparing the yeast, fruit fly and human proteomes, it is evident that the general distribution of N-termini is largely unaltered between the different classes, ‘NatA’, ‘NatB’, ‘NatC’ and ‘Other’ (Figure 1A). However, when considering all different subgroups based on the first two N-terminal amino acids, some significant alterations (p<0.01) appeared. Besides the general difference of the amino acid usage in yeast versus human N-termini in agreement with recent observations [34], the occurrence of (Met-)Ala- N-termini increased from 8% in yeast to 23% in humans, while Met-Glu- N-termini increased from 5% to 10%. On the other hand, (Met-)Ser- N-termini have decreased in occurrence from 23% in yeast to 11% in humans (Figure 1B). Interestingly, for these major trends, the occurrences in fruit-fly are intermediate between yeast and humans, indicating that these might be characteristic of the evolution to multicellular and more complex organisms. The next question is thus whether these changes in N-terminal sequences are causing a shift in N-Ac. In the current work, we performed COFRADIC-based N-terminal acetylation analyses of yeast and HeLa proteomes and present datasets covering 868 and 1,497 unique yeast and human N-termini, respectively (Tables S1 and S2). An overview of the occurrence of N-Ac of the different classes of assigned N-termini in the yeast (n = 648) and human (n = 1345) control samples is presented in Table 1. When relating the occurrence of N-Ac in yeast to the distribution of human N-termini and vice versa (based on the first two amino acids of the identified N-termini), we found no overall significant changes in N-Ac levels (Table S3). Thus, alteration in usage of the first two N-terminal amino acids, which are the major determinants for N-Ac, is not a significant cause for the observed shift from lower to higher eukaryotes.
Since it was shown that amino acid usage at protein N-termini differs significantly from what is expected [34], and differences in dipeptide composition have been used to predict protein expression levels [35], thermostability [36] and subcellular localization [37], we further characterized the residue contact order at protein N-terminal parts by studying dipeptide frequencies in the theoretical proteomes of Homo sapiens, Drosophila melanogaster and Saccharomyces cerevisiae (UniProt/SwissProt entries (version 2011-05)). Therefore, the occurrence of the 400 possible dipeptides from the 20 amino acids in all proteins was estimated for randomly selected human dipeptides and N-terminal (amino acids 2–11) dipeptides by Monte-Carlo sampling. Further, a z-score was applied to correct for differences in database size. Contacting residues in a random, non-N-terminal set correlate well with the expected theoretical contact order (data not shown). In sharp contrast, the overall dipeptide composition deviates significantly for database-annotated N-termini. A heatmap visualization centered and scaled by species mean and standard deviation for Homo sapiens, Drosophila melanogaster and Saccharomyces cerevisiae is shown for the ten dipeptides with the highest and lowest z-scores (union of n = 49) (Figure 1C). Overall, these data strengthen the observation that N-terminal sequences not only display altered patterns of amino acid frequencies but deviate extensively in their residue contact order in a species-specific manner, which might additionally impose yet undetected constraints in determining N-Ac.
When considering each type of N-terminus, it is evident that several of these are more acetylated in humans while some are mainly unchanged, but none are less N-Ac. The major groups of protein N-termini with an increase in N-Ac in humans as compared to yeast include (Met-)Ala-, (Met-)Val-, and Met-Lys- N-termini and thus represent major contributors to the overall evolutionary shift (Table 1 and Table S3).
Another potential cause for the evolution towards the higher level of N-Ac is a shift in the substrate specificity between species-specific NATs. For NatA, which is responsible for N-Ac of two of the important N-terminal types mentioned above, (Met-)Ala- and (Met-)Val- this seemed not to be the case as both human NatA and yeast NatA acetylated the very same subset of N-termini in yeast [18]. For the final group, Met-Lys- N-termini, no information is available since such N-termini have not been linked to any of the NAT classes previously characterized.
In search of novel human NATs, we used the sequences of known human NATs in NCBI BLAST queries (search set: Swiss-Prot database restricted to human proteins). We identified one protein with a significant similarity to several of the known NATs, namely NAT15/Q9H7X0/Naa60p (Figure 2). NAT15/Naa60p is highly conserved among animals (Figure 2B) and homologues are also potentially present in plants (for instance At5g16800). In order to assess whether NAT15 was an N-terminal acetyltransferase, the NAT15 ORF was recombinantly expressed and purified from Escherichia coli and applied to a newly developed in vitro proteome-derived peptide library N-terminal acetylation assay [38]. In brief, natural proteomes are used to generate Nα-free peptide substrate pools (libraries) by enrichment with strong cation exchange (SCX). When such a peptide library is incubated with a NAT enzyme, the newly Nα-acetylated peptides are enriched by a second SCX fractionation step, resulting in a positive selection of NAT-specific peptide substrates. Subsequently, the NAT-oligopeptide substrates are identified by LC-MS/MS, and the in vitro substrate specificity profile of the NAT in question is analyzed using IceLogo [39], an analytical tool that uses probability theory to visualize significant conserved sequence patterns in multiple peptide sequence alignments by comparing against a chosen background (reference) sequence set. Using this proteome-derived peptide assay, NAT15 Nα-acetylated numerous peptides in vitro and displayed a distinct substrate specificity profile (Figure 3A). Thus, according to the revised NAT-nomenclature system [15], we named this protein Naa60p and its activity NatF. Remarkably, the preferred N-termini included Met-Lys-, Met-Ala-, Met-Val-, and Met-Met-, categories for which there are currently no known N-terminal acetyltransferase(s). Of particular interest, recent data revealed that several Met-Lys- N-termini were acetylated in humans and fruit fly while no such N-Ac events of Met-Lys- N-termini were found in yeast, pointing to the presence of (a) NAT(s) specific for higher eukaryotes or an altered specificity profile of (a) higher eukaryotic NAT(s) as compared to yeast NAT(s) [18], [32]. To expand these observations to higher eukaryotes in general, we purified the predicted fruit fly homologue dNaa60p (CG18177) and confirmed this protein to be a NAT with a nearly indistinguishable specificity profile as compared to hNaa60p (Figure 3B). As deduced from the in vitro specificity profile, besides Met-; Leu- was also preferred at the first position, which, as we described previously [38], is expected since both Met and Leu share similar physiochemical characteristics [40], [41]. However, for co-translational Nα-acetylation, Leu at the first position appears physiologically irrelevant as it is not expected as the first amino acid, since when it follows the initiator methionine, its size precludes the removal of this initiator methionine by MAPs [14]. When only including Met residues at the first position, the specificity profile remains largely unchanged (Figure 3C, 3D and Figure S1). Given its in vitro specificity, we considered Naa60p a qualified candidate for the Met-Lys- acetylation activities observed in higher eukaryotes.
In order to assess whether hNaa60p represents a NAT in vivo and to address its potential role in the evolutionary N-Ac shift, we generated a yeast strain expressing hNaa60p. We were not able to observe any differences in growth rates or plating efficiencies between yeast control strains and yeast strains expressing hNaa60p (data not shown). Since yeast does not have an obvious homolog of hNaa60p, ectopic expression was expected to reveal whether hNaa60p endows yeast with a greater acetylating potential. Indeed, when comparing N-terminal acetylation in the proteome of control yeast (yeast control) to the yeast expressing hNaa60p (yeast+NatF), significant alterations in the Nα-acetylome were observed (Figure 4). For example the Smr domain-containing protein YPL199C and uncharacterized protein YGR130C, with respectively Met-Lys- and Met-Leu- N-termini, were unacetylated in control yeast while 82% and 48% acetylated in the strain expressing hNaa60p/NatF (Figure 5). In total, for 464 of the 544 (or 85%) unique N-termini identified in both proteomes, the N-acetylation status could univocally be determined. Of these, 72 N-termini were more acetylated in the hNaa60p expressing strain, while none were less acetylated, indicating that at least 16% of the identified yeast proteome was acetylated by hNaa60p (Figure 4 and Table S4). 44 of the 72 hNaa60p acetylated N-termini were completely unacetylated in control yeast, while 28 were partially acetylated. For the latter group, hNaa60p increased the degree of acetylation with at least 10%. It should be noted that this may represent an underestimation of hNaa60p's capacity since fully acetylated N-termini (53%) in the control strain may also represent targets, which would be masked by redundancy with the yeast NAT-machinery. The hNaa60p yeast substrates identified in vivo were in agreement with the in vitro determined substrate specificities. The most common in vivo substrate classes were Met-Lys- (n = 14), Met-Ser- (n = 9), Met-Val- (n = 8), Met-Leu- (n = 8), Met-Gln- (n = 6), Met-Ile- (n = 5), Met-Tyr- (n = 5), and Met-Thr- (n = 5) (Table 2).
Among those acetylated by hNaa60p were proteins with Met-Lys- starting N-termini, which are of particular interest because these are acetylated in humans by an unknown NAT, while only rarely acetylated in yeast [18]. When considering the yeast control dataset, only 13% of the Met-Lys- N-termini are fully or partially acetylated, while the corresponding number for the yeast+NatF strain increases to 48%. In striking resemblance, 40% to 70% of Met-Lys- N-termini are N-terminally acetylated in human cell lines as respectively demonstrated previously [18] and in the current dataset (Table 1). Met-Leu-, Met-Ile-, and Met-Phe- starting N-termini, a class of N-termini considered NatC substrates, are other types of N-termini frequently found to be acetylated by hNaa60p. Finally, many substrate N-termini without a proper NAT-classification (including initiator Met-retaining N-termini of which the iMet is only partially removed) were acetylated: Met-Ser-, Met-Val-, Met-Thr- and Met-Met-, and Met-Gln-. Thus, hNaa60p acetylates both N-free besides partially acetylated protein N-termini in yeast, some without any known corresponding yeast NAT, as well as N-termini for which there is a putative NAT (NatC). This indicates that Naa60p may mediate a significant part of the shift in N-terminal acetylation from lower to higher eukaryotes. Furthermore, in contrast to the current opinion, this also strongly suggests redundancy in the Nα-acetylation system, meaning that different NATs may have (partially) overlapping substrates. The effect of hNaa60p on overall N-terminal acetylation in yeast is shown in Figure 5C. Overall, the expression of hNaa60p increased the fraction of Nα-acetylated yeast proteins from 68% to 78%, in particular affecting the groups ‘yNatC’ and ‘Other’ (Figure 4 and Figure 5).
Overexpression or knockdown of hNAA60 in HeLa cells was found to increase or decrease, respectively, the N-terminal acetylation of proteins matching the above defined in vitro and in vivo substrate specificity of hNaa60p (Table S5). Examples include the proteins STIP1 homology and U-box containing protein1 (1MKGKEEKEGGAR12) and mediator of RNA polymerase II transcription subunit 25 (1MVPGSEGPAR10) where the Nα-acetylation status is shifted as a consequence of hNAA60 overexpression (from 18% to 32% acetylation) or knockdown (from 26% to 17% acetylation), respectively (Figure 6). These data strongly point to the fact that hNaa60p in human cells can act on the classes of N-termini deduced from the in vitro and in vivo yeast analyses described above (Table 2). Obviously, overexpression analysis will be limited by the redundancy among NATs and by the fact that naturally hNaa60p-acetylated N-termini may be fully acetylated and as such do not appear as substrates for the overexpressed hNaa60p. Furthermore and in line with previous knockdown analyses of NatA in HeLa cells, the semi-effective nature of siRNA-mediated knockdown as well as the long time period needed for a clear effect on N-terminal acetylations to occur, make such analyses indicative rather that providing the full picture of acetylation events mediated via a specific NAT and as shown previously, primarily affects the least efficiently acetylated N-termini [18]. Thus, the real number of Naa60p substrates in human cells is likely to be significantly higher as compared to the substrates identified in these particular analyses.
Finally, two of the acetylated N-termini of the predicted NatF class picked up from the HeLa dataset (Table S2) were tested by a direct in vitro approach. Synthetic peptides derived from the Met-Lys- and Met-Ala- N-termini of Septin 9 and Protein phosphatase 6, respectively, were subjected to an in vitro acetylation assay with purified hNaa60p followed by an HPLC-based analysis of acetylated and unacetylated peptides. In agreement with the human and yeast in vivo data and in vitro substrate profiles obtained above, hNaa60p acetylated both these peptides, as well as representatives of NatC and NatE class substrates (Figure 7). Thus, we confirmed the N-terminal acetylation of human substrates as well as the potential redundancy with NatC and NatE enzyme classes (Table 2).
In order to assess the cellular function of dNaa60p, its expression was knocked down in Drosophila Dmel2 cells by RNAi. Similarly to dNAA50-depleted cells [30], [31] (data not shown), dNAA60-depleted cells showed chromosomal segregation defects during anaphase (Figure 8A–C, 8F, 8G, 8J, 8K). However, while dNAA50-depleted cells exhibit abnormal metaphases with an obvious mitotic arrest, control and dNAA60-depleted cells exhibited normal metaphases, with all chromosomes perfectly aligned within the spindle equator and without any mitotic arrest (Figure 8D, 8E, 8H, 8I and Figure S2). In contrast, during anaphase we consistently observed chromosome segregation defects in dNAA60-depleted cells, which included lagging chromosomes (Figure 8K, highlighted by asterisk) and chromosomal bridges (Figure 8B, 8G, highlighted by asterisk; quantification of abnormal anaphases is shown in Figure 8C). Chromosome lagging and bridging in dNAA60-depleted cells may be explained by kinetochore abnormalities; however we failed to detect any obvious defect in the localization of the Centromere identifier protein (Cid) during metaphase or anaphase (Figure 8D–G). We also failed to detect any obvious cohesion defect since the distance between kinetochores during metaphase was normal according to Cid localization (Figure 8D, 8E). Chromosome lagging could also be explained by centrosome/mitotic spindle defects. Yet, we did not detect any obvious defect in the localization of Centrosomin (Cnn), and the mitotic spindle was bipolar and correctly attached to chromosomes and centrosomes (Figure 8D–G). Furthermore, dNAA60-depleted cells showed no obvious defects in the actin and microtubule cytoskeleton in both mitotic and interphase cells (Figure 8H–M). Since dNAA60-depleted cells were otherwise normal, our data suggest that dNaa60p is required for chromosome segregation during anaphase. Naa60p-dependent N-terminal acetylation of one or more substrates is therefore likely to be required for chromosome segregation in vivo.
The basic co-translational machinery performing N-Ac in eukaryotes was believed to be fully identified and mostly characterized, with five NATs, NatA-NatE, each of which composed of specific subunits and acetylating its own subset of substrates [15]. However, the significant shift in occurrence of N-Ac from lower to higher eukaryotes, clearly points to the fact that species-specific factors are major determinants for N-Ac. Indeed, in the current study we revealed that higher eukaryotes express NatF/Naa60p, a unique NAT responsible for N-Ac of a large subset of eukaryotic proteins. These N-termini include Met-Lys-, Met-Met-, Met-Val- and Met-Ser- to which so far no NAT has been assigned. Also N-termini like Met-Leu- and Met-Ile-, previously believed to be solely NatC substrates, may be acetylated by NatF. Thus, the previous clear-cut classification between Nat substrate classes based on the N-terminal sequences should be re-evaluated when in vivo datasets are considered. The current knowledge on the NATs of higher eukaryotes and their corresponding substrates is presented in Figure 9.
In contrast to the N-termini acetylated by NatF, for the increased N-Ac of the processed (Met-)Ala- and (Met-)Val- N-termini there is presently no explanation. The intrinsic enzymatic properties of human and yeast NatA appeared to be very similar when expressed in yeast [18]. Co-determining factors that should be elaborated upon concerning the NatA substrates are interaction partners specific for NatA of higher eukaryotes, like HYPK which was demonstrated to modulate N-terminal acetylation [33]. Notwithstanding the generally lower expression levels, the existence of higher eukaryotic paralogues of Naa15p and Naa10p, being Naa16p and Naa11p respectively [42], [43], might additionally account for modulators of the observed Nα-acetylome. However, information regarding their potential proteome-wide contribution to N-Ac is currently lacking.
We found that evolution of N-Ac prone N-termini most likely contributes only to a very small degree to the overall evolutionary shift in the occurrence of N-Ac. Furthermore, there might be a shift in the substrate specificity between species-specific NATs, for instance for the NatB, NatC and NatE activities, requiring further experimental validation. However viewing their strict evolutionary conservation, this may be rather unlikely.
The current data are more comprehensive as compared to previous analyses [18], and overall the 648 unique yeast and 1345 unique human N-termini identified were analysed for their acetylation status (Table 1, Tables S1 and S2). 68% of the yeast N-termini and 85% of the human N-termini are partially or fully N-terminally acetylated. Previously, we determined that 57% of yeast proteins and 84% of human proteins were N-terminally acetylated, thus implicating some shift in the N-Ac of the yeast N-termini between experiments. We believe the current dataset likely holds a more representative picture since more N-termini were sampled and since yeast was grown under slightly different deprivating (SILAC) conditions in the previous setup. Nevertheless, still a significant difference between yeast (68%) and humans (85%) can be observed and as demonstrated, this difference is significantly diminished in yeast expressing NatF (78%) (Figure 4 and Figure 5).
The current study provides to the best of our knowledge, the first evidence shedding light on the molecular basis of the evolutionary shift in the Nα-acetylome from lower to higher eukaryotes. With the presence of NatF, higher eukaryotes are enforced in their capacity to acetylate Met-Lys-, Met-Leu- and other Met- starting N-termini, thus explaining in part the increased occurrence of N-Ac. This additional NAT may have evolved to meet the increased demands of more complex proteomes with a higher level of regulation. In light of the recent suggestion that N-Ac generates degrons and thus acts as a destabilizer [5], these issues will be of particular importance. Our results suggested that dNaa60 activity is likely to be specifically required for chromosome segregation during anaphase, as cells depleted for dNaa60 showed normal alignment of chromosomes during metaphase plates and progressed normally through mitosis, without any obvious cell cycle arrest (Figure 8 and Figure S2). With an increasing support for N-Ac in controlling protein stability, function and subcellular localization, it is very likely that Naa60p will emerge as a key regulator for several proteins. Future investigations will aim at elucidating these specific Naa60p substrates.
The random dipeptide frequencies (n = 400) were estimated by Monte Carlo sampling of one randomly selected decapeptide per protein in the databases of; Homo sapiens, Drosophila melanogaster and Saccharomyces cerevisiae (UniProt/SwissProt entries (version 2011-05)). After 100 sampling rounds, the mean and standard deviation for each dipeptide were estimated. Thereafter, the N-terminal dipeptide frequency of all decapeptides from position 2 to 11 were calculated, and the obtained frequencies compared with the random frequencies. The corresponding species-specific z-score, reflecting the amino acid dipetide frequency differences between the protein N-terminal and overall protein sequence were calculated as follows:
Sequences of the known human catalytic NAT units/subunits, hNaa10p (P41227), hNaa11p (Q9BSU3), hNaa20p (P61599), hNaa30p (Q147X3) and hNaa50p (Q9GZZ1), were used in the search of novel human NATs by making use of NCBI BLAST (blastp) queries (search set: ‘Swiss-Prot protein sequences’ restricted to organism: ‘Homo sapiens’ and otherwise the predefined parameters). Besides the known human NATs, there was in particular one significant hit, the uncharacterized NAT15 (Q9H7X0), which held sequence similarity to all query NATs with E-values between 3×10−6 and 0.24. NAT15 is an automatically annotated name due to the presence of a N-acetyltransferase domain (pfam00583) in the protein sequence. When using hNaa30p and hNaa50p as query sequences, NAT15 scored even better than some of the known human NATs (hNaa20p and hNaa10p/hNaa11p/hNaa20p, respectively). When using hNaa30p as query sequence, some other human proteins scored equally well as NAT15: NAT8 (Q9UHE5), NAT8B (Q9UHF3), NAT8L (Q8N9F0) and ATAC2/CRP2BP (Q9H8E8) with E-values ranging from 7×10−5 to 0.034. However, all these candidates were biochemically characterized as members of the GNAT family (pfam00583) with functions distinct from protein N-terminal acetylation. NAT8 is a cysteinyl-S-conjugate N-acetyltransferase catalyzing the last step of mercapturic acid formation while NAT8B is a likely pseudogene of NAT8 [44]. NAT8L catalyses the synthesis of N-acetylaspartate [45] and ATAC2 catalyses lysine acetylation on histone H4 [46]. Thus, NAT15 was the only uncharacterized protein with a significant similarity to the known human NATs (Figure 2) and was therefore further pursued.
Plasmid encoding V5-tagged NAT15/hNaa60p (Gene ID: 79903) used for mammalian expression was constructed from human HEK293 cDNA by use of Transcriptor Reverse Transcriptase (Roche). The PCR product containing the CDS plus four 5′ nucleotides (gaga) was inserted into the TOPO TA vector pcDNA 3.1/V5-His TOPO Invitrogen) according to the instruction manual. An E. coli expression vector encoding MBP-His-tagged hNaa60p was constructed by subcloning hNAA60 from phNAA60-V5 to the pETM-41 vector using the Acc65I and NcoI sites. pETM-41-dNAA60, encoding the predicted fruit fly Naa60p was made by subcloning the CG18177 CDS from pOT2-CG18177 (clone LD27619 from the Drosophila Genomics Resource Centre, Indiana University) to pETM-41. pETM-41 was generously provided by G. Stier, EMBL, Heidelberg. A yeast expression vector, pBEVY-U-hNAA60 encoding untagged hNaa60p was constructed by subcloning hNAA60 from phNAA60-V5 to the pBEVY-U vector [47] using the BamHI and SalI sites.
The plasmid pETM-41-hNAA60 or pETM-41-dNAA60 was transformed into E. coli BL21 Star™ (DE3) cells (Invitrogen) by heat shock. A 200 ml cell culture was grown in LB (Luria Bertani) medium to an OD600 nm of 0.6 at 37°C and subsequently transferred to 20°C. After 30 min of incubation, protein expression was induced by IPTG (1 mM). After 17 h of incubation, the cultures were harvested by centrifugation and the pellets stored at −20°C. E. coli pellets containing recombinant proteins were thawed at 4°C and the cells lysed by sonication in lysis buffer (1 mM DTT, 50 mM Tris-HCl (pH 7.5 or 8.3 for MBP-dNAA60p and MBP-hNAA60p, respectively), 300 mM NaCl, 1 tablet EDTA-free protease Inhibitor cocktail per 50 ml (Roche)). Following centrifugation, the cell extracts were applied on a metal affinity FPLC column (HisTrap HP, GE Healthcare, Uppsala, Sweden). MBP-hNaa60p and MBP-dNaa60p were eluted with 300 mM Imidazole in 50 mM Tris (pH 7.5 or 8.3 for MBP-dNAA60 or MBP-hNAA60, respectively), 300 mM NaCl and 1 mM DTT. Fractions containing recombinant protein were pooled and further purified using size exclusion chromatography (Superdex™ 75, GE Healthcare) until apparent purity as analysed by Coomassie stained SDS-PAGE gels. The protein concentrations were determined by OD280 nm measurements.
Preparation of proteome derived peptide libraries. Proteome-derived peptide libraries were generated from human K-562 cells. Cells were repeatedly (3×) washed in D-PBS and then re-suspended at 7×106 cells per ml in lysis buffer (50 mM sodium phosphate buffer pH 7.5, 100 mM NaCl, 1% CHAPS and 0.5 mM EDTA) in the presence of protease inhibitors (Complete protease inhibitor cocktail tablet (Roche Diagnostics, Mannheim, Germany)). After lysis for 10 min on ice, the lysate was cleared by centrifugation for 10 min at 16,000× g and solid guanidinum hydrochloride was added to the supernatant to a final concentration of 4 M. The protein samples were reduced and S-alkylated, followed by tri-deuteroacetylation of primary amines and digestion with trypsin as described previously [48], [49]. The resulting peptide mixtures were vacuum dried. The dried peptides were re-dissolved in 500 µl 50% acetonitrile. The sample was acidified to pH 3.0 using a stock solution of 1% TFA in 50% acetonitrile and further diluted with 10 mM sodium phosphate in 50% acetonitrile to a final volume of 1 ml. This peptide mixture was then loaded onto an AccuBONDII SCX SPE cartridge (Agilent Technologies, Waldbronn, Germany) and SCX separation (SCX fractionation 1) of Nα-blocked N-terminal peptides (and C-terminal peptides) from Nα-free peptides was performed as described previously [48], [50]. The flow-through containing the Nα-blocked N-terminal peptides and C-terminal peptides was discarded and the SCX-bound fraction (containing the Nα-free peptides) was collected by elution with 4 ml of 400 mM NaCl and 10 mM sodium phosphate in 40% of acetonitrile (pH 3.0). Eluted peptides were vacuum dried and re-dissolved in 1 ml of HPLC solvent A (10 mM ammonium acetate in 2% acetonitrile, pH 5.5). C18 solid-phase extraction (SPE desalting step) of the Nα-free peptides was performed by loading the peptide mixture onto a AccuBONDII ODS-C18 SPE cartridge (1 ml tube, 100 mg, Agilent Technologies). This cartridge has a binding capacity of 1 mg of peptides and thus for each mg of material, a separate cartridge was used. Prior to sample loading, the cartridges were washed with 2 ml of 50% acetonitrile and then washed with 5 ml of HPLC solvent A. Sample loading was followed by washing the C18 cartridge with 5 ml of 2% acetonitrile. Peptides were eluted with 3 ml of 70% acetonitrile and subsequently vacuum dried.
In vitro peptide library-based NAT assay. 100 nmol of the desalted Nα-free peptide pool was reconstituted in acetylation buffer (50 mM Tris-HCl (pH 8.5), 1 mM DTT, 800 µM EDTA, 10% glycerol) together with equimolar amounts of a stable isotope encoded variant of acetyl-CoA, 13C2-acetyl CoA, (99% 13C2-acetyl CoA, ISOTEC-Sigma (lithium salt)) and 1 nmol of enzyme (i.e. recombinant hNaa60p or dNaa60p) was added to a final reaction volume of 1 ml. The reaction was allowed to proceed for 1 h at 37°C and stopped by addition of acetic acid to a 5% final concentration. SPE was then performed as described above.
NAT oligopeptide-substrate recovery and RP-HPLC based separation. Peptides starting with pyroglutamate were unblocked prior to the second SCX fractionation step. Here, 25 µl of pGAPase (25 U/ml) (TAGZyme kit, Qiagen, Hilden, Germany) was activated for 10 min at 37°C by addition of 1 µl of 50 mM EDTA (pH 8.0), 1 µl of 800 mM NaCl, and 11 µl of freshly prepared 50 mM cysteamine-HCl. 25 µl of Qcyclase (50 U/ml, TAGZyme) was then added to the pGAPase solution. The dried peptides were re-dissolved in 212 µl of buffer containing 16 mM NaCl, 0.5 mM EDTA, 3 mM cysteamine, and 50 µM aprotinin. The activated pGAPase and Q-cyclase mixture was added to the peptide sample and the mixture (275 µl total volume) was incubated for 60 min at 37°C. 275 µl acetonitrile was then added and the sample was acidified to pH 3 using a 1% TFA stock solution in 50% acetonitrile. The sample was further diluted with 10 mM sodium phosphate in 50% acetonitrile to a final volume of 1 ml. SCX enrichment of Nα-blocked N-terminal peptides was performed as described [48] (SCX fractionation 2). The SCX fraction containing the newly blocked N-terminal peptides was vacuum dried and re-dissolved in 100 µl of HPLC solvent A. To prevent oxidation of methionine between the primary and secondary RP-HPLC separations (and thus unwanted segregation of methionyl peptides [51], methionines were uniformly oxidized to sulfoxides prior to the primary RP-HPLC run by adding 2 µl of 30% (w/v) H2O2 (final concentration of 0.06%) for 30 min at 30°C. This peptide mixture was injected onto a RP-column (Zorbax 300SB-C18 Narrowbore, 2.1 mm (internal diameter)×150 mm length, 5 µm particles, Agilent Technologies) and the RP-HPLC separation was performed as described previously [48]. Fractions of 30 s wide were collected from 20 to 80 min after sample injection (120 fractions). To reduce LC-MS/MS analysis time, fractions eluting 12 min apart were pooled, vacuum dried and re-dissolved in 40 µl of 2% acetonitrile. In total, 24 pooled fractions per setup were subjected to LC-MS/MS analysis (see below).
LC-MS/MS analysis. LC-MS/MS analysis was performed using an Ultimate 3000 HPLC system (Dionex, Amsterdam, The Netherlands) in-line connected to a LTQ Orbitrap XL mass spectrometer (Thermo Electron, Bremen, Germany) and, per LC-MS/MS analysis, 2 µl of sample was consumed. LC-MS/MS analysis and generation of MS/MS peak lists were performed as described [52]. These MS/MS peak lists were then searched with Mascot using the Mascot Daemon interface (version 2.2.0, Matrix Science). The Mascot search parameters were set as follows. Searches were performed in the Swiss-Prot database with taxonomy set to human (UniProtKB/SwissProt database version 2010_05 containing 20,286 human protein sequences). Trideutero-acetylation at lysines, carbamidomethylation of cysteine and methionine oxidation to methionine-sulfoxide were set as fixed modifications. Variable modifications were trideutero-acetylation, acetylation and 13C2-acetylation of protein N-termini and pyroglutamate formation of N-terminal glutamine. Endoproteinase Arg-C/P (Arg-C specificity with arginine-proline cleavage allowed) was set as enzyme allowing no missed cleavages. The mass tolerance on the precursor ion was set to 10 ppm and on fragment ions to 0.5 Da. The estimated false discovery rate by searching decoy databases was typically found to lie between 2 and 4% on the spectrum level [48]. Quantification of the degree of N-Ac was done as previously explained [18].
Purified MBP-hNaa60p (500 nM) was mixed with selected oligopeptide substrates (200 µM) and 300 µM of acetyl-CoA in a total volume of 50 µl acetylation buffer (50 mM Tris (pH 8.5), 800 µM EDTA, 10% glycerol, 1 mM DTT) and incubated at 37°C for 35 min. The enzyme activities were quenched by the addition of 5 µl of 10% TFA. Peptide acetylation was quantified using RP-HPLC as described previously [23].
Peptides were custom-made (Biogenes) to a purity of 80–95%. All peptides contain 7 unique amino acids at their N-terminus, as these are the major determinants influencing N-terminal acetylation. The next 17 amino acids are essentially identical to the ACTH peptide sequence (RWGRPVGRRRRPVRVYP) however; lysines were replaced by arginines to minimize any potential interference by Nε-acetylation. Oligopeptide sequences:
SYSM-RRR (ACTH (aa138–161, P01189): [H] SYSMDHFRWGRPVGRRRRPVRVYP [OH]; MDEL-RRR (NF-kκB p65, Q04206): [H] MDELFPLRWGRPVGRRRRPVRVYP [OH]; MLGT-RRR (hnRNP H, P31943): [H] MLGTEGGRWGRPVGRRRRPVRVYP [OH]; MAPL-RRR (Prot phosphatase 6, O00743): [H] MAPLDLDRWGRPVGRRRRPVRVYP [OH]; MLGP-RRR (hnRNP F, P52597): [H] MLGPEGGRWGRPVGRRRRPVRVYP [OH]; SESS-RRR (High mob. gr. prot A1, P17096): [H] SESSSKSRWGRPVGRRRRPVRVYP [OH]; MKKS-RRR (Septin 9, Q9UHD8): [H] MKKSYSGRWGRPVGRRRRPVRVYP [OH].
The S. cerevisiae MATalpha strain BY4742 (Euroscarf) was transformed with pBEVY-U or pBEVY-U-hNAA60 and transformants were selected on plates lacking uracil. The two strains generated, BY4742-pBEVY-U (yeast normal) and BY4742-pBEVY-U-hNAA60 (yeast+NatF), were cultivated in 300 ml synthetic medium lacking uracil (Sigma) to an OD600nm of ∼3. After harvesting, cells were washed twice in lysis buffer (50 mM Tris, 12 mM EDTA, 250 mM NaCl, 140 mM Na2HPO4 (pH 7.6) supplemented with a complete protease inhibitor mixture tablet (1 tablet per 100 mL) (Roche Diagnostics) and glass beads were added before several rounds of vortex/ice (10×). One milliliter of lysis buffer was used for a pellet resulting from 300 mL of yeast culture. The lysates were centrifuged at 5000× g for 10 min and the retained supernatants were analyzed by COFRADIC analyses. Aliquots were analysed by SDS-PAGE and Western blotting using anti-hNaa60p. Solid guanidinium hydrochloride was added to a final concentration of 4 M in order to inactivate proteases and denature all proteins. Subsequently, proteins were reduced and alkylated simultaneously, using TCEP.HCl (1 mM final concentration (f.c.)) and IAA (2 mM f.c.) respectively, for 1 h at 30°C. Subsequent steps of the N-terminal COFRADIC protocol were performed as described previously [48]. Aliquots were analysed by SDS-PAGE and Western blotting using anti-hNaa60p.
HeLa cells (epithelial cervix adenocarcinoma, ATCC CCL-2) were cultured in Glutamax-containing DMEM medium supplemented with 10% dialyzed foetal bovine serum (Invitrogen, Carlsbad, CA, USA), 100 units/ml penicillin (Invitrogen) and 100 µg/ml streptomycin (Invitrogen). Cells were grown in media containing either natural (12C6) or 13C615N4 L-arginine (Cambridge Isotope Labs, Andover, MA, USA) [53] at a concentration of 80 µM (i.e. 20% of the suggested concentration present in DMEM at which L-arginine to proline conversion was not detectable for HeLa cells). Cells were cultured for at least six population doublings to ensure complete incorporation of the labeled arginine. Human K-562 cells (ATCC CCL-243) were grown in Glutamax-containing RPMI-1640 medium supplemented with 10% foetal calf serum, 100 units/ml penicillin and 100 µg/ml streptomycin. Cells were cultured at 37°C and in 5% CO2.
Plasmid transfections were performed using Fugene6 (Roche) according to the instruction manual. siRNA transfections were performed using Dharmafect 1 (Dharmacon). In the overexpression experiment, 10×10 cm dishes of cells cultivated in 13C615N4 L-arginine were transfected with control vector and cells cultivated in 12C6 L-arginine were transfected with phNAA60-V5. Cells were harvested 48 hours post-transfection. Aliquots were analysed by SDS-PAGE and Western blotting using anti-V5 (Invitrogen) to confirm efficient overexpression (See Figure 6A). In the knockdown experiment, 10×10 cm dishes of control control cells cultivated in 12C6 L-arginine were transfected with 50 nM si-non-targeting control (D-001810, Dharmacon) and cells cultivated in 13C615N4 L-arginine were transfected with 50 nM sihNAA60 pool (D-014479, Dharmacon). After 48 hours of transfection, the medium was replaced by new SILAC medium containing 5 µM zVAD-fmk. After 84 hours, cells were harvested, lysed and subjected to COFRADIC analysis as described previously [18]. Aliquots were analysed by SDS-PAGE and Western blotting using anti-hNaa60p (Custom made affinity purified rabbit antibody targeting a peptide corresponding to aa 69–82 of hNaa60p, Biogenes) to confirm efficient knockdown (See Figure 6B). Each sample of the knockdown- and overexpression experiments resulted from 10×10 cm dishes of cells and was processed further for N-terminal COFRADIC analyses as described previously [18].
The ratios of Nα-acetylation for all N-termini were quantified using MASCOT Distiller. The extent of Nα-acetylation was calculated after extracting the corresponding peak intensities (extracted from the resulting rov-files). The modified peptide sequences were used to calculate the theoretical isotope peak distribution using the MS-isotope pattern calculator (http://prospector.ucsf.edu). For both variants (i.e., in vivo Nα-acetylated (peak at m/z) and in vitro 13C2D3-Nα-acetylated (peak at m/z+5 Da)), the predicted intensity of the 5th contributing isotope was subtracted from the measured intensity of the corresponding monoisotopic peak of the other overlapping isotopic envelopes in order to correct for overlapping isotopic envelopes. Only the corresponding highest scoring MS/MS-spectra were withheld and inspected to evaluate the calculated Nα-acetylation degree (in case of inconsistencies, whenever possible the second, third or next highest scoring MS/MS-spectra were inspected to evaluate the calculated Nα-acetylation degree, if inconclusive the status was set as “N.D.”). When unclear MS-spectra were observed, the N-Ac status was also documented as “N.D.”. When no clear isotopic envelope was present for one of the possible variants, the status was set at 0% and 100% or 100% and 0% respectively. Further, if the Nα-acetylation calculated was ≤2% of ≥98%, the overall N-Ac status was accounted for as being free or fully N-Ac respectively.
When comparing the degrees of Nα-acetylation from two independent control experiments (with the degrees of Nα-acetylation of more than 1,000 unique N-termini calculated) and taking into account a [x−10%, x+10%] interval around the calculated x-value (the x-value being the degree (%) of Nα-acetylation for the calculated data point in one dataset), the p-value was calculated to be p<0.01, indicating that upon setting these limits, less than 1% of all measured N-Ac values differed more than 10%. Therefore, a significant variation in the degree of Nα-acetylation was set to 10% or more. In the case of free N-termini identified in a control setup however, significance was set to 5% since in this case two isotopic envelopes could clearly be distinguished.
Dmel2 cells were cultured at 25°C and RNAi was performed according to standard procedures. To deplete dNaa60 (CG18177), Dmel2 cells were separately transfected with two different double-stranded RNAs (dsRNA) corresponding to fragments of dNaa60 defined by the set of primers (Forward-1) CAACAAACACAGTGCGCC and (Reverse-1) CACATTTCGATAGGGTTTGATTTC or (Forward-2) GACTCGATGGGTCGTTCCGC and (Reverse-2) GTGGATGGCCGCCGTTAAT. GFP-targeting dsRNA was used as control. Each primer incorporates a T7 RNA polymerase binding site. All PCR products were used as template to synthesize dsRNA by use of the T7 RiboMAX Express kit (Promega). Drosophila Dmel2 cells were grown in SFM Medium (GIBCO) supplemented with 1× glutamine and 1× PenStrep (GIBCO). Cells were counted and diluted to 2×106 cells/ml in SFM medium supplemented with glutamine. Cells were incubated during 1 h with 40 µg for each dsRNA at a concentration of 1 µg/µl. After 1 h incubation with dsRNA, 3 ml of SFM media supplemented with glutamine and PenStrep (GIBCO) was added back. After 93 h dsRNA treatment, 2×106 cells were added to coverslips by 1 h incubation at 25°C. Cells were fixed with 4% formaldehyde, 0.03 M PIPES, 0.11 M HEPES, 0.01 M EGTA and 4 mM MgSO4 for 10 min, followed by two washes in 1× PBS. Permeabilization and blocking was performed for 1 h with PBS-TB (PBS, 0.1% Triton X-100, 1% fetal bovine serum). Primary antibody incubations were done in blocking solution for 2 h at room temperature or overnight at 4°C, followed by three 5 min washes in PBS-TB. Secondary antibody incubations were performed as described for the primary antibodies, including three 5 min washes. Primary antibodies included mouse anti-α-tubulin DM1A (1∶500; Sigma), rabbit anti-pSer10-Histone H3 (1∶500; Upstate Biotechnology), chicken anti-Cid (1∶500; kindly provided by David Glover's laboratory) and rabbit anti-Cnn (1∶500; kindly provided by Jordan Raff). F-actin was stained with rhodamine-conjugated phalloidin (Sigma) and DNA was stained with DAPI at 1∶1000 (stock concentration 1 mg/ml), with the addition of 5 µg/ml RNAse A. Visualization of fixed cells was performed using a Delta Vision Core System (Applied Precision) using a 100× UplanSApo objective and a cascade2 EMCCD camera (Photometrics). Images were acquired as a series of z-sections separated by 0.2-µm intervals. Deconvolution was performed using the conservative ratio method in softWoRx software. Phenotypic quantification was performed using a regular Epifluorescent microscope Leica DMRA2.
|
10.1371/journal.ppat.1006272 | The aromatic amino acid hydroxylase genes AAH1 and AAH2 in Toxoplasma gondii contribute to transmission in the cat | The Toxoplasma gondii genome contains two aromatic amino acid hydroxylase genes, AAH1 and AAH2 encode proteins that produce L-DOPA, which can serve as a precursor of catecholamine neurotransmitters. It has been suggested that this pathway elevates host dopamine levels thus making infected rodents less fearful of their definitive Felidae hosts. However, L-DOPA is also a structural precursor of melanins, secondary quinones, and dityrosine protein crosslinks, which are produced by many species. For example, dityrosine crosslinks are abundant in the oocyst walls of Eimeria and T. gondii, although their structural role has not been demonstrated, Here, we investigated the biology of AAH knockout parasites in the sexual reproductive cycle within cats. We found that ablation of the AAH genes resulted in reduced infection in the cat, lower oocyst yields, and decreased rates of sporulation. Our findings suggest that the AAH genes play a predominant role during infection in the gut of the definitive feline host.
| Toxoplasma gondii is an intracellular parasite that infects up to one-quarter of humans worldwide. Although it can infect virtually any warm-blooded animal, its definitive host is the cat where the sexual cycle occurs in enterocytes of the small intestine, producing microscopic, durable oocysts that are shed in feces and can remain infectious for extended periods of time in the environment. Two parasite genes, AAH1 and AAH2, code for aromatic amino acid hydroxylases, which produce L-DOPA, the precursor to dopamine. However, L-DOPA is also a precursor of other structural molecules including dityrosine, which may play a role in the wall of the oocyst. We investigated the effect of AAH deletion on the ability of the parasites to undergo sexual reproduction in cats, and found that AAH-deficient parasites were defective in their ability to produce oocysts, and those oocysts were partially defective in their ability to undergo maturation once produced. Collectively, these results suggest that the AAH genes play their primary role in transmission through the definitive host.
| Toxoplasma gondii is an obligate intracellular parasite and a member of the phylum Apicomplexa. It is related to Plasmodium spp., the causative agents of malaria, as well as parasites of human and veterinary importance including Cryptosporidium spp., Eimeria spp., and Neospora spp. T. gondii is one of the most widely distributed parasites in the world, and can be found on every continent and in virtually every species of warm-blooded animal investigated [1]. The definitive host of T. gondii is the cat, including all members of the family Felidae [2]. Within enterocytes of the cat intestine, T. gondii is capable of producing oocysts that are shed in the feces [3]. Oocysts are spheroid, 10–12 μm in size, and are comprised of an outer wall encapsulating two sporocysts that each contain four infectious sporozoites [4]. Oocysts are structurally robust with an elasticity and strength similar to common plastics [5]. They are very environmentally resilient, able to withstand a wide range of physical and chemical challenges including bleach, ethanol, acids, and bases [6], can stay infectious for years in the environment [7], and represent a significant source of dissemination for the parasite [8]. Omnivorous and herbivorous animals such as livestock can become infected by eating oocysts that contaminate rangeland, or by ingestion of contaminated water supplies [1]. Humans can also be infected by accidental ingestion of oocysts in contaminated food sources such as vegetables [9], or by ingestion of oocysts in water [10].
The walls of T. gondii oocysts are highly proteinaceous, composed of >90% protein [6], as well as β 1–3 glucan carbohydrates [11], and acid-fast lipids [12]. Large-scale proteomic analyses have identified 1,031 [13] or 1,304 [14] individual, non-redundant proteins associated with the oocyst. Although the function and localization of many remain unknown, two classes of oocyst wall structural proteins have been identified in other apicomplexans. In Cryptosporidium parvum, cysteine-rich COWPs (Cryptosporidium oocyst wall proteins) form a proteinaceous structure through extensive disulfide bridges [15]. Alternatively, tyrosine-rich EmGam (Eimeria gametocyte) proteins form a proteinaceous structure through extensive dityrosine linkages in the oocyst walls of Eimeria maxima [16–18]. The T. gondii genome contains seven cysteine-rich TgOWP proteins that are thought to be homologous to the COWPs. TgOWP proteins TgOWP1-3, were characterized and described in the outer oocyst walls but not the inner sporocyst walls [19]. Although T. gondii does not contain clear homologues of Eimeria’s EmGam proteins, many tyrosine-rich proteins have been identified in both outer oocyst wall and inner sporocyst wall fractions by mass spectrometry [13, 14], although they have not been definitively identified as structural components in the oocyst wall [5].
The genome of T. gondii contains two genes encoding aromatic amino acid hydroxylases referred to as AAH1 and AAH2 [20]. These genes encode predicted secretory proteins that catalyze conversion of phenylalanine to tyrosine, and tyrosine to 3,4 dihydroxyphenylalanine (L- DOPA) [20]. Conversion of tyrosine to L-DOPA is the rate-limiting step of dopamine synthesis in metazoans [21]. Although initial studies suggested that these enzymes are involved in modulating dopamine production in mammalian hosts [20, 22, 23], we were unable to replicate these findings in our previous work that focused on generating a knockout of AHH2 [24]. Moreover, our findings failed to reveal an elevated level of dopamine in chronically infected animals or in dopaminergic cells infected in vitro [24], consistent with recent reports by other authors [25, 26]. Hence, we sought to investigate other pathways that could require aromatic amino acid hydroxylase activity by T. gondii.
L-DOPA serves as a precursor to many structural components across other branches of eukaryotes, including helminths, molluscs, annelids, ascidians [27], and insects [28], and in coccidian apicomplexan parasites. In E. maxima, L-DOPA has been identified in the oocyst, where conversion of tyrosine to 3,4 dihydroxyphenylalanine (i.e. L-DOPA) on the tyrosine-rich EmGam precursor glycoproteins is an intermediate step in the formation of dityrosine crosslinks that provide structural strength to the Eimeria oocyst wall [16, 17]. Dityrosine has a strong blue auto fluorescence under UV light, a fluorescence observed in the oocysts of both Eimeria and T. gondii [29]. Furthermore, microarray data indicates that the AAH genes are upregulated during oocyst development in T. gondii [30], and protein mass spectrometry identifies both tyrosine-rich proteins and the tyrosine hydroxylases AAH1 and AAH2 in the oocyst of T. gondii [13, 14]. In contrast, these hydroxylases are not found in similar mass spectrometric analyses of tachyzoites or bradyzoites [14].
Here we sought to investigate the role of the T. gondii AAH genes in oocyst development using a combination of genetic, cellular, and biochemical studies. Although deletion of AAH2 alone caused a mild defect, ablation of AAH1, or loss of both genes, caused a severe defect in infection of the intestine and oocyst yield. Together, our results show that the AAH genes play an important role in parasite development during the sexual cycle in the intestinal epithelium of the cat.
Animal studies on mice were approved by the Institutional Animal Studies Committee (School of Medicine, Washington University in St. Louis). All procedures on cats were carried out in accordance with relevant guidelines and regulations following a protocol approved by the Beltsville Area Animal Care and Use Committee (BAACUC), United States Department of Agriculture, Beltsville, MD, USA.
Parasites were propagated by serial passage in human foreskin fibroblast (HFF (obtained from the laboratory of Dr. John Boothroyd, Stanford University School of Medicine)) cells grown in Dulbecco’s Modified Eagle Medium (DMEM) (Life Technologies, Carlsbad, CA) containing 10% fetal bovine serum (FBS) (Hyclone, Logan, UT) 10mM HEPES, pH 7.4, 1mM glutamine, 10 μg/mL gentamycin, under 5% CO2 at 37°C (D10 media). The parental ME49Δhxg::Luc strain was obtained from Laura Knoll (University of Wisconsin, Madison) [31]. A complete list of strains and clones used or generated in this study is provided in S1 Table. Tachyzoites were maintained by serial passage in HFF cells, grown as above. For induction of bradyzoites, cultures were switched to Roswell Park Memorial Institute 1640 medium (RPMI 1640), 50 mM HEPES pH 8.2 (Thermo Fisher Scientific, Grand Island, NY) and grown at 37°C without CO2, as described previously [32]. Cultures were determined to be free of mycoplasma using the e-Myco plus kit (iNtron Biotechnology). Parasites were harvested for experiments by scraping infected HFF monolayers into suspension, lysing HFFs and liberating tachyzoites by passage through a 20 g needle, and purifying tachyzoites with a 3.0 micron polycarbonate filter.
CNV data was obtained from an Illumina sequencing dataset of sixteen T. gondii reference strains and 46 non-reference strains, aligned using Bowtie2 using the end-to-end option [40].
A complete list of plasmids used or generated in this study is provided in S2 Table. CRISPR/Cas9 plasmids were adapted from the pSAG1:CAS9,U6:sgUPRT plasmid previously generated by our lab [33]. The guide RNA of the plasmid was modified to target the AAH2 5’ UTR by Q5 mutagenesis (New England Biolabs, Ipswich, MA), creating the plasmid pSAG1:CAS9,U6:sgAAH2. A second guide RNA expression cassette targeting the AAH2 3’ UTR was inserted into the same plasmid backbone by traditional cloning steps to create the CRISPR/Cas9 AAH2 double cut plasmid pSAG1:CAS9,U6:dgAAH2. The same plasmid backbone was similarly adapted to target the AAH1 5’ and 3’ UTRs (pSAG1:CAS9,U6:sgAAH1 and pSAG1:CAS9,U6:dgAAH1). The pSAG1:CAS9,U6:sgUPRT plasmid described previously [33] was also modified to create a double-cutting CRISPR/Cas9 plasmid targeting the HXGPRT gene pSAG1:CAS9,U6:dgHXGPRT [34]. Plasmids used to generate the Δaah2 knockout using the HXGPRT selectable marker to replace the gene in the ME49Δhxg::Luc strain [31], and to restore expression of AAH2 were described previously [24]. Plasmids used to generate the Δaah1 mutant by replacement with the selectable marker DHFR-Ts, and to complement expression with a cDNA construct targeted to the uracil phosphoribosyl transferase (UPRT) locus, were created using Gibson assembly (New England Biolabs).
To generate transgenic ME49 parasites, 107 tachyzoites, harvested as described above, were mixed with 5μg of CRISPR plasmids and 15μg of the appropriate homologous repair construct as plasmids linearized by restriction digest. Parasites were transfected by electroporation, and allowed to recover on HFF monolayers for 24 h. Positive selection for the HXGPRT cassette was done with 25 μg/mL mycophenolic acid (Sigma-Aldrich, St. Louis, MO) supplemented with 50 μg/mL xanthine (Sigma-Aldrich) [34]. Negative selection against the HXGPRT cassette was done with 340 μg/mL 6-Thioxanthine (Toronto Research Chemicals, Toronto, ON) [35]. Positive selection for the DHFR-Ts construct was done with 5μM pyrimethamine (Sigma-Aldrich) [36]. Negative selection against the UPRT locus was done with 10μM 5-fluorodeoxyuracil (FUDR) (Sigma-Aldrich) [37]. Clones were isolated by limiting dilution in 96-well plates containing HFF monolayers, grown as above. Clones were screened by PCR against the selectable marker and the AAH genes (S3 Table).
Parasites were seeded into T-25s containing monolayers of HFF cells and allowed to invade and grow for 24 h. Infected T-25s were then rinsed three times with PBS to remove any extracellular parasites. Intracellular parasites were harvested as previously described, counted by hemocytometer and seeded into 96-well plates containing monolayers of HFFs with fresh D10 media at a concentration of 105 parasites per well. Plates were allowed to grow for 24 h before being lysed with 30uL of 1x Cell Culture Lysis Reagent (Promega, Madison, WI). Luminescence was developed with the Luciferase Assay Kit (Promega), and imaged on a Cytation 3 imaging system (Biotek, Winooski, VT).
Parasites were harvested as previously described, counted by hemocytometer, and diluted into PBS. Eight-week old female CD1 mice (Charles River Laboratories, Wilmington, MA) were injected i.p. in a volume of 200 μL PBS containing 103 parasites and monitored daily. One month post-infection, mice were euthanized by CO2 asphyxiation followed by cervical dislocation. Brains were removed, homogenized by passage through a 20 g needle, and stained with Dolicos biflorus lectin (DBL) as previously described [38]. Fifteen μL of stained homogenate was examined using a Zeiss wide-field epifluorescence microscope. Three separate aliquots were counted per brain sample, and total brain cyst load was determined based on the total volume of the brain homogenate and the average count per 15 μL.
All procedures described here were carried out in accordance with relevant guidelines and regulations following a protocol approved by the Beltsville Area Animal Care and Use Committee (BAACUC), United States Department of Agriculture, Beltsville, MD, USA. T. gondii -free kittens (10- to 12-week old) were used to study T. gondii infections. Briefly, T. gondii infected mouse brains were homogenized by syringe and fed to the cats by placing them at the back of the tongue. All feces for each cat were collected daily after feeding infected mouse brains, and examined for T. gondii oocysts. The screening and harvesting of T. gondii oocysts were done between 3 to 21 days after infection by following procedures as described previously [1]. Cats were euthanized on day 21 post infection and blood was collected to do modified agglutination tests (MAT) to test for immunological reactivity to T. gondii antigens. Oocysts were collected by floatation methods using sucrose solution with a specific gravity of 1.15 or higher. Concentrated oocyst pellets were suspended in an aqueous solution containing 2% H2SO4, and aerated on the shaker for 7 days at room temperature (20–22°C) to allow for oocyst sporulation. Oocysts were counted using a disposable hemocytometer. Total oocysts shed by individual cats were calculated based on total counts, dilution factor, and total volume.
For histological studies, infected cats were euthanized at day 6/7 and portions of intestinal ileum were fixed in 10% buffered neutral formalin. Fixed tissues were cut into sections (2.5 x 0.7 cm), placed in cassettes, embedded in paraffin, and sectioned 4–5 μm thick. Slides were deparaffinized, rehydrated, and stained with hematoxylin and eosin (Leica Microsystems, Buffalo Grove, IL), or by immunohistochemistry with Rabbit anti-RH polyclonal antibody [39] and Streptavidin-HRP (Jackson Labs, West Grove, PA), according to standard protocols [1].
Images were taken on a Zeiss AxioSkop wide field epifluorescence microscope equipped with AxioCam CCD camera and images were captured using AxioVision v3.1 (Carl Zeiss Inc., Thornwood, NY). For each image, 10 μL of oocyst-laden cat fecal suspension were placed on a slide and imaged with a DAPI filter (300–390 nm excitation, 420 nm emission).
Statistical analysis was done in Prism 6 for Mac OSX (GraphPad Software, La Jolla, CA). One-way and two-way ANOVAs for parametric data sets and Kruskal-Wallis tests for nonparametric data sets were conducted with a threshold of P ≤ 0.05 considered significant.
Previous studies have described two genes AAH1 and AAH2 that are very closely related and located on chromosome V (ToxoDB ver. 8 ME49 genome) [40]. Analysis of copy number variation (CNV) of AAH2 TgME49_212740 showed approximately two copies in the type 1 strain GT1, the type 2 strain Pru, the type 3 strain VEG and the type 10 strain VAND (Fig 1A). In contrast, the type 2 strain ME49 had a CNV level consistent with three copies (Fig 1A). Although AAH1 and AAH2 genes appeared as tandem loci in ToxoDBv8, v9 and subsequent assemblies placed AAH1 (TgME49_087510) on an unassembled contig (KE139705), and contains an additional third gene consistent with AAH2 (TgME49_212710) on another unassembled contig (KE139818), while recognizing only one tyrosine hydroxylase AAH2 within the parasite genome itself, located on chromosome V.
Mapping reads across each base pair of the AAH2 locus showed a consistent CNV of approximately 3 across the coding region of AAH2 (Fig 1B). To further examine the nature of the predicted third copy, we amplified the 3’ region of AAH1/ AAH2 using primers common to both genes (Fig 1C). We then interrogated the nature of the alleles present in the ME49 strain using Sanger sequencing. Inspection of the chromatographs from Sanger sequencing indicated a 2:1 ratio of AAH2 to AAH1 single nucleotide polymorphisms (SNPs), consistent with a duplication of AAH2 in ME49 (Fig 1D). These sequencing results also confirmed the ToxoDB ver. 8 arrangement of flanking regions for AAH1 and AAH2.
We previously reported that deletion of AAH2 in the type 2 Pru strain has no effect on growth in vitro or development of bradyzoites in vivo [24]. To examine the ability of Δaah2 mutants to be passaged through cats, we decided to generate a similar Δaah2 deletion in the type 2 ME49 strain, which has a high capacity for oocyst generation. We targeted the AAH2 gene for replacement with the HXGPRT selectable marker in the ME49Δhxg::Luc strain (referred to as wild type (WT)), which has a deletion on the hxgprt locus and is also tagged with firefly luciferase. To efficiently delete the AAH2 gene, a CRISPR/Cas9 plasmid containing two guide RNAs targeting the 5’ and 3’ UTRs of AAH2 was created (Fig 2A) (S2 Table). This double-cutting plasmid was co-transfected into the parental WT strain with an HXGPRT drug resistance cassette targeted to the AAH2 locus to create the clone Δaah2::HXG (S1 Table). Sanger sequencing of this clone revealed that both copies of AHH2 had been removed, while the AHH1 gene remained intact (Fig 1D). To remove the HXGPRT selectable marker, a CRISPR/Cas9 double-cutter of HXGPRT was co-transfected with an aah2-null fusion construct of the AAH2 5’ and 3’ UTRs (pΔaah2) or a complement construct of its 5’ and 3’ UTRs appended to a cDNA copy of AAH2 (pAAH2) to make the clean knockout clone Δaah2 (referred to as Δh2) and the complement clone Δaah2::AAH2 (referred to as Δh2-H2), which restores expression of AAH2 (Fig 2A).
Subsequently, to knock out AAH1, we created a double-cutting CRISPR/Cas9 construct targeted to the UTRs of the AAH1 gene, and co-transfected it with a Δaah1::DHFR-Ts construct (pΔaah1::DHFR-Ts) (S2 Table) into WT or Δh2 strains to make the clones Δaah1 (referred to as Δh1) and Δaah1Δaah2 (referred to as Δh1Δh2) (Fig 2B) (S1 Table). To restore AAH1, we co-transfected the pSAG1:CAS9,U6:sgUPRT CRISPR plasmid with a repair construct containing HXGPRT and a cDNA copy of AAH1 to create the clones Δaah1-AAH1 (referred to as Δh1-H1) and Δaah1Δaah2-AAH1 (referred to as Δh1Δh2-H1) (S1 Table).
Having generated a single knockout of each of the Δaah1 and Δaah2, as well as the double Δaah1Δaah2 knockout and several complemented strains, we decided to compare their growth and differentiation abilities in vitro and in vivo. Consistent with the fact that we were able to obtain the mutants readily in culture without any apparent growth defect, their growth as tachyzoites was similar when compared using a highly quantitative luciferase assay (Fig 2D). We also tested their ability to differentiate to bradyzoites in vitro under conditions of pH 8.2 stress, as assessed by staining with Dolichos biflorus lectin, which stains carbohydrates in the cyst wall. We observed that the ability of the knockout and complemented strains to differentiate into bradyzoites was unaffected (Fig 3A and 3B). Additionally, these strains were injected into mice in order to assess their ability to form cysts in the brains of chronically infected mice. Loss of the AAH1 or AAH2 genes did not affect the ability to produce cysts in the mouse brain, and although the complementation of the double Δaah1Δaah2 (Δh1Δh2) with the AAH1 gene showed slightly higher cyst burdens, this was not significant (Fig 3C). The lack of a discernable phenotype on the development of bradyzoites is consistent with our previous studies in the Pru strain, albeit this was previously only tested with the Δaah2 mutant [24].
To investigate development during the sexual cycle, tissue cysts contained in mouse brain homogenate were fed orally to cats and oocyst shedding was monitored. The normal prepatent period for oocyst shedding following infection with bradyzoites is 3–5 days with peak shedding from 5–8 days [2]. Consistent with this, cats that showed oocysts shedding commenced within the first week. However, to be sure we collected all of the oocysts produced, we extended the observation period to 21 days. Infection with the WT strain consistently yielded around 106−107 total oocysts shed during this time period (Fig 4A). Although the Δaah2 (Δh2) mutant yielded much lower levels of oocyst in two of three cats, a third animal showed only ~ 10 fold reduction to ~105 total oocysts (Fig 4A). In contrast, the Δaah1 mutant (Δh1) and Δaah1Δaah2 double mutant (Δh1Δh2) showed a severe defect in oocyst yield in two of two cats tested, leading to only ~103 total oocysts per animal (Fig 4A). The differences observed in these animals were significant when the knockout strains were compared as a whole to the wild type (Fig 4A). However, they did not reach statistical significance when compared individually to the wild type (Fig 4A), due to the low sample sizes used. Given the magnitude of the phenotype, and the consistency among mutants, we did not feel it was worthwhile to use more animals simply to achieve an arbitrary level of statistical significance. The moderate defect in the Δaah2 (Δh2), and the very severe defect in both the Δaah1 (Δh1) and the double Δaah1Δaah2 (Δh1Δh2) knockouts, were fully restored in the respective complemented strains (Fig 4A).
We also tested the ability of shed oocysts to undergo sporulation, since meiosis occurs after oocyst shedding. The sporulation rate is a measure of viability as unless oocyst mature to form sporozoites, they remain non-infectious [4]. Wild type oocysts showed a successful sporulation rate of 75–80% and this dropped significantly to ~ 60% in the Δaah2 (Δh2) (Fig 4B). Oocyst shedding was so low that we were not able to adequately quantify the efficiency of sporulation in the single Δaah1 (Δh1) and double Δaah1Δaah2 (Δh1Δh2) mutants (Fig 4B); however, based upon very limited counts, the sporulation success rate of these strains varied from 10–50% across samples. Complementation of AAH1 to the Δaah1 (Δh1) single knockout or the Δaah1Δaah2 (Δh1Δh2) double knockout partially rescued sporulation efficiency (Fig 4B). Dityrosine fluorescence is normally much stronger on the inner sporocyst walls, and consequently the intensity of fluorescence under UV illumination was lower in unsporulated oocysts (Fig 5). Although the single and double mutants showed variable defects in the extent of sporulation (Fig 4B), when oocyst sporulation was normal, the resulting fluorescence of the inner sporocyst walls was similar among all the strains tested (Fig 5). We successfully hatched Δaah2 oocysts and recovered them back into in vitro culture as tachyzoites, indicating that the oocysts that appeared to develop successfully were viable. However, the yield of the Δaah1 and Δaah1Δaah2 knockouts was too low to allow for this method of recovery.
We reasoned that any defect during asexual expansion in the cat intestine or during the sexual cycle could cause a block that resulted in fewer oocysts being formed. Infection in the cat intestine initially proceeds though asexual expansion, termed A-E forms, which divide by endodyogeny and schizogony, before sexual development commences with the formation of macro and microgamocytes [2]. This process culminates with the exflaggelation of microgametes followed by fertilization of the macrogamete to yield a zygote that matures into an oocyst [2]. To examine the parasite infectivity and development of stages that occur in the cat intestine, we euthanized animals during the initial phase of oocyst shedding and examined tissue sections by conventional histology. In tissue sections from cats infected with the wild type (WT), Δaah1 (Δh1) and Δaah2 (Δh2) parasites taken at 6–7 days post-infection, parasite infection of the intestinal ileum was readily seen (Fig 6A–6C). However, the Δaah1 parasites showed a significant defect in overall density of infection (Fig 6D). We were readily able to recognize merozoites, schizonts, microgamonts and macrogamonts, indicating that these lines grow well in the gut (Fig 7). Although the density of infection was lower in the Δaah1 mutant (Fig 8A), the relative distribution of parasite stages was not significantly different (Fig 8B, S4 Table), ruling out the possibility of a defect in any specific stage of parasite sexual development inside the intestinal ileum. Collectively, these findings indicate that AAH1 plays a role in infection in the cat intestine, and that both AAH1 and AAH2 affect the efficiency of oocyst formation in vivo, and to a lesser extent the sporulation efficiency, and that these phenotypes are partially penetrant.
Previous studies have suggested that the presence of aromatic amino acid hydroxylase genes AAH1 and AAH2 in T. gondii may be an adaptation for altering host dopamine levels and thereby affecting behavior [20, 22, 23]. However, in prior studies [24] we were not able to replicate the association between T. gondii infection and elevated dopamine that was seen in mice [41] or in dopaminergic cell lines [22]. Additionally, alternative explanations for the AAH genes are provided by studies showing that oocyst walls of E. maxima [16, 17] contain dityrosine crosslinks, and fluorescence under UV illumination suggests similar modifications exist in T. gondii oocyst walls [29]. To resolve the potential role of the T. gondii AAH genes in oocyst formation, we disrupted one or both genes using CRIPSR-based genome editing [33]. Our findings reveal that AAH2 plays a moderate role, while AAH1 plays a much stronger role in formation of oocysts during infection in the cat. Additionally, AAH1 may play a role in parasite survival inside the cat intestinal epithelium as it showed a defect in infectivity even at early stages of merogony and schizogony. It is possible that dityrosine or other L-DOPA derived products produced by these AAH genes play a protective role in shielding or cloaking the parasite from the host’s innate immune response in a manner analogous to the role of melanin in Cryptococcus neoformans [42], or the AAH genes may play an additional role in nutrient availability for the parasite, converting scavenged phenylalanine to tyrosine or vice-versa. Although these findings do not rule out a CNS role for the AAH genes, they suggest that one primary function is during infection in the cat intestine, leading to formation of mature oocysts.
Although the AAH genes of T. gondii have been proposed as candidate effectors for the parasite’s ability to manipulate host behavior via manipulating dopamine in the host [23, 25, 41, 43–48], our previous work failed to reproduce the parasite’s described ability to exert effects upon host dopamine levels [24], consistent with other reports [25, 26]. Further, inconsistencies in cat-aversive behavior and other reported behavioral changes including anxiety, activity level, learning, memory, and more, challenge the robustness of this behavioral manipulation [25, 26, 46, 49–53]. Finally, the hypothesis that tissue cysts of brain-resident parasites actively alter host dopamine to exert behavioral control faces exceptional challenge from the observation that parasites defective in their ability to establish lifelong residency in the brain still result in abnormal cat attraction [51]. Additionally, the expression of the AAH genes is relatively low in both the lytic and chronic asexual stages [24] and is only upregulated in the sexual stages [30] and mass spectrometry has failed to find evidence of these proteins in tachyzoite or bradyzoite stages but identified them in the oocyst [14]. Hence if the AAH gene products are involved in altering dopamine levels in the CNS of infected rodents, they would need to do so based on exceedingly low expression levels, and in a localized region. We are presently examining neurotransmitter levels and behavioral change in mice infected with AAH mutants describe here, and such studies could potentially resolve the role of these genes in such pathways.
Because of the high variability in findings regarding the effects of T. gondii infection on brain neurotransmitters and behaviors, we sought to explore alternate roles for these genes in the parasite life cycle. One obvious candidate would be the contribution of L-DOPA to the formation of protein-protein dityrosine crosslinks in the proteinaceous oocyst wall, analogous to what has been described in E. maxima [16, 17]. Recently, the oocyst wall proteins TgOWP1-7, which are cysteine-rich structural proteins analogous to the Cryptosporidium oocyst wall proteins, were characterized and shown to localize to the outer oocyst wall, but not the inner sporocyst walls [19]. Mass spectrometry data also reveal that tyrosine rich proteins are found in oocysts [13, 14], but as yet there is not direct biochemical evidence for dityrosine cross linked proteins in the oocyst wall. However, consistent with the presence of such crosslinks, both the outer oocyst wall and inner sporocyst walls show dityrosine fluorescence, although the signal is significantly brighter in the sporocyst walls. Using the efficiency of CRISPR/Cas9 to direct genetic disruption, we demonstrated that ablation of AAH1 or both AAH1 and AAH2 causes a severe defect in oocyst yield, as well as a maturation defect in the oocysts that do emerge. Parasites ablated for AAH1 were compromised in replication and development during growth in the cat intestine, and parasites ablated for AAH2 were able to develop normally within the cat intestine but were compromised in their yield and maturation efficiency after shedding into the environment.
One potential function for the AAH genes is in generating modified tyrosine residues (i.e. 3,4 dihydroxyphenylalanine) that are the precursor for dityrosine crosslinks in oocyst wall proteins. This modification is expected to increase oocyst resistance to environmental conditions. The observed decrease in oocyst yield from the aah mutants following purification from cat feces is consistent with them being more fragile and prone to loss during the intensive process of osmolar, physical, and chemical treatments that are used during isolation. Although we were able to recover a small number of oocysts from the mutants, they underwent sporulation less efficiently. Since sporulation is associated with increased levels of UV fluorescence, the reduced rate of sporulation in the aah mutants is consistent with formation of fewer dityrosine crosslinks. However, some oocysts shed by the mutants were able to undergo sporulation and form oocysts with normal UV fluorescence, although at a much lower total numbers than the wild type. This suggests that if the AAH enzymes normally participate in dityrosine crosslinks, this function can be rescued in the absence of the parasite genes, albeit inefficiently. In this regard, there are at least two other potential sources for 3,4 dihydroxyphenylalanine that serves as a precursor for this reaction: the host cell and the microbiome. Hence, it is possible that salvage from these other sources may enable T. gondii to generate dityrosine crosslinks at a lower frequency in the absence of AAH genes.
Combined with previous findings, our results suggest that T. gondii builds its oocyst walls using a hybrid strategy combining features of Cryptosporidium’s cysteine-cross-linked walls and Eimeria’s dityrosine-cross-linked walls. We hypothesize that the proteinaceous part of the outer oocyst wall of T. gondii is predominantly Cryptosporidium-like, composed of TgOWPs cross-linked by disulfides. A secondary Eimeria-like component of tyrosine-rich proteins cross-linked by dityrosines comprises the proteinaceous inner sporocyst walls in T. gondii oocysts. In this model, the aromatic amino acid hydroxylases AAH1 and AAH2 are expected to catalyze the conversion of tyrosine residues on wall proteins into 3,4 dihydroxyphenylalanine residues for subsequent dityrosine bond formation. The final conversion of these residues into cross-linked proteins is also likely to require a peroxidase, and a putative oxidoreductase that reliably emerges as the most abundant protein in mass spectrometry analyses provides a candidate for this activity [13, 14, 30]. The reduction in infectivity in the AAH1 mutant suggests that dityrosine or secondary quinones may also play a role as a virulence factor throughout earlier stages of development, analogous to the role of melanin in the neurotropic yeast Cryptococcus neoformans [42]. Alternately, the AAH genes may be involved in the conversion of phenylalanine to tyrosine to cope with nutrient limitations for growth in vivo. To test these models, further studies would be needed to define the localization of the putative tyrosine-rich protein precursors, confirm the presence of dityrosine crosslinks, and investigate the interaction of the AAH enzymes with such substrates during sexual stage and oocyst development. However, at present such studies are hindered by the necessity for sexual development to take place in the complex environment of the cat intestine. However, further exploration of these pathways may also be of value for defining attenuated mutants of T. gondii that are unable to yield infectious oocysts and yet which may induce protective immunity in the cat, thus potentially breaking transmission of the life cycle.
|
Subsets and Splits